arxiv_id
stringlengths 11
13
| markdown
stringlengths 2.09k
423k
| paper_doi
stringlengths 13
47
⌀ | paper_authors
listlengths 1
1.37k
| paper_published_date
stringdate 2014-06-03 15:22:49
2024-08-02 17:59:51
| paper_updated_date
stringdate 2014-06-03 15:22:49
2025-04-29 20:34:17
| categories
listlengths 1
7
| title
stringlengths 15
236
| summary
stringlengths 57
2.54k
|
|---|---|---|---|---|---|---|---|---|
1408.0739v1
|
## The sensitivity of pulsar timing arrays
## C J Moore
Institute of Astronomy, Madingley Road, Cambridge, CB3 0HA, United Kingdom
E-mail:
[email protected]
Abstract. Simple analytic expressions are derived for the sensitivity curve of a pulsar timing array (PTA) to both a monochromatic source of gravitational waves and an isotropic stochastic background of gravitational waves. These derivations are performed in both frequentist and Bayesian frameworks and the results verified with numerical injections and recovery in mock PTA datasets. The consistency of the frequentist and Bayesian approaches is demonstrated and the results are used to emphasise the fact that the sensitivity curve of a PTA depends not only on the properties of the PTA itself, but also on the properties of the gravitational wave source being observed.
## 1. Introduction
Pulsar timing arrays (PTAs) are ongoing attempts to detect low frequency gravitational waves (GWs) by using the timing of a network of galactic millisecond pulsars. The chosen millisecond pulsars are extremely stable rotators which allow a timing model of sufficient accuracy to be constructed to account for every single rotation of the pulsar over the many years of observation. After subtracting the timing model from the pulse arrival times the remaining timing residuals include any unmodeled effects; e.g. timing noise and (possibly) GWs. All of the pulsars reside in the same GW field and hence the induced timing residuals in different pulsars are correlated by an amount that depends on their angular separation. It is this correlation which allows a network of pulsars spread across the sky to be exploited as a GW detector.
PTAs are sensitive to GWs in the range 10 -9 Hz glyph[lessorsimilar] f glyph[lessorsimilar] 10 -6 Hz and are typically most sensitive to the lowest frequencies in this range. Potential sources of GWs in this frequency band are individual supermassive black hole binaries with chirp masses around M≈ 10 9 M glyph[circledot] out to redshifts of z ≈ 2 ((1; 2)). If there are many sources per unit frequency then they cease to be individually resolvable and instead overlap to form a stochastic background of GWs ((3; 4; 5)). It is currently unclear whether the first detection of GWs by PTAs will be an individual binary or a stochastic background. Stochastic backgrounds may also be caused by other mechanisms such as cosmic string networks (6), or primordial GWs ((7; 8)).
There are currently three PTAs operating: the European Pulsar Timing Array (EPTA , (9)), 1 the Parkes Pulsar Timing Array (PPTA , (10)) and the North American Nanohertz Observatory 2 for Gravitational Waves (NANOGrav , (11)). 3 There are plans to merge the seperate PTAs to form the International Pulsar Timing Array (IPTA , (12)). 4
These proceedings are based on a talk given at the 10th International LISA Symposium, this talk in turn was based largely on the work done in (13). In Sec. 2 of this paper the properties of the PTA which will be assumed throughout the remainder of the paper are summarised, Sec.
1 http://www.epta.eu.org/
2 http://www.atnf.csiro.au/research/pulsar/ppta/
3 http://nanograv.org/
4 http://www.ipta4gw.org/
3 describes the sensitivity of this PTA to an individual binary, Sec. 4 describes the sensitivity to a stochastic background and final concluding remarks are given in Sec. 5.
## 2. Our canonical PTA
In order to produce definite values for the sensitivity curves it is necessary to assume particular values for the PTA parameters. For all calculations in this paper the PTA was assumed to consist of 36 pulsars randomly placed on the sky all timed with a cadence of 1 / (2 weeks) for a total of 5 years with a precision of 100 ns. This is comparable to the Open1 dataset in the recent IPTA data challenge . It should be noted that the performance of this canonical PTA is 5 significantly better than any of the current PTAs.
## 3. Sensitivity to an individual binary
Binary black holes, in contrast to the stochastic backgrounds considered in Sec. 4, are a GW source which can be accurately modelled over the entire duration of observations; hence matched templates may be constructed for the purpose of detection and parameter estimation. PTAs are sensitive to individual binary black holes with very large chirp masses (10 8 M glyph[circledot] glyph[lessorsimilar] M glyph[lessorsimilar] 10 10 M glyph[circledot] ) in the early, adiabatic stage of their inspiral. These systems are typically in circular orbits evolving on timescales longer than the total length of our observations and appear in the data as quasi-monochromatic signals. Therefore, for simplicity, in this section we consider the sensitivity of our PTA to a monochromatic source of GWs as a proxy for an individual binary.
## 3.1. Frequentist detection
The frequentist procedure for claiming a detection of a signal in the PTA dataset is to define a scalar detection statistic, S , which is a function of the dataset. The signal to noise ratio (SNR) of this statistic, glyph[rho1] , is then defined as the expectation of this statistic in the presence of a signal divided by the square root of the variance of the statistic in the absence of a signal. A detection is claimed if the value of the detection statistic evaluated on the actual measured data exceeds some predetermined threshold, glyph[rho1] th (the value of this threshold may be tuned to fix the false alarm rate).
Here the detection statistic was chosen to be the cross correlation of two data streams with the optimal (or 'matched') filter function, ˜ ( K f ); see Eq. 1. This statistic has the advantage that it gives an SNR which is quartic in the signal amplitude (see Eq. 1, we tentatively venture to call this the 'power SNR' in contrast to the usual 'amplitude' SNR) and hence may be directly compared with the stochastic background SNR (see Sec. 4). It should be noted that in the case of a single binary it is also possible to use the usual matched filter statistic and associated 'amplitude' SNR for the purposes of detection, but in the case of the stochastic background one is forced to use the cross correlation statistic because the signal in each pulsar is swamped by the intrinsic pulsar red-noise.
$$\mathcal { S } & = \int \, \mathrm d t \, \int \, \mathrm d t ^ { \prime } \, \mathbf s ( t ) ^ { T } \mathbf K ^ { \dagger } ( t - t ^ { \prime } ) \mathbf s ( t ^ { \prime } ) = \int \, \mathrm d f \, \tilde { \mathbf s } ( f ) ^ { T } \tilde { \mathbf K } ^ { \dagger } ( f ) \tilde { \mathbf s } ( f ) \\ \tilde { \mathbf K } ( f ) & = \frac { \tilde { \mathbf h } ( f ) \tilde { \mathbf h } ( f ) ^ { T } } { S _ { n } ^ { 2 } } \Big | _ { \text{diag} \rightarrow 0 } \, \Rightarrow \, \varrho ^ { 2 } = \sum _ { y } \sum _ { x > y } \frac { 8 } { T } \, \mathrm d f \, \frac { \left | \tilde { h } _ { x } ( f ) \right | ^ { 2 } \left | \tilde { h } _ { y } ( f ) \right | ^ { 2 } } { S _ { n } ^ { 2 } } \,.$$
In evaluating the SNR in Eq. 1 the noise terms in the timing residuals from pulsars x and y have been assumed to be white, Gaussian and uncorrelated between pulsars. If pulsar x is timed with precision σ x and with a cadence 1 /δt x then this leads to the familiar expression
$$\langle \tilde { n } _ { x } ^ { * } ( f ) \tilde { n } _ { y } ( f ^ { \prime } ) \rangle = \frac { 1 } { 2 } S _ { n } ( f ) \delta _ { x y } \delta ( f - f ^ { \prime } ) \quad \text{where} \ \ S _ { n } = 2 \sigma _ { x } ^ { 2 } \delta t _ { x }.$$
The red curve in Fig. 1 shows the sensitivity curve obtained by setting glyph[rho1] = glyph[rho1] th in Eq. 1 and rearranging to find the amplitude as a function of frequency. This sensitivity curve has
5 http://www.ipta4gw.org/?page\_id=89
the obvious problem that it tends to a constant at low frequencies. This is because the need to use the measured data to fit for the free parameters in the pulsar timing model hasn't been accounted for. Heuristically the effect of this fitting may be understood by approximating the timing model as a quadratic in time. To motivate the choice of a quadratic notice that a constant offset in the timing residuals is degenerate with mis-estimating the distance to the pulsar, a linear drift in the timing residuals is degenerate with the pulsar spin and a quadratic trend in the timing residuals is degenerate with the pulsar spin-down rate. Expanding the sinusoidal signal at low frequencies ( ft glyph[lessmuch] 1) as a Taylor series in powers of ft allows us to include the effect of fitting for the quadratic timing model (see (13)) by subtracting off the first three terms in this series (up to and including O ( ft ) 2 ). The resulting sensitivity curve is shown in Fig. 1. This procedure is rather ad-hoc , a much more satisfactory way of accounting for the timing model is provided by the Bayesian approach, which we now discuss.
## 3.2. Bayesian detection
The Bayesian procedure for claiming a detection of a signal in the PTA dataset is to consider two competing hypotheses which attempt to describe the measured data, s : the noise hypothesis, {H n : s = n + m } ; and the signal hypothesis, {H h : s = n + m + h } . Here n denotes the random noise, m the timing model and h the signal; any of n m , or h may depend on a number of free parameters, glyph[vector] θ . For each hypothesis i ∈ { n, h } the likelihood, L i , may be evaluated for a given choice of the parameters. The evidence for each hypothesis, O i , may be calculated by integrating, or marginalising , the likelihood over all of the free parameters in that hypothesis (weighted by any prior knowledge of each parameter). The odds ratio (or Bayes factor) is then defined as the ratio of the evidences B = O h / O n and a detection is claimed if the Bayes factor exceeds some predetermined threshold, B th . The threshold Bayes factor is the Bayesian equivalent of the frequentist threshold SNR, and B th may be chosen to fix a desired false alarm rate.
Here the timing model was taken to be a quadratic in each pulsar, m t x ( ) = α x + β t x + γ t x 2 , where α x , β x and γ x are free parameters with flat priors. These parameters are then marginalised over when evaluating the evidence for each hypothesis, this provides a rigorous method for accounting for the loss of sensitivity at low frequencies caused by fitting for the timing model. In reality the timing model is significantly more complex than a simple quadratic; it must account for effects as diverse as the Roemer delay, dispersion of the pulse by the interstellar medium and the orbital motion of the pulsar (if it happens to be in a binary) ((14; 15)). However the simple quadratic timing model considered here captures well the qualitative effect of fitting for the pulsar parameters and has the attractive feature that the free parameters ( α x , β x and γ x ) may be marginalised over analytically.
The likelihood functions for both hypotheses may be written in terms of the noise covariance matrix, Σ n = σ 1 , where it has been assumed for simplicity that σ x = σ, ∀ x ,
$$\log \mathcal { L } _ { n } ( \vec { \Theta } ) = \log A - \frac { 1 } { 2 } \left ( s - m ( \vec { \Theta } ) \right ) ^ { T } \Sigma _ { n } ^ { - 1 } \left ( s - m ( \vec { \Theta } ) \right ) \, \\ \log \mathcal { L } _ { h } ( \vec { \Theta }, \vec { \Psi } ) = \log A - \frac { 1 } { 2 } \left ( s - m ( \vec { \Theta } ) - h ( \vec { \Psi } ) \right ) ^ { T } \Sigma _ { n } ^ { - 1 } \left ( s - m ( \vec { \Theta } ) - h ( \vec { \Psi } ) \right ) \.$$
The evidence for each hypothesis may be evaluated by marginalising over the free parameters; for the pulsar parameters, Θ, flat priors were used, for the source parameters, Ψ, delta function glyph[vector] glyph[vector] priors at the true values were used. The choice of delta function priors may seem unreasonable at first, but in reality it is simply a convenient way of imposing a large signal approximation. A detection of GWs is only claimed if the Bayes factor exceeds some threshold, this threshold must be chosen to be large if we are to have high confidence in the detection. In the limit of a large Bayes factor the prior becomes uninformative compared to the data and the value of the Bayes factor obtained in the experiment is independent of the choice of prior. Therefore a delta-function prior is chosen to ease the analytic calculations. For a further discussion of this
point see (13). The evidence for hypothesis H i is given by
$$\mathcal { O } _ { i } ( \mathbf s ) = \, \int \mathrm d \vec { \lambda } _ { i } \ \mathcal { L } _ { i } ( \mathbf s, \vec { \lambda } _ { i } ) P _ { i } ( \vec { \lambda } _ { i } ) \, \quad \text{where} \ \lambda _ { n } = \vec { \theta } \text{ and } \lambda _ { h } = \left \{ \vec { \theta }, \vec { \psi } \right \} \,.$$
The Bayes factor, B , may now be evaluated and the expectation over noise realisations calculated,
$$\overline { \mathcal { B } } = \int \mathrm d \mathbf n \, P ( \mathbf n ) \mathcal { B } \quad \text{where} \, P ( \mathbf n ) = \frac { \exp \left ( - \frac { 1 } { 2 } \mathbf n ^ { T } \Sigma _ { n } ^ { - 1 } \mathbf n \right ) } { \sqrt { ( 2 \pi ) ^ { N _ { p } T / \delta t } } \det \left ( \Sigma _ { n } \right ) } \.$$
The integral in Eq. 5 can be performed analytically and inverted to give an analytic expression for the minimum detectable amplitude as a function of frequency. This expression is too lengthy to reproduce here (see (13)) but is plotted in Fig. 1. Both of the analytic curves shown in Fig. 1 should be compared with the results of the numerical injections and recovery shown in Fig. 2.
Figure 1: The sensitivity of our canonical PTA to a monochromatic source. The lefthand panel shows the prediction of the frequentist analytic formula in Sec. 3.1 with (red) and without (black) the effect of fitting for the timing model. The right-hand panel shows the prediction of the Bayesian analytic formula in Sec. 3.2 with (red) and without (black) the effect of marginalising over the timing model parameters. In both panels it can be seen that the effect of including the timing model is a reduction in sensitivity at low frequencies. (Figure reproduced with modifications from (13).)
GLYPH<144>
GLYPH<144>
## 4. Sensitivity to a stochastic background of GWs
A stochastic background of gravitational waves is completely specified by the power spectral density, S h ( f ), which is usually taken to be a power law in frequency,
$$\left \langle \tilde { h } _ { x } ( f ) \tilde { h } _ { y } ^ { * } ( f ^ { \prime } ) \right \rangle = \frac { 1 } { 2 } \delta ( f - f ^ { \prime } ) \Gamma _ { x y } S _ { h } ( f ) \, \quad \text{where} \ S _ { h } ( f ) = \frac { A ^ { 2 } } { 1 2 \pi ^ { 2 } f _ { 0 } ^ { 3 } } \left ( \frac { f } { f _ { 0 } } \right ) ^ { - \gamma } \.$$
The constant Γ xy is called the overlap reduction function and, for an isotropic background, depends only on the relative sky positions of pulsars x and y . As can be seen from Eq. 6 the signal has contributions from all frequencies, in contrast to the monochromatic signal in Sec. 3. When calculating the sensitivity curve it no longer makes sense to calculate the minimum detectable amplitude as a function of frequency, h c ( f ). Instead the minimum detectable amplitude of the background in terms of the power-law slope may be determined, A γ ( ). Unfortunately the interpretation of this quantity is somewhat opaque, so an approach suggested by (16) can be used. The parameter γ is gridded with n values selected between some large-negative and largepositive limits. For each value of γ the value A γ ( ) may be calculated, giving the set of ordered 4
@ GLYPH<144>
D
Figure 2: The results of the recovery of numerical injections of a monochromatic signal into the mock dataset. The left-hand panel shows a surface plot of the Bayes factor against the amplitude and frequency of a monochromatic source with the contour B = B th indicated by a solid black line. The right-hand panel shows the contour B = B th in a form more comparable to Fig. 1. For more detail on the numerical calculations used to produce these plots see (13). (Figure reproduced from (13).)
GLYPH<144>
pairs ( γ , A i i ) for i ∈ { 1 2 , , . . . n } . For each pair the power-law curve for characteristic strain, h c = A γ ( ) ( f/f 0 ) ( γ -3) / 2 , may be drawn. This gives a set of power-law curves (straight lines on a log-log plot) for characteristic strain, the envelope of this set of curves is then interpreted as the power-law-integrated sensitivity curve (16), See Figs. 3 and 4.
## 4.1. Frequentist detection
Similar calculations to those performed in Sec. 3.1 may now be performed for the stochastic background. The principal difference is that in the expression for the filter function ˜ ( K f ) (Eq. 1) the product ˜ h x ( f ) ˜ h y ( f ) is replaced by the expectation value 〈 ˜ h x ( f ) ˜ h y ( f ) 〉 . The resulting expression for the SNR is (see (13))
$$\rho ^ { 2 } = \sum _ { x > y } \sum _ { y } 8 T \int \mathrm d f \, \frac { \Gamma _ { x y } ^ { 2 } S _ { h } ^ { 2 } ( f ) } { S _ { n } ^ { 2 } ( f ) } \.$$
Setting glyph[rho1] = glyph[rho1] th in Eq. 7 and inverting gives the expression A γ ( ) (13) and the power-lawintegrated sensitivity curve may be drawn following the procedure outlined above, see Fig. 3.
## 4.2. Bayesian detection
Similar calculations to those performed in Sec. 3.2 may now be performed for the stochastic background. This yields an analytic expression for the expectation value of the Bayes factor, B . Setting B = B th and inverting gives an analytic expression for A γ ( ). The analytic expressions are too lengthy to reproduce here, but can be found in (13). Using the expression A γ ( ) the power-law-integrated sensitivity curve may be drawn following the procedure outlined above, see Fig. 3.
## 5. Conclusions
The sensitivity of a PTA depends not only on the properties of the PTA itself, but also on the properties of the source being observed. This statement is also true for more traditional interferometric detectors such as aLIGO and eLISA ((17; 18)) however the effect is particularly important for PTAs because their sensitivity is determined by the sampling properties of the
12
13
14
15
16
17
-
10
9
## Bayesian Analytic Sensitivity Curve
-
10
8
10
-
7
10
-
6
10
-
Frequency f Hz
Figure 3: The sensitivity of our canonical PTA to a stochastic background. The left-hand panel shows the prediction of the frequentist analytic formula in Sec. 3.1, the curves drawn in red are for comparison with Fig. 4. The right-hand panel shows the prediction of the Bayesian analytic formula in Sec. 3.2. (Figure reproduced from (13).)
GLYPH<144>
5
GLYPH<144>
GLYPH<144>
Figure 4: The results of the recovery of numerical injections of a stochastic background signal into the mock dataset. The left-hand panel shows a surface plot of the Bayes factor against the amplitude and power-law index of a stochastic background with the contour B = B th indicated by a solid black line. The right-hand panel shows a power-law-integrated sensitivity curve formed by using points off the contour B = B th . The range of values of the power-law index was limited for reasons of numerical stability however this plot may be compared with the curves drawn in red in Fig. 3. For more detail on the numerical calculations used to produce these plots see (13). (Figure reproduced from (13).)
dataset rather than properties of the noise. For example, in the case of LIGO the lowest frequency which may be detected is determined to a large extent by the properties of the seismic noise, whereas in the case of PTAs it is determined by the total length of the observations.
The sensitivity curve of a canonical PTA was evaluated for both a monochromatic source and a stochastic background, and this was done in both the frequentist and Bayesian frameworks, with consistent results found. The analytic calculations were also verified by numerical injections into a mock dataset. In the simplified case of regularly sampled data with white, Gaussian, uncorrelated noise, simple analytic expressions for the sensitivity of a PTA have been derived. It is hoped that these expressions will be useful going forward in order to assess the prospects for detection and astronomical inference using PTA observations.
Strain hc
-
10
-
10
-
10
-
10
-
10
-
10
## Acknowledgments
The author is supported by the STFC. This talk was a partial summary of work done in collaboration with S. R. Taylor and J. R. Gair. The author would like to thank the organisers of the 10th international LISA symposium for the opportunity to give this talk.
## References
- [1] Jaffe A H and Backer D C 2003 ApJ 583 616-631 ( Preprint arXiv:astro-ph/0210148 )
- [2] Wyithe J S B and Loeb A 2003 ApJ 590 691-706 ( Preprint arXiv:astro-ph/0211556 )
- [3] Sesana A, Vecchio A and Colacino C N 2008 MNRAS 390 192-209 ( Preprint 0804.4476 )
- [4] Sesana A, Vecchio A and Volonteri M 2009 MNRAS 394 2255-2265 ( Preprint 0809.3412 )
- [5] Sesana A and Vecchio A 2010 Classical and Quantum Gravity 27 084016 ( Preprint 1001.3161 )
- [6] ¨ lmez S, Mandic V and Siemens X 2010 O Phys. Rev. D 81 104028 ( Preprint 1004.0890 )
- [7] Grishchuk L P 1976 Pis ma Zhurnal Eksperimental noi i Teoreticheskoi Fiziki 23 326-330
- [8] Grishchuk L P 2005 Physics Uspekhi 48 1235-1247 ( Preprint arXiv:gr-qc/0504018 )
- [9] Kramer M and Champion D J 2013 Classical and Quantum Gravity 30 224009
- [10] Hobbs G 2013 Classical and Quantum Gravity 30 224007 ( Preprint 1307.2629 )
- [11] McLaughlin M A 2013 Classical and Quantum Gravity 30 224008 ( Preprint 1310.0758 )
- [12] Manchester R N and IPTA 2013 Classical and Quantum Gravity 30 224010
- [13] Moore C J, Taylor S R and Gair J R 2014 ArXiv e-prints ( Preprint 1406.5199 )
- [14] Hobbs G B, Edwards R T and Manchester R N 2006 MNRAS 369 655-672 ( Preprint arXiv:astro-ph/0603381 )
- [15] Edwards R T, Hobbs G B and Manchester R N 2006 MNRAS 372 1549-1574 ( Preprint arXiv:astro-ph/0607664 )
- [16] Thrane E and Romano J D 2013 Phys. Rev. D 88 124032 ( Preprint 1310.5300 )
- [17] Harry G M and LIGO Scientific Collaboration 2010 Classical and Quantum Gravity 27 084006
- [18] Amaro-Seoane P, Aoudia S, Babak S, Bin´ etruy P, Berti E, Boh´ A, Caprini C, Colpi M, e Cornish N J, Danzmann K, Dufaux J F, Gair J, Jennrich O, Jetzer P, Klein A, Lang R N, Lobo A, Littenberg T, McWilliams S T, Nelemans G, Petiteau A, Porter E K, Schutz B F, Sesana A, Stebbins R, Sumner T, Vallisneri M, Vitale S, Volonteri M and Ward H 2012 cqg 29 124016 ( Preprint 1202.0839 )
|
10.1088/1742-6596/610/1/012018
|
[
"Christopher J. Moore"
] |
2014-08-04T16:36:48+00:00
|
2014-08-04T16:36:48+00:00
|
[
"gr-qc",
"astro-ph.IM"
] |
The sensitivity of pulsar timing arrays
|
Simple analytic expressions are derived for the sensitivity curve of a pulsar
timing array (PTA) to both a monochromatic source of gravitational waves and an
isotropic stochastic background of gravitational waves. These derivations are
performed in both frequentist and Bayesian frameworks and the results verified
with numerical injections and recovery in mock PTA datasets. The consistency of
the frequentist and Bayesian approaches is demonstrated and the results are
used to emphasise the fact that the sensitivity curve of a PTA depends not only
on the properties of the PTA itself, but also on the properties of the
gravitational wave source being observed.
|
1408.0740v2
|
## Gravitational-wave sensitivity curves
## C J Moore , R H Cole 1 1 & C P L Berry 1 2 ,
- 1 Institute of Astronomy, Madingley Road, Cambridge, CB3 0HA, United Kingdom
- 2 School of Physics and Astronomy, University of Birmingham, Edgbaston, Birmingham B15 2TT,
United Kingdom
E-mail:
[email protected]
Abstract. There are several common conventions in use by the gravitational-wave community to describe the amplitude of sources and the sensitivity of detectors. These are frequently confused. We outline the merits of and differences between the various quantities used for parameterizing noise curves and characterizing gravitational-wave amplitudes. We conclude by producing plots that consistently compare different detectors. Similar figures can be generated on-line for general use at http://rhcole.com/apps/GWplotter .
PACS numbers: 04.30.-w, 04.30.Db, 04.80.Nn, 95.55.Ym
## 1. Introduction
The next few years promise to deliver the first direct detection of gravitational waves (GWs). This will most likely be achieved by the advanced versions of the Laser Interferometer Gravitational-wave Observatory (LIGO; Harry, 2010) and Virgo (Acernese et al., 2009) detectors operating in the frequency range (10-10 ) Hz. 3 By the end of the decade, it is expected that pulsar timing arrays (PTAs; Foster and Backer, 1990) will also detect very low frequency GWs around 10 -8 Hz. Further into the future, space-based detectors, such as the evolving Laser Interferometer Space Antenna (eLISA; Seoane et al., 2013) ‡ These advances shall herald the beginning of multi-wavelength GW astronomy as a means of observing the Universe.
There already exists an extensive literature assessing the potential of all of these detectors to probe the astrophysics of various sources. There are several different methods commonly used to describe the sensitivity of a GW detector and the strength of a GW source. It is common practice to summarise this information on a sensitivity-curve plot. When producing these plots, it is desirable to have a consistent convention between detectors and sources that allows information about both to be plotted on the same graph. Ideally, the detectors
- ‡ The e in eLISA originally stood for evolved, but the preference now is for evolving.
and sources are represented in such a way that the relative detectability of the signals is immediately apparent.
In this work, we discuss the differing conventions commonly used in GW astronomy. The amplitude of a GW is a strain, a dimensionless quantity h . This gives a fractional change in length, or equivalently light travel time, across a detector. The strain is small, making it a challenge to measure: we are yet to obtain a direct detection of a GW. To calibrate our expectations for future detections it is necessary to quantify the sensitivity of our instruments and the strength of their target signals. When discussing the loudness of sources and the sensitivity of detectors there are three commonly used parametrizations based upon the strain: the characteristic strain, the power spectral density (PSD) and the spectral energy density. We aim to disambiguate these three and give a concrete comparison of different detectors. It is hoped that this will provide a useful reference for new and experienced researchers in this field alike.
We begin by expounding the various conventions and the relationships between the conventions in sections 2 and 3. A review of GW detectors (both current and proposed) is given in section 4 and a review of GW sources is given in section 5. In Appendix A, several example sensitivity curves are presented. A website where similar figures can be generated is available at http://rhcole.com/apps/GWplotter . Here, the user may select which sources and detectors to include to tailor the figure to their specific requirements.
## 2. Signal parametrization
## 2.1. Signal analysis preliminaries
Gravitational radiation has two independent polarization states denoted + and × ; a general signal can be described as a linear combination of the two polarization states, h = A h + + + A h × × . The sensitivity of a detector to these depends upon the relative orientations of the source and detector. The output of a gravitational wave detector s t ( ) contains a superposition of noise n t ( ) and (possibly) a signal h t ( ),
$$s ( t ) = n ( t ) + h ( t ) \.$$
We shall have recourse to work with the Fourier transform of the signal, using the conventions that
$$\tilde { x } ( f ) = \mathcal { F } \left \{ x ( t ) \right \} ( f ) \quad = \int _ { \substack { - \infty \\ \omega \infty } } ^ { \infty } d t \, x ( t ) \exp ( - 2 \pi i f t )$$
$$x ( t ) \, = \, \mathcal { F } ^ { - 1 } \left \{ \tilde { x } ( f ) \right \} ( t ) = \int _ { - \infty } ^ { \nu \, - \infty } \mathrm d f \, \tilde { x } ( f ) \exp ( 2 \pi i f t ) \.$$
For simplicity, it is assumed that the noise in the GW detector is stationary and Gaussian (with zero mean); under these assumptions the noise is fully characterised via the one-sided
noise PSD S n ( f ),
$$\langle \tilde { n } ( f ) \tilde { n } ^ { * } ( f ^ { \prime } ) \rangle = \frac { 1 } { 2 } \delta ( f - f ^ { \prime } ) S _ { n } ( f )$$
where angle brackets 〈 . . . 〉 denote an ensemble average over many noise realisations (Cutler and Flanagan, 1994). In reality, we have only a single realisation to work with, but the ensemble average can be replaced by a time average for stationary stochastic noise. The procedure is to measure the noise over a sufficiently long duration T and then compute the Fourier transform ˜( n f ) with a frequency resolution ∆ f = T -1 ; this is repeated many times to give an average. The noise PSD S n ( f ) has units of inverse frequency.
Since the GW signal and detector output are both real, it follows that ˜ ( h -f ) = ˜ h ∗ ( f ) and ˜( n -f ) = ˜ ( n ∗ f ); therefore, S n ( f ) = S n ( -f ). The fact that S n ( f ) is an even function means that Fourier integrals over all frequencies can instead be written as integrals over positive frequencies only, e.g., (9) and (14); it is for this reason that S n ( f ) is called the one-sided PSD. §
When integrated over all positive frequencies, the PSD gives the mean square noise amplitude. Starting by taking the time average of the square of the detector noise:
$$\text{sim} \ \overline { | n ( t ) | ^ { 2 } } & = \lim _ { T \to \infty } \frac { 1 } { 2 T } \int _ { - T } ^ { T } d t \ n ( t ) n ^ { * } ( t ) \\ & = \lim _ { T \to \infty } \frac { 1 } { 2 T } \int _ { - T } ^ { T } d t \ \int _ { - \infty } ^ { \infty } d f \ \int _ { - \infty } ^ { \infty } d f ^ { \prime } \, \tilde { n } ( f ) \tilde { n } ^ { * } ( f ^ { \prime } ) \exp { ( 2 \pi i f t ) } \exp { ( - 2 \pi i f ^ { \prime } t ) } \\ & = \lim _ { T \to \infty } \frac { 1 } { 2 T } \int _ { - T } ^ { T } d t \ \int _ { - \infty } ^ { \infty } d f \ \int _ { - \infty } ^ { \infty } d f ^ { \prime } \, \mathcal { F } \{ n ( \tau ) \} \left ( f \right ) [ \mathcal { F } \{ n ( \tau ) \} \left ( f ^ { \prime } ) ] ^ { * } \\ & \quad \times \exp { ( 2 \pi i f t ) } \exp { ( - 2 \pi i f ^ { \prime } t ) } \, \\ \quad \text{who} \quad \text{has} \ \text{when} \ \text{without} \ \text{in} \ \text{in} \ \text{the} \ \text{definition} \ of} \ \text{the} \ \text{Env} \ \text{from} \ \text{and} \ \text{it} }$$
where we have substituted in using the definitions of the Fourier transform and its inverse. A property of Fourier transforms is that a time-domain translation by amount t is equivalent to a frequency-domain phase change 2 πft ; if F { n τ ( ) } ( f ) = ˜( n f ), then F { n τ ( -t ) } ( f ) = ˜( n f ) exp(2 π ft i ). Therefore, the exponential factors in (6) may be absorbed as
$$\overline { \left | n ( t ) \right | ^ { 2 } } = \lim _ { T \rightarrow \infty } \frac { 1 } { 2 T } \int _ { - T } ^ { T } \mathrm d t \int _ { - \infty } ^ { \infty } \mathrm d f \, \int _ { - \infty } ^ { \infty } \mathrm d f ^ { \prime } \, \mathcal { F } \left \{ n ( \tau - t ) \right \} \left ( f \right ) \left [ \mathcal { F } \left \{ n ( \tau - t ) \right \} \left ( f ^ { \prime } \right ) \right ] ^ { * } \,.$$
Since the noise is a randomly varying signal, we can use the ergodic principle to equate a time average, denoted by ( . . . ), with an ensemble average, denoted by 〈 . . . 〉 . The noise is stationary, consequently, its expectation value is unchanged by the time-translation performed above.
§ An alternative convention is to use the two-sided PSD S (2) n ( f ) = S n ( f ) / 2.
Therefore, using (4), the mean square noise amplitude is given by
$$\overline { | n ( t ) | ^ { 2 } } & = \int _ { - \infty } ^ { \infty } \mathrm d f \, \int _ { - \infty } ^ { \infty } \mathrm d f ^ { \prime } \, \langle \tilde { n } ( f ) \tilde { n } ^ { * } ( f ^ { \prime } ) \rangle \\ & = \int _ { - \infty } ^ { \infty } \mathrm d f \, \int _ { - \infty } ^ { \infty } \mathrm d f ^ { \prime } \, \frac { 1 } { 2 } S _ { n } ( f ) \delta ( f - f ^ { \prime } ) \\ & = \int _ { 0 } ^ { \infty } \mathrm d f \, S _ { n } ( f ) \,. \\ \text{Given a detector output, the challenge is to extract the signal. There is a well known}$$
Given a detector output, the challenge is to extract the signal. There is a well known solution to this problem that involves constructing a Wiener optimal filter (Wiener, 1949). Let K t ( ) be a real filter function with Fourier transform ˜ ( K f ). Convolving this with the detector output gives a contribution from the signal and a contribution from the noise,
$$\left ( s * K \right ) \left ( \tau \right ) = \int _ { - \infty } ^ { \infty } \text{d} t \, \left [ h ( t ) + n ( t ) \right ] K ( t - \tau ) \approx \mathcal { S } + \mathcal { N } \,.$$
The signal contribution S is defined as the expectation of the convolution in (10) when a signal is present, maximised by varying the offset to achieve the best overlap with the data. Since the expectation of pure noise is zero it follows that
$$\mathcal { S } = \int _ { - \infty } ^ { \infty } \mathrm d t \ h ( t ) K ( t ) = \int _ { - \infty } ^ { \infty } \mathrm d t \ h ( t ) K ^ { * } ( t ) = \int _ { - \infty } ^ { \infty } \mathrm d f \, \tilde { h } ( f ) \tilde { K } ^ { * } ( f ) \.$$
The squared contribution from noise N 2 is defined as the mean square of the convolution in (10) when no signal is present,
$$( 1 0 ) & \text{ when no signal is present}, \\ \mathcal { N } ^ { 2 } & = \int _ { - \infty } ^ { \infty } d t \, \int _ { - \infty } ^ { \infty } d t ^ { \prime } \, K ( t ) K ( t ^ { \prime } ) \, \langle n ( t ) n ( t ^ { \prime } ) \rangle \\ & = \int _ { - \infty } ^ { \infty } d t \, \int _ { - \infty } ^ { \infty } d t ^ { \prime } \, K ( t ) K ^ { * } ( t ^ { \prime } ) \int _ { - \infty } ^ { \infty } d f \, \int _ { - \infty } ^ { \infty } d f ^ { \prime } \, \langle \tilde { n } ( f ) \tilde { n } ^ { * } ( f ^ { \prime } ) \exp \left [ 2 \pi i ( f t - f ^ { \prime } t ^ { \prime } ) \right ] \\ & = \int _ { - \infty } ^ { \infty } d f \, \frac { 1 } { 2 } S _ { n } ( f ) \tilde { K } ( f ) \tilde { K } ^ { * } ( f ) \, \\ \text{ using the definition of } S _ { n } ( f ) \, \text{ from } ( 4 ). \, \text{ Hence the signal-to-noise ratio (SNR)} \, \wp \text{ is given by}$$
using the definition of S n ( f ) from (4). Hence the signal-to-noise ratio (SNR) glyph[rho1] is given by
$$\varrho ^ { 2 } = \frac { \mathcal { S } ^ { 2 } } { \mathcal { N } ^ { 2 } } = \frac { \left ( \frac { 1 } { 2 } S _ { n } ( f ) \tilde { K } ( f ) \Big | \tilde { h } ( f ) \right ) ^ { 2 } } { \left ( \frac { 1 } { 2 } S _ { n } ( f ) \tilde { K } ( f ) \Big | \frac { 1 } { 2 } S _ { n } ( f ) \tilde { K } ( f ) \right ) },$$
where we have introduced the inner product between signal ˜ and A ˜ as (Finn, 1992) B
$$\left ( \tilde { A } ( f ) \Big | \tilde { B } ( f ) \right ) = 4 \Re \left \{ \int _ { 0 } ^ { \infty } \mathrm d f \, \frac { \tilde { A } ^ { * } ( f ) \tilde { B } ( f ) } { S _ { n } ( f ) } \right \} \,.$$
The optimum filter is that function ˜ ( K f ) which maximises the SNR in (13). From the Cauchy-Schwarz inequality, it follows that the optimum filter is
$$\tilde { K } ( f ) = \frac { \tilde { h } ( f ) } { S _ { n } ( f ) } \.$$
This is the Wiener filter, which may be multiplied by an arbitrary constant since this does not change the SNR. Using this form for ˜ ( K f ), the squared SNR is
$$\varrho ^ { 2 } = \int _ { 0 } ^ { \infty } \mathrm d f \frac { 4 | \tilde { h } ( f ) | ^ { 2 } } { S _ { n } ( f ) } = \left ( \tilde { h } ( f ) | \tilde { h } ( f ) \right ).$$
In order to construct the Wiener filter, it is necessary to know a priori the form of the signal, ˜ ( h f ), for this reason the Wiener filter is sometimes called the matched filter.
Whilst the magnitude of the Fourier transform of the signal | ˜ ( h f ) | provides a simple quantification of the GW amplitude as a function of frequency, it has one main deficiency. For an inspiralling source, the instantaneous amplitude can be orders of magnitude below the noise level in a detector; however, as the signal continues over many orbits, the SNR can be integrated up to a detectable level. It is useful to have a quantification of the GW amplitude that accounts for this effect; we shall now describe how this can be achieved.
## 2.2. Characteristic strain
The characteristic strain h c is designed to include the effect of integrating an inspiralling signal. Its counterpart for describing noise is the noise amplitude h n . These are defined as
$$[ h _ { c } ( f ) ] ^ { 2 } \, = \, 4 f ^ { 2 } \left | \tilde { h } ( f ) \right | ^ { 2 } \,,$$
$$[ h _ { n } ( f ) ] ^ { 2 } \, = f S _ { n } ( f )$$
such that the SNR in (16) may be written
$$\varrho ^ { 2 } = \int _ { - \infty } ^ { \infty } \text{d} \left ( \log f \right ) \, \left [ \frac { h _ { c } ( f ) } { h _ { n } ( f ) } \right ] ^ { 2 } \,.$$
The strain amplitudes h c ( f ) and h n ( f ) are dimensionless. Using this convention, when plotting on a log-log scale, the area between the source and detector curves is related to the SNR via (19). This convention allows the reader to integrate by eye to assess the detectability of a given source (see figure A1).
An additional advantage of this convention is that the values on the strain axis for the detector curve h n ( f ) have a simple physical interpretation: they correspond to the root-mean-square noise in a bandwidth f . One downside to plotting characteristic strain is that the values on the strain axis h c ( f ) do not directly relate to the amplitude of the waves from the source. Another disadvantage is that applying (17) to a monochromatic source
gives a formally undefined answer. The correct identification of characteristic strain for a monochromatic source is the amplitude of the wave times the square root of the number of periods observed (see section 3.1).
## 2.3. Power spectral density
A second commonly used quantity for sensitivity curves is the square root of the PSD or the amplitude spectral density (see figure A2). When discussing a detector, rearranging (18) gives
$$\sqrt { S _ { n } ( f ) } = h _ { n } ( f ) f ^ { - 1 / 2 } \ ;$$
by analogy, we can define an equivalent for source amplitudes
$$\sqrt { S _ { h } ( f ) } = h _ { c } ( f ) f ^ { - 1 / 2 } = 2 f ^ { 1 / 2 } \left | \tilde { h } ( f ) \right |$$
where we have used (17). Both √ S n ( f ) and √ S h ( f ) have units of Hz -1 2 / . The root PSD is the most frequently plotted quantity in the literature.
The PSD, as defined by (4), has the nice property, demonstrated in (9), that integrated over all positive frequencies it gives the mean square amplitude of the noise in the detector. However, in one important regard it is less appealing than characteristic strain: the height of the source above the detector curve is no longer directly related to the SNR.
## 2.4. Energy density
A third way of describing the amplitude of a GW is through the energy carried by the radiation. This has the advantage of having a clear physical significance. The energy density is most commonly used in sensitivity curves showing stochastic backgrounds of GWs (see section 3.2).
The energy in GWs is described by the Isaacson stress-energy tensor (Misner et al., 1973, section 35.15)
$$T _ { \mu \nu } = \frac { c ^ { 4 } } { 3 2 \pi G } \left \langle \partial _ { \mu } \bar { h } _ { \alpha \beta } \partial _ { \nu } \bar { h } ^ { \alpha \beta } \right \rangle \,,$$
where the angle brackets denote averaging over several wavelengths or periods, and ¯ h αβ is the transverse-traceless metric perturbation. The energy density ρc 2 is given by the T 00 component of this tensor. Consequently (cf. Berry and Gair, 2013),
$$\rho c ^ { 2 } & = \frac { c ^ { 2 } } { 1 6 \pi G } \int _ { - \infty } ^ { \infty } \text{d} f \ ( 2 \pi f ) ^ { 2 } \tilde { h } ( f ) \tilde { h } ^ { * } ( f ) \\ & = \int _ { 0 } ^ { \infty } \text{d} f \ \frac { \pi c ^ { 2 } } { 4 G } f ^ { 2 } S _ { h } ( f ) \,,$$
$$= \, \int _ { 0 } ^ { \infty } \mathrm d f \, \frac { \vec { \pi c } ^ { 2 } } { 4 G } f ^ { 2 } S _ { h } ( f )$$
where the definition (21) has been used. The integrand in (24) is defined as the spectral energy density, the energy per unit volume of space, per unit frequency (Hellings and Downs, 1983)
$$S _ { \text{E} } ( f ) = \frac { \pi c ^ { 2 } } { 4 G } f ^ { 2 } S _ { h } ( f ) \, ;$$
a corresponding expression for the noise can be formulated by replacing S h ( f ) with S n ( f ).
Cosmological studies often work in terms of the dimensionless quantity Ω GW , the energy density per logarithmic frequency interval, normalised to the critical density of the Universe ρ c ,
$$\Omega _ { \text{GW} } ( f ) = \frac { f S _ { \text{E} } ( f ) } { \rho _ { \text{e} } c ^ { 2 } } \.$$
The critical density is
$$\rho _ { c } = \frac { 3 H _ { 0 } ^ { 2 } } { 8 \pi G }$$
where H 0 is the Hubble constant, commonly parametrized as
$$H _ { 0 } = h _ { 1 0 0 } \times 1 0 0 \, \text{km} \, s ^ { - 1 } \, \text{Mpc} ^ { - 1 }.$$
The reduced Hubble parameter h 100 has nothing to do with strain. The most common quantity related to energy density to be plotted on sensitivity curves is Ω GW 100 h 2 (figure A3) as this removes sensitivity to the (historically uncertain) measured value of the Hubble constant.
This quantity has one aesthetic advantage over the others: it automatically accounts for there being less energy in low frequency waves of the same amplitude. However, unlike characteristic strain, the area between the source and detector curves is no longer simply related to the SNR.
## 2.5. Relating the descriptions
The dimensionless energy density in GWs Ω GW , spectral energy density S E , one-sided PSD S h , characteristic strain h c and frequency-domain strain ˜ ( h f ) are related via
$$H _ { 0 } ^ { 2 } \Omega _ { \text{GW} } ( f ) = \frac { 8 \pi G } { 3 c ^ { 2 } } f S _ { \text{E} } ( f ) = \frac { 2 \pi ^ { 2 } } { 3 } f ^ { 3 } S _ { h } ( f ) = \frac { 2 \pi ^ { 2 } } { 3 } f ^ { 2 } \left [ h _ { c } ( f ) \right ] ^ { 2 } = \frac { 8 \pi ^ { 2 } } { 3 } f ^ { 4 } \left | \tilde { h } ( f ) \right | ^ { 2 } \,,$$
using (17), (21), (25), (26) and (27). Corresponding expressions for the noise are obtained by substituting S n ( f ) for S h ( f ), h n ( f ) for h c ( f ) and ˜( n f ) for ˜ ( h f ).
## 3. Types of source
GW signals can be broadly split into three categories: those from well-modelled sources, for which we have a description of the expected waveform; stochastic backgrounds, for which we can describe the statistical behaviour; and unmodelled (or poorly-modelled) transient sources. The classic example of a well-modelled source is the inspiral of two compact objects, this is discussed in section 3.1. Stochastic backgrounds can either be formed from many overlapping sources, which could be modelled individually, or from some intrinsically random process, these are discussed in section 3.2. An example of an unmodelled (or poorly-modelled) transient source is a supernova; searches for signals of this type are often called burst searches and are discussed in section 3.3.
## 3.1. Inspirals
Inspiralling binaries may the most important GW source. They spend a variable amount of time in each frequency band. If φ is the orbital phase, then the number of cycles generated at frequency f can be estimated as
$$N _ { \text{cycles} } = \frac { f } { 2 \pi } \frac { \text{d} \phi } { \text{d} f } = \frac { f ^ { 2 } } { \dot { f } }$$
where an overdot represents the time derivative and ˙ φ = 2 πf . The squared SNR scales with N cycles , so it would be expected that h c ( f ) ≈ √ N cycles | ˜ ( h f ) . |
The form for h c can be derived from the Fourier transform in the stationary-phase approximation. Consider a source signal with approximately constant (root-mean-square) amplitude h 0 and central frequency f ′ . In this case,
$$\begin{array} { c } \dots _ { x } \dots \dots _ { v } \dots \dots \dots _ { x } \dots \dots _ { x } \dots \dots _ { x } \dots \\ h ( t ) & = \sqrt { 2 } h _ { 0 } \cos \left [ \phi ( t ) \right ] \,, & \dots _ { x } \dots _ { x } \dots _ { x } \dots _ { x } \dots _ { x } \dots \\ \end{array}$$
$$\tilde { h } ( f ) = \frac { \partial ^ { \prime } } { \sqrt { 2 } } \int _ { - \infty } ^ { \times } \partial ^ { \prime \prime \prime } _ { - \infty } \partial t \, \exp \left \{ 2 \pi i \left [ \frac { \phi ( t ) } { 2 \pi t } - f \right ] t \right \} + \exp \left \{ - 2 \pi i \left [ \frac { \phi ( t ) } { 2 \pi t } + f \right ] t \right \} \.$$
Without loss of generality, we can assume an initial phase of zero, such that φ (0) = 0. The largest contribution to the integral comes from where the argument of the exponentials is approximately zero. For the first term, this occurs when f = f ′ , then the term in brackets is
$$\left [ \frac { \phi ( t ) } { 2 \pi t } - f \right ] _ { f = f ^ { \prime } } = f ^ { \prime } + \dot { f ^ { \prime } } t + \mathcal { O } \left ( t ^ { 2 } \right ) - f ^ { \prime } = \dot { f ^ { \prime } } t + \mathcal { O } \left ( t ^ { 2 } \right ).$$
The higher-order terms cause the expoential to oscillate rapidly, such that these terms integrate to zero and may be neglected. Performing a similar expansion about f = -f ′ for
the second term in (32), and then evaluating the Gaussian integrals gives
$$\ t n e { \text{second term in (32), and then evaluation the cause savings} } \, & \tilde { h } ( f ) \simeq \frac { h _ { 0 } } { \sqrt { 2 } } \int _ { - \infty } ^ { \infty } d t \, \exp \left ( 2 \pi i \dot { f } ^ { \prime } t ^ { 2 } \right ) + \exp \left ( - 2 \pi i \dot { f } ^ { \prime } t ^ { 2 } \right ) \\ & \simeq \frac { h _ { 0 } } { \sqrt { 2 \dot { f } ^ { \prime } } }. \\ \tau = \tau \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \,.$$
From (17) and (34), the characteristic strain for inspiralling sources is given by (Finn and Thorne, 2000)
$$h _ { \text{c} } ( f ) = \sqrt { \frac { 2 f ^ { 2 } } { \dot { f } } } h _ { 0 } \.$$
Equation (17) should be considered as the definition of characteristic strain and (35) a consequence of it for inspirals. Equation (35) is the relation between h c ( f ) and the instantaneous root-mean-square amplitude h 0 for an inspiralling source; for other types of source a new relation satisfying (19) has to be found.
## 3.2. Stochastic backgrounds
Another important source of GWs is that of stochastic backgrounds, which can be produced from a large population of unresolvable sources. These can be at cosmological distances, where it is necessary to distinguish the frequency in the source rest frame f r from the measured frequency f ; the two are related through the redshift z via f r = (1 + z f ) . The comoving number density of sources ν producing the background is also a function of redshift; if the sources producing the background are all in the local Universe, then simply set ν z ( ) = ν δ z 0 ( ) and replace d L ( z ) with d in all that follows, where d L ( z ) and d are respectively the luminosity and comoving distances to the source, d L ( z ) = (1 + z d ) .
We shall assume that the individual sources are binaries, in which case the number density of sources is also a function of the component masses. It is convenient to work in terms of the chirp mass, defined as M = µ 3 5 / M 2 5 / , where µ is the reduced mass and M is the total mass of the binary. The comoving number density of sources shall be represented by ν z, ( M ).
Equation (29) gives an expression for the energy density in GWs per logarithmic frequency interval,
$$f S _ { E } ( f ) = \frac { \pi c ^ { 2 } } { 4 G } f ^ { 2 } \left [ h _ { c } ( f ) \right ] ^ { 2 } \,.$$
The total energy emitted in the logarithmic frequency interval d (log f r ) by a single binary in the population is [d E GW / d(log f r )] d(log f r ); the energy density may be written as
$$f S _ { E } ( f ) = \int _ { 0 } ^ { \infty } d z \, \frac { d \nu } { d z } \frac { 1 } { ( 1 + z ) } \frac { 1 } { \left [ d _ { L } ( z ) \right ] ^ { 2 } } \frac { d E _ { G W } } { d \left ( \log f _ { r } \right ) }$$
where the factor of (1 + z ) -1 accounts for the redshifting of the energy.
For simplicity, consider the background to comprise of binaries in circular orbits, with frequencies f GW = f / r 2, which are far from their last stable orbit. The energy radiated may then be calculated using the quadrupole approximation (Peters and Mathews, 1963). The energy in GWs from a single binary per logarithmic frequency interval is
$$\frac { \mathrm d E _ { \mathrm G W } } { \mathrm d \left ( \log f _ { \mathrm r } \right ) } = \frac { G ^ { 2 / 3 } \pi ^ { 2 / 3 } } { 3 } \mathcal { M } ^ { 5 / 3 } f _ { \mathrm r } ^ { 2 / 3 }$$
between minimum and maximum frequencies set by the initial and final radii of the binary orbit respectively. Here, we assume that the maximum and minimum frequencies are outside of the range of our detector and hence can be neglected. Using (36), (37) and (38), an expression for characteristic strain can now be found (Sesana et al., 2008)
$$[ h _ { c } ( f ) ] ^ { 2 } = \frac { 4 G ^ { 5 / 3 } } { 3 \pi ^ { 1 / 3 } c ^ { 2 } } f ^ { - 4 / 3 } \int _ { 0 } ^ { \infty } \mathrm d z \, \int _ { 0 } ^ { \infty } \mathrm d \mathcal { M } \, \frac { \mathrm d ^ { 2 } \nu } { \mathrm d z \, \mathrm d \mathcal { M } } \frac { 1 } { [ \mathrm d _ { \mathrm L } ( z ) ] ^ { 2 } } \left ( \frac { \mathcal { M } ^ { 5 } } { 1 + z } \right ) ^ { 1 / 3 } \,.$$
From (39) it can be seen that the characteristic strain due to a stochastic background of binaries is a power law in frequency with spectral index α = -2 3. / The amplitude of the background depends on the population statistics of the binaries under consideration via ν z, ( M ). The power law is often parametrised as
$$h _ { \text{c} } ( f ) = A \left ( \frac { f } { f _ { 0 } } \right ) ^ { \alpha } \,,$$
and constraints are then placed on A . In practice, this power law also has upper and lower frequency cut-offs related to the population of source objects.
A stochastic background from other sources, such as cosmic strings or relic GWs from the early Universe, can also be written in the same form as (40), but with different spectral indices: α = -7 / 6 for cosmic strings or α in the range -1 to -0 8 for relic GWs (Jenet et al., . 2006).
An alternative method for graphically representing the sensitivity of a GW detector to stochastic backgrounds, called the power-law-integrated sensitivity curve , was suggested by Thrane and Romano (2013). This method accounts for the there being power across all frequencies in the sensitivity band by integrating the noise-weighted signal over frequency. As our aim here is to present stochastic backgrounds alongside other types of sources for comparison, we will not use this approach.
## 3.3. Burst sources
Some sources of GWs can produce signals with large amplitudes, greater than the detector noise. The typical duration of such a signal is short, of the order of a few wave periods, and so
there is not time to accumulate SNR in each frequency band as for inspirals. As a consequence, waveform models are not required for detection; we simply rely on identifying the excess power produced by these burst sources. Typically, we may be looking for signals from core-collapse supernovae (Ott, 2009), the late stages of merging compact binaries (Abadie et al., 2012 b ), cosmic strings (Aasi et al., 2014 a ), or more generally, signals from any unexpected or poorly modelled sources.
Burst searches are often carried out using time-frequency techniques. The data stream from a detector is temporally split into segments, the length of which can be tuned to give greater sensitivity to particular sources. Each segment is then transformed into the frequency domain, whitened and normalised to the noise spectrum of the detector to produce a time-frequency plot. Potential GW signals are identified by searching for clusters of pixels that contain an excess of power (e.g., Abadie et al., 2012 a ).
The presence of excess power across a number of pixels eliminates modelled noise sources, but such a cluster may also be caused by atypical noise within a detector. We can improve our confidence of a GW signal by making use of information obtained from other GW detectors. Signals across a network of detectors should have compatible arrival times (given the sky direction) as well as consistent amplitudes, frequencies and shapes of the waveform. Different pipelines are currently in use that analyse the signal consistency in different ways: both coincidence searches (Chatterji et al., 2004) and fully coherent methods (Klimenko et al., 2008) are used.
An important aspect of burst search algorithms is to accurately estimate the noise properties within each time segment. To this extent, null data streams can be constructed that are insensitive to real GW signals. In order to estimate the false alarm rate, the data from different detectors can be shifted in time to remove any genuine coincident GW signals (Cannon, 2008). These time-shifts are then analysed to simulate the potential occurence of coincident noise events. The algorithms are tuned using time-shifted data to ensure there is no bias in the final search.
As discussed in 3.1, the expected relation between h c ( f ) and a typical waveform ˜ ( h f ) is
$$h _ { \text{c} } ( f ) = \sqrt { N _ { \text{cycles} } } \left | \tilde { h } ( f ) \right | \,,$$
where N cycles is the number of cycles of radiation generated by the source, which is of order unity for bursts.
An alternative characterisation of the signal amplitude commonly used for burst sources is the root-sum-square of the waveform polarisations:
$$h _ { \text{rss} } ^ { 2 } = \, \int \text{d} t \, | h _ { + } ( t ) | ^ { 2 } + | h _ { \times } ( t ) | ^ { 2 } \.$$
For a linearly polarised GW, with ˜ ( h f ) constant across the bandwidth ∆ , f this is
approximately related to the characteristic strain via
$$h _ { \text{rss} } \simeq \left | \tilde { h } ( f ) \right | \sqrt { \Delta f }$$
where we have neglected the detector response functions (see section 4.1), which are of order unity. In this work, we favour a constant h c ( f ) rather than h rss for consistency with the other types of source where the bandwidth is detector specific.
## 4. Detectors
In this section we introduce the detector noise curves used in Appendix A. We begin with a description of the basic operation of detectors. We then discuss ground-based detectors, space-based detectors and PTAs in turn. The latter function somewhat differently than conventional interferometers, so we include a brief introduction to PTA analysis. References for the noise curves used for individual detectors can be found in the relevant subsections and further information about the detectors can be discovered in these. Detectors are frequently upgraded and redesigned, hence, while these curves are believed to be correct at the time of writing, it is best to check for updates from the appropriate science teams before relying on the details given here, although we hope that they shall remain accurate enough for illustrative purposes.
## 4.1. Operating principle of an interferometric detector
All of the man-made detectors discussed in this section utilise the principle of interferometry. Such detectors work by taking a beam of monochromatic light and splitting it into two beams travelling at some angle to each other. Each beam is passed in to an optical cavity where it undergoes a number of round trips before being recombined to form an interference pattern. The ends of the cavity are, in the ideal case, freely floating test masses which move relative to each other in response to a passing GW, this effect is measured by observing the changing interference pattern.
The response of a detector to an incident plane-fronted GW depends upon the relative orientations of the detector and the incoming wave. Let us choose the origin of our coordinate system to be the beam-splitter of the interferometer, and l i 1 and l i 2 to be unit 3-vectors pointing along the two arms. In the absence of noise the output of the detector is the difference in strain between the two arms (Thorne, 1987)
$$h ( t ) = \frac { 1 } { 2 } h _ { i j } \left ( l _ { 1 } ^ { i } l _ { 1 } ^ { j } - l _ { 2 } ^ { i } l _ { 2 } ^ { j } \right ) \,,$$
where h ij are the spatial components of the GW metric perturbation. Let ˆ r i be the unit 3-vector pointing towards the source of the GWs, with spherical polar angles ( θ, φ ) relative
to some axes fixed to the detector, and let p i and q i be unit vectors orthogonal to ˆ . r i We can now define the basis tensors
$$H _ { i j } ^ { + } = p _ { i } p _ { j } - q _ { i } q _ { j }$$
$$H _ { i j } ^ { \times } = p _ { i } q _ { j } + q _ { i } p _ { j } \.$$
There remains a freedom in the coordinates described, a rotation of p i and q i through an angle ψ about ˆ r i known as the polarization angle. For a single frequency component, the strain induced by a GW may be written as
$$h _ { i j } = A _ { + } H _ { i j } ^ { + } \cos \left ( 2 \pi f t \right ) + A _ { \times } H _ { i j } ^ { \times } \cos \left ( 2 \pi f t + \Delta \phi \right )$$
where A + and A × are the amplitudes of the two polarisation states. Combining (44) and (47) allows the detector output to be written as
$$h ( t ) = F ^ { + } ( \theta, \phi, \psi ) A _ { + } \cos { ( 2 \pi f t ) } + F ^ { \times } ( \theta, \phi, \psi ) A _ { \times } \cos { ( 2 \pi f t } + \Delta \phi )$$
where the response functions inherit their angular dependence from the choice of coordinates
$$F ^ { + } ( \theta, \phi, \psi ) = \frac { 1 } { 2 } H _ { i j } ^ { + } \left ( l _ { 1 } ^ { i } l _ { 1 } ^ { j } - l _ { 2 } ^ { i } l _ { 2 } ^ { j } \right ) \,,$$
$$& F ^ { \times } ( \theta, \phi, \psi ) = \frac { \overline { 1 } } { 2 } H _ { i j } ^ { \times } \left ( l _ { 1 } ^ { i } l _ { 1 } ^ { j } - l _ { 2 } ^ { i } l _ { 2 } ^ { j } \right ) \,. \\ & \quad \quad \quad$$
The response function of a two-arm interferometric detector is quadrupolar, an example is plotted in figure 1. Throughout this paper, detector sensitivity refers to the polarisation and sky averaged sensitivity F , where
$$F ^ { 2 } = \int _ { 0 } ^ { 2 \pi } \frac { \mathrm d \psi } { 2 \pi } \, \int _ { 0 } ^ { 2 \pi } \frac { \mathrm d \phi } { 2 \pi } \, \int _ { 0 } ^ { \pi } \frac { \sin \theta \, \mathrm d \theta } { 2 } \, \$$
For a single 90 -interferometer, such as LIGO, the sky and polarisation averaged response is ◦ F = √ 1 / 5 ≈ 0 447. .
A detector may consist of several interferometers. Let F a be the averaged response of the a -th interferometer, the average response of a network of k detectors is obtained by adding in quadrature,
$$F _ { \text{Total} } ^ { 2 } = \frac { 1 } { k } \sum _ { a = 1 } ^ { k } F _ { a } ^ { 2 } \.$$
The averaging in (51) assumes a uniform distribution of polarisation angles ψ . This is the case for a stochastic background; however, for a non-inspiralling circular binary, the polarisation is a function of the two spherical polar angles ( ι, ξ ) specifying the orientation of the binary's orbital angular momentum. Here, ι is the inclination angle, the polar angle between the orbital angular momentum and the line joining the source to the detector ( -ˆ) r
Figure 1: The angular response function of an interferometric detector shown both as a surface plot and in an Aitoff-Hammer projection. The quantity that is plotted is is the polarisation average, [ (1 / 2 π ) ∫ d ψ ( F + 2 ) +( F × ) 2 ] 1 2 / . The response is a function of two sky angles, θ and φ , and varies between 0 and 1. The two detector arms lie in the x y -plane either side of one of the zeros in the response.
and ξ is the azimuthal angle around the same line. In this case, to characterise the detector sensitivity we average over all four angles ( θ, φ, ι, ξ ). If the binary is inspiralling, then the polarisation depends on still more parameters which need to be averaged over. These more complicated averages all have the property that they depend on both the detector and the source, hence they are unhelpful for our present purpose separating the source amplitude from the detector sensitivity. Additionally, the different averages do not work out to be so different from each other: Finn and Chernoff (1993) calculated the sensitivity for a detector with the LIGO geometry averaged over the four angles ( θ, φ, ι, ξ ) as √ 4 / 25 = 0 4 times . peak sensitivity, which should be compared with the value √ 1 / 5 ≈ 0 447 above. The effect . of replacing the true sensitivity with the sky-averaged sensitivity for the LISA detector was considered in detail by Vallisneri and Galley (2012); they also found there is a small difference when considering an entire population of sources. For the remainder of this paper the three-angle average defined in (51) is used.
## 4.2. Ground-based detectors
Ground-based detectors are the most numerous. A collection of interferometric detectors are listed in table 1, these are sensitive to GWs in the frequency range O (10-10 ) Hz. They all 3 simulate free-floating test masses by suspending a mass from a pendulum system with natural frequency far removed from than that of the GW. Their sensitivity curves include narrow lines that arise from noise sources in the instrument, including resonances in the suspension system and electrical noise at multiples of 60 Hz: these have been removed in the Appendix A
Table 1: Summary of ground-based laser interferometers.
| Detector | Country | Arm length | Approximate date | Generation |
|---------------------------------------------------------------|-------------------------------------------|--------------------------------------------|---------------------------------------------------------------------------------------|----------------------------------------------------|
| GEO600 a TAMA300 b iLIGO c Virgo d aLIGO e AdV f KAGRA g ET h | Germany Japan USA Italy USA Italy Japan - | 600 m 300 m 4 km 3 km 4 km 3 km 3 km 10 km | 2001-present 1995-present 2004-2010 2007-2010 est. 2015 est. 2016 est. 2018 est. 2025 | First First First First Second Second Second Third |
a Grote (2010), b Ando (2002), c Abbott et al. (2009),
d Accadia et al. (2012), e Harry (2010), f Acernese et al. (2009), g Somiya (2012), h Hild et al. (2011).
figures for clarity. The detectors fall broadly into three categories: first-generation detectors, which have already operated; second-generation detectors currently under construction; and third-generation detectors at the planning stage.
The most notable ground-based detectors are LIGO and Virgo, which work in collaboration, supported by GEO600. LIGO, Virgo and GEO600 have completed science runs as first-generation detectors (e.g., Abadie et al., 2010). As LIGO and Virgo are currently being upgraded their initial configurations are now referred to as Initial LIGO (iLIGO) and Initial Virgo (iVirgo) respectively. The upgraded, second-generation versions are referred to as Advanced LIGO (aLIGO) and Advanced Virgo (AdV) respectively. LIGO has two observatories: one at Hanford, Washington, which has two detectors; and another at Livingston, Louisiana. There is an agreement to move one of the upgraded Hanford detector systems to a location in India (Iyer et al., 2011; Unnikrishnan, 2013). The GEO600 detector is not subject to a major upgrade plan, but since summer 2009 it has been enhanced by a series of smaller improvements, notably improving high-frequency sensitivity (GEO-HF; Willke et al., 2006). The advanced detectors should start operation in the next couple of years, with LIGO-India following further in the future.
TAMA300 is a Japanese first-generation detector. Its successor, currently under construction, is the Kamioka Gravitational Wave Detector (KAGRA), formerly the Largescale Cryogenic Gravitational wave Telescope (LCGT), which is located underground in the Kamioka mine. It employs more sophisticated noise-reduction techniques than LIGO or Virgo, such as cryogenic cooling.
The Einstein Telescope (ET) is an ambitious proposal to construct an underground third-generation detector. Its location would provide shielding from seismic noise, allowing it to observe frequencies of (10-10 ) Hz. 4
We use an interpolation to the data published on https://wwwcascina.virgo.infn.it/
advirgo/ (2013) for the AdV sensitivity curve, an interpolation to the data for version D of the KAGRA detector published on http://gwcenter.icrr.u-tokyo.ac.jp/en/researcher/ parameter (2013) and analytic fits to the sensitivity curves from Sathyaprakash and Schutz (2009) for the remaining detectors.
## 4.3. Space-based detectors
Space-based detectors work on similar principles to ground based detectors, but with the test masses residing inside of independent, widely separated satellites. Space-based detectors are sensitive to lower frequency GWs than their ground-based counterparts; this is partly because space-based detectors can have much longer arms, and partly because they are unaffected by seismic noise which limits the low frequency performance of ground-based detectors.
The canonical design for a space-based detector is the Laser Interferometer Space Antenna (LISA), which is sensitive to millihertz GWs. LISA would consist of three satellites flying in a triangular constellation with arms of length 5 × 10 9 m in a 1 AU orbit around the Sun, trailing the Earth by 20 . ◦ The laser arms in a LISA-like detector are not a cavity, the light only travels once along each arm. eLISA is a rescoped version of LISA designed to probe the same frequency range, while proposals such as the Advanced Laser Interferometer Antenna (ALIA), Big Bang Observer (BBO) and Deci-hertz Interferometer GW Observatory (DECIGO) are designed to probe decihertz GWs.
4.3.1. LISA and eLISA The instrumental noise curves for LISA are approximated by the analytic fit given by Sathyaprakash and Schutz (2009), which we use for the plots in Appendix A. When observing individual sources with LISA there is an additional contribution to the noise from a background of unresolvable binaries. This is not included here as we consider the background as a source of GWs (see section 5.2.2). eLISA is a rescoped version of the classic LISA mission, the main differences are shorter arms (10 9 m instead of 5 × 10 9 m), two laser arms instead of three, and a different orbit (drifting away from Earth instead of 20 ◦ Earth trailing). The effect of these changes is a slightly reduced peak sensitivity and a shift to higher frequencies. We use an analytic fit to the instrumental noise curve given by Amaro-Seoane et al. (2013).
4.3.2. DECIGO, ALIA and BBO These missions are designed to probe the decihertz region of the GW spectrum; they are considerably more ambitious than the LISA or eLISA mission and their launches will be further into the future. We use a simple analytic fit to the sensitivity curve for ALIA (Bender et al., 2013), while for DECIGO and BBO, fits to the sensitivity curves given by Yagi and Seto (2011) are used.
## 4.4. Pulsar timing arrays
PTAs can be thought of as naturally occurring interferometers with galactic-scale arm lengths. Accordingly, they are sensitive to much lower frequencies than the detectors previously discussed. Each pulsar is a regular clock and the measured pulse arrival time can be compared against a prediction, leaving a residual which includes the effects of passing GWs. Using an array of these pulsars spread across the sky allows us to correlate residuals between different pulsars, to exploit the fact that GWs influence all pulsars whereas intrinsic pulsar noise does not. The correlation between different pulsars depends only on their angular separation on the sky, and has a distinctive shape, known as the Hellings and Downs curve (Hellings and Downs, 1983).
The redshift of the rate of arrival of pulses for a pulsar at a distance L from the SolarSystem barycentre (SSB), in the direction of the unit spatial vector ˆ induced by a GW p travelling in direction of the unit vector Ω is (Anholm et al., 2009) ˆ
$$z ( t, \hat { \Omega } ) = \frac { 1 } { 2 } \frac { \hat { p } ^ { j } \hat { p } ^ { i } } { 1 + \hat { \Omega } \cdot \hat { p } } \left [ h _ { i j } ^ { \text{Pulsar} } \left ( t - \frac { L } { c }, \hat { \Omega } \right ) - h _ { i j } ^ { \text{Earth} } ( t, \hat { \Omega } ) \right ] = \frac { 1 } { 2 } \frac { \hat { p } ^ { j } \hat { p } ^ { i } } { 1 + \hat { \Omega } \cdot \hat { p } } \Delta h _ { i j } ( t, \hat { \Omega } ) \.$$
The redshift includes two terms: the pulsar term and the Earth term. The pulsar term is often neglected in PTA analysis as it can be considered as an extra noise term which averages to zero across the array. The experimentally measured quantity is not the redshift but the timing residual, the two are related via
$$R ( t, \hat { \Omega } ) = \int _ { 0 } ^ { t } \mathrm d t ^ { \prime } \ z ( t ^ { \prime }, \hat { \Omega } ) \.$$
All of the pulsars, and the Earth, are subject to the same metric-perturbation field. This may be expressed in terms of its Fourier transform
$$h _ { i j } ( t, \vec { r } ) = \sum _ { A = +, \times } \int \mathrm d f \ \iint _ { \mathbb { S } _ { 2 } } \mathrm d \hat { \Omega } \ \tilde { h } _ { A } ( f, \hat { \Omega } ) e _ { i j } ^ { A } ( \hat { \Omega } ) \exp \left [ 2 \pi i f \left ( t - \frac { \hat { \Omega } \cdot \vec { x } } { c } \right ) \right ]$$
where e A ij ( ˆ ) Ω is the A polarisation basis tensor for direction Ω, and ˆ glyph[vector] r is the spatial position. Choosing the SSB as the origin of our coordinate system, so the pulsar is at position Lp ˆ, gives
$$\Delta h _ { i j } ( t, \hat { \Omega } ) = \sum _ { A = +, \, \times } \int \mathrm d f \, \tilde { h } _ { A } ( f, \hat { \Omega } ) e _ { i j } ^ { A } ( \hat { \Omega } ) \exp \left ( 2 \pi i f t \right ) \left \{ \exp \left [ - 2 \pi i f L \left ( 1 + \hat { p } \cdot \hat { \Omega } \right ) \right ] - 1 \right \} \.$$
From (53) and (56), the Fourier transform of the redshift ˜( z f, ˆ ) can be identified as Ω
$$\tilde { z } ( f, \hat { \Omega } ) = \left \{ \exp \left [ - 2 \pi i f L \left ( 1 + \hat { p } \cdot \hat { \Omega } \right ) \right ] - 1 \right \} \sum _ { A = +, \, \times } \tilde { h } _ { A } ( f, \hat { \Omega } ) F ^ { A } ( \hat { \Omega } )$$
where
$$F ^ { A } ( \hat { \Omega } ) = \frac { e _ { i j } ^ { A } ( \hat { \Omega } ) \hat { p } ^ { j } \hat { p } ^ { i } } { 2 \left ( 1 + \hat { \Omega } \cdot \hat { p } \right ) } \,.$$
The function F A ( ˆ ) may be regarded as the PTA equivalent of the detector response functions Ω in (49). The stochastic background of GWs is fully characterised by the one-sided PSD via the expectation value
$$\left \langle \tilde { h } _ { A } ^ { * } ( f, \hat { \Omega } ) \tilde { h } _ { A ^ { \prime } } ( f ^ { \prime }, \hat { \Omega } ^ { \prime } ) \right \rangle = \frac { 1 } { 2 } S _ { h } ( f ) \delta ^ { ( 2 ) } ( \hat { \Omega }, \hat { \Omega } ^ { \prime } ) \delta _ { A A ^ { \prime } } \delta ( f - f ^ { \prime } )$$
where δ (2) ( ˆ Ω ˆ ) is the delta-function on the sphere. , Ω ′ From (57) and (59), the expectation of the product of signals from two different pulsars in directions ˆ p 1 and ˆ p 2 may be evaluated as
$$\langle \tilde { z } _ { 1 } ( f ) \tilde { z } _ { 2 } ^ { * } ( f ^ { \prime } ) \rangle = \frac { 1 } { 2 } S _ { h } ( f ) \delta ( f - f ^ { \prime } ) \Gamma ( f ) \,,$$
where
$$\Gamma ( f ) = \sum _ { A = +, \, \times } & \iint _ { \mathbb { S } ^ { 2 } } \mathrm d \hat { \Omega } \left \{ \exp \left [ 2 \pi i f L _ { 1 } \left ( 1 + \hat { \Omega } \cdot \hat { p } _ { 1 } \right ) \right ] - 1 \right \} \\ & \times \left \{ \exp \left [ - 2 \pi i f L _ { 2 } \left ( 1 + \hat { \Omega } \cdot \hat { p } _ { 2 } \right ) \right ] - 1 \right \} F _ { 1 } ^ { A } ( \hat { \Omega } ) F _ { 2 } ^ { A } ( \hat { \Omega } ) \.$$
The overlap function Γ( f ) tends to a constant value in the limit that the distances to the pulsars are large compared to the wavelength of GWs; PTAs operate in this limit (Mingarelli and Sidery, 2014), so the overlap may be approximated as a constant,
$$\Gamma ( f ) \approx \Gamma _ { 0 } = \sum _ { A = +, \, \times } \iint _ { \mathbb { S } ^ { 2 } } \mathrm d \hat { \Omega } \ F _ { 1 } ^ { A } ( \hat { \Omega } ) F _ { 2 } ^ { A } ( \hat { \Omega } ) \.$$
Neglecting the exponential terms in the overlap is the frequency-domain equivalent of neglecting the pulsar term in (53). The integral may be evaluated to give an expression depending only on the angle θ between the two pulsars; this is the famous Hellings and Downs curve, shown in figure 2,
$$\Gamma _ { 0 } = \frac { 1 } { 2 } + \frac { 3 \xi } { 2 } \left ( \ln \xi - \frac { 1 } { 6 } \right ) \,,$$
where ξ = (1 -cos θ / ) 2 .
The sensitivity bandwidth of a PTA is set by the sampling properties of the data set. If measurements are spaced in time by δt and taken for a total length of time T , then the PTA is sensitive to frequencies in the range (1 /T ) < f < (1 /δt ). The characteristic strain that the PTA is sensitive to scales linearly with f in this range. This gives the wedge-shaped
Angular separation GLYPH<144> Figure 2: The Hellings and Downs (1983) curve, the correlation between two pulsars separated on the sky by an angle θ .
curves plotted in Appendix A. The absolute value of the sensitivity is fixed by normalising to a calculated limit at a given frequency for each PTA. For a discussion of the sensitivities of PTA to both individual sources and stochastic background, see Moore et al. (2014).
There is a discrepancy between the treatment of PTA sensitivity curves here and the higher frequency detectors discussed in sections 4.2 and 4.3. When observing a long-lived source, such as an inspiral, with a high frequency detector, the convention was to define a characteristic strain to satisfy (19). Here, the convention is to leave the strain untouched and instead adjust the PTA sensitivity curve with observation time, again to satisfy (19). This discrepancy is an unfortunate result of the conventions in use by the different GW communities; however, it is natural given the sources under observation. When observing a transient source, such as a burst or inspiral, which changes within the lifetime of the detector, it is natural to consider the detector as performing constantly while the signal changes. However, when observing a monochromatic source or a stochastic background, which is unchanging over the detector lifetime, it is more natural to consider the source as being fixed and the sensitivity of the detector gradually improving. All that is required by the definition in (19) is that the ratio h c ( f ) /h n ( f ) is constant.
4.4.1. Current PTAs The PTAs currently in operation are the European Pulsar Timing Array (EPTA ; Kramer and Champion, 2013), the Parkes Pulsar Timing Array (PPTA ‖ ¶ ; Hobbs, 2013) based in Australia, and the North American Nanohertz Observatory for Gravitational waves (NANOGrav + ; McLaughlin, 2013). There are published limits on the amplitude of the
‖ http://www.epta.eu.org/
¶ http://www.atnf.csiro.au/research/pulsar/ppta/
+ http://nanograv.org/
stochastic background from all three detectors: the most recent from EPTA is van Haasteren et al. (2011), from PPTA is Shannon et al. (2013) and from NANOGrav is Demorest et al. (2013). The limits from the existing detectors are all of a similar magnitude.
In Appendix A we use the EPTA limit based on an analysis of 5 pulsars over approximately 10 years. The curve labelled 'EPTA' assumes that all the pulsar were timed identically every 2 weeks with an r.m.s. error in each timing residual of 100 ns. The NANOGrav limit is comparable to this, and the PPTA limit is a factor ∼ 2 5 lower. .
4.4.2. The International Pulsar Timing Array Combining the existing arrays would yield a single PTA using approximately 4 times as many pulsars; this consortium of consortia is known as the International Pulsar Timing Array (IPTA ∗ ; Manchester and IPTA, 2013). The IPTA curve plotted in the figures assumes 20 pulsars timed every 2 weeks for 15 years with an r.m.s. error in each timing residual of 100 ns.
4.4.3. SKA The next great advancement in radio astronomy shall come with the completion of the Square Kilometre Array (SKA; Dewdney et al., 2009). This shall greatly increase the sensitivity of pulsar timing (Kramer et al., 2004). The sensitivity curves plotted in Appendix A for the SKA assumes 50 pulsars timed every 2 weeks for 20 years with an r.m.s. error in each timing residual of 30 ns. Our choice of 50 pulsars might be conservative, it is possible that the SKA will discover many more suitable milli-second pulsars than this throughout its operation, in which case the corresponding sensitivity curve would become lower.
## 5. Astrophysical sources
All the sources described here are represented by shaded boxes in Appendix A. Sources with short durations (i.e., burst sources) and sources that evolve in time over much longer timescales than our observations are drawn with flat-topped boxes for h c ( f ). Inspiraling binaries, or stochastic backgrounds of binaries, are drawn with a sloping top proportional to f -2 3 / for h c ( f ), which is result derived in section 3.2. The width of the box gives the range of frequencies sources of a given type can have while remaining at a detectable amplitude. The question of the height of the box is more problematic; it would be desirable to normalise each box so that there was a fixed event rate, say one event per year with an amplitude lying within the box. However, for many of the sources considered, the event rate is subject to a large degree of uncertainty (and estimates change rapidly as our understanding of the astrophysics improves). Consequently, we take the more definite, but somewhat arbitrary, approach that for each type of source a fiducial event is nominated (with parameters or amplitude detailed in the relevant section below) and the amplitude of this event is used to position the top of the box. The parameters for each fiducial event are chosen such that the
∗ http://www.ipta4gw.org/
resulting amplitude is roughly consistent with the optimistic end of current predictions: the boxes indicate where we could plausibly find GW sources; the actual event rates could turn out to be substantially lower than this upper bound.
## 5.1. Sources for ground-based detectors
5.1.1. Neutron-star binaries The inspiral and merger of a pair of neutron stars is the primary target for ground-based detectors. The expected event rate for this type of source is uncertain, but estimates centre around γ NS -NS = 1 3 . × 10 -4 Mpc -3 yr -1 (Abadie et al., 2012 c ). Plotted in Appendix A are boxes labelled ' compact binary inspirals ' with amplitudes such that a ratio h /h c n = 16 is produced for Advanced LIGO at peak sensitivity and a width between (3-300) Hz, corresponding to expected observable frequencies.
5.1.2. Supernovae Simulations of core-collapse supernovae show that GWs between (10 2 10 ) Hz can be produced (Kotake et al., 2006). The GW signal undergoes 3 O (1) oscillation and is hence burst-like. Dimmelmeier et al. (2002) calculate the average maximum amplitude of GWs for a supernova at distance r as
$$h _ { \max } = 8. 9 \times 1 0 ^ { - 2 1 } \left ( \frac { 1 0 \ k p c } { r } \right ) \,.$$
The event rate for supernovae is approximately γ SN = 5 × 10 -4 Mpc -3 yr -1 . The boxes labelled ' supernova ' plotted in Appendix A correspond to a distance r = 300 kpc with the frequency range quoted above. The LIGO and Virgo detectors have already placed bounds on the event rate for these sources (Abadie et al., 2012 a ).
5.1.3. Continuous waves from rotating neutron stars Rotating neutron stars are a source of continuous GWs if they possess some degree of axial asymmetry (Abbott et al., 2007; Prix, 2009; Aasi et al., 2013). The signals are near monochromatic with a frequency twice the rotation frequency of the neutron star, and are a potential source for ground-based detectors. The amplitude of the GWs depends upon the deformation of the neutron star, and its spin frequency. The magnitude of the distortion depends upon the neutron star equation of state and the stiffness of the crust, which are currently uncertain (Chamel and Haensel, 2008; Lattimer, 2012). Deformations can also be supported by internal magnetic fields (Haskell et al., 2008). Several known pulsars could be sources for the advanced detectors and upper limits from the initial detectors help to constrain the deformations.
The boxes labelled ' pulsars ' plotted in Appendix A correspond to the upper limits placed on a GW signal from the Crab pulsar (Aasi et al., 2014 ), extrapolated across a b frequency range between (20-10 ) Hz. The extrapolation was performed using the scaling 3 h 0 ∝ √ ˙ f/f (Aasi et al., 2014 b ) (this is the spin-down limit, which assumes that the only
loss of energy from the system is due to GW emission) and (35). This gives h c ∝ f 1 2 / ; higher frequency sources are observed at a louder SNR because they undergo more cycles.
## 5.2. Sources for space-based detectors
For a review of the GW sources for space-based missions see, for example, Amaro-Seoane et al. (2013), Gair et al. (2013) or Jennrich et al. (2011).
5.2.1. Massive black-hole binaries Space-based detectors shall be sensitive to equal-mass mergers in the range (10 -10 ) 4 7 M glyph[circledot] . Predictions of the event rate for these mergers range from O (10-100) yr -1 for eLISA with SNRs of up to 10 3 (Seoane et al., 2013). The large range in the rate reflects our uncertainty in the growth mechanisms of the supermassive black hole population (Volonteri, 2010). A 10 6 M glyph[circledot] binary can be observed out to a redshift z ∼ 8 with an SNR of 100 (Seoane et al., 2013). This fiducial source gives the amplitude of the boxes labelled ' ≈ 10 6 solar mass binaries ' in Appendix A. The range of frequencies is (3 × 10 -4 -3 × 10 -1 ) Hz, extrapolated using the slope h c ( f ) ∝ f -2 3 / derived in section 3.2.
5.2.2. Galactic white-dwarf binaries For space-based detectors, these are the most numerous GWsources; they are also the only guaranteed source since several detectable systems (known as verification binaries) have already been identified by electromagnetic observations (Stroeer and Vecchio, 2006).
Galactic binaries divide into two classes: the unresolvable and the resolvable galactic binaries. The unresolvable binaries overlap to form a stochastic background as discussed in section 3.2. The distinction between resolvable and unresolvable is detector specific; here we choose LISA. This boundary shall not be too different for eLISA, but would move substantially for decihertz detectors. Plotted in Appendix A with the label ' unresolvable galactic binaries ' is the estimate of this background due to Nelemans et al. (2001) where an observation time of one year has been assumed,
$$h _ { \text{c} } ( f ) = 5 \times 1 0 ^ { - 2 1 } \left ( \frac { f } { 1 0 ^ { - 3 } \ H z } \right ) ^ { - 2 / 3 } \,.$$
Estimates for the event rate of resolvable binaries centre around O (10 ) events for eLISA. 3 The boxes plotted in Appendix A with the label ' resolvable galactic binaries ' have a ratio h /h c n = 50 for eLISA at its peak sensitivity. The frequency range of the box is (3 × 10 -4 -10 -2 ) Hz, estimated from Monte Carlo population simulation results presented in Amaro-Seoane et al. (2013).
5.2.3. Extreme mass-ratio inspirals EMRIs occur when a compact stellar-mass object inspirals into a supermassive black hole. There is extreme uncertainty in the event rate for
EMRIs due to the poorly constrained astrophysics in galactic centres (e.g., Merritt et al., 2011); the best guess estimate is around 25 events per year with eLISA with SNR ≥ 20 (Seoane et al., 2013). The boxes labelled ' extreme mass ratio inspirals ' plotted in Appendix A have a characteristic strain of h c = 3 × 10 -20 at 10 -2 Hz, which corresponds to a 10 M glyph[circledot] black hole inspiralling into a 10 6 M glyph[circledot] black hole at a luminosity distance of 1 Gpc. The frequency width of the box is somewhat uncertain; EMRI events can occur around a black hole of any mass, and hence EMRIs can, in principal, occur at any frequency. The boxes in Appendix A are drawn with a width comparable to that of the LISA sensitivity curve.
## 5.3. Sources for PTAs
5.3.1. Supermassive black hole binaries The main target for PTAs is a stochastic background of GWs produced by a population of supermassive black-hole binaries at cosmological distances (Sesana et al., 2008). Supermassive black holes are known to lie at the centres of most galaxies and black-hole mergers are associated with the mergers of their host galaxies (Volonteri et al., 2003; Ferrarese and Ford, 2005). The current best 95% confidence limit for the amplitude of the stochastic background is h c = 2 7 . × 10 -15 at a frequency of f 0 = 1 yr -1 (Shannon et al., 2013). There is strong theoretical evidence that the actual background lies close to the current limit (Sesana, 2013). The distribution of source amplitudes with frequency is dependent on the chirp mass distribution of binaries, which is astrophysically uncertain. Reflecting our lack of information about the true shape of this box, we draw it with a flat top in characteristic strain.
5.3.2. Stochastic background of supermassive binaries Supermassive black-hole binaries at higher frequencies are inspiralling faster and, hence, there are fewer of them per frequency bin. At a certain frequency, these sources cease to be a background and become individually resolvable (Sesana et al., 2008; Sesana et al., 2009). It is currently unclear whether PTAs shall detect an individual binary or a stochastic background first. Plotted in Appendix A with the label ' stochastic background ' is a third of the current limit with a cut off frequency of f = 1 yr -1 . This is suggested by Monte Carlo population studies (Sesana et al., 2008), which give a range of plausible amplitudes the mean of which is plotted here. For the resolvable sources, labelled ' ≈ 10 9 solar mass binaries ', the amplitude of the current limit is plotted between (3 × 10 -9 -3 × 10 -7 ) Hz.
## 5.4. Cosmological sources
In addition to the sources above, early Universe processes, such as inflation (Grishchuk, 2005) or a first-order phase transition (Bin´truy et al., 2012), could have created GWs. e More speculatively, it has been hypothesised that cosmic strings could also be a potential source (Damour and Vilenkin, 2005; Bin´ etruy et al., 2012; Aasi et al., 2014 a ). These relic GWs
allow us to explore energy scales far beyond those accessible by other means, providing insight into new and exotic physics. The excitement surrounding the tentative discovery by BICEP2 of the imprint of primordial GWs (generated during inflation) in the cosmic microwave background (Ade et al., 2014), and the subsequent flurry of activity, has shown the scientific potential of such cosmological GWs. These GW signals are so alluring because they probe unknown physics; this also makes them difficult to predict. Cosmological stochastic backgrounds have been predicted across a range of frequencies with considerable variation in amplitude. As a consequence of this uncertainty, although we could learn much from measuring these signals, we have not included them amongst the sources in Appendix A.
The sensitivity curves plotted in Appendix A would change substantially for stochastic backgrounds. This is because searches for stochastic backgrounds utilise the cross correlation of the outputs of multiple detectors for detection (Allen and Romano, 1999; Maggiore, 2000), instead of the individual outputs. Therefore, the sensitivity of a network of ground based detectors is much greater than that of any individual detector. The shape of the sensitivity curve for a stochastic background would also differ from the curves plotted here, for a discussion of this point see Thrane and Romano (2013).
## 6. Concluding remarks
When quantifying the sensitivity of a GW detector and the loudness of a GW source, there are three commonly used quantities: the characteristic strain, the power spectral density, and the spectral energy density. We have carefully defined these quantities and derived the relationships between them. The characteristic strain (section 2.2) is most directly related to the SNR, the PSD is mostly closely related to the mean square amplitude in the detector, and the energy density has a clear physical interpretation. We have produced example plots using each of these quantities for a wide range of detectors and sources. The predicted source amplitudes are based on astrophysical estimates of the event rates and are subject to varying degrees of uncertainty. glyph[sharp] Interactive versions of these plots, with user-specified detectors and sources, are available on-line at http://rhcole.com/apps/GWplotter . Trying to summarise an entire field of astronomy on one plot is an impossible task; however, we hope that the figures and analysis presented here provide useful insight.
## Acknowledgments
CJM, RHC and CPLB are supported by the Science & Technology Facilities Council. We are indebted to Jonathan Gair, for useful advice and proof-reading the manuscript. We also thank Stephen Taylor and Justin Ellis for useful conversations about pulsar timing. The authors thanks a number of members of the LIGO-Virgo Scientific Collaboration, especially
glyph[sharp] There is also the exciting possibility of sources yet to be considered.
Chris Messenger, for useful comments and suggestions. This paper has been assigned LIGO document LIGO-P1400129.
## Appendix A. Sensitivity curves
The plots in this section show all of the detectors and sources described in the main text. Clearer, interactive versions of these plots, allowing for removal of any of the curves, may be created and downloaded on-line, http://rhcole.com/apps/GWplotter . The detector noise curves all have their resonance spikes removed for clarity.
Figure A1: A plot of characteristic strain against frequency for a variety of detectors and sources.
Figure A2: A plot of the square root of power spectral density against frequency for a variety of detectors and sources.
Figure A3: A plot of the dimensionless energy density in GWs against frequency for a variety of detectors and sources.
## References
Aasi J et al. 2013. Einstein@Home all-sky search for periodic gravitational waves in LIGO S5 data. Physcial Review D 87 (4), 042001. arXiv:1207.7176 .
Aasi J et al. 2014 a . Constraints on Cosmic Strings from the LIGO-Virgo Gravitational-Wave Detectors. Physical Review Letters 112 (13), 131101. arXiv:1310.2384 .
Aasi J et al. 2014 b . Gravitational waves from known pulsars: results from the initial detector era. The Astrophysical Journal 785 (2), 119. arXiv:1309.4027 .
Abadie J et al. 2010. All-sky search for gravitational-wave bursts in the first joint LIGOGEO-Virgo run. Physical Review D 81 (10), 102001. arXiv:1002.1036 .
Abadie J et al. 2012 a . All-sky search for gravitational-wave bursts in the second joint LIGO-Virgo run. Physical Review D 85 (12), 122007. arXiv:1202.2788 .
Abadie J et al. 2012 b . Search for Gravitational Waves Associated with Gamma-Ray Bursts during LIGO Science Run 6 and Virgo Science Runs 2 and 3. The Astrophysical Journal 760 (1), 12. arXiv:1205.2216 .
Abadie J et al. 2012 . c Search for gravitational waves from low mass compact binary coalescence in LIGO's sixth science run and Virgo's science runs 2 and 3. Physcial Review D 85 (8), 082002. arXiv:1111.7314 .
Abbott B P et al. 2009. LIGO: the Laser Interferometer Gravitational-Wave Observatory. Reports on Progress in Physics 72 (7), 076901. arXiv:0711.3041 .
Abbott B et al. 2007. Searches for periodic gravitational waves from unknown isolated sources and Scorpius X-1: Results from the second LIGO science run. Physical Review D 76 (8), 082001. arXiv:gr-qc/0605028 .
Accadia T et al. 2012. Virgo: a laser interferometer to detect gravitational waves. Journal of Instrumentation 7 (03), 03012.
Acernese F et al. 2009 Advanced virgo baseline design Virgo Technical Report VIR-0027A-09. http://tds.ego-gw.it/itf/tds/file.php?callFile=VIR-0027A-09.pdf
Ade P A R et al. 2014. Detection of B-Mode Polarization at Degree Angular Scales by BICEP2. Physical Review Letters 112 (24), 241101. arXiv:1403.3985 .
Allen B and Romano J D 1999. Detecting a stochastic background of gravitational radiation: Signal processing strategies and sensitivities. prd 59 (10), 102001. arXiv:gr-qc/9710117 .
Amaro-Seoane P et al. 2013. eLISA: Astrophysics and cosmology in the millihertz regime. GW Notes 6 , 4-110. arXiv:1201.3621 .
Ando M 2002. Current status of TAMA. Classical and Quantum Gravity 19 , 1409-1419.
Anholm M, Ballmer S, Creighton J D E, Price L R and Siemens X 2009. Optimal strategies for gravitational wave stochastic background searches in pulsar timing data. Physcial Review D 79 (8), 084030. arXiv:0809.0701 .
| Bender P L, Begelman M C and Gair J R 2013. Possible LISA follow-on mission scientific objectives. Classical and Quantum Gravity 30 (16), 165017. |
|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| Berry C P L and Gair J R 2013. Observing the Galaxy's massive black hole with gravitational wave bursts. Monthly Notices of the Royal Astronomical Society 429 (1), 589- 612. arXiv:1210.2778 . |
| Bin´truy e P, Boh´ e A, Caprini C and Dufaux J F 2012. Cosmological backgrounds of gravitational waves and eLISA/NGO: phase transitions, cosmic strings and other sources. Journal of Cosmology and Astroparticle Physics 2012 (06), 027. arXiv:1201.0983 . |
| Cannon K C 2008. A Bayesian coincidence test for noise rejection in a gravitational-wave burst search. Classical and Quantum Gravity 25 (10), 105024. |
| Chamel N and Haensel P 2008. Physics of Neutron Star Crusts. Living Reviews in Relativity 11 (10). arXiv:0812.3955 . |
| Chatterji S, Blackburn L, Martin G and Katsavounidis E 2004. Multiresolution techniques for the detection of gravitational-wave bursts. Classical and Quantum Gravity 21 (20), S1809. arXiv:gr-qc/0412119 . |
| Cutler C and Flanagan ´ E E 1994. Gravitational waves from merging compact binaries: How accurately can one extract the binary's parameters from the inspiral waveform? Physical Review D 49 (6), 2658-2697. arXiv:gr-qc/9402014 . |
| Damour T and Vilenkin A 2005. Gravitational radiation from cosmic (super)strings: Bursts, stochastic background, and observational windows. Physical Review D 71 (6), 063510. arXiv:hep-th/0410222 . |
| Demorest P B et al. 2013. Limits on the Stochastic Gravitational Wave Background from the North American Nanohertz Observatory for Gravitational Waves. ApJ 762 , 94. arXiv:1201.6641 . |
| Dewdney P E, Hall P J, Schilizzi R T and Lazio T J L W2009. The Square Kilometre Array. Proceedings of the IEEE 97 (8), 1482-1496. |
| 393 , 523-542. arXiv:astro-ph/0204289 . L and Ford H 2005. Supermassive Black Holes in Galactic Nuclei: Past, Present and |
| Ferrarese Future Research. Space Science Reviews 116 (3-4), 523-624. arXiv:astro-ph/0411247 . |
| Finn L S 1992. Detection, measurement, and gravitational radiation. Physical Review D 46 (12), 5236-5249. arXiv:gr-qc/9209010 . |
| Finn L S and Chernoff D F 1993. Observing binary inspiral in gravitational radiation: One interferometer. Physcial Review D 47 , 2198-2219. arXiv:gr-qc/9301003 . |
| Finn L S and Thorne K S 2000. Gravitational waves from a compact star in a circular, inspiral orbit, in the equatorial plane of a massive, spinning black hole, as observed by LISA. Physical Review D 62 (12), 124021. arXiv:gr-qc/0007074 . |
|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| Foster R S and Backer D C 1990. Constructing a pulsar timing array. Astrophysical Journal 361 , 300-308. |
| Gair J R, Vallisneri M, Larson S L and Baker J G 2013. Testing General Relativity with Low-Frequency, Space-Based Gravitational-Wave Detectors. Living Reviews in Relativity 16 , 7. arXiv:1212.5575 . |
| Grishchuk L P 2005. Relic gravitational waves and cosmology. Physics-Uspekhi 48 (12), 1235- 1247. arXiv:gr-qc/0504018 . |
| Grote H 2010. The GEO 600 status. Classical and Quantum Gravity 27 (8), 084003. |
| Harry G M 2010. Advanced LIGO: the next generation of gravitational wave detectors. Classical and Quantum Gravity 27 (8), 084006. |
| Haskell B, Samuelsson L, Glampedakis K and Andersson N 2008. Modelling magnetically deformed neutron stars. Monthly Notices of the Royal Astronomical Society 385 (1), 531-542. arXiv:0705.1780 . |
| Hellings R W and Downs G S 1983. Upper limits on the isotropic gravitational radiation background from pulsar timing analysis. Astrophyiscal Journal Letters 265 , L39-L42. |
| Hild S et al. 2011. Sensitivity studies for third-generation gravitational wave observatories. Classical and Quantum Gravity 28 (9), 094013. arXiv:1012.0908 . |
| Hobbs G 2013. The Parkes Pulsar Timing Array. Classical and Quantum Gravity 30 (22), 224007. arXiv:1307.2629 . |
| Iyer B et al. 2011 LIGO-India, Proposal of the Consortium for Indian Initiative in Gravitational-wave Observations (IndIGO) Technical Report M1100296. https://dcc. ligo.org/cgi-bin/DocDB/ShowDocument?docid=75988 Jenet F A et al. 2006. Upper Bounds on the Low-Frequency Stochastic Gravitational Wave |
| Background from Pulsar Timing Observations: Current Limits and Future Prospects. Astrophysical Journal 653 , 1571-1576. arXiv:astro-ph/0609013 . O et al. 2011 NGO Revealing a hidden Universe: opening a new chapter of discovery |
| Jennrich Technical Report ESA/SRE(2011)19 European Space Agency Noordwijk. |
| Klimenko S, Yakushin I, Mercer A and Mitselmakher G 2008. A coherent method for detection of gravitational wave bursts. Classical and Quantum Gravity 25 (11), 114029. arXiv:0802.3232 . |
| Kotake K, Sato K and Takahashi K 2006. Explosion mechanism, neutrino burst and gravitational wave in core-collapse supernovae. Reports on Progress in Physics 69 (4), 971- 1143. arXiv:astro-ph/0509456 . |
| Kramer M and Champion D J 2013. The European Pulsar Timing Array and the Large European Array for Pulsars. Classical and Quantum Gravity 30 (22), 224009. |
|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| Kramer M et al. 2004. Strong-field tests of gravity using pulsars and black holes. New Astronomy Reviews 48 (11-12), 993-1002. arXiv:astro-ph/0409379 . |
| Lattimer J M 2012. The Nuclear Equation of State and Neutron Star Masses. Annual Review of Nuclear and Particle Science 62 (1), 485-515. arXiv:1305.3510 . |
| Maggiore M 2000. Gravitational wave experiments and early universe cosmology. Physics Reports 331 , 283-367. arXiv:gr-qc/9909001 . |
| Manchester R N and IPTA 2013. The International Pulsar Timing Array. Classical and Quantum Gravity 30 (22), 224010. |
| McLaughlin MA2013. The North American Nanohertz Observatory for Gravitational Waves. Classical and Quantum Gravity 30 (22), 224008. arXiv:1310.0758 . |
| Merritt D, Alexander T, Mikkola S and Will C M2011. Stellar dynamics of extreme-mass-ratio inspirals. Physical Review D 84 (4), 044024. arXiv:1102.3180 . |
| Mingarelli C M F and Sidery T 2014. The effect of small inter-pulsar distance variations in stochastic gravitational wave background searches with Pulsar Timing Arrays. arXiv:1408.6840 . |
| Misner C W, Thorne K S and Wheeler J A 1973 Gravitation W.H. Freeman and Co. San Fansisco. |
| Moore C J, Taylor S R and Gair J R 2014. Estimating the sensitivity of pulsar timing arrays. arXiv:1406.5199 . |
| Nelemans G, Yungelson L R and Portegies Zwart S F 2001. The gravitational wave signal from the Galactic disk population of binaries containing two compact objects. Astronomy and Astrophysics 375 , 890-898. arXiv:astro-ph/0105221 . |
| Ott C D 2009. The gravitational-wave signature of core-collapse supernovae. Classical and Quantum Gravity 26 (6), 063001. arXiv:0809.0695 . |
| Peters P C and Mathews J 1963. Gravitational Radiation from Point Masses in a Keplerian Orbit. Physical Review 131 , 435-440. |
| Prix R 2009 in WBecker, ed., 'Neutron Stars and Pulsars' Astrophysics and Space Science Library Springer Berlin chapter 24, pp. 651-685. |
| Sathyaprakash B S and Schutz B F 2009. Physics, Astrophysics and Cosmology with Gravitational Waves. Living Reviews in Relativity 12 , 2. arXiv:0903.0338 . |
| Sesana A 2013. Systematic investigation of the expected gravitational wave signal from supermassive black hole binaries in the pulsar timing band. Monthly Notices of the Royal Astronomical Society 433 , L1-L5. arXiv:1211.5375 . |
| Sesana A, Vecchio A and Colacino C N 2008. The stochastic gravitational-wave background from massive black hole binary systems: implications for observations with Pulsar Timing Arrays. Monthly Notices of the Royal Astronomical Society 390 , 192-209. |
|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| arXiv:0804.4476 . A, Vecchio A and Volonteri M 2009. Gravitational waves from resolvable massive |
| Sesana black hole binary systems and observations with Pulsar Timing Arrays. Monthly Notices of the Royal Astronomical Society 394 (4), 2255-2265. arXiv:0809.3412 . |
| Shannon R Met al. 2013. Gravitational-wave limits from pulsar timing constrain supermassive black hole evolution. Science 342 (6156), 334-337. arXiv:1310.4569 . |
| Somiya K 2012. Detector configuration of KAGRA-the Japanese cryogenic gravitational-wave detector. Classical and Quantum Gravity 29 (12), 124007. arXiv:1111.7185 . |
| Stroeer A and Vecchio A 2006. The LISA verification binaries. Classical and Quantum Gravity 23 , 809. arXiv:astro-ph/0605227 . |
| Thorne K S 1987 in S. WHawking and WIsrael, eds, 'Three Hundred Years of Gravitation' Cambridge University Press Cambridge chapter 9, pp. 330-458. |
| Thrane E and Romano J D 2013. Sensitivity curves for searches for gravitational-wave backgrounds. Physical Review D 88 (12), 124032. arXiv:1310.5300 . |
| Unnikrishnan C S 2013. IndIGO and Ligo-India Scope and Plans for Gravitational Wave Research and Precision Metrology in India. International Journal of Modern Physics D 22 (1), 1341010. |
| Vallisneri M and Galley C R 2012. Non-sky-averaged sensitivity curves for space- based gravitational-wave observatories. Classical and Quantum Gravity 29 (12), 124015. arXiv:1201.3684 . |
| van Haasteren R et al. 2011. Placing limits on the stochastic gravitational-wave background using European Pulsar Timing Array data. Monthly Notices of the Royal Astronomical Society 414 (4), 3117-3128. arXiv:1103.0576 . |
| Volonteri M 2010. Formation of supermassive black holes. The Astronomy and Astrophysics Review 18 (3), 279-315. arXiv:1003.4404 . |
| Volonteri M, Haardt F and Madau P 2003. The Assembly and Merging History of Supermassive Black Holes in Hierarchical Models of Galaxy Formation. The Astrophysical Journal 582 (2), 559-573. arXiv:astro-ph/0207276 . |
| Wiener N 1949 The extrapolation, interpolation and smoothing of stationary time series The M.I.T. Press Cambridge, Massachusetts. |
| Willke B et al. 2006. The GEO-HF project. Classical and Quantum Gravity 23 , 207. |
|
10.1088/0264-9381/32/1/015014
|
[
"Christopher J. Moore",
"Robert H. Cole",
"Christopher P. L. Berry"
] |
2014-08-04T16:37:05+00:00
|
2014-12-11T16:35:39+00:00
|
[
"gr-qc",
"astro-ph.HE"
] |
Gravitational-wave sensitivity curves
|
There are several common conventions in use by the gravitational-wave
community to describe the amplitude of sources and the sensitivity of
detectors. These are frequently confused. We outline the merits of and
differences between the various quantities used for parameterizing noise curves
and characterizing gravitational-wave amplitudes. We conclude by producing
plots that consistently compare different detectors. Similar figures can be
generated on-line for general use at \url{http://rhcole.com/apps/GWplotter}.
|
1408.0741v2
|
## Effective models of membranes from symmetry breaking
Omar Zanusso 1, ∗
1 Institute for Mathematics, Astrophysics and Particle Physics, Radboud University Nijmegen, Heyendaalseweg 135, 6525 AJ Nijmegen, The Netherlands
We show how to obtain all the models of the continuous description of membranes by constructing the appropriate non-linear realizations of the Euclidean symmetries of the embedding. The procedure has the advantage of giving a unified formalism with which the models are generated and highlights the relevant order parameters in each phase. We use our findings to investigate a fluid description of both tethered and hexatic membranes, showing that both the melting and the loss of local order induce long range interactions in the high temperature fluid phase. The results can be used to understand the appearance of intrinsic ripples in crystalline membranes in a thermal bath.
PACS numbers:
## I. INTRODUCTION.
An effective model of membrane is the continuous description of an intrinsically two-dimensional object. Examples of nature's realized membranes are ubiquitous and are of both biological and non biological nature. On the biological side, cells' membranes, which are characterized by both fluid (amphiphilic bilayers) and crystalline (cytoskeleton) properties, are a significant example [1]. On the other hand, the interest in nonbiological two-dimensional crystals has been renewed in recent years due to the discovery of graphene and similar materials with enormous potential technological applications [2]. From both a theoretical and a phenomenological point of view, it is of interest to investigate the phasediagram of these structures to gain better knowledge of their behavior under thermal and mechanical stresses.
In a statistical mechanics framework the phasediagram can be obtained assigning a microscopic (bare) Hamiltonian to the membrane model and computing the corresponding free-energy via a path-integral formalism. The formalism allows to identify the critical points of the model, which separate the salient mechanical phases of the membrane. For the sake of this introduction, it is useful to consider the membrane as an effective description of a layer of fundamental constituents (monomers) linked together by a bonding interaction of fixed connectivity (crystal). The phase-diagram of a theory of membranes is generally very complex because two-dimensional geometry allows for various definitions of local order with corresponding order-parameters.
In a continuous formulation the membrane is the image of a map
$$r \, \colon \mathbb { R } ^ { d } \to \mathbb { R } ^ { D } \,, \quad \quad \ \ ( 1 ) \quad \ \ e v.$$
where in the physically interesting case of a twodimensional membrane embedded in three-dimensional
∗ Electronic address: [email protected]
space one has d = 2 and D = 3, but for the sake of generality in most of this work we will leave the couple ( d, D ) general. As a general requirement, we demand any membrane model to be invariant under the isometries of the embedding
$$r ^ { \mu } \rightarrow R ( \alpha ) ^ { \mu } _ { \, \nu } r ^ { \nu } + b ^ { \mu } \,,$$
where R α ( ) is any D -dimensional rotation parametrized by some angles α and b is a general translation vector in R D . The continuous formulation is of course only an effective description of the fundamental monomers' interactions. If a is the typical intra-monomer length, the effective continuous description will generally work well for hcβ glyph[lessmuch] a where β = 1 /k B T is the inverse of the temperature in units of the Boltzmann constant. In fact, the examples cited above are characterized by a ∼ 1 nm (bilayers and graphene) and a ∼ 1 µ m (cytoskeleton), and admit an accurate continuous description at the higher length scales of 10 a -100 a .
We refer to the study of statistical field theories of (1) as the Kosterlitz-Thouless-Nelson-Halperin-Young (KTNHY) description [3-5]. To summarize it, it is convenient to introduce the binding energy scale E b which might or might not be of order 1 /a . One can distinguish lowT glyph[lessmuch] E /k b B , highT glyph[greatermuch] E /k b B and intermediatetemperature T glyph[similarequal] E /k b B phases. Ideally, at low temperatures the bindings are intact and the monomers are in their 'natural' crystalline phase. In this phase the local connectivity is unaltered, thus the thermal fluctuations are elastic and can affect only the bindings' lengths. At high temperatures the bindings melt and the membrane undergoes a fluid phase that is characterized by the absence of both local connectivity and order. In the intermediate temperatures phase, the binding interactions are assumed to be relevant, but not dominant. The intermediate regime describes the melting and the local connectivity plays the role of the order parameter [6, 7].
In this paper we want to show how to derive the three main effective models of the KTNHY description in a unified formalism by considering the breaking of the global symmetries of the embedding space R D due to the pres-
ence of the membrane [8]. For this purpose, we will use the formalism of the Maurer-Cartan form (MCF) which proves very efficient in the investigation of symmetry breaking patterns. The MCF formalism has the advantage of clearly identifying the order parameters of the broken symmetries through a coset construction. The order parameters obtained in this way thus enjoy a formal definition as outlined in [9].
## II. THE COSET CONSTRUCTION AND ISO( D ) .
In this section we want to briefly review the use of the MCF in the construction of a coset model that represents the symmetry breaking pattern of a theory of membranes (1) and involves the breaking of both internal and embedding symmetries [10]. The interested reader can find more details in the excellent recent review in [11]. Let G be the symmetry group of a membrane theory and let it be the direct product of ISO( D ), which is the Euclidean isometry group of the embedding (2), and some compact internal symmetry group G int related to the crystalline structure of the membrane
$$G = \text{ISO} ( D ) \times G _ { \text{int} } \,. \quad \quad \ ( 3 ) \quad \text{wh}$$
The group G is thus generated by the standard Euclidean translation P µ and rotation J µν operators that enjoy the algebra
$$\begin{array} { c } [ J _ { \mu \nu }, J _ { \rho \sigma } ] & = i \left ( \delta _ { \mu \rho } J _ { \nu \sigma } + \text{perm.} \right ), \\ [ J _ { \mu \nu }, P _ { \rho } ] & = i \left ( \delta _ { \mu \rho } P _ { \nu } - \delta _ { \nu \rho } P _ { \mu } \right ), \\ [ P _ { \mu }, P _ { \rho } ] & = 0 \,, \end{array} \quad ( 4 )$$
as well as by some generator T m of the internal symmetry group G int such that
$$[ T _ { m }, J _ { \mu \nu } ] = 0 \,, \ [ T _ { m }, P _ { \mu } ] = 0 \,. \quad \ \ ( 5 ) \quad t h ^ { \cdot \cdot }$$
We assume that a general membrane configuration breaks G spontaneously to the unbroken subgroup H ⊂ G . We also assume that H might be non-compact, implying that some translations of ISO( D ) might not be broken by the physical configurations. Excluding the unbroken translations from H , it is possible to identify a compact subgroup H 0 ⊂ H , which is generated by the unbroken rotations and internal symmetries of G . In this case an effective membrane model can be obtained as the coset
$$G / H _ { 0 } \,.$$
From the general theory of cosets, the effective model will be manifestly invariant under the subgroup H 0 , but it will maintain the full G symmetry, albeit it will be realized, at least partly, non-linearly.
Any element of the coset (6) can be parametrized by an equivalence class of elements of G under the right-action of H 0 . It is convenient to choose a representative for the equivalence class of the form
$$\omega ( r, \xi ) = e ^ { i r ^ { \mu } P _ { \mu } } \, e ^ { i \xi ^ { A } X _ { A } } \,, \quad \quad \ \ ( 7 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
where X A are all the broken generators of both embedding and internal symmetries. The equivalence class is thus related to ω r, ξ ( ) as
$$[ \omega ( r, \xi ) ] = \{ \omega ( r, \xi ) \cdot h ; \, \forall h \in H _ { 0 } \} \, \quad \quad ( 8 )$$
and by construction any of its elements has the same physical field content. In general, ω r, ξ ( ) is parametrized by the field content of the model r µ and by the Goldstone fields ξ A associated with the broken symmetries. By construction, the element (7) transforms under G by the leftaction of a generic element g ∈ G as ω r, ξ ( ) → · g ω r, ξ ( ). The action of G may bring one coset representative into another inequivalent one, thus defining implicitly the transformation properties of r µ and ξ A via
$$\omega ( r, \xi ) \to g \cdot \omega ( r, \xi ) \equiv \omega ( r ^ { \prime }, \xi ^ { \prime } ) \cdot h ^ { \prime } \,, \quad \quad ( 9 )$$
where h ′ belongs to the unbroken H 0 and returns the representative to the form (7).
The effective model for the field arguments of (7) can be constructed using the MCF, which is defined as
$$L = \omega ( r, \xi ) ^ { - 1 } \mathrm d \omega ( r, \xi ) \,, \, \text{ \quad \ \ } ( 1 0 )$$
which by construction belongs to the algebra of G . Once the physical content of L is identified via its algebra components an effective Hamiltonian for the theory with broken symmetries can be written down. In the following we will follow this method, adopting physically motivated symmetry breaking patterns for all the models of the KTNHY description.
## III. THE TETHERED MEMBRANE.
At low temperatures the bonding interactions among the monomers are intact and favor a crystalline structure [6, 12]. In this situation each monomer fluctuates under mechanical stresses as a point-like object as in Fig. 1. A point-like object has a definite position in space, therefore its configurations break all translations, while leaving the Euclidean rotations invariant. For simplicity, in this Section we neglect any internal symmetry group of the monomers' lattice. The full group of symmetries is thus G = ISO( D ), while the unbroken symmetries are H 0 = SO( D ). The coset is thus
$$\text{ISO} ( D ) \, / \text{SO} ( D ) \,.$$
As representative of the coset we choose the simple
$$\omega = \mathrm e ^ { i r ^ { \mu } P _ { \mu } } \,.$$
We now introduce a set of coordinates x α of R d and compute the MCF as
$$L _ { \alpha } = \omega ^ { - 1 } \partial _ { \alpha } \omega = i \partial _ { \alpha } r ^ { \mu } P _ { \mu } \,. \quad \quad ( 1 3 )$$
The components e µ α ≡ ∂ α r µ of the MCF are the Goldstone fields of the broken translations and are formally
FIG. 1: Tethered membrane. Each monomer (black dot) is regarded as a point-like object that breaks fully the translational invariance of the embedding. In the crystalline phase the bindings (lines) are intact, it is relevant to deform the membrane by displacing each single monomer in any direction.
defined as the order parameters of the model in the sense of [9]. By construction, the components e µ α transform linearly under the SO( D ) rotation subgroup of (2), but are scalars under the broken translations:
$$\begin{array} { c } e ^ { \mu } _ { \alpha } \to R ( \alpha ) ^ { \mu } _ { \nu } \, e ^ { \nu } _ { \alpha } & \text{for } r ^ { \mu } \to R ( \alpha ) ^ { \mu } _ { \nu } \, r ^ { \nu } \,, \quad & ( 1 4 ) \\ e ^ { \mu } _ { \alpha } \to e ^ { \mu } _ { \alpha } & \text{for } r ^ { \mu } \to r ^ { \mu } + b ^ { \mu } \,. \end{array} \quad \text{of } t l$$
A general Hamiltonian for a crystalline phase can thus be constructed only from e µ α and its derivatives. Up to fourth order in a derivative expansion and expressing e µ α in terms of the membrane configuration r µ , the most general Hamiltonian is
$$\beta \mathcal { H } _ { t } \left [ r \right ] = & \int \mathrm d ^ { d } x \left \{ \epsilon _ { 0 } + \frac { \mu _ { t } } { 2 } ( \partial _ { \alpha } r ^ { \mu } ) ^ { 2 } + \frac { \kappa _ { t } } { 2 } ( \partial ^ { 2 } r ^ { \mu } ) ^ { 2 } \\ & + u ( \partial _ { \alpha } r ^ { \mu } \partial _ { \beta } r ^ { \mu } ) ^ { 2 } + v ( \partial _ { \alpha } r ^ { \mu } \partial _ { \alpha } r ^ { \mu } ) ^ { 2 } \right \},$$
which is known as the tethered membrane model [6, 12]. We neglected possible boundary terms and adopted a rather general notation for the couplings that have been introduced: glyph[epsilon1] 0 is a chemical potential for the monomer number that can be neglected at fixed membrane volume, µ t is the surface tensions, κ t is the extrinsic rigidity, and u and v are Lam´ e coefficients parametrizing the elastic properties of the membrane.
The fields e µ α are the order parameters of the model and distinguish the possible mechanical phases of the theory. To be more precise we introduce a path integral with which Boltzmann averages of any operator O [ r ] can be computed as
$$\langle \mathcal { O } [ r ] \rangle = \int D r \, \mathcal { O } [ r ] \, e ^ { - \beta \mathcal { H } _ { t } [ r ] } \,, \quad \ \ ( 1 6 ) \quad \underset { \text{nor} } { \text{plac} }$$
where we defined a properly normalized measure D r that is invariant under (14). Two mechanical phases can be roughly distinguished as glyph[negationslash]
$$\begin{array} { c } \left \langle \i d ^ { d } x \, e _ { \alpha } ^ { 2 } \right \rangle \Big / \i d ^ { d } x = 0 & \text{crumpled} \,, \quad & \left ( 1 7 \right ) & \text{app} \\ \left \langle \i d ^ { d } x \, e _ { \alpha } ^ { 2 } \right \rangle \Big / \i d ^ { d } x \neq 0 & \text{flat} \,, \end{array} \right ) \, \stackrel { \lambda \, \iota } { \exp }$$
where e 2 α indicates the square of the order parameter e 2 = ∑ µ e µ α α e µ (no summation over α ). Each order parameter e µ α enjoys a separate treatment, thus anisotropic phases in which the membrane is crumpled along some directions while extended along the others are possible [13]. For a more detailed description of these phases we refer to [1]. The order of the phase-transition separating crumpled and flat phases is still subject to investigation and is potentially very relevant for applications in the developing physics of graphene. The most recent nonperturbative treatment as well as lattice simulations suggests that the crumpled-to-flat transition is of first order [14, 15].
## IV. THE FLUID MEMBRANE.
At high temperatures the bindings among the monomers melt, which are then free to diffuse along the membrane [16]. At macroscopic equilibrium, two membrane configurations cannot be distinguished if a monomer is translated along the tangential directions of the membrane as in Fig. 2. To picture the phase,
FIG. 2: Fluid membrane. The monomers are free to diffuse, so the in-plane displacements do not change the (macroscopic) continuous description of the membrane.
it is convenient to consider the infinitesimal plaquette of the membrane in position r µ defined by r µ + d r µ = r µ + ∂ α r µ d x α . Infinitesimally, the plaquette breaks spontaneously the translations perpendicular to the plane generated by ∂ α r µ d x α and the rotations that do not leave the same plane invariant. Again, we shall neglect any possible internal symmetry group in the course of this Section.
The (compact) unbroken subgroup for the fluid membrane is therefore H 0 = SO( ) d × SO( D -d ) and includes both the SO( d ) subgroup of local frame rotations on the plaquette and the subgroup SO( D -d ) of rotations of its normal space. The coset is thus
$$\text{ISO} ( D ) \, \slash \text{SO} ( d ) \times \text{SO} ( D - d ) \. \quad \quad ( 1 8 )$$
In this Section we will follow closely the construction of [11], which was developed in a different context, but applies here with minor modifications. Before choosing a coset representative, it is convenient to switch to a system
of coordinates in R D that locally aligns the first d axes with the tangents of the membranes and the remaining D -d ones with their normal space. In the new system of coordinates the membrane is described by the couple r µ = ( r α , r i ). The coset representative is then chosen to include all the broken generators as well as the unbroken translations
$$\omega = e ^ { i r ^ { \alpha } P _ { \alpha } + i r ^ { i } P _ { i } } e ^ { i \xi ^ { \alpha i } J _ { \alpha i } } \,. \quad \quad \quad ( 1 9 )$$
For later use, we introduce a further set of coordinates y a on the membrane whose purpose will become clear below.
In order to correctly identify the field content of the model, we first choose a general parametrization of the MCF of the form
$$L _ { a } = i e _ { a } ^ { \, \alpha } ( P _ { \alpha } + \nabla _ { \alpha } \pi ^ { i } P _ { i } + \nabla _ { \alpha } \xi ^ { \gamma i } J _ { \gamma i } + A _ { \alpha } ^ { \, \beta \gamma } J _ { \beta \gamma } ) \,, \, ( 2 0 ) \quad \stackrel { \lambda } { \text{fields} } ^ { \, \lambda }$$
where both π i and ξ γi have to be thought of as Goldstone fields of the broken translations and rotations, respectively, while A α βγ are unimportant tensors corresponding to unbroken local rotations on the membrane. The components ∇ α π i correspond to the broken normal translations. Whenever translations are broken in a system the usual counting of the degrees of freedom of the Goldstone theorem is more subtle, as there is not necessarily one Goldstone field for each broken symmetry generator. The mismatch in the counting follows what is known as the inverse Higgs mechanism. The redundant Goldstone fields can be eliminated by imposing a so-called inverse Higgs constraint [17]. The simplest possible constraint for our system is
$$\nabla _ { \alpha } \pi ^ { i } = 0 \,, \quad \quad \ \ ( 2 1 ) \quad d \epsilon$$
while all other possible choices can be related to this one by a redefinition of the couplings. The solution of (21) can then be used to express the Goldstone fields of the broken translations π i in terms of those of the broken rotations ξ αi . Through this mechanism the correct counting of the Goldstone fields is achieved. From a physical point of view, it is possible to interpret the theory constrained by (21) in various complementary ways [11]. In this context, we can understand the constrained theory as the one in which the modes π i have been effectively integrated out and refer to Sect. V for further insights on this interpretation.
Using (19) the explicit computation of the components of the form (20) corresponding to the translations gives
$$e _ { a } ^ { \, \alpha } & = \partial _ { a } r ^ { \mu } R ( \xi ) _ { \mu } ^ { \, \alpha } \\ e _ { a } ^ { \, \alpha } \nabla _ { \alpha } \pi ^ { i } & = \partial _ { a } r ^ { \mu } R ( \xi ) _ { \mu } ^ { \, \, i } \,, \quad \text{(2)} \quad \text{with} \quad \text{ext} \quad \text{the} \quad \text{$\quad$}$$
where we introduced the orthogonal rotation R ξ ( ) parametrized by the Goldstone fields of the broken rotations. The condition (21) simply implies
$$\partial _ { a } r ^ { \mu } R ( \xi ) _ { \mu } ^ { \, \, i } = 0 \,, \quad \quad \ \ ( 2 3 ) \quad \text{m} \epsilon$$
which defines the orthonormal basis R ξ ( ) µ i = n i µ . The normal basis is obviously directly related to the Goldstone fields of the broken rotations.
The fields e a α transform covariantly under the unbroken rotations and can be used as frames on the membrane. To show this and give a geometrical meaning to the construction we introduce the metric
$$g _ { a b } = e _ { a } ^ { \, \alpha } e _ { b } ^ { \, \beta } \delta _ { \alpha \beta } \,.$$
Using (22) and the invariance of the embedding space metric δ αβ under orthogonal rotations it is very easy to see that the metric corresponds to the induced metric on the membrane g ab = ∂ r a µ ∂ r b ν δ µν using the new coordinate patch y a . From now on, Latin indices from the beginning of the alphabet will be raised and lowered using g ab and its inverse. The derivatives of the Goldstone fields can be interpreted geometrically too. We first compute the components of the form (20) corresponding to the broken rotations
$$e _ { a } ^ { \, \alpha } \nabla _ { \alpha } \xi ^ { \gamma i } \, = \, [ R ( \xi ) ^ { - 1 } \partial _ { a } R ( \xi ) ] ^ { \gamma i } \,. \quad \ \ ( 2 5 )$$
Introducing the inverse frame e α a ≡ e -1 α a , it can be directly inverted and related to the normal basis as
$$\nabla _ { \alpha } \xi ^ { \gamma i } \, = \, e _ { \alpha } ^ { \, a } R ( \xi ) ^ { - 1 \gamma \mu } \partial _ { a } n _ { \mu } ^ { i } \,. \, \quad \ \ ( 2 6 )$$
This expression should be compared with the definition of extrinsic curvature of a membrane, which we manipulate using the constraint (23) as
$$K _ { a b } ^ { i } \equiv \partial _ { b } r ^ { \mu } \partial _ { a } n _ { \mu } ^ { i } = e _ { b } ^ { \beta } R ( \xi ) _ { \beta } ^ { \mu } \partial _ { a } n _ { \mu } ^ { i } \,. \quad \ \ ( 2 7 )$$
Combining (26) and (27) we immediately see that the derivatives of the Goldstone fields of the broken rotations are directly related to the extrinsic curvatures as
$$K _ { a b } ^ { i } \, = \, e _ { a } ^ { \, \alpha } e _ { b } ^ { \, \beta } \nabla _ { \alpha } \xi ^ { i } _ { \, \beta } \,. \, \text{ \quad \ \ } ( 2 8 )$$
The most general Hamiltonian that can be constructed from the MCF can only be a function of e a α , ∇ α ξ i β and their derivatives. Due to the geometrical meaning of these quantities it can thus be written as a function of metric g ab and extrinsic curvatures K i ab . The Hamiltonian will also be reparametrization invariant for transformations of the arbitrary coordinate y a . Neglecting boundary terms, the most general invariant Hamiltonian up to second order in the derivatives is
$$\beta \mathcal { H } _ { f } \left [ r \right ] \, = \, \int \mathrm d ^ { d } y \, \sqrt { g } \left \{ \mu _ { f } + \frac { \kappa _ { f } } { 2 } K ^ { 2 } + \frac { \bar { \kappa } } { 2 } R \right \}, \ \ ( 2 9 )$$
where K 2 is the square K 2 = K K i i of the traces of the extrinsic curvatures K i = g ab K i ab , g ab is the inverse of the induced metric (24), R = g ab R ab = g ab R ca c b is the (intrinsic) curvature scalar, R ab c d is the Riemann tensor defined by [ ∇ a , ∇ b ] v c = R ab c d v d , ∇ a is the metriccompatible connection defined by ∇ a v b = ∂ v a b +Γ a b c v c with the Christoffel symbols obtained from the induced metric Γ a b c = 1 2 g bd ( ∂ g a dc + ∂ g c ad -∂ g d ac ) and g = det g ab
is the determinant of the metric. This is the so-called fluid membrane model, which is also well known in string theory [18]. The coupling µ f is the surface tensions, κ f is the (fluid) extrinsic rigidity, and ¯ is known as the κ Gaussian rigidity of the membrane.
In the fluid model reparametrization invariance has a clear physical origin that we shall outline before concluding the Section. The general mechanical deformation of the fluid membrane would perturb it along both tangential and normal directions like in the tethered model of the previous Section. It can be parametrized as
$$r ^ { \mu } \rightarrow r ^ { \mu } + \nu ^ { a } \partial _ { a } r ^ { \mu } + \nu ^ { i } n _ { i } ^ { \mu } \,. \quad \quad ( 3 0 ) \quad \text{cha}$$
From the discussion above, however, it is clear that at its equilibrium the fluid membrane macroscopic state is unaffected by tangential deformations ν a corresponding to translations of the monomers along the membrane. Reparametrization invariance ensures that the transformations ν a can be absorbed by a corresponding reparametrization of the coordinates y a which is infinitesimally parametrized by a vector field tangent to the membrane itself. A path-integral can be constructed using (29) as
$$\langle \mathcal { O } [ r ] \rangle = \int D r _ { \text{rep} } \, \mathcal { O } [ r ] \, \text{e} ^ { - \beta \mathcal { H } _ { f } [ r ] - \beta \mathcal { H } _ { g. \ell. } } \,, \quad ( 3 1 ) \quad \text{the s}$$
where D r rep is a new normalized measure that respects reparametrization invariance and H g f . . is an opportune gauge fixing term. When constructing (31) it is often convenient to adopt the background field method and choose the physical gauge for which ν a = 0, which is analogous to the Landau gauge of Yang-Mills theories. Albeit quite different from the point of view of the coset construction, the fluid membrane shares the same mechanical phases of the tethered membrane and the same discussion of (17) applies [19]. It can however manifest a phase-transition of different order. We will return to this point in the next Section, highlighting a substantial difference between the two models.
## V. FROM TETHERED TO FLUID MEMBRANE.
In this Section we attempt a fluid model description of the tethered membrane. We will achieve it by properly integrating the appropriate Goldstone fields. For simplicity we will deal with the Hamiltonian of the tethered model (15) truncated to the second order of the derivative expansion
$$\beta \mathcal { H } _ { t } \left [ { \mathbf r } \right ] \, = \, \int \mathrm d ^ { d } x \left \{ \epsilon _ { 0 } + \frac { \mu _ { t } } { 2 } ( \partial _ { \alpha } r ^ { \mu } ) ^ { 2 } \right \}, \quad ( 3 2 ) \quad \text{ that.}$$
which has to be regarded as a toy model that neglects both the rigidity and the elastic properties. We notice that this Hamiltonian is evidently not reparametrization invariant and this is the main difference from the fluid model of the previous Section. In this sense we interpret the coordinates x α as a fiducial set that describes the position of the monomers on the membrane and has a special role if compared to the arbitrary coordinate patch y a introduced for the fluid model.
The task is now to rewrite this action in terms of the fields appearing in the nonlinear realization (20) before imposing the inverse Higgs constraint. This can be done performing the symmetry transformation
$$\partial _ { a } r ^ { \mu } \rightarrow R ( \xi ) ^ { \mu } _ { \, \nu } \partial _ { a } r ^ { \nu } \,,$$
however it is also necessary to perform a coordinate change from the 'crystalline' patch x α to the arbitrary coordinate patch y a with the transformation
$$\partial x ^ { \alpha } / \partial y ^ { a } = e _ { a } ^ { \, \alpha } \,.$$
The result of the manipulations is
$$\cdot \quad \beta \mathcal { H } _ { t } \left [ r \right ] \, = \, \int \mathrm d ^ { d } y \sqrt { g } \left \{ \epsilon _ { 0 } + \frac { d } { 2 } \mu _ { t } + \frac { \mu _ { t } } { 2 } ( \nabla _ { a } \pi ^ { i } ) ^ { 2 } \right \}, ( 3 5 )$$
where g ab is again the induced metric (24) and π i was introduced in (20).
The task is now to integrate the field π i , which here plays the role of the order parameter for the breaking of the symmetry
$$\text{SO} ( D ) \rightarrow \text{SO} ( d ) \times \text{SO} ( D - d )$$
which characterizes the compact symmetry content of the transition from the tethered to the fluid descriptions. The fields π i transform as D -d scalars under SO( d ). The path integration of the Goldstone fields π i is performed using the properly normalized measure D π as
$$e ^ { - \beta \mathcal { H } _ { t - \ell } [ r ] } = \int \text{D} \pi \, \text{e} ^ { - \beta \mathcal { H } _ { t } [ r, \pi ] } \,, \quad \quad ( 3 6 )$$
and provides a physical realization of the inverse Higgs mechanism (21) of Sect. IV. For simplicity we shall consider the physical model with d = 2 and D = 3. In this case π i = π is a single scalar and thus an integration of its kinetic term in (35) that preserves reparametrization invariance gives the well known Liouville action (see for example [20]). We obtain an effective fluid Hamiltonian for the tethered model as
$$\beta \mathcal { H } _ { t - f } \left [ r \right ] \, = \, \int \mathrm d ^ { 2 } y \sqrt { g } \left \{ \mu _ { R } - \frac { 1 } { 9 6 \pi } R \frac { 1 } { \Delta } R \right \}, \ \ ( 3 7 )$$
where we introduced a renormalized tension
$$\mu _ { R } = \epsilon _ { 0 } + \frac { d } { 2 } \mu _ { t }$$
that could be easily evinced from (35), the Laplacian operator ∆ = -∇ a ∂ a , and the curvature scalar of the induced metric R . The new effective Hamiltonian enjoys reparametrization invariance as required for a fluid description. Interestingly, the integration of the local order parameter controlling the breaking of the symmetry
interfacing a tethered-fluid transition gives rise to long range interactions among the intrinsic curvatures of the fluid phase.
Long range interactions such as that of (37) are of particular importance in two-dimensional systems, as they are known to provide a way out of the Mermin-Wagner theorem [21] and are responsible for the non-triviality of systems that undergo a Kosterlitz-Thouless type of transitions [3]. The sole presence of the long range interactions casts the question on whether the tethered and fluid models should share the same phase-diagram. A complete answer to this question is still unknown, even though in the past it has been investigated at length using lattice methods [15, 22]. To gain a better insight physical quantities should be compared among the different phases of the two models at least qualitatively. The advantage of the formalism described in this paper is that it makes possible a quantitative comparison of the results for the phase-diagrams of the two models.
One particularly important observable in the continuum is the fractal (Hausdorff) dimension d f of the membrane [7], which here represents how fuzzy a crystalline membrane becomes in the process of melting. Given (37) we can compute an estimate of the membrane's fractal dimension using results originally developed for the hexatic model, which will be discussed in the next Section. Using either the one-loop result of David et al. [7] or the nonperturbative renormalization group result of [23] we obtain
$$\begin{smallmatrix} d _ { f } = 1 4 & \text{one-loop} \,, \\ d _ { f } = 2. 5 7 & \text{non-perturbative} \,. \end{smallmatrix} \quad ( 3 8 )$$
The high numerical difference between the two numbers is due to the fact that an expansion in the inverse coupling of the Liouville action is performed in the one-loop result, making the non-perturbative result more reliable. The physical understanding of these numbers goes as follows: The Liouville interaction of (37) tends to attract intrinsic curvatures of different sign, making the surface more and more fuzzy as different curvatures are packed together as hinted by Fig. 3.
FIG. 3: Naive visualization of the long range interactions in the fluid description of tethered and hexatic membranes. Configurations of the membrane with alternating sign of the scalar curvature are energetically favored.
The non-perturbative result resums the curvature interactions to a higher extent and thus includes some higher-order screening that considerably lowers the value of d f . Since we are neglecting any internal symmetry the melting tethered surface under consideration is not characterized by local order. Our result thus predicts that an unstructured crystal undergoes a crinkled phase characterized by a definite fractal dimension in the process of melting. In the crinkled phase the membrane is characterized by intrinsic ripples of the surface. The physics of ripples is relevant for the understanding of the stability of two-dimensional crystals [24].
## VI. THE HEXATIC MEMBRANE.
The hexatic membrane can be understood as a melting tethered membrane or as a fluid membrane with the residual effect of the local order that breaks the internal rotations, thus describing the intermediate temperature phase separating the crystalline low-temperature phase with the high-temperature fluid one [7]. In the physical case, the crystalline structure has to break the local SO(2) rotations tangent to the membrane as depicted in Fig. 4. To generalize this situation while maintaining the desired d = 2 limit, we assume that the crystal has an internal G int = SO( ) d cr symmetry with generators T A = S αβ emerging from the continuous description of some discrete lattice group, but more general breaking patterns can be studied in similar ways. Differently from the previous Sections, the full group of symmetries is enhanced to
$$G = \text{ISO} ( D ) \times \text{SO} ( d ) _ { \text{cr} } \,.$$
The local rotations on the membrane with generators J αβ are broken, but a combined rotation of the membrane and lattice leaves the configuration invariant. We thus require that the subgroup
$$\text{SO} ( d ) \times \text{SO} ( d ) _ { \text{cr} } \subset \text{ISO} ( D ) \times \text{SO} ( d ) _ { \text{cr} } \quad ( 4 0 )$$
is broken to the diagonal SO( ) generated by d M αβ = J αβ -S αβ .
As coset representative we choose a structure that enhances the breaking pattern (20) to the internal rotations
$$\omega = e ^ { i r ^ { \alpha } P _ { \alpha } + i r ^ { i } P _ { i } } e ^ { i \xi ^ { \alpha i } J _ { \alpha i } + ( i / 2 ) \xi ^ { \alpha \beta } J _ { \alpha \beta } } \,. \quad ( 4 1 )$$
The MCF is then parametrized as
$$L _ { a } = i e _ { a } ^ { \, \alpha } \left ( P _ { \alpha } + \nabla _ { \alpha } \pi ^ { i } P _ { i } + \nabla _ { \alpha } \xi ^ { \gamma i } J _ { \gamma i } + \frac { 1 } { 2 } \nabla _ { \alpha } \xi ^ { \beta \gamma } J _ { \beta \gamma } \right ),$$
where the fields e a α still play the role of frames for the membrane and new Goldstone fields ξ βγ for the breaking of SO( ) appear. d The first three components are in a form equal to (22) and (25), while the remaining one can be computed as
$$e _ { a } ^ { \, \alpha } \nabla _ { \alpha } \xi _ { \beta } ^ { \, \gamma } = ( R ( \xi ) ^ { - 1 } \partial _ { a } R ( \xi ) ) _ { \beta } ^ { \, \gamma } \,, \quad \ \ ( 4 2 )$$
for a rotation R ξ ( ) that is parametrized by ξ β i and ξ β γ .
FIG. 4: Hexatic membrane. The crystalline structure breaks the local rotations of the membrane. The arrows represent the order parameter for the associated phase-transition. A disordered phase is depicted.
As in the fluid case, the constraint ∇ α π i = 0 identifies a set of orthonormal vectors defined by n i µ = R ξ ( ) µ i . In this case, however, an additional set of orthonormal tangent vector fields is identified as N γ µ = R ξ ( ) µ γ and describes the breaking of the local rotations. They are related to the new Goldstone fields through the MCF as
$$\nabla _ { \alpha } \xi ^ { \beta \gamma } \, = \, e _ { \alpha } ^ { \ a } R ( \xi ) ^ { - 1 \beta \mu } \partial _ { a } N _ { \mu } ^ { \gamma } \,. \quad \ \ ( 4 3 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
The construction of the Hamiltonian follows closely the fluid case, but includes the new order parameter for the broken rotations. At second order in the derivative expansion the Hamiltonian contains (29) and new terms involving the Goldstone fields ξ αβ that can only be of the form
$$\beta \mathcal { H } \left [ r, \xi \right ] \, = \, \frac { K _ { A } } { 2 } \int \mathrm d ^ { d } y \sqrt { g } \, I _ { \alpha \beta \gamma \delta } \nabla _ { a } \xi ^ { \alpha \beta } \nabla ^ { a } \xi ^ { \gamma \delta } \,, \, ( 4 4 )$$
where I αβγδ is a tensor containing the details of the breaking of the combination of rotations and internal symmetries. The coupling K A has been introduced to parametrize the strength of the new Goldstone fields interaction and is known as hexatic rigidity. For the case of diagonal breaking the symmetry requirement implies that we simply have I αβγδ = δ αγ δ βδ .
In the physically interesting case of d = 2 and D = 3 there is only one local rotation on the membrane, which is fully broken by almost all possible discrete lattice rotations. The corresponding Goldstone field is an angular SO(2) variable θ = ξ 12 , which we can associate with a tangent vector on the membrane defining N α = cos θ e 1 α +sin θ e 2 α . When expressed in terms of N α , the Hamiltonian becomes a well-known formulation of the hexatic term
$$\beta \mathcal { H } _ { \text{hex} } \left [ r, N \right ] \, = \, \frac { K _ { A } } { 2 } \int \mathrm d ^ { d } y \sqrt { g } \, \nabla _ { a } N ^ { \alpha } \nabla ^ { a } N ^ { \alpha } \,. \, \ ( 4 5 ) \, \underset { \mod \epsilon } { \text{and} } }$$
At second order the vector N appears quadratically and can thus be integrated away following the procedure of [7] that brings the integration over the new field in a form similar to (36). The result of the integration is again a
Liouville action contribution to the Hamiltonian of fluid model of the form
$$\beta \mathcal { H } _ { h - f } \left [ \mathbf r \right ] \, = \, - \frac { \bar { K } _ { A } } { 8 } \int \mathrm d ^ { 2 } y \sqrt { g } \, R \frac { 1 } { \Delta } R \,, \quad ( 4 6 )$$
where the coupling ¯ K A underwent only a finite renormalization ¯ K A = K A -1 / 12 π . The same comments of the previous section on the long range interactions induced by a Liouville action apply in this context as well. The term (46) is fundamental to circumvent the fact that a fluid membrane has an extended phase only at zero coupling (1 /κ f = 0) [25]. The hexatic model, in fact, is known to display a non-trivial extended phase for a finite value of κ f . The fractal dimension of this phase is known perturbatively [7] in an expansion in 1 /K ¯ A as
$$d _ { f } = 2 + \frac { D ( D - 2 ) } { 3 \pi } \bar { K } _ { A } ^ { - 1 } + \mathcal { O } ( \bar { K } _ { A } ^ { - 2 } ) \, \quad ( 4 7 )$$
and has been investigated non-perturbatively in [23].
The nontrivial spectral dimension (47) is a manifestation of the fact that the hexatic membrane has equilibrium configurations characterized by intrinsic ripples. Ripples are well known and studied in the physics of graphene [24]. While the results of Sect. V predict a definite spectral dimension when interfacing a tethered and a fluid membrane, it is clear from (47) that the characterization of the fractal properties of the hexatic model require the experimental determination of the hexatic rigidity, which enters as an effective parameter in (46).
## VII. SUMMARY.
In a unified formalism, we obtained all the models of the KTNHY description of continuous membranes as field theories that nonlinearly realize the Euclidean symmetries of the embedding space. Each specific coset was constructed using the very efficient formalism of the Maurer-Cartan form and is motivated by the physical properties that the fundamental constituents of the membrane are assumed to posses according to the temperature of the thermal bath. The inclusion of the physical properties of the constituents and of their crystalline structure is essential for reproducing all the distinct phases of the KTNHY description, therefore distinguishing our effective models of membranes from those in which the membrane is considered a fundamental object such as string theory [8].
Immediately identified in the effective description obtained through the Maurer-Cartan form are both the Goldstone fields of the broken symmetries of embedding and the relevant order parameters for the phases of each model, which in these models have immediate physical interpretations [11]. In all models, the order parameters obtained via the coset construction coincide with those that were already well-established in the literature [1], but within the coset construction they acquire a formal definition in the sense of [9].
The construction of the KTNHY models shows clearly the degrees of freedom that have to be effectively integrated in the low- and intermediate-temperature regimes to obtain effective fluid descriptions for all of them. The effective integration of both the low-temperature crystalline phase and of the intermediate hexatic phase give effective fluid models that possess long range interactions among intrinsic curvatures on the membrane, which are believed to induce crinkled phases at equilibrium [25]. The crinkled phase is determined by the presence and interactions of intrinsic ripples of the surface characterizing the membrane [24]. While the importance of role of long range interactions among displacements of the membrane is already well known for the characterization of the flat phase of a crystalline membrane [14], we showed how these effectively manifest in a fluid description and gave a clear geometrical interpretation.
The long range interactions are induced by the thermal fluctuations, by the melting of the crystalline structure and by the loss of local order among the fundamental constituents. The sole presence of the long range interactions casts doubts on whether a crystalline and a fluid descriptions may share the same phase-diagram. For a physical two-dimensional membrane embedded in a three-dimensional space there is no conclusive proof whether the phase-diagram is shared or not. This work can be considered as a step forward in the direction of
- [1] M. J. Bowick and A. Travesset, Phys. Rept. 344 , 255 (2001) [cond-mat/0002038].
- [2] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov and A. K. Geim, Rev. Mod. Phys. 81 , 109 (2009).
- [3] J. M. Kosterlitz and D. J. Thouless, J. Phys. C 6 , 1181 (1973).
- [4] D. R. Nelson and B. I. Halperin, Phys. Rev. B 19 , 2457 (1979).
- [5] A. P. Young, Phys. Rev. B 19 , 1855 (1979).
- [6] D. R. Nelson and L. Peliti, J. Physique 48 (1987) 10851092;
- [7] F. David, E. Guitter and L. Peliti, J. Physique 48 (1987) 2059-2066.
- [8] P. C. West, JHEP 0002 , 024 (2000) [hep-th/0001216].
- [9] S. Weinberg, Cambridge, UK: Univ. Pr. (1996) 489 p. (Chapter 19).
- [10] J. Gomis, K. Kamimura and P. C. West, Class. Quant. Grav. 23 , 7369 (2006) [hep-th/0607057]; J. Gomis, K. Kamimura and J. M. Pons, Nucl. Phys. B 871 , 420 (2013) [arXiv:1205.1385 [hep-th]].
- [11] L. V. Delacretaz, S. Endlich, A. Monin, R. Penco and F. Riva, arXiv:1405.7384 [hep-th].
- [12] M. Paczuski, M. Kardar and D. R. Nelson, Phys. Rev. Lett. 60 , 2638 (1988).
- [13] K. Essafi, J. -P. Kownacki and D. Mouhanna, Phys. Rev. Lett. 106 , 128102 (2011).
- [14] K. Essafi, J. -P. Kownacki and D. Mouhanna, Phys. Rev. E 89 , 042101 (2014) [arXiv:1402.0426 [cond-mat.statmech]].
- [15] A. Tr¨ster, Phys. Rev. B o 87 , 104112 (2013); A. Tr¨ster, o
a better understanding of this question and a further indication that it has to be addressed non-perturbatively [14, 19].
Our results are especially relevant for the melting of both regular and unstructured two-dimensional crystals into a fluid phase. In such a scenario, the crystal undergoes a crinkled phase that is characterized by an anomalous fractal dimension of the membrane due to the presence of effective long range interactions among the monomers. In our computations, the spectral dimension is a genuine prediction of the formalism when the crystal does not exhibit local order even in the crystalline phase, while it depends on the hexatic rigidity when local order is present.
## Acknowledgments.
The author thanks M. Demmel, G. D'Odorico, S. Endlich and R. Penco for illuminating discussions and valuable comments on the draft of this paper. The work benefited by the hospitality of the Perimeter Institute for Theoretical Physics during the workshop 'Low Energy Challenges for High Energy Physicists', where a first draft of ideas was developed. This work is supported by the DFG within the Emmy-Noether program (Grant SA/1975 1-1).
- J. Phys.: Conf. Ser. 454 , 012032 (2013).
- [16] L. Peliti and S. Leibler, Phys. Rev. Lett. 54 , 1690 (1985).
- [17] E. A. Ivanov and V. I. Ogievetsky, Teor. Mat. Fiz. 25 , 164 (1975); I. Low and A. V. Manohar, Phys. Rev. Lett. 88 , 101602 (2002) [hep-th/0110285].
- [18] A. M. Polyakov, Nucl. Phys. B 268 , 406 (1986).
- [19] A. Codello and O. Zanusso, Phys. Rev. D 83 , 125021 (2011) [arXiv:1103.1089 [hep-th]].
- [20] E. Mottola, J. Math. Phys. 36 , 2470 (1995) [hepth/9502109]; A. Codello, Annals Phys. 325 , 1727 (2010) [arXiv:1004.2171 [hep-th]].
- [21] N. D. Mermin and H. Wagner, Phys. Rev. Lett. 17 , 1133, 1966.
- [22] S. M. Catterall, Phys. Lett. B 220 , 207 (1989); J. Ambjorn, J. Jurkiewicz, S. Varsted, A. Irback and B. Petersson, Phys. Lett. B 275 , 295 (1992); J. Ambjorn, A. Irback, J. Jurkiewicz and B. Petersson, Nucl. Phys. B 393 , 571 (1993); M. J. Bowick, P. Coddington, L. -p. Han, G. R. Harris and E. Marinari, Nucl. Phys. B 394 , 791 (1993); K. Anagnostopoulos, M. J. Bowick, P. Coddington, M. Falcioni, L. -p. Han, G. R. Harris and E. Marinari, Phys. Lett. B 317 , 102 (1993).
- [23] A. Codello and O. Zanusso, Phys. Rev. E 88 , 022135 (2013) [arXiv:1304.4953 [cond-mat.stat-mech]].
- [24] A. Fasolino, J. H. Los, M. I. Katsnelson, Nature Materials 6, 858-861 (2007).
- [25] J. -M. Park and T. C. Lubensky, Phys. Rev. E 53 , 2648 (1996); J. -M. Park and T. C. Lubensky, Phys. Rev. E 53 , 2665 (1996).
|
10.1103/PhysRevE.90.052110
|
[
"O. Zanusso"
] |
2014-08-04T16:43:07+00:00
|
2014-10-31T15:37:00+00:00
|
[
"cond-mat.stat-mech",
"hep-th"
] |
Effective models of membranes from symmetry breaking
|
We show how to obtain all the models of the continuous description of
membranes by constructing the appropriate non-linear realizations of the
Euclidean symmetries of the embedding. The procedure has the advantage of
giving a unified formalism with which the models are generated and highlights
the relevant order parameters in each phase. We use our findings to investigate
a fluid description of both tethered and hexatic membranes, showing that both
the melting and the loss of local order induce long range interactions in the
high temperature fluid phase. The results can be used to understand the
appearance of intrinsic ripples in crystalline membranes in a thermal bath.
|
1408.0742v1
|
Using the local gyrokinetic code, GS2, to investigate global ITG modes in tokamaks. (I) s -α model with profile and flow shear effects
P. A. Abdoul ∗ 1 , D. Dickinson , C. M. Roach 1 2 and H. R. Wilson 1
1 York Plasma Institute, Department of Physics, University of York, Heslington, York, YO10 5DD,
UK CCFE, Culham Science Centre, Abingdon, Oxfordshire, OX14 3DB, UK
2
August 5, 2014
## Abstract
This paper combines results from a local gyrokinetic code with analytical theory to reconstruct the global eigenmode structure of the linearly unstable ion-temperature-gradient (ITG) mode with adiabatic electrons. The simulations presented here employ the s α -tokamak equilibrium model. Local gyrokinetic calculations, using GS2 have been performed over a range of radial surfaces, x , and for ballooning phase angle, p , in the range -π ≤ p ≤ π , to map out the complex local mode frequency, Ω ( 0 x, p ) = ω 0 ( x, p ) + iγ 0 ( x, p ). Assuming a quadratic radial profile for the drive, namely η i = L /L n T , (holding constant all other equilibrium profiles such as safety factor, magnetic shear etc.), Ω 0 ( x, p ) has a stationary point. The reconstructed global mode then sits on the outboard mid plane of the tokamak plasma, and is known as a conventional or isolated mode, with global growth rate, γ ∼ Max[ γ 0 ( x, p )], where γ 0 ( x, p ) is the local growth rate. Taking the radial variation in other equilibrium profiles (e.g safety factor q x ( )) into account, removes the stationary point in Ω ( 0 x, p ) and results in a mode that peaks slightly away from the outboard mid plane with a reduced global growth rate. Finally, the influence of flow shear has also been investigated through a Doppler shift, ω 0 → ω 0 + n Ω ′ x , where n is the toroidal mode number and Ω ′ incorporates the effect of flow shear. The equilibrium profile variation introduces an asymmetry to the growth rate spectrum with respect to the sign of Ω , consistent with recent global gyrokinetic calculations. ′
## 1 Introduction
Tokamaks [1] provide one of the most stable and promising configurations for magnetic confinement fusion. However, confinement in tokamaks is not perfect. There are a number of mechanisms by which energy and particles can be transported across the magnetic flux surfaces from the core confinement region to the plasma edge. The main contribution is typically due to turbulent transport, which is widely believed to originate primarily from microinstabilities, driven by density and temperature gradients. The drift-type modes with low frequency compared to the cyclotron frequency, are the dominant tokamak microinstabilities [1, 2], and the turbulence they derive often determines the minimum size of magnetic confinement fusion reactors, such us ITER and DEMO. Therefore, it is important to understand them and find a way to suppress them and/or reduce their effects. Previous theoretical studies have shown that flow shear can control the stability of drift waves, providing a mechanism to suppress or even stabilise them completely [3, 4, 5, 6, 7, 8, 9]. Furthermore, understanding the structure of these instabilities and associated threshold gradients is likely to be very important for developing simplified models for heat and particle transport. Microinstabilities are typically investigated via numerical simulations of the gyrokinetic equations [10, 11] using both local and global approaches, as we now discuss.
For high toroidal mode numbers, n , the rational flux surfaces (i.e. those where the safety factor, q, is rational) are closely packed so equilibrium quantities vary only weakly from one to the next. Two length scales can be identified: the equilibrium length scale, characterised by the minor radius, r , and the distance
∗ [email protected]
between rational surfaces, ∆ = 1 /nq ′ where a prime denotes derivative with respect to r . The separation between these length scales is exploited in ballooning theory [12, 13] to perform an expansion in the small parameter, ∆ /r . All rational surfaces spanned by the mode are then equivalent to leading order, and local gyrokinetic codes, like GS2 [14, 15], exploit this 'ballooning symmetry to reduce the intrinsic 2D eigenvalue problem to 1D in the extended ballooning coordinate that is aligned with magnetic field lines, denoted by η . Local gyrokinetic codes do not directly yield the radial mode structure, but they do provide the structure along the magnetic field line, together with the local mode eigenvalues, Ω ( 0 x, p ), depending on radial coordinate x = r -r 0 a , where a is the minor radius of the tokamak plasma, and the ballooning angle parameter, p . At this order, both x and p are free parameters in the local gyrokinetic simulations. However, they are constrained in the higher order ballooning theory, which then yields the radial mode structure and global eigenvalue, Ω (which is a complex mode frequency).
In this paper, we shall reconstruct the global mode structures purely from local solutions, employing the Fourier-Ballooning representation [16] described in section 2. According to this equation, the global mode structure is related to both the local mode structure ξ x ( p, θ, t ) and its corresponding complex mode frequency Ω ( 0 x, p ). It has previously been established that under the very special conditions where Ω ( 0 x, p ) has a stationary point, a so-called isolated or conventional ballooning mode is obtained from the higher order analytical theories. This type of mode, originally studied in the context of ideal MHD theory [17], is usually located at the outboard mid plane, where the poloidal angle θ = 0 [3]. Profile variations, however, are usually present in realistic situations, and global gyrokinetic simulations do not always find isolated modes. For example, global gyrokinetic simulations of the linear electrostatic ion temperature gradient (ITG) mode in ASDEX Upgrade plasmas, predicted that the mode was slightly shifted relative to the outboard mid plane [18]. However, in exceptional situations, e.g. when a critical value of flow shear compensates the effect of profile variation, an isolated mode can exist [19]. More generally, linear global gyrokinetic simulations have shown that the radial profile variations break the up-down poloidal symmetry of the mode, providing a mechanism for toroidal momentum transport [20]. This transport influences so-called intrinsic rotation [21], which is of crucial importance in generating plasma flows in a machine like ITER for which external torque is very small.
Using a simplified fluid model of ITG modes, it has been shown that global mode structures can be accurately constructed from local ballooning results for given radial profiles [3]. Our work builds on that by applying the theory to a more realistic and generally applicable kinetic plasma model. Specifically the global mode structures of an ion temperature gradient driven microinstability have been reconstructed directly from local gyrokinetic solutions from the GS2 code. To achieve this, numerical solutions from GS2, have been used with the higher order analytical theory presented in reference [12]. We have carried out these calculations using the s -α equilibrium model [22, 23], in which circular magnetic flux surfaces and high aspect ratio have been assumed.
This paper is organised as follows. Section 2 presents the theoretical formalism on which this paper is based. The results, from applying this technique to our example problem, are presented in section 3. Finally, section 4 summarises our conclusions, and plans for future work.
## 2 The technique: From local to global gyrokinetic calculations
We employ the initial value local gyrokinetic code, GS2 [14, 15], to solve the linearised gyrokinetic equations numerically [10, 11]. GS2 provides the mode structure along a given magnetic field line with its local eigenvalue, Ω ( 0 x, p ) for a prescribed radial coordinate, x = r -r 0 a , and ballooning angle, p . Here, r is the minor radius, and x is a normalised distance from the rational surface at r = r 0 , with r = a being the plasma edge. To investigate the linear global mode properties by using only solutions from the local gyrokinetic code, we employ the Fourier-Ballooning (FB) representation [16]
$$\phi ( x, \theta, t ) = \int _ { - \infty } ^ { \infty } \xi _ { x } ( p, \theta, t ) \exp ( - i n q _ { 0 } \theta ) \exp ( - i n q ^ { \prime } x ( \theta - p ) ) A ( p ) d p$$
Here, φ x, θ, t ( ) is the global mode. The parameter q 0 is the safety factor evaluated at x = 0. The function ξ x ( p, θ, t ), obtained from the local gyrokinetic code, GS2, represents the local mode structure where the poloidal angle, θ measures the distance along a field line. Note that ξ is invariant under the transformation
p → p + 2 πl and θ → θ + 2 πl for any integer l . The envelope A p ( ) is required to reconstruct the radial mode structure and is obtained from the higher order theory. From equation(1) and the aforementioned symmetry properties of ξ , A must be periodic in p in order for φ to be periodic in θ .
We seek a global eigenmode satisfying:
$$\frac { d \phi ( x, \theta, t ) } { d t } = - i \Omega \phi ( x, \theta, t )$$
Where, Ω = ω + iγ represents the true, global complex mode frequency, with ω and γ , corresponding to the frequency and growth rate, respectively. Substituting equation(1) into (2), and noting dξ x ( p,θ,t ) dt = -i Ω ( 0 x, p ) ξ x ( p, θ, t ) for the local eigenfunction ξ with local complex mode frequency Ω ( 0 x, p ), gives:
$$\Omega \phi ( x, \theta, t ) = \int _ { - \infty } ^ { \infty } \Omega _ { 0 } ( x, p ) \xi _ { x } ( p, \theta, t ) \exp ( - i n q _ { 0 } \theta ) \exp ( - i n q ^ { \prime } x ( \theta - p ) ) A ( p ) d p$$
Note that, the local complex mode frequency Ω ( 0 x, p ) = ω 0 ( x, p ) + iγ 0 ( x, p ), where ω 0 and γ 0 are the local frequency and growth rate, respectively, is measured in unit of ( v ti /R ). It follows that time is measured in unit of ( R/v ti ). Using equation(1) for φ , we can then write our eigenmode condition in the form:
$$\int _ { - \infty } ^ { \infty } ( \Omega - \Omega _ { 0 } ( x, p ) ) \xi ( p, \theta, t ) \exp ( - i n q _ { 0 } \theta ) \exp ( - i n q ^ { \prime } x ( \theta - p ) ) A ( p ) d p = 0$$
Here, the local complex mode frequency, Ω ( 0 x, p ), is mapped out by running the local gyrokinetic code, GS2, many times over the required range of x and a period of p . We then Taylor expand in x and Fourier expand in p to derive:
$$\Omega _ { 0 } ( x, p ) = \sum _ { k = 0 } ^ { N _ { k } } \sum _ { m = 0 } ^ { 2 } a _ { k } ^ { ( m ) } x ^ { m } c o s ( k p )$$
where N k is the number of Fourier modes retained. The coefficients a ( m ) k are complex numbers, derived by fitting to the Ω 0 ( x, p ) results from GS2. In the limit of high toroidal mode number, n → ∞ , exploited by local gyrokinetic codes, the radial extents of the reconstructed global modes are very small compared to the equilibrium scale lengths. Therefore, we do not expect the complex mode frequency to have radial components higher than second order, i.e m ≤ 2. Now substituting our fit for Ω 0 ( x, p ) from equation(5) into equation(4), we have
$$\sum _ { k = 0 } ^ { N _ { k } } \int _ { - \infty } ^ { \infty } \ \left [ ( \Omega - a _ { k } ^ { ( 0 ) } \cos ( k p ) ) A - \frac { i } { n q ^ { \prime } } a _ { k } ^ { ( 1 ) } \cos ( k p ) \frac { d A } { d p } + \frac { 1 } { ( n q ^ { \prime } ) ^ { 2 } } a _ { k } ^ { ( 2 ) } \cos ( k p ) \frac { d ^ { 2 } A } { d p ^ { \prime } } \right ] \\ \xi _ { x } ( p, \theta, t ) \exp ( - i n q _ { 0 } \theta ) \exp [ - i n q ^ { \prime } x ( \theta - p ) ] d p = 0$$
where we have integrated by parts and assumed that A p ( ) has the fastest variation with p to derive the mapping x A k → -( i/nq ′ ) k d A/dp k k . Equation(6) must hold for all θ , which then provides our final equation for A p ( ):
$$\ddots \upsilon r ^ { \prime } \cdot \\ a _ { k } ^ { ( 2 ) } \cos ( k p ) \frac { 1 } { ( n q ^ { \prime } ) ^ { 2 } } \frac { d ^ { 2 } A } { d p ^ { 2 } } - a _ { k } ^ { ( 1 ) } \cos ( k p ) \frac { i } { n q ^ { \prime } } \frac { d A } { d p } + [ \Omega - a _ { k } ^ { ( 0 ) } \cos ( k p ) ] A = 0$$
This is to be solved subject to a periodicity boundary condition to determine the eigenfunction A p ( ) and the global complex mode frequency, Ω as an eigenvalue. Knowledge of A p ( ), together with ξ x ( p, θ, t ) from GS2 then allows the full 2D eigenfunction, φ x, θ, t ( ), to be reconstructed from equation(1).
glyph[negationslash]
Equation(7) can be solved analytically in two limiting cases: either a (1) k = 0 or a (2) k = 0 for all k ; the former implies that Ω ( 0 x, p ) is stationary at x = 0, i.e. ∂ Ω 0 /∂x | x =0 = 0. These limits were considered in references [3, 12], for example, to derive the two different eigenmode classes which they referred to as 'isolated modes' in the special cases when a (1) k = 0 and 'general modes' in the more usual situation that a (1) k = 0 (in which case the terms in a (2) k can often be neglected as they are small in the limit of large n ). In this paper, we shall retain both terms and solve equation(7) numerically.
Figure 1: From left to right: The variation of (a) real frequency, ω 0 , (b) linear growth rate, γ 0 , with k ρ y i and (c) the local mode structure, ξ η, p ( = 0), for the most unstable mode ( k ρ y i = 0 45), along the magnetic . field line, η . Note that, both ω 0 and γ 0 are measured in unit of ( v ti /R ) and these simulations have been carried out at mid-radius, i.e x = 0 ( r = r 0 ).
Figure 2: Radial profile for the temperature gradient, R/L T = 17 00 . -1500 x 2 . Note that, x , normalized to minor radius a , measures the distance from a particular mode rational surface located at the centre of the domain.
## 3 Results
Before considering the global calculations, we first start with the local ballooning analysis. Here, only Ion Temperature Gradient (ITG) modes [1, 2, 24, 25], believed to be one of the causes for turbulent transport in tokamak plasmas, have been considered. In this paper for convenience we have limited ourselves to linear electrostatic calculations with adiabatic electrons. The model equilibrium parameters that we have used are given in table 1. Fig 1 shows the real frequency, ω 0 , and growth rate, γ 0 , as a function of k ρ y i for the mode with p = 0, together with the local mode structure, ξ , in ballooning space, η . To map from the infinite domain, η , to the poloidal angle, θ , we make use of the relation ξ θ ( +2 lπ, p +2 lπ ) = ξ θ, p ( ) for any integer l . Here, k y is the binormal perpendicular wave number and ρ i is the ion Larmor radius. The most unstable mode is found at k ρ y i = 0 45. In the following, we apply the technique presented in section 2 to reconstruct . the global mode structures for this most unstable mode.
## 3.1 Global Calculations: The quadratic η i profile
To reconstruct the global mode structure, we first consider a quadratic radial temperature gradient profile, R/L T = 17 00 . -1500 x 2 , that peaks radially at the centre of the domain, i.e at x = 0, see Fig 2. We hold all other equilibrium profiles constant at those values given in table 1. In particular we have assumed that R/L n = 2 2 over the entire simulation so that the profile variation in the driving source, . η i , is determined by the R/L T profile. GS2 calculations across the radial domain, x , and full range in ballooning angle -π ≤ p ≤ π , then provide the local complex mode frequency, Ω ( 0 x, p ). Fig 3 shows the contour plot for the Ω ( 0 x, p ) numerical data, taken from GS2, and the model fit using the parametrization given in equation(5). As expected, due to the quadratic radial variation of the temperature gradient profile R/L T , the coefficients with m = 1 are all zero. To obtain A p ( ) and its eigenvalue Ω, we solve equation(7), numerically, with a
Table 1: The model parameters that have been used throughout this paper. Note that subscript 0 indicates that the parameter has been evaluated at the rational surface r = r 0 .
| Parameter | ˆ s | q 0 | R/L T | R/L n | k y ρ i | glyph[epsilon1] | a | R | β | nq ′ | ν ii × R v ti | T i T e |
|-------------|-------|-------|---------|---------|-----------|-------------------|-------|-------|-----|--------|-----------------|-----------|
| Value | 1.5 | 1.4 | 17 | 2.2 | 0.45 | 0.18 | 0.27m | 1.50m | 0 | 300 | 0.8 | 1 |
Figure 3: Contour plots of real and imaginary parts of the local complex mode frequency, measured in unit of ( v ti /R ), as a functions of radius, x , and ballooning angle, p , for quadratic R/L T radial profile, while excluding other profile variations. (a) and (b) are, respectively, the frequency and growth rate obtained from the local gyrokinetic code, GS2. The corresponding frequency and growth rate from the fitted model, using Ω ( 0 x, p ) = ∑ 10 k =0 ∑ 2 m =0 a m k cos kp ( ), are presented in (c) and (d) respectively. The ∗ symbol indicate the marginal stability contour.
periodic boundary condition, i.e A p ( -π ) = A p ( + ). Note that including Fourier harmonics above π N k ≥ 7, the global eigenvalue converges to a constant value, Ω = ω + iγ = 1 093 + 0 4609 . . . i For this reason, in what follows, we have used N k = 10.
Using this numerical solution for A p ( ), and the solution for ξ x, p, θ ( ) from GS2, equation(1) is solved to obtain the global mode structure, φ x, θ ( ). The function A p ( ) and the corresponding solution for φ x, θ ( ) in the poloidal cross-section are shown in Fig 4. We see that A p ( ) is highly localized about p = 0. This leads to a mode that balloons on the outboard mid-plane, where θ = 0, with growth rate of γ =Max[ γ 0 ( x, p )] = 0 4609. . This mode is the isolated mode identified in [3, 12, 13]. Isolated modes are only possible when both the local real frequency, ω 0 ( x, p ), and the local growth rate, γ 0 ( x, p ) are stationary at the same radial position, i.e where x = 0 for our case, see Fig 3.
The radial mode width, ∆ w r , and its variation with the toroidal mode number, n , has also been calculated and presented in Fig 5. We have taken the magnitude of the radial slice from the constructed mode structure at θ = 0, where the mode peaks, i.e φ x, ( 0) and then fitted a Gaussian function to it. The radial width, ∆ w r , has then been calculated as the full width at half maximum, FWHM, of the Gaussian. We find that the radial width of the mode scales inversely with the square root of n , ∆ w r ∼ n -0 54 . . This is expected for these isolated modes [17].
Figure 4: From left to right: (a) presents the numerical solution of the envelope function, A p ( ), obtained from equation(7), as a function of ballooning angle, p , and (b) reconstructed electrostatic potential global mode structure, φ x, θ ( ), in the poloidal plane. Note that profile variations other than L T are excluded, and the mode structure, centred on the outboard mid plane, is aligned radially where θ = 0.
Figure 5: The radial mode width, ∆ w r , as a function of toroidal mode number, n . It scales, approximately, inversely with square root of n according to: ∆ w r = 0 289 . n -0 54 . .
Figure 6: The global real frequency ω (a) and growth rate γ (b) as functions of toroidal mode number, n . The ω and γ are measured in unit of ( v ti /R ). They scale with n as follow: ω = ω 0 (0 0) , -1 35 . n -1 15 . , γ = γ 0 (0 , 0) -377100 n -0 97 . , where Ω (0 0) = 0 , ω 0 (0 , 0) + iγ 0 (0 0) = 1 1086 + , . i 0 4690 is the value of local . complex mode frequency, indicated by two horizontal lines, calculated at radial position x = 0 and ballooning angle p = 0. Note that, as n increases the value of Ω converges to Ω . 0
Changing the toroidal mode number also affects the global mode frequency, Ω. Fig 6 shows that both the real frequency, ω , and the linear growth rate, γ , scale with n according to: ω = ω 0 (0 0) , -1 35 . n -1 15 . and γ = γ 0 (0 0) , -377100 n -0 97 . , respectively. With ω 0 (0 0) = 1 1086 and , . γ 0 (0 , 0) = 0 4690, respectively, being . the values of the local frequency and growth rate calculated at x = 0 and p = 0. This indicates that the finite n correction scales inversely with the toroidal mode number as expected from conventional ballooning
Figure 7: From left to right: (a) Radially varying q profile, q = 3 923( . r/a ) ˆ s , with constant magnetic shear ˆ = 1 5, and (b) reconstructed global mode in the poloidal cross-section, s . φ x, θ ( ). Note that the mode is now tilted on the outboard mid plane, where θ = 0, due to the profile variation (c.f. Fig 4).
theory [17], i.e ∼ n -1 . In the limit of n goes to ∞ , the global growth rate is equal to the local growth rate calculated at p = 0 and x = 0, i.e γ =Max[ γ 0 ( x, p )]. This is only true for this special case where Ω ( 0 x, p ) has a stationary point at x = 0. Our results are consistent with the predictions of analytical theory of the higher order ballooning calculations presented elsewhere in the literature, e.g in references [12, 3, 24]. As we shall see in the following section, taking into account the radial variation of other equilibrium profiles, e.g safety factor q , introduces a small, but significant, deviation from conventional ballooning theory.
## 3.2 Global Calculations: Profile variations and shear flow effects
We repeat the previous analysis, but now introduce a radially varying safety factor profile, q = 3 93( . r/a ) ˆ s , with constant magnetic shear parameter ˆ = 1 5. The s . q profile varies over the length scale of the instability and gives rise to coefficients which are linear in x into equation(5). This modifies the local mode frequency, Ω ( 0 x, p ), i.e. introduces terms with m = 1 in the Taylor expansion of Ω ( 0 x, p ). Furthermore, this profile variation affects the real frequency, ω 0 ( x, p ), and the linear growth rate, γ 0 ( x, p ), differently, breaking their symmetry about x = 0. It introduces a radial shift in the position of the maximum of ω 0 ( x, p ) with respect to γ 0 ( x, p ), which affects the poloidal position of the mode peak and its stability.
Fig 7 presents both the radially varying q profile and the reconstructed global mode structure in the poloidal cross-section, φ x, θ ( ). The mode shifts slightly downward with respect to the out-board mid plane, breaking the up-down poloidal symmetry with a slight reduction, ≈ 3%, in the growth rate, compared to the flat q profile case of the previous section. As we can see, the radial q profile variation tilts the structure on the outboard mid plane. This finding agrees qualitatively with other published results for full global simulations of linear ITG modes presented in [20]. They have shown that the radial profile variation can induce toroidal momentum transport in the plasma, and this transport consequently can influence intrinsic rotation [21] which is of particular interest for a machine like ITER for which the external source of torque will be negligible.
We have also studied the influence of a small, constant flow shear on the reconstructed mode structures and their stability, by introducing a Doppler shift in Ω 0 ,
$$\Omega _ { 0 } ( x, p ) \rightarrow \Omega _ { 0 } ( x, p ) + n \Omega ^ { \prime } x = ( \omega _ { 0 } ( x, p ) + n \Omega ^ { \prime } x ) + i \gamma _ { 0 } ( x, p ),$$
where Ω ′ is constant and controls the flow shear effect. Thus Ω ′ x represents the rotation frequency of the magnetic flux surfaces measured relative to the mode rational surface at x = 0 where the reconstructed global mode sits. We have considered flow shear cases, with -4 0 . ≤ n Ω ′ ≤ +4 0. .
From equation8 it is clear that the flow shear introduces linear, m = 1, terms into the a ( m ) k coefficients of equation5. This indicates that both flow shear and the radial q -profile variation are expected to have similar effects on the global mode structures and their instabilities.
Figure 8: The linear growth rate, γ , is measured in unit of ( v ti /R ), as a function of flow shear, n Ω , calculated ′ for the most unstable mode with k ρ y i = 0 45. . The toroidal mode number, n = 50 and q ′ ≈ 6. The radial variation of the q profile is excluded in (a) and included in (b).
Figure 9: Reconstructed global mode structure in a small region of the poloidal cross-section at the outboard mid plane, under the combined effects of both radially varying q -profile and flow shear, n Ω . ′ From left to right, n Ω = ′ -4 5 . , -1 5 . , 0 and +1 5, respectively. . For n Ω = 0, the mode is already tilted, due to the profile ′ variation effect, while a critical value of flow shear, occurs at n Ω = ′ ∼ -1 5, cancels out the effect of profile . variation and, once again, the mode structure is aligned radially (as for a conventional ballooning mode). From 8, we see that this is the maximally unstable flow shear.
Depending on the sign of the flow shear the reconstructed global mode shifts upward or downward relative to the outboard mid plane with a structure in the poloidal plane that is very similar to that in Fig 7. For a mode peaking at θ = 0 in the absence flow shear, increasing flow shear tilts the mode structure on the outboard mid plane and lowers the linear growth rate. Fig 8 shows how the growth rate varies with flow shear. Excluding the q -profile variation, see Fig 8a, the growth rate curve is symmetric about Ω ′ = 0. However, the q -profile variation introduces an asymmetry to the growth rate with respect to the sign of flow shear, as shown in Fig 8b. The effect of this asymmetry on the reconstructed global mode structure is illustrated in Fig 9. For n Ω = 0 the structure is already tilted, but increasing flow shear in the negative ′ direction acts to re-align the mode radially. For a critical value of flow shear, n Ω = ′ ∼ -1 5, the effect of the . profile variation is completely compensated, allowing an isolated mode again to form with largest growth
rate, γ =Max[ γ 0 ( x, p )]. Increasing flow shear even further, below the critical value, tilts the mode structure in the negative direction and lowers its linear growth rate again. These results obtained purely from solutions of GS2, are in very good qualitative agreement with global calculations of linear electrostatic ITG modes presented in reference [19].
For more realistic cases where we take other profile variations into account we do not in general expect an isolated mode to be observed in the global calculations, except for a critical value of flow shear. This is a special case, and generally the linear growth rate of microinstabilities will be lower.
## 4 Conclusion
In this work we have reconstructed radial and poloidal mode structures for linearly unstable electrostatic ITG microinstabilities, purely from solutions of the local gyrokinetic code, GS2. Firstly, we introduce a quadratic radial profile for the mode driving source, while all other equilibrium parameters are held constant. We found that the complex mode frequency Ω ( 0 x, p ) is stationary at the mode rational surface, at x = 0, where the global mode sits. This leads to an isolated type of mode that peaks at the outboard mid plane with a large growth rate, γ ∼ Max[ γ 0 ( x, p )]. These results are in very good qualitative agreement with the simplified fluid model of ITG modes presented in [3]. Moreover, introducing a radial variation in the q profile, we have found that the radial position of the stationary points in both local frequency, ω 0 , and growth rate, γ 0 , are slightly shifted with respect to each other. In this case, the reconstructed global mode becomes less unstable and shifts slightly downward with respect to the outboard mid plane. A constant flow shear, introduced as a Doppler shift in the real frequency, influences the global mode, reducing its linear growth rate and tilting the mode structure at the out board mid plane. Taking the q -profile variation effect into account causes an asymmetry in the growth rate with respect to the sign of flow shear. These findings are qualitatively similar to those captured in global gyrokinetic calculations [19, 20].
Due to the radial variation of the equilibrium profiles, except for a critical value of flow shear, we expect the high growth rate isolated modes to be suppressed. Therefore, in general, the growth rate will be lower than the maximum value obtained from the local codes even in the n →∞ limit. This might have important implications for the quasilinear predictions of the heat and particle fluxes.
Finally, the procedure used in this work is quite general and can be used to explore more realistic tokamak equilibria. In our future work we will investigate the effect of shaping (eg elongation and triangularity) on the behaviour and stability of the linear global ITG modes.
## Acknowledgement
The main author is extremely grateful to the Ministry of Higher Education in Kurdistan region of Iraq for the opportunity and funding they provided to study for a PhD at University of York. This work has also received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement number 633053 and from the RCUK Energy Programme [grant number EP/I501045]. The views and opinions expressed herein do not necessarily reflect those of the European Commission. The simulations presented were carried out using supercomputing resources on HECToR (from the Plasma HEC Consortium EPSRC grant number EP/L000237/1). A part of this work was also carried out using the HELIOS supercomputer system at Computational Simulation Centre of International Fusion Energy Research Centre (IFERC-CSC), Aomori, Japan, under the Broader Approach collaboration between Euratom and Japan, implemented by Fusion for Energy and JAEA.
## References
- [1] J. Wesson, Tokamaks (Clarendon Press-Oxford, New York, 2004).
- [2] W. Horton, Rev. Mod. Phys. 71, 735-778 (1999)
- [3] D. Dickinson, C. M. Roach, J. M. Skipp and H. R. Wilson, Physics of Plasmas 21, 010702 (2014) .
- [4] J. W. Connor and T. J. Martin, Plasma Phys. Control. Fusion 49(2007) 1497-1507
- [5] P. Terry, Rev. Mod. Phys. 72, 109 (2000)
- [6] Y. Kishimoto and et al , Plasma Phys. Control. Fusion 41(A663) 1999
- [7] C. M. Roach and et al , Plasma Phys. Control. Fusion 51, 124090 (2009)
- [8] H. Biglari, P. H. Diamond and P. W. Terry, Phys. Fludis B 2(1) January 1990
- [9] R. E. Waltz, G. D. Kerbel and J. Milovich, Phys. Plasmas 1, 2229 (1994)
- [10] P. H. Rutherford and E. A. Frieman. Physics of Fluids, 11(8):569-585, 1968 .
- [11] E. A. Frieman and L. Chen. Physics of Fluids, 25:502-508, 1982 .
- [12] J. B. Taylor, H. R. Wilson and J. W. Connor, Plasma Phys. Control. Fusion 38, 243-250 (1996) .
- [13] R.L. Dewar, Plasma Phys. Control. Fusion 39, 453-470 (1997) .
- [14] M. Kotschenreuther, G. Rewoldt and W. M. Tang. Computer Physics Communications, 88, 1995 .
- [15] W. Dorland, F. Jenko, M. Kotschenreuther, and B.N. Rogers. Phys. Rev. Lett. 85, 5579 (2000) .
- [16] Y. Z. Zhang and S. M. Mahajan, Phys. Lett. A 157, 133 (1991) .
- [17] J. B. Taylor, J. W. Connor and R. J. Hastie. In booktitle, 365 of the proceeding of the Royal Society of London A, Mathematical and Physical Sciences, pages 1-17 ( The Royal Society, Feb. 19 1979) .
- [18] A. Bottino and et al. Phys. Plasma, 11:198-2006, 2004 .
- [19] P. Hill and et al , Plasma Phys. Control. Fusion 54, 1-8 (2012) .
- [20] Y. Camenen, Y. Idomura, S. Jolliet and A. G. Peeters, Nucl. Fusion 51 (2011) 073039 .
- [21] A.G. Peeters and et al , Nucl. Fusion 51 (2011) 094027 .
- [22] J. W. Connor, R. J. Hastie and J. B. Taylor. Phys. Rev. Lett. 40, 396 (1978) .
- [23] X. Lapillonne et al . Phys. Plasmas 16 (2009) 1 .
- [24] F. Romanelli and F. Zonca. Phys. Fluids B 5 (11), November 1993 .
- [25] A. G. Peeters and D. Strintzi. Phys. Plasmas, Volume 11, Issue 8, pp. 3748-3751 (2004) .
| null |
[
"P. A. Abdoul",
"D. Dickinson",
"C. M. Roach",
"H. R. Wilson"
] |
2014-08-04T16:48:53+00:00
|
2014-08-04T16:48:53+00:00
|
[
"physics.plasm-ph"
] |
Using the local gyrokinetic code, GS2, to investigate global ITG modes in tokamaks. (I) s-$α$ model with profile and flow shear effects
|
This paper combines results from a local gyrokinetic code with analytical
theory to reconstruct the global eigenmode structure of the linearly unstable
ion-temperature-gradient (ITG) mode with adiabatic electrons. The simulations
presented here employ the s-${\alpha}$ tokamak equilibrium model. Local
gyrokinetic calculations, using GS2 have been performed over a range of radial
surfaces, x, and for ballooning phase angle, p, in the range -${\pi} {\leq} p
{\leq\pi}$, to map out the complex local mode frequency, ${\Omega_0(x, p) =
\omega_0(x, p) + i\gamma_0(x, p)}$. Assuming a quadratic radial profile for the
drive, namely ${\eta_i = L_n/L_T}$, (holding constant all other equilibrium
profiles such as safety factor, magnetic shear etc.), ${\Omega_0(x, p)}$ has a
stationary point. The reconstructed global mode then sits on the outboard mid
plane of the tokamak plasma, and is known as a conventional or isolated mode,
with global growth rate, ${\gamma}$ ~ Max[${\gamma_0(x, p)}$], where
${\gamma_0(x, p)}$ is the local growth rate. Taking the radial variation in
other equilibrium profiles (e.g safety factor q(x)) into account, removes the
stationary point in ${\Omega_0(x, p)}$ and results in a mode that peaks
slightly away from the outboard mid-plane with a reduced global growth rate.
Finally, the influence of flow shear has also been investigated through a
Doppler shift, ${\omega_0 \rightarrow \omega_0 + n\Omega^{\prime}x}$, where n
is the toroidal mode number and ${\Omega^{\prime}}$ incorporates the effect of
flow shear. The equilibrium profile variation introduces an asymmetry to the
growth rate spectrum with respect to the sign of ${\Omega^{\prime}}$,
consistent with recent global gyrokinetic calculations.
|
1408.0743v1
|
## HIGH-SCHOOL ALGEBRA OF THE THEORY OF DICRITICAL DIVISORS: ATYPICAL FIBRES FOR SPECIAL PENCILS AND POLYNOMIALS
E. ARTAL BARTOLO, I. LUENGO, AND A. MELLE-HERNÁNDEZ
In this work we got a revival of our discussions about dicriticals with Ram. Dedicated to the memory of S.S. Abhyankar.
Abstract. In this work we deal with dicritical divisors, curvettes and polynomials. These objects have been one of the main research interests of S.S. Abhyankar during his last years. In this work we provide some elementary proofs of some S.S. Abhyankar and I. Luengo results for dicriticals in the framework of formal power series. Based on these ideas we give a constructive way to find the atypical fibres of a special pencil and give bounds for its number, which are sharper than the existing ones. Finally, we answer a question of J. Gwoździewicz finding polynomials that reach his bound.
## Introduction
The study of the topology and geometry of polynomial maps is of great interest in Affine Algebraic Geometry, for instance for the cancellation problem or affine exotic spaces. The Jacobian problem is one of the main open problems in this area. Recently the local theory of algebraic dicritical divisors and curvettes has been developed ([13, 8, 9]) and applied to get some control on the fibers of a Jacobian pair. Dicritical divisors have been studied by S.S. Abhyankar either alone, [2, 4, 3, 5, 6, 7], or with co-authors, [8, 9, 11, 10, 12, 13, 14]. He has developed an algebraic theory which starts from the geometric intuition coming from analytic geometry and extends the result to the more general setting: starting from C { x, y } he developed (with his collaborators) a general theory valid for general regular local rings.
In this work we want to apply this theory to the study of special pencils, i.e., elements of the quotient field of a regular ring whose denominator is a power of a regular element of the ring. The fundamental reason to study these pencils is that they appear naturally when working with polynomial maps at infinity. Moreover, the strategy to study these pencils is through the resolution of the base points of the pencil where dicriticals appear in a natural way. With their algebraic techniques, several results about dicriticals are proved in [13, 14]: the restriction of the pull-back of the pencil to each dicritical is a polynomial, dicriticals are in one-to-one correspondence with the irreducible factors of the pencil, see §3 for details.
The core of the paper is to provide elementary algebraic proofs, valid also in positive characteristic, for rings of power series over a field by high-school algebra methods following the mathematical
Date : August 5, 2014.
2010 Mathematics Subject Classification. 14A05, 14R15.
Key words and phrases. dicritical divisor, special pencil.
philosophy of S.S Abhyankar. In order to achieve the proof, we proceed with a variation of the Newton-Puiseux process realized by birational transformations, see also [18] for similar approaches. Using Newton polygon techniques we describe a finite recursive argument which presents in an explicit case a toric resolution of the pencil which is combinatorially much less complex than the resolution via standard blow-ups or quadratic transformations. With this method, the dicritical divisors are in bijection with some edges of a sequence of Newton polygons, from which we keep two important data: a 1 -variable polynomial coming from the edge and a positive integer which is related to a quotient singularity coming from a toric blowing-up.
We will apply these techniques in order to improve some bounds for the number of atypical values of special pencils given by J. Gwoźdiewicz in [22].
Theorem 1.1. ([22]) Let f ( x, y ) , l ( x, y ) ∈ C { x, y } , f (0 , 0) = l (0 , 0) = 0 , be convergent power series without common factor. Assume that the curve l ( x, y ) = 0 is smooth and that the curve f ( x, y ) = 0 has d components counted without multiplicities. Then, the pencil f ( x, y ) -tl ( x, y ) M = 0 , where M is a positive integer, has at most d nonzero atypical values.
Our main result provides a more accurately defined bound for the number of atypical values for a special pencil which is given by the sum of the number of dicriticals plus the number of non-zero roots of the derivatives of the polynomials associated to the dicriticals, see Theorem 2.11. Moreover, this result is true for formal power series over algebraically closed fields without restrictions on the characteristic (except a mild separability hypothesis), following Abhyankar's style. Example 2.14 shows that our bound is sharp.
This local bound is also extended to the polynomial setting, see also [23]. Since at each base point at infinity the polynomial defines a local special pencil then the number of atypical values at infinity is bounded by the sum of the corresponding local bounds we got in Theorem 2.11. Therefore, as a consequence, an algebraic proof of the next Theorem is given.
Theorem 1.2. ([22]) Assume that the complex algebraic curve f ( x, y ) = 0 has n branches at infinity. Then the polynomial f has at most n critical values at infinity different from 0 .
We also provide examples showing that our bound is also sharper than the one of [22, Theorem 1.2]. Notice that Gwoźdiewicz's result is in the same spirit as the following Moh's Theorem [26] as quoted by Ephraim's version [21].
Theorem 2.2. ([21]) Assume that the complex algebraic curve f ( x, y ) = 0 has only one branch at infinity. Then f has no critical values at infinity. In particular, all curves f ( x, y ) = t for t ∈ C are equisingular at infinity.
As T.T. Moh pointed out in [26], S.S. Abhyankar gave another proof of this result by applying [15, (3.4)].
The number of branches at infinity is related with the Jacobian problem:
if f , f 1 2 ∈ K [ x, y ] , char( K ) = 0 , is a Jacobian pair, i.e. its Jacobian determinant is equal to 1 , then K [ f , f 1 2 ] = K [ x, y ] .
T.T. Moh remarks in [26] that the following Engel's statement was a main tool in W. Engel's attempted proof of the Jacobian conjecture, see [20]:
For a special member of the pencil f ( x, y ) + c = 0 , the number of branches at infinity cannot be greater than the corresponding number for the general member.
In 1971 S.S. Abyhankar found a counterexample to Engel's statement.
Abhyankar and Moh, see e.g [1] for details, translated the Jacobian condition into conditions on the resulting special expansions getting the following result:
The Two Point Theorem. ([1]) If f 1 and f 2 is a Jacobian pair, then f 1 and f 2 have at most two points at infinity. Moreover, it can be deduced that if the Jacobian condition implies that f 1 and f 2 have at most one point at infinity then the Jacobian problem has an affirmative answer.
In fact if f 1 and f 2 ∈ K [ x, y ] is a Jacobian pair with two points at infinity it follows from H. Żołądek in [29] that f 1 and f 2 have some common dicriticals. In fact, not all the dicritical components can be in common because in such a case the degree of the polynomial map from C 2 to C 2 vanishes, hence the Jacobian is identically zero (private communication to the authors of Pierrette Cassou-Noguès).
As we explain in §4, the conditions to reach this number of branches at infinity are quite involved (in particular Moh-Ephraim result shows that it is not possible when there is only one branch at infinity). The last part of §4 is devoted to construct two examples. Example 4.2 is the polynomial version of Example 2.14. Example 4.1 answers positively the following question proposed by J. Gwoźdiewicz [22].
Question. Does there exist a polynomial f ( x, y ) with n nonzero critical values at infinity such that the curve f ( x, y ) = 0 has n branches at infinity?
Example 4.1 is a polynomial where the generic fiber has two branches at infinity. Following a referee's comment we provide in Example 4.3 a way to construct such examples with an arbitrary number of branches at infinity for the generic fiber.
## 1. Toric-Newton transforms of special meromorphic functions
For convenience we work over an algebraically closed field K . Nevertheless, the results are valid over any field since it is well known that one can get the resolution of the base points of a pencil over a finite extension of the base field K . Let R = K [[ x, y ]] be the formal power series ring over K ; note that most of the results are also valid for convergent power series in case of complex numbers and some of them will also be valid for more general (almost complete) two-dimensional local rings (without restriction on the characteristic and even in mixed characteristic) especially if they have analytical properties, see [9]. Following Abhyankar we will study regular local rings contained in L (the fraction field of R ) and dominating R though we will replace these rings by their completion for simplicity. We will denote M R ( ) the maximal ideal of R .
A formal power series p x, y ( ) ∈ R can be evaluated at the only closed point 0 ∈ Spec R , giving an element p (0 , 0) ∈ K . For an element r x, y ( ) := p x,y ( ) q x,y ( ) the evaluation at 0 can be defined on P 1 K = K ∪ {∞} , with one important exception. If p, q ∈ M R ( ) are coprime, then r (0) is not defined, it is undetermined . It is also useful to treat r as the pencil of curves { C t : p = tq } , for t ∈ K ∪ {∞} having 0 as base point.
It is well-known that one can eliminate this indetermination via a birational map π : S → Spec( R ) , which is the composition of a sequence of closed points blow-ups, also called quadratic transformations,
such that π ∗ ( r ) : S → P 1 K is a well defined morphism. This means that from the point of view of pencils, the strict transforms of the curves C t are disjoint.
Let E = π -1 (0) be the exceptional divisor of the map π , with irreducible components E , . . . , E 1 s . A divisor E i ⊂ E is called dicritical (or some authors called them horizontal ) if π ∗ ( r ) | E i is not a constant map, that is π ∗ ( r )( E i ) = P 1 K .
Let us define
$$P ( x, y, T ) \coloneqq p ( x, y ) - T q ( x, y ) & = \sum _ { i, j } A _ { i, j } x ^ { i } y ^ { j } \in \mathbb { K } ( T ) [ [ x, y ] ] \ ( T \text{ an indeterminate} ). \\ & \quad \text{-}$$
We have two main interests: To study the curve ˜ C given by P ∈ K ( T )[[ x, y ]] and to study the curves C t = { P x, y, t ( ) = 0 } for t ∈ K , both generic and atypical.
Definition 1.1 . The Newton polygon NP( ) r of r is the Newton polygon of P ∈ K ( T )[[ x, y ]] i.e. the compact faces of the convex closure of NR( P ) := Supp( P ) + N 2 ⊂ N 2 ⊂ R 2 .
We are interested in giving several algebraic characterizations of dicritical divisors in a particular class of pencils, specially important for polynomial maps.
Definition 1.2 . A meromorphic germ r ∈ L (or its corresponding pencil) is called special if r x, y ( ) = p x,y ( ) x U x,y c ( ) for some local parameters x, y ∈ R c > , 0 and a unit U x, y ( ) ∈ K [[ x, y ]] (we always assume that x does not divide p x, y ( ) ).
Remark 1.3 . Since x does not divide p , the y -order d of p x, y ( ) is well-defined, i.e. the unique positive integer such that p (0 , y ) is a series of order d .
## Example 1.4 . The pencil
$$p _ { x } ( x, z, T ) = ( x ^ { 3 } - z ^ { 5 } ) ^ { 2 } - x ^ { 6 } + x ( x ^ { 5 } - z ^ { 2 } ) ^ { 5 } + 5 x z ^ { 7 } \left ( x - \frac { 3 } { 4 } z ^ { 2 } \right ) - T z ^ { 1 1 } \\ \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \end{array}$$
is special in K ( T )[[ z, x ]] .
These pencils are called Ephraim pencils in [22], based on [21]. It was shown in [13, Theorem A] that for a special pencil r , the restriction of the pull-back π ∗ ( r ) to any dicritical divisor is a polynomial, for arbitrary two dimensional local regular rings, not necessarily equicharacteristic. In this paper an elementary proof of this result for R = K [[ x, y ]] is given; the tools used in the proof detect the so-called atypical fibers of the pencil which are also studied in this work.
From now on we assume that r ∈ L is special. We are going to give a recursive method to solve a special pencil r by means of toric transformations and translations associated to NP( ) r .
We introduce some notation. Fix an edge /lscript of NP( ) r which is contained in the line nx + my = ω ( m,n ∈ N coprime). We denote by ω /lscript the weight ω /lscript ( i, j ) := ni + mj . This edge supports a ω /lscript -quasihomogeneous polynomial of degree ω
$$P _ { \ell } ( x, y ) = \sum _ { \omega _ { \ell } ( i, j ) = \omega } A _ { i, j } x ^ { i } y ^ { j } = x ^ { u } y ^ { v } q _ { \ell } ( x ^ { m }, y ^ { n } ),$$
where q /lscript ( s , s 1 2 ) ∈ K [ T ][ s , s 1 2 ] is a homogeneous polynomial of degree d /lscript with at least two monomials and coprime with s s 1 2 . Note that
$$P ( x, y, T ) = P _ { \ell } ( x, y ) + \text{ monomials with } \omega _ { \ell } \text{-degree } > \omega.$$
The coefficients of q /lscript ( s , s 1 2 ) are in K with only one eventual exception: if v = 0 and u = c , i.e., the vertex ( c, 0) is in /lscript ⊂ NP( ) r . Bezout identity allows to choose
$$( 1. 2 ) & & a, b \in \mathbb { Z } _ { > 0 } \text{ such that } b n - a m = 1.$$
Notation 1.5 . The coprime weights ( n, m ) will be denoted if necessary as ( n , m /lscript /lscript ) ; we will refer to n as the v-ratio and m as the h-ratio of the edge /lscript .
The following concept appears also in [18].
Definition 1.6 . An edge /lscript of NP( ) r is called a dicritical edge if ( c, 0) is a vertex of /lscript
Remark 1.7 . We assume that if ( n, m ) = (1 , m ) then P /lscript is not proportional to ( y -Ax m e ) , A ∈ K . If it is the case, the change of variables y = y 1 + Ax m makes the edge /lscript disappear. The polygon NP( ) r has at most one dicritical edge.
Example 1.8 . Let us consider p x as in Example 1.4. Its Newton polygon is in Figure 1. There is only one edge /lscript and P /lscript = ( x 3 -z 5 ) 2 ( x plays the role of y , we keep these variables for further use in §4). The edge is not dicritical.
T
Figure 1. Newton polygon of p x
Proposition 1.9. Assume that /lscript is not a dicritical edge. The monomial transformation
$$\varphi _ { M } ( x _ { 1 }, y _ { 1 } ) \coloneqq ( x _ { 1 } ^ { n } y _ { 1 } ^ { a }, x _ { 1 } ^ { m } y _ { 1 } ^ { b } ), \ \ s e e \ ( 1. 2 ),$$
is birational (i.e. it is a composition of quadratic transformations) and the polynomial P /lscript is transformed as
$$P _ { \ell } ( x _ { 1 } ^ { n } y _ { 1 } ^ { a }, x _ { 1 } ^ { m } y _ { 1 } ^ { b } ) = \beta x _ { 1 } ^ { \omega } y _ { 1 } ^ { a u + b v + a m d _ { \ell } } q _ { \ell } ( 1, y _ { 1 } ).$$
Proof. Note that
$$P _ { \ell } ( x _ { 1 } ^ { n } y _ { 1 } ^ { a }, x _ { 1 } ^ { m } y _ { 1 } ^ { b } ) = \beta x _ { 1 } ^ { n u + m v } y _ { 1 } ^ { a u + b v } q _ { \ell } ( x _ { 1 } ^ { m n } y _ { 1 } ^ { a m }, x _ { 1 } ^ { m n } y _ { 1 } ^ { b n } ) = \beta x _ { 1 } ^ { \omega } y _ { 1 } ^ { a u + b v + a m d _ { \ell } } q _ { \ell } ( 1, y _ { 1 } ).$$
We use that ω = nu + mv + mnd /lscript , bn -am = 1 , and the fact that q /lscript is homogeneous of degree d /lscript . /square
Figure 2. L M for the edge BC .
This means that the image of the NP( ) r by the affinity
↦
$$L _ { M } \colon \mathbb { Z } ^ { 2 } \to \mathbb { Z } ^ { 2 }, \begin{pmatrix} u \\ v \end{pmatrix} \mapsto \begin{pmatrix} n & m \\ a & b \end{pmatrix} \begin{pmatrix} u \\ v \end{pmatrix}$$
has a vertical edge and Supp( ϕ ∗ M P ) , where ϕ ∗ M P = P x y , x ( n 1 a 1 m b 1 y , T 1 ) ∈ K ( T )[[ x , y 1 1 ]] , is contained in L M (NR( P )) , see Figure 2. Let us factor
$$q _ { \ell } ( s _ { 1 }, s _ { 2 } ) & = \beta \prod _ { j = 1 } ^ { e } ( s _ { 2 } - \alpha _ { j } s _ { 1 } ) ^ { m _ { j } }, \quad \beta, \alpha _ { j } \in \mathbb { K } \ \{ 0 \}, \quad m _ { j } > 0, \text{ i.e., } d _ { \ell } = \sum _ { j = 1 } ^ { e } m _ { j }. \\ \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ { j = 1 } ^ { e } \prod _ }$$
Definition 1.10 . For /lscript a non-dicritical edge and α j a root of q /lscript (1 , s ) , the toric-Newton transformation associated to ( /lscript, α j ) is the toric transformation ϕ M followed by the translation y 1 = ¯ y 1 + α j .
Definition 1.11 . The strict transform P /lscript,α j ( x , y 1 ¯ 1 , T ) of P by the toric-Newton transformation associated to ( /lscript, α j ) is
$$P _ { \ell, \alpha _ { j } } ( x _ { 1 }, \bar { y } _ { 1 }, T ) = \frac { P ( x _ { 1 } ^ { n } ( \bar { y } _ { 1 } + \alpha _ { j } ) ^ { a }, x _ { 1 } ^ { m } ( \bar { y } _ { 1 } + \alpha _ { j } ) ^ { b }, T ) } { x _ { 1 } ^ { \omega } ( \bar { y } _ { 1 } + \alpha _ { j } ) ^ { \tilde { v } } }. \\. \. \. \. \. \. \. \. \. \. \. \. \. \. \. \. \.$$
where ˜ v ≤ au + bv + amd is the minimum of the powers of y 1 which appear from the pull-back by ϕ M .
Example 1.12 . Let us study the strict transform for the toric-Newton transformation of Example 1.4. The Newton polygon of this strict transform is shown in Figure 3a; the quasihomogeneous polynomial is ( x 1 -5 2 z 1 ) 2 and we are in the situation of Remark 1.7. We perform the translation and we obtain a special pencil whose Newton polygon, in Figure 3b has only one edge and it is dicritical since the quasihomogeneous polynomial is x 2 2 -( T + 5 8 ) z 3 .
Proposition 1.13. The strict transform P /lscript,α j ( x , y 1 ¯ 1 , T ) is a special pencil in K [[ x , y 1 ¯ ]] 1 such that its ¯ y 1 -order is m j .
Proof. The part of the strict transform corresponding to P /lscript is
/negationslash
$$\beta \bar { y } _ { 1 } ^ { m _ { j } } \prod _ { k \neq j } ( \bar { y } _ { 1 } + \alpha _ { j } - \alpha _ { k } ) ^ { m _ { k } }.$$
The rest of the strict transform is divided by x 1 . The monomial Tx U x,y c ( ) is transformed into
$$T x ^ { n c - \omega } ( \alpha _ { j } + \bar { y } _ { 1 } ) ^ { a c - ( a u + b v + a m d ) } U ( x _ { 1 } ^ { n } ( \bar { y } _ { 1 } + \alpha _ { j } ) ^ { a }, x _ { 1 } ^ { m } ( \bar { y } _ { 1 } + \alpha _ { j } ) ^ { b } )$$
and the result follows.
$$a m d ) U ( x _ { 1 } ^ { n } ( \bar { y } _ { 1 } + \alpha _ { j } ) ^ { a }, x _ { 1 } ^ { m } ( \bar { y } _ { 1 } + \alpha _ { j } ) ^ { b } )$$
We will study later what to do if /lscript is a dicritical edge. Because of Proposition 1.13, this process can be also applied to the strict transforms of P by the toric-Newton transformations.
Definition 1.14 . The toric-Newton process of P is the sequence of special pencils obtained by applying toric-Newton transformations recursively. The tree of Newton polygons of P is the family of all Newton polygons in the toric-Newton process. An edge of such a Newton polygon is called a dicritical edge if it is at the bottom of the polygon and the coefficient for ( ∗ , 0) depends on T .
## Proposition 1.15. The toric-Newton process is finite.
/negationslash
Proof. Note that the y -order of the special pencils decreases unless we are in the situation of Remark 1.7. Since the pencils are special only a finite number of translations may arise until we reach the T -monomial. Note that while the term Tx c is not present in NP r ( ) one is following the resolution (of one branch) of the fibre p x, y ( ) = 0 . This means that after a finite number of toric maps and translations we arrive to a point Q where the branch is non-singular and eventually non-reduced. Then the local equation of the total transform of P is h k ( x , y 1 1 ) u x , y ( 1 1 ) + Tx e 1 1 with u (0 , 0) = 0 and h x , y ( 1 1 ) = ( y 1 + . . . ) . It is now clear we can make a change of coordinates y 1 = y + a x 1 1 such that h x , y ( 1 )) = y + a e +1 x e +1 /square
Remark 1.16 . Note that this is the case for the pencil in Example 1.4.
## 2. Dicritical edges
Let us study now what happens with dicritical edges. We start with a simple proof of [13, Theorem A] when the regular local ring is a formal power series ring.
Proposition 2.1. Let P x, y, T ( ) := p x, y ( ) -Tx U x,y c ( ) be a special pencil, then at each dicritical divisor E the function π ∗ ( r x, y ( )) | E is a polynomial.
Proof. The previous process allows to resolve the base points of the pencil by toric maps and translations and moreover pencils arising at the process are still special. Let us study what happens at a dicritical edge /lscript . We keep the notation of (1.1) and we get that
$$q _ { \ell } ( 1, s ) = a _ { 0 } s ^ { d _ { \ell } } + a _ { 1 } s ^ { d _ { \ell } - 1 } + \dots + a _ { d _ { \ell } - 1 } s - ( T - a _ { d _ { \ell } } )$$
where a j ∈ K . We denote again π x , y ( 1 1 ) = ( x y , x n 1 a 1 m b 1 y 1 ) the toric transformation associated to /lscript . Then
$$P _ { \ell } ( x _ { 1 } ^ { n } y _ { 1 } ^ { a }, x _ { 1 } ^ { m } y _ { 1 } ^ { b } ) = x _ { 1 } ^ { \omega } y _ { 1 } ^ { \tilde { v } } \left ( q _ { \ell } ( 1, y _ { 1 } ) + x _ { 1 } G ( x _ { 1 }, y _ { 1 } ) \right ),$$
and x 1 = 0 is the equation of E and G x , y ( 1 1 ) is some series. Notice that
$$\frac { \pi ^ { * } ( p ) } { \pi ^ { * } ( x ^ { c } U ( x, y ) ) } = \frac { x _ { 1 } ^ { \omega } y _ { 1 } ^ { \tilde { v } } \left ( q _ { \ell } ( 1, y _ { 1 } ) + x _ { 1 } G ( x _ { 1 }, y _ { 1 } ) \right ) } { x _ { 1 } ^ { \omega } y _ { 1 } ^ { \tilde { v } } ( U ( 0, 0 ) + x _ { 1 } H ( x _ { 1 }, y _ { 1 } ) ) } = q _ { \ell } ( 1, y _ { 1 } ) + x _ { 1 } G ( x _ { 1 }, y _ { 1 } ),$$
/negationslash where U (0 , 0) = 0 and H x ,y ( 1 1 ) is some series. Restricting to x 1 = 0 we obtain the desired result.
The computations above also prove that the corresponding polynomial map q E : E → P 1 , where q E ( z ) := q /lscript (1 , z ) -T , has degree d E := d /lscript . /square
It is not hard to check that the dicritical divisors of r are in one to one correspondence with the dicritical edges of NP( ) r and its transforms. We study now the toric-Newton transformations for dicritical edges. Note that the toric part behaves as in the non-dicritical case, as shown in the proof of Proposition 2.1, but the translation part depends on the particular values of t . Moreover, separability properties of the polynomial q E ( z ) have a strong influence on the behavior of the fibers of the pencil near the dicritical E .
Proposition 2.2. Let P x, y, T ( ) be a special pencil as above and let E be a dicritical divisor of r associated to a dicritical edge /lscript of the toric-Newton process of P . Assume that q E ( z ) is a separable polynomial, i.e its derivative is not identically zero.
Let A ∗ E := { q E ( α ) | q ′ E ( α ) = 0 } and let t 0 ,E := q E (0) . Then, the strict transform of the germ of the curve p x, y ( ) -tx U x, y c ( ) contains exactly d E non-singular transversal curvettes meeting at d E distinct points of E , in the following cases:
/negationslash
- (1) If t / A ∈ ∗ E and t = t 0 ,E .
- (2) If t = t 0 ,E , t / A ∈ ∗ E and n = 1 .
Proof. We start with the first case. Since t / A ∈ ∗ E and the polynomial q E ( z ) is separable, we have that gcd( q E ( z ) -t, q ′ E ( z )) = 1 and all the roots of q E ( z ) -t are simple roots, i.e.:
/negationslash
/negationslash
$$q _ { E } ( z ) - t = \prod _ { i = 1 } ^ { d _ { E } } ( z - \alpha _ { i } ), \quad \alpha _ { i } \neq \alpha _ { j }, \text{ if } i \neq j.$$
Hence, the quasi-homogeneous polynomial associated to the edge /lscript for the suitable strict transform of P x, y, t ( ) = 0 is
$$\prod _ { \cdot \cdot } ^ { d _ { E } } ( y _ { 1 } ^ { n } - \alpha _ { i } x _ { 1 } ^ { m } ).$$
/negationslash
$$\prod _ { i = 1 } ^ { a _ { E } } ( y _ { 1 } ^ { n } - \alpha _ { i } x _ { 1 } ^ { m } ). \\ \text{ove factors look simil}.$$
Since α i = 0 and since t = t 0 ,E , all the above factors look similar. Hence if we consider the (non trivial) translation y 1 = ¯ y 1 + α i
/negationslash
/negationslash
$$q _ { E } ( \bar { y } _ { 1 } + \alpha _ { i } ) - t = b _ { 0 } \bar { y } _ { 1 } ^ { d _ { E } } + b _ { 1 } \bar { y } _ { 1 } ^ { d _ { E } - 1 } \dots + b _ { d _ { E } - 1 } \bar { y } _ { 1 }, \ \ b _ { d _ { E } - 1 } \neq 0.$$
If we compose the toric map of the proof of Proposition 2.1 with the above translation, we obtain then, up to terms of higher degree, that the strict transform is written as
$$b _ { 0 } \bar { y } _ { 1 } ^ { d _ { E } } + b _ { 1 } \bar { y } _ { 1 } ^ { d _ { E } - 1 } \dots + b _ { d _ { E } - 1 } \bar { y } _ { 1 } + x _ { 1 } ( \dots )$$
and one gets d E non-singular curves intersecting transversally the dicritical divisor E : { x 1 = 0 } at different points.
If t = t 0 ,E is not a root of q ′ E ( z ) and n = 1 , though the Newton polygon is changing, the factor corresponding to α i = 0 is again a curvette. /square
Remark 2.3 . With this method, along the exceptional dicritical divisor there will be no base points of the pull-back of the pencil. By this process we get a log-canonical resolution (with quotient singularities) the base points of the pencil. Since at each step we perform toric quadratic transformations we must be careful with the behavior when no translation is needed.
/negationslash
From now on we assume that the map q E ( z ) is separable, i.e. either char( K ) = 0 or char( K ) = p and q ′ E ( z ) = 0 .
Definition 2.4 . A value t ∈ K is called a typical value for P x, y, T ( ) at E if the strict transform of the curve P t, x, y ( ) has exactly d E non-singular branches (curvettes) intersecting E and is called an atypical value for P x, y, T ( ) at E otherwise.
If t ∈ K is a typical value for P x, y, T ( ) at all dicritical divisors E then t ∈ K will be called a typical value for P x, y, T ( ) , and an atypical one otherwise.
Example 2.5 . In Figure 3b, we have the Newton polygon of the unique dicritical edge for p x in Example 1.4. If we fix t = t 0 ,E = -5 8 , the vertex (0 , 3) disappears. The corresponding Newton polygon is in Figure 4. Since the general fiber is an ordinary cusp and for t 0 ,E we have a tacnode, we conclude that
2
Figure 4. Final Newton polygon for the special fiber
this value is atypical at E .
Remark 2.6 . In char( K ) = 0 this definition is equivalent to the standard definition, see, for instance, the first definition in [24, Section 3]. Note that the cases ( ) i and ( iii ) in that definition are not possible for special pencils: ( ) i in this case is only valid for t 0 = q E (0) and ( iii ) is not possible because the first time ones gets a dicritical divisor the linear system has no base points.
We are going to prove a sort of reciprocal of Proposition 2.2.
Theorem 2.7. Let P x, y, T ( ) be a special pencil as in Proposition 2.2 . Then
- (1) If t ∈ A ∗ E then t is an atypical value for P x, y, T ( ) at E .
- (2) If n > 1 ( n the v-ratio) then t 0 ,E is atypical at E regardless the value of q ′ /lscript ( t 0 ,E ) .
Remark 2.8 . From the interpretation of dicriticals of Lê-Weber, the case n > 1 corresponds exactly with the dicriticals which admit a bamboo, see [24], which will be called dicriticals with bamboo .
/negationslash
Proof. For the proof of (1), we follow the ideas in Proposition 2.2. Let α i be a multiple root of q E ( s ) -t . In (2.2), the condition b d E -1 = 0 fails and the corresponding point cannot be a curvette.
For (2), the Newton polygon of P x, y, t ( 0 ,E ) has a bottom edge which is non parallel to /lscript and of height n > 1 , so there are some branches of this curve which do not meet E , see Figure 5 for a typical behavior of Newton polygons. /square
y
y
T
x
Figure 5. Left-hand-side polygon for generic T , right-hand-side for t 0 ,E .
Example 2.9 . Let us describe some examples.
- (1) Consider the special pencil P x, y, T ( ) = y 4 + y x 2 3 + yx 7 + x 12 -Tx 6 , see NP( P ) in Figure 6a. The edge /lscript = [(0 , 4) (6 , , 0)] is a dicritical edge such that P /lscript ( x, y ) = y 4 + y x 2 3 -Tx 6 , q E ( z ) = z 2 + z -T and q E ( z ) is separable. Since the v-ratio n equals 2 > 1 , t = 0 is an atypical value, see its Newton polygon in Figure 6b. On the other side -1 2 is the only root of q ′ /lscript and then t = -1 4 is the other atypical value at E , see the Newton polygon after the toric-Newton transformation in Figure 6c. In this case a generic fibre has two branches at E while there are 3 branches for t = 0 and only one branch for t = -1 4 .
- (2) For the special pencil P x, y, T ( ) = y 3 + y x 2 -x 4 -Tx 3 the edge /lscript = [(0 , 3) (3 , , 0)] is dicritical and q E ( z ) = z 3 + z 2 -T , see NP( P ) in Figure 7a. The derivative has two roots 0 , -2 3 , and then 0 , 4 27 are the atypical values. Since the v-ratio is 1 , t = 0 is atypical only for being a critical
value of q E , see its Newton polygon in Figure 7b. In order to study the fiber for t = 4 27 , we can check that the quasi-homogeneous polynomial has one simple root and one double root. It is enough to study what happens on the double root; instead of the toric-Newton transformation we can do the change y = y 1 -2 3 x , and we obtain again the Newton polygon of Figure 7b. All the typical fibres have 3 branches while the atypical ones have 2 branches.
y
y
- (3) For the special pencil P x, y, T ( ) = y 3 + yx 2 -x 4 -Tx 3 , the value t = 0 is typical at the unique dicritical, even if the Newton polygons do not coincide, see Figure 8.
Remark 2.10 . Note that Proposition 2.2 and Theorem 2.7 gives a complete characterization of atypical values of a special pencil in terms of the polynomials q E ( z ) if they are separable. In the inseparable case the atypical values cannot be computed just from q E as the following examples, in char( K ) = p , show:
y
y
- (a) y p + x p +1 -Tx p , t = 0 behaves as the other values of K .
- (b) y p + y x 2 p -1 + Tx p , t = 0 does not behave as the other values of K .
In both cases the generic members of the pencil have the singularity type of y p + x p +1 . In particular it is not a curvette, as curvettes are smooth, and following our definition all values would be atypical. A natural extension of our definition to the non-separable case would imply that t = 0 is typical for
- (a) and atypical for (b). See [25] for a more complete description of pencils in positive characteristic.
In the separable case, we can recover algebraically the results of [22]. More precisely it is possible to recover the number of atypical fibers only in terms of the Newton polygons. The type of the atypical fibers needs the part behind the Newton polygons, but for the number, these Newton polygons are enough, compare with Remark 2.10.
We would like to estimate the number of atypical values at a dicritical. Let us collect the relevant information from the Newton process. We have E , . . . , E 1 r dicriticals coming from dicritical edges /lscript 1 , . . . , /lscript r , each one carries a polynomial q i ( z ) := q E i ( z ) of degree d i and from the weight ω /lscript i we keep the number n i . The separability hypothesis asserts that q i ( z ) is separable.
Theorem 2.11. Let P x, y, T ( ) be a special pencil satisfying the separability hypothesis. Let E be a dicritical and let n be its v-ratio. Let A E be the set { q E ( α ) | q ′ E ( α ) = 0 } . Then, the set of atypical values for P x, y, T ( ) at E is
$$\begin{cases} A _ { E } \cup \{ q _ { E } ( 0 ) \} & \text{if $n>1$} \\ A _ { E } & \text{if $n=1$} \\ \text{if $\alpha1$ was $focal $focal $D(\nu$ $s.T\nmid\alpha1$} \ E$ is out} \end{cases}$$
In particular, the number of atypical values for P x, y, T ( ) at E is at most
$$M _ { E } \coloneqq \# \{ n o n \text{-zero roots of} \, q _ { E } ^ { \prime } \} + 1 \,,$$
and the number of atypical values for P x, y, T ( ) is at most ∑ E dicritical M E .
Proof. This is a direct consequence of Theorem 2.7.
The following result is an easy consequnece of Theorem 2.11.
Corollary 2.12 (Gwoździewicz [22]) . Let P x, y, T ( ) be a special pencil.
- (1) If E is a dicritical divisor of degree d E , then there are at most d E atypical values at E .
/square
- (2) If there is a value t 0 such that C red t 0 has r branches at a dicritical divisor E then there are at most r atypical values at E (besides eventually t 0 ).
- (3) The number of atypical values of the pencil is bounded by min( ν gen , ν min + 1) , where ν gen is the number of branches of the generic value and ν min is the minimal number of branches of the fibers.
/negationslash
Remark 2.13 . In order to reach the bound ν gen , the following conditions must happen. For every dicritical E , one has n > 1 , q ′ E ( t 0 ,E ) = 0 , q ′ E has simple roots, and these roots have distinct values by q E . Moreover, the sets of atypical values for each dicritical are pairwise disjoint.
Example 2.14 . Let us consider the special pencil
$$P ( x, y, T ) = y ^ { 4 } - 2 x ^ { 2 } y ^ { 2 } + ( y ^ { 2 } - x ^ { 2 } ) y x ^ { 2 } + x ^ { 7 } - T x ^ { 4 }$$
which, for all t ∈ K has 4 branches; the bound proposed in Corollary 2.12, see [22], for the number of atypical values is at most 4. Let us compute the bound of Theorem 2.11. The unique edge /lscript of the Newton polygon is dicritical and for its dicritical E we have q E ( z ) = z 4 -2 z 2 -T . The roots of q ′ E ( z ) are α = 0 1 , , -1 , hence the bound equals 3 . Since q E (0) = 0 and q E (1) = q E ( -1) = 1 , there are exactly two atypical values, t = 0 1 , . Figure 9a shows NP( P x, y, T ( )) , while Figure 9b shows NP( ( p x, y )) . Note that Figure 9b shows also NP( ( p x, y ± x ) -1) .
y
y
Figure 9
## 3. Factors of a special pencil over K ( T )
Let us interpret a result of [14] in this language, always in the special case of power series, namely that the dicritical divisors of r are in one-to-one correspondence with the factors of P x, y, T ( ) in K ( T )[[ x, y ]] . Fix a dicritical edge and keep the notations of Proposition 2.1.
Proposition 3.1. Let /lscript be a dicritical edge of the NP( ) r corresponding to a dicritical divisor E . Then there exists an irreducible factor Q x,y,T /lscript ( ) ∈ K ( T )[[ x ]][ y ] ⊂ K ( T )[[ x, y ]] of an element P x, y, T ( ) such that its weighted initial form for ω /lscript equals q /lscript ( x m /lscript , y n /lscript ) .
Proof. Note first that using Weierstraß Preparation Theorem, P x, y, T ( ) can be decomposed as a product of a unit and a Weierstraß polynomial in y (recall that P is y -regular of order d ). We apply the version of Hensel's Lemma in A to this Weierstraß polynomial and the result follows. /square
Remark 3.2 . Instead of using Hensel's Lemma one can follow the ideas in [17, Section 2]
↦
If the dicritical edge /lscript is in another special pencil r 1 of the toric-Newton process with coordinates ( x , y 1 ¯ ) 1 , then Proposition 3.1 allow us to construct an irreducible factor ˜ Q x ,y /lscript ( 1 ¯ ) 1 of r 1 in K ( T )[[ x 1 ]][ y 1 ] ; this factor is ¯ y 1 -regular of order d /lscript . Let us see the effect of the inverse of the toric-Newton transformation in this element which produced r 1 . The toric Newton transformation has two parts; the inverse of the translation is ¯ y 1 → ¯ y 1 + α j = y 1 while ϕ -1 M ( x , y 1 1 ) = ( x y b 1 -m 1 , x -a 1 y n 1 ) = (¯ x, y ¯) . Hence, the inverse of the toric-Newton transformation will is
↦
$$( x _ { 1 }, \bar { y } _ { 1 } ) \mapsto ( x _ { 1 } ^ { b } ( \bar { y } _ { 1 } + \alpha _ { j } ) ^ { - m }, x _ { 1 } ^ { - a } ( \bar { y } _ { 1 } + \alpha _ { j } ) ^ { n } ) = ( \bar { x }, \bar { y } ).$$
Taking out denominators we obtain ¯ Q x,y /lscript (¯ ¯) which is a divisor of the special pencil ¯ (¯ P x, y, T ¯ ) at this level. It is not hard to see that ¯ Q x,y /lscript (¯ ¯) is ¯ y -regular of order nd /lscript . The contribution of this factor to the ¯ y -degree is the expected one. We continue till we arrive to the first level; at each step the degree on the y -coordinate is multiplied by the corrsponding v-ratio. The final pull-back Q /lscript of ˜ Q /lscript (taking out denominators) to K ( T )[[ x, y ]] is an irreducible factor of P x, y, T ( ) .
Let /lscript 1 , . . . , /lscript s be the dicritical edges of the toric-Newton process. For each dicritical edge /lscript i we consider the sequence of v-ratios n , . . . , n i 1 i h i ( h i is the number of steps till /lscript i appear) and its degree d /lscript i . The factor Q /lscript i has y -order
$$d _ { i } \coloneqq d _ { \ell _ { i } } \cdot \prod _ { j = 1 } ^ { h _ { i } } n _ { j } ^ { i } \\ \text{nd we conclude the $n$}.$$
Theorem 3.3. Let P x, y, T ( ) be a special pencil. Then there is a one-to-one correspondence between dicritical edges of the pencil and irreducible factors of P ∈ K [ T ][[ x, y ]] . By this correspondence to an edge /lscript j we associate the factor Q /lscript j .
If d = ord ( y P ) , note that d = ∑ s j =1 d j and we conclude the next Theorem, see [14] in more generality.
For typical t ∈ K the irreducible components of ˜ C t := { P x, y, t ( ) = 0 } , i.e. Spec( R/ P x,y,t ( ( ))) , are in one-to-one correspondence with the factors of P x, y, T ( ) in K ( T )[[ x, y ]] and the factors corresponding to a given factor in K [ T ][[ x, y ] are the curvettes of the corresponding dicritical (as many as the degree).
## 4. Special pencils, polynomials and atypical fibers
In this section we recall the well-known relationship between special pencils and polynomials. The polynomial f ( x, y ) ∈ K [ x, y ] , D := deg f , defines a polynomial map f : A 2 K → A 1 K , where A j := A j K is the affine space of dimension j over K . We consider ( x, y ) the affine coordinates of A 2 and [ X : Y : Z ] the homogeneous coordinates of P 2 := P 2 K with the inclusion ( x, y ) ↪ → [ x : y : 1] . Let us consider the rational extension of f to a map ˜ : f P 2 /axisshort/axisshort/arrowaxisright P 1 ≡ K ∪ {∞} . If f ( x, y ) = ∑ D j =0 f j ( x, y ) is the decomposition in homogeneous components then
$$B \coloneqq \{ [ u \colon v \colon 0 ] \, | \, f _ { D } ( u, v ) = 0 \}$$
is the set of base points of ˜ f . At every base point P 0 ∈ B (at the line at infinity) the corresponding pencil is an special pencil.
/negationslash
Assume that P 0 := [1 : 0 : 0] is one of these points. In the affine chart X = 0 (with affine coordinates y, z ) this map looks like
$$\frac { f _ { y } ( y, z ) } { z ^ { D } }, \ \ f _ { y } ( y, z ) \coloneqq z ^ { D } f \left ( \frac { 1 } { z }, \frac { y } { z } \right )$$
and the fibers of ˜ f near P 0 are of the form f y ( y, z ) -tz D = 0 , for t ∈ K ∪{∞} , hence a special pencil. By definition the dicriticals of the polynomial f at infinity are the dicriticals of the corresponding special pencils at all base points P 0 ∈ B . We define accordingly the atypical values at infinity at a dicritical of the polynomial f , see also [19].
In [22], Gwoździewicz finds that the number of atypical values at infinity of a polynomial is bounded above by the minimum of the two following numbers:
- · The number ν gen ∞ of the branches at infinity of a generic fiber.
- · The number ν min ∞ +1 where ν min ∞ is the minimal number of branches at infinity for any fiber.
Therefore an algebraic proof of these results follows immediately from our algebraic proof of Corollary 2.12.
In the same work, Gwoździewicz asked if it is possible to reach the bound ν gen ∞ (or ν min ∞ + 1 ). As we have observed in Remark 2.13, to reach this bound imposes strong conditions on the special pencils over all the dicriticals E :
- · n E > 1 .
- · q ′ E must have simple roots.
- · q E must pairwise separate the values of 0 and the roots of q ′ E .
- · The sets of atypical values are disjoint for any pair of dicriticals.
When we deal with polynomials the last condition must be applied to any dicritical at infinity. Besides this difficulty the geometry of the polynomials imposes more difficulties to find an example reaching the bound.
Namely, no polynomial with only one dicritical reaches the bound. Assume for simplicity that the polynomial is primitive. Then, the only dicritical is of degree 1 , see e.g. [16]. Hence, by [27] all the fibers have only one branch at infinity and by [21], there is no atypical value at infinity. It is not hard to find polynomials with two dicriticals E , E 1 2 both of multiplicity one but n i > 1 . These polynomials have two branches at infinity and one atypical value for each dicritical. The problem is that most obvious examples satisfy that the set of atypical values is the same for both dicriticals.
Gwoździewicz's question. Does there exist a polynomial f ( x, y ) with n nonzero critical values at infinity such that the curve f ( x, y ) = 0 has n branches at infinity?
Example 4.1 . No polynomial of degree ≤ 10 and two dicriticals reaches the bound. The polynomial
$$p ( x, y ) = x ^ { 6 } y ^ { 5 } - 5 x ^ { 5 } y ^ { 4 } + 1 0 x ^ { 4 } y ^ { 3 } - 2 x ^ { 3 } y ^ { 3 } - 1 0 x ^ { 3 } y ^ { 2 } + 5 x ^ { 2 } y ^ { 2 } + 5 x ^ { 2 } y - \frac { 1 5 } { 4 } x y - x + y$$
does. We will show that this polynomial p x, y ( ) has two non-zero critical values at infinity and the curve p x, y ( ) = 0 has two branches at infinity. This polynomial can be written as
$$p ( x, y ) = \left ( x ^ { 3 } y ^ { 2 } - 1 \right ) ^ { 2 } y + ( x y - 1 ) ^ { 5 } x - x ^ { 6 } y ^ { 5 } + 5 x y \left ( x y - \frac { 3 } { 4 } \right ).$$
y
(a) Newton polygon of p
(b) Newton polygon of p y
Figure 10
In order to obtain the resolution of the polynomial we have to study the special pencils located at the two points at infinity of p . The first one is given by
$$p _ { x } ( x, z ) = ( x ^ { 3 } - z ^ { 5 } ) ^ { 2 } + \dots - t z ^ { 1 1 }$$
and it is the one in Example 1.4 (see also Example 2.5). We have seen that it has only one dicritical which is of degree one and v-ratio 2 . There is only one atypical value for this dicritical, namely t = -5 8 . Let us study now the special pencil associated to the other point at infinity:
$$p _ { y } ( y, z ) = ( y - z ^ { 2 } ) ^ { 5 } + \dots - t z ^ { 1 1 }$$
We are in the situation of Remark 1.7, hence we perform a translation as a change of variables, y = y 1 + z 2 1 , z = z 1 . In Figure 11a we see the new Newton polygon where the coefficient of z 11 equals -( t + 3 4 ) . The Newton polygon for t = -3 4 is in Figure 11b. Hence, there is one atypical value for this polynomial associated to this dicritical.
Then, the two atypical values for each dicritical are different and the polynomial p reaches the bound: as many non-zero atypical fibers at infinity as branches at infinity for the fiber p x, y ( ) = 0 . The two atypical fibers at infinity have three branches. The polynomial p has only one (affine) singular fiber p -1 ( -20 27 ) which has an ordinary double point at ( -900 , -4 3375 ) .
Example 4.2 . In the same way as in the local case, see Example 2.14, the following polynomial shows that our bounds are better than the ones in [22]. Consider the following polynomial of degree 10 (see
y
(a) Newton polygon of f
y
Figure 11
y
(b) Newton polygon of f y
Figure 12
its Newton polygon in Figure 12a):
$$\text{ts Newton polygon in Figure 12a} \colon \\ f ( x, y ) = y ^ { 6 } - 4 ( x ^ { 2 } + 1 ) y ^ { 5 } + \left ( 1 2 x ^ { 2 } + 6 x ^ { 4 } + \frac { 4 1 } { 4 } \right ) y ^ { 4 } - \left ( 4 x ^ { 6 } + \frac { 2 5 } { 2 } + 1 2 x ^ { 4 } + \frac { 9 9 } { 4 } x ^ { 2 } \right ) y ^ { 3 } \\ + \left ( x ^ { 8 } + 4 x ^ { 6 } + \frac { 7 5 } { 4 } x ^ { 4 } + \frac { 5 9 } { 4 } x ^ { 2 } \right ) y ^ { 2 } + \left ( - \frac { 1 7 } { 4 } x ^ { 6 } + \frac { 7 5 } { 4 } x ^ { 2 } + 4 x ^ { 4 } + \frac { 2 5 } { 4 } \right ) y \\ - \frac { 2 5 } { 2 } x ^ { 2 } - \frac { 2 5 } { 4 } x ^ { 6 } - \frac { 7 1 } { 4 } x ^ { 4 }. \\ \text{This polynomial has two points at infinity, that is } P _ { 0 } = [ 1 \colon 0 \colon 0 ] \text{ and } P _ { 1 } \, = \, [ 0 \, \colon 1 \, \colon 0 ] \text{ Thus the}$$
This polynomial has two points at infinity, that is P 0 = [1 : 0 : 0] and P 1 = [0 : 1 : 0] Thus the corresponding special pencil at P 0 is given by
$$\text{ corresponding special pencil at $P_{0}$ is given by} \\ f _ { y } ( z, y ) - T z ^ { 1 0 } = y ^ { 6 } z ^ { 4 } - 4 y ^ { 5 } z ^ { 3 } - 4 y ^ { 5 } z ^ { 5 } + 1 2 y ^ { 4 } z ^ { 4 } + 6 y ^ { 4 } z ^ { 2 } + \frac { 4 1 } { 4 } y ^ { 4 } z ^ { 6 } - 4 y ^ { 3 } z - \frac { 2 5 } { 2 } y ^ { 3 } z ^ { 7 } \\ - 1 2 y ^ { 3 } z ^ { 3 } - \frac { 9 9 } { 4 } y ^ { 3 } z ^ { 5 } + 4 y ^ { 2 } z ^ { 2 } + \frac { 7 5 } { 4 } y ^ { 2 } z ^ { 4 } + \frac { 5 9 } { 4 } y ^ { 2 } z ^ { 6 } + y ^ { 2 } - \frac { 1 7 } { 4 } y z ^ { 3 } \\ + \frac { 7 5 } { 4 } y z ^ { 7 } + 4 y z ^ { 5 } + \frac { 2 5 } { 4 } y z ^ { 9 } - \frac { 2 5 } { 2 } z ^ { 8 } - \frac { 2 5 } { 4 } z ^ { 4 } - \frac { 7 1 } { 4 } z ^ { 6 } - T z ^ { 1 0 }$$
Let us see that this special pencil has 2 branches for all t ∈ K and it has two dicriticals E 1 and E 2 of degree 1 . Its Newton polygon (see Figure 12b) has only one edge /lscript which is not dicritical and such that
$$P _ { \ell } = y ^ { 2 } - \frac { 2 5 } { 4 } z ^ { 4 } _ { \rho } = \frac { ( 2 y - 5 z ^ { 2 } ) ( 2 y + 5 z ^ { 2 } ) } { 4 }.$$
Thus q /lscript has degree 2 and two simple roots ± 5 2 . Making the toric-Newton transformation associated to each root ( /lscript, ± 5 2 ) one gets two dicriticals, each one of degree 1 (which are sections with no bamboo). Moreover, these two dicriticals have no atypical value associated.
- (a) Newton polygon of f x
x
1
(b) Newton polygon of f x, 1
Figure 13
The other special pencil at P 1 is given by
$$\text{$i ne owner special pencil at $r_{1}$ is given by} \\ f _ { x } ( z, x ) - T z ^ { 1 0 } = - \frac { 2 5 } { 4 } z ^ { 4 } x ^ { 6 } - \frac { 1 7 } { 4 } z ^ { 3 } x ^ { 6 } + 4 z ^ { 2 } x ^ { 6 } - 4 x ^ { 6 } z - 1 2 z ^ { 3 } x ^ { 4 } + 6 x ^ { 4 } z ^ { 2 } \\ - \frac { 7 1 } { 4 } z ^ { 6 } x ^ { 4 } + 4 z ^ { 5 } x ^ { 4 } + \frac { 7 5 } { 4 } z ^ { 4 } x ^ { 4 } - \frac { 2 5 } { 2 } z ^ { 8 } x ^ { 2 } + \frac { 7 5 } { 4 } z ^ { 7 } x ^ { 2 } + \frac { 5 9 } { 4 } z ^ { 6 } x ^ { 2 } - \frac { 9 9 } { 4 } z ^ { 5 } x ^ { 2 } \\ + 1 2 x ^ { 2 } z ^ { 4 } - 4 x ^ { 2 } z ^ { 3 } + z ^ { 4 } - 4 z ^ { 5 } + \frac { 4 1 } { 4 } z ^ { 6 } - \frac { 2 5 } { 2 } z ^ { 7 } + \frac { 2 5 } { 4 } z ^ { 9 } + x ^ { 8 } - T z ^ { 1 0 }. \\ \text{$t us check that this special pencil has $4$ branches for all $t\in\mathbb{K}$ and one dictric $E$. Its Newt}$$
Let us check that this special pencil has 4 branches for all t ∈ K and one dicritical E. Its Newton polygon is in Figure 13a; there is only one edge /lscript , which is not dicritical and the quasihomogenous polynomial associated to the edge is P /lscript = ( x 2 -z ) 4 . We need only one toric-Newton transformation at this stage:
↦
$$\varphi _ { M } ( z _ { 1 }, x _ { 1 } ) = ( z _ { 1 } ^ { 2 } x _ { 1 }, z _ { 1 } x _ { 1 } ), \ \ x _ { 1 } \mapsto \bar { x } _ { 1 } + 1$$
The Newton polygon of the strict transform f x, 1 ( z , x 1 ¯ ) 1 is in Figure 13b. We have only one edge /lscript 1 , which is non-dicritical with P /lscript 1 = (¯ x 1 + z 2 1 ) 4 . If we perform the translation of Remark 1.7 we obtain a new special pencil f x, 2 ( z , x 2 2 ) . The Newton polygon is in Figure 14. We have only one edge /lscript 2 , which is dicritical, since
$$P _ { \ell _ { 2 } } = x _ { 2 } ^ { 4 } - 2 x _ { 2 } ^ { 2 } z _ { 2 } ^ { 5 } + ( 1 - T ) z _ { 2 } ^ { 1 2 },$$
i.e., its v-ratio equals 2 , q /lscript ( z ) = z 4 -2 z 2 + 1 -T and q E ( z ) = z 4 -2 z 2 +1 . The roots of q ′ E ( z ) are α = 0 1 , , -1 , hence the bound equals 3 . Since q E (0) = 0 and q E (1) = q E ( -1) = 1 , there are exactly two atypical values, t = 0 1 , .
x
Figure 14. Newton polygon of f x, 2
Example 4.3 . The referee asked whether Gwoździewicz's question has an affirmative answer for other positive integer n ≥ 3 . In this example we provide a polynomial family which confirms the required positive answer.
For any d , we consider two monic polynomials q t , Q t ( ) ( ) ∈ K [ t ] of degrees 2 d and 2 d +1 , respectively, such that:
(C1) deg( Q t ( ) -tq ( )) t ≤ d . (C2) q t ( ) = ∏ m j =1 ( t -a j ) m j , ∑ m j =1 m j = 2 d , m j ≥ 2 . (C3) Q t ( ) = ∏ n j =1 ( t -b j ) n j , ∑ n j =1 n j = 2 d +1 , n j ≥ 2 . Let f ( x, y ) be the polynomial
$$f ( x, y ) = \left ( y + 1 \right ) \left ( x q ( x y ) + \left ( y + 1 \right ) Q ( x y ) \right ).$$
Its Newton polygon has four edges whose vertices are given by
$$[ ( 0, 0 ), ( 0, 1 ), ( 2 d + 1, 2 d + 2 ), ( 2 d + 1, 2 d ), ( 1, 0 ) ].$$
Let /lscript 1 = [(0 , 1) (2 , d +1 2 , d +2)] , /lscript 2 = [(1 , 0) (2 , d +1 2 )] , d and /lscript 3 = [(2 d +1 2 ) (2 , d , d +1 2 , d +2)] be the edges not passing through the origin.
The support polynomial f /lscript 1 is yQ xy ( ) . Because of condition (C3), one can see that each root b j induces a dicritical section with bamboo, producing exactly one atypical value.
The support polynomial f /lscript 2 is xq xy ( ) . As above, condition (C2) implies that each root a j induces a dicritical section with bamboo, producing exactly one atypical value.
The support polynomial of the vertical edge /lscript 3 is f /lscript 3 = x 2 d +1 y 2 d ( y +1) 2 . The condition (C1) implies that the translation y = y 1 -1 produces a new edge /lscript ′ 3 = [(0 0) (2 , , d +1 2)] , . Hence /lscript ′ 3 is a dicritical edge with bamboo ( v /lscript ′ 3 = 2 ) and only one atypical value.
Of course the conditions (C1), (C2) and (C3) impose restrictions but one can see that solutions exist and for generic choices, the atypical values for each dicritical are distinct, providing the required affirmative answer.
For example, this is the case if
$$Q ( t ) = ( t + 1 ) ^ { 2 d + 1 } \text{ and } p ( t ) = \left ( \prod _ { j = 1 } ^ { d } ( t - a _ { j } ) \right ) ^ { 2 }. \\ \triangleright \text{ give the coefficients of } p ( t ). \text{ A tedious verification ensures that } j$$
Then (C1) allows to give the coefficients of p t ( ) . A tedious verification ensures that f has d +2 dicritical sections with bamboo, and the generic fiber has d +2 branches at infinity, its genus being d . The fact that they have d + 2 different atypical values has been checked for small values of d ( ≤ 20 ) with SAGE [28].
Remark 4.4 . Note that for d = 1 , we can obtain a polynomial with three branches and degree 7 , while Example 4.1, with two branches, has degree 11 . Surprisingly, this is the smallest degree for a two-branch polynomial reaching the bound. Note that all the examples have only dicritical sections.
Example 4.5 . Both Examples 4.1 and 4.3 have only dicritical sections. We have found also an example of degree 18 , with two dicriticals (with bamboo), one of them E with multiplicity 2 , hence having also three branches at infinity for the generic fiber and three atypical values. The fiber corresponding to the value in A ∗ E has only two branches at infinity, i.e., ν min ∞ +1 = ν gen ∞ , check the bounds in Page 15.
## Acknowledgements
First author is partially supported by MTM2010-21740-C02-02; the last two authors are partially supported by the grant MTM2010-21740-C02-01. The authors would like to thank Pierrette CassouNoguès and referees for useful remarks and valuable comments.
## Appendix A. Hensel's Lemma
In order to be clear which flavor of Hensel's Lemma we are going to use, we state and prove the following elementary result
/negationslash
Let K be a field and fix a weight ω x, y ( ) := nx + my for n, m ∈ N . Given 0 = F ∈ K [[ x, y ]] , we will consider its decomposition in ω -quasihomogeneous forms
$$( \text{A.1} ) & & F ( x, y ) = F _ { a + b } ( x, y ) + F _ { a + b + 1 } ( x, y ) + \dots,$$
where the subindex means the ω -weight
Lemma A.1 (Hensel's Lemma) . Asume that F a + b ( x, y ) = f a ( x, y ) g b ( x, y ) , f , g a b ∈ K [ x, y ] quasihomogeneous and coprime.h Then, there exist
$$f, g \in \mathbb { K } [ [ X, Y ] ], \ \ f = f _ { a } + f _ { a + 1 } + \dots, \ \ g = g _ { b } + g _ { b + 1 } + \dots$$
such that F = fg . Moreover if f a is an irreducible polynomial, then f is an irreducible power series.
Proof. We need to find recursively ω -quasihomogeneous polynomials f a + k , g b + k , k ∈ N such that
$$( A. 2 ) & f _ { a } ( x, y ) g _ { b + k } ( x, y ) + g _ { b } ( x, y ) f _ { a + k } ( x, y ) = F _ { a + b + k } ^ { * } ( x, y )$$
where g b + k , f a + k are the unknowns and F ∗ a + + b k is obtained from F a + + b k and the previous solutions for k ′ < k .
Let us decompose the above polynomials (where now the subindex correspond now to the homogeneous degree for the weigh ω 0 defined by n = m = 1 ):
$$\begin{array} {$$
The decompositions of a, b, c, d, e are unique if we assume that the all indices are non-negative, the coefficient of n is less than m and the coefficient of m is less than n . We prove it in several steps.
Claim 1 . The statement holds for ω 0 , i.e., the homogeneous case.
It is an immediate consequence of the properties of the resultant.
Claim 2 . The statement holds if f a ( x, y ) is a power of x or y .
Assume that f a is a power of x . In this case, we have
- · a = n a ( x + ma ′ ) , 0 ≤ a x < m .
/negationslash
- · g b (0 , 1) = 0 , i.e., b x = 0 .
The following equalities hold:
$$n ( a _ { x } + d _ { x } ) + m d _ { y } +$$
We deduce that e x = c x = a x + d x -αm e , y = d y = b y + d y -βn , where α, β ∈ { 0 1 , } and ′ ′
$$e = a ^ { \prime } + d + \alpha = b ^ { \prime } + c$$
Equation (A.2) is equivalent to
$$x ^ { \alpha + a ^ { \prime } } \tilde { g } _ { d$$
which follows from Claim 1, and Claim 2 holds.
Claim 3 . The statement holds if both f a and g b are coprime with x, y .
In this case a x = a y = b x = b y = 0 and
$$d _ { x } = e _ { x }, \ \ d _ { y } = e _$$
Hence (A.2) is transformed again in its homogeneous version and Claim 3 follows from again from Claim 1. Combining these claims, the statement is proved. /square
## E. ARTAL, I. LUENGO, AND A. MELLE
## References
- [1] S.S. Abhyankar, Lectures on expansion techniques in algebraic geometry , Tata Institute of Fundamental Research Lectures on Mathematics and Physics, vol. 57, Tata Institute of Fundamental Research, Bombay, 1977, Notes by Balwant Singh.
[2] , Dicritical divisors and Jacobian problem , Indian J. Pure Appl. Math. 41 (2010), no. 1, 77-97.
[3] , More about dicriticals , Proc. Amer. Math. Soc. 139 (2011), no. 9, 3083-3097.
[4] , Pillars and towers of quadratic transformations , Proc. Amer. Math. Soc. 139 (2011), no. 9, 3067-3082.
[5] , Quadratic transforms inside their generic incarnations , Proc. Amer. Math. Soc. 140 (2012), no. 12, 4111- 4126.
[6] , Dicriticals of pencils and Dedekind's Gauss lemma , Rev. Mat. Complut. 26 (2013), no. 2, 735-752.
[7] , Generic incarnations of quadratic transforms , Proc. Amer. Math. Soc. 141 (2013), no. 12, 4103-4117.
[8] S.S. Abhyankar and E. Artal, Algebraic theory of curvettes and dicriticals , Proc. Amer. Math. Soc. 141 (2013), no. 12, 4087-4102.
[9] , Analytic theory of curvettes and dicriticals , to appear in Rev. Mat. Complut., doi:10.1007/s13163-014-0148- 4, 2014.
[10] S.S. Abhyankar and W.J. Heinzer, Existence of dicritical divisors revisited , Proc. Indian Acad. Sci. Math. Sci. 121 (2011), no. 3, 267-290.
[11] , Existence of dicritical divisors , Amer. J. Math. 134 (2012), no. 1, 171-192.
[12] , Rees valuations , Proc. Indian Acad. Sci. Math. Sci. 122 (2012), no. 4, 525-546.
[13] S.S. Abhyankar and I. Luengo, Algebraic theory of dicritical divisors , Amer. J. Math. 133 (2011), no. 6, 1713-1732.
[14] , Spiders and multiplicity sequences , Proc. Amer. Math. Soc. 141 (2013), no. 12, 4071-4085.
[15] S.S. Abhyankar and T.T. Moh, Newton-Puiseux expansion and generalized Tschirnhausen transformation. I, II , J. Reine Angew. Math. 260 (1973), 47-83; ibid. 261 (1973), 29-54.
[16] E. Artal, Une démonstration géométrique du théorème d'Abhyankar-Moh , J. Reine Angew. Math. 464 (1995), 97-108.
[17] E. Artal, Pi. Cassou-Noguès, I. Luengo, and A. Melle-Hernández, On ν -quasi-ordinary power series: factorization, Newton trees and resultants , Topology of algebraic varieties and singularities, Contemp. Math., vol. 538, Amer. Math. Soc., Providence, RI, 2011, pp. 321-343.
[18] Pi. Cassou-Noguès, Newton trees at infinity of algebraic curves , Affine algebraic geometry, CRM Proc. Lecture Notes, vol. 54, Amer. Math. Soc., Providence, RI, 2011, pp. 1-19.
[19] A.H. Durfee, Five definitions of critical point at infinity , Singularities (Oberwolfach, 1996) (Basel), Progr. Math., no. 162, Birkhäuser, 1998, pp. 345-360.
[20] W. Engel, Ein Satz über ganze Cremona-Transformationen der Ebene , Math. Ann. 130 (1955), 11-19.
[21] R. Ephraim, Special polars and curves with one place at infinity , Singularities, Part 1 (Arcata, Calif., 1981), Proc. Sympos. Pure Math., vol. 40, Amer. Math. Soc., Providence, RI, 1983, pp. 353-359.
[22] J. Gwoździewicz, Ephraim's pencils , Int. Math. Res. Not. IMRN (2013), no. 15, 3371-3385.
[23] Z. Jelonek and M. Tibar, Bifurcation locus and branches at infinity of a polynomial f : C 2 → C , Preprint available at arXiv:1401.6544v1 [math.AG] .
[24] D.T. Lê and C. Weber, Équisingularité dans les pinceaux de germes de courbes planes et C 0 -suffisance , Enseign. Math. (2) 43 (1997), no. 3-4, 355-380.
[25] A. Melle-Hernández and C.T.C. Wall, Pencils of curves on smooth surfaces , Proc. London Math. Soc. (3) 83 (2001), no. 2, 257-278.
[26] T.T. Moh, On analytic irreducibility at ∞ of a pencil of curves , Proc. Amer. Math. Soc. 44 (1974), 22-24.
[27] , On the Jacobian conjecture and the configurations of roots , J. Reine Angew. Math. 340 (1983), 140-212.
[28] W.A. Stein et al., Sage Mathematics Software (Version 5.10) , The Sage Development Team, 2013, http://www.sagemath.org .
[29] H. Żołądek, An application of Newton-Puiseux charts to the Jacobian problem , Topology 47 (2008), no. 6, 431-469.
IUMA, Departamento de Matemáticas, Facultad de Ciencias, Universidad de Zaragoza, c/ Pedro Cerbuna 12, 50009 Zaragoza, SPAIN
E-mail address : [email protected]
ICMAT (CSIC-UAM-UC3M-UCM), Departamento de Álgebra, Facultad de Ciencias Matemáticas, Universidad Complutense, 28040 Madrid, SPAIN
E-mail address : [email protected],[email protected]
|
10.1142/S0219498815400095
|
[
"E. Artal Bartolo",
"I. Luengo",
"A. Melle-Hernández"
] |
2014-08-04T16:54:25+00:00
|
2014-08-04T16:54:25+00:00
|
[
"math.AG",
"14A05, 14R15"
] |
High-school algebra of the theory of dicritical divisors: atypical fibres for special pencils and polynomials
|
In this work we deal with dicritical divisors, curvettes and polynomials.
These objects have been one of the main research interests of S.S. Abhyankar
during his last years. In this work we provide some elementary proofs of some
S.S. Abhyankar and I. Luengo results for dicriticals in the framework of formal
power series. Based on these ideas we give a constructive way to find the
atypical fibres of a special pencil and give bounds for its number, which are
sharper than the existing ones. Finally, we answer a question of J.
Gwo\'zdziewicz finding polynomials that reach his bound.
|
1408.0744v1
|
## AN L p THEORY OF SPARSE GRAPH CONVERGENCE II: LD CONVERGENCE, QUOTIENTS, AND RIGHT CONVERGENCE
CHRISTIAN BORGS, JENNIFER T. CHAYES, HENRY COHN, AND YUFEI ZHAO
Abstract. We extend the L p theory of sparse graph limits, which was introduced in a companion paper, by analyzing different notions of convergence. Under suitable restrictions on node weights, we prove the equivalence of metric convergence, quotient convergence, microcanonical ground state energy convergence, microcanonical free energy convergence, and large deviation convergence. Our theorems extend the broad applicability of dense graph convergence to all sparse graphs with unbounded average degree, while the proofs require new techniques based on uniform upper regularity. Examples to which our theory applies include stochastic block models, power law graphs, and sparse versions of W -random graphs.
## Contents
| 1. | Introduction | 1 |
|---------------------------------------------------|--------------------------------------------------------|-----|
| 2. | Definitions and main results | 4 |
| 3. | Further definitions, remarks, and examples | 15 |
| 4. | Convergence without the assumption of upper regularity | 20 |
| 5. | Convergent sequences of graphons | 24 |
| 6. | Convergent sequences of uniformly upper regular graphs | 36 |
| 7. | Inferring uniform upper regularity | 42 |
| Appendix A. Proof of the rearrangement inequality | Appendix A. Proof of the rearrangement inequality | 46 |
| References | References | 47 |
## 1. Introduction
In the companion paper [3], we developed a theory of graph convergence for sequences of sparse graphs whose average degrees tend to infinity. These results fill a major gap in the theory of convergent graph sequences, which dealt primarily with either bounded degree graphs or dense graphs. While progress in this direction was made by Bollob´s and Riordan in [2], their approach required a 'bounded density' a condition that excludes many graphs of interest. For example, it cannot handle graphs with heavy-tailed degree distributions such as power laws. To accommodate these and other graphs excluded by the bounded density condition, we generalized the Bollob´s-Riordan approach in [3] to graphs obeying a condition we called a L p upper regularity. We then showed that when p > 1, every sequence of L p upper
Zhao was supported by a Microsoft Research PhD Fellowship and internships at Microsoft Research New England.
regular graphs contains a subsequence converging to a symmetric, measurable function W : [0 , 1] 2 → R that is in L p ([0 , 1] 2 ). Such a function is an L p graphon . Conversely, only L p upper regular sequences can converge to L p graphons, and so our results characterize these limits. The work of Bollob´s and Riordan in [2] and a the prior work on dense graph sequences amount to the special case p = ∞ , while L p graphons with p < ∞ describe limiting behaviors that occur only in the sparse setting. Thus, the L p theory of graphons completes the previous L ∞ theory to provide a rich setting for limits of sparse graph sequences with unbounded average degree.
One attractive feature of dense graph limits is that many definitions of convergence coincide, and it is natural to ask whether the same is true for sparse graphs. After all, there are many ways to formulate the idea that two graphs are similar. For example, one could base convergence on subgraph counts or quotients. Furthermore, statistical physics provides many numerical measures for similarity, such as ground state energies or free energies.
Let us first address the question of subgraph counts. For dense graphs, the sequence ( G n n ) ≥ 0 converges under the cut metric if and only if the F -density in G n converges for all graphs F , where the F -density is the probability that a random map from F to G n is a homomorphism [7]. One might guess that suitably normalized F -densities would characterize sparse graph convergence as well, but this fails dramatically: for sparse graphs, cut metric convergence does not determine subgraph densities (see Section 2.9 of [3]). This is not merely a technicality, but rather a fundamental fact about sparse graphs. We must therefore give up on convergence of subgraph counts as a criterion for sparse graph convergence.
By contrast, we show in this paper that several other widely studied forms of convergence are indeed equivalent to cut metric convergence in the sparse setting. Thus, with the exception of subgraph counts, the scope and consequences of sparse graph convergence are comparable with those of dense graph convergence.
We will consider several notions of convergence motivated by statistical physics and the theory of graphical models from machine learning, such as convergence of ground state energies and free energies, as well as convergence of quotients, 1 which encode 'global' graph properties of interest to computer scientists, such as max-cut and min-bisection. We will also analyze the notion of large deviation (LD) convergence, which was recently introduced for graph sequences with bounded degrees [4] and can easily be adapted to our more general context. For bounded degree graphs, LD convergence was strictly stronger than convergence of quotients or other notions introduced before, but we will see that in our setting it is equivalent to these other forms of convergence.
All these question can be studied for L p upper regular sequences of sparse graphs, but they can also be studied directly for L p graphons. While the former might be more interesting from the point of view of applications, the latter turns out to be more elegant from an abstract point of view. We therefore first develop the theory for sequences of graphons, and then prove our results for sparse graph sequences.
We begin in Section 2 with motivation, definitions, and precise statements of our results, with some ancillary results stated in Section 3. We begin the proofs in Section 4 by completing the cases that do not require the notion of upper regularity. We then make use of upper regularity to deal with graphons in Section 5 and graphs
1 Quotient convergence is also called partition convergence in some of the literature.
in Section 6. Finally, in Section 7, we show that any sequence whose quotients, microcanonical free energies, or ground state energies converge to those of a graphon must be upper regular, which completes the proofs.
Before turning to these details, though, we will explain the motivations behind the different types of convergence analyzed in this paper.
- 1.1. Motivation. When formulating a notion of convergence for growing sequences of graphs, one is immediately faced with the problem of deciding when to consider two large graphs on different numbers of vertices to be similar.
One natural approach is to compare summary statistics, such as weighted counts of homomorphisms to or from small graphs. Convergence based on these statistics is called left convergence if it uses homomorphisms from small graphs and right convergence if it uses homomorphisms to small graphs. Left convergence amounts to using subgraph counts, and as discussed in the previous section it is not a useful tool for characterizing sparse graph convergence. By contrast, right convergence is far more useful in the sparse setting. It amounts to using statistical physics models, and it encompasses quantities such as max-cut, min-bisection, etc. that are important in combinatorial optimization.
The advantage of using summary statistics is that they can easily be normalized to compare graphs on different numbers of nodes. For a more direct approach, one must find other ways to compare such graphs.
One way to deal with this is to blow up both graphs to obtain two new graphs on a common, much larger set of vertices. Conceptually, the most elegant way to do this is probably an infinite blow-up, replacing the vertex sets of both graphs by the interval [0 , 1] and the adjacency matrices by appropriate step functions on [0 , 1] 2 . Comparing the two graphs then reduces to comparing two functions on [0 , 1] 2 , leading to the notion of convergence in the cut norm. A priori, this has the problem that relabeling the nodes of a graph would change its representation as a function on [0 , 1] 2 , but this can be cured by defining the distance as the cut distance of 'aligned' step functions, where alignments are formalized as measure preserving transformations from [0 1] , → [0 , 1], chosen in such a way that the resulting two functions are as close to each other as possible. The resulting definition is known as cut metric convergence , and it was analyzed for sparse graphs in [3].
Another way to deal with the different vertex sets is to 'squint your eyes' and look at whether the results are similar. More formally, one divides the vertex sets of both graphs into q blocks, and then averages the adjacency matrices over the respective blocks, leading to two q × q matrices representing the edge densities between various blocks (we call these matrices q -quotients). One might want to call two graphs similar if their q -quotients are close, but we are again faced with an alignment problem, now of a slightly different kind: different ways of dividing the vertex set of a graph into blocks produce different quotients. While some of the quotients of a graph contain useful information about the graph (for example those corresponding to Szemer´di partitions), others might not. e Unfortunately, it is not a priori clear which of the q -quotients of a graph represent its properties well and which do not. We solve this problem by defining two graphs to be similar if the sets of their q -quotients are close, measured in the Hausdorff distance between subsets of the metric space of weighted graphs on q nodes.
The four notions of convergence describe informally above, namely left convergence, right convergence, convergence in metric, and convergence of quotients, were
Figure 1. Implications between different notions of sparse graph convergence.
already introduced in [7, 8] in the context of sequences of dense graphs. But we felt it to be useful to review the motivation behind these notions, before addressing the extra complications stemming from the fact that we want to analyze sparse graphs.
In this paper we also discuss a fifth notion of convergence: large deviation convergence (LD convergence), which was recently introduced [4] to discuss convergence of bounded degree graphs. Roughly speaking, LD convergence keeps track of not just the possible quotients of a graph but also how often they occur.
Figure 1 illustrates the implications among these concepts. In the upper half of the figure, we see that LD convergence is the strongest notion and ground state energy convergence is the weakest. To complete the cycle and prove that they are all equivalent to metric convergence, we require one hypothesis, namely uniform upper regularity. This notion first arose in [3], and we review its definition below; intuitively, it ensures that subsequential limits are graphons rather than more subtle objects. Indeed, it is possible to state our results using just the fact that limits can be expressed in terms of graphons, without explicitly referring to upper regularity. We will chose this approach when stating our results in Theorem 2.10.
## 2. Definitions and main results
glyph[negationslash]
2.1. Notation. We begin with some notation. As usual, a weighted graph G = ( V, α, β ) consists of a set V = V ( G ) of vertices, vertex weights α x ≥ 0 for x ∈ V , and edge weights β xy = β yx ∈ R for x, y ∈ V . We use E G ( ) to denote the set of edges of G , i.e., the set of pairs { x, y } such that β xy = 0. If we consider several graphs at the same time, then we make the dependence on G explicit, denoting the edge weights by β xy ( G ) and the vertex weights by α x ( G ). The maximal node weight of G will be denoted by
$$\alpha _ { \max } ( G ) = \max _ { x \in V ( G ) } \alpha _ { x } ( G ),$$
and the total node weight of G will be denoted by
$$\alpha _ { G } = \sum _ { x \in V ( G ) } \alpha _ { x } ( G ).$$
We will always assume that α G is strictly positive. If U is a subset of V ( G ), we will use α U ( G ) to denote the total weight of U , i.e., α U ( G ) = ∑ x ∈ U α x ( G ). We say a sequence ( G n n ) ≥ 0 of graphs has no dominant nodes if α max ( G n ) /α G n → 0 as n →∞ . Finally, if c ∈ R , we often use cG to denote the weighted graph with vertex weights identical to those of G and edge weights β xy ( cG ) = cβ xy ( G ). If G is a simple graph with edge set E G ( ), we often identify it with the weighted graph with vertex weights 1 and edge weights β xy = 1 xy ∈ E G ( ) . In this case, α G is just the number of vertices in G , and ( β xy ( G )) x,y ∈ V ( G ) is the adjacency matrix. As usual, we use [ n ] to denote the set [ n ] = { 1 , . . . , n } and N to denote the set of positive integers. Finally, we define the density of a weighted graph G to be
$$\| G \| _ { 1 } = \sum _ { x, y \in V ( G ) } \frac { \alpha _ { x } ( G ) \alpha _ { y } ( G ) } { \alpha _ { G } ^ { 2 } } | \beta _ { x y } ( G ) |.$$
Note that for an unweighted graph without self-loops, ‖ G ‖ 1 is just the edge density 2 | E G ( ) | / V | ( G ) | 2 .
- 2.2. Convergence in metric. One of the main topics studied in [3] is the conditions under which a sequence of sparse graphs contains a subsequence that converges in metric. This question led us to the notion of L p upper regularity, and more generally uniform upper regularity. Upper regularity plays an important role in the proofs in the present paper, but it is not essential for stating our main results. We therefore defer the discussion of uniform upper regularity to Section 2.7 and restrict ourselves here to just defining convergence in metric. For examples, see Section 3.3.
As already discussed, it is convenient to define this distance by embedding the space of graphs into the set of functions from [0 , 1] 2 into the reals.
Definition 2.1. An L p graphon is a measurable, symmetric function W : [0 , 1] 2 → R such that
$$\| W \| _ { p } \coloneqq \left ( \int | W ( x, y ) | ^ { p } \, d x \, d y \right ) ^ { 1 / p } < \infty.$$
Here symmetry means W x,y ( ) = W y,x ( ) for all ( x, y ) ∈ [0 , 1] 2 . If we do not specify p , we assume that W is in L 1 and call it simply a graphon, rather than an L 1 graphon.
On the set of graphons, one defines the cut norm ‖ · ‖ glyph[square] by
$$\left \| W \right \| _ { \Box } = \sup _ { S, T \subseteq [ 0, 1 ] } \left | \int _ { S \times T } W ( x, y ) \, d x \, d y \right |,$$
where the supremum is over measurable sets S, T ⊆ [0 , 1]; this notion goes back to the classic paper of Frieze and Kannan [10] on the 'weak regularity' lemma. One then defines the cut distance between two graphons U and W by
$$\delta _ { \Box } ( U, W ) = \inf _ { \phi } \| U - W ^ { \phi } \| _ { \Box },$$
where the infimum is over all invertible maps φ : [0 , 1] → [0 , 1] such that both φ and its inverse are measure preserving, and W φ is defined by W φ ( x, y ) = W φ x ,φ y ( ( ) ( )) (see [6, 12, 7]); such a map φ is called a measure-preserving bijection . After identifying graphons with cut distance zero, the space of graphons equipped with the metric δ glyph[square] becomes a metric space.
To define the cut distance between two weighted graphs, we assign a graphon W G to a weighted graph G as follows: let n = | V ( G ) , | identify V ( G ) with [ n ], and
let I 1 , . . . , I n be consecutive intervals in [0 , 1] of lengths α 1 ( G /α ) G , . . . , α n ( G /α ) G , respectively. We then define W G to be the step function that is constant on sets of the form I u × I v with
$$( 2. 2 ) \quad \ \ W ^ { G } ( x, y ) = \beta _ { u v } ( G ) \quad \text{if} \quad ( x, y ) \in I _ { u } \times I _ { v }.$$
Informally, we consider the adjacency matrix of G and replace each entry ( u, v ) by a square of size α u ( G α ) v ( G /α ) 2 G with the constant function β uv on this square.
With this definition, one easily checks that the density of a weighted graph G can be expressed as ‖ G ‖ 1 = ‖ W G ‖ 1 . For dense graphs, one can define a distance δ glyph[square] ( G,G ′ ) between two graphs by just considering the cut distance between W G and W G ′ . But for sparse graphs, the inequality
$$\delta _ { \Box } ( W ^ { G }, W ^ { G ^ { \prime } } ) \leq \| W ^ { G } - W ^ { G ^ { \prime } } \| _ { \Box } \leq \| W ^ { G } \| _ { 1 } + \| W ^ { G ^ { \prime } } \| _ { 1 }$$
means the cut distance is not very informative, since under this metric all sparse graph sequences are Cauchy sequences.
To overcome this problem, we identify weighted graphs whose edge weights only differ by a multiplicative factor. 2 Explicitly, we introduce the distance
$$\delta _ { \Box, \text{norm} } ( G, G ^ { \prime } ) = \delta _ { \Box } \left ( \frac { 1 } { \| G \| _ { 1 } } W ^ { G }, \frac { 1 } { \| G ^ { \prime } \| _ { 1 } } W ^ { G ^ { \prime } } \right ),$$
where in the degenerate case of a graph G with ‖ G ‖ 1 = 0 we define 1 ‖ G ‖ 1 W G to be zero. As an example, with this definition, two random graphs G n,p for different p can be shown to be close in the metric δ glyph[square] , norm , as are two random graphs with different numbers of nodes, at least as long as pn →∞ as n →∞ (see Section 3.3).
Definition 2.2 ([3]) . Let ( G n n ) ≥ 0 be a sequence of weighted graphs, and let W be a graphon. We say that ( G n n ) ≥ 0 is convergent in metric if ( G n n ) ≥ 0 is a Cauchy sequence in the metric δ glyph[square] , norm ( G,G ′ ) defined in (2.3), and we say that G n converges to W in metric if δ glyph[square] ( 1 ‖ G n ‖ 1 W G n , W ) → 0. (Again, we set 1 ‖ G n ‖ 1 W G n = 0 if ‖ G n ‖ 1 = 0.)
2.3. Convergence of quotients. The next object we define is convergence of quotients. To formalize this, consider a weighted graph G and a partition P = ( V , . . . , V 1 q ) of V ( G ) into q parts, some of which could be empty. Equivalently, consider a map φ : V ( G ) → [ q ] (related to P by setting φ x ( ) = i iff x ∈ V i ). We will define a quotient G/φ = G/ P as a pair ( α, β ) = ( α G/φ ,β G/φ ( ) ( )), where α ∈ R q is a vector encoding the total vertex weights of the classes in P and β ∈ R q × q is a matrix encoding the number of edges (weighted by their edge weights) between different classes. Explicitly,
$$\alpha _ { i } ( G / \phi ) = \frac { \alpha _ { V _ { i } } ( G ) } { \alpha _ { G } }$$
and
$$\beta _ { i j } ( G / \phi ) = \frac { 1 } { \| G \| _ { 1 } } \sum _ { ( u, v ) \in V _ { i } \times V _ { j } } \frac { \alpha _ { u } ( G ) } { \alpha _ { G } } \frac { \alpha _ { v } ( G ) } { \alpha _ { G } } \beta _ { u v } ( G ).$$
(In the degenerate case where G has no edges and ‖ G ‖ 1 = 0, we set β G/φ ( ) = 0.) We call G/φ a q -quotient of G , and we denote the set of all q -quotients of G
2 Of course, this slightly decreases our ability to distinguish between dense graphs.
by S q ( G ). Note that without the normalization factor 1 ‖ G ‖ 1 in (2.5), the weights β ij ( G/φ ) would scale with the density of G , which means that all quotients of a sparse sequence would tend to zero. We have chosen this factor in such a way that
$$\| \beta ( G / \phi ) \| _ { 1 } = \sum _ { i, j } | \beta _ { i j } ( G / \phi ) | \leq 1,$$
with equality if and only if G has non-negative weights and density ‖ G ‖ 1 > 0. We will consider S q ( G ) as a subset of
$$( 2. 6 ) \quad \mathcal { S } _ { q } = \left \{ ( \alpha, \beta ) \in [ 0, 1 ] ^ { q } \times [ - 1, 1 ] ^ { q \times q } \colon \sum _ { i \in [ q ] } \alpha _ { i } = 1 \text{ and } \sum _ { i, j \in [ q ] } | \beta _ { i j } | \leq 1 \right \},$$
equipped with the usual glyph[lscript] 1 distance on R q + q 2 ,
$$d _ { 1 } ( ( \alpha, \beta ), ( \alpha ^ { \prime }, \beta ^ { \prime } ) ) = \sum _ { i \in [ q ] } | \alpha _ { i } - \alpha _ { i } ^ { \prime } | + \sum _ { i, j \in [ q ] } | \beta _ { i j } - \beta _ { i j } ^ { \prime } |,$$
which turns S q into a compact metric space ( S q , d 1 ), a fact we will use repeatedly in this paper. For a ∈ glyph[triangle] q , we define the subspace
$$\mathcal { S } _ { \mathfrak a } = \{ ( \alpha, \beta ) \in \mathcal { S } _ { q } \colon \alpha = \mathfrak a \},$$
which is closed and hence also compact. Note that our normalizations are a little different from those in [8], in order to ensure compactness. 3
The quotients of a graph G allow one to express many properties of interest to combinatorialists and computer scientists in a compact form. For example, the size of a maximal cut in a simple graph G ,
$$\max \text{Cut} ( G ) = \max _ { W \subseteq V ( G ) } \sum _ { ( x, y ) \in W \times ( V ( G ) \langle W ) } \beta _ { x y } ( G ),$$
can be expressed as
$$\text{MaxCut} ( G ) = | E ( G ) | \max _ { ( \alpha, \beta ) \in \mathcal { S } _ { 2 } ( G ) } \left ( \beta _ { 1 2 } + \beta _ { 2 1 } \right ).$$
Restricting oneself to the subset of quotients ( α, β ) ∈ S 2 ( G ) such that α 1 = α 2 = 1 2, / one can express quantities like min- or max-bisection, and considering S q ( G ) for q > 2, one obtains weighted versions of max-cut for partitions into more than two sets.
To define convergence of quotients, we need the Hausdorff metric on subsets of a metric space ( X,d ). As usual, it is a metric d Hf on the set of nonempty compact subsets of X , defined by
$$d ^ { H f } ( S, S ^ { \prime } ) = \max \left \{ \sup _ { x \in S } d ( x, S ^ { \prime } ), \sup _ { y \in S ^ { \prime } } d ( y, S ) \right \},$$
where
$$d ( x, S ) = \inf _ { y \in S } d ( x, y ).$$
If d is a complete metric, then so is d Hf (see [13]), and the same holds for total boundedness. Thus, starting from the metric space ( S q , d 1 ), this gives a metric d Hf 1
3 Specifically, the analogue of (2.4) and (2.5) in [8] would be to use β ij ( G/φ / α ) ( i ( G/φ α ) j ( G/φ ) ) instead of β ij ( G/φ ), while modifying the definition of d 1 accordingly. This would encode essentially the same information, but the analogue of S q would not be compact.
on the space of nonempty compact subsets of S q , and this space inherits compactness from the compactness of ( S q , d 1 ).
Definition 2.3. Let ( G n n ) ≥ 0 be a sequence of weighted graphs. We say that the sequence ( G n n ) ≥ 0 has convergent quotients if for each q , there exists a closed set S ∞ q ⊆ S q such that S q ( G n ) converges to S ∞ q in the Hausdorff metric.
Remark 2.4 . Note that the closedness of the set S ∞ q ⊆ S q can be assumed without loss of generality (every set has Hausdorff distance zero from its closure, which is why the Hausdorff metric is restricted to closed sets). Furthermore, because of the compactness of the Hausdorff metric, convergence of quotients is equivalent to the statement that the quotients S q ( G n ) form a Cauchy sequence. It is then easy to verify that the limiting set S ∞ q can be expressed as 4
$$\mathcal { S } _ { q } ^ { \infty } = \{ ( \alpha, \beta ) \in \mathcal { S } _ { q } \, \colon d _ { 1 } \left ( ( \alpha, \beta ), \mathcal { S } _ { q } ( G _ { n } ) \right ) \rightarrow 0 \}.$$
2.4. Statistical physics and multiway cuts. Next we define some notions motivated by concepts from statistical physics (or, for a different audience, by the concept of graphical models in machine learning).
Consider a weighted graph G . We will randomly color the vertices of G with q colors; i.e., we will consider random maps φ : V ( G ) → [ q ]. We allow for all possible maps, not just proper colorings, and call such a map a spin configuration . To make the model nontrivial, different spin configurations get different weights, based on a symmetric q × q matrix J with entries J ij ∈ R called the coupling matrix . Given G and J , a map φ : V ( G ) → [ q ] then gets an energy
$$E _ { \phi } ( G, J ) = - \frac { 1 } { \| G \| _ { 1 } } \sum _ { u, v \in V ( G ) } \frac { \alpha _ { u } ( G ) \alpha _ { v } ( G ) } { \alpha _ { G } ^ { 2 } } \beta _ { u v } ( G ) J _ { \phi ( v ) \phi ( u ) }.$$
(If G has no edges, we set this term equal to zero.) Given a vector a = ( a , . . . , a 1 q ) of nonnegative real numbers adding up to 1 (we denote the set of these vectors by glyph[triangle] q ), we consider configurations φ such that the (weighted) fraction of vertices mapped onto a particular color i ∈ [ q ] is near to a i . More precisely, we consider configurations φ in
$$\Omega _ { \mathfrak { a }, \varepsilon } ( G ) = \left \{ \phi \colon [ q ] \to V ( G ) \, \colon \left | \frac { \alpha _ { \phi ^ { - 1 } ( \{ i \} ) } ( G ) } { \alpha _ { G } } - a _ { i } \right | \leq \varepsilon \text{ for all } i \in [ q ] \right \}.$$
On Ω a ,ε ( G ) we then define a probability distribution
$$\mu _ { G, J } ^ { ( \mathfrak { a }, \varepsilon ) } ( \phi ) = \frac { 1 } { Z _ { G, J } ^ { ( \mathfrak { a }, \varepsilon ) } } e ^ { - | V ( G ) | E _ { \phi } ( G, J ) },$$
where Z ( a ,ε ) G,J is the normalization factor
(2.10)
$$Z _ { G, J } ^ { ( \mathbf a, \varepsilon ) } = \sum _ { \phi \in \Omega _ { \mathbf a, \varepsilon } ( G ) } e ^ { - | V ( G ) | E _ { \phi } ( G, J ) }.$$
4 To see why, note that if ( α, β ) ∈ S ∞ q , then
$$d _ { 1 } ( ( \alpha, \beta ), \mathcal { S } _ { q } ( G _ { n } ) ) \leq d _ { 1 } ^ { H f } ( \mathcal { S } _ { q } ^ { \infty }, S _ { q } ( G _ { n } ) ) \to 0,$$
while if d 1 (( α, β ) , S q ( G n )) → 0, then combining this limit with d Hf 1 ( S ∞ q , S q ( G n )) → 0 and the fact that S ∞ q is closed shows that ( α, β ) ∈ S ∞ q .
The distribution µ ( a ,ε ) G,J is usually called the microcanonical Gibbs distribution of the model J on G , and Z ( a ,ε ) G,J is called the microcanonical partition function .
In this paper, we will not analyze the particular properties of the distribution µ ( a ,ε ) G,J , but we will be interested in the normalization factor, or more precisely its normalized logarithm
$$F _ { \mathfrak { a }, \varepsilon } ( G, J ) = - \frac { 1 } { | V ( G ) | } \log Z _ { G, J } ^ { ( \mathfrak { a }, \varepsilon ) },$$
which is called the microcanonical free energy . We will also be interested in the dominant term contributing to Z ( a ,ε ) G,J , or more precisely its normalized logarithm, the microcanonical ground state energy
$$E _ { \mathfrak { a }, \varepsilon } ( G, J ) = \min _ { \phi \in \Omega _ { \mathfrak { a }, \varepsilon } ( G ) } E _ { \phi } ( G, J ).$$
glyph[negationslash]
Note that the energy E φ ( G,J ) has been normalized in such a way that | E φ ( G,J ) | ≤ ‖ J ‖ ∞ (where ‖ J ‖ ∞ = max i,j ∈ [ q ] | J ij | ), and Ω a ,ε ( G ) = ∅ as long as ε ≥ α max ( G /α ) G , implying that under this condition, Z ( a ,ε ) G,J ≥ e -| V ( G E ) | a ,ε ( G,J ) ≥ e -| V ( G ) |‖ J ‖ ∞ . Thus, for fixed J the microcanonical energies and free energies are of order one.
Definition 2.5. Let ( G n n ) ≥ 0 be a sequence of weighted graphs. We say that
- (i) ( G n n ) ≥ 0 has convergent microcanonical ground state energies if the limit
$$E _ { \mathfrak { a } } ( J ) = \lim _ { \varepsilon \to 0 } \lim \sup _ { n \to \infty } E _ { \mathfrak { a }, \varepsilon } ( G, J ) = \lim _ { \varepsilon \to 0 } \lim \inf _ { n \to \infty } E _ { \mathfrak { a }, \varepsilon } ( G, J )$$
exists for all q ∈ N , a ∈ glyph[triangle] q , and symmetric J ∈ R q × q , and (ii) ( G n n ) ≥ 0 has convergent microcanonical free energies if the limit
$$F _ { \mathbf a } ( J ) = \lim _ { \varepsilon \to 0 } \lim \sup _ { n \to \infty } F _ { \mathbf a, \varepsilon } ( G, J ) = \lim _ { \varepsilon \to 0 } \lim \inf _ { n \to \infty } F _ { \mathbf a, \varepsilon } ( G, J )$$
exists for all q ∈ N , a ∈ glyph[triangle] q , and symmetric J ∈ R q × q .
Recall that the microcanonical ground state energy describes the largest term contributing to the microcanonical partition function Z ( a ,ε ) G,J . Using the fact that this partition function contains at least one and at most q | V ( G ) | terms, we will see that a scaling argument shows that convergence of the microcanonical free energies implies convergence of the microcanonical ground state energies. On the other hand, the energy of a configuration φ can be expressed in terms of the quotient G/φ as
$$E _ { \phi } ( G, J ) = - \langle \beta ( G / \phi ), J \rangle,$$
where
$$\langle \beta, J \rangle = \sum _ { i, j } \beta _ { i j } J _ { i j }.$$
Using this identity, we will express the microcanonical ground state energy as a minimum over quotients, which in turn can be used to show that convergence of quotients implies convergence of the microcanonical ground state energies.
The following theorem gives a precise statement of these facts. We will restate the theorem as part of Lemma 3.2 and Theorem 3.3 and prove it in Section 4.
Theorem 2.6. Let q ∈ N and let ( G n n ) ≥ 0 be a sequence of weighted graphs.
- (i) If S q ( G n ) converges to a closed set S ∞ q in the Hausdorff metric, then the limit (2.13) exists for all a ∈ glyph[triangle] q and all symmetric J ∈ R q × q and can be expressed as
$$E _ { \mathfrak { a } } ( J ) = - \max _ { ( \alpha, \beta ) \in \mathcal { S } _ { q } ^ { \infty } \cap \mathcal { S } _ { \mathfrak { a } } } \langle \beta, J \rangle.$$
- (ii) Let a ∈ glyph[triangle] q . If | V ( G n ) | →∞ and the limit (2.14) exists for all symmetric J ∈ R q × q , then the limit (2.13) exists for all such J and
$$E _ { \mathbf a } ( J ) = \lim _ { \lambda \rightarrow \infty } \frac { 1 } { \lambda } F _ { \mathbf a } ( \lambda J ).$$
Remark 2.7 . Definition 2.5 differs from that given in [8] for dense graphs in that we are taking the double limit of first sending n → ∞ and then sending ε → 0, rather than a single limit with an n -dependent ε = ε n . (In [8], ε n was chosen to be α max ( G n ) /α G n , even though all theorems involving the microcanonical free energies required the additional assumption that G n has vertex weights one, corresponding to ε n = 1 / V | ( G n ) ). | While there is some merit to the simplicity of a single limit, here we decided to follow the spirit of the definitions from mathematical statistical physics, where the formulation of a double limit is standard; it is also more consistent with the double limits usually taken in the theory of large deviations, where an n -dependent ε usually makes no sense.
However, the two definitions are equivalent if G n is dense with bounded edge weights and vertex weights one (this follows from Theorem 2.15 below, because such graphs are L ∞ upper regular). Thus, as far as the results of [8] are concerned, there is no difference between the two definitions.
- 2.5. Large deviation convergence. As we have seen in the last section, the quotients of a graph G provide enough information to calculate the microcanonical ground state energies (2.12), since the quotients tell us which energies E φ ( G,J ) can be realized. However, to calculate the microcanonical free energies (2.11) we need to know a little more, namely how often a term with given energy appears in the sum (2.10).
This leads to the notion of large deviation convergence (LD convergence), which was first introduced in the context of bounded degree graphs [4], where it turned out to be strictly stronger than convergence of quotients. Roughly speaking, this notion codifies how often a given quotient ( α, β ) ∈ S q ( G ) appears in a sum of the form (2.10). Or, put differently, it specifies the probability that for a uniformly random map φ : V ( G ) → [ q ], the quotient G/φ is approximately equal to ( α, β ). The precise definition is as follows:
Definition 2.8. Let q ∈ N , let ( G n n ) ≥ 0 be a sequence of weighted graphs, and let P q,G n be the probability distribution of G /φ n when φ : V ( G n ) → [ q ] is chosen uniformly at random. We say that ( G n n ) ≥ 0 is q -LD convergent if | V ( G n ) | → ∞ and
$$& \lim _ { \varepsilon \to 0 } \liminf _ { n \to \infty } \frac { \log \mathcal { P } _ { q, G _ { n } } \left [ d _ { 1 } ( ( \alpha, \beta ), G _ { n } / \phi ) \leq \varepsilon \right ] } { | V ( G _ { n } ) | } \\ & \quad = \lim _ { \varepsilon \to 0 } \lim _ { n \to \infty } \frac { \log \mathcal { P } _ { q, G _ { n } } \left [ d _ { 1 } ( ( \alpha, \beta ), G _ { n } / \phi ) \leq \varepsilon \right ] } { | V ( G _ { n } ) | }$$
and say it is q -LD convergent with rate function I q : S q → [0 , ∞ ] if the above limit is equal to -I q (( α, β )). We say that ( G n n ) ≥ 0 is LD convergent if it is q -LD convergent for all q ∈ N .
The following theorem states that LD convergence is at least as strong as convergence of quotients and convergence of the microcanonical free energies. We prove it in Sections 4.3 and 4.4.
Theorem 2.9. Let q ∈ N and let ( G n n ) ≥ 0 be a sequence of weighted graphs. If ( G n n ) ≥ 0 is q -LD convergent with rate function I q , then the following hold:
- (i) The sets of quotients S q ( G n ) converge to the closed set
$$\mathcal { S } _ { q } ( I _ { q } ) = \{ ( \alpha, \beta ) \in \mathcal { S } _ { q } \colon I _ { q } ( ( \alpha, \beta ) ) < \infty \}$$
in the Hausdorff metric.
- (ii) For all a ∈ glyph[triangle] q and all symmetric J ∈ R q × q , the microcanonical free energies converge to
$$F _ { \mathbf a } ( I _ { q }, J ) = \inf _ { ( \alpha, \beta ) \in \mathcal { S } _ { \mathbf a } } \Big ( - \langle \beta, J \rangle + I _ { q } ( ( \alpha, \beta ) ) \Big ) - \log q.$$
2.6. Limiting expressions for convergent sequences of graphs. The results stated so far, namely Theorems 2.6 and 2.9, raise the question of whether the four notions of convergence considered in these theorems are equivalent. They also raise the question of whether the limits of the quotients, microcanonical ground state energies, and free energies as well as the rate functions I q can be expressed in terms of a limiting graphon. It turns out that the answers to these two questions are related, and that we have equivalence if we postulate convergence to a graphon W ∈ L 1 .
We need some definitions. All of them rely on the notion of a fractional partition of [0 , 1] into q classes (briefly, a fractional q -partition ), which we define as a q -tuple of measurable functions ρ , . . . , ρ 1 q : [0 , 1] → [0 , 1] such that ρ 1 ( x )+ · · · + ρ q ( x ) = 1 for all x ∈ [0 , 1]. We denote the set of fractional q -partitions by FP q . To each fractional partition ρ ∈ FP q , we assign a weight vector α ρ ( ) = ( α 1 ( ρ , . . . , α ) q ( ρ )) ∈ glyph[triangle] q and an entropy Ent ( ρ ) ∈ [0 , log q ] by setting
$$\alpha _ { i } ( \rho ) = \int _ { 0 } ^ { 1 } \rho _ { i } ( x ) \, d x$$
and
$$\text{and} \quad \text{Ent} ( \rho ) = \int _ { 0 } ^ { 1 } \text{Ent} _ { x } ( \rho ) \, d x \text{ \ with \ } \text{Ent} _ { x } ( \rho ) = - \sum _ { i = 1 } ^ { q } \rho _ { i } ( x ) \log \rho _ { i } ( x )$$
(with 0 log 0 = 0). Let
$$\widehat { \mathcal { S } } _ { q } = \left \{ ( \alpha, \beta ) \in [ 0, 1 ] ^ { q } \times \mathbb { R } ^ { q \times q } \colon \sum _ { i \in [ q ] } \alpha _ { i } = 1 \right \}$$
(in comparison with the definition (2.6) of S q , we do not restrict β ). Given a graphon W and a fractional q -partition ρ ∈ FP q , we then define the quotient W/ρ to be the pair ( α, β ) ∈ S ̂ q where
$$\alpha _ { i } ( W / \rho ) = \alpha _ { i } ( \rho )$$
and
$$\beta _ { i j } ( W / \rho ) = \int _ { [ 0, 1 ] ^ { 2 } } \rho _ { i } ( x ) \rho _ { j } ( y ) W ( x, y ) \, d x \, d y$$
for i, j ∈ [ q ]. We call W/ρ a fractional q -quotient of W . Let ̂ S q ( W ) denote the set of all fractional q -quotients of W , and for a ∈ glyph[triangle] q , let ̂ S a ( W ) denote the set of pairs in ̂ S q ( W ) whose first coordinate equals a . It will be shown in Proposition 5.5 that ̂ S q ( W ) is compact.
Next we define the microcanonical ground state energies and free energies of a graphon W . Given an integer q ≥ 1 and a symmetric matrix J ∈ R q × q , we define the energy of a fractional partition ρ ∈ FP q to be
$$\mathcal { E } _ { \rho } ( W, J ) = - \sum _ { i, j } J _ { i j } \, \int _ { [ 0, 1 ] ^ { 2 } } \rho _ { i } ( x ) \rho _ { j } ( y ) W ( x, y ) \, d x \, d y.$$
For a ∈ glyph[triangle] q , the microcanonical ground state energy is defined as
$$\mathcal { E } _ { \mathfrak a } ( W, J ) = \inf _ { \rho \colon \alpha ( \rho ) = \mathfrak a } \mathcal { E } _ { \rho } ( W, J ),$$
while the microcanonical free energy is defined as
$$\mathcal { F } _ { \mathfrak a } ( W, J ) = \inf _ { \rho \colon \alpha ( \rho ) = \mathfrak a } \left ( \mathcal { E } _ { \rho } ( W, J ) - \text{Ent} ( \rho ) \right ).$$
The infima in these equations are over all fractional q -partitions of [0 , 1] such that α ρ ( ) = a . Note that all these quantities are well defined because 0 ≤ Ent x ( ρ ) ≤ log q and
$$| \mathcal { E } _ { \rho } ( W, J ) | \leq \| J \| _ { \infty } \| W \| _ { 1 }.$$
Finally, the LD rate function I q ( F, W ) is defined as
$$I _ { q } ( ( \alpha, \beta ), W ) = \inf _ { \rho \in F P _ { q } \colon W / \rho = ( \alpha, \beta ) } ( \log q - \text{Ent} ( \rho ) ),$$
Note that I q (( α, β ) , W ) ∈ [0 , log q ] if ( α, β ) ∈ S ̂ q ( W ) and I q (( α, β ) , W ) = ∞ if ( α, β ) / ∈ S ̂ q ( W ).
We are now ready to state the main theorem of this paper. Recall that graphons are assumed to be L 1 (and not necessarily L ∞ , as in some papers in the literature).
Theorem 2.10. Let W be a graphon, and let ( G n n ) ≥ 0 be a sequence of weighted graphs with no dominant nodes, in the sense that α max ( G n ) /α G n → 0 . Then the following statements are equivalent:
- (i) ( G n n ) ≥ 0 converges to W in metric.
- (ii) For all q ∈ N , S q ( G n ) →S ̂ q ( W ) in the Hausdorff metric d Hf 1 .
- (iii) The microcanonical ground state energies of ( G n n ) ≥ 0 converge to those of W .
If all the vertices of G n have weight one, then the following two statements are also equivalent to (i):
- (iv) ( G n n ) ≥ 0 is LD convergent with rate function I q = I q ( · , W ) .
- (v) The microcanonical free energies of ( G n n ) ≥ 0 converge to those of W .
We prove this theorem in Section 6.
2.7. Uniform upper regularity. It is natural to ask whether one can state Theorem 2.10 without reference to the limiting graphon W . It turns out that the answer is yes, and in fact this reformulation (Theorem 2.15) will play a key role in the proof. To state this theorem, we need the notion of upper regularity, which first arose in our study of subsequential metric convergence in [3] and plays a key role both in that paper and in this one.
To define this concept, we define the L p norm of a weighted graph G to be
$$\| G \| _ { p } = \left ( \sum _ { x, y \in V ( G ) } \frac { \alpha _ { x } ( G ) \alpha _ { y } ( G ) } { \alpha _ { G } ^ { 2 } } | \beta _ { x y } ( G ) | ^ { p } \right ) ^ { 1 / p },$$
and for p = ∞ we set
$$\| G \| _ { \infty } = \max _ { \substack { x, y \in V ( G ) \\ \alpha _ { x } ( G ), \alpha _ { y } ( G ) > 0 } } | \beta _ { x y } ( G ) |.$$
As we already have seen in Section 2.2, when studying graph convergence for sparse graphs, it is natural to reweight the edge weights by 1 ‖ G ‖ 1 to obtain a weighted graph which does not go to zero for trivial reasons. In order to control the now possibly large entries of the adjacency matrix of the weighted graph 1 ‖ G ‖ 1 G , one might want to require the L p norm of 1 ‖ G ‖ 1 G to be bounded, but this turns out to be too restrictive. Instead, we will use a weaker condition, which requires the L p norm of 1 ‖ G ‖ 1 G to be bounded 'on average,' at least when the averages are taken over sufficiently large blocks. To make this precise, we need some additional notation.
Given a weighted graph G and a partition P = { V , . . . , V 1 q } of V ( G ) into disjoint sets V , . . . , V 1 q , we define G P to be the weighted graph with the same vertex weights as G and edge weights which are defined by averaging over the blocks V i × V j , suitably weighted by the vertex weights:
$$\beta _ { x y } ( G _ { \mathcal { P } } ) = \frac { 1 } { \alpha _ { V _ { i } } ( G ) \alpha _ { V _ { j } } ( G ) } \sum _ { ( u, v ) \in V _ { i } \times V _ { j } } \alpha _ { u } ( G ) \alpha _ { v } ( G ) \beta _ { u v } ( G )$$
if ( x, y ) ∈ V i × V j and α V i ( G α ) V j ( G > ) 0, while we set β xy ( G P ) = 0 if either x or y lie in a block V k ( G ) with total node weight α V k ( G ) = 0.
Definition 2.11. Let G be a weighted graph, let C, η > 0, and let p > 1. We say that G is ( C, η ) -upper L p regular if α max ( G ) ≤ ηα G and
$$\| G _ { \mathcal { P } } \| _ { p } \leq C \| G \| _ { 1 }$$
for all partitions P = { V , . . . , V 1 q } for which min i α V i ≥ ηα G . We say that a sequence of graphs ( G n n ) ≥ 0 is C -upper L p regular if there exists a sequence η n → 0 such that G n is ( C, η n )-upper regular, and we say that ( G n n ) ≥ 0 is L p upper regular if there exists a C < ∞ such that ( G n n ) ≥ 0 is C -upper L p regular.
The definition of L 1 upper regularity always holds vacuously, but the following definition of uniform upper regularity turns out to be the correct L 1 analogue, as described in Appendix C of [3]. It is closely related to the notion of uniform integrability of a set of graphons (see Section 5.2), and it is the notion we will need in this paper.
Definition 2.12. Let η > 0 and let K : (0 , ∞ → ) (0 , ∞ ) be any function. We say that a weighted graph G is ( K,η ) -upper regular if α max ( G ) ≤ ηα G and
$$\sum _ { x, y \in V ( G ) } \frac { \alpha _ { x } ( G ) \alpha _ { y } ( G ) } { \alpha _ { G } ^ { 2 } } \frac { | \beta _ { x y } ( G _ { \mathcal { P } } ) | } { \| G \| _ { 1 } } \mathbf 1 _ { | \beta _ { x y } ( G _ { \mathcal { P } } ) | \geq K ( \varepsilon ) \| G \| _ { 1 } } \leq \varepsilon$$
for all ε > 0 and all partitions P = { V , . . . , V 1 q } for which min i α V i ≥ ηα G . We say that a sequence of graphs ( G n n ) ≥ 0 is K -upper regular if there exists a sequence η n → 0 such that G n is ( K,η n )-upper regular, and we say that ( G n n ) ≥ 0 is uniformly upper regular if there exists a function K : (0 , ∞ → ) (0 , ∞ ) such that ( G n n ) ≥ 0 is K -upper regular.
Note that the properties of L p upper regularity and uniform upper regularity require ( G n n ) ≥ 0 to have no dominant nodes , a property we already encountered in Theorem 2.10. One of the main results of [3] is the following theorem.
Theorem 2.13 (Theorem C.7 in [3]) . Let ( G n n ) ≥ 0 be a uniformly upper regular sequence of weighted graphs. Then ( G n n ) ≥ 0 contains a subsequence that is convergent in metric. Furthermore, if ( G n n ) ≥ 0 is convergent in metric, then there exists a graphon W such that G n converges to W in metric.
Conversely, it was shown in [3] that every sequence of weighted graphs which converges in metric to a graphon and has no dominant nodes must be upper regular. The precise statement is given by the following theorem, which follows immediately from Corollary 2.11 and Proposition C.5 in [3].
Theorem 2.14 ([3]) . Let ( G n n ) ≥ 0 be a sequence of weighted graphs without dominant nodes, and assume that G n converges to some graphon W in metric. Then ( G n n ) ≥ 0 is uniformly upper regular. If W is in L p , then ( G n n ) ≥ 0 is L p -upper regular.
A uniformly upper regular sequence of simple graphs must have unbounded average degree, by Proposition C.15 in [3]. This corresponds to the fact that graphons are not the appropriate limiting objects for graphs with bounded average degree (although they apply to all other sparse graphs).
Returning to the subject of this paper, the question of whether the five versions of convergence defined in Sections 2.2 through 2.5 are equivalent, we are now ready to state our results without reference to a limiting graphon.
Theorem 2.15. Let ( G n n ) ≥ 0 be a uniformly upper regular sequence of weighted graphs. Then the following three statements are equivalent:
- (i) ( G n n ) ≥ 0 is convergent in metric.
- (ii) ( G n n ) ≥ 0 has convergent quotients.
- (iii) ( G n n ) ≥ 0 has convergent microcanonical ground state energies.
If all the vertices of G n have weight one, then the following two statements are also equivalent to (i):
- (iv) ( G n n ) ≥ 0 is LD convergent.
- (v) ( G n n ) ≥ 0 has convergent microcanonical free energies.
Note that by Theorems 2.6 and 2.9, we already know that (iv) implies both (v) and (ii), and that both (v) and (ii) imply (iii); in fact, we need neither node weights one, nor the assumption of upper regularity. So the important part of this theorem is that under the assumption of uniform upper regularity, convergence in metric implies convergence of quotients (and LD convergence, if we assume node
weights one), and convergence of the microcanonical ground state energies implies convergence in metric. We prove Theorem 2.15 in Section 6.
One may want to know whether the assumption of upper regularity is actually necessary for these conclusions to hold. The answer is yes, by the following example.
Example 2.16. Let c n ∈ N be such that c n →∞ and c n /n → 0 as n →∞ , and let G n be the disjoint union of a complete graph on c n nodes with n -c n isolated nodes. Then ( G n n ) ≥ 0 is LD convergent (and hence has convergent quotients, microcanonical free energies and microcanonical ground state energies); see Section 3.3.6 below. However, ( G n n ) ≥ 0 is not a Cauchy sequence in the normalized cut metric δ glyph[square] , norm from (2.3) and hence does not converge to any graphon in metric (see the proof of Proposition 2.12(a) in [3]).
The following theorem states that convergence of the quotients, microcanonical ground state energies, or microcanonical free energies to those of a graphon W all imply upper regularity, as does LD convergence with a rate function I q ( · , W ) given in terms of a graphon W . It is the analogue of Theorem 2.14 for these notions of convergence.
Theorem 2.17. Let ( G n n ) ≥ 0 be a sequence of weighted graphs with no dominant nodes, and let W be a graphon. Then any of the following conditions implies that ( G n n ) ≥ 0 is uniformly upper regular.
- (i) The microcanonical ground state energies of ( G n n ) ≥ 0 converge to those of W .
- (ii) For all q ∈ N , S q ( G n ) →S ̂ q ( W ) in the Hausdorff metric d Hf 1 .
- (iii) The microcanonical free energies of ( G n n ) ≥ 0 converge to those of W .
- (iv) ( G n n ) ≥ 0 is LD convergent with rate function I q = I q ( · , W ) .
Note that the first two assertions in this theorem already follow by combining Theorem 2.14 with Theorem 2.10(i)-(iii). However, this is not how our proofs of Theorems 2.10 and 2.17 proceed. Instead of proving Theorem 2.10 directly, we use uniform upper regularity to prove Theorem 2.15 in Section 6. Then Theorem 2.17 is exactly what we need to deduce Theorem 2.10 from Theorem 2.15, and we prove Theorem 2.17 in Section 7.
## 3. Further definitions, remarks, and examples
3.1. Convergence of free energies and ground state energies. In addition to the microcanonical quantities introduced in Section 2.4, statistical physicists often analyze the unrestricted probability measure
$$\mu _ { G, J, h } ( \phi ) = \frac { 1 } { Z _ { G, J, h } } e ^ { - | V ( G ) | E _ { \phi } ( G, J ) + | V ( G ) | \langle h, \alpha ( G / \phi ) \rangle },$$
where h is a vector in R q called the magnetic field ,
$$\langle h, \alpha \rangle = \sum _ { i \in [ q ] } h _ { i } \alpha _ { i },$$
and Z G,J,h is the normalization factor
$$Z _ { G, J, h } = \sum _ { \phi \colon V ( G ) \to [ q ] } e ^ { - | V ( G ) | E _ { \phi } ( G, J ) + | V ( G ) | \langle h, \alpha ( G / \phi ) \rangle },$$
usually called the partition function . The normalized logarithm of the partition function is the free energy
$$F ( G, J, h ) = - \frac { 1 } { | V ( G ) | } \log Z _ { G, J, h }$$
of the model ( J, h ) on G , and the maximizer in the sum (3.1), or more precisely its normalized logarithm, is the ground state energy
$$E ( G, J, h ) = \min _ { \phi \colon [ q ] \to V ( G ) } \left ( E _ { \phi } ( G, J, h ) - \langle h, \alpha ( G / \phi ) \rangle \right ).$$
## Definition 3.1.
- (i) ( G n n ) ≥ 0 has convergent ground state energies if the limit
$$( 3. 2 )$$
$$E ( J, h ) = \lim _ { n \to \infty } E ( G _ { n }, J, h )$$
exists for all q ∈ N , symmetric J ∈ R q × q , and h ∈ R q .
(ii) ( G n n ) ≥ 0 has convergent free energies if the limit
$$F ( J, h ) = \lim _ { n \to \infty } F ( G _ { n }, J, h )$$
$$\begin{matrix} \end{matrix}$$
exists for all q ∈ N , symmetric J ∈ R q × q , and h ∈ R q .
These notions are implied by the microcanonical versions, and convergence of free energies implies convergence of ground state energies. This is the content of the following lemma, which we will prove in Section 4.1. Note that part (iii) is a restatement of Theorem 2.6(ii).
Lemma 3.2. Let ( G n n ) ≥ 0 be a sequence of weighted graphs with | V ( G n ) | → ∞ , and let q ∈ N . Then the following hold:
- (i) Let J be a symmetric matrix in R q × q , and assume that the limit (2.14) exists for all a ∈ glyph[triangle] q . Then the limit (3.3) exists for all h ∈ R q , and
$$F ( J, h ) = \inf _ { \mathfrak { a } \in \triangle _ { q } } \left ( F _ { \mathfrak { a } } ( J ) - \langle \mathfrak { a }, h \rangle \right ).$$
- (ii) Let J be a symmetric matrix in R q × q , and assume that the limit (2.13) exists for all a ∈ glyph[triangle] q . Then the limit (3.2) exists for all h ∈ R q , and
$$E ( J, h ) = \inf _ { \mathfrak { a } \in \triangle _ { q } } \left ( E _ { \mathfrak { a } } ( J ) - \langle \mathfrak { a }, h \rangle \right ).$$
- (iii) Let a ∈ glyph[triangle] q , and assume that the limit (2.14) exists for all symmetric J ∈ R q × q . Then the limit (2.13) exists for all such J , and
$$E _ { \mathbf a } ( J ) = \lim _ { \lambda \rightarrow \infty } \frac { 1 } { \lambda } F _ { \mathbf a } ( \lambda J ).$$
- (iv) Assume that the limit (3.3) exists for all h ∈ R d and all symmetric J ∈ R q × q . Then the limit (3.2) exists for all h ∈ R d and all symmetric J ∈ R q × q , and
$$E ( J, h ) = \lim _ { \lambda \to \infty } \frac { 1 } { \lambda } F ( \lambda J, \lambda h ).$$
Convergence of the ground state and free energies is strictly weaker than that of the microcanonical versions. See Section 3.3.5 for an example.
On the other hand, we can use (2.15) to express both the microcanonical ground state energies E a ,ε ( G,J ) and the unrestricted ground state energies E G,J,h ( ) as minima over quotients. Using this fact, it is not hard to show that convergence
of quotients implies convergence of the ground state energies as well as the microcanonical ground state energies. This is the content of the following theorem, which again holds for an arbitrary sequence, with no assumption about upper regularity. We prove the theorem (which encompasses Theorem 2.6(i)) in Section 4.2.
Theorem 3.3. Let q ∈ N and let ( G n n ) ≥ 0 be a sequence of weighted graphs such that S q ( G n ) converges to a closed set S ∞ q in the Hausdorff metric. Then the limit (2.13) exists for all a ∈ glyph[triangle] q and all symmetric J ∈ R q × q and can be expressed as
$$E _ { \mathbf a } ( J ) = - \max _ { ( \alpha, \beta ) \in \mathcal { S } _ { q } ^ { \infty } \cap \mathcal { S } _ { \mathbf a } } \langle \beta, J \rangle.$$
and the limit (3.2) exists for all symmetric J ∈ R q × q and all h ∈ R q and can be expressed as
$$E ( J, h ) = \min _ { ( \alpha, \beta ) \in \mathcal { S } _ { q } ^ { \infty } } \left ( - \langle \beta, J \rangle - \langle \alpha, h \rangle \right ),$$
Much as in Section 2.6, we can write down limiting expressions for a graphon W . The ground state energy of the model ( J, h ) on W is
$$\mathcal { E } ( W, J, h ) = \inf _ { \rho \in \text{FP} _ { q } } \left ( \mathcal { E } _ { \rho } ( W, J ) - \sum _ { i } h _ { i } \int _ { [ 0, 1 ] } \rho _ { i } ( x ) \, d x \right )$$
and its free energy is defined as
$$\mathcal { F } ( W, J, h ) = \inf _ { \rho \in \text{FP} _ { q } } \left ( \mathcal { E } _ { \rho } ( W, J ) - \sum _ { i } h _ { i } \int _ { [ 0, 1 ] } \rho _ { i } ( x ) \, d x - \text{Ent} ( \rho ) \right ).$$
It follows from Lemma 3.2 and Theorem 2.10 that if ( G n n ) ≥ 0 has no dominant nodes and converges to W in metric, then its ground state energies converge to those of W , and if all the vertices of G n have weight one, then the free energies also converge to those of W .
## 3.2. LD convergence.
Remark 3.4 . It is not hard to see that ( G n n ) ≥ 0 is q -LD convergent if and only if P q,G n obeys a large deviation principle with speed | V ( G n ) , | i.e., if there exists a lower semicontinuous function I q : S q → [0 , ∞ ] such that
$$& \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \\ & \quad - \inf _ { ( \alpha, \beta ) \in \tilde { S } } I _ { q } ( ( \alpha, \beta ) ) \leq \liminf _ { n \to \infty } \frac { \log \mathcal { P } _ { q, G _ { n } } \left [ G _ { n } / \phi \in \tilde { S } \right ] } { | V ( G _ { n } ) | } \\ & \quad \leq \lim _ { n \to \infty } \frac { \log \mathcal { P } _ { q, G _ { n } } \left [ G _ { n } / \phi \in \tilde { S } \right ] } { | V ( G _ { n } ) | } \leq - \inf _ { ( \alpha, \beta ) \in \tilde { S } } I _ { q } ( ( \alpha, \beta ) ) \\ & \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdelta \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdcl \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdsclimfrak { } }$$
for all sets S ⊆ S q . Here ¯ denotes the closure of S S and ˚ its interior. S
Indeed, assume that (3.4) holds for some lower semicontinuous function I q : S q → [0 , ∞ ]. By the lower semicontinuity of I q ,
$$I _ { q } ( ( \alpha, \beta ) ) = \lim _ { \varepsilon \to 0 } \inf \{ I _ { q } ( ( \alpha ^ { \prime }, \beta ^ { \prime } ) ) \colon d _ { 1 } ( ( \alpha, \beta ), ( \alpha ^ { \prime }, \beta ^ { \prime } ) ) < \varepsilon \},$$
which implies (2.16) when inserted into (3.4). It turns out that (2.16) is also sufficient for (3.4) to hold. Indeed, under the assumption that the underlying metric space is compact (which is the case here), the equality of the two limits in (2.16) implies that P q,G n obeys a large deviation principle with rate function given by I q ; see, for example, Theorem 4.1.11 in [9] for the proof.
- 3.3. Examples. In this section we give some examples of convergent graph sequences, as well as a few counterexamples in which the equivalences in Theorem 2.15 fail (of course because uniform upper regularity does not hold).
- 3.3.1. Erd˝ os-R´nyi e random graphs. The simplest example of a uniformly upper regular sequence-in fact an L ∞ -upper regular sequence-is the standard Erd˝so R´ enyi random graphs G n,p obtained by connecting each pair of distinct vertices in [ n ] independently with probability p . Here p can depend on n , as long as pn →∞ as n →∞ . Under this condition, G n,p converges with probability one to the constant graphon W = 1. This can proved in several ways, for example by showing that in expectation all the quotients in S q ( G n,p ) converge to the corresponding quotients in ̂ S q ( W ) and proving concentration with the help of Azuma's inequality.
- 3.3.2. Stochastic block models. Next we consider the block models obtained as follows. Fix k ∈ N , a symmetric matrix B = ( b ij ) i,j ∈ [ k ] with entries b ij ≥ 0 satisfying k -2 ∑ i,j b ij = 1, and a target density ρ n ≤ 1 / max b ij . Divide [ n ] into k blocks V , . . . , V 1 k of equal size (or, in the case where n is not divisible by k of sizes differing by at most 1) and define p uv = ρ n ij b if ( u, v ) ∈ V i × V j . Then we connect vertices u and v with probability p uv . If nρ n →∞ as n →∞ , then the resulting graph converges with probability one to the step function W that is equal to b ij on the block ( i -1 n , i n ] × ( j -1 n , j n ]. The proof can again be obtained by proving that the quotients converge in expectation, followed by a concentration argument.
glyph[negationslash]
- 3.3.3. Power law graphs. Starting again with the vertex set [ n ], connect i = j with probability min(1 , n β ( ij ) -α ), where 0 < α < 1 and 0 ≤ β < 2 α . In other words, the expected degree distribution follows an inverse power law with exponent α , while the n β scaling factor ensures that the probabilities do not become too small. If β > 2 α -1, then the expected number of edges is superlinear, and a similar argument to the one used in the above two examples shows convergence, this time to a graphon that is not in L ∞ , namely W x,y ( ) = (1 -α ) 2 ( xy ) -α .
- 3.3.4. W -random graphs. Our fourth example provides a construction of a sequence ( G n n ) ≥ 0 of simple graphs that converge to a given graphon W with non-negative entries W x,y ( ) ≥ 0. Normalizing W so that ∫ [0 1] , 2 W = 1 and fixing a target density ρ n , we proceed by first choosing n i.i.d. variables x , . . . x 1 n uniformly in [0 , 1], and then defining a random graph G n ( W,ρ n ) on { 1 , . . . , n } by connecting each pair { i, j } ∈ ( [ n ] 2 ) independently with probability min { 1 , ρ n W x ,y ( i i ) } . Assuming that ρ n → 0 and nρ n → ∞ , the graphs G n converge to W under the normalized cut metric with probability one, by Theorem 2.14 in [3]. If W is a step function, this is more or less equivalent to the convergence of stochastic block models, while for general graphons W , one can proceed by first approximating W by a step function.
- 3.3.5. Convergence of free energies without convergence of microcanonical free energies. Our next example is a generalization of Example 6.3 from [8] to the sparse setting, and is based on the observation that for an arbitrary sequence of graphs G n , the free energies of G n and a disjoint union of G n with itself are identical (this follows from the fact that for two disjoint graphs G and G ′ , the partition function on G ∪ G ′ factors into that of G times that of G ′ ). If we take G n to be equal to G n,p if n is odd, and equal to a disjoint union of two copies of G n,p if n is even, then we get convergence of the free energies. By contrast, in the notions of convergence
from Theorem 2.10, the odd subsequence converges to W = 1, while the even one converges to the block graphon W ′ that is equal to 2 on [0 , 1 2] / 2 ∪ [1 / 2 1] , 2 and 0 elsewhere. In particular, the min-bisection of the even subsequence converges to zero, while the min-bisection of the odd sequence converges to 1 / 2. This shows that the microcanonical ground state energies are not convergent, which implies that the microcanonical free energies don't converge either.
3.3.6. LD convergence without metric convergence. This is Example 2.16 from Section 2.7, consisting of a graph G n that is the disjoint union of a complete graph on c n nodes with n -c n isolated nodes. A random q -quotient is then determined by how many elements of the clique there are in each part and how many elements of the non-clique. Calling these numbers b 1 , . . . , b q and a , . . . , a 1 q , we have b 1 + · · · + b q = c n and a 1 + · · · + a q = n -c n , and this occurs with probability
$$q ^ { - n } \binom { c _ { n } } { b _ { 1 }, \dots, b _ { q } } \binom { n - c _ { n } } { a _ { 1 }, \dots, a _ { q } }.$$
glyph[negationslash]
Everything else is determined from this data: α i = ( a i + b i ) /n β , ij = b b i j / c ( n ( c n -1)) if i = j , and β ii = b i ( b i -1) / c ( n ( c n -1)). If c n ∈ N is such that c n → ∞ and c n /n → 0 as n →∞ , then in the rate function, the choice of b 1 , . . . , b q gets wiped out by the choice of a , . . . , a 1 q , leading to LD convergence with rate function
$$I _ { q } ( ( \alpha, \beta ) ) = \log q + \sum _ { i = 1 } ^ { q } \alpha _ { i } \log \alpha _ { i }$$
as long as β ∈ R q × q satisfies β ii ≥ 0, β ij = √ β ii β jj , and ∑ i √ β ii = 1 (while I q (( α, β )) = ∞ otherwise). On the other hand, ( G n n ) ≥ 0 is not a Cauchy sequence in the normalized cut metric δ glyph[square] , norm from (2.3) and hence does not converge to any graphon in metric (see the proof of Proposition 2.12(a) in [3]).
3.3.7. Convergence of quotients without convergence of the microcanonical free energies. We close our example section with an example from [4] (Example 5 from that paper) which shows that without the assumption of upper regularity, convergence of quotients does not imply convergence of the microcanonical free energies, and hence does not imply LD-convergence either. Before stating this example, we note that whenever H n is a sequence of regular bipartite graphs, c n →∞ , and G n is the union of c n disjoint copies of H n , then the quotients of G n converge to the convex hull of the quotients of a graph consisting of a single edge. To see why, consider a map from the vertex set of G n into [ q ]. Since G n is regular, the corresponding quotient does not change if we replace G n by a disjoint union of | E G ( n ) | edges (and map each of the split vertices to the same element of [ q ] as its original vertex in G n ). Thus, the quotient is in the convex hull of the quotients of a single edge. On the other hand, each quotient of a single edge can be realized in the bipartite graph H n , showing that each quotient in the convex hull can be arbitrarily well approximated in G n if c n →∞ .
To get a sequence G n without convergent microcanonical free energies we specialize to the case where H n consists of a 4-cycle when n is even and a 6-cycle when n is odd. The free energies of G n are then equal to the free energies of the 4-cycle when n is even and those of the 6-cycle when n is odd. But it is easy to check that the 4-cycle has different free energies from a 6-cycle, implying that G n does not have convergent free energies, and hence does not have convergent microcanonical
free energies either (an alternative proof was given in [4], where it was used that G n does not converge in the sense of Benjamini and Schramm [1], which in turn is necessary for convergence of the free energies, as proved in [5]).
## 4. Convergence without the assumption of upper regularity
In this section, we consider general sequences of weighted graphs G n without any additional assumptions (except that G n has at least one edge with nonzero edge weight). We will prove Lemma 3.2, Theorem 3.3, and Theorem 2.9.
4.1. Free energies and ground state energies. In this section, we prove Lemma 3.2. We start with the proof of (i). To this end, we note that for all a ∈ glyph[triangle] q we have the lower bound
$$Z ( G _ { n }, J, h ) ^ { 1 / | V ( G _ { n } ) | } \geq e ^ { \langle \mathfrak { a }, h \rangle - \varepsilon \| h \| _ { 1 } } \left ( Z _ { G, J } ^ { ( \mathfrak { a }, \varepsilon ) } \right ) ^ { 1 / | V ( G _ { n } ) | },$$
from which we conclude that
$$\lim _ { n \to \infty } F ( G _ { n }, J, h ) \leq F _ { \mathbf a } ( J ) - \langle \mathbf a, h \rangle.$$
Since a ∈ glyph[triangle] q was arbitrary, this gives
$$\lim _ { n \to \infty } F ( G _ { n }, J, h ) \leq \inf _ { \mathfrak { a } \in \bigtriangleup _ { q } } \left ( F _ { \mathfrak { a } } ( J ) - \langle \mathfrak { a }, h \rangle \right ).$$
To get a matching lower bound, we use the fact that glyph[triangle] q can be covered by glyph[ceilingleft] 1 (2 ) / ε glyph[ceilingright] q ≤ ε -q cubes of the form ∏ q i =1 [ a i -ε, a i + ]. Explicitly, let ε glyph[triangle] ( ε ) q be the set of points a where each coordinate is an odd multiple of ε . Then
$$Z ( G _ { n }, J, h ) ^ { 1 / | V ( G _ { n$$
implying that
$$\begin{array} { l l$$
where in the second step, we used that the minimum is over a finite set. Let ε k be a sequence going to zero, let a k be the minimizer on the right hand side, and assume (by taking a subsequence, if necessary) that a k converges to some a . Let ˜ ε k = ε k + ‖ a -a k ‖ ∞ . Since F a k ,ε k ( G ,J n ) ≥ F a ,ε ˜ k ( G ,J n ),
$$\liminf _ { n \to \infty } F ( G _ { n }, J, h$$
Sending k →∞ , we conclude that
$$\liminf _ { n \to \infty } F ( G _ { n }, J, h$$
as desired.
The proof of (ii) starts from the observations that
$$\overset { \cdot } { E } _ { \mathbf a ^ { \prime },$$
for all a ′ ∈ glyph[triangle] and all ε > 0. Using these two bounds, the proof of (ii) is now identical to the proof of (i).
To prove (iii) and (iv), we note that the number of terms in (3.1) and (2.10) is at most q | V ( G ) | , implying that
$$F ( G, J, h ) \leq E ( G, J, h ) \leq F ( G, J, h ) + \log q$$
and
$$F _ { \mathfrak { a }, \varepsilon } ( G, J ) \leq E _ { \mathfrak { a }, \varepsilon } ( G, J ) \leq F _ { \mathfrak { a }, \varepsilon } ( G, J ) + \log q.$$
Rescaling J and h by a factor λ → ∞ , and using that both the energies and microcanonical energies are linear in λ , we obtain the claimed implications. glyph[square]
## 4.2. Convergence of quotients implies convergence of microcanonical ground state energies. In this section, we prove Theorem 3.3.
To this end, we use (2.15) to express the microcanonical ground state energies as
$$E _ { \mathfrak { a }, \varepsilon } ( G, J ) = - \max _ { ( \alpha, \beta ) \in \mathcal { S } _ { \mathfrak { a }, \varepsilon } ( G ) } \langle \beta, J \rangle.$$
where
$$\mathcal { S } _ { \mathbf a, \varepsilon } ( G ) = \mathcal { S } _ { q } ( G ) \cap \mathcal { S } _ { \mathbf a, \varepsilon } \ \text{ with } \ \mathcal { S } _ { \mathbf a, \varepsilon } = \left \{ ( \alpha, \beta ) \in \mathcal { S } _ { q } \colon \| \alpha - \mathbf a \| _ { \infty } \leq \varepsilon \right \}.$$
Proof of Theorem 3.3. In view of Lemma 3.2 it is enough to prove convergence of the microcanonical ground state energies.
Let ε > 0. Since S q ( G n ) is assumed to converge to S ∞ q , we can find an n 0 ∈ N such that
$$d _ { 1 } ^ { \text{Hf} } ( \mathcal { S } _ { q } ( G _ { n } ), \mathcal { S } _ { q } ^ { \infty } ) \leq \varepsilon \text{ \ for all \ } n \geq n _ { 0 }.$$
For n ≥ n 0 , choose ( α ( n ) , β ( n ) ) ∈ S a ,ε ( G n ) ⊆ S q ( G n ) such that
$$E _ { \mathfrak { a }, \varepsilon } ( G _ { n }, J ) = - \langle \beta ^ { ( n ) }, J \rangle,$$
and choose (˜ α ( n ) , β ˜ ( n ) ) ∈ S ∞ q such that d 1 ((˜ α ( n ) , β ˜ ( n ) ) , ( α ( n ) , β ( n ) )) ≤ ε . Then
$$E _ { \mathfrak { a }, \varepsilon } ( G _ { n }, J ) \geq - \langle \tilde { \beta } ^ { ( n ) }, J \rangle - \varepsilon \| J \| _ { \infty }.$$
Since | ˜ α ( n ) i -a i | ≤ | α ( n ) i -a i | + d 1 ((˜ α ( n ) , β ˜ ( n ) ) , ( α ( n ) , β ( n ) )) ≤ 2 , ε we have that (˜ α ( n ) , β ˜ ( n ) ) ∈ S a , 2 ε , proving in particular that
$$E _ { \mathfrak { a }, \varepsilon } ( G _ { n }, J ) \geq - \langle \tilde { \beta } ^ { ( n ) }, J \rangle - \varepsilon \| J \| _ { 1 } \geq - \sup _ { ( \alpha, \beta ) \in \mathcal { S } _ { q } ^ { \infty } \cap \mathcal { S } _ { \mathfrak { a }, 2 \varepsilon } } \langle \beta, J \rangle - \varepsilon \| J \| _ { \infty }.$$
Taking first the lim inf as n →∞ and then the limit ε → 0, this shows that
$$\lim _ { \varepsilon \to 0 } \liminf _ { n \to \infty } E _ { \mathfrak a, \varepsilon } ( G _ { n }, J ) \geq - \lim _ { \varepsilon \to 0 } \sup _ { ( \alpha, \beta ) \in \mathcal { S } _ { q } ^ { \infty } \cap \mathcal { S } _ { \mathfrak a, \varepsilon } } \langle \beta, J \rangle = - \max _ { ( \alpha, \beta ) \in \mathcal { S } _ { q } ^ { \infty } \cap \mathcal { S } _ { \mathfrak a } } \langle \beta, J \rangle,$$
where the final step is due to compactness. The proof of the matching upper bound
$$\lim _ { \varepsilon \to 0 } \lim _ { n \to \infty } \sup _ { n \to \infty } E _ { \mathbf a, \varepsilon } ( G _ { n }, J ) \leq - \lim _ { \varepsilon \to 0 } \sup _ { ( \alpha, \beta ) \in \mathcal { S } _ { q } ^ { \infty } \cap \mathcal { S } _ { \mathbf a, \varepsilon } } \langle \beta, J \rangle = - \max _ { ( \alpha, \beta ) \in \mathcal { S } _ { q } ^ { \infty } \cap \mathcal { S } _ { \mathbf a } } \langle \beta, J \rangle,$$
proceeds along the same lines, now using that for any ( α, β ) ∈ S ∞ q with ‖ α - ‖ a ∞ ≤ ε we can find ( α ( n ) , β ( n ) ) ∈ S a , 2 ε ( G n ) with d 1 (( α, β ) , ( α ( n ) , β ( n ) )) ≤ ε . glyph[square]
- 4.3. LD convergence implies convergence of quotients. In this section, we prove part (i) of Theorem 2.9, which is statement (iii) of the following lemma.
Lemma 4.1. Let q ∈ N , assume that ( G n n ) ≥ 0 is a q -LD convergent sequence of weighted graphs with rate function I q , and let S q ( I q ) = { ( α, β ) ∈ S q : I q (( α, β )) < ∞} . Then the following are true:
- (i) The set S q ( I q ) is closed with respect to the metric d 1 .
- (ii) The set S q ( I q ) is equal to the set S ∞ q = { ( α, β ) : d 1 ( ( α, β ) , S q ( G n ) ) → 0 } .
- (iii) S q ( G n ) converges to S q ( I q ) in the Hausdorff distance.
Proof. (i) For each a ∈ R , the set { ( α, β ) ∈ S q : I q (( α, β )) ≤ a } is closed by the lower semicontinuity of I q . To prove closedness of the set S q ( I q ), we observe that
$$I _ { q, \varepsilon, n } ( ( \alpha, \beta ) ) = - \frac { \log \mathcal { P } _ { q, G _ { n } } \left [ d _ { 1 } ( ( \alpha, \beta ), G _ { n } / \phi ) \leq \varepsilon \right ] } { | V ( G _ { n } ) | }$$
takes values in [0 , log q ] ∪{∞} , which in turn implies that I q takes values in [0 , log q ] ∪ {∞} and shows that S q ( I q ) = { ( α, β ) : I q (( α, β )) ≤ log q } .
- (ii) Let us first assume that ( α, β ) ∈ S q ( I q ). Then
$$\lim _ { n \to \infty } \sup _ { \substack { n \to \infty \\ } } I _ { q, \varepsilon, n } ( ( \alpha, \beta ) ) \leq I _ { q } ( ( \alpha, \beta ) ) \leq \log q \text{ \ for all \ } \varepsilon > 0,$$
because I q,ε,n (( α, β )) is non-increasing in ε . Since I q,ε,n takes values in [0 , log q ] ∪ {∞} , this implies that for all ε > 0 we can find an n 0 such that
$$I _ { q, \varepsilon, n } ( ( \alpha, \beta ) ) \leq \log q \ \text{ if } \ n \geq n _ { 0 },$$
which in turn implies that
$$d _ { 1 } \left ( ( \alpha, \beta ), \mathcal { S } _ { q } ( G _ { n } ) \right ) \leq \varepsilon \ \text{ if } \ n \geq n _ { 0 }.$$
This proves that ( α, β ) ∈ S q ( I q ) implies d 1 ( ( α, β ) , S q ( G n ) ) → 0.
Assume on the other hand that d 1 ( ( α, β ) , S q ( G n ) ) → 0, and by contradiction, assume further that ( α, β ) / ∈ S q ( I q ), i.e., assume that I q (( α, β )) = ∞ . Since I q,ε,n (( α, β )) takes values in [0 , log q ] ∪ {∞} , this implies that there exists an ε > 0 such that
$$\liminf _ { n \to \infty } I _ { q, \varepsilon, n } ( ( \alpha, \beta ) ) = \infty.$$
which in turn implies that there exists an n 0 < ∞ such that I q,ε,n (( α, β )) = ∞ for all n ≥ n 0 . As a consequence,
$$d _ { 1 } \left ( ( \alpha, \beta ), \mathcal { S } _ { q } ( G _ { n } ) \right ) > \varepsilon \ \text{ if } \ n \geq n _ { 0 },$$
contradicting the assumption that d 1 ( ( α, β ) , S q ( G n ) ) → 0.
(iii) Using the fact that S q ( I q ) is compact, one easily transforms the statement that d 1 ( ( α, β ) , S q ( G n ) ) → 0 for all ( α, β ) ∈ S q ( I q ) into the uniform statement that
$$\sup _ { ( \alpha, \beta ) \in \mathcal { S } _ { q } ( I _ { q } ) } d _ { 1 } \Big ( ( \alpha, \beta ), \mathcal { S } _ { q } ( G _ { n } ) \Big ) \to 0.$$
To prove convergence in the Hausdorff distance we have to prove the matching bound
$$\sup _ { ( \alpha, \beta ) \in \mathcal { S } _ { q } ( G _ { n } ) } d _ { 1 } \left ( ( \alpha, \beta ), \mathcal { S } _ { q } ( I _ { q } ) \right ) \to 0.$$
Fix ε > 0, and let S be the set S = ( { α, β ) ∈ S q : d 1 (( α, β ) , S q ( I q )) ≥ } ε . Since S is closed, we may use (3.4) to conclude that
$$\lim _ { n \to \infty } \frac { \log \mathcal { P } _ { q, G _ { n } } [ G _ { n } / \phi \in S ] } { | V ( G _ { n } ) | } \leq - \inf _ { ( \alpha, \beta ) \in S } I _ { q } ( ( \alpha, \beta ) ) = - \infty.$$
Since the probability on the left hand side takes values in { 0 } ∪ [ q -| V ( G n ) | , 1] this shows there must exists an n 0 = n 0 ( ε, q ) such that P q,G n [ G /φ n ∈ S ] = 0 if n ≥ n 0 , showing that S q ( G n ) ∩ S = ∅ when n ≥ n 0 . Expressed differently, for all ε > 0 we can find an n 0 such that for n ≥ n 0 ,
$$\mathcal { S } _ { q } ( G _ { n } ) \subseteq \{ ( \alpha, \beta ) \in \mathcal { S } _ { q } \colon d _ { 1 } ( ( \alpha, \beta ), \mathcal { S } _ { q } ( I _ { q } ) ) < \varepsilon \}.$$
Or still expressed differently, we can find a sequence ε n → 0 as n →∞ such that
$$d _ { 1 } ( ( \alpha, \beta ), \mathcal { S } _ { q } ( I _ { q } ) ) \leq \varepsilon _ { n } \text{ \ for all } ( \alpha, \beta ) \in \mathcal { S } _ { q } ( G _ { n } ). \text{ \quad \ } \Box$$
## 4.4. LD convergence implies convergence of free energies. In this section, we prove part (ii) of Theorem 2.9.
(ii) Given δ, ε > 0, chose an arbitrary ( α, β ) ∈ S a ,δ , and let Ω ( α,β ,ε ) be the set of configurations φ : V ( G n ) → [ q ] such that d 1 (( α, β ) , G n /φ ) ≤ ε . If φ ∈ Ω ( α,β ,ε ) , then | a i -α i ( G /φ n ) | ≤ ε + , implying that Ω δ ( α,β ,ε ) ⊆ Ω a ,ε + δ ( G n ). Using further that d 1 (( α, β ) , G n /φ ) ≤ ε implies that |〈 β, J 〉 - 〈 β G /φ ,J ( n ) 〉| ≤ ε ‖ J ‖ ∞ , and we then bound
$$& \text{bound} \\ & q ^ { | V ( G _ { n } ) | } e ^ { \langle \beta, J \rangle | V ( G _ { n } ) | } \mathcal { P } _ { q, G _ { n } } \left [ d _ { 1 } ( ( \alpha, \beta ), G _ { n } / \phi ) \leq \varepsilon \right ] \\ & \quad \leq \sum _ { \phi \in \Omega _ { \mathfrak { a }, \varepsilon + \delta } ( G _ { n } ) } e ^ { \left ( \langle \beta ( G _ { n } / \phi ), J \rangle + \varepsilon \| J \| _ { \infty } \right ) | V ( G _ { n } ) | } \\ & \quad = Z _ { G _ { n }, J } ^ { ( \mathfrak { a }, \varepsilon + \delta ) } e ^ { \varepsilon \| J \| _ { \infty } | V ( G _ { n } ) | }, \\ & \text{where the last sten follows from the definition (2.10) and the fact that $-F\cap (G...J)$}$$
where the last step follows from the definition (2.10) and the fact that -E φ ( G ,J n ) can be expressed as 〈 β G /φ ,J ( n ) 〉 . Using (2.16) plus monotonicity in ε to guarantee the existence of the limit ε → 0, this implies that
$$\lim _ { \varepsilon \to 0 } \liminf _ { n \to \infty } \frac { \log Z _ { G _ { n }, J } ^ { ( \mathfrak { a }, \varepsilon + \delta ) } } { | V ( G _ { n } ) | } \geq \log q + \langle \beta, J \rangle - I _ { q } ( ( \alpha, \beta ) ).$$
Since δ > 0 and ( α, β ) ∈ S a ,δ were arbitrary, this shows that
$$\sum _ { \varepsilon \to 0 } \times \sum _ { \lambda, \varepsilon } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ {\lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times \sum _ { \lambda, \alpha } \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \% 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0 \times 0$$
To get a matching upper bound we again fix a and ε, δ > 0. Since S a ,ε is closed and hence compact, we can find a finite set S δ ⊂ S a ,ε such that d Hf 1 ( S , δ S a ,ε ) ≤ δ . For s ∈ S δ , let B δ ( s ) be the set of pairs ( α, β ) ∈ S a ,ε such that d 1 ( s, ( α, β )) ≤ δ . Then S a ,ε = ⋃ s ∈ S δ B δ ( s ). As a consequence,
$$\sqrt { \dots } \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon \\ Z _ { G _ { n }, J } ^ { ( \mathfrak { a }, \varepsilon ) } & = \sum _ { \phi \colon V ( G _ { n } ) \to \left \{ q \right \} } e ^ { \langle G _ { n } / \phi, J \rangle \left \| V ( G _ { n } ) \right \| } \mathbf 1 _ { G _ { n } / \phi \in S _ { \mathfrak { a }, \varepsilon } } \\ & \leq q ^ { \left | V ( G _ { n } ) \right | } \sum _ { s \in S _ { \delta } } \left ( \sup _ { ( \alpha, \beta ) \in B _ { \delta } ( s ) } e ^ { \langle \beta / \phi, J \rangle \left | V ( G _ { n } ) \right | } \mathcal { P } _ { q, G _ { n } } \left [ G _ { n } / \phi \in B _ { \delta } ( s ) \right ] \right ).$$
Since S δ is finite and does not depend on n , we find that
$$& \text{Since $S_{\delta}$ is finite and does not depend on $n$, we find that} \\ & \limsup _ { n \to \infty } \frac { \log Z _ { G _ { n }, J } ^ { ( a, \varepsilon ) } } { | V ( G _ { n } ) | } \\ & \leq \log q + \limsup _ { n \to \infty } \max _ { s \in S _ { \delta } } \left ( \sup _ { ( \alpha, \beta ) \in B _ { \delta } ( s ) } \langle \beta, J \rangle + \frac { \log \mathcal { P } _ { q, G _ { n } } \left [ G _ { n } / \phi \in B _ { \delta } ( s ) \right ] } { | V ( G _ { n } ) | } \right ) \\ & = \log q + \max _ { s \in S _ { \delta } } \left ( \sup _ { ( \alpha, \beta ) \in B _ { \delta } ( s ) } \langle \beta, J \rangle + \limsup _ { n \to \infty } \log \mathcal { P } _ { q, G _ { n } } \left [ G _ { n } / \phi \in B _ { \delta } ( s ) \right ] { | V ( G _ { n } ) | } \right ) \\ & \leq \log q + \max _ { s \in S _ { \delta } } \left ( \sup _ { ( \alpha, \beta ) \in B _ { \delta } ( s ) } \langle \beta, J \rangle - \inf _ { ( \alpha, \beta ) \in B _ { \delta } ( s ) } I _ { q } ( ( \alpha, \beta ) ) \right ), \\ & \text{where we used } ( 3. 4 ) \text{ in the last step.} \\ & \alpha \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots.$$
where we used (3.4) in the last step.
Since sup ( α,β ) ∈ B δ ( s ) 〈 β, J 〉 ≤ 〈 β , J ′ 〉 +2 ‖ J ‖ ∞ δ for all ( α , β ′ ′ ) ∈ B δ ( s ), we conclude that
$$\text{raw} \\ \lim _ { n \to \infty } \sup _ { \substack { n \to \infty \\ | V ( G _ { n }, J } ) } \frac { \log Z _ { G _ { n }, J } ^ { ( a, e ) } } { | V ( G _ { n } ) | } & \leq \log q + \max _ { s \in S _ { \delta } } \sup _ { ( \alpha, \beta ) \in B _ { \delta } ( s ) } \left ( \langle \beta, J \rangle - I _ { q } ( ( \alpha, \beta ) ) \right ) + 2 \delta \| J \| _ { \infty } \\ & = \log q + \sup _ { ( \alpha, \beta ) \in S _ { \alpha, e } } \left ( \langle \beta, J \rangle - I _ { q } ( ( \alpha, \beta ) ) \right ) + 2 \delta \| J \| _ { \infty }. \\.$$
Sending δ → 0 and using the fact that Z ( a ,ε ) G ,J n is monotone in ε , this gives
$$\lim _ { \varepsilon \to 0 } \lim \sup _ { n \to \infty } \frac { \log Z _ { G _ { n }, J } ^ { ( \mathbf a, \varepsilon ) } } { | V ( G _ { n } ) | } \leq \sup _ { ( \alpha, \beta ) \in \mathcal { S } _ { \mathbf a, \varepsilon } } \left ( \log q + \langle \beta, J \rangle - I _ { q } ( ( \alpha, \beta ) ) \right ).$$
Choose an arbitrary sequence ε k going to zero, and choose ( α , β k k ) ∈ S a ,ε such that the supremum on the right hand side is bounded by log q + 〈 β , J k 〉-I q (( α , β k k ))+ ε k . Going to a subsequence if needed, assume that ( α , β k k ) converges to some ( α, β ) ∈ S a in the d 1 distance. Then
$$\lim _ { \varepsilon \to 0 } \lim \sup _ { n \to \infty } \frac { \log Z _ { G _ { n }, J } ^ { ( \mathbf a, \varepsilon ) } } { | V ( G _ { n } ) | } & \leq \log q + \langle \beta, J \rangle - \liminf _ { k \to \infty } I _ { q } ( ( \alpha _ { k }, \beta _ { k } ) ) \\ & \leq \log q + \langle \beta, J \rangle - I _ { q } ( ( \alpha, \beta ) )$$
where in the last step we used that I q is lower semi-continuous. Since ( α, β ) ∈ S a , the right hand side is bounded by as desired.
$$\overset { \cdot } { \sup } _ { ( \alpha, \beta ) \in \mathcal { S } _ { \alpha } } \left ( \log q + \langle \beta, J \rangle - I _ { q } ( ( \alpha, \beta ) ) \right ) = - F _ { \mathfrak { a } } ( I _ { q }, J ), \\ \underset { ( \alpha, \beta ) \in \mathcal { S } _ { \alpha } } { \sup }$$
## 5. Convergent sequences of graphons
In this section, we formulate and prove our main results in the language of graphons. Several of these results are generalizations of the corresponding results for L ∞ graphons proved in [8]; the exceptions are those involving LD convergence, which was not considered in [8]. It turns out, however, that most of our proofs are quite different from those of [8], most notably the proof that convergence of ground state energies implies convergence in metric (Section 5.5), which involves some new ideas not present in [8] such as the use of rearrangement inequalities.
5.1. Upper regularity for graphons. First we review the notion of upper regularity for graphons from [3].
Given a graphon W and a partition P = ( Y , . . . , Y 1 m ) of the interval [0 , 1] into finitely many measurable sets, we define W P to be the step function whose value on Y i × Y j equals to the average of W over Y i × Y j , i.e.,
$$W _ { \mathcal { P } } = \frac { 1 } { \lambda ( Y _ { i } ) \lambda ( Y _ { j } ) } \int _ { Y _ { i } \times Y _ { j } } W ( x, y ) \, d x \, d y \text{ \ on } Y _ { i } \times Y _ { j },$$
↦
where λ denotes the Lebesgue measure. An easy fact is that W → W P is contractive with respect to the L p norms ‖ · ‖ p and the cut norm ‖ · ‖ glyph[square] , i.e.,
$$( 5. 1 ) \quad \| W _ { \mathcal { P } } \| _ { \square } \leq \| W \| _ { \square } \quad \text{and} \quad \| W _ { \mathcal { P } } \| _ { p } \leq \| W \| _ { p } \quad \text{for all $p\geq 1$.}$$
Another standard fact is that up to a factor of 2, W P is the best step function approximation to W with steps in P , in the sense that
$$\| W - W _ { \mathcal { P } } \| _ { \square } \leq 2 \| W - U _ { \mathcal { P } } \| _ { \square }$$
for all graphons U . To see why, note that
$$\cdot \quad \cdot \quad \begin{array} { c } \cdot \cdot \\ \| W - W _ { \mathcal { P } } \| _ { \Box } & \leq \| W - U _ { \mathcal { P } } \| _ { \Box } + \| U _ { \mathcal { P } } - W _ { \mathcal { P } } \| _ { \Box } \\ & = \| W - U _ { \mathcal { P } } \| _ { \Box } + \| ( W - U _ { \mathcal { P } } ) _ { \mathcal { P } } \| _ { \Box } \\ & \leq 2 \| W - U _ { \mathcal { P } } \| _ { \Box }. \end{array}$$
Definition 5.1. Let K : (0 , ∞ → ) (0 , ∞ ) be any function. We say that a graphon W has K -bounded tails if for each ε > 0,
$$\int _ { [ 0, 1 ] ^ { 2 } } | W ( x, y ) | \mathbf 1 _ { | W ( x, y ) | \geq K ( \varepsilon ) } \, d x \, d y \leq \varepsilon.$$
A graphon W is ( K,η ) -upper regular if W P has K -bounded tails whenever P is a partition of the interval [0 , 1] into sets of measure at least η . A sequence ( W n n ) ≥ 0 of graphons is uniformly upper regular if there exist K : (0 , ∞ → ) (0 , ∞ ) and a sequence η n → 0 such that W n is ( K,η n )-upper regular for all n .
A key result from [3] is that every uniformly upper regular sequence of graphons contains a subsequent that converges in cut distance to some graphon. This is stated below.
Theorem 5.2 (Theorem C.7 in [3]) . If ( W n n ) ≥ 0 is a sequence of uniformly upper regular graphons, then there exists a graphon W and a subsequence ( W ′ n n ) ≥ 0 of ( W n n ) ≥ 0 such that δ glyph[square] ( W ,W ′ n ) → 0 .
5.2. Equivalent notions of convergence for graphons. The main theorem of this section, Theorem 5.3 below, is the analogue of the first four statements of Theorem 2.15. To state it, we need the analogue of the microcanonical ground state energies and microcanonical free energies define in (2.12) and (2.11), namely the quantities
$$\mathcal { E } _ { \mathfrak { a }, \varepsilon } ( W, J ) = \inf _ { \substack { \rho \in \text{FP} _ { q } \\ \| \alpha ( \rho ) - \mathfrak { a } \| _ { \infty } \leq \varepsilon } } \mathcal { E } _ { \rho } ( W, J ),$$
and
$$\mathcal { F } _ { \mathfrak { a }, \varepsilon } ( W, J ) = \inf _ { \substack { \rho \in \text{FP} _ { q } \\ \| \alpha ( \rho ) - \mathfrak { a } \| _ { \infty } \leq \varepsilon } } \left ( \mathcal { E } _ { \rho } ( W, J ) - \text{Ent} ( \rho ) \right ).$$
The following theorem is the main theorem of this section, and will be proved in Sections 5.4, 5.5, and 5.6 below.
Theorem 5.3. Let ( W n n ) ≥ 0 be a sequence of uniformly upper regular graphons. Then the following statements are equivalent:
- (i) ( W n n ) ≥ 0 is a Cauchy sequence in the cut metric δ glyph[square] .
- (ii) For every q ∈ N , the sequence ( ̂ S q ( W n )) n ≥ 0 is a Cauchy sequence under the Hausdorff distance d Hf 1 .
- (iii) For every q ∈ N , a ∈ glyph[triangle] q , and symmetric matrix J ∈ R q × q ,
$$\lim _ { \varepsilon \to 0 } \liminf _ { n \to \infty } \mathcal { E } _ { \mathfrak a, \varepsilon } ( W _ { n }, J ) = \lim _ { \varepsilon \to 0 } \lim _ { n \to \infty } \sup _ { n \to \infty } \mathcal { E } _ { \mathfrak a, \varepsilon } ( W _ { n }, J ).$$
- (iv) For every q ∈ N , a ∈ glyph[triangle] q , and symmetric matrix J ∈ R q × q ,
$$\lim _ { \varepsilon \to 0 } \liminf _ { n \to \infty } \mathcal { F } _ { \mathbf a, \varepsilon } ( W _ { n }, J ) = \lim _ { \varepsilon \to 0 } \lim _ { n \to \infty } \sup _ { n \to \infty } \mathcal { F } _ { \mathbf a, \varepsilon } ( W _ { n }, J ).$$
## Remark 5.4 .
- (i) We will prove the theorem by showing that (i) ⇒ (ii) ⇒ (iii) ⇒ (i) and that (i) ⇒ (iv) ⇒ (iii). It turns out that the assumption of uniform upper regularity is only needed for the proof that (iii) ⇒ (i). All other implications hold for arbitrary sequences of graphons.
- (ii) Under the assumption (ii) of the above theorem, ̂ S q ( W n ) converges to the compact set 5 S ∞ q = { ( α, β ): d 1 (( α, β ) , ̂ S q ( W n )) → } 0 . Proceeding as in the proof of Theorem 3.3, this in turn implies that (iii) holds with the limit given as
$$E _ { \mathbf a } ( J ) = - \max _ { \substack { ( \alpha, \beta ) \in \mathcal { S } _ { q } ^ { \infty } \\ \alpha = \mathbf a \\ \beta \beta } } \langle \beta, J \rangle,$$
again without the assumption of uniform upper regularity.
- (iii) Under the assumption (i) of the above theorem, the sequences ( E a ( W ,J n )) n ≥ 0 and ( F a ( W ,J n )) n ≥ 0 are convergent for all q, a , J . (In particular, the use of ε in the theorem statement is just for comparison with the case of graphs, and not because it is truly needed.) Finally, under the assumption of uniform upper regularity, each of these two statements is not only necessary but also sufficient for convergence in metric to hold, as we will show in Sections 5.5 and 5.6.
5.3. Compactness of quotient space. Before jumping into the proof of Theorem 5.3, we prove some compactness results about quotients of W , thereby shedding light on the quantities E a ( W,J ), F a ( W,J ), and I q (( α, β ) , W ).
Recall the glyph[lscript] 1 distance d 1 from (2.7) as well as the definitions of fractional graphon quotients from Section 2.6.
Proposition 5.5. Let W be a graphon, let q ∈ N , and let a ∈ glyph[triangle] q . Then ̂ S q ( W ) and ̂ S a ( W ) are compact under the metric d 1 .
We will prove this proposition after we develop a few preliminaries.
In (2.17)-(2.19), E a ( W,J ), F a ( W,J ), and I q (( α, β ) , W ) were originally defined as infima over some subset of fractional partitions. We will see that the infima are attained by some fractional partitions, so that the 'inf' can be replaced by 'min'.
5 The nonempty compact subsets of ̂ S q form a complete metric space under d Hf 1 (see [13]).
Furthermore, we will see that
$$\mathcal { E } _ { \mathfrak a } ( W, J ) = \lim _ { \varepsilon \to 0 } \mathcal { E } _ { \mathfrak a, \varepsilon } ( W, J ),$$
$$\mathcal { F } _ { \mathfrak { a } } ( W, J ) = \lim _ { \varepsilon \to 0 } \mathcal { F } _ { \mathfrak { a }, \varepsilon } ( W, J ), \quad \text{and}$$
$$I _ { q } ( ( \alpha, \beta ), W ) = \lim _ { \varepsilon \to 0 } \inf _ { \substack { \rho \in \text{FP} _ { q } \\ d _ { 1 } ( W / \rho, ( \alpha, \beta ) ) \leq \varepsilon } } ( \log q - \text{Ent} ( \rho ) ).$$
The quantities E a ( W,J ) and F a ( W,J ) are both continuous with respect to a , W, J . On the other hand, I q (( α, β ) , W ) is lower semicontinuous in its arguments (Proposition 5.10), and it is not continuous (it takes values in [0 , log q ] when ( α, β ) ∈ S ̂ q ( W ) and is infinite otherwise).
It follows as an immediate corollary of Proposition 5.5 that
$$\mathcal { E } ( W, J, h ) = - \max _ { ( \alpha, \beta ) \in \widehat { \mathcal { S } } _ { q } ( W ) } \left ( \langle \beta, J \rangle + \langle \alpha, h \rangle \right )$$
and
$$\mathcal { E } _ { \mathfrak { a } } ( W, J ) = - \max _ { ( \alpha, \beta ) \in \widehat { \mathcal { S } } _ { \mathfrak { a } } ( W ) } \langle \beta, J \rangle,$$
↦
since ( α, β ) →〈 β, J 〉 and ( α, β ) →〈 α, h 〉 are continuous in the d 1 metric. This gives an alternate representation of ground state energies of W in terms of its quotients.
↦
5.3.1. Approximations by step functions. One way to approximate a graphon W by step functions is given by the following lemma, which is an immediate consequence of the almost everywhere differentiability of the integral function.
Lemma 5.6. Let p ≥ 1 . For a positive integer n , let P n be the partition of [0 , 1] into consecutive intervals of length 1 /n . If W is a graphon, then W P n → W almost everywhere. In addition, W P n → W in L p whenever W is an L p graphon.
Proof. Almost everywhere convergence follows the Lebesgue differentiation theorem. To get convergence in L p we approximate W by the bounded graphon W M = W 1 | W |≤ M , where M > 0. By triangle inequality and (5.1),
$$\| W - W _ { \mathcal { P } _ { n } } \| _ { p } & \leq \| W _ { M } - ( W _ { M } ) _ { \mathcal { P } _ { n } } \| _ { p } + \| W - W _ { M } \| _ { p } + \| ( W - W _ { M } ) _ { \mathcal { P } _ { n } } \| _ { p } \\ & \leq \| W _ { M } - ( W _ { M } ) _ { \mathcal { P } _ { n } } \| _ { p } + 2 \| W - W _ { M } \| _ { p }.$$
The second term on the right can be made arbitrarily small by setting M to be sufficiently large, and for any fixed M , the first term on the right goes to zero as n →∞ . This shows that ‖ W -W P n ‖ p → 0. glyph[square]
The lemma does not, however, give any information on the speed of convergence. If instead of almost everywhere convergence we content ourselves with convergence in the cut metric, the situation is different, as is well known in the case of L 2 graphons, where one can apply the weak version of the regularity lemma first established in [10]. This lemma can be generalized to L p graphons for p > 1 and more generally to any graphon with K -bounded tails (see [3]), but we will not need this here, where we use only the corresponding version for uniformly upper regular sequences of graphs (see Theorem 6.1 in Section 6).
5.3.2. Limits of fractional partitions. We say that a sequence ρ ( n ) ∈ FP q of fractional partitions converges to ρ ∈ FP q over rational intervals if ∫ D ρ ( n ) i ( x dx ) → ∫ D ρ i ( x dx ) for every interval D ⊆ [0 , 1] with rational endpoints.
Lemma 5.7. Fix q ∈ N . Every sequence ρ ( n ) ∈ FP q of fractional partitions contains a subsequence that converges to some ρ ∈ FP q over rational intervals.
Proof. By restricting to a subsequence, we may assume that ∫ D ρ ( n ) i ( x dx ) converges for every interval D ⊆ [0 , 1] with rational endpoints, and let us denote this limiting value by µ i ( D ). Then, by the extension theorem for measures (Proposition 5.5 in [11]), µ i can be extended to a measure on [0 1] such that , ∑ i ∈ [ q ] µ i ( D ) = λ D ( ) for every measurable D ⊆ [0 , 1], where λ is the Lebesgue measure. It is easy to see that µ i is absolutely continuous with respect to λ . Defining ρ i to be the density of µ i with respect to λ and changing ρ i on a set of measure zero, we obtain the desired fractional partition ρ = ( ρ i ) i ∈ [ q ] . glyph[square]
Lemma 5.8. If ρ ( n ) ∈ FP q converges to ρ ∈ FP q over rational intervals and ‖ W n -W ‖ glyph[square] → 0 , then d 1 ( W /ρ n ( n ) , W/ρ ) → 0 .
Proof. We have α i ( ρ ( n ) ) = ∫ [0 1] , ρ ( n ) i → ∫ [0 1] , ρ i = α i ( ρ ). Next we have
$$\left | \beta _ { i j } ( W _ { n } / \rho ^ { ( n ) } ) - \beta _ { i j } ( W / \rho ^ { ( n ) } ) \right | & = \left | \int \rho _ { i } ^ { ( n ) } ( x ) \rho _ { j } ^ { ( n ) } ( y ) ( W _ { n } ( x, y ) - W ( x, y ) ) \, d x \, d y \right | \\ & \leq \| W _ { n } - W \| _ { \Box } \,.$$
It remains to show that β ij ( W/ρ ( n ) ) → β ij ( W/ρ ), i.e.,
$$\int \rho _ { i } ^ { ( n ) } ( x ) \rho _ { j } ^ { ( n ) } ( y ) W ( x, y ) \, d x \, d y \to \int \rho _ { i } ( x ) \rho _ { j } ( y ) W ( x, y ) \, d x \, d y,$$
which follows from Lemma 5.6 as we can approximate W arbitrarily well in L 1 using step functions with rational steps, and ρ ( n ) converges to ρ over rational intervals. glyph[square]
Now we can prove the compactness of the set of quotients.
Proof of Proposition 5.5. Let ( W/ρ ( n ) ) n ≥ 1 be a sequence of quotients in ̂ S q ( W ) (or ̂ S a ( W )). By Lemma 5.7 we can restrict to a subsequence so that ρ ( n ) converges to some ρ ∈ FP q over rational intervals. By Lemma 5.8, we have d 1 ( W/ρ ( n ) , W/ρ ) → 0, thereby proving that the space of quotients is closed and hence compact. glyph[square]
The claim (5.4) has a similar proof: we have E ρ ( W,J ) = -〈 β W/ρ ,J ( ) 〉 , so that d 1 ( W/ρ ( n ) , W/ρ ) → 0 implies E ρ ( n ) ( W,J ) →E ρ ( W,J ).
5.3.3. Entropy and lower semicontinuity. Now we prove (5.5) and (5.6), and furthermore the claim that in the definitions (2.18) and (2.19) for F a ( W,J ) and I q (( α, β ) , W ) the infimum is attained by some fractional partition. In fact, they are all immediate consequences of Lemma 5.7 along with the following lemma.
Lemma 5.9. If ρ ( n ) ∈ FP q converges to ρ ∈ FP q over rational intervals, then
$$\lim \sup E n t ( \rho ^ { ( n ) } ) \leq E n t ( \rho ).$$
$$n \rightarrow \infty$$
Proof. For any positive integer k , let ρ [ k ] ∈ FP q (and similarly ρ ( n ) [ k ] ) denote the fractional partition obtained from ρ [ k ] by averaging over the interval [( j -1) /k, j/k ) for each integer j ∈ [ k ]. Specifically, we set the value of ρ [ k ,i ] on [( j -1) /k, j/k ) to be k ∫ j/k ( j -1) /k ρ i ( x dx ) .
Since -x log x is concave, Ent ( ρ ( n ) ) ≤ Ent ( ρ ( n ) [ k ] ) by Jensen's inequality. For a fixed k we have
$$\lim _ { n \to \infty } \sup _ { n \to \infty } E _ { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } { \L } }$$
where the last equality follows from ρ ( n ) converging to ρ on rational intervals. Finally, we have ρ [ k ,i ] → ρ i almost everywhere as k →∞ by the Lebesgue differentiation theorem, and thus Ent ( ρ [ k ] ) → Ent ( ρ ) as k → ∞ by the bounded convergence theorem. This proves the lemma. glyph[square]
Proposition 5.10. Let q ∈ N . The function I q (( α, β ) , W ) is lower semicontinuous (with the metric d 1 on the first argument and δ glyph[square] on the second).
Proof. We need to show that if ( α ( n ) , β ( n ) ) → ( α, β ) in d 1 and W n → W in δ glyph[square] as n →∞ , then
$$\liminf _ { n \to \infty } I _ { q } ( ( \alpha ^ { ( n ) }, \beta ^ { ( n ) } ), W _ { n } ) \geq I _ { q } ( ( \alpha, \beta ), W ).$$
We may restrict to a subsequence so that I q (( α ( n ) , β ( n ) ) , W n ) converges to the original lim inf. Since I q is invariant under measure preserving bijections for the graphon, we may assume that ‖ W n -W ‖ glyph[square] → 0. The result is automatic if the limit is infinity, so we might as well assume that I q (( α ( n ) , β ( n ) ) , W n ) < ∞ (and hence at most log q ) for all n , so that there is some ρ ( n ) ∈ FP q with W/ρ ( n ) = ( α ( n ) , β ( n ) ) and I q (( α ( n ) , β ( n ) ) , W n ) = log q -Ent ( ρ ( n ) ). By Lemma 5.7 we can further restrict to a subsequence so that ρ ( n ) converges to some ρ ∈ FP q over rational intervals. By Lemma 5.8 we have W/ρ = lim n →∞ W /ρ n ( n ) = lim n →∞ ( α ( n ) , β ( n ) ) = ( α, β ), so I q (( α, β ) , W ) ≤ log q -Ent ( ρ ). By Lemma 5.9,
$$\liminf _ { n \to \infty } I _ { q } ( ( \alpha ^ { ( n ) }, \beta ^ { ( n ) } ), W _ { n } ) & = \liminf _ { n \to \infty } ( \log q - \text{Ent} ( \rho _ { n } ) ) \\ & \geq \log q - \text{Ent} ( \rho ) \geq I _ { q } ( ( \alpha, \beta ), W ),$$
as desired.
$$\sqcup$$
- 5.4. Proof of (i) ⇒ (ii) ⇒ (iii) in Theorem 5.3. The claim that (i) implies (ii) follows from Lemma 5.11 below. The claim that (ii) implies (iii)-with the limit expressed as described in Remark 5.4(ii)-is essentially identical to the proof of Theorem 3.3 from Section 4.2 and is left to the reader.
Lemma 5.11. Let q ∈ N , U and W be graphons, and ρ ∈ FP q . Then d 1 ( U/ρ,W/ρ ) ≤ q 2 ‖ U -W ‖ glyph[square] and hence
$$d _ { 1 } ^ { \text{Hf} } ( \widehat { \mathcal { S } } _ { q } ( U ), \widehat { \mathcal { S } } _ { q } ( W ) ) \leq q ^ { 2 } \delta _ { \Box } ( U, W ).$$
Proof. We have α U/ρ ( ) = α ρ ( ) = α W/ρ ( ). Also for i, j ∈ [ q ] we have
$$\ell _ { 1 } \quad \partial \cdots \quad \partial \cdots \quad \partial \cdots \quad \partial \cdots \quad \partial \cdots \quad \partial \cdots \\ ( 5. 9 ) \quad & | \beta _ { i j } ( U / \rho ) - \beta _ { i j } ( W / \rho ) | = \left | \int _ { [ 0, 1 ] ^ { 2 } } ( U - W ) ( x, y ) \rho _ { i } ( x ) \rho _ { j } ( y ) \, d x \, d y \right | \\ & \leq \left \| U - W \right \| _ { \Box }.$$
Summing over all i, j ∈ [ q ] gives d 1 ( U/ρ,W/ρ ) ≤ q 2 ‖ U -W ‖ glyph[square] . The claim (5.8) follows immediately. glyph[square]
We next observe that microcanonical ground state energies are continuous in the cut metric. To state this result, we define the norm ‖ J ‖ 1 of a matrix J ∈ R q × q to be ∑ i,j ∈ [ q ] | J ij | .
Proposition 5.12. Let q ∈ N , a ∈ glyph[triangle] q , and h ∈ R q , and let J ∈ R q × q be a symmetric matrix. If U and W are arbitrary graphons, then
$$( 5. 1 0 ) & & | \mathcal { E } _ { \mathbf a } ( U, J ) - \mathcal { E } _ { \mathbf a } ( W, J ) | \leq \| J \| _ { 1 } \delta _ { \square } ( W, U ).$$
Proof. Since the left side of this bound does not change if we replace U by U φ for some measure preserving bijection φ : [0 , 1] → [0 , 1], it is enough to prove it in terms of ‖ W -U ‖ glyph[square] instead of δ glyph[square] ( W,U ). Let ρ ∈ FP q . Using (5.9) we obtain
$$\wedge \, \| J \| _ { 1 } \| _ { 2 } \cdots \, \| \, \wedge \, \| J \| _ { 2 } \cdots \, \| \, \wedge \, \| J \| _ { 2 } \cdots \, \| \, \wedge \, \| J \| _ { 2 } \cdots \, \| \, \wedge \, \| J \| _ { 2 } \cdots \, \| \, \wedge \, \| J \| _ { 2 } \cdots \, \| \, \wedge \, \| J \| _ { 2 } \cdots \, \| \, \wedge \, \| J \| _ { 2 } \cdots \, \| \, \wedge \, \| J \| _ { 2 } \cdots \, \| \, \wedge \, \| J \| _ { 2 } \cdots \, \| \, \\ ( 5. 1 1 ) & \left | \mathcal { E } _ { \rho } ( U, J, h ) - \mathcal { E } _ { \rho } ( W, J, h ) \right | = \left | \sum _ { i, j \in [ q ] } ( \beta _ { i j } ( U / \rho ) - \beta _ { i j } ( W / \rho ) ) J _ { i j } \right | \\ & \leq \| J \| _ { 1 } \| _ { 2 } \| U - W \| _ { \Box }, \\ \text{as desired.}$$
5.5. Proof of (iii) ⇒ (i) in Theorem 5.3. By the bound (5.10), convergence in metric implies convergence of the microcanonical ground state energies. To prove the converse, we will establish the following proposition, one of the main results of this section.
Theorem 5.13. Let W and U be two graphons. If
$$\mathcal { E } _ { \mathbf a } ( U, J ) = \mathcal { E } _ { \mathbf a } ( W, J )$$
for all q ∈ N , every symmetric matrix J ∈ R q × q , and all a of the form
$$\mathbf a _ { q } & = ( 1 / q, \dots, 1 / q ),$$
then δ glyph[square] ( W,U ) = 0 .
This theorem proves the implication (iii) ⇒ (i) in Theorem 5.3, as well as the fact that convergence of E a ( W ,J n ) is sufficient for convergence in metric (see Remark 5.4(iii)). Indeed, for the second of these assertions, assume first that the ground state energies E a ( W ,J n ) converge for all q ∈ N , all a of the form (5.13), and all J , while W n does not converge in the cut metric. Since ( W n n ) ≥ 0 is assumed to be uniformly upper regular, we may use Theorem 5.2 to find two subsequences W ′ n and W ′′ n of W n that converge to two graphons W and U in the cut distance δ glyph[square] , while δ glyph[square] ( W,U ) > 0. But convergence in the cut distance implies convergence of the ground state energies by (5.10), which means that U and W have identical ground state energies, a contradiction. The proof of (iii) ⇒ (i) in Theorem 5.3 is similar, since convergence of W ′ n to W in metric implies that ̂ S q ( W ′ n ) →S ̂ q ( W ) in the Hausdorff distance, which in turn can easily be seen to give convergence of the quantities in (iii) to E a ( W,J ) (the proof is the same as that of Theorem 3.3), and similarly for the convergence along the subsequence W ′′ n to E a ( U, J ).
To prove Theorem 5.13, we will work with the quasi-inner product
$$\mathcal { C } ( W, Y ) = \sup _ { j } E [ W Y ^ { \phi } ],$$
where the supremum is taken over all measure preserving bijections φ : [0 , 1] → [0 , 1] and the expectation E [ ] · is with respect to the Lebesgue measure on [0 1] , 2 , i.e.,
$$E [ W Y ^ { \phi } ] = \underset { [ 0, 1 ] ^ { 2 } } { \int } \, W ( x, y ) Y ^ { \phi } ( x, y ) \, d x \, d y = \underset { [ 0, 1 ] ^ { 2 } } { \int } \, W ( x, y ) Y ( \phi ( x ), \phi ( y ) ) \, d x \, d y.$$
This quantity was defined in [8], where it was assumed that both W and Y are in L ∞ . But the definition makes sense in our more general context, where we will assume that W is an arbitrary graphon and Y is bounded.
Lemma 5.14. Let W and U be two graphons such that (5.12) holds for all q ∈ N and all a of the form (5.13) . Then
$$\mathcal { C } ( W, Y ) = \mathcal { C } ( U, Y )$$
for all bounded graphons Y .
Proof. If Y = W H for a weighted graph H on q nodes, where H has edge weights J ∈ R q × q and vertex weights a of the form (5.13), then
$$- \mathcal { E } _ { \mathfrak a } ( W, J ) = \mathcal { C } ( W, W ^ { H } ),$$
and the claim follows directly from the assumption (5.12). For general Y , we use Lemma 5.6 to approximate Y by step functions. More explicitly, let P n be as in Lemma 5.6, and let Y n = Y P n . Then Y n → Y in L 1 , and ‖ Y n ‖ ∞ ≤ ‖ Y ‖ ∞ . Hence
$$\left | E [ W Y _ { n } ^ { \phi } ] - E [ W Y ^ { \phi } ] \right | & \leq 2 \| Y \| _ { \infty } | \| W \mathbf 1 _ { | W | \geq K } \| _ { 1 } + \| W \mathbf 1 _ { | W | \leq K } \| _ { \infty } \| Y ^ { \phi } - Y _ { n } ^ { \phi } \| _ { 1 } \\ & \leq 2 \| Y \| _ { \infty } | \| W \mathbf 1 _ { | W | \geq K } \| _ { 1 } + K \| Y - Y _ { n } \| _ { 1 }.$$
The right side can be made as small as desired by first choosing K large enough and then n large enough. Observing that the resulting convergence is uniform in φ , this implies that C ( W,Y n ) →C ( W,Y ) and similarly for C ( U, Y n ). From this, the claim follows. glyph[square]
Lemma 5.15. Let W and U be two graphons such that (5.14) holds for all bounded graphons Y . Consider the real valued random variables ̂ W = W x,y ( ) and ̂ U = U x,y ( ) where x, y are chosen independently uniformly at random from [0 , 1] . Then ̂ W and ̂ U have the same distribution.
To prove this lemma, we will use some notions and results from the theory of monotone rearrangement.
5.5.1. Monotone rearrangements and proof of Lemma 5.15. Throughout this section, we identify graphons W with the real-valued random variables W x,y ( ) obtained by choosing x, y independently uniformly at random from [0 1]; if , W is such a random variable, we use E W [ ] to denote its expectation. For s ∈ R we use { W > s } to denote the event that W > s , namely { W > s } = { ( x, y ) : W x,y ( ) > s } , and Pr[ W > s ] to denote the probability of this event.
For a graphon W , we define the monotone rearrangement as the function
$$W ^ { * } ( x _ { 1 }, x _ { 2 } ) = \sup \{ t \in \mathbb { R } \, \colon \Pr [ W > t ] > \| x \| _ { \infty } ^ { 2 } \},$$
where ‖ x ‖ ∞ = max { x , x 1 2 } . Then W ∗ ( x , x 1 2 ) is a weakly decreasing function of ‖ x ‖ ∞ , and it has the same distribution as W ; i.e.,
$$\Pr [ W > t ] = \Pr [ W ^ { * } > t ].$$
To see why, note that
$$\text{To see why, note that} \\ \Pr [ W ^ { * } > t _ { 0 } ] & = \Pr [ \sup \{ t \in \mathbb { R } \, \colon \Pr [ W > t ] > \| x \| _ { \infty } ^ { 2 } \} > t _ { 0 } ] \\ & = \Pr [ \text{there exists $t>t_{0}$ such that $Pr[W>t] > \| x \| _ { \infty } ^ { 2 } ] \\ & = \Pr [ \Pr [ W > t _ { 0 } ] > \| x \| _ { \infty } ^ { 2 } ] \\ & = \Pr \left [ x _ { 1 } < \sqrt { \Pr [ W > t _ { 0 } ] } \text{ and } x _ { 2 } < \sqrt { \Pr [ W > t _ { 0 } ] } \right ] \\ & = \Pr [ W > t _ { 0 } ], \\ \text{where the third line follows from the fact that $t\mapsto Pr[W>t|$ is right-continuous.}$$
↦
where the third line follows from the fact that t → Pr W > t [ ] is right-continuous.
Define two graphons W and U to be aligned if their level sets are nested, in the sense that for all s, t ∈ R , either { W > s } ⊆ { U > t } or { U > s } ⊆ { W > s } . It is easy to see that for any two graphons U , W , the monotone rearrangements U ∗ and W ∗ are aligned.
Let W be an L 1 graphon, and let Y be a bounded graphon. Then we have the rearrangement inequality
$$E [ W Y ] \leq E [ W ^ { * } Y ^ { * } ].$$
See Appendix A for a proof. The proof also tells us that
$$E [ W Y ] = E [ W ^ { * } Y ^ { * } ]$$
if Y and W are aligned. Before delving into the proof of Lemma 5.15 we observe that
$$E [ W Y ] \leq \mathcal { C } ( W, Y ) \leq \mathbf E [ W ^ { * } Y ^ { * } ].$$
The first inequality follows immediately from the definition of C ( W,Y ), and the second follows from the definition and the fact that Y and Y φ have the same distribution, which in turn implies that Y ∗ = ( Y φ ) ∗ .
Proof of Lemma 5.15. Define top λ ( W ) ⊆ [0 , 1] 2 in such a way that the Lebesgue measure of top λ ( W ) is λ , and W u,v ( ) ≤ inf ( x,y ) ∈ top λ ( W ) W x,y ( ) whenever ( u, v ) / ∈ top λ ( W ). Explicitly, let M = sup { s ∈ R : Pr[ W > s ] ≥ λ } . If Pr { W = M } = 0, then we have that Pr { W > M } = Pr { W ≥ M } = λ , and we define top λ ( W ) = { W > M } . Otherwise, Pr { W > M } ≤ λ ≤ Pr { W ≥ M } , in which case we chose top λ ( W ) in such a way that { W > M } ⊆ top λ ( W ) ⊆ { W ≥ M } and µ [top λ ( W )] = λ , where µ denotes the Lebesgue measure on [0 1] , 2 . In either case, we have M = inf ( x,y ) ∈ top λ ( W ) W x,y ( ) and W u,v ( ) ≤ M whenever ( u, v ) / ∈ top λ ( W ).
It is easy to see that W and the indicator function 1 top λ ( W ) are aligned, implying that E [ W 1 top λ ( W ) ] = E [ W ∗ 1 top λ ( W ) ∗ ] . Consider now the L 1 graphon Y = 1 top λ ( W ) . With the help of (5.15) and the fact that E [ WY ] = E W Y [ ∗ ∗ ] we have
$$\mathbf E [ W Y ] = \mathcal { C } ( W, Y ) = \mathcal { C } ( U, Y ) \leq \mathbf E [ U ^ { * } Y ^ { * } ].$$
Let ˜ Y = 1 top λ ( U ) . Then Y and ˜ have the same distribution, implying that Y ˜ Y ∗ = Y ∗ . On the other hand, ˜ Y and U are aligned, implying that E [ UY ˜ ] = E [ U Y ∗ ˜ ∗ ]. Putting everything together, we conclude that
$$E [ W Y ] \leq E [ U \tilde { Y } ].$$
In a similar way, we show that E [ UY ˜ ] ≤ E [ WY ]. We thus have shown that for all λ ∈ [0 , 1],
$$E [ W { \mathbf 1 } _ { \mathrm t o p _ { \lambda } ( W ) } ] = E [ U { \mathbf 1 } _ { \mathrm t o p _ { \lambda } ( U ) } ].$$
This in turn implies that W and U have the same distribution.
glyph[square]
5.5.2. Proof of Theorem 5.13. Theorem 5.13 follows immediately from Lemma 5.14 and the following proposition. We remark that when W is bounded, or even in L 2 , the proof of the proposition is much easier as one can consider C ( W,W ) = E [ W 2 ]. This does not work when W is only assumed to be in L 1 . The proof begins by transforming W into a bounded graphon. In what follows, we use the metric
$$\delta _ { p } ( U, W ) = \inf _ { \phi } \| U - W ^ { \phi } \| _ { p },$$
where the infimum is over all measure-preserving bijections φ : [0 , 1] → [0 , 1]. It clearly satisfies δ glyph[square] ( U, W ) ≤ δ 1 ( U, W ).
Proposition 5.16. If U and W are graphons such that C ( U, Y ) = C ( W,Y ) for all L ∞ graphons Y , then δ 1 ( U, W ) = 0 .
Proof. We know that U and W have the same distribution. Let ˜ W = arctan W and ˜ U = arctan U (note that ˜ W and ˜ U are both bounded). Since both arctan x and x -arctan x are increasing in x , for every measure preserving bijection σ : [0 , 1] → [0 , 1] the rearrangement inequality implies that
$$\mathbf E [ ( W - \widetilde { W } ) \widetilde { W } ] \geq \mathbf E [ ( U - \widetilde { U } ) ^ { \sigma } \widetilde { W } ]$$
and
$$E [ \widetilde { W } ^ { 2 } ] \geq E [ \widetilde { U } ^ { \sigma } \widetilde { W } ].$$
Thus
$$E [ W \widetilde { W } ] - E [ U ^ { \sigma } \widetilde { W } ] \geq E [ \widetilde { W } ^ { 2 } ] - E [ \widetilde { U } ^ { \sigma } \widetilde { W } ] \geq 0,$$
By assumption we have C ( U, W ˜ ) = C ( W,W ˜ ), which must equal E [ WW ˜ ] by the rearrangement inequality. Taking the infimum over all σ in (5.16) and using the facts from the previous sentence yields C ( ˜ ˜ W,W ) = C ( ˜ U,W ˜ ). A similar argument shows that C ( ˜ U,U ˜ ) = C ( ˜ U,W ˜ ). Therefore
$$\delta _ { 2 } ( \widetilde { U }, \widetilde { W } ) ^ { 2 } = \inf _ { \sigma } \mathbf E [ ( \widetilde { U } ^ { \sigma } - \widetilde { W } ) ^ { 2 } ] = \mathbf E [ \widetilde { U } ^ { 2 } ] + \mathbf E [ \widetilde { W } ^ { 2 } ] - 2 \sup _ { \psi } \mathbf E [ \widetilde { U } ^ { \sigma } \widetilde { W } ]$$
$$^ { \prime }, w _ { \, \lambda } \, = \, \underset { \sigma } { \min } \, \mathfrak { E } _ { [ \{ \mathcal { U } \, - w _ { \, \lambda } \, ] } } & = \mathfrak { E } _ { [ \mathcal { U } \, ] } + \mathfrak { E } _ { [ \mathcal { V } \, ] } - 4 \sup _ { \sigma } \mathfrak { E } _ { [ \mathcal { U } \, w _ { \, \lambda } \, ] } \\ & = \mathcal { C } ( \widetilde { U }, \widetilde { U } ) + \mathcal { C } ( \widetilde { W }, \widetilde { W } ) - 2 \mathcal { C } ( \widetilde { U }, \widetilde { W } ) = 0.$$
Since δ 1 ( ˜ U,W ˜ ) ≤ δ 2 ( ˜ U,W ˜ ), we must have δ 1 ( ˜ U,W ˜ ) = 0 as well.
Finally we need to deduce that δ 1 ( U, W ) = 0. Let K > 0. From the mean value theorem, we know that
$$| x - y | \leq ( 1 + K ^ { 2 } ) \, | \arctan x - \arctan y |$$
whenever x, y ∈ -[ K,K ]. It follows that for every σ ,
$$\| ( U ^ { \sigma } - W ) \mathbf 1 _ { | U ^ { \sigma } | \leq K, | W | \leq K } \| _ { 1 } \leq ( 1 + K ^ { 2 } ) \| \widetilde { U } ^ { \sigma } - \widetilde { W } \| _ { 1 }.$$
The left side differs from ‖ U σ -W ‖ 1 by at most 4 ∥ ∥ U 1 | U >K | ∥ ∥ 1 (here we use the triangle inequality, and the fact that U and W have the same distribution, so that ∥ ∥ W 1 | U σ | >K ∥ ∥ 1 ≤ ∥ ∥ U 1 | U >K | ∥ ∥ 1 and ∥ ∥ U σ 1 | W >K | ∥ ∥ 1 ≤ ∥ ∥ U 1 | U >K | ∥ ∥ 1 by the rearrangement inequality). Thus
$$\| U ^ { \sigma } - W \| _ { 1 } \leq ( 1 + K ^ { 2 } ) \| \widetilde { U } ^ { \sigma } - \widetilde { W } \| _ { 1 } + 4 \left \| U \mathbf 1 _ { | U | > K } \right \| _ { 1 }.$$
Taking the infimum over σ and using δ 1 ( ˜ U,W ˜ ) = 0, we find that δ 1 ( U, W ) ≤ 4 ∥ ∥ U 1 | U >K | ∥ ∥ . Since K can be made arbitrarily large, δ 1 ( U, W ) = 0. glyph[square]
5.6. Proof of (i) ⇒ (iv) ⇒ (iii) in Theorem 5.3. The following result gives the implication (i) ⇒ (iv) in Theorem 5.3 as well as the statement that F a ( W ,J n ) converges whenever W n converges in metric.
Proposition 5.17. Let q ∈ N , a ∈ glyph[triangle] q , ε > 0 , and h ∈ R q , and let J ∈ R q × q be a symmetric matrix. For any two graphons U and W ,
$$\left | \mathcal { F } _ { \mathbf a } ( U, J ) - \mathcal { F } _ { \mathbf a } ( W, J ) \right | \leq \| J \| _ { 1 } \delta _ { \square } ( U, W )$$
and
$$\left | \mathcal { F } _ { \mathfrak a, \varepsilon } ( U, J ) - \mathcal { F } _ { \mathfrak a, \varepsilon } ( W, J ) \right | \leq \| J \| _ { 1 } \delta _ { \square } ( U, W ).$$
Proof. Since the left sides of the above bounds do not change if we replace U by U φ for a measure preserving bijection φ : [0 , 1] → [0 , 1], it is enough to prove the lemma with a bound in terms of ‖ U -W ‖ glyph[square] instead of δ glyph[square] ( U, W ). The result then follows immediately from (5.11) and the definitions (2.18) and (5.3). glyph[square]
Next we show that convergence of F a ( W ,J n ) for all J implies convergence of E a ( W ,J n ). Together with our results from the last section, this shows that convergence of F a ( W ,J n ) for all q ∈ N , a ∈ glyph[triangle] q , and symmetric J ∈ R q × q is sufficient for metric convergence, which concludes the proof of Remark 5.4(ii).
Lemma 5.18. Let q ∈ N , a ∈ glyph[triangle] q , and c > 0 , let J ∈ R q × q be symmetric, and let U and W be two graphons. Then
$$\left | \mathcal { E } _ { \mathbf a } ( W, J ) - \mathcal { E } _ { \mathbf a } ( U, J ) \right | \leq \frac { 1 } { c } \left | \mathcal { F } _ { \mathbf a } ( W, c J ) - \mathcal { F } _ { \mathbf a } ( U, c J ) \right | + \frac { 2 \log q } { c }.$$
Proof. Using the fact that Ent ( ρ ) ≤ log q , we get by (2.17) and (2.18)
$$\left | \mathcal { E } _ { \mathbf a } ( W, J ) - \mathcal { F } _ { \mathbf a } ( W, J ) \right | \leq \log q,$$
for all J , and similarly for U . Hence
$$\dots \dots \dots, \dots \dots \dots \dots \\ \left | \mathcal { E } _ { \mathfrak { a } } ( W, J ) - \mathcal { E } _ { \mathfrak { a } } ( U, J ) \right | & = \frac { 1 } { c } \left | \mathcal { E } _ { \mathfrak { a } } ( W, c J ) - \mathcal { E } _ { \mathfrak { a } } ( U, c J ) \right | \\ & \leq \frac { 1 } { c } \left ( \left | \mathcal { F } _ { \mathfrak { a } } ( W, c J ) - \mathcal { F } _ { \mathfrak { a } } ( U, c J ) \right | + 2 \log q \right ), \\ \text{black} \quad \text{the main } \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \end{cases}$$
which proves the claim.
Proof of (iv) ⇒ (iii) in Theorem 5.3. As in the proof above, one sees that
$$\frac { 1 } { c } \mathcal { F } _ { \mathfrak { a }, \varepsilon } ( W _ { n }, c J ) - \frac { \log q } { c } \leq \mathcal { E } _ { \mathfrak { a }, \varepsilon } ( W _ { n }, J ) \leq \frac { 1 } { c } \mathcal { F } _ { \mathfrak { a }, \varepsilon } ( W _ { n }, c J ) + \frac { \log q } { c }.$$
But this clearly shows that (iv) ⇒ (iii) in Theorem 5.3.
glyph[square]
5.7. Quantitative bounds on distance between fractional quotients. In this section, we prove a quantitative bound on the distance between two different quotients of the same graphon, which will be used in the next section.
We define the L 1 distance in FP q to be
$$d _ { 1 } ( \rho, \rho ^ { \prime } ) = \sum _ { i \in [ q ] } \int _ { 0 } ^ { 1 } | \rho _ { i } ( x ) - \rho _ { i } ^ { \prime } ( x ) | \, d x ;$$
note that d 1 ( ρ, ρ ′ ) ≤ 2. We also need the definition of K -bounded tails from (5.2).
glyph[square]
Lemma 5.19. Let q ∈ N , and let W be a graphon with K -bounded tails for some function K : (0 , ∞ → ) (0 , ∞ ) . Then there exists a weakly increasing function ε K : [0 , 2] → [0 , ∞ ) such that ε K ( x ) → 0 as x → 0 and
$$d _ { 1 } ( W / \rho, W / \rho ^ { \prime } ) \leq \varepsilon _ { K } ( d _ { 1 } ( \rho, \rho ^ { \prime } ) )$$
for all ρ, ρ ′ ∈ FP q . For a b , ∈ glyph[triangle] q ,
$$d _ { 1 } ^ { H f } \left ( \widehat { \mathcal { S } } _ { \mathfrak { a } } ( W ), \widehat { \mathcal { S } } _ { \mathfrak { b } } ( W ) \right ) \leq \varepsilon _ { K } ( \| \mathfrak { a } - \mathfrak { b } \| _ { 1 } ),$$
where ‖ a -b ‖ 1 = ∑ i | a i -b i | .
Proof. Fix ε > 0. Clearly
$$\sum _ { i } | \alpha _ { i } ( \rho ) - \alpha _ { i } ( \rho ^ { \prime } ) | \leq \sum _ { i } \int _ { 0 } ^ { 1 } | \rho _ { i } ( x ) - \rho _ { i } ^ { \prime } ( x ) | \, d x = d _ { 1 } ( \rho, \rho ^ { \prime } ).$$
On the other hand, using the fact that ∑ i ρ i ( x ) = ∑ j ρ ′ i ( x ) = 1 for all x ∈ [0 , 1], we have
$$\text{On the other hand, using the fact that } \sum _ { i, j } \rho _ { i } ( x ) = \sum _ { j } \rho _ { i } ^ { \prime } ( x ) = 1 \text{ for all } x \in [ 0, 1 ], \\ \text{we have} \\ \sum _ { i, j } \left | \beta _ { i j } ( W / \rho ) - \beta _ { i j } ( W / \rho ^ { \prime } ) \right | = & \sum _ { i, j } \right | \int _ { [ 0, 1 ] ^ { 2 } } W ( x, y ) \left ( \rho _ { i } ( x ) \rho _ { j } ( y ) - \rho _ { i } ^ { \prime } ( x ) \rho _ { j } ^ { \prime } ( y ) \right ) d x \, d y \right | \\ & \leq \int _ { [ 0, 1 ] ^ { 2 } } \left | W ( x, y ) \right | \sum _ { i, j } \left | \rho _ { i } ( x ) \rho _ { j } ( y ) - \rho _ { i } ( x ) \rho _ { j } ^ { \prime } ( y ) \right | d x \, d y \\ & \quad + \int _ { [ 0, 1 ] ^ { 2 } } \left | W ( x, y ) \right | \sum _ { i, j } \left | \rho _ { i } ( x ) \rho _ { j } ^ { \prime } ( y ) - \rho _ { i } ^ { \prime } ( x ) \rho _ { j } ^ { \prime } ( y ) \right | d x \, d y \\ & = 2 \int _ { [ 0, 1 ] ^ { 2 } } \left | W ( x, y ) \right | \sum _ { i } \left | \rho _ { i } ( x ) - \rho _ { i } ^ { \prime } ( x ) \right | d x \, d y \\ & \leq 2 \int _ { [ 0, 1 ] ^ { 2 } } K ( \varepsilon / 8 ) \sum _ { i } \left | \rho _ { i } ( x ) - \rho _ { i } ^ { \prime } ( x ) \right | d x \, d y \\ & \quad + 4 \int _ { [ 0, 1 ] ^ { 2 } } \left | W ( x, y ) \right | 1 _ { | W ( x, y ) | \geq K ( \varepsilon / 8 ) } d x \, d y \\ & \leq 2 K ( \varepsilon / 8 ) d _ { 1 } ( \rho, \rho ^ { \prime } ) + \varepsilon / 2, \\ \text{showing that}$$
≤ 2 K ε/ ( 8) d 1 ( ρ, ρ ′ ) + ε/ 2 , showing that
d 1 ( W/ρ,W/ρ ′ ) ≤ (1 + 2 K ε/ ( 8)) d 1 ( ρ, ρ ′ ) + ε/ 2 , which is at most ε provided that d 1 ( ρ, ρ ′ ) ≤ ε/ (2 + 4 K ε/ ( 8)). Since ε > 0 was arbitrary, this immediately implies the existence of the desired function ε K .
The claim (5.18) follows from noting that for any a b , ∈ glyph[triangle] q and ρ ∈ FP q with α ρ ( ) = a , we can find a ρ ′ ∈ FP q with α ρ ( ′ ) = b such that d 1 ( ρ, ρ ′ ) = ‖ a -b ‖ 1 . (In fact, we can choose ρ ′ so that ρ i ≤ ρ ′ i if and only if a i ≤ b i .) glyph[square]
The above lemma can be used to show that E a ( W,J ) is continuous in a . The continuity of F a ( W,J ) then follows from noting that Ent ( ρ ) is uniformly continuous in ρ ∈ FP q with respect to d 1 (as -x log x is continuous on [0 , 1]). More explicitly, we have the following lemma.
↦
Lemma 5.20. Let q ∈ N . Then the function Ent : FP q → [0 , log q ] with ρ → Ent ( ρ ) is uniformly continuous in the metric d 1 defined in (5.17) . Explicitly,
$$| \text{Ent} ( \rho ) - \text{Ent} ( \rho ^ { \prime } ) | \leq q \tilde { f } \left ( \frac { 1 } { q } d _ { 1 } ( \rho, \rho ^ { \prime } ) \right ),$$
$$\ w h e r e \, \tilde { f } ( x ) = x ( 1 - \log x ).$$
↦
Proof. Because the function f ( x ) = -x log x is concave, for each t > 0 the function x → f ( x ) -f ( x + ) is weakly increasing. t It follows from this and f (0) = f (1) = 0 that for any x, y ∈ [0 , 1],
$$\max \, \text{on any } x, y \in \{ \mathfrak { v }, \mathfrak { v } \}, \\ | f ( x ) - f ( y ) | & \leq \max \{ | f ( 0 ) - f ( | x - y | ) |, | f ( 1 - | x - y | ) - f ( 1 ) | \} \\ & = \max \{ f ( | x - y | ), f ( 1 - | x - y | ) \} \\ & \leq f ( | x - y | ) + f ( 1 - | x - y | ) \\ & \leq f ( | x - y | ) + | x - y | \\ & = \tilde { f } ( | x - y | ), \\ \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \, \text{thumb} \end{cases}$$
where the last inequality holds because f (1 -t ) ≤ t for all t ∈ [0 , 1]. Since d 1 ( ρ, ρ ′ ) = ∑ ∫ i [0 1] , | ρ i ( x ) -ρ ′ i ( x ) | dx and ˜ is concave, we have f
$$\rho, \rho ^ { \prime } ) = \sum _ { i } J _ { [ 0, 1 ] } \left | \rho _ { i } ( x ) - \rho _ { i } ( x ) \right | d x \text{ and } f \text{ is concave, we have} \\ \frac { 1 } { q } \left | \text{Ent} ( \rho ) - \text{Ent} ( \rho ^ { \prime } ) \right | & = \frac { 1 } { q } \left | \sum _ { i = 1 } ^ { q } \int _ { 0 } ^ { 1 } \left ( f ( \rho _ { i } ( x ) ) - f ( \rho _ { i } ^ { \prime } ( x ) ) \right ) d x \right | \\ & \leq \frac { 1 } { q } \sum _ { i = 1 } ^ { q } \int _ { 0 } ^ { 1 } \tilde { f } ( | \rho _ { i } ( x ) - \rho _ { i } ^ { \prime } ( x ) | ) \, d x \\ & \leq \tilde { f } \left ( \frac { 1 } { q } d _ { 1 } ( \rho, \rho ^ { \prime } ) \right ). \\ \text{cause } \tilde { f } \text{ is continuous at } 0. \text{ this completes the proof.}$$
Because ˜ is continuous at 0, this completes the proof. f glyph[square]
- 6. Convergent sequences of uniformly upper regular graphs
In this section we prove Theorem 2.15. Theorem 2.10 will follow from this theorem and Theorem 2.17, which we prove in Section 7.
- 6.1. Preliminaries. We start by stating some of the results from [3], which will allow us to replace uniformly upper regular sequences of weighted graphs by sequences of weighted graphs with K -bounded tails. To state them, we will use the cut distance between two weighted graphs G G , ′ with identical node sets V ( G ) = V ( G ′ ) = V and identical node weights α x = α x ( G ) = α x ( G ′ ), defined as
$$d _ { \square } ( G, G ^ { \prime } ) = \max _ { S, T \subseteq V } \Big | \sum _ { ( x, y ) \in S \times T } \frac { \alpha _ { x } \alpha _ { y } } { \alpha _ { G } ^ { 2 } } \Big ( \beta _ { x y } ( G ) - \beta _ { x y } ( G ^ { \prime } ) \Big ) \Big |.$$
Note that this distance is equal to ‖ W G -W G ′ ‖ glyph[square] , where W G and W G ′ are the step functions defined in (2.2) and ‖ · ‖ glyph[square] is the cut norm defined in (2.1). Indeed, for W = W G , the supremum in (2.1) can easily be shown to be a maximum that is attained for sets S and T which are both unions of the intervals I i .
We will also use the notion of an equipartition of the vertex set V ( G ) of a weighted graph G , defined by requiring that the weights of the parts of the partition differ from an equal distribution by at most α max ( G ). Explicitly, a partition P = ( V , . . . , V 1 k ) of V ( G ) is called an equipartition if | α V i -1 k α G | ≤ α max ( G ) for all i ∈ [ k ]. The following version of the weak regularity lemma was proved in [3].
Theorem 6.1 (Theorem C.12 in [3]) . Let K : (0 , ∞ → ) (0 , ∞ ) and 0 < ε < 1 . Then there exist constants N = N K,ε ( ) and η 0 = η 0 ( K,ε ) such that the following holds
for all η ≤ η 0 : for every ( K,η ) -upper regular graph G and each natural number k ≥ N , there exists a equipartition P = ( V , . . . , V 1 k ) of V ( G ) into k parts, such that
$$d _ { \square } ( G, G _ { \mathcal { P } } ) \leq \varepsilon \| G \| _ { 1 }.$$
As a consequence, given a K -upper regular sequence of weighted graphs G n with no dominant nodes, we can find a sequence of equipartitions P n of V ( G n ) into k n classes such that the sequence of weighted graphs 6 ̂ G n = 1 ‖ G n ‖ 1 ( G n ) P n satisfies
$$( 6. 2 ) \quad d _ { \square } \left ( \frac { 1 } { \| G _ { n } \| _ { 1 } } G _ { n }, \widehat { G } _ { n } \right ) \to 0, \quad k _ { n } \frac { \alpha _ { \max } ( G _ { n } ) } { \alpha _ { G _ { n } } } \to 0, \quad \| \widehat { G } _ { n } \| _ { 1 } \leq 1$$
and
$$W ^ { \widetilde { G } _ { n } } \text{ has } K \text{-bounded tails}.$$
We call a sequence ( ̂ G n n ) ≥ 0 with these properties a regularized version of ( G n n ) ≥ 0 , and P n a regularizing partition for G n .
To see that all these conditions can be simultaneously achieved, let η n → 0 be such that G n is ( K,η n )-upper regular. Assume that ε n goes to zero slowly enough that η n ≤ η 0 ( K,ε n ) and η n N K,ε ( n ) → 0 in Theorem 6.1. Choosing k n = N K,ε ( n ), the theorem then gives a sequence of equipartitions P n of V ( G n ) into k n classes such that (6.2) holds. The weight of each class of P n is bounded from below by α G n /k n -α max ( G n ), which is asymptotically greater than η n α G n because η n k n → 0 and k n α max ( G n ) /α G n → 0. Thus, the bound (2.21) holds for P = P n , establishing that
$$\sum _ { x, y \in V ( G _ { n } ) } \frac { \alpha _ { x } ( G _ { n } ) \alpha _ { y } ( G _ { n } ) } { \alpha _ { G _ { n } } ^ { 2 } } | \beta _ { x y } ( \hat { G } _ { n } ) | { \mathbf 1 } _ { | \beta _ { x y } ( \hat { G } _ { n } ) | \geq K ( \varepsilon ) } \leq \varepsilon$$
for all ε > 0. In other words, W ̂ G n has K -bounded tails.
- 6.2. Comparing sequence of graphs to sequences of graphons. In this section we prove three lemmas, which are the main technical lemmas used to reduce many statements in Section 2 to those in Section 5. We define
$$\tilde { \mathcal { S } } _ { \mathbf a, \varepsilon } ( W ) = \{ ( \alpha, \beta ) \in \tilde { \mathcal { S } } _ { q } ( W ) \, \colon \| \alpha - \mathbf a \| _ { \infty } \leq \varepsilon \}.$$
↦
Lemma 6.2. Let G be a weighted graph, let P be an equipartition of V ( G ) into k classes such that G = G P , and let q ∈ N . Then there exist two maps φ → ρ φ and ρ → ¯ ρ from the set of configurations φ : V ( G ) → [ q ] into the set of fractional partitions FP q and from FP q to FP q , respectively, such that the following hold:
- (i) W /ρ G = W /ρ G ¯ for all ρ ∈ FP q .
- (ii) ‖ G ‖ 1 ( G/φ ) = W /ρ G φ for all φ : V ( G ) → [ q ] .
- (iii) For each ρ ∈ FP q there exists a φ : V ( G ) → FP q such that
$$d _ { 1 } ( \rho _ { \phi }, \bar { \rho } ) \leq q k \frac { \alpha _ { \max ( G ) } } { \alpha _ { G } }.$$
Proof. Let P = ( V , . . . , V 1 k ), and assume that the vertices in G are ordered in such a way that V 1 = { 1 2 , , . . . , | V 1 |} , V 2 = {| V 1 | + 1 , . . . , | V 1 | + | V 2 |} , etc. Let x µ = α V µ ( G /α ) G for µ ∈ [ k ], and let I 1 , . . . , I k ⊆ [0 , 1] be consecutive intervals of length x 1 , . . . , x k .
6 If ‖ G n ‖ 1 = 0, we choose ̂ G n to have edge weights 0.
For ρ ∈ FP q define ¯ by averaging ρ ρ over the intervals I µ , i.e., ¯ ( ρ i x ) = 1 x µ ∫ I µ ρ i ( y ) dy if x ∈ I µ . Since W G is constant on sets of the form I µ × I ν for µ, ν ∈ [ k ], we clearly have W /ρ G = W /ρ G ¯, proving (i).
Next, given φ : V ( G ) → [ q ], let ρ φ be the fractional q -partition defined by
$$( 6. 5 ) \quad \ \ ( \rho _ { \phi } ) _ { i } ( x ) = \frac { 1 } { \alpha _ { V _ { \mu } ( G ) } } \sum _ { u \in V _ { \mu } ( G ) } \alpha _ { u } ( G ) { \mathbf 1 } _ { \phi ( u ) = i } \quad \text{ when } x \in I _ { \mu }.$$
Using the fact that β uv ( G ) is constant on sets of the form V µ × V ν , it is then easy to check that ‖ G ‖ 1 G/φ = W /ρ G φ , proving (ii).
To prove (iii), consider ρ ∈ FP q , and let α µ,i = ∫ I µ ρ i ( x dx ) . We then decompose V µ into q sets V µ,i such that
$$| \alpha _ { \mu, i } - \alpha _ { V _ { \mu, i } } ( G ) / \alpha _ { G } | \leq \frac { \alpha _ { \max } ( G ) } { \alpha _ { G } }$$
for all i and µ . Setting φ = i on ⋃ µ V µ,i , we then get a map φ : V ( G ) → [ q ] such that (6.4) holds. glyph[square]
Lemma 6.3. Let q ∈ N and let ( G n n ) ≥ 0 be a uniformly upper regular sequence of weighted graphs. If ̂ G n is a regularized version of G n , then
$$d _ { 1 } ^ { \text{Hf} } \Big ( \mathcal { S } _ { q } ( G _ { n } ), \widehat { \mathcal { S } } _ { q } ( W ^ { \widehat { G } _ { n } } ) \Big ) & \to 0.$$
Proof of Lemma 6.3. We start by showing that
$$d _ { 1 } ( G _ { n } / \phi, \| \widehat { G } _ { n } \| _ { 1 } ( \widehat { G } _ { n } / \phi ) ) \leq q ^ { 2 } d _ { \Box } \left ( \frac { 1 } { \| G _ { n } \| _ { 1 } } G _ { n }, \widehat { G } _ { n } \right )$$
whenever ̂ G n is a regularized version of G n . Indeed, from the definition of the cut distance (6.1) it is easy to see that d 1 ( G/φ,G /φ ′ ) ≤ q 2 d glyph[square] ( 1 ‖ G ‖ 1 G, 1 ‖ G ′ ‖ 1 G ′ ) whenever G and G ′ have identical node sets and node weights and for any φ : V ( G ) → [ q ]. More generally, for any λ ≥ 0, we have d 1 ( G/φ,λ G /φ ( ′ )) ≤ q 2 d glyph[square] ( 1 ‖ G ‖ 1 G, λ ‖ G ′ ‖ 1 G ′ ).
Setting λ n = ‖ ̂ G n ‖ 1 proves (6.7).
Next we observe that by Lemma 6.2(ii),
$$\| \widehat { G } _ { n } \| _ { 1 } \mathcal { S } _ { q } ( \widehat { G } _ { n } ) \subseteq \widehat { \mathcal { S } } _ { q } ( W ^ { \widehat { G } _ { n } } ).$$
On the other hand, given W ̂ G n /ρ ∈ S ̂ q ( W ̂ G n ), we may use Lemma 6.2 and Lemma 5.19 to find a quotient ̂ G /φ n ∈ S q ( ̂ G n ) such that
$$d _ { 1 } \left ( W ^ { \widehat { G } _ { n } } / \rho, \| \widehat { G } _ { n } \| _ { 1 } \widehat { G } _ { n } / \phi \right ) & = d _ { 1 } ( W ^ { \widehat { G } _ { n } } / \bar { \rho }, W ^ { \widehat { G } _ { n } } / \rho _ { \phi } ) \\ & \leq \varepsilon _ { K } \left ( q k _ { n } \frac { \alpha _ { \max ( G _ { n } ) } } { \alpha _ { G _ { n } } } \right ),$$
where k n is the number of classes in the regularizing partition corresponding to ̂ G n and K is the function from (6.3).
Taking into account the second bound in (6.2), the bound (6.9) together with (6.8) implies that
$$d _ { 1 } ^ { \text{Hf} } \Big ( \| \widehat { G } _ { n } \| _ { 1 } \mathcal { S } _ { q } ( \widehat { G } _ { n } ), \widehat { \mathcal { S } } _ { q } \Big ( W ^ { \widehat { G } _ { n } } \Big ) \Big ) \to 0,$$
while (6.7) and the fact that d glyph[square] ( 1 ‖ G n ‖ 1 G ,G n ̂ n ) → 0 show that
$$d _ { 1 } ^ { \text{Hf} } ( \mathcal { S } _ { q } ( G _ { n } ), \| \widehat { G } _ { n } \| _ { 1 } \mathcal { S } _ { q } ( \widehat { G } _ { n } ) ) \to 0.$$
Together, these two bounds imply (6.6).
glyph[square]
6.3. Proof of (i) ⇒ (ii) and (ii) ⇒ (iii) in Theorem 2.10. Recall that by Theorem 2.14, convergence in metric implies uniform upper regularity. As a consequence, (i) ⇒ (ii) follows immediately from Lemma 6.3. Indeed, δ glyph[square] ( 1 ‖ G n ‖ 1 W G n , W ) → 0
by (6.2), which implies that δ glyph[square] ( W ̂ G n , W ) → 0. We then use the bound (6.6) from Lemma 6.3 in conjunction with (5.8) from Lemma 5.11 to conclude that
$$d _ { 1 } ^ { H f } \Big { ( } \mathcal { S } _ { q } ( G _ { n } ), \widehat { \mathcal { S } } _ { q } ( W ) \Big { ) } \rightarrow 0,$$
completing the proof of (i) ⇒ (ii) in Theorem 2.10.
The implication (ii) ⇒ (iii) follows with the help of Theorem 3.3, the closedness of ̂ S q ( W ), and the representation (5.7).
- 6.4. Proof of the equivalence (i) ⇔ (ii) ⇔ (iii) in Theorem 2.15. Recall that by Theorem 2.13, for a uniformly upper regular sequence convergence in metric implies convergence in metric to some graphon W , so the above proof of the implication (i) ⇒ (ii) for Theorem 2.10 also proves it for Theorem 2.15. The implication (ii) ⇒ (iii) follows from Theorem 3.3.
It remains to show (iii) ⇒ (i); i.e., under the assumption of uniform upper regularity, convergence of the microcanonical ground state energies implies convergence in metric. Assume for the sake of contradiction that G n does not converge in metric. By Theorem 2.13 this implies that there are two subsequences G ′ n and G ′′ n and two graphons U and W such that G ′ n → U and G ′′ n → W in metric while δ glyph[square] ( U, W ) > 0. By the already proved (i) ⇒ (iii) in Theorem 2.10, the microcanonical ground state energies of G ′ n and G ′′ n converge to those of U and W , and by our assumption that G n has convergent ground state energies, this implies that U and W have identical ground state energies, contradicting Theorem 5.13.
- 6.5. Convergence in metric implies LD convergence. Our main result in this section is the following theorem, which by (5.6) proves the implication (i) ⇒ (iv) in Theorem 2.10 and hence also Theorem 2.15.
Theorem 6.4. Let q ∈ N and let ( G n n ) ≥ 0 be a uniformly upper regular sequence of weighted graphs such that G n converges to a graphon W in metric and G n has vertex weights one. Then the limit (2.16) exists with
$$I _ { q } ( ( \alpha, \beta ) ) = \log q - \lim _ { \varepsilon \to 0 } \sup _ { \substack { \rho \in \text{FP} _ { q } \\ d _ { 1 } ( W / \rho, ( \alpha, \beta ) ) \leq \varepsilon } } \text{Ent} ( \rho ).$$
Proof. To prove the theorem, we will need to calculate probabilities of the form P q,G [ d 1 (( α, β ) , G/φ ) ≤ ε ] for graphs G that are near to W in the normalized cut distance. Assume for the moment that G is well approximated by a weighted graph whose edge weights are constant over large blocks. More precisely, assume that there exists an equipartition P = ( V , . . . , V 1 k ) such that G and G P are close in the cut norm. Under such a condition, the quotients of G are close to those of G P , provided they are suitably normalized. More precisely, if we define ̂ G to be the normalized weighted graph ̂ G = 1 ‖ G ‖ 1 ( G ) P , the quotients of G are close
to those of ̂ G multiplied by ‖ G ‖ 1 (see (6.7)). Consider thus the probabilities P q,G [ d 1 (( α, β ) , ‖ G ‖ 1 ( ̂ G/φ )) ≤ ε ] .
Since the edge weights of ̂ G are constant on sets of the form V µ × V ν , we have that ̂ G/φ = ̂ G/φ ′ if for all i ∈ [ q ] and all µ ∈ [ k ], the number of vertices in V µ that are mapped to i ∈ [ q ] is the same in φ and φ ′ . Denote this number by k i,µ = k i,µ ( φ ). The number of configurations φ : V ( G ) → [ q ] with given k i,µ is
$$N ( \{ k _ { i, \mu } \} ) & = \prod _ { \mu = 1 } ^ { k } \frac { n _ { \mu }! } { k _ { 1, \mu }! \dots k _ { q, \mu }! },$$
where n µ = | V µ | . Approximating k ! as ( k/e ) k (we analyze the error term below), we have
$$N ( \{ k _ { i, \mu } \} ) \approx \exp \Big ( - \sum _ { \mu = 1 } ^ { k } \sum _ { i = 1 } ^ { q } - k _ { i, \mu } \log \Big ( \frac { k _ { i, \mu } } { n _ { \mu } } \Big ) \Big ) = e ^ { | V ( G ) | \text{Ent} ( \rho _ { \phi } ) },$$
where ρ φ is the fractional partition defined in (6.5). Observing that the number of choices for { k µ,i } is polynomial in | V ( G ) , | and hence will not contribute to I q (again we bound the error later), and noting further that ‖ ̂ G ‖ 1 ( ̂ G/φ ) = W /ρ ̂ G φ by Lemma 6.2, we then approximate P q,G [ d 1 (( α, β ) , ‖ G ‖ 1 ( ̂ G/φ )) ≤ ε ] by
$$q ^ { - | V ( G ) | } \max _ { \substack { \phi \colon V ( G ) \to [ q ] \\ d _ { 1 } ( ( \alpha, \beta ), W ^ { \hat { G } } / \rho _ { \phi } ) \leq \varepsilon } } e ^ { | V ( G ) | \text{Ent} ( \rho _ { \phi } ) }.$$
Taking into account that (again by Lemma 6.2) any fractional quotient W /ρ ̂ G can be well approximated by a quotient of the form W /ρ ̂ G φ , we obtain the theorem.
The formal proof proceeds as follows. Let K : (0 , ∞ → ) (0 , ∞ ) and ( η n n ) ≥ 0 be such that lim n →∞ η n = 0 and G n is ( K,η n )-upper regular. Fix ε > 0. By Theorem 6.1 and Definition 2.12, there are constants k ∈ N and n 0 < ∞ such that for each n ≥ n 0 there exists an equipartition P = P n = ( V , . . . , V 1 k ) of V ( G n ) into k parts such that
$$q ^ { 2 } d _ { \square } \left ( G _ { n }, ( G _ { n } ) _ { \mathcal { P } } \right ) \leq \frac { \varepsilon } { 2 } \| G _ { n } \| _ { 1 }$$
and W ̂ G n with ̂ G n = 1 ‖ G n ‖ 1 ( G n ) P has K -bounded tails. Let G = G and G = G . By the bound (6.7), n ̂ ̂ n
$$d _ { 1 } ( G / \phi, \| \widehat { G } \| _ { 1 } ( \widehat { G } / \phi ) ) \leq q ^ { 2 } d _ { \square } \left ( \frac { 1 } { \| G \| _ { 1 } } G, \widehat { G } \right ) \leq \frac { \varepsilon } { 2 },$$
implying that
$$\mathcal { P } _ { q, G } [ d _ { 1 } ( ( \alpha, \beta ), \| \widehat { G } \| _ { 1 } ( \widehat { G } / \phi ) ) \leq \varepsilon / 2 ] & \leq \mathcal { P } _ { q, G } [ d _ { 1 } ( ( \alpha, \beta ), G / \phi ) \leq \varepsilon ] \\ & \leq \mathcal { P } _ { q, G } [ d _ { 1 } ( ( \alpha, \beta ), \| \widehat { G } \| _ { 1 } ( \widehat { G } / \phi ) ) \leq 3 \varepsilon / 2 ].$$
Given a configuration φ : V ( G ) → [ q ], let k i,µ ( φ ) be the number of vertices v ∈ V µ such that φ v ( ) = i , and let N ( { k i,µ } ) be the number of maps φ leading to the same k i,µ ; see (6.10) above. Bounding the number of choices for { k µ,i } by | V ( G ) | qk and
observing that ‖ ̂ G ‖ 1 ( ̂ G/φ ) = W /ρ ̂ G φ , we then bound
$$\max _ { \phi \colon d _ { 1 } ( ( \alpha, \beta ), W ^ { \widehat { G } } / \rho _ { \phi } ) \leq \varepsilon / 2 } N ( \{ k _ { i, \mu } ( \phi ) \} ) \leq q ^ { | V ( G ) | } \mathcal { P } _ { q, G } \left [ d _ { 1 } ( ( \alpha, \beta ), G / \phi ) \leq \varepsilon \right ]$$
$$\leq | V ( G ) | ^ { q k } \max _ { \phi \colon d _ { 1 } ( ( \alpha, \beta ), W ^ { \hat { G } } / \rho _ { \phi } ) \leq 3 \varepsilon / 2 } N ( \{ k _ { i, \mu } ( \phi ) \} ).$$
$$& \vdots \varepsilon \, ] \\ & N ( \{ k _ { i, \mu } ( \phi ) \} ).$$
Since ( k/e ) k ≤ k ! ≤ ek k/e ( ) k ,
$$\left ( \frac { 1 } { e | V ( G ) | } \right ) ^ { k q } e ^ { \text{Ent} ( \rho _ { \phi } ) | V ( G ) | } \leq N ( \{ k _ { i, \mu } \} ) \leq ( e | V ( G ) | ) ^ { k } e ^ { \text{Ent} ( \rho _ { \phi } ) | V ( G ) | },$$
implying
$$\cfrac { \ c c } { \ c c } \ sigma _ { r } \sim$$
and
$$\overset { \dots } { \overset { \quad } { \quad } } q$$
Next we use Lemma 6.2 to approximate an arbitrary fractional partition ρ ∈ FP q by a fractional partition of the form ρ φ , with a error of qk/ V | ( G ) | in the d 1 distance. With the help of Lemmas 5.19 and 5.20, we can ensure that for n (and hence | V ( G ) | = | V ( G n ) ) large enough, the resulting errors in | d 1 (( α, β ) , W ̂ G /ρ ) and Ent ( ρ ) are bounded by ε/ 4 and ε , respectively, leading to the lower bound
$$q ^ { | V ( G ) | } \mathcal { P } _ { q, G }$$
To conclude the proof, we note that if G n → W in metric, then δ glyph[square] ( W ̂ G n , W ) → 0. Taking into account Lemma 5.19 and the fact that the entropy Ent ( ρ ) is invariant under measure preserving transformations, we get that for n sufficiently large
$$\begin{array} {$$
As a consequence
$$- \log q - \varepsilon + \sup _ { \substack { \rho$$
$$- \log q - \varepsilon + \sup _ { \substack { \rho \in \text{FP} _ { q } \\ d _ { 1 } ( ( \alpha, \beta ), W / \rho ) \leq \varepsilon / 8 } } \text{Ent} ( \ell$$
$$\text{Ent} ( \rho ) \leq \liminf _ { n \to \infty } \frac { \log \mathcal { P } _ { q, G _ { n } } \left [ d _ { 1 } ( ( \alpha, \beta ), G _ { n } / \phi ) \leq \varepsilon \right ] } { | V ( G _ { n } ) | }$$
$$& \leq \lim \sup _ { n \to \infty } \frac { \log \mathcal { P } _ { q, G _ { n } } \left [ d _ { 1 } ( ( \alpha, \beta ), G _ { n } / \phi ) \leq \varepsilon \right ] } { | V ( G _ { n } ) | } \\ & \leq - \log q + \sup _ { \substack { \rho \in \text{FP} _ { q } \\ d _ { 1 } ( ( \alpha, \beta ), W / \rho ) \leq 2 \varepsilon } } \text{Ent} ( \rho ). \\ \bigcap \in \varepsilon$$
Sending ε → 0 completes the proof.
glyph[square]
6.6. Completion of the proofs of Theorems 2.15 and 2.10. To complete the proof of Theorem 2.15, we still need to show for graphs with node weights one, statements (iv) and (v) are equivalent to the other statements of the theorem. We will also have to establish the limit expressions given in Theorem 2.10.
By Theorem 6.4, we know the implication (i) ⇒ (iv) in Theorem 2.15, and we also know that the rate function is given by (2.19), as claimed in Theorem 2.10(iv). Finally, by Theorem 2.9(ii), in Theorem 2.15 statement (iv) implies statement (v), and by Lemma 3.2(iii), this in turn implies statement (iii) of Theorem 2.15, completing the proof of Theorem 2.15.
Theorem 2.17, which we prove in the next section, is nearly enough to deduce Theorem 2.10 from Theorem 2.15. The only missing piece is the explicit limit expressions stated in Theorem 2.10 for how limiting quotients, ground state energies, free energies, and large deviations rate function depend on W . So far, we have dealt with all of them except for the microcanonical free energies. Since we already have shown that convergence in metric for graphs with node weights one implies LD convergence with rate function given by (2.19), this follows from the following lemma.
Lemma 6.5. Let W be a graphon, and let ( G n n ) ≥ 0 be a sequence of weighted graphs. If ( G n n ) ≥ 0 is LD convergent with rate function I q = I q ( · , W ) as defined in (2.19) , then the microcanonical free energies of ( G n n ) ≥ 0 converge to those of W , as defined in (2.18) .
Proof. By Theorem 2.9(ii), the assumption implies convergence of the microcanonical free energies, with the limiting free energies given by
$$\text{rgies, with the limiting free energies given by} \\ F _ { \mathbf a } ( J ) & = \inf _ { ( \alpha, \beta ) \in \widehat { \mathcal { S } } _ { \mathbf a } } \left ( - \langle \beta, J \rangle + I _ { q } ( ( \alpha, \beta ), W ) \right ) - \log q. \\ & = \inf _ { ( \alpha, \beta ) \in \widehat { \mathcal { S } } _ { \mathbf a } } \inf _ { \rho \in \mathbf F P _ { q } } \left ( - \langle \beta, J \rangle - \mathbf E n t ( \rho ) \right ) \\ & = \inf _ { \rho \in \mathbf F P _ { q } } \left ( - \langle \beta ( W / \rho ), J \rangle - \mathbf E n t ( \rho ) \right ) \\ & = \mathcal { F } _ { \mathbf a } ( W, J ), \\ \text{ed.}$$
as desired.
glyph[square]
## 7. Inferring uniform upper regularity
In this section we prove Theorem 2.17. We have already proved a number of implications between the four conditions in the theorem statement. Specifically, from
Lemma 6.5 we know that (iv) implies (iii). From Lemma 3.2(iii) (and the analogous assertion that E a ( W,J ) = lim λ →∞ λ -1 F a ( λJ ) with an essentially identical proof) we know that (iii) implies (i). From Theorem 3.3 we deduce that (ii) implies (i). Thus, it remains to show that (i) implies that ( G n n ) ≥ 0 is uniformly upper regular, which is the statement of the following proposition.
Proposition 7.1. Let ( G n n ) ≥ 0 be a sequence of weighted graphs with no dominant nodes and W a graphon. If the microcanonical ground state energies of G n converge to those of W , then ( G n n ) ≥ 0 is uniformly upper regular.
In order to prove this proposition, we introduce a notion of equipartition upper regularity , where instead of considering all partitions of the vertex set with no part having weight smaller than ηα G , we consider all equipartitions into q parts. Following the definition, we prove a lemma which says that the two notions of upper regularity are qualitatively equivalent.
Definition 7.2. Let K : (0 , ∞ → ) (0 , ∞ ) be any function and let q ∈ N . A weighted graph G is ( K,q ) -equipartition upper regular if α max ( G ) ≤ α G / (2 q ) and for every ε > 0 and equipartition P of V ( G ) into q parts,
$$\sum _ { i, j \in [ q ] } | \beta _ { i j } ( G / \mathcal { P } ) | \mathbf 1 _ { | \beta _ { i j } ( G / \mathcal { P } ) | \geq K ( \varepsilon ) \alpha _ { i } ( G / \mathcal { P } ) \alpha _ { j } ( G / \mathcal { P } ) } \leq \varepsilon.$$
Equipartitions are defined in Section 6.1. We use β ij ( G/ P ) and β ij ( G/φ ) as synonyms (see (2.5) for the definition), where the function φ : V ( G ) → [ q ] defines the partition P as the preimages of the points in [ q ], but recall from (2.20) that β ij ( G P ) is normalized differently from β ij ( G/ P ).
Using (2.4), (2.5), (2.20), and (2.21), it is clear that if G is ( K,η )-upper regular, then it is ( K,q )-equipartition upper regular for q ≤ 1 (2 / η ), since in every equipartition of V ( G ) into q parts, the weight of each part is at least
$$\alpha _ { G } / q - \alpha _ { \max } ( G ) \geq \alpha _ { G } / q - \eta \alpha _ { G } \geq \eta \alpha _ { G }.$$
Conversely, we also have the following.
Lemma 7.3. Let K ′ : (0 , ∞ → ) (0 , ∞ ) be any function, and let
$$K ( \varepsilon ) = \max \{ 4 \varepsilon ^ { - 1 } K ^ { \prime } ( \varepsilon / 4 ), 1 6 \varepsilon ^ { - 2 } \}.$$
Let η > 0 and q 0 = η -2 . If a weighted graph G is ( K , q ′ ′ ) -equipartition upper regular for some q ′ ≥ q 0 , then G is ( K,η ) -upper regular.
Proof. By scaling the vertex weights, we may assume without loss of generality that α G = 1. Let P = ( V , . . . , V 1 q ) be a partition of V ( G ) into q classes, where α V i ≥ η for each i ∈ [ q ], and let ε > 0. Define
$$S \coloneqq \{ ( i, j ) \in [ q ] \times [ q ] \, \colon | \beta _ { i j } ( G / \mathcal { P } ) | \geq K ( \varepsilon ) \alpha _ { i } ( G / \mathcal { P } ) \alpha _ { j } ( G / \mathcal { P } ) \}.$$
We need to prove that
$$\sum _ { ( i, j ) \in S } | \beta _ { i j } ( G / \mathcal { P } ) | \leq \varepsilon.$$
$$\varepsilon.$$
Since α i ( G/ P ) ≥ η for all i ∈ [ q ] and ∑ i,j ∈ [ q ] | β ij ( G/ P | ≤ ) 1, we have
$$( 7. 2 ) \quad \eta ^ { 2 } | S | \leq \sum _ { ( i, j ) \in S } \alpha _ { i } ( G / \mathcal { P } ) \alpha _ { j } ( G / \mathcal { P } ) \leq \frac { 1 } { K ( \varepsilon ) } \sum _ { ( i, j ) \in S } | \beta _ { i j } ( G / \mathcal { P } ) | \leq \frac { 1 } { K ( \varepsilon ) }.$$
Thus if η K ε 2 ( ) > 1, then S = ∅ and (7.1) trivially holds. So assume from now on that η K ε 2 ( ) ≤ 1. Since K ε ( ) ≥ 16 ε -2 by assumption, we have
$$\varepsilon & \geq 4 \eta.$$
For each x ∈ V ( G ), define the (weighted) degree of x to be
$$\deg _ { x } ( G ) = \sum _ { y \in V ( G ) } \alpha _ { y } ( G ) \left | \beta _ { x y } ( G ) \right |.$$
We construct an equipartition P ′ of V ( G ) as follows. For each i ∈ [ q ], we partition V i into subsets V i, 0 , V i, 1 , . . . , V i,k i such that
$$\alpha _ { V _ { i, k } } ( G ) \in \left [ \frac { 1 } { q ^ { \prime } } - \alpha _ { \max } ( G ), \frac { 1 } { q ^ { \prime } } + \alpha _ { \max } ( G ) \right ] \quad \text{for } 1 \leq k \leq k _ { i },$$
α V i, 0 ( G < ) 1 /q ′ , and the vertices in V i, 0 all have the lowest degree present in V i . We will do this in such a way that for all i = 1 . . . , q and all j = 1 , . . . , k i ,
$$\frac { 1 } { q ^ { \prime } } \left ( \sum _ { i ^ { \prime } = 1 } ^ { i - 1 } k _ { i ^ { \prime } } + j \right ) \leq \sum _ { i ^ { \prime } = 1 } ^ { i - 1 } \sum _ { j ^ { \prime } = 1 } ^ { k _ { i ^ { \prime } } } \alpha _ { V _ { i ^ { \prime }, j ^ { \prime } } ( G ) } + \sum _ { j ^ { \prime } = 1 } ^ { j } \alpha _ { V _ { i, j ^ { \prime } } ( G ) } \leq \frac { 1 } { q ^ { \prime } } \left ( \sum _ { i ^ { \prime } = 1 } ^ { i - 1 } k _ { i ^ { \prime } } + j \right ) + \alpha _ { \max } ( G ).$$
For example, we can do this greedily by sorting all the vertices in V i according to their degrees and then placing them into V i,j for j = 1 2 , , . . . in decreasing order by degree until the lower bound in the above inequality is satisfied. Since the last vertex added contributed at most α max ( G ) to the sum, we are guaranteed to have the upper bound as well. When the total weight left in V i is too small to fill another V i,j , we are necessarily left with a remainder V i, 0 that has weight less than 1 /q ′ and contains only vertices with the lowest degree from V i .
Next, consider the remainder sets V 1 0 , , . . . , V q, 0 , whose union we denote by V 0 . By construction, either V 0 is empty, in which case we do nothing, or α V 0 lies between k /q 0 ′ -α max ( G ) and k /q 0 ′ , where k 0 = q ′ -∑ i ≥ 1 k i . Proceeding again greedily (this time ignoring the degrees), we decompose V 0 into k 0 sets V 0 1 , , . . . , V 0 ,k 0 with weights between 1 /q ′ -α max ( G ) and 1 /q ′ + α max ( G ). The sets V i,j with 0 ≤ i ≤ q and j ≥ 1 then form an equipartition P ′ of V ( G ) into q ′ sets.
Define S ′ to be the set of pairs ( u, v ) ∈ [ q ′ ] × [ q ′ ] for which we can find an ( i, j ) ∈ S such that u is the label of a subclass of V i and v is the label of a subclass of V j . In other words, S ′ refines the set S from [ q ] × [ q ] to [ q ′ ] × [ q ′ ], except that it does not necessarily contain pairs ( u, v ) for which u or v is in V 0 (since the remainder sets used to form V 0 do not necessarily come from a single part of P ). Thus, we have
$$\sum _ { ( i, j ) \in S } | \beta _ { i j } ( G / \mathcal { P } ) | \leq \sum _ { ( u, v ) \in S ^ { \prime } } | \beta _ { u v } ( G / \mathcal { P } ^ { \prime } ) | + \frac { 1 } { \| G \| _ { 1 } } \sum _ { \substack { x, y \in V ( G ) \\ r \in V \cdot \, \text{or} \, u \in V \cdot } } \alpha _ { x } ( G ) \alpha _ { y } ( G ) | \beta _ { x y } ( G ) |$$
$$\leq \sum _ { ( u, v ) \in S ^ { \prime } } | \beta _ { u v } ( G / \mathcal { P } ^ { \prime } ) | + \frac { 2 } { \| G \| _ { 1 } } \sum _ { x \in V _ { 0 } } \alpha _ { x } ( G ) \deg _ { x } ( G ).$$
$$\sum _ { ( i, j ) \in S } | \beta _ { i j } ( G / \mathcal { P } ) | & \leq \sum _ { ( u, v ) \in S ^ { \prime } } | \beta _ { u v } ( G / \mathcal { P } ^ { \prime } ) | + \frac { 1 } { \| G \| _ { 1 } } \sum _ { \substack { x, y \in V ( G ) \\ x \in V _ { 0 } \text{ or } y \in V _ { 0 } } } \alpha _ { x } ( G ) \alpha _ { y } ( G ) | \beta _ { x y } ( G ) | \\ ( 7. 4 ) & \leq \sum _ { \substack { \iota, \iota, \iota \searrow \circ \mathfrak { e } ^ { \prime } } } | \beta _ { u v } ( G / \mathcal { P } ^ { \prime } ) | + \frac { 2 } { \| G \| _ { 1 } } \sum _ { \substack { \iota \sim V \iota } } \alpha _ { x } ( G ) \deg _ { x } ( G ).$$
It remains to prove that (7.4) is at most ε .
We begin with the second term. For each i , since the vertices in V i, 0 are among the lowest degree vertices of V i , α V i, 0 ( G < ) 1 /q ′ , and α V i ( G ) ≥ η , we have
$$\sum _ { x \in V _ { i, 0 } } \alpha _ { x } ( G ) \deg _ { x } ( G ) \leq \frac { \alpha _ { V _ { i, 0 } } ( G ) } { \alpha _ { V _ { i } } ( G ) } \sum _ { x \in V _ { i } } \alpha _ { x } ( G ) \deg _ { x } ( G ) \leq \frac { 1 } { q ^ { \prime } \eta } \sum _ { x \in V _ { i } } \alpha _ { x } ( G ) \deg _ { x } ( G ).$$
Summing over i ∈ [ q ] and using ∑ x ∈ V ( G ) α x ( G ) deg x ( G ) = ‖ G ‖ 1 , we see that the second term in (7.4) is at most
$$\frac { 2 } { q ^ { \prime } \eta } \leq \frac { 2 } { q _ { 0 } \eta } = 2 \eta \leq \frac { \varepsilon } { 2 }$$
by (7.3).
To bound the first term in (7.4), we decompose the sum into a sum of those terms for which | β u,v ( G/ P ′ ) | is larger than K ε/ ′ ( 4) α u ( G/ P ′ ) α v ( G/ P ′ ) and a sum of those for which it is at most this large. Since G is ( K , q ′ ′ )-equipartition upper regular, we can bound the first sum by ε/ 4, while the second is clearly bounded by the sum of K ε/ ′ ( 4) α u ( G/ P ′ ) α v ( G/ P ′ ) over ( u, v ) ∈ S ′ . Taking into account that K ε/ ′ ( 4) ≤ ε 4 K ε ( ), this proves that
$$\sum _ { ( u, v ) \in S ^ { \prime } } | \beta _ { u v } ( G / \mathcal { P } ^ { \prime } ) | \leq \frac { \varepsilon } { 4 } + \frac { \varepsilon K ( \varepsilon ) } { 4 } \sum _ { ( u, v ) \in S ^ { \prime } } \alpha _ { u } ( G / \mathcal { P } ^ { \prime } ) \alpha _ { v } ( G / \mathcal { P } ^ { \prime } ).$$
Since
$$\sum _ { ( u, v ) \in S ^ { \prime } } \alpha _ { u } ( G / \mathcal { P } ^ { \prime } ) \alpha _ { v } ( G / \mathcal { P } ^ { \prime } ) \leq \sum _ { ( i, j ) \in S } \alpha _ { i } ( G / \mathcal { P } ) \alpha _ { j } ( G / \mathcal { P } ) \leq \frac { 1 } { K ( \varepsilon ) }$$
by (7.2), we have shown that the first term in (7.4) is bounded by ε/ 2, which completes our proof. glyph[square]
Proof of Proposition 7.1. To show that ( G n n ) ≥ 0 is uniformly upper regular by Lemma 7.3, it suffices to show that there is some K : (0 , ∞ → ) (0 , ∞ ) and some sequence of integers q n →∞ such that G n is ( K,q n )-equipartition upper regular.
Equivalently, this amounts to showing that we can find some q n →∞ so that α max ( G n ) /α G n ≤ 1 / (2 q n ) and for every ε > 0, there is some real K > 0 so that for sufficiently large 7 n , we have
$$\max _ { \substack { \phi \colon V ( G _ { n } ) \to [ q _ { n } ] \\ \text{equipartition} } } \sum _ { i, j \in [ q _ { n } ] } | \beta _ { i j } ( G _ { n } / \phi ) | \mathbf 1 _ { | \beta _ { i j } ( G _ { n } / \phi ) | \geq K \alpha _ { i } ( G _ { n } / \phi ) \alpha _ { j } ( G _ { n } / \phi ) } \leq \varepsilon.$$
For any weighted graph G with α max ( G /α ) G ≤ 1 / (2 q ) and equipartition φ : V ( G ) → [ q ], we have
$$\sum _ { i, j \in [ q ] } | \beta _ { i j } ( G / \phi ) | \mathbf 1 _ { | \beta _ { i j } ( G / \phi ) | \geq K \alpha _ { i } ( G / \phi ) \alpha _ { j } ( G / \phi ) } = - E _ { \phi } ( G, J )$$
by the definition (2.9) of E φ ( G,J ), where J ∈ {-1 0 1 , , } q × q is given by
$$J _ { i j } = \text{sign} ( \beta _ { i j } ( G / \phi ) ) \mathbf 1 _ { | \beta _ { i j } ( G / \phi ) | \geq K \alpha _ { i } ( G / \phi ) \alpha _ { j } ( G / \phi ) }.$$
7 This is equivalent to the same claim for all n since we can increase K to account for the first finitely many values of n .
Using α i ( G/φ ) ≥ 1 /q -α max ( G /α ) G ≥ 1 / (2 q ), we obtain
$$\mathfrak { z } / \phi ) \geq 1 / q - \alpha _ { \max } ( G ) / \alpha _ { G } \geq 1 / ( 2 q ), \, \text{we obtain} \\ \sum _ { i, j \in [ q ] } | J _ { i j } | & = \sum _ { i, j \in [ q ] } \mathfrak { 1 } _ { | \beta _ { i j } ( G / \phi ) | \geq K \alpha _ { i } ( G / \phi ) \alpha _ { j } ( G / \phi ) } \\ & \leq \sum _ { i, j \in [ q ] } \frac { | \beta _ { i j } ( G / \phi ) | } { K \alpha _ { i } ( G / \phi ) \alpha _ { j } ( G / \phi ) } \\ & \leq \frac { 4 q ^ { 2 } } { K } \sum _ { i, j \in [ q ] } | \beta _ { i j } ( G / \phi ) | \leq \frac { 4 q ^ { 2 } } { K }. \\ \text{rom the definition } ( 2. 1 2 ) \, \text{of } E _ { \text{a.e} } ( G, J ) \text{ that}$$
It follows from the definition (2.12) of E a ,ε ( G,J ) that
$$\emph { n o w o m e u e m m o n } \{ 2. 1 2 \} \text{$E_{a,\varepsilon}(\mathcal{O}$,$J$} \max} \\ \text{left side of } ( 7. 5 ) & \leq \max _ { \phi \colon V ( G _ { n } ) \to \{ q _ { n } \} } \max _ { \substack { \phi \colon V ( G _ { n } ) \to \{ q _ { n } \} } \substack { ( - E _ { \phi } ( G _ { n }, J ) ) \\ \text{equipartition symmetric} \\ \sum _ { i, j } | J _ { i j } | \leq 4 q _ { n } ^ { 2 } / K } } ( - E _ { \max } \max _ { \substack { \jmath \in \{ - 1, 0, 1 \} ^ { q _ { n } \times q _ { n } } \\ \text{ symmetric} \\ \sum _ { i, j } | J _ { i j } | \leq 4 q _ { n } ^ { 2 } / K } } ( G _ { n }, J ) ) \\ \text{here } \mathfrak { a }. = ( 1 / a. \quad 1 / a. \, ) \, \cong \, \wedge ^ { q _ { n } } \text{ and we also write } \mathfrak { a } = ( 1 / a \quad 1 / a ) \, \cong \, \wedge ^ { q }$$
(here q n = (1 /q n , . . . , 1 /q n ) ∈ glyph[triangle] q n , and we also write q = (1 /q, . . . , 1 /q ) ∈ glyph[triangle] q below).
Since the microcanonical ground state energies of G n converge to those of W , we know that for every q ∈ N , we can find ε 0 ( q ) > 0 and n 0 ( q ) so that
$$- E _ { \mathbf q, \varepsilon } ( G _ { n }, J ) \leq - \mathcal { E } _ { \mathbf q } ( W, J ) + 1 / q$$
for all 0 < ε < ε 0 ( q ), all n > n 0 ( q ), and every symmetric matrix J ∈ {-1 0 1 , , } q × q (as there are only finitely many such J for each q ). Set ε n = α max ( G n ) /α G n , so that ε n → 0 because there are no dominant nodes. It follows that we can find a slowly growing sequence q n → ∞ so that (for sufficiently large n ) we have ε n < ε 0 ( q n ), n > n 0 ( q n ), and q n ε n ≤ 1 / 2, from which it follows that
$$- E _ { \mathbf q _ { n }, \varepsilon _ { n } } ( G _ { n }, J ) \leq - \mathcal { E } _ { \mathbf q _ { n } } ( W, J ) + 1 / q _ { n }$$
for all symmetric matrices J ∈ {-1 0 1 , , } q n × q n . Hence for sufficiently large n
$$\text{metrics rules } J \in \{ - 1, U, 1 \} ^ { q _ { n } \times q _ { n } }. \text{ Hence for sumciently } \text{a} \\ ( 7. 6 ) & \leq \max _ { J \in \{ - 1, 0, 1 \} ^ { q _ { n } \times q _ { n } } } \left ( - \mathcal { E } _ { q _ { n } } ( W, J ) + 1 / q _ { n } \right ) \\ & \sum _ { i, j } | J _ { i j } | \leq 4 q _ { n } ^ { 2 } / K \\ & \leq \sup _ { \substack { S \subseteq [ 0, 1 ] ^ { 2 } \\ \lambda ( S ) \leq 4 / K } } \int _ { S \\ \lambda ( S ) \leq 4 / K } } | W ( x, y ) | \, d x \, d y + 1 / q _ { n }. \\ \text{ose } K \text{ large enough that the first term in the final bound } \colon$$
We can choose K large enough that the first term in the final bound above is at most ε/ 2. Since q n →∞ , the second term is also at most ε/ 2 for sufficiently large n . This proves (7.5), showing that ( G n n ) ≥ 0 is uniformly upper regular. glyph[square]
Appendix A. Proof of the rearrangement inequality
In this appendix, we prove that
$$\mathbf E [ W U ] \leq \mathbf E [ W ^ { * } U ^ { * } ].$$
when W is an L p graphon and U is an L p ′ graphon with 1 p + 1 p ′ = 1, with equality holding whenever U and W are aligned.
If W,U ≥ 0, the proof of the rearrangement inequality is standard, and can, e.g., be deduced from the following level-set representations
$$U ( x, y ) = \int _ { 0 } ^ { \infty } d t \, { \mathbf 1 } [ U ( x, y ) > t ] \ \text{ and } \ W ( x, y ) = \int _ { 0 } ^ { \infty } d s \, { \mathbf 1 } [ U ( x, y ) > s ].$$
Indeed, with the help of this representation, we get
$$\sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim \sim$$
where in the last step we used that U and U ∗ as well as W and W ∗ have the same distribution. Since the U ∗ and W ∗ have nested level sets, the expression in the last line is equal to
$$\text{l to} \int _ { 0 } ^ { \infty } d s \int _ { 0 } ^ { \infty } d t \Pr \left [ W ^ { * } > s \text{ and } U ^ { * } > t \right ] \\ = \text{E} \left [ \int _ { 0 } ^ { \infty } d s \text{ 1} [ W ^ { * } > s ] \int _ { 0 } ^ { \infty } d t \text{ 1} [ U ^ { * } > t ] \right ] \\ = \text{E} [ W ^ { * } U ^ { * } ]. \\ \text{ are aligned the max-loss the only inequality in the above row}$$
If W and U are aligned themselves, the only inequality in the above proof becomes an equality, showing that E [ WU ] = E [ W U ∗ ∗ ] if W and U are aligned.
If W and U are bounded below, say by W ≥ -M and U ≥ -M for some M < ∞ , then we just use that E [ W U ∗ ∗ ] -E [ WU ] = E [( W + M ) ∗ ( U + M ) ∗ ] -E [( W + M ) ( U + M )], which follows from linearity of expectations and the fact that E [ W ] = E [ W ∗ ] and E [ U ] = E [ U ∗ ]. Finally, to control the tails as M →∞ , we bound
$$\imath \underline { E } [ \overline { U } ] = \underline { E } [ \overline { U } ]. \ r \text{ may, to control one tax as $M\to\infty$, we bounds} \\ \left | \underline { E } [ W U ] - \underline { E } \left [ W 1 [ W \geq - M ] \underline { U } 1 [ U \geq - M ] \right ] \right | \\ \leq \underline { E } \left [ | W U | 1 [ | U | \geq M ] \right ] + \underline { E } \left [ | W U | 1 [ | W | \geq M ] \right ] \\ \leq \| W \| _ { p } \left \| \underline { U } 1 [ | U | \geq M ] \right \| _ { p ^ { \prime } } + \| U | \| _ { p ^ { \prime } } \left \| W 1 [ | W | \geq M ] \right \| _ { p }. \\ \text{w the right side goes to zero as $M\to\infty$} \text{$hv$ our assumption that $W\in\mathbb{P}$} \text{an}$$
Now the right side goes to zero as M →∞ by our assumption that W ∈ L p and U ∈ L p ′ .
## References
- [1] I. Benjamini and O. Schramm, Recurrence of distributional limits of finite planar graphs , Electron. J. Probab. 6 (2001), no. 23, 13 pp. arXiv:math/0011019 doi:10.1214/EJP.v6-96 MR1873300
- [2] B. Bollob´s and O. Riordan, a Metrics for sparse graphs , in S. Huczynska, J. D. Mitchell, and C. M. Roney-Dougal, eds., Surveys in combinatorics 2009 , pages 211-287, London Math. Soc. Lecture Note Ser. 365 , Cambridge University Press, Cambridge, 2009. arXiv:0708.1919 MR2588543
- [3] C. Borgs, J. T. Chayes, H. Cohn, and Y. Zhao, An L p theory of sparse graph convergence I: limits, sparse random graph models, and power law distributions , preprint, 2014. arXiv:1401.2906
- [4] C. Borgs, J. Chayes, and D. Gamarnik, Convergent sequences of sparse graphs: a large deviations approach , preprint, 2013. arXiv:1302.4615
- [5] C. Borgs, J. Chayes, J. Kahn, and L. Lovasz, Left and right convergence of graphs with bounded degree , Random Struct. Alg. 42 (2013), 1-28. arXiv:1002.0115 doi:10.1002/rsa.20414 MR2999210
- [6] C. Borgs, J. Chayes, L. Lov´sz, V. T. S´s, and K. Vesztergombi, a o Counting graph homomorphisms , in M. Klazar, J. Kratochv´ ıl, M. Loebl, J. Matouˇek, R. Thomas, and P. Valtr, eds., s Topics in discrete mathematics , pages 315-371, Algorithms Combin. 26 , Springer, Berlin, 2006. doi:10.1007/3-540-33700-8 18 MR2249277
- [7] C. Borgs, J. T. Chayes, L. Lov´sz, V. T. S´s, and K. Vesztergombi, a o Convergent sequences of dense graphs I: Subgraph frequencies, metric properties and testing , Adv. Math. 219 (2008), 1801-1851. arXiv:math/0702004 doi:10.1016/j.aim.2008.07.008 MR2455626
- [8] C. Borgs, J. T. Chayes, L. Lov´sz, V. T. S´s, and K. Vesztergombi. a o Convergent sequences of dense graphs II: Multiway cuts and statistical physics , Ann. of Math. (2) 176 (2012), 151-219. doi:10.4007/annals.2012.176.1.2 MR2925382
- [9] A. Dembo and O. Zeitouni, Large deviations techniques and applications , corrected reprint of the second (1998) edition, Stochastic Modelling and Applied Probability 38 , Springer-Verlag, Berlin, 2010. doi:10.1007/978-3-642-03311-7 MR2571413
- [10] A. Frieze and R. Kannan, Quick approximation to matrices and applications , Combinatorica 19 (1999), 175-220. doi:10.1007/s004930050052 MR1723039
- [11] A. W. Knapp, Basic real analysis , Cornerstones, Birkh¨user Boston, Inc., Boston, MA, 2005. a doi:10.1007/0-8176-4441-5 MR2155259
- [12] L. Lov´sz and B. Szegedy, a Szemer´ edi's lemma for the analyst , Geom. Funct. Anal. 17 (2007), 252-270. doi:10.1007/s00039-007-0599-6 MR2306658
- [13] G. B. Price, On the completeness of a certain metric space with an application to Blaschke's selection theorem , Bull. Amer. Math. Soc. 46 (1940), 278-280. doi:10.1090/S0002-9904-194007195-2 MR0002010
Microsoft Research, One Memorial Drive, Cambridge, MA 02142 E-mail address : [email protected]
Microsoft Research, One Memorial Drive, Cambridge, MA 02142 E-mail address : [email protected]
Microsoft Research, One Memorial Drive, Cambridge, MA 02142 E-mail address : [email protected]
Department of Mathematics, Massachusetts Institute of Technology, Cambridge,
MA 02139
E-mail address :
[email protected]
|
10.1214/17-AOP1187
|
[
"Christian Borgs",
"Jennifer T. Chayes",
"Henry Cohn",
"Yufei Zhao"
] |
2014-08-04T16:59:39+00:00
|
2014-08-04T16:59:39+00:00
|
[
"math.CO",
"math.PR"
] |
An $L^p$ theory of sparse graph convergence II: LD convergence, quotients, and right convergence
|
We extend the $L^p$ theory of sparse graph limits, which was introduced in a
companion paper, by analyzing different notions of convergence. Under suitable
restrictions on node weights, we prove the equivalence of metric convergence,
quotient convergence, microcanonical ground state energy convergence,
microcanonical free energy convergence, and large deviation convergence. Our
theorems extend the broad applicability of dense graph convergence to all
sparse graphs with unbounded average degree, while the proofs require new
techniques based on uniform upper regularity. Examples to which our theory
applies include stochastic block models, power law graphs, and sparse versions
of $W$-random graphs.
|
1408.0745v3
|
## TOPOS LOGIC IN MEASUREMENT-BASED QUANTUM COMPUTATION
Leon Loveridge 1 2 , , Raouf Dridi , Robert Raussendorf 2 2
- 1 Department of Computer Science, University of Oxford, Oxford, UK.
- 2 Department of Physics and Astronomy, University of British Columbia, Vancouver, Canada.
Abstract. We report first steps towards elucidating the relationship between contextuality, measurement-based quantum computation (MBQC) and the non-classical logic of a topos associated with the computation. We show that, in a class of MBQC, classical universality requires non-classical logic, which is 'consumed' during the course of the computation, thereby pinpointing another potential quantum computational resource.
## 1. Introduction
The quantum mechanical description of the world has startling features which are quite alien from a classical point of view. Epitomising the structural differences from classical physics includes the existence of entangled states, non-commutative observable algebras and the necessary disturbances caused by measurement. Less well known is the impossibility of assigning values to quantum observables whilst preserving some basic functional relations a trait known as contextuality and a central concept in this paper.
An upshot of the mathematical structure of quantum theory is the possibility of attaining certain advantages over classical physics in regard to computation; there now exist numerous examples of quantum protocols which far outperform their classical analogues, which are well documented in the literature. However, the origin of the quantum advantage remains unclear; 'largeness' of Hilbert space, superposition, interference, and entanglement have been suggested as candidates in the past. Whilst being supported in a variety of situations [1, 2, 3], each of these has also met with objection [4, 5, 6, 7]. Recently, contextuality has been proposed as a contender for the source of the speedup in quantum computation; it has been observed that for quantum computation with a restricted gate set and injection of magic states [8], contextuality is necessary for both quantum computational universality and the hardness of classical simulation of quantum computation [9, 10].
Contextuality also plays a role in measurement-based quantum computation (MBQC). If an MBQC on qubits evaluates a non-linear Boolean function (in which case the computation is classically universal) with high probability of success, this computation is also contextual [11]. Therefore, for MBQC, contextuality is important not only for what can be computed efficiently, but for what can be computed at all. The simplest example of this is the execution of an OR-gate using Pauli measurements on a 3-qubit Greenberger-Horne-Zeilinger (GHZ) state [12]. This gate is Mermin's proof [13] of the Kochen-Specker theorem in Hilbert spacedimension 8, recast as a simple quantum computation.
In this paper we offer a perspective related to contextuality but distinct from it: that a quantum-over-classical advantage arises from the non-classical internal logic of the given computation (we shall treat 'classical' and 'Boolean' as synonyms). Specifically, we associate a particular topos to the computation which comes equipped with a natural logical structure, and show that the topos logic is non-Boolean whenever contextuality is present. With [11] we find that if the topos logic is Boolean, the MBQC may only compute linear functions. Hence non-Booleanness is a necessary condition for quantum computational advantage in the class of computations we discuss. We also demonstrate explicitly how the non-Booleanness is
depleted by the act of computation, endorsing the view that non-Booleanness is a quantum computational resource.
Boolean logic was devised in the first instance as a formalisation of the 'laws of Thought' [14], i.e., as a way of consistently reasoning about the (classical) world. Many years have passed since Boole's seminal contribution; a modern connection to Boolean logic is the theory of (classical, digital) computation. The quantum analogue-the relationship between quantum computation and 'quantum logic'-has thus far remained largely unexplored (though there are exceptions; see e.g. [15] for a version of 'quantum computational logic'). In this paper we make some inroads into this territory.
Soon after the mathematical structure of quantum mechanics was essentially crystallised it was realised that the logical structure of the theory was fundamentally different from that found in classical physics [16]. Though there are now many candidates for the 'logic of quantum theory' (see, e.g.,[15]), we choose to follow the topos approach initiated by Isham and Butterfield and developed by others (see, e.g., [17], [18]; there are numerous other topos approaches - Adelman and Corbett [19] seem to have been the first on the scene of 'topos quantum mechanics', and the work of Heunen, Landsman, and Spitters [20] takes a covariant, rather than presheaf, approach). Our reasons for this choice are manifold. The logic of a topos is distributive, in contrast to the (propositional) logic ingeniously fashioned from projections by Birkhoff and von Neumann [16], which seems to have permanently acquired the name 'quantum logic'.
Moreover, topos quantum logic is a first order predicate logic, thus reaching well beyond the propositional quantum logic and allowing for the expression of a much larger class of statements (involving quantifiers, for instance). It is in the topos approach that the connection between logic and contextuality is most clearly seen; we have a simple argument demonstrating that wherever there is contextuality there is non-Boolean topos logic, and that this applies to both state-independent and state-dependent versions of contextuality. Furthermore, for the topoi of interest in this investigation, the quantification of non-Booleanness takes a simple form, allowing for a direct demonstration in the Anders-Browne OR-gate that the 'amount' of non-Booleanness present is related to computational power at each stage of the computation.
## 2. Survey of Results
Before proceeding to the provision of the mathematical and physical frameworks within which our work is carried out, we believe it helpful to survey the main results of this paper, in part as useful reference material and summary, and in part as a guide for what follows.
Following the pioneering work of Isham, Butterfield and D¨ring, we begin by discussing o the Kochen-Specker theorem in terms of the 'spectral presheaf' Σ, which associates to each Abelian observable algebra (or context ) a classical phase space, viewed as the space of valuations of the given observables. The Kochen-Specker theorem provides a negative answer to the question 'is it possible that observables in quantum mechanics have predetermined measurement outcomes which are preserved under functional relations between these observables?'. The statement of the Kochen-Specker theorem can be recast in terms of the (non-)existence of global sections of Σ. We are also able, through the pseudostate w | ϕ 〉 , to give state-dependent Kochen-Specker proofs from within the presheaf perspective, providing a simple version of a proof of Mermin in the new language. We note that what is required for such a proof is a specific subposet W H ( ) of the poset V H ( ) of all Abelian subalgebras of B ( H ) -the space of all linear operators on a finite dimensional Hilbert space H . W H ( ) is also related to computation, constructed by taking intersections of lists of local observables whose product has the computational resource state as an eigenstate, and then forming the Abelian algebras at each node via the double commutant.
Understanding how Σ and w | ϕ 〉 relate to their 'environment' amounts to constructing the topos ̂ W H ( ) of all set-valued presheaves on W H ( ). A topos is a set-like universe; from a
logical perspective any topos behaves similarly to the category of sets but fails in two crucial respects: there are many possible truth values, arranged in a so-called Heyting algebra , and the law of excluded middle fails. The latter is equivalent to the statement that the internal logic of the given topos is non-Boolean . Moreover, the axiom of choice (AC) fails in most topoi and, by a theorem of Diaconescu, choice and excluded middle are related; in any topos choice implies excluded middle and hence Booleanness. In the topoi of interest to us, namely the topoi of presheaves on a poset, the above implication is promoted to an equivalence: Boolean logic is equivalent to AC. By considering the appropriate categorical formulation of choice, we arrive at our first main observation: the presence of state-dependent or state-independent contextuality implies non-Boolean logic in ̂ W H ( ).
Through the intermediary of contextuality, the relationship between MBQC and logic is considered. Anders and Browne [12] noticed that the state-dependent Kochen-Specker proof of Mermin can be recast as an MBQC OR-gate, for which they observe that contextuality is a necessary requirement. For a general /lscript -2 MBQC (see section 4), it has been shown [11] that non-linear function evaluation implies contextuality in the given computation, i.e., the existence of a Kochen-Specker proof. By constructing the poset W H ( ) canonically associated to the MBQC and identifying the corresponding topos, we may call upon the results stated in the previous paragraph to demonstrate that:
For the computation of non-linear functions in /lscript -2 MBQC, the associated topos is necessarily non-Boolean.
In other words, just as entanglement, superposition and contextuality may be understood to be computational resources, we see that non-Boolean topos logic may be viewed similarly. By considering each stage of the computation in the Anders-Browne example, and calling upon a construction which quantifies the non-Booleanness of the associated topos at each stage, we show that the non-Booleanness is depleted by the computation and is directly related to the class of functions which can be computed after each measurement. Many of our findings are generic in the class of /lscript -2 MBQC we consider.
The rest of this paper is organised as follows. In section 3 we give definitions of stateindependent and state-dependent contextuality in terms of Σ and w | ϕ 〉 , respectively, including Mermin's state-dependent proof from this perspective. Section 4 provides the definition of /lscript -2 MBQC as well as Anders and Browne's contextual OR-gate. We then turn to topoi and logic in section 5. Motivating the notion of a Heyting algebra from the propositional structure governed by the open sets of a topological space, we then provide details of the topos Sets required for understanding the logical structures of a general topos which will be important in later sections. We provide some important definitions in the general theory of topoi and give the topos version of AC.
A consideration of the topos ̂ P of presheaves on a generic poset P and its quantum realisation W H ( ) constitutes section 6. We present an important isomorphism between ̂ P and the topos of sheaves Sh ( P ) on a topological space P constructed from the poset P , which allows for a simple quantification of the non-Booleanness of ̂ P . Section 6 also establishes the necessity of non-Boolean logic for non-linear function evaluation in /lscript -2 MBQC, and section 7 demonstrates the consumption of the non-Booleanness during the execution of the computation.
Sections 3, 4 and 5 contain review material on contextuality, MBQC and topoi respectively as one or more topic may be unfamiliar. We include a summary paragraph at the beginning to highlight the essential content.
## 3. Contextuality in terms of Σ and w | ϕ 〉
In this section we formulate the state-independent and state-dependent Kochen-Specker theorems from the perspective of the Isham-Butterfield presheaves Σ and w | ϕ 〉 , and demonstrate that a simple proof of Mermin [13] may be cast in this language. The main observation is that only a small subposet of the poset of all Abelian subalgebras of B ( H ) is required to demonstrate a contradiction with the assumption of pre-assigned measurement outcomes.
We work in a complex Hilbert space H for which dim H > 2, equipped with the standard inner product 〈·|·〉 : H×H → C , linear in the second argument, pure states being given as unit vectors in H . In order to obviate various technical topological issues we also assume that dim H < ∞ . This is also justified on physical grounds; the quantum computational protocols we consider all take place in the finite dimensional setting.
Let B ( H ) denote the algebra of all linear mappings H → H , and V H ( ) the collection of all unital Abelian subalgebras of B ( H ), (partially) ordered by inclusion. To each V ∈ V H ( ) ( V viewed as a C ∗ algebra) we form the set of characters Σ( V );
$$\Sigma ( V ) \coloneqq \{ \lambda _ { V } \colon V \to \mathbb { C } \ | \ \lambda _ { V } \neq 0, \ \lambda _ { V } \text{ is linear and } \lambda _ { V } ( a b ) = \lambda _ { V } ( a ) \lambda _ { V } ( b ) \}. \quad \quad ( 1 ) \\ - \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \end{$$
/negationslash
Following Isham et al. (e.g., [17]) we also define how Σ acts on inclusion relations between different Abelian subalgebras: for λ V ∈ Σ( V ), if U ⊆ V , define
$$\Sigma ( U \subseteq V ) ( \lambda _ { V } ) \coloneqq \lambda _ { V } | _ { U } \text{ (as function restriction)} & & ( 2 ) \\ \Lambda \dots \llcorner \Lambda \dots & & \omega \, \vdots \, \shortmid \, \longrightarrow \, \Lambda \, \end{cases}$$
$$\begin{matrix} & \lambda _ { V } \mapsto \lambda _ { V } | _ { U } ; & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & \\ & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & && & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &
& & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & \ & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &. & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & - & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & } & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & ; & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &.. & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &. \ & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &.
& & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &. } & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &. \\ & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &. ; & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &. = & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & = & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &$$
↦
the mapping Σ( U ⊆ V ) is called a restriction map . It is clear that, if U ⊆ V ⊆ W (all Abelian),
$$\Sigma ( V \subseteq W ) \circ \Sigma ( U \subseteq V ) = \Sigma ( U \subseteq W ) ; \quad \quad \quad ( 4 ) \\ \vdots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots.$$
in other words, it makes no difference if one restricts in stages, or 'all at once'. In mathematical terms, Σ is then referred to as a contravariant, set-valued functor (or presheaf ), called the spectral presheaf and represents the 'state space', as introduced in [21].
↦
For each V ∈ V H ( ), Σ( V ) can be viewed both topologically and logically as a classical phase space. It is topologically compact and Hausdorff, and the algebra of observables V can be recovered as the space of continuous functions Σ( V ) → C (in other words, V ∼ = C (Σ( V )) and the isomorphism is isometric; this is the Gelfand duality arising from the Gelfand transform A → ¯ ; A λ A ( ) := ¯ ( A λ )). The algebra of physical propositions is given by the clopen (closed and open) subsets of Σ( V ) which is necessarily Boolean, reflecting the standard propositional structure corresponding to classical physical systems.
3.1. Valuations and Sections. The set Σ( V )-the Gelfand spectrum of V -is in bijection with the set of simultaneous eigenvectors of self-adjoint operators in V , and therefore, for any self-adjoint a ∈ V and any λ V ∈ Σ( V ), λ V ( a ) is an eigenvalue of a (we call this condition SPEC ). Furthermore, any λ V obeys the functional constraint f ( λ V ( a )) = λ V ( f ( a )) for any (continuous) function f : R → R and self-adjoint a in V . Hence λ V acting on V satisfies the 'functional composition principle' FUNC (see [22], ch. 5, for details) and can therefore be viewed as a valuation of the self-adjoint operators in V (and the collection of all λ V are all the valuations). Such valuations are called local sections (above V ). FUNC states that if an observable a (energy, for example) has value a 0 , then f ( a ) has value f ( a 0 ) (i.e., the value of the square of the energy should be the square of its value, for instance), and is (along with the spectrum requirement SPEC) an ostensibly mild condition even when considered on the whole of B ( H ). However, V H ( ) is woven together in a complicated way (recall the partial order) and, as we shall see, this prohibits the existence of 'global' valuations, i.e., those defined on all observables in B ( H ).
The existence of a valuation on B ( H ) is of clear importance in attempts to restore any kind of realism to quantum theory: if observables can be said to have values prior to and
independently of measurement, the status of measurement can be downgraded to mere revelation of values/properties 'held' by the quantum object in question. If such a valuation were to exist, quantum theory would then retain some essence of classical physics where all observables have values in all states, and where propositions about a physical system are either true or false. To reiterate, a valuation λ : B ( H → ) C must satisfy:
- · For a = a ∗ and f : R → R , f ( λ a ( )) = λ f a ( ( )) (FUNC);
- · For a = a ∗ , λ a ( ) ∈ σ a ( ) (i.e., the value must be an eigenvalue) (SPEC)
any such valuation is necessarily linear and multiplicative on commuting operators and corresponds to sections above Abelian algebras. It is automatic, of course, that for a projection P , λ P ( ) ∈ { 0 1 , } .
If such a global value assignment λ existed, we would be able to choose for each V ∈ V H ( ) a point λ V in Σ( V ) in such a way that, for any U ⊆ V , Σ( U ⊆ V )( λ V ) = λ U . From equation (4), we see that local valuations, on U , say, are not sensitive to which other Abelian algebras U is contained in if they come from a global valuation, intuitively corresponding to a non-contextuality requirement.
- 3.2. The Kochen-Specker theorem. As we have seen, a global assignment λ would specify an eigenvalue of each self-adjoint operator in B ( H ) (SPEC) whilst respecting functional relations (FUNC). Keeping the same notation we now view λ as giving rise to a local section above each V , in such a way as to respect the inclusions in V H ( ).
Definition 1. A global section λ of Σ is an assignment of a point λ V ( ) ≡ λ V ∈ Σ( V ) for each V ∈ V H ( ) , such that for U ⊆ V , Σ( U ⊆ V )( λ V ) = λ V | U = λ U .
There is a categorical description in which a global section looks like a 'point' of a set. Consider the singleton set {∗} -the terminal object in the category Sets of abstract sets and functions-characterised by the property that for any set A (including the empty set ∅ ), there is precisely one map A →{∗} . Clearly any point of a set A can be given as a map {∗} → A , and the collection of all such maps determines A completely.
In the category of all presheaves V H ( ) → Sets =: E (which is also a topos - more on this later), in order to define 'points' of objects we will need to also consider maps between presheaves, which are the arrows in E :
Definition 2. Let F G , : V H ( ) → Sets be two presheaves. A natural transformation τ : F → G is a map of presheaves, given context-wise as τ V : F V ( ) → G V ( ) , which for each U ⊆ V , τ U ◦ F ( U ⊆ V ) = G U ( ⊆ V ) ◦ τ V .
The constant presheaf 1 : V H ( ) → Sets for which 1( V ) = {∗} for each V (and on inclusions 1( U ⊆ V )( {∗} ) = {∗} ) is a terminal object in E ; for any presheaf A in E there is precisely one map (natural transformation) A → 1. Hence one can view a (generalised) point , or element , of Σ as a map (natural transformation) 1 → Σ and this map is precisely λ . Hence a global section of an object A in E is a generalised element of A . In this language the Kochen-Specker theorem [23], or the existence of contextuality , reads [21]:
Theorem 1. (Kochen-Specker, Isham-Butterfield) For Σ : V H ( ) → Sets , there exists no global section 1 → Σ , and hence Σ has no points at all in a generalised sense! An alternative view presents itself. For sets, surjective maps are right cancellable; if g 1 ◦ f = g 2 ◦ f then g 1 = g 2 . This provides the categorical version of a surjective map, called an epic map, or an epimap or an epimorphism . Just as any function A →{∗} is surjective, in any topos of presheaves the map A → 1 is an epimap. Hence one may view the section 1 → Σ as the 'cross-section' (i.e., right inverse) of an epimap Σ → 1; following standard terminology we also call this map a section (indeed, this is the reason for calling 1 → Σ a section in the first place). As shall become evident, the nonexistence of such a section is not only critical for a realist description of quantum theory,
it is also of crucial importance to logic. Considering any subposet W H ( ) of V H ( ) we may make the following definition:
Definition 3. W H ( ) is contextual if Σ : W H → ( ) Sets has no global section.
This allows us to look for contextual 'scenarios', i.e., instances of contextuality, by considering a small number of observables, constructing the corresponding subposet of V H ( ), and looking for global sections of Σ there (we always refer to Σ despite the different domains; the context should make clear the poset to which we refer).
Theorem 1 does not depend on any choice of (pure) state; there are numerous examples of so-called 'state-dependent' proofs of the Kochen-Specker theorem which do invoke a particular state, physically interpreted as the impossibility of observables possessing values in that given state . Such proofs can also be displayed in terms of (the lack of) global sections of the 'pseudostate' w | ϕ 〉 : V H ( ) → Sets (or subposets of V H ( ) thereof) - this provides a unified perspective of state-dependent and state-independent versions of the Kochen-Specker theorem. With P V ( ) denoting the projection lattice of V (seen as a von Neumann algebra) and with ϕ ∈ H fixed, we first define the presheaf S | ϕ 〉 (neglecting the restrictions) by
$$\pi \text{ use, we must require one presneau } \vartheta _ { | \varphi \rangle } ( \text{megting the restrictions by} \\ S _ { | \varphi \rangle } ( V ) \coloneqq & \bigwedge \left \{ P \in P ( V ) \ | \ P \geq | \varphi \rangle \langle \varphi | \right \}, \\ \text{list projector } P \, \in \, P ( V ) \text{ which dominates } | \varphi \rangle \langle \varphi |. \, \text{ The Gelfand transform}$$
↦
viz. the smallest projector P ∈ P V ( ) which dominates | ϕ 〉 〈 ϕ | . The Gelfand transform P → ¯ , P defined as λ V ( P ) = ¯ ( P λ V ), dictates that ¯ P can be viewed as the characteristic function on Σ( V ) of those λ V for which ¯ ( P λ V ) = 1. This motivates the definition of w | ϕ 〉 , given by
$$\mathfrak { w } ^ { | \varphi \rangle } ( V ) & = \{ \lambda _ { V } \in \Sigma ( V ) \ | \ \lambda _ { V } ( P ) = 1 \}, \quad \text{(6)} \\ \text{atlv a sheet of $\Sigma/V$} \ \text{for each $V$ and hence $\mathfrak{m}|\varphi$ is a shallow effect of $\Sigma$ } \text{the}$$
which is manifestly a subset of Σ( V ) for each V , and hence w | ϕ 〉 is a subobject of Σ (the restriction maps for w | ϕ 〉 are of course the same as for Σ; w | ϕ 〉 ( U ⊆ V )( λ V ) = λ U ). Continuing to write W H ( ) for some fixed subposet of V H ( ), we have the following definition:
Definition 4. State-dependent contextuality is present for ( | ϕ 〉 ∈ H W H , ( )) if there is no global section 1 → w | ϕ 〉 .
The following simple observation demonstrates the relationship between state-dependent and state-independent contextuality:
Proposition 1. For any W H ( ) , if Σ has no global section then there does not exist a | ϕ 〉 ∈ H for which w | ϕ 〉 has a global section.
Since w | ϕ 〉 picks out a subset of all possible sections above each V ∈ W H ( ), the proof is clear. We end this subsection by noting that connections between contextuality andglobal sections have also been given elsewhere; for instance sheaf theoretic methods in contextuality and non-locality have been employed to great effect by Abramsky and coworkers (e.g., [24, 25]), who demonstrate that contextuality can be displayed by an operational consideration of measurement outcomes, without recourse to the quantum formalism.
3.3. Mermin's state-dependent proof. We let H ≡ H 1 ⊗H ⊗H 2 3 := C 2 ⊗ C 2 ⊗ C 2 with the Pauli operators σ , σ x y , σ z having the standard definitions, and write X 1 ≡ σ x ⊗ /BD ⊗ /BD , Y 2 ≡ ⊗ σ y ⊗ , etc . Consider the following putative valuation:
$$\lambda ( X _ { 1 } X _ { 2 } X _ { 3 } ) = + 1$$
/BD
/BD
$$\lambda ( X _ { 1 } Y _ { 2 } Y _ { 3 } ) & = - \, 1 \\ \vee / V. \ V. \vee V. \ \, = \quad 1$$
$$\lambda ( Y _ { 1 } Y _ { 2 } X _ { 3 } ) = - 1. \quad \quad ( 1 0 ) \\ \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \end{array}$$
$$\lambda ( Y _ { 1 } X _ { 2 } Y _ { 3 } ) & = - \, 1 \\ \vee / V. V. V. \, V. \, \vee & = \, 1 & / 1 \infty$$
By definition λ is multiplicative on commuting operators (and strictly is defined on the algebras generated by the given qubit observables). The product of the four terms on the left
hand sides of (7)-(10) is 1, whereas the product of the four values on the right hand side is -1, thus ruling out the mapping λ as a viable valuation.
$$\text{$\text{$me$} \, $save_{[av]}$.} \\ | \Psi \rangle \equiv 1 / \sqrt { 2 } \left ( | 0 0 0 \rangle + | 1 1 1 \rangle \right ) & & ( 1 1 ) \\ \text{and to uncomment one of $\sigma$} \cdot \cos [97] \text{ for $\text{Mom}in\,\text{etc}$ to $\text{don}$} \text{$\text{$\text{$coh}$} \, $\text{$coh}$}$$
The values taken by λ on the right hand side of (7) (10) are precisely those given by (expectation values in) the GHZ state [26]:
(the 0's and 1's correspond to eigenvalues of σ z ; see [27] for Mermin's state-dependent KochenSpecker proof). The GHZ state is not multiplicative on the given observables and therefore does not satisfy the definition of a section. In order to prove that we are in the presence of state-dependent contextuality in the language of this paper, i.e., in accordance with definition 4, it must be shown that w | Ψ 〉 has no global section, i.e., that the local sections, compatible with the constraints imposed by the GHZ state, cannot be appropriately 'glued' to give a global section with the required restrictions.
$$\text{ above } V _ { i } \text{ for } i \in \{ 1 \dots 4 \}. \text{ Hence we have the constraints:} \\ \tilde { \lambda } _ { V _ { 1 } } ( * ) ( X _ { 1 } ) \tilde { \lambda } _ { V _ { 1 } } ( * ) ( X _ { 2 } ) \tilde { \lambda } _ { V _ { 1 } } ( * ) ( X _ { 3 } ) = + 1 \\ \tilde { \lambda } _ { V _ { 2 } } ( * ) ( X _ { 1 } ) \tilde { \lambda } _ { V _ { 2 } } ( * ) ( Y _ { 2 } ) \tilde { \lambda } _ { V _ { 2 } } ( * ) ( Y _ { 3 } ) = - 1 & & ( 1 2 ) \\ \tilde { \lambda } _ { V _ { 3 } } ( * ) ( Y _ { 1 } ) \tilde { \lambda } _ { V _ { 3 } } ( * ) ( X _ { 2 } ) \tilde { \lambda } _ { V _ { 3 } } ( * ) ( Y _ { 3 } ) = - 1 \\ \tilde { \lambda } _ { V _ { 4 } } ( * ) ( Y _ { 1 } ) \tilde { \lambda } _ { V _ { 4 } } ( * ) ( Y _ { 2 } ) \tilde { \lambda } _ { V _ { 4 } } ( * ) ( X _ { 3 } ) = - 1. \\ \Re \, \text{ we write } \lambda _ { V _ { i } } \coloneqq \tilde { \lambda } _ { V _ { i } } ( * ). \text{ The local sections } \lambda _ { V _ { i } } \text{ have restrictions, for example}$$
Let ˜ λ : 1 → w | Ψ 〉 denote a hypothetical global section of w | Ψ 〉 , and define V 1 to be the observable algebra generated by { X ,X ,X 1 2 3 } and, following equations (7) - (10), mutatis mutandis , do the same for V , V 2 3 and V 4 . We denote by V X 1 the algebra generated by X 1 (which is contained in V 1 ), etc . We have that ˜ λ V i : {∗} → w | Ψ 〉 ( V i ) so that ˜ λ V i ( {∗} ) is a local section of w | Ψ 〉 above V i for i ∈ { 1 ... 4 . } Hence we have the constraints:
For convenience we write λ V i := ˜ λ V i ( ∗ ). The local sections λ V i have restrictions, for example λ V 1 | V X 1 = λ V X 1 etc. It is thus required that the compatibility conditions hold: λ V 1 | V X 1 = λ V 2 | V X 1 , λ V 1 | V X 2 = λ V 3 | V X 2 and so on. These are manifestly requirements of non-contextuality. Appropriate substitution back into (12) gives rise to the same contradiction found on λ in (7) - (10). The following Hasse diagram illustrates the poset structure of the GHZ scenario:
4. Measurement Based Quantum Computation
In this section /lscript -2 MBQC is introduced and its connection with contextuality is emphasised. The main result is theorem 2, which demonstrates the necessity of contextuality for non-linear function evaluation, the simplest manifestation of which is the contextual OR-gate of Anders and Browne [12] constructed from Mermin's state-dependent Kochen Specker proof.
4.1. /lscript -2 MBQC. Measurement based quantum computation (MBQC) employs single-particle (qubit or qudit, projective or positive operator valued) measurements on a fixed initial state to achieve a computational output. The state is altered by the local measurements (in particular any entanglement is destroyed); the initial state is thus viewed as a resource, and for suitable choices such as cluster states [28, 29], MBQC is universal for quantum computation.
The measurements generating the computation have individually random but correlated outcomes, and the computational result depends crucially on these correlations. In addition,
the observables to be measured must typically be adjusted according to previous measurement outcomes. Therefore, classical side-processing is required both to direct the computation and to produce the desired output. This classical processing is typically limited, and is not universal for classical computation. For example, in the universal scheme of [30], only addition modulo 2 is permitted.
We note that universal MBQC schemes with different classical side-processing, different resource states and different sets of allowed measurements have been devised; see, e.g., [31, 32, 33, 34, 35, 36]. However, for the present purpose, we consider MBQC on quantum states of multiple qubits, where the classical side-processing is restricted to addition modulo 2, see [30, 12, 11].
The relevant Hilbert space for our work is C 2 n ∼ = ⊗ n k =1 C 2 and measured observables are non-trivial on only one factor in the tensor product; these are referred to as local observables . Local measurements are defined in the obvious way. We now give a brief definition of /lscript -2 MBQC and refer to [11] for details; for clarity of exposition we restrict attention to the case where there is only one output bit, though our results pertain to the general case. We also restrict ourselves to computations which are temporally flat , i.e., joint measurements of local observables or, in other words, where the choice of measured local observable does not depend on the outcome of a previous measurement.
↦
Definition 5. A temporally flat, deterministic /lscript -2 MBQC M → consists of a resource state | Φ 〉 ∈ C 2 n , classical input i ∈ Z m 2 and classical output o : Z m 2 Z 2 , i → o i ( ) , a collection of (maximal) local observables { O k ( q k ) | k ∈ { 1 ...n } ; q k ∈ Z 2 } for which the measurement outcomes of a given O k ( q k ) are labelled s k ( q k ) ∈ { 0 1 , } for each k . The computed output o i ( ) = ∑ k s k ( q k ) mod 2, and, with q = ( q ...q 1 n ) , the measured observable O k ( q k ) is related to the outcome s k ( q k ) by q = Qi where Q : Z m 2 → Z n 2 .
Since all the computations we consider are of the above form, for brevity we occasionally omit the qualifiers 'deterministic' and 'temporally flat', and only refer to such conditions when necessary.
Given B ( C 2 n ), we must construct the poset W ( C 2 n ), yielding a diagram similar to (13). We proceed by considering all strings { ( O 1 ( q 1 ) , ..., O n ( q n )) } for which O 1 ( q 1 ) ...O n ( q n ) | Ψ = 〉 ( -1) o i ( ) | Ψ . Given such a string ( 〉 O 1 ( q 1 ) , ..., O n ( q n )) the Abelian algebra (context) is generated by the local observables, written { O 1 ( q 1 ) , ..., O n ( q n ) } ′′ (the prime denotes the commutant; see, e.g., [37] theorem 5.3.1.).
Definition 6. Let { O i } denote the collection of strings of local observables given above with O i ∩ O j the intersection of any pair, giving morphisms O i ∩ O j → O i and O i ∩ O j → O j , etc, and call this poset W H ( ) 0 . Then W H ( ) is the poset defined by taking the algebra generated by each object in W H ( ) 0 .
Hence the collection { O i } forms the 'top layer' of the poset, with the second layer given as the algebras generated by elements in the intersection of the generating sets of the maximal algebras, etc. (note that any { O 1 ( q 1 ) , ..., O n ( q n ) } ′′ is maximal since O 1 ( q 1 ) ...O n ( q n ) has 2 n mutually orthogonal rank-1 projections in its spectral resolution, and the algebra generated by these projections coincides with { O 1 ( q 1 ) , ..., O n ( q n ) } ′′ ). A maximal Abelian subalgebra is called a masa . The main result of [11] reads:
Theorem 2. [11] Every /lscript -2 MBQC which deterministically computes a non-linear Boolean function is contextual.
The version of contextuality used in [11] is the 'strong contextuality' of [24], which, though different from our definition, coincides with it in the cases we consider.
We conclude this subsection with two remarks. Firstly, there are quantum algorithms with super-polynomial speed-up in the class of deterministic MBQCs, exhibiting contextuality. Specifically, the quantum algorithm for the 'discrete log' problem [38] breaks a crypto system. It is deterministic in the circuit model [39], and can be converted into deterministic MBQC
by standard techniques [30]. Secondly, the condition of determinism in theorem 2 can be relaxed. For probabilistic function evaluation, for every given non-linear function there is a threshold in the success probability above which the corresponding MBQC is contextual. This threshold can be low; for the bent functions [40] it approaches 1/2 in the limit of large number of input bits [11].
- 4.2. Anders and Browne's OR-gate. The first and simplest example that illustrates the connection between contextuality and non-linear function evaluation in MBQC is the ORgate of Anders and Browne [12], which is directly constructed from Mermin's state-dependent proof of the Kochen-Specker theorem.
In order to consider computation, it is necessary to rewrite the proof of contextuality arising from the GHZ state in computational variables. Consider once more the 'attempted GHZ-valuation' λ given in (7) - (10). Define λ X ( i ) = ( -1) x i and λ Y ( i ) = ( -1) y i . Direct substitution into (7) - (10) yields the following constraints (with all sums modulo 2):
$$\begin{array} { r c l } x _ { 1 } + x _ { 2 } + x _ { 3 } & = & 0, \\ x _ { 1 } + y _ { 2 } + y _ { 3 } & = & 1, \\ y _ { 1 } + x _ { 2 } + y _ { 3 } & = & 1, \\ y _ { 1 } + y _ { 2 } + x _ { 3 } & = & 1, \\ \end{array}$$
resulting, once again, in an insoluble set of equations (this time seen by summing the left hand side of (14), summing the right hand side, and comparing). The above proof of the Kochen-Specker theorem can be adapted to MBQC. There is no contradiction if in (7) - (10), or equivalently their computational counterparts in (14), the x , y i i are regarded as actual measurement outcomes rather than predetermined (noncontextual) values.
The right hand side of (14) can be viewed as the output of an OR-gate in /lscript -2 MBQC (recall definition 5) as follows. The input i = ( i 1 , i 2 ) ∈ Z 2 × Z 2 and o i ( ) ∈ Z 2 . If q k = 0, X k is measured; if q k = 1, Y k is measured. The outcomes s k ( q k ) are written s k (0) = x k , s k (1) = y k . The map Q effects q 1 = i 1 , q 2 = i 2 , q 3 = i 1 + i 2 mod 2. Hence o i ( ) = s 1 ( q 1 )+ s 2 ( q 2 )+ s 3 ( q 3 ) so, for instance, if ( i 1 , i 2 ) = (0 0), , q (0 0) = , s 1 (0) + s 2 (0) + s 3 (0) = x 1 + x 2 + x 3 = 0. Computing the outputs for the remaining inputs verifies that o i ( 1 , i 2 ) = OR i ,i ( 1 2 ), and, given equation (14), we see how the above Kochen-Specker proof gives rise to an OR-gate in /lscript -2 MBQC.
The OR-gate has the property of being a non-linear Boolean function (compare OR (1 1)+ , OR (0 1) with , OR (1 0)) which, as theorem 2 shows, implies the presence of contextuality. , While not of practical use, the ability to evaluate an OR-gate within MBQC is of fundamental relevance. The classical control computer in /lscript -2 MBQC is limited to addition mod 2, and is thus by itself not classically universal. Access to a GHZ state and Pauli measurements promotes it to classical universality, vastly increasing its computational power.
## 5. Topoi and logic
In this section some basic elements of topos theory are given, with particular emphasis on logic. The subobject classifier is defined and the result that the topos version of the axiom of choice implies Booleanness is stated.
Topos theory emerged from two distinct places (a good history can be found in the introduction of [41]): Grothendieck's work on sheaves and Lawvere and Tierney's work on categorical set theory and the foundations of mathematics; it is here that the logical aspects of topos theory are most visible.
A topos resembles a 'universe of sets'; certain universal constructions available for sets such as Cartesian products, terminal objects, exponentials ( B A := { f | f : A → B } is a set) and so on, are available in any topos. However, most topoi are non-Boolean, manifested by the failure of the excluded middle, and therefore constructive arguments must be used. Moreover, AC fails in general and, as we shall see, this also has an impact upon the logic of the topos.
- 5.1. Logic and Heyting algebras. The motivating example of a Heyting algebra is the lattice O ( X ) of open sets of a topological space X . Interpreting the open sets as propositions, the logical 'and' ( ∧ ) and 'or' ( ∨ ) are, just as for a Boolean algebra, set theoretic intersection ( ∩ ) and union ( ∪ ), respectively, since these operations preserve openness. However, logical negation ¬ cannot be interpreted as set-theoretic complementation as the complement fails to be open in general, and rather ¬ U := intU c ( c to denote set-theoretic complement.) Any derived operation involving ¬ must also be modified. The lattice of open sets of a topological space, under these operations, forms a Heyting algebra . However, there are Heyting algebras not of this form, and so we provide the general definition.
Definition 7. A Heyting algebra H is a bounded distributive lattice (like a Boolean algebra) with a top element 1, a bottom element 0, and a binary operation = ⇒ which satisfies, for all a, b, c in H , c ≤ a = ⇒ b if and only if ( c ∧ a ) ≤ b .
The negation ¬ a is defined by a = ⇒ 0 and is called the pseudo-complement (of a ); it is the greatest element b of H for which a ∧ b = 0. An element a ∈ H is regular if ¬¬ a = a and is complemented if the pseudo-complement satisfies
$$a & \vee \neg a = 1 ; & & ( 1 5 ) \\ \intertext { a b a m b a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a l d a r d a l d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d a r d }$$
in general a ∨ ¬ a ≤ 1 and thus the excluded middle is not available in a general Heyting algebra. This is the principal difference between Heyting algebras and Boolean algebras. Indeed (see [42], proposition 1.2.11 and the accompanying proof),
Proposition 2. A Heyting algebra H is Boolean if, and only if, for each a ∈ H , ¬¬ a = a , i.e., if each element is regular which holds if and only if for all a ∈ H , a ∨¬ a = 1 (i.e., each element is complemented).
Heyting algebras arise in topos theory in governing both the structure of subobjects of a given object, and the structure of the global elements of the subobject classifer.
- 5.2. The topos Sets . The category Sets of sets and functions provides an archetypal topos. The logical structure of Sets encompasses how subsets (of a given set) relate to each other. Given S ⊆ A , the categorical 'philosophy' dictates that one should rephrase this only in terms of functions. One can view the subset relation S ⊆ A alternatively as an injective map S /arrowtailright A (or, more properly, an equivalence class of injective maps). These maps are 'classified' by characteristic functions A →{ 0 1 , } , i.e., given any s : S /arrowtailright A there is a unique map χ S whose inverse image χ -1 S (1) = s S ( ). Thus with subA denoting the collection of injective maps into A , P ( A ) the power set of A and 2 A the collection of all functions A → 2,
As a case in point, consider Σ : V → Sets , where V is a finite Abelian algebra. Then Σ( V ) is the finite analogue of a classical phase space; sub (Σ) is a Boolean algebra corresponding to subsets of the phase space, and a pure state is either in or out of a given subset, as witnessed by the 'truth value set' 2 = { 0 1 , } .
$$s u b A & \cong \mathcal { P } ( A ) \cong 2 ^ { A }. & ( 1 6 ) \\ \intertext { a n e } \nabla \, \mathbb { N } \,. \, \mathbb { N } \quad \mathfrak { C } \quad \mathfrak { C } \quad \text{when} \, \mathbb { N } \, \mathbb { N } \, \text{is a bit} \, \infty \, \text{for} \, \mathbb { N } \, \mathbb { N } \, \text{$\mathbb{ }$} \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb{ N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb { N } \, \mathbb$$
It is possible to give an analogous correspondence between subobjects of a given object A and maps A → Ω in any topos, where Ω represents a 'truth value object' generalising 2. The required definitions can be given without reference to points, which is crucial given that objects in a topos may have many subobjects but no points at all (e.g., Σ in the presence of contextuality).
- 5.3. General topoi. The full definition of a topos, which is technical, will not be required (see, e.g., [42] ch. 5). Instead we provide only the concepts necessary for the definition of the subobject classifier which embodies the logical structure of a topos, and discuss the relationship between the logic and the topos formulation of AC. The appropriate generalisation of subset to subobject is given as a monic (left-cancellable) map:
Definition 8. A subobject A of an object B in E is defined as a monic arrow A /arrowtailright B .
Of course, 'monic' is the categorical version of 'injective' in Sets . The definition of subobject classifier in a topos E mirrors the correspondence in Sets between subsets and functions with co-domain 2:
Definition 9. Let E be a topos. A ubobject classifier is an object s Ω together with a morphism true : 1 → Ω such that for every object A of E and every monic s : S /arrowtailright A , there is a unique morphism χ S : A → Ω such that the following square is a pullback (i.e., commutes and is universal among such squares):
/d47
/d47
/d15
/d15
$$S \xrightarrow { \longrightarrow } 1 \\ s \Big [ \begin{array} { c } 1 \\ \Big | \\ A \xrightarrow { \ x s } \end{array} \Big ] t r u e \quad \quad \quad$$
/d15
/d15
/d15
/d15
/d47
/d47
This sets up a Heyting algebra isomorphism between subobjects of some object X and characteristic morphisms X → Ω;
$$s u b X \cong H o m ( X, \Omega ) \cong \Omega ^ { X }$$
where Hom X, ( Ω) denotes the set of maps (morphisms) X → Ω. We state without proof the following fact, referring the reader to [42], proposition 6.2.1:
Proposition 3. For any object X in E , subX is a Heyting algebra.
For example, the collection of subobjects of Σ : V H ( ) → Sets forms a Heyting algebra which is not Boolean, giving the topos version of phase space a fundamentally different structure from that which is usual for classical systems.
Replacing X by 1-the terminal object in E -immediately yields
$$s u b ( 1 ) & \cong \Gamma \Omega,$$
demonstrating at once that the global elements of Ω have the structure of a Heyting algebra, given as the subobjects of the terminal object 1.
Definition 10. E is called Boolean if subΩ (the Heyting algebra of subobjects of Ω ) is a Boolean algebra.
Definition 11. If every epimap in E has a section, then we say the AC holds in E .
This is a straightforward generalisation from the standard scenario (of set theory) where it is usually maintained that AC holds. AC fails in 'most' topoi, and has a surprising connection to logic.
Theorem 3. (Diaconescu) . In a topos E if AC holds then E is Boolean.
Though the converse fails in general, in any ( localic ) topos, for example the topos of presheaves on a poset, the converse does hold (see [42], proposition 7.5.5 or theorem 4 below). Thus we may tie the non-existence of a global section of Σ → 1 or w | ϕ 〉 → 1, and hence contextuality, to non-Boolean topos logic.
$$6. \text{ PRESHEAVES ON A POSET AND } \widehat { \mathcal { W } ( \mathcal { H } ) } \\ \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cd$$
In this section we present some basic facts about topoi of presheaves on a poset, and provide a construction for quantifying the non-Booleanness of the subobject classifier and therefore the topos as a whole. The quantum topos ̂ W H ( ) is introduced and physically motivated as representing a subtopos of the Isham-Butterfield topos ̂ V H ( ).
- 6.1. General case. Let ( P, ≤ ) denote a poset, henceforth written P . The following demonstrates that P is Boolean if and only if P consists of only one point:
Thus it is trivial to observe when a given topos of presheaves on a poset is Boolean. If C were just a single Abelian algebra (a poset with one point), ̂ C ∼ = Sets , i.e., one recovers the topos of sets and functions. However, we also wish to quantify the non-Booleanness of a presheaf topos on a poset, and to do this we first view ̂ P as a topos of sheaves on a topological space. This equivalence also allows for the demonstration that contextuality gives rise to non-Boolean logic.
̂ Proposition 4. ([43], lemma 1.4.12.) Let C be an arbitrary (locally small) category. ̂ C is Boolean if and only if C is a groupoid.
Definition 12. A down-set A on a poset P is a subset A ⊆ P for which x ∈ A and y ≤ x implies y ∈ A .
The down-sets ↓ P on P define the open sets of the anti-Alexandrov topology on P ; we write P for this space.
Definition 13. Let X be a topological space and F : O ( X ) → Sets a presheaf on the open sets O ( X ) . F is a sheaf if for any open set U and any open cover U = ∪ i U i of U and each collection f i ∈ F ( U i ) (of local sections) for which f i | U i ∩ U j = f j | U i ∩ U j , there exists a unique f ∈ F ( U ) such that f | U i = f i for each i .
We denote the collection of (set-valued) sheaves on X by Sh X ( ). It turns out (e.g., [44]) that the presheaves on P seen as a poset are exactly the sheaves on P :
Proposition 5. ̂ P ∼ = Sh ( P ) where ∼ = denotes isomorphism of categories. Furthermore (see [45], ch. II, proposition 4), with 1 denoting the terminal object in Sh ( P ) and with (19) and proposition 5,
$$\mathcal { O } ( P ) \cong \downarrow P \cong s u b ( 1 ) \cong \Gamma \Omega \quad \left ( 2 0 \right ) \\.. \frac { 1 } { 2 } \dots \dots \dots$$
This enables us to understand the logical structure of the truth values ΓΩ in ̂ P by directly looking at the poset P , which will be of use when we consider the topoi involved in quantum computation. Since subB is a Heyting algebra for any B in E , the complement of a subobject A /arrowtailright B is defined as the Heyting algebra complement, and the existence of complements is strongly tied to the logic.
Theorem 4. Let E be a topos of the form E = Sh X ( ) for X a topological space. The following are equivalent ([42], proposition 7.5.5, and [41], proposition 5.14) :
- (1) Ω is a Boolean Algebra (i.e., E is Boolean);
- (3) All subobjects have complements ;
- (2) AC holds in E ;
- (4) O ( X ) is a complete Boolean algebra.
One can see, for example, that when defined on the poset 13, w | ϕ 〉 /arrowtailright Σ has no complement, showing the non-classical structure of the phase space Σ. Theorem 4 also highlights the interrelatedness of many of category/topos theoretic concepts: non-Booleanness arises from the (failure of) AC, or via the lack of complements of subobjects, or by the failure of the base category to be a groupoid.
However, we wish to not only indicate whether a topos is non-Boolean, but to be able to say how non-Boolean it is. Equation (20) shows that ΓΩ is Boolean if and only if O ( P ) is, and by theorem 4, this is equivalent to ̂ P being Boolean. Hence the (non-)Booleanness of ΓΩ governs the (non)-Booleanness of P .
̂ Since we aim to show that non-Boolean logic is a resource for quantum computation, the quantification of such a resource is crucial, and making such a notion precise is the subject of the next subsection.
- 6.2. Quantifying non-Booleanness. In light of the fact that ̂ P is Boolean precisely when the Heyting algebra ΓΩ is, which by proposition 2 is the case precisely when each element of ΓΩ is complemented, the non-Booleanness of ΓΩ can be captured as the ratio of noncomplemented elements in ΓΩ to the total number of elements. In turn, this serves to indicate the non-Booleanness of P .
̂ We write comp(ΓΩ) for the sub-Heyting algebra of ΓΩ consisting of complemented elements, which is also a Boolean algebra, and write | comp(ΓΩ) | for its cardinality.
## Definition 14. The non-Booleanness q is given by
$$q \Gamma \Omega = 1 - \frac { \cdot } { | \Gamma \Omega | } & & ( 2 1 ) \\ \text{ents of $\Omega$ are commented $a\Gamma=0$ for a nose of fixed size $a$}$$
Thus, if all global elements of Ω are complemented, q ΓΩ = 0. For a poset of fixed size, q therefore defines a monotone function with range [ 0 1 , -2 | ΓΩ | ] . Only in the limit | ΓΩ | →∞ can q ΓΩ become 1.
- 6.3. The quantum topoi. Only in the last 20 years or so has topos theory been applied to problems in the foundations of quantum mechanics; topos quantum theory has attracted a large number of contributions in recent years, most focussing on the role of the many classical 'pictures' embedded in the quantum description (e.g. [46] and references therein).
The topos Sets captures the logical structure of classical physics, where Boolean logic plays a fundamental role both in the structure of propositions (i.e., distinguished subsets of the phase space) and in the truth evaluation of the given propositions - any such proposition is true or false, and the two truth values are arranged as a Boolean algebra. It has been argued by Isham and collaborators that one should use topoi other than Sets for construction of physical theories in general. Isham and D¨ oring in [17] provide a thorough discussion of the topos ̂ V H ( ) (which we recall is the topos of presheaves on the poset of Abelian von Neumann subalgebras of B ( H )). The spectral presheaf Σ carries a logical structure of its own and represents the phase space in the topos ̂ V H ( ).
The topos ̂ W H ( ) of interest to us is a subtopos of ̂ V H ( ) and these topoi have many features in common. We may view ̂ W H ( ) as the restriction to a specific physical situation of interest, with the given logical structure as the logic governing that situation.
Consider once more the poset W H ⊆ V H ( ) ( ) and the topos ̂ W H ( ). Combining definition 11 with theorem 4 we thus immediately see that non-Booleanness arises from sectionless epimaps; we thus have as a corollary of theorem 4 the following observation:
Corollary 1. Let E = ̂ W H ( ) and let Σ and w have the given definitions. If Σ has no global section or if there exists a | ϕ 〉 ∈ H for which w | ϕ 〉 has no global section, then the logic of E is non-Boolean.
Therefore,
Corollary 2. State-dependent or state-independent contextuality implies non-Boolean logic in E .
Thus state-dependent or independent contextuality may be viewed as a sufficient condition for the given topos to be non-Boolean, thus demonstrating a link between the Kochen Specker theorem and the logic of a topos closely related to a concrete physical situation . We now explore the connection between the non-classical logic of a topos and the promotion of nonuniversal to universal computation in MBQC.
## 7. Non-Boolean logic and quantum computation
Before analysing the role played by non-Boolean topos logic in the class of /lscript -2 MBQC of interest, we briefly summarise the various topics that have been presented thus far, and
how they are related. The state-dependent and state-independent Kochen-Specker theorems were rephrased in terms of the non-existence of global sections of the presheaves w | ϕ 〉 and Σ respectively, both defined on a poset W H ( ) ⊆ V H ( ), the latter representing the collection of Abelian von Neumann subalgebras of B ( H ), ordered by inclusion, and we presented Mermin's state-dependent proof in this language. Temporally flat deterministic /lscript -2 MBQC was introduced, with emphasis on a result in [11] that demonstrates the necessity of contextuality in the evaluation of non-linear Boolean functions, and we presented the simple example of Anders and Browne's OR-gate [12].
The connection to logic was then made by introducing some basic machinery of topos theory, where the Boolean logic of set theory is replaced by (the non-Boolean) 'intuitionistic logic', with Heyting algebras governing the structure of the multivalued truth object and the structure of subobjects of the 'phase space' Σ. The special case of (the topos of) presheaves on a poset, which encompasses the quantum case, was then analysed in greater detail, and we delineated conditions under which such a topos is Boolean. Via the topos version AC, we then established the major link between contextuality and logic: by constructing the topos ̂ ̂ W H ( ) from the poset W H ( ), we showed (corollary 2) that contextuality implies non-Boolean logic in W H ( ).
We are therefore presented with a link between non-Boolean logic and computation. With Φ a resource state for an /lscript -2 MBQC M as defined in section 4, the main result of [11], in the language of this paper, reads
Proposition 6. If M computes a non-linear Boolean function, then m | Φ 〉 has no global section, i.e., there is state-dependent contextuality for ( Φ | 〉 , W H ( )) .
Hence we immediately arrive at one of our main results:
Theorem 5. If M computes a non-linear Boolean function then ̂ W H ( ) is a non-Boolean topos.
In other words, non-classical logic is a necessary condition for non-linear function evalutation, and hence universal classical computation in /lscript -2 MBQC, the model of computation we consider. It does not seem unwarranted to view the non-Boolean logic of ̂ W H ( ) as affording a new potential 'resource' for quantum computation. Given other qualities previously mooted as computational resources-entanglement of | Φ 〉 for instance-we may view nonBooleanness in the same light - a view which is to be strengthened in subsection 7.1 where we will demonstrate the 'consumption' of the non-Booleanness by the execution of the Anders and Browne OR-gate. Just as the entanglement is 'used up' in each step of the computation, yielding a product state after the computational output has been produced, we will show that the non-Booleanness follows a similar trajectory, with the 'final topos' being a Boolean one.
We turn attention once more to the example of Anders and Browne's OR-gate, in order to demonstrate explicitly the consumption of non-Booleanness in the course of the computation, and how this is correlated to the decrease in computational power at each stage. Our main conclusions, however, pertain also to any MBQC satisfying definition 5.
7.1. Non-Booleanness as a computational resource. We again refer to the AndersBrowne OR-gate, and consider the situation after each measurement with regards to the non-Booleanness and computational power. The initial state is | Ψ 〉 = 1 √ 2 ( 000 | 〉 + 111 ), | 〉 with W H ( ) diagram 13 and associated topos ̂ W H ( ).
Due to the fact that ( Ψ | 〉 , W H ( )) computes an OR-gate, all Boolean functions can be evaluated. The quantity | ΓΩ for the topos | ̂ W H ( ) can be found by counting the down-sets in diagram 13, giving | ΓΩ = 113. | The only complemented elements are 0 and 1 (i.e., the empty set and the whole space), yielding q ΓΩ = 111 113. /
Without loss of generality we assume that X 1 is measured and the +1 outcome is registered, resulting in the state transformation | Ψ 〉 ↦→ | Ψ , with ′ 〉
This is the new resource state for the rest of the computation. It also specifies a subposet W H ( ) ′ ⊂ W H ( ): for deterministic computation in line with definition 5 the local observables are specified by
$$| \Psi ^ { \prime } \rangle = | + \rangle \otimes ( \frac { | 0 0 \rangle + | 1 1 \rangle } { \sqrt { 2 } } ). & & ( 2 2 ) \\ \intertext { c e state for the rest of the computation. } \text{ It also specifies a subpost}$$
$$X _ { 1 } X _ { 2 } X _ { 3 } \left | \Psi ^ { \prime } \right \rangle = \left | \Psi ^ { \prime } \right \rangle & & ( 2 3 ) \\ X _ { 1 } Y _ { 2 } Y _ { 3 } \left | \Psi ^ { \prime } \right \rangle = - \left | \Psi ^ { \prime } \right \rangle, \\ \text{the subposet defined by } V _ { 1 } \text{ and } V _ { 2 } \colon$$
i.e., we are restricted to the subposet defined by V 1 and V 2 :
The existence of a valuation ν compatible with the constraints (23) shows that ( | Ψ ′ 〉 , W H ( ) ′ ) is non-contextual (i.e., w | Ψ ′ 〉 has sections). However, ̂ W H ( ) ′ is non-Boolean, with nonBooleanness q ΓΩ = 3 5. Referring to definition 5 and its realisation in the Anders-Browne ′ / example (discussion subsequent to (14)), that Y 1 is missing in (23) constrains the possible computational inputs and therefore the computation as a whole. q 1 is fixed at q 1 = 0, and hence i 1 = 0 and the only computable function is o (0 0) = 0 and , o (0 1) = 1, i.e., a partial , function on Z 2 × Z 2 , or the identity function on one input.
Supposing, again without loss of generality, that X 2 is then measured and the outcome +1 is registered. Then the state after this measurement is
$$| \Psi ^ { \prime \prime } \rangle = | + + \rangle \otimes ( \frac { | 0 \rangle + | 1 \rangle } { \sqrt { 2 } } ) \equiv | + + + \rangle \,. \quad \text{(25)} \\ \text{ything is determined after just two measurements, as would be expected}$$
Therefore, everything is determined after just two measurements, as would be expected due to the constraints on the outcomes. This specifies the poset V 1 , i.e., local observables { X ,X ,X 1 2 3 } . The topos ̂ V 1 ≡ W H ̂ ( ) ′′ ∼ = Sets is Boolean, and of course q ΓΩ ′′ = 0. With the constraints on the computation i 1 = i 2 = q 1 = q 2 = q 3 = 0, the only possible computation is o (0 0) = 0. ,
The situation may be summarised as follows. Measurements 0, 1 and 2 gives rise, for the particular choice of measurements in the above discussion, to the sequence of states
↦
$$\left ( | \Psi \rangle = \frac { 1 } { \sqrt { 2 } } \left ( | 0 0 0 \rangle + | 1 1 1 \right ) \right ) \mapsto \left ( | \psi ^ { \prime } \rangle = \frac { 1 } { \sqrt { 2 } } | + \rangle \otimes \left ( | 0 0 \rangle + | 1 1 \right ) \right ) \mapsto \left ( | \Psi ^ { \prime \prime } \rangle = | + + + \rangle \right ), \\ \text{serviceif} v i n o \text{ as someone of nonsets}$$
↦
specifying a sequence of posets
$$\begin{array} { c c } & & \mathcal { W } ( \mathcal { H } ) \mapsto \mathcal { W } ( \mathcal { H } ) ^ { \prime } \mapsto \mathcal { W } ( \mathcal { H } ) ^ { \prime \prime }, & & ( 2 7 ) \\ \kappa \,. & &. \end{array}$$
↦
$$\widehat { \mathcal { W } ( \mathcal { H } ) } \mapsto \widehat { \mathcal { W } ( \mathcal { H } ) } ^ { \prime } \mapsto \widehat { \mathcal { W } ( \mathcal { H } ) } ^ { \prime \prime },$$
↦
specifying a sequence of topoi
specifying a sequence of Heyting algebras
$$\Gamma \Omega \mapsto \Gamma \Omega ^ { \prime } \mapsto \Gamma \Omega ^ { \prime \prime },$$
↦
↦
specifying a sequence of 'non-Booleannesses'
↦
$$\alpha \, \text{sequence} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \,
\quad \left ( q \Gamma \Omega = \frac { 1 1 1 } { 1 1 3 } \right ) & \mapsto \left ( q \Gamma \Omega ^ { \prime } = \frac { 3 } { 5 } \right ) & \mapsto \left ( q \Gamma \Omega ^ { \prime \prime } = 0 \right ) ; \\ \quad \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{end} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \,
\quad \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text`\, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text` \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \Text`\, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \, \text{and} \,
\end{$$
↦
accompanied by a sequence of 'computability classes'
$$( \text{All functions on $\mathbb{Z}_{2} \times \mathbb{Z}_{2}) \mapsto (identity function on $\mathbb{Z}_{2}) \mapsto (0 \mapsto 0)$.}$$
↦
↦
↦
Not only have entanglement and state dependent contextuality been exhausted by the computation, we may say that the pre-computation non-Booleanness has been 'consumed' during its course. Since the computational power manifestly decreases at every stage, we put forward the view that non-Boolean logic is a computational resource .
Any deterministic, temporally flat /lscript -2 MBQC specifies analogous sequences to those given above, always resulting in a final topos which is Boolean; the relevant post-computation poset in /lscript -2 MBQC will be given by a single point of the form { O 1 ( q 1 ) , ..., O n ( q n ) } ′′ (prime denoting commutant here). Hence our findings in the Anders and Browne example have a natural generalisation to the computations considered in this paper.
- 7.2. Internal perspective. It is also worth observing that the presheaf w | Ψ 〉 specifies the computation 'internally' to the topos ̂ W H ( ). For the context V 1 for example, w | Ψ 〉 gives rise to the family of sections { λ +++ , λ + --, λ -+ -, λ --+ } , each of which takes value 1 on X X X 1 2 3 which coincides with the computed output. Identical behaviour occurs for the three remaining contexts (on V 2 , for instance, we find the family of sections which evaluate X X X 1 2 3 as -1).
This raises the possibility of a fully internal description of /lscript -2 MBQC, along with an exploitation of the full power and machinery of topos theory proper. Any topos gives rise to a first order intuitionistic predicate logic - the Mitchell-Benabou language . The availability of the quantifiers ∀ and ∃ allows for the expression of a rich class of theorems; though the techniques required for a proof lie beyond the scope of this paper, it is possible to show that the Anders and Browne OR-gate takes the form of a theorem within the Mitchell-Benabou language. A full investigation of the scope of this observation remains a goal for the future.
## 8. Concluding Remarks and Outlook
This paper offers a footbridge between the thus far separate disciplines of quantum computation and 'quantum logic', whichever flavour of the latter one wishes to consider. We have established that non-Boolean logic is required for universal classical computation in /lscript -2 MBQC, consituting a new perspective on classical-over-quantum improvement. Furthermore, the non-Booleanness appears to be a commodity which is consumed through the act of computation. Interestingly, the specific logical rules which allow for the construction and truth evaluation of sentences in a topos have not been explicitly used; therefore further examination of the specific structure of the intuitionistic language in the topoi arising from MBQC, and what the language tells us about computation, seems a worthy goal.
## Acknowledgements
Thanks are due to Chris Heunen and Rebecca Ronke for valuable feedback on an earlier draft of this manuscript. LL, RD and RR acknowledge support from NSERC, CIFAR and IARPA; LL also acknowledges support from the John Templeton Foundation.
## References
- [1] Cleve, R., Ekert, A., Macciavello, C. & Mosca, M. 1998 Quantum algorithms revisited. Proc. R. Soc. A 454 , 339-354. (DOI 10.1098/rspa.1998.0164.)
- [2] Vidal, G. 2003 Efficient Classical Simulation of Slightly Entangled Quantum Computations. Phys. Rev. Lett. 91 . (DOI 0.1103/PhysRevLett.91.147902.)
- [3] Van den Nest, M., D¨r, W., Vidal, G. & Briegel, H.-J. 2007 Classical simulation versus universality in u measurement-based quantum computation. Phys. Rev. A 75 . (DOI 10.1103/PhysRevA.75.012337.)
- [4] Gottesman, D. 1997 Stabilizer Codes and Quantum Error Correction Ph.D. thesis, California Institute of Technology, 8232.
- [5] Bremner, M. J., Mora, C. & Winter, A. 2009 Are random pure states useful for quantum computation? Phys. Rev. Lett. 102 . (DOI 10.1103/PhysRevLett.102.190502.)
- [6] Gross, D., Flammia, S. T. & Eisert, J. 2009 Most Quantum States Are Too Entangled To Be Useful As Computational Resources. Phys. Rev. Lett. 102 . (DOI 10.1103/PhysRevLett.102.190501.)
- [7] Van den Nest, M. 2013 Universal Quantum Computation with Little Entanglement. Phys. Rev. Lett. 110 . (DOI 10.1103/PhysRevLett.110.060504.)
- [8] Bravyi, S. & and Kitaev, A. 2005 Universal quantum computation with ideal Clifford gates and noisy ancillas. Phys. Rev. A 71 . (DOI 10.1103/PhysRevA.71.022316.)
- [9] Veitch, V., Ferrie, C., Gross, D., & Emerson, J. 2012 Negative quasi-probability as a resource for quantum computation. N. J. Phys. 14 . (DOI 10.1088/1367-2630/14/11/113011.)
- [10] Howard, M., Wallman, J. J., Veitch, V. & Emerson, J. 2014 Contextuality supplies the magic for quantum computation. Nature 510 , 351-355. (DOI 10.1038/nature13460.)
- [11] Raussendorf, R. 2013 Contextuality in measurement based quantum computation. Phys. Rev. A 88 (DOI 10.1103/PhysRevA.88.022322.)
- [12] Anders, J. & Browne, D. 2009 Computational Power of Correlations. Phys. Rev. Lett . 102 (DOI 10.1103/PhysRevLett.102.050502.)
- [13] Mermin, N. D. 1993 Hidden variables and the two theorems of John Bell. Rev. Mod. Phys. 65 (DOI 10.1103/RevModPhys.65.803.)
- [14] Boole, G. 1958 An Investigation of the Laws of Thought , New Edition. Dover.
- [15] Dalla Chiara, M., Giuntini, R. & Greechie, R. 2004 Reasoning in Quantum theory , Springer.
- [16] Birkhoff, G. & von Neumann, J. 1936 The Logic of Quantum Mechanics. Ann. Math. 37 , 823843.
- [17] Isham, C. J. & D¨ring, A. 2011 'What is a thing'?: Topos Theory in the Foundations of Physics. In o New Structures for Physics, Springer Lecture Notes in Physics (ed. Bob Coecke). 813 , pp. 753-940. Springer, Heidelberg.
- [18] Flori, C. 2013 A first course in topos quantum theory . Springer.
- [19] Adelman, M. & Corbett, J. 1995 A sheaf model for intuitionistic quantum mechanics. App. Cat. Str. 3 , 79-104. (DOI 10.1007/BF00872949.)
- [20] Heunen, C., Landsman, N.P. & Spitters, S. 2009 A topos for algebraic quantum theory. Comm. Math. Phys. 291 , 63-110. (DOI 10.1007/s00220-009-0865-6.)
- [21] Isham, C. J., Butterfield, J. & Hamilton, A. 2000 Topos Perspective on the Kochen-Specker Theorem: III. Von Neumann Algebras as the Base Category. Int. J. Th. Phys. 39 , 1413-1436. (DOI 10.1023/A:1003667607842.)
- [22] Redhead, M. 1993 Incompleteness, Nonlocality and Realism , Reprint Edition. Clarendon Press.
- [23] Kochen, S. & Specker, E. P. 1967 The problem of hidden variables in quantum mechanics. J. Math. Mech. 17 , 59-87.
- [24] Abramsky, S. & Brandenburger, A. 2011 The sheaf-theoretic structure of non-locality and contextuality. N. J. Phys. 13 . (DOI 10.1088/1367-2630/13/11/113036.)
- [25] Abramsky, S., Mansfield, S. & Barbosa, R. S. 2012 The Cohomology of Non-Locality and Contextuality. Proceedings 8th International Workshop on Quantum Physics and Logic Nijmegen, Netherlands 95 , 1-14. Op. Pub. Ass. (DOI 10.4204/EPTCS.95.1.)
- [26] Greenberger, D. M., Horne, M. A. & Zeilinger, A. 1989 Going Beyond Bell's Theorem. In Bell's Theorem, Quantum Theory, and Conceptions of the Universe (ed. M. Kafatos), pp. 69 - 72. Kluwer, Dordrecht.
- [27] Mermin, N. D. 1990 Quantum Mysteries Revisited. Am. J. Phys. 58 , 731. (DOI 10.1119/1.16503)
- [28] Briegel, H.-J. & Raussendorf, R. 2001 Persistent Entanglement in Arrays of Interacting Particles. Phys. Rev. Lett . 86 , 910. (DOI 10.1103/PhysRevLett.86.910.)
- [29] Schlingemann, D. & Werner, R. F. 2001 Quantum error-correcting codes associated with graphs. Phys. Rev. A 65 . (DOI 10.1103/PhysRevA.65.012308.)
- [30] Raussendorf, R. & Briegel, H.-J. 2001 A one-way quantum computer. Phys. Rev. Lett. 86 . (DOI 10.1103/PhysRevLett.86.5188.)
- [31] Leung, D. W. 2004 Quantum Computation by Measurements. Int. J. Quant. Inf. 2 . (DOI 10.1142/S0219749904000055.)
- [32] Gross, D. & Eisert, J. 2007 Novel Schemes for Measurement-Based Quantum Computation. Phys. Rev. Lett. 98 . (DOI 10.1103/PhysRevLett.98.220503.)
- [33] Chen, X., Duan, R., Ji, Z. & Zeng, B. 2010 Quantum State Reduction for Universal Measurement Based Computation. Phys. Rev. Lett. 105 . (DOI 10.1103/PhysRevLett.105.020502.)
- [34] Miyake, A. 2011 Quantum computational capability of a 2D valence bond solid phase. Ann. Phys. 326 , 16561671 (DOI 10.1016/j.aop.2011.03.006.)
- [35] Wei, T. C., Affleck, I. & and Raussendorf, R. 2011 Affleck-Kennedy-Lieb-Tasaki State on a Honeycomb Lattice is a Universal Quantum Computational Resource. Phys. Rev. Lett. 106 . (DOI 10.1103/PhysRevLett.106.070501.)
- [36] Else, D. V., Schwarz, I., Bartlett, S. D. & Doherty, A. C. 2012 Symmetry-Protected Phases for Measurement-Based Quantum Computation. Phys. Rev. Lett. 108 . (DOI 10.1103/PhysRevLett.108.240505.)
- [37] Kadison, R. V. & Ringrose, J. R. 1997 Fundamentals of the theory of operator algebras volume I: elementary theory . American Mathematical Society.
- [38] Shor, P. 1994 Algorithms for quantum computation: discrete logarithms and factoring. Proceedings of the 35th Annual Symposium on Foundations of Computer Science, Santa Fe, NM , 124-134. IEEE Press, New York. (DOI 10.1109/SFCS.1994.365700.)
- [39] Mosca, M. & Zalka, C. 2004 Exact quantum Fourier transforms and discrete logarithm algorithms. Int. J. Quant. Inf. , 91-100. (DOI 10.1142/S0219749904000109.)
- [40] MacWilliams, F. J. & Sloane, N. A. 1977 The Theory of Error-Correcting Codes . North-Holland.
- [41] Johnstone, P. 2013 Topos Theory , Reprint Edition. Dover.
- [42] Borceux, F. 2008 Handbook of categorical algebra 3 Categories of Sheaves . CUP.
- [43] Johnstone, P. 2002 Sketches of an Elephant: A Topos Theory Compendium vol. I . OUP.
- [44] Johnstone, P. 2002 Sketches of an Elephant: A Topos Theory Compendium vol. II . OUP.
- [45] Moerdijk, I. & Mac Lane, S. 1992 Sheaves in Geometry and Logic, A First Introduction to Topos Theory , 1st edn. Springer.
- [46] Heunen, C. 2014 The many classical faces of quantum structures. arXiv:1412.2177
|
10.1098/rspa.2014.0716
|
[
"Leon Loveridge",
"Raouf Dridi",
"Robert Raussendorf"
] |
2014-08-04T17:02:33+00:00
|
2015-10-23T15:24:48+00:00
|
[
"quant-ph"
] |
Topos logic in measurement-based quantum computation
|
We report first steps towards elucidating the relationship between
contextuality, measurement-based quantum computation (MBQC) and the
non-classical logic of a topos associated with the computation. We show that,
in a class of MBQC, classical universality requires non-classical logic, which
is 'consumed' during the course of the computation, thereby pinpointing another
potential quantum computational resource.
|
1408.0746v2
|
## Effect of spin-orbit interactions on the 0.7 anomaly in quantum point contacts
Olga Goulko, 1, 2 Florian Bauer, 1 Jan Heyder, 1 and Jan von Delft 1
1 Physics Department, Arnold Sommerfeld Center for Theoretical Physics, and Center for NanoScience,
Ludwig-Maximilians-Universit¨t, a Theresienstraße 37, 80333 Munich, Germany
2 Present address: Department of Physics, University of Massachusetts, Amherst, MA 01003, USA
We study how the conductance of a quantum point contact is affected by spin-orbit interactions, for systems at zero temperature both with and without electron-electron interactions. In the presence of spin-orbit coupling, tuning the strength and direction of an external magnetic field can change the dispersion relation and hence the local density of states in the point contact region. This modifies the effect of electron-electron interactions, implying striking changes in the shape of the 0.7-anomaly and introducing additional distinctive features in the first conductance step.
PACS numbers: 71.70.Ej, 73.40.-c
Spin-orbit interactions (SOI) play an important role in a variety of fields within mesoscopic physics, such as spintronics and topological quantum systems. In this Letter we study the effects of SOI on the conductance of a quantum point contact (QPC), a one-dimensional constriction between two reservoirs [1, 2]. The linear conductance G of a QPC is quantized in multiples of G Q = 2 e /h 2 , showing the famous staircase as a function of gate voltage. In addition, at the onset of the first plateau, measured curves show a shoulderlike structure near 0 7 . G Q [3]. In this regime QPCs exhibit anomalous behavior in the electrical and thermal conductance, noise, and thermopower [3-11]. The microscopic origin of this 0 7-anomaly has . been the subject of a long debate [12-18]. It has recently been attributed to a strong enhancement of the effects of electron-electron interactions (EEI) by a smeared van Hove singularity in the local density of states (LDOS) at the bottom of the lowest QPC subband [15, 18]. While this explains the 0 7-anomaly without evoking SOI, the . presence of SOI can change the dispersion relation and hence the LDOS, thus strongly affecting the shape of the 0 7-anomaly. Previous studies of SOI in QPCs exist [19-. 23], but not with the present emphasis on their interplay with the QPC barrier shape and EEI, which are crucial for understanding the effect of SOI on the 0 7-anomaly. .
Setup. We consider a heterostructure forming a twodimensional electron system (2DES) in the xy -plane. Gate voltages are used to define a smooth, symmetric potential which splits the 2DES into two leads, connected by a short, one-dimensional channel along the x -axis: the QPC [1, 2]. The transition between the leads and the QPCis adiabatic. We also assume the confining potential in the transverse direction to be so steep that the subband spacing is much larger than all other energy scales relevant for transport, in particular those related to the magnetic field and SOI, and consider only transport in the first subband, corresponding to the lowest transverse mode. This can be described by a one-dimensional model with a smooth potential barrier and local EEI [18]. The magnetic field B is assumed to be in the xy -plane, acting as a pure Zeeman field, without orbital effects.
A moving electron in an electric field can experience an effective magnetic field B SOI proportional to its momen- tum glyph[planckover2pi1] k . Depending on the origin of the electric field one distinguishes between Rashba and Dresselhaus terms, the former resulting from the gradient of the external potential, and the latter from the asymmetry of the ionic lattice [24]. To be able to rotate B through any angle ϕ w.r.t. B SOI we require that B SOI also lies in the xy -plane. W.l.o.g. (see Supplement) we choose the y -axis to be parallel to B SOI , such that the SOI contribution to the Hamiltonian is -ασ k y , where α characterizes the strength of the (Rashba) SOI and σ y is a Pauli matrix [25]. We only consider the leading SOI contribution proportional to k and choose the spin quantization direction along B .
Without SOI, the dispersion relation glyph[planckover2pi1] 2 k / m 2 2 of a homogeneous one-dimensional model with effective mass m splits in the presence of a Zeeman field into two identical branches offset in energy by ± B/ 2. On the other hand, without a Zeeman field, the momentum-dependent SOI splits the dispersion in k -direction and also yields a negative spin-independent energy offset of magnitude ∆ E SOI = α m/ 2 2 glyph[planckover2pi1] 2 . In the following we shift the energy origin by -∆ E SOI and quote all energies w.r.t. the new origin. If both B and B SOI are non-zero, their interplay depends on ϕ , as illustrated in Fig. 1(a1-a3). In (a1), where the fields are parallel ( ϕ = 0), the energy offsets simply add, while for nonparallel fields a spin mixing occurs, resulting in an avoided crossing [26]. For orthogonal fields ( ϕ = π/ 2), the lower dispersion branch exhibits either one broader minimum at k = 0 if B ≥ 4 E SOI , or two minima at finite k and a maximum at k = 0 otherwise. The latter case is shown in Fig. 1(a2-a3).
Model. For the lowest subband we model the QPC by a symmetric potential barrier which is quadratic around its maximum,
$$V ( x ) \simeq V _ { g } + \mu - C _ { b } x ^ { 2 } / 2, \text{ \quad \ \ } ( 1 )$$
and vanishes smoothly at the boundary of the QPC. The barrier height V g , measured w.r.t. the chemical potential µ , mimics the role of the gate-voltage. If V g is swept downwards through zero, the conductance g = G/G Q increases from 0 to 1. For B = 0 this occurs in a single step whose width is given by the energy scale Ω x = √ C C b d ,
√
glyph[negationslash]
FIG. 1. Effect of SOI on the model without EEI, left columns (a1-d3), and with EEI, right column (e1-e4). The left columns (a1-d1), (a2-d2) and (a3-d3) represent different combinations of SOI strength R and angle ϕ between B and B SOI . They highlight the correspondence between the dispersion relation ω k ( ) in a homogeneous system (a1-a3), the LDOS for fixed ω = µ as function of V g on the central site of a QPC with potential barrier (b1-b3), the conductances of the two QPC transmission channels (c1-c3), and the total conductance of the QPC for several equally-spaced magnetic field values between B = 0 and B = 0 88Ω . x (d1-d3). In (a1-c3) the magnetic field is fixed at B = 0 88Ω , with dashed lines showing the case . x B = 0 for comparison. The line colors in (a1-a3) quantify the contribution of each spin state (red= ↑ , blue= ) in the dispersion branches, ↓ to illustrate the spin mixing at ϕ = 0. The right column (e1-e4) shows the total conductance for U > 0, with ϕ = π/ 2 and several combinations of R and B (the latter were chosen smaller than in (d1-d3), since EEI enhance the g -factor [18]).
which is set by the fixed curvature of the barrier, C b , and the curvature of the bulk dispersion at its minimum, C d [27]. For ϕ = 0, C d = glyph[planckover2pi1] 2 /m .
SOI in a QPC is determined by the dimensionless parameter
For numerical purposes we discretize real space and obtain an infinite tight-binding chain with spacing a , taking B and α constant throughout the chain. The noninteracting Hamiltonian is
$$H _ { 0 } & = \sum _ { j, \sigma, \sigma ^ { \prime } } d _ { j \sigma } ^ { \dagger } \left [ ( V _ { j } + 2 \tau ) \delta _ { \sigma \sigma ^ { \prime } } - \frac { 1 } { 2 } ( \sigma \cdot B ) _ { \sigma \sigma ^ { \prime } } \right ] d _ { j \sigma ^ { \prime } } \quad ( 2 ) \quad \text{A stro} \\ & \quad + \sum _ { j, \sigma, \sigma ^ { \prime } } \left [ d _ { j + 1 \sigma } ^ { \dagger } \left ( - \tau _ { 0 } \delta _ { \sigma \sigma ^ { \prime } } + \frac { i \alpha } { 2 } ( \sigma _ { y } ) _ { \sigma \sigma ^ { \prime } } \right ) d _ { j \sigma ^ { \prime } } + \text{h.c.} \right ], \quad [ 2 6 ] \text{ a} \\ & \quad \text{Small} \text{.}$$
where d jσ annihilates an electron with spin σ ∈{↑ ↓} ≡ , { + , -} at site j . The effective mass of the charge carrier is m = glyph[planckover2pi1] 2 / 2 τa 2 with τ = √ τ 2 0 + α 2 [28]. We keep τ fixed when varying α . The QPC barrier potential V j = V ( ja ) (and later EEI) are nonzero only in a region of length L = 2 Na centered around j = 0, representing the QPC. All results shown are for N = 50. We use the smooth function V ( x ) = ( V g + µ ) exp[ -(2 x/L ) 2 / (1 -(2 x/L ) 2 )] for the potential, with µ = 2 . τ Sites j< -N and j>N represent two leads with bandwidth 4 τ . The strength of
$$R = \sqrt { \frac { \Delta E _ { S O I } } { \Omega _ { x } } } = \frac { \alpha } { \hbar } \sqrt { \frac { m } { 2 \Omega _ { x } } }.$$
SOI strengths of up to α glyph[similarequal] 10 -11 eVm have been reported in the literature [19, 26, 29, 30]. Typical values of Ω x glyph[similarequal] 1meV and m glyph[similarequal] 0 05 . m e for InGaAs yield R glyph[similarequal] 0 2. . A stronger spin-orbit effect due to an enhancement of the anisotropic Lande g -factor is reported in [31]. Hole quantum wires have been used to observe the spin-orbit gap [26] and the anisotropic Zeeman splitting [32]. For hole QPCs, the larger effective hole mass and the resulting smaller Ω x imply larger values of R . Here we consider both small and large R , where R glyph[lessorsimilar] 0 4 is a realistic scale . for electron systems and R glyph[greaterorsimilar] 1 is accessible using hole systems [33], for QPCs with small barrier curvature C b and hence small Ω x .
System without EEI. Many insights on the interplay between SOI and geometry can already be gained from the model without EEI, as shown in the left part (a1d3) of Fig. 1. We discuss exact results for two physical quantities, which we also relate to the bulk dispersion relation: the linear conductance g and the LDOS
glyph[negationslash]
A σ j ( ω ) = -Im G σσ jj ( ω /πa ) , where G σσ ′ jj ′ is the retarded propagator from site j ′ with spin σ ′ to site j with spin σ . Due to SOI, spin is not conserved for ϕ = 0 and hence G σσ ′ jj is not spin-diagonal. However at j = 0 its off-diagonal elements turn out to be negligible compared to the diagonal ones. Thus it is meaningful to analyze the LDOS at j = 0 for given σ . The linear conductance at zero temperature can be calculated via g = g 1 + g 2 ∝ Tr( t † t ) [34], where t σσ ′ = G σσ ′ -N,N ( µ ) is the transmission matrix of the QPC and Tr( t † t ) equals the sum of the eigenvalues of t † t . The spin structure of t depends on N , but the eigenvalues of t † t , which yield the conductances g 1 and g 2 of the two transmission channels, do not.
For ϕ = 0 (Fig. 1, left column) spin is conserved and SOI have no influence on the LDOS and the conductance. This case is analogous to the one discussed in [18]. The bulk (i.e. V ( x ) = 0) LDOS,
$$\mathcal { A } _ { \text{bulk} } ^ { \sigma } ( \omega ) \, \infty \, \frac { \partial k } { \partial \omega } \Big | _ { \sigma } = \sqrt { \frac { m } { 2 \hslash ^ { 2 } ( \omega + \sigma B / 2 ) } }, \quad ( 4 ) \, \begin{pmatrix} 4 & \mathcal { A } _ { 0 } ^ { * } ( \\ \text{Betv} \\ \min _ { \cdot \,, } \end{pmatrix}$$
has a van Hove singularity, diverging at the minimum ω = -σB/ 2 of the corresponding dispersion branch, where the electron velocity vanishes. In the QPC, the x -dependent LDOS is shifted in energy by the barrier potential V ( x ). Since the barrier breaks translational invariance, the van Hove singularity is smeared out on a scale set by Ω x [15], forming a ridgelike structure, called van Hove-ridge in [18]. The LDOS height becomes finite, of order O ( √ m/ ( glyph[planckover2pi1] 2 Ω )), determined by Ω x x and the curvature glyph[planckover2pi1] 2 /m of the bulk dispersion. At a given position x , the LDOS maximum occurs at an energy which is O (Ω ) larger than the corresponding potential energy x V ( x ) -σB/ 2. Here and henceforth we quote the LDOS as a function of V g at fixed ω = µ . Figure 1(b1) shows it at the central site j = 0; the spatially resolved LDOS is shown in Fig. 1 of the Supplement. The LDOS has the same shape for both spins. Its structure is clearly inherited from that of the dispersion in (a1), with peak energies aligned with the dispersion minima up to the shift of O (Ω ). x Similarly, the conductances g 1 ( V g ) and g 2 ( V g ) of the two channels in (c1) show steps of the same shape with widths ∝ Ω x [27], split by B and aligned with the dispersion minima. This causes the total conductance g V ( g ) in (d1) to split symmetrically into a double step with increasing field, just as for a QPC without SOI.
Next consider the case ϕ = π/ 2 shown in Fig. 1(a2-d3). Spin mixing leads to an avoided crossing with spin gap ∝ B , which splits the dispersion into an upper branch with a narrow minimum and a lower branch with two minima and one maximum (for B < 4 E SOI ). Note that bulk LDOS structures separated in energy by less than Ω x are not resolved within the QPC. In the following we give an intuitive explanation of how the dispersion minima relate to the properties of the LDOS peaks and the conductance steps. The curvatures of the lower and upper dispersion branches are, respectively, smaller or larger than in (a1), C d1 < C d < C d2 (loosely speaking, C d1 is the effective curvature obtained by smearing the double dispersion minimum by Ω x , yielding a single minimum). Because the barrier curvature C b is fixed, this results in two modified energy scales Ω xi = √ C C b d i , with Ω x 1 < Ω x < Ω x 2 , which determine the LDOS peak heights and widths, as well as the conductance step widths. Consequently, in (b2) the LDOS peak for A ↓ 0 is lower and wider than for A ↑ 0 . Likewise, in (c2) the conductance step for g 2 ( V g ) is wider than for g 1 ( V g ), causing g V ( g ) in (d2) to show a striking asymmetry for its B -induced evolution from a single to a double step. This asymmetry is reminiscent of but unrelated to that known for the 0.7-anomaly - the latter is driven by EEI, as discussed below - but should be observable in higher conductance steps, where EEI are weaker.
For R glyph[greaterorsimilar] 1 more structures emerge, see Fig. 1(a3-b3). Spin-mixing produces an additional 'emergent' peak in A ↓ 0 (b3) and an additional step in g 2 ( V g ) (c3) near V g glyph[similarequal] 0. Between the two steps, the transmission g 2 ( V g ) has a minimum, corresponding to the spin gap, and the total conductance g V ( g ) in (d3) likewise develops a spin gap minimum with increasing B . These features can be understood by looking at the spin composition of the two bulk dispersion branches, depicted quantitatively through the colors in Fig. 1(a1-a3). At k = 0 the SOI field is zero and we have pure spin-states w.r.t. the chosen quantization. At larger | k | the SOI field increases, leading to spin-mixing. In fact in the limit k → ∞ we find a fully mixed state with equal up/down contributions. Since the upper branch minimum at k = 0 is in a pure spin-down state it corresponds to a peak only in A ↓ . But the minima of the lower branch are shifted away from k = 0 and have a spin down share besides the dominant spin up contribution. This causes the emergent peak in A ↓ at low frequencies, whose height increases with R , due to the stronger spin-mixing.
Interacting system. We now include EEI via H int = ∑ j U d j † j ↑ d j ↑ d † j ↓ d j ↓ . The on-site interaction U j = U ja ( ) is switched on smoothly over the QPC according to U x ( ) = U exp( -(2 x/L ) 6 / (1 -(2 x/L ) 2 )). We set U j = 0 for | j | > N , because outside the QPC region transverse confinement is weak or absent, and screening strong [18, 35]. We calculate the conductance at zero temperature with the functional Renormalization Group technique in the one-particle irreducible version [28, 3639] using the coupled ladder approximation, which was presented in [35] for a model without SOI. Generalizations necessary in the presence of SOI are described in the Supplement.
The B -dependence of the conductance for ϕ = π/ 2 and different R in the presence of EEI is shown in the right column (e1-e4) of Fig. 1 and the corresponding transconductance d g/ d V g in Fig. 2(b-f). The case R = 0, see Figs. 1(e1) and 2(a-c), which is equivalent to ϕ = 0, has been discussed in [18, 35]: once a finite magnetic field breaks the spin degeneracy a surplus of spin-up electrons
|
|
FIG. 2. fRG results for the conductance g and transconductance dg/dV g , for U = 3 36 . √ Ω x τ at zero temperature. Top row: 3d or colorscale plots of the conductance (a) and the transconductance (b,c) as functions of V g and B , for R = 0. Three bottom rows (d-l): Colorscale plots of the transconductance for three choices of R (three columns), plotted as a function of V g and either B for fixed ϕ = π/ 2, (second row), or of ϕ for fixed B = 0 18Ω . x (third row) and B = 0 88Ω . x (fourth row).
develops in the QPC, so that spin-down electrons experience both a Zeeman and a Coulomb energy cost. This Stoner-type effect depends on the LDOS at µ and hence is strongest when the apex of the van Hove ridge touches the chemical potential, i.e. when V g is within glyph[similarequal] 0 5Ω . x below 0 [18]. This causes an asymmetry w.r.t. V g = 0 in the B -induced evolution of g V ( g ) from a single to a double step in Fig. 1(e1), in contrast to the case without EEI in Fig. 1(d1). This asymmetry is characteristic of the 0.7-anomaly. The corresponding transconductance in Fig. 2(b-c) shows a double peak whose spacing increases roughly linearly with B (with an EEI-enhanced g-factor), as seen in numerous experiments [3, 10, 18].
glyph[negationslash]
The Stoner-type Coulomb enhancement of a fieldinduced population imbalance is amplified when R = 0, as shown in Figs. 1(e2-e4) and 2(d-f), because of the height imbalance for the spin-up and spin-down LDOS peaks caused by SOI. Correspondingly, with increasing R the double-step structure in the conductance becomes more pronounced, the second substep becoming much broader than the first, see Figs. 1(e2-e3), and the transconductance in Fig. 2(d-e) shows a weakening of the lowerV g peak with increasing R . This reflects the increasing curvature C d2 of the upper dispersion branch (and hence larger step width Ω x 2 ). For R glyph[greaterorsimilar] 1, additional features, inherited from the noninteracting case, emerge for g V ( g ) in Fig. 1(e4): a local maximum (marked by an arrow), followed by a spin gap minimum at lower V g . For the transconductance, Fig. 2(f), these features show up as a strong secondary peak around V / g Ω x glyph[similarequal]1 (marked by an arrow), followed by a region of negative transconductance (black). EEI also induce a secondary 0.7-type double-step feature in g V ( g ) for V / g Ω x between 0 and -1, Fig. 1(e4), which is similar to, but narrower than that for R = 0. It originates from the main LDOS peak in A ↑ 0 and the emergent peak in A ↓ 0 . Unlike the regular A ↓ 0 peak aligned with the upper dispersion branch, whose V g -position is governed by the magnetic field, the emergent A ↓ 0 peak occurs, due to strong spin-mixing, at nearly B -independent energy close to the A ↑ 0 peak. As a result, the two transconductance maxima in Fig. 2(f) remain parallel with increasing B , in strong contrast to the situation for R < 1 in Fig. 2(c-e).
Figures 2(g-l) show, for two fixed values of B , how the transconductance evolves as | ϕ | is increased from 0 to π/ 2, thus switching on the effects of SOI. The decrease in peak spacing with increasing | ϕ | in Fig. 2(l) strikingly reflects the increasing importance of spin mixing. The strong angle-dependence predicted here is a promising candidate for an experimental test of our theory [40].
At small nonzero temperature, inelastic scattering causes a Fermi-liquid-type reduction of the conductance, g T, V ( g ) /g (0 , V g ) = 1 -( T/T ∗ ) 2 for T glyph[lessmuch] T ∗ , with a V g -dependent low-energy scale T ∗ ( V g ). We expect its magnitude to be similar to the case without SOI, typically glyph[similarequal] 1K [18]. Thus, for T glyph[lessorsimilar] 0 1K, the . T -dependence should be very weak and the T = 0 predictions applicable.
In summary, we have shown that in the presence of SOI, the changes in the dispersion induced by the interplay of B and B SOI can strongly affect the shape of the 0.7-anomaly. In the absence of EEI, SOI cause an anisotropic response of the spin splitting to the applied in-plane magnetic field. With EEI, the 0.7-anomaly also develops an anisotropic response to magnetic field, and if SOI are strong, the conductance develops additional features due to the interplay of EEI and SOI: for ϕ = π/ 2 these include a field-induced double step in the conductance that does not split linearly with B , followed by a spin gap minimum. The dependence of the conductance on the angle between B and B SOI is already apparent for R glyph[similarequal] 0 4, which is accessible in experiments with electron . QPCs. Hole QPCs with R glyph[greaterorsimilar] 1 would allow access to regimes with strong SOI.
An experimental verification of our predictions would highlight the influence of LDOS features on the conductance and thus lend further support to the van Hove scenario of Ref. [18] as microscopic explanation for the 0 7-anomaly. . More generally, our work lays out a conceptual framework for analyzing the interplay of SOI,
EEI and barrier shape in quasi-1D geometries: examine how SOI and barrier shape modify the (bare) LDOS near µ - whenever the LDOS is large, EEI effects are strong. We expect this to be relevant for the more complicated hybrid superconductor-semiconductor junctions currently studied by seekers of Majorana fermions [4143]. A proper analysis of such systems would require a generalization of our approach to include superconduct-
- [1] D. A. Wharam, T. J. Thornton, R. Newbury, M. Pepper, H. Ahmed, J. E. F. Frost, D. G. Hasko, D. C. Peacock, D. A. Ritchie, and G. A. C. Jones, Journal of Physics C: Solid State Physics 21 , L209 (1988).
- [2] B. J. van Wees, H. van Houten, C. W. J. Beenakker, J. G. Williamson, L. P. Kouwenhoven, D. van der Marel, and C. T. Foxon, Phys. Rev. Lett. 60 , 848 (1988).
- [3] K. J. Thomas, J. T. Nicholls, M. Y. Simmons, M. Pepper, D. R. Mace, and D. A. Ritchie, Phys. Rev. Lett. 77 , 135 (1996).
- [4] N. J. Appleyard, J. T. Nicholls, M. Pepper, W. R. Tribe, M. Y. Simmons, and D. A. Ritchie, Phys. Rev. B 62 , R16275 (2000).
- [5] A. Kristensen, H. Bruus, A. E. Hansen, J. B. Jensen, P. E. Lindelof, C. J. Marckmann, J. Nyg˚rd, a C. B. Sørensen, F. Beuscher, A. Forchel, and M. Michel, Phys. Rev. B 62 , 10950 (2000).
- [6] S. M. Cronenwett, H. J. Lynch, D. Goldhaber-Gordon, L. P. Kouwenhoven, C. M. Marcus, K. Hirose, N. S. Wingreen, and V. Umansky, Phys. Rev. Lett. 88 , 226805 (2002).
- [7] L. DiCarlo, Y. Zhang, D. T. McClure, D. J. Reilly, C. M. Marcus, L. N. Pfeiffer, and K. W. West, Phys. Rev. Lett. 97 , 036810 (2006).
- [8] O. Chiatti, J. T. Nicholls, Y. Y. Proskuryakov, N. Lumpkin, I. Farrer, and D. A. Ritchie, Phys. Rev. Lett. 97 , 056601 (2006).
- [9] L. W. Smith, A. R. Hamilton, K. J. Thomas, M. Pepper, I. Farrer, J. P. Griffiths, G. A. C. Jones, and D. A. Ritchie, Phys. Rev. Lett. 107 , 126801 (2011).
- [10] A. P. Micolich, Journal of Physics: Condensed Matter 23 , 443201 (2011).
- [11] R. Danneau, O. Klochan, W. R. Clarke, L. H. Ho, A. P. Micolich, M. Y. Simmons, A. R. Hamilton, M. Pepper, and D. A. Ritchie, Phys. Rev. Lett. 100 , 016403 (2008).
- [12] D. J. Reilly, T. M. Buehler, J. L. O'Brien, A. R. Hamilton, A. S. Dzurak, R. G. Clark, B. E. Kane, L. N. Pfeiffer, and K. W. West, Phys. Rev. Lett. 89 , 246801 (2002).
- [13] K. F. Berggren and I. I. Yakimenko, Phys. Rev. B 66 , 085323 (2002).
- [14] Y. Meir, K. Hirose, and N. S. Wingreen, Phys. Rev. Lett. 89 , 196802 (2002).
- [15] C. Sloggett, A. I. Milstein, and O. P. Sushkov, The European Physical Journal B 61 , 427 (2008).
- [16] A. M. Lunde, A. D. Martino, A. Schulz, R. Egger, and K. Flensberg, New J. Phys. 11 , 023031 (2009).
- [17] K. Aryanpour and J. E. Han, Phys. Rev. Lett. 102 , 056805 (2009).
- [18] F. Bauer, J. Heyder, E. Schubert, D. Borowsky, D. Taubert, B. Bruognolo, D. Schuh, W. Wegscheider, J. von Delft, and S. Ludwig, Nature 501 , 73 (2013).
ing effects.
We thank S. Fischer, A. Hamilton, K. Hudson, S. Ludwig, C. Marcus, A. Micolich and A. Srinivasan for interesting and useful discussions and acknowledge support from the DFG via SFB-631, SFB-TR12, De730/4-3, and the Cluster of Excellence Nanosystems Initiative Munich . O.G. acknowledges support from the NSF under the grant PHY-1314735.
- [19] M. Kohda, S. Nakamura, Y. Nishihara, K. Kobayashi, T. Ono, J.-i. Ohe, Y. Tokura, T. Mineno, and J. Nitta, Nat Commun 3 , 1082 (2012).
- [20] J.-H. Hsiao and T.-M. Hong, Phys. Rev. B 82 , 115309 (2010).
- [21] M. P. Nowak, K. Kolasi´ski, n and B. Szafran, Phys. Rev. B 90 , 035301 (2014), arXiv:1401.1707 [cond-mat.meshall].
- [22] R. Danneau, W. R. Clarke, O. Klochan, A. P. Micolich, A. R. Hamilton, M. Y. Simmons, M. Pepper, and D. A. Ritchie, Applied Physics Letters 88 , 012107 (2006).
- [23] F. Nichele, S. Chesi, S. Hennel, A. Wittmann, C. Gerl, W. Wegscheider, D. Loss, T. Ihn, and K. Ensslin, Phys. Rev. Lett. 113 , 046801 (2014).
- [24] R. Winkler, Spin-Orbit Coupling Effects in TwoDimensional Electron and Hole Systems (Springer Berlin Heidelberg, 2003).
- [25] L. Meier, G. Salis, I. Shorubalko, E. Gini, S. Sch¨n, o and K. Ensslin, Nature Physics 3 , 650 (2007).
- [26] C. H. L. Quay, T. L. Hughes, J. A. Sulpizio, L. N. Pfeiffer, K. W. Baldwin, K. W. West, D. Goldhaber-Gordon, and R. de Picciotto, Nature Physics 6 , 336 (2010).
- [27] M. B¨ uttiker, Phys. Rev. B 41 , 7906(R) (1990).
- [28] J. E. Birkholz, Spin-orbit interaction in quantum dots and quantum wires of correlated electrons - A way to spintronics? , Ph.D. thesis, Georg-August-Universit¨t zu a G¨ ottingen (2008).
- [29] J. Nitta, T. Akazaki, H. Takayanagi, and T. Enoki, Phys. Rev. Lett. 78 , 1335 (1997).
- [30] T. Kita, T. Kakegawa, M. Akabori, and S. Yamada, Solid State Communications 136 , 479 (2005).
- [31] T. P. Martin, A. Szorkovszky, A. P. Micolich, A. R. Hamilton, C. A. Marlow, R. P. Taylor, H. Linke, and H. Q. Xu, Phys. Rev. B 81 , 041303 (2010).
- [32] J. C. H. Chen, O. Klochan, A. P. Micolich, A. R. Hamilton, T. P. Martin, L. H. Ho, U. Z¨licke, D. Reuter, u and A. D. Wieck, New J. Phys. 12 , 033043 (2010).
- [33] A. R. Hamilton, T. Li, O. P. Sushkov, and R. Winkler, For weakly 1D-confined holes, it can be argued that the dominant SOI term has the same kσ y form as used here. Private communication. (2014).
- [34] S. Datta, Electronic Transport in Mesoscopic Systems (Cambridge University Press, 1995).
- [35] F. Bauer, J. Heyder, and J. von Delft, Phys. Rev. B 89 , 045128 (2014).
- [36] C. Wetterich, Physics Letters B 301 , 90 (1993).
- [37] J. E. Birkholz and V. Meden, Phys. Rev. B 79 , 085420 (2009).
- [38] F. Bauer, 0.7 Anomaly of Quantum Point Contacts: Treatment of Interactions with Functional Renormalization Group. , Master's thesis, LMU-M¨nchen (2008). u
- [39] W. Metzner, M. Salmhofer, C. Honerkamp, V. Meden, and K. Sch¨nhammer, Rev. Mod. Phys. o 84 , 299 (2012).
- [40] A. R. Hamilton, K. Hudson, A. P. Micolich, and A. Srinivasan, private communication (2014).
- [41] R. M. Lutchyn, J. D. Sau, and S. Das Sarma, Phys. Rev. Lett. 105 , 077001 (2010).
- [42] Y. Oreg, G. Refael, and F. von Oppen, Phys. Rev. Lett. 105 , 177002 (2010).
- [43] V. Mourik, K. Zuo, S. M. Frolov, S. R. Plissard, E. P. A. M. Bakkers, and L. P. Kouwenhoven, Science 336 , 1003 (2012).
## SUPPLEMENTAL MATERIAL TO 'EFFECT OF SPIN-ORBIT INTERACTIONS ON THE 0.7 ANOMALY IN QUANTUM POINT CONTACTS'
## 1. Geometric details of the model
In our model the 2DES is in the xy -plane and the QPC is directed along the x -axis (this is the direction of motion of the charge carrier). For the directions of the B and B SOI fields we impose the following restrictions. To avoid orbital effects we require the magnetic field B to be in the xy -plane of the 2DES. We also want to be able to rotate B through any angle ϕ w.r.t. B SOI , which implies that B SOI also must lie in the xy -plane. With the latter condition, the B SOI field can be either parallel to the direction of motion of the electrons (pure Dresselhaus contribution), or orthogonal to it (pure Rashba contribution), or a combination of the two. But for our mathematical model, the end results depend only on the relative angle ϕ between B and B SOI . This means that we can choose the direction of B SOI without loss of generality. We choose B SOI to be parallel to the y -axis, c.f. Eq. (2) of the main text.
## 2. The spatially resolved LDOS
In Fig. 1(b1-b3) of the main text we show the LDOS at fixed µ as a function of V g on the central site of the QPC. The behavior at the center captures all relevant features. For completeness we include here in Fig. 3 the spatially resolved plots of the LDOS A σ j for both spin states and the same parameter values as in Fig. 1 of the main text.
## 3. Second order fRG
The functional Renormalization Group (fRG) method is an improved perturbation technique [28, 36-39]. Rather than expanding the Green's function in orders of the coupling and truncating the expansion, fRG introduces a flow parameter Λ into the free Green's function G 0 . At zero temperature we define
$$\mathcal { G } _ { 0 } ( i \omega ) \to \theta ( | \omega | - \Lambda ) \mathcal { G } _ { 0 } ( i \omega ) \equiv \mathcal { G } _ { 0 } ^ { \Lambda } ( i \omega ), \quad ( 5 ) \quad \text{$r$} \quad \text{$tion}$$
where θ ω ( ) is the Heaviside step function. At the beginning of the flow, setting Λ = ∞ yields G Λ 0 = 0, which means that the only contribution to the full Green's function comes from the bare vertex. At the end of the flow, Λ = 0 recovers the full model. The technical details of the one-particle irreducible version of the fRG employed in this work are presented in depth in [35]. We use the static approximation at zero temperature, which will be described below in Sec. 3 b. Since [35] does not deal with spin-orbit interactions, no spin-mixing is possible, which introduces additional symmetries into the system.
FIG. 3. Spatially resolved plots of the noninteracting LDOS A σ j at fixed ω = µ , plotted as a function of gate voltage V g and site index j , for B = 0 88Ω . x and for spin σ = ↑ (top row) and σ = ↓ (bottom row). Left column: R = 0 84, . ϕ = 0. Middle column: R = 0 84, . ϕ = π/ 2. Right column: R = 1 26, . ϕ = π/ 2. All results shown are for N = 50.
In our case these symmetries are no longer present. In this Supplement we focus on the generalizations necessary to account for such spin-mixing terms.
The second-order fRG flow equations in the oneparticle irreducible version and in the static approximation are given by
$$\frac { d } { d \Lambda } \gamma _ { 1 } ^ { \Lambda } ( \vec { q } _ { 1 } ^ { \prime }, q _ { 1 } ) = \frac { 1 } { 2 \pi } \sum _ { \omega = \pm \Lambda } \sum _ { \vec { q } _ { 2 } ^ { \prime }, q _ { 2 } } \widetilde { \mathcal { G } } _ { q _ { 2 }, \vec { q } _ { 2 } ^ { \prime } } ^ { \Lambda } ( i \omega ) \gamma _ { 2 } ^ { \Lambda } ( \vec { q } _ { 2 } ^ { \prime }, q _ { 1 } ^ { \prime } ; q _ { 2 }, q _ { 1 } ),$$
$$\frac { d } { d \Lambda } \gamma _ { 2 } ^ { \Lambda } = \frac { d } { d \Lambda } ( \gamma _ { p } ^ { \Lambda } + \gamma _ { x } ^ { \Lambda } + \gamma _ { d } ^ { \Lambda } ), \text{ where }$$
$$\frac { \text{d} } { \text{d} \varUpsilon _ { p } ^ { \Lambda } ( q _ { 1 } ^ { \prime }, q _ { 2 } ^ { \prime } ; q _ { 1 }, q _ { 2 } ) } = \frac { 1 } { 2 \pi } \sum _ { \omega = \pm \Lambda } \sum _ { \alpha _ { 0 } ^ { \prime }, q _ { 0 }, q _ { 0 } ^ { \prime }, q _ { 4 } } \frac { 1 } { 2 } \gamma _ { 2 } ^ { \Lambda } ( q _ { 1 } ^ { \prime }, q _ { 2 } ^ { \prime } ; q _ { 3 }, q _ { 4 } ) \widetilde { \mathcal { G } } _ { q _ { 3 }, q _ { 4 } ^ { \prime } } ^ { \Lambda } ( i \omega ) \widetilde { \mathcal { G } } _ { q _ { 4 }, q _ { 4 } ^ { \prime } } ^ { \Lambda } ( - i \omega ) \gamma _ { 2 } ^ { \Lambda } ( q _ { 0 } ^ { \prime }, q _ { 4 } ^ { \prime } ; q _ { 1 }, q _ { 2 } ),$$
$$\frac { d } { d \Lambda } \gamma _ { x } ^ { \Lambda } ( \tilde { \gamma } _ { 1 }, \, q _ { 2 } ^ { \prime } ; q _ { 1 }, q _ { 2 } ) = \frac { 1 } { 2 \pi } \sum _ { \omega = \pm \Lambda } \sum _ { \alpha _ { 3 } ^ { \prime }, \, q _ { 3 }, \, q _ { 4 } ^ { \prime }, \, q _ { 4 } } \gamma _ { 2 } ^ { \Lambda } ( \tilde { \gamma } _ { 1 } ^ { \prime }, \, q _ { 4 } ^ { \prime } ; q _ { 3 }, q _ { 2 } ) \widetilde { \mathcal { G } } _ { q _ { 3 }, \, \alpha _ { 3 } ^ { \prime } } ^ { \Lambda } ( i \omega ) \widetilde { \mathcal { G } } _ { q _ { 4 }, \, \alpha _ { 4 } ^ { \prime } } ^ { \Lambda } ( i \omega ) \gamma _ { 2 } ^ { \Lambda } ( q _ { 3 } ^ { \prime }, \, \alpha _ { 2 } ^ { \prime } ; q _ { 1 }, \, q _ { 4 } ),$$
$$\frac { \text{d} } { \text{d} } \gamma _ { d } ^ { \Lambda } ( q _ { 1 } ^ { \prime }, q _ { 2 } ^ { \prime } ; q _ { 1 }, q _ { 2 } ) = - \frac { 1 } { 2 \pi } \sum _ { \omega = \pm \Lambda \, q _ { 3 }, q _ { 3 }, q _ { 4 }, q _ { 4 } } ^ { \bullet \, \bullet \, \bullet \, \bullet } \gamma _ { 2 } ^ { \Lambda } ( q _ { 1 } ^ { \prime }, q _ { 3 } ^ { \prime } ; q _ { 1 }, q _ { 4 } ) \widetilde { \mathcal { G } } _ { q _ { 4 }, q _ { 4 } } ^ { \Lambda } ( i \omega ) \widetilde { \mathcal { G } } _ { q _ { 3 }, q _ { 3 } ^ { \prime } } ^ { \Lambda } ( i \omega ) \gamma _ { 2 } ^ { \Lambda } ( q _ { 4 } ^ { \prime }, q _ { 2 } ^ { \prime } ; q _ { 3 }, q _ { 2 } ),$$
where -γ Λ 1 is the self-energy and γ Λ 2 is the two-particle irreducible vertex. All higher order vertices γ n ≥ 3 have been set to zero. Here ˜ G Λ is defined as
## b. Approximations
$$\widetilde { \mathcal { G } } ^ { \Lambda } = \left [ \mathcal { G } _ { 0 } ^ { - 1 } + \gamma _ { 1 } ^ { \Lambda } \right ] ^ { - 1 } = \frac { 1 } { i \omega - H _ { 0 } + \gamma _ { 1 } ^ { \Lambda } }, \quad ( 1 1 ) \quad \text{denc} \\. \quad \text{$\mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathioth \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \mathfrak{ } \text{strude}$$
where H 0 is the (known) Hamiltonian of the noninteracting system. The quantum numbers q i encode the spin and spatial degrees of freedom q ≡ ( σ, j ). The flow of γ Λ 2 was split into three contributions called the particleparticle channel ( P ), and the exchange ( X ) and direct ( D ) contributions to the particle-hole channel, respectively. This will allow us to simplify the flow equations later.
For a derivation of Eqs. (6-10) see for instance [18, 35, 38].
## a. Initial condition
For the numerical treatment we cannot set the initial value of the flow parameter Λ init to infinity, but it is sufficient that it is much larger than all relevant energy scales. We have the following initial condition at Λ init [18, 35, 38],
$$\gamma _ { 2 } ^ { \Lambda _ { \text{init} } } ( q _ { 1 }, q _ { 2 }, q _ { 3 }, q _ { 4 } ) = v _ { q _ { 1 }, q _ { 2 }, q _ { 3 }, q _ { 4 } }, \quad \quad ( 1 2 ) \quad t r i e s$$
$$\nu _ { 1 } ^ { \Lambda _ { \text{init} } } ( q _ { 1 }, q _ { 2 } ) = - \frac { 1 } { 2 } \sum _ { q } v _ { q _ { 1 }, q, q _ { 2 }, q }, \quad ( 1 3 ) \quad \text{of} \, \nu _ { \text{fee} }$$
where the vertex v q 1 ,q 2 ,q 3 ,q 4 is site diagonal and at site j ≡ j 1 = j 2 = j 3 = j 4 is given by
$$v _ { q _ { 1 }, q _ { 2 }, q _ { 3 }, q _ { 4 } } = U _ { j } \delta _ { \sigma _ { 1 } \bar { \sigma } _ { 2 } } \left ( \delta _ { \sigma _ { 1 } \sigma _ { 3 } } \delta _ { \sigma _ { 2 } \sigma _ { 4 } } - \delta _ { \sigma _ { 1 } \sigma _ { 4 } } \delta _ { \sigma _ { 2 } \sigma _ { 3 } } \right ). \ \ ( 1 4 ) \quad \text{higher} \quad \text{too } \text{$\lambda$}$$
This means that the spins q 1 and q 2 , as well as the spins q 3 and q 4 must be opposite. This leaves two possibilities: σ 1 = σ 3 = ¯ σ 2 = ¯ σ 4 that has positive sign, and σ 1 = σ 4 = ¯ σ 2 = ¯ σ 3 that has negative sign. Inserting this into the initial condition for γ Λ 1 yields
$$\gamma _ { 1 } ^ { \Lambda _ { \text{init} } } ( q ^ { \prime }, q ) = - ( U _ { j } / 2 ) \delta _ { \sigma ^ { \prime } \sigma }. \quad \quad ( 1 5 )$$
We use the following approximations, see [18] and references thereof. Firstly, we neglect the frequency dependence of γ Λ 2 . This is called the static approximation and is known to give good results at T = 0 [35]. Given the structure of the flow equation for γ Λ 2 above, it is natural to divide the flowing vertex into four parts as follows:
$$\gamma _ { 2 } ^ { \Lambda } = v + \gamma _ { p } ^ { \Lambda } + \gamma _ { x } ^ { \Lambda } + \gamma _ { d } ^ { \Lambda }.$$
Here v is shorthand for the bare vertex, and the flows of γ Λ p , γ Λ x and γ Λ d were given above. The initial value for γ Λ 2 is simply the bare vertex. If we insert the bare vertex into the flow equations for the components of γ Λ 2 we observe that some of its symmetries remain preserved in the derivative on the left hand side. For instance, for γ Λ p we see that the first two and the second two site indices must be identical and also that the first and the second pair of spin indices must be opposite, respectively. Similarly, for γ Λ x we see that the first and fourth site index, as well as the the second and third site index must be equal. For γ Λ d the first and the third, as well as the second and the fourth site indices are equal. There is no restriction on the any of the spin indices for either γ Λ x or γ Λ d .
The next iteration would break the remaining symmetries, since all three channels contribute to the derivative of γ Λ 2 and then back-feed into the differential equations for each channel. If instead we choose to only backfeed each channel into its own differential equation, we can preserve the symmetries described above. This immensely simplifies the treatment of the equations. By doing so, we only neglect contributions of order v 3 and higher, which justifies their neglect as long as U j is not too large. Altogether we obtain the following contributions:
$$P _ { j i } ^ { \sigma \bar { \sigma } } \coloneqq \gamma _ { p } ^ { \Lambda } \left ( j \sigma, j \bar { \sigma } ; i \sigma, i \bar { \sigma } \right ) \text{ \quad \ \ } ( 1 7 )$$
$$\bar { P } _ { j i } ^ { \sigma \bar { \sigma } } \coloneqq \gamma _ { p } ^ { \Lambda } \left ( j \sigma, j \bar { \sigma } ; i \bar { \sigma }, i \sigma \right ) \text{ \quad \ \ } ( 1 8 )$$
$$X _ { j i } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } \coloneqq \gamma _ { x } ^ { \Lambda } \left ( j \sigma _ { 1 }, i \sigma _ { 2 } ; i \sigma _ { 3 }, j \sigma _ { 4 } \right ) \quad ( 1 9 )$$
$$D _ { j i } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } \coloneqq \gamma _ { d } ^ { \Lambda } \left ( j \sigma _ { 1 }, i \sigma _ { 2 } ; j \sigma _ { 3 }, i \sigma _ { 4 } \right ) \quad ( 2 0 )$$
Note that some elements of a channel can also satisfy the symmetries of another channel. So for instance the diagonal element P σσ ¯ jj has the same symmetries as the diagonal elements X σσσσ ¯ ¯ jj and D σσσσ ¯ ¯ jj . If we back-feed such elements too we preserve the symmetries in each channel, but obtain a more accurate approximation. Therefore, in each of the three flow equations for the channels of γ Λ 2 we replace γ Λ 2 on the right hand side by the appropriate channels plus the site diagonal contributions of the other channels that obey the same symmetries. The initial conditions for the three channels follow immediately from (16): γ Λ init p = γ Λ init x = γ Λ init d = 0. Of course, for the differential equation for γ Λ 1 we need the full γ Λ 2 which is the sum of all three channels and the bare vertex.
## c. Symmetries
Due to the hermiticity of the Hamiltonian the following relation holds for the Green's function
$$\mathcal { G } ( i \omega ) = \mathcal { G } ^ { \dagger } ( - i \omega ) \Leftrightarrow \mathcal { G } _ { i j } ( i \omega ) = \mathcal { G } _ { j i } ^ { * } ( - i \omega ). \quad ( 2 1 ) \quad \text{ if fol} \\ \text{also} \ \,.$$
We assume that this relation also holds for ˜ G Λ . If γ Λ 1 is hermitian then the assumption is obviously justified. Numerical results indeed confirm that γ Λ 1 is hermitian. We also have the following symmetries of γ 2 :
$$\gamma _ { 2 } ( q _ { 1 }, q _ { 2 }, q _ { 3 }, q _ { 4 } ) & = - \gamma _ { 2 } ( q _ { 2 }, q _ { 1 }, q _ { 3 }, q _ { 4 } ) \quad \ ( 2 2 ) \\ & - \sim \lambda ( a, a, a ) \cdot \quad \ ( 2 3 )$$
$$= \gamma _ { 2 } ( q _ { 2 }, q _ { 1 }, q _ { 4 }, q _ { 3 } ) \quad \ ( 2 4 )$$
$$& = - \gamma _ { 2 } ( q _ { 1 }, q _ { 2 }, q _ { 4 }, q _ { 3 } ) \quad \ ( 2 3 ) \\ & - \sim \omega l a s \ a, \ a, \ a s \, \cdots \quad \ ( 2 4 \, \cdots )$$
This follows directly from the equation defining the twoparticle vertex, see e.g. [38]. Again we assume that these relations hold also for γ Λ 2 and moreover for each of the separate channels. Their consistency with the numerical results will be demonstrated below. Altogether this yields the following symmetry relations for the different channels:
$$\begin{array} { c } P _ { j i } ^ { \sigma \bar { \sigma } } = P _ { j i } ^ { \bar { \sigma } \sigma } = - \bar { P } _ { j i } ^ { \sigma \bar { \sigma } } \\ D _ { j } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } = D _ { j } ^ { \sigma _ { 2 } \sigma _ { 1 } \sigma _ { 4 } \sigma _ { 3 } } = - X _ { j } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 4 } \sigma _ { 3 } } \end{array} \quad ( 2 5 ) \quad \text{Tra} \\ P _ { \ } m \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ } \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ }. \quad \Phi _ { \ } } \end{array}$$
$$X _ { j i } ^ { \mu } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } = X _ { i j } ^ { \mu } \sigma _ { 2 } \sigma _ { 1 } \sigma _ { 4 } \sigma _ { 3 } } = - D _ { j i } ^ { \mu } \sigma _ { 2 } \sigma _ { 4 } \sigma _ { 3 } \quad ( 2 7 ) \quad \text{true}$$
$$& D _ { j i } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } ^ { \nu } = D _ { i j } ^ { \sigma _ { 2 } \sigma _ { 1 } \sigma _ { 4 } \sigma _ { 3 } } = - X _ { j i } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 4 } \sigma _ { 3 } } \quad \begin{pmatrix} \nu \\ ( 2 6 ) \end{pmatrix} \quad \text{$P$ m} \\ & Y ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } = Y ^ { \sigma _ { 2 } \sigma _ { 1 } \sigma _ { 4 } \sigma _ { 3 } } - D ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 4 } \sigma _ { 3 } } \quad \text{$P$ v} \quad \text{$two$}$$
We observe that P ↑↓ = P ↓↑ and hence the spin indices for P will be dropped from now on, leaving only the site index. The alternative configuration ¯ follows completely P from P and does not need to be kept track of separately. Same applies to X and D which completely define each other. We choose to work with D . There are various symmetries of D but there is no restriction on the spin index. This means that there are 2 4 = 16 different submatrices corresponding to 16 different spin configurations of D . We choose to arrange them as follows
$$D ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } = \begin{pmatrix} \frac { \uparrow \uparrow \uparrow \uparrow | \uparrow \uparrow \downarrow | \uparrow \uparrow | \uparrow | \uparrow | \downarrow | \uparrow | } { \uparrow \uparrow | \uparrow | \downarrow | \uparrow | \uparrow | \downarrow | } \\ \frac { \uparrow \uparrow \uparrow | \downarrow | \uparrow | \downarrow | \uparrow | \downarrow | \uparrow | \downarrow | \uparrow | \downarrow | } { \downarrow | \uparrow | \uparrow | \downarrow | \downarrow | \downarrow | } \end{pmatrix} \quad ( 2 8 )$$
Note that the first and third spin index are fixed along a row and correspondingly the second and fourth index are fixed along one column. This form of the matrix will prove convenient later. From the symmetries of D it follows that this matrix is symmetric. Numerically we also confirm the following relations between the different blocks, schematically
| A s | B | B ∗ | C |
|-----------|-------|-----------|-----|
| B T | D s | E h | F |
| ( B ∗ ) T | E T h | D ∗ s | F ∗ |
| C T | F T | ( F ∗ ) T | G s |
where identical symbols denote equal blocks and symmetric (hermitian) submatrices are labeled by the subscript s ( h ). There are only seven different blocks in total. Numerically we also show that the corner submatrices A s , G s , C and C T are real. The other submatrices are complex in general. For a hermitian γ Λ 1 , the first flow equation implies that γ Λ 2 ( q ′ 2 , q ′ 1 ; q 2 , q 1 ) = γ Λ ∗ 2 ( q 2 , q 1 ; q ′ 2 , q ′ 1 ). Translated to the separate channels this confirms that P must indeed be hermitian, since P ij = P ∗ ji , as well as all the remaining relations between the different submatrices of D .
Restricting γ Λ p according to the symmetries of the P -channel we obtain the following simplified equation for the derivative of P :
$$\text{derive of $P$} \colon \\ \frac { d } { d \Lambda } P _ { j i } = \frac { d } { d \Lambda } \gamma _ { p } ^ { \Lambda } ( j \sigma, j \sigma ; i \sigma ) & & ( 2 9 ) \\ = \frac { 1 } { 2 \pi } \sum _ { \omega = \pm \Lambda } \sum _ { k, l } \frac { 1 } { 2 } \cdot \left [ \gamma _ { l } ^ { \Lambda } ( j \sigma, j \sigma ; k \sigma, k \bar { \sigma } ) \widetilde { \mathcal { G } } _ { k l } ^ { \Lambda \sigma \sigma } ( i \omega ) \widetilde { \mathcal { G } } _ { k l } ^ { \Lambda \bar { \sigma } \sigma } ( - i \omega ) \gamma _ { l } ^ { \Lambda } ( l \sigma, l \bar { \sigma } ; i \sigma ) \\ + \gamma _ { l } ^ { \Lambda } ( j \sigma, j \bar { \sigma } ; k \bar { \sigma }, k \sigma ) \widetilde { \mathcal { G } } _ { k l } ^ { \Lambda \bar { \sigma } \sigma } ( i \omega ) \widetilde { \mathcal { G } } _ { k l } ^ { \Lambda \bar { \sigma } \sigma } ( - i \omega ) \gamma _ { l } ^ { \Lambda } ( l \bar { \sigma }, l \bar { \sigma } ; i \sigma ) \\ + \gamma _ { l } ^ { \Lambda } ( j \sigma, j \bar { \sigma } ; k \bar { \sigma }, k \sigma ) \widetilde { \mathcal { G } } _ { k l } ^ { \Lambda \bar { \sigma } \sigma } ( i \omega ) \widetilde { \mathcal { G } } _ { k l } ^ { \Lambda \bar { \sigma } \sigma } ( - i \omega ) \gamma _ { l } ^ { \Lambda } ( l \bar { \sigma } ; i \sigma, i \bar { \sigma } ) \right ] & & ( 3 0 ) \\ \text{Note that the first two terms and the last two terms in the sum are equivalent after summation over $\omega$, due to the $
$. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $. $........................................................................................................................................................................................................$$
Note that the first two terms and the last two terms in the sum are equivalent after summation over ω , due to the symmetry relations (23) and (24). We can thus keep one of the terms respectively and cancel the factor of 1 / 2. With the definitions
$$\Pi _ { k l } ^ { p \Lambda ( 1 ) } = \frac { 1 } { 2 \pi } \sum _ { \omega = \pm \Lambda } \widetilde { \mathcal { G } } _ { k l } ^ { \Lambda \partial \sigma } ( i \omega ) \widetilde { \mathcal { G } } _ { k l } ^ { \Lambda \sigma \sigma } ( - i \omega )$$
$$\Pi _ { k l } ^ { p \Lambda ( 2 ) } = \frac { 1 } { 2 \pi } \sum _ { \omega = \pm \Lambda } ^ { - } \widetilde { \mathcal { G } } _ { k l } ^ { \Lambda \bar { \sigma } \sigma } ( i \omega ) \widetilde { \mathcal { G } } _ { k l } ^ { \Lambda \sigma \bar { \sigma } } ( - i \omega )$$
the flow equation can be written more succinctly as
$$\hat { \frac { d } { d \Lambda } } P _ { j i } & = \sum _ { k l } \left [ \gamma _ { 2 } ^ { \Lambda } ( j \sigma, j \bar { \sigma } ; k \sigma, k \bar { \sigma } ) \Pi _ { k l } ^ { p \Lambda ( 1 ) } \gamma _ { 2 } ^ { \Lambda } ( \sigma, l \bar { \sigma } ; i \bar { \sigma } ) + \gamma _ { 2 } ^ { \Lambda } ( j \sigma, j \bar { \sigma } ; k \sigma, k \bar { \sigma } ) \Pi _ { k l } ^ { p \Lambda ( 2 ) } \gamma _ { 2 } ^ { \Lambda } ( l \bar { \sigma }, l \bar { \sigma } ; i \bar { \sigma } ) \right ] } \\ & = \sum _ { k l } \gamma _ { 2 } ^ { \Lambda } ( j \sigma, j \bar { \sigma } ; k \sigma, k \bar { \sigma } ) \left [ \Pi _ { k l } ^ { p \Lambda ( 1 ) } - \Pi _ { k l } ^ { p \Lambda ( 2 ) } \right ] \gamma _ { 2 } ^ { \Lambda } ( l \bar { \sigma } ; i \bar { \sigma } ), \\ \hat { \, } \, \quad \, \, \,$$
where in the last step we used symmetry relation (23). If we now define
$$\Pi ^ { p \Lambda } _ { k l } \equiv \Pi ^ { p \Lambda ( 1 ) } _ { k l } - \Pi ^ { p \Lambda ( 2 ) } _ { k l } = \frac { 1 } { 2 \pi } \sum _ { \omega = \pm \Lambda } \left [ \widetilde { \mathcal { G } } ^ { \Lambda \partial \sigma } _ { k l } ( i \omega ) \widetilde { \mathcal { G } } ^ { \Lambda \sigma \sigma } _ { k l } ( - i \omega ) - \widetilde { \mathcal { G } } ^ { \Lambda \partial \sigma } _ { k l } ( i \omega ) \widetilde { \mathcal { G } } ^ { \Lambda \sigma \sigma \sigma } _ { k l } ( - i \omega ) \right ]$$
we arrive at
$$\frac { \text{d} } { \text{d} \Lambda } P _ { j i } = \tilde { P } _ { j k } \Pi _ { k l } ^ { p \Lambda } \tilde { P } _ { l i },$$
where ˜ P equals P plus the diagonal contributions from the other channels which have the same symmetries as P . Explicitly we get
$$\tilde { P } _ { j k } = P _ { j k } + \delta _ { j k } \left ( X _ { j j } ^ { \sigma \bar { \partial } \sigma \bar { \partial } } + D _ { j j } ^ { \sigma \bar { \partial } \sigma \bar { \partial } } + U _ { j } \right ) = P _ { j k } + \delta _ { j k } \left ( - D _ { j j } ^ { \sigma \bar { \partial } \bar { \partial } \sigma } + D _ { j j } ^ { \sigma \bar { \partial } \sigma \bar { \partial } } + U _ { j } \right )$$
Note also that the matrix Π p Λ kl is hermitian, due to the symmetry (21) of the Green's function.
## e. Flow equation for the D-channel
Restricting γ Λ d according to the symmetries of the D -channel we obtain the following simplified equation for the derivative of D :
$$\frac { d } { d \Lambda } D _ { j i } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } & = \frac { d } { d \Lambda } \gamma _ { d } ^ { \Lambda } ( j \sigma _ { 1 }, i \sigma _ { 2 } ; j \sigma _ { 3 }, i \sigma _ { 4 } ) \\ & = - \frac { 1 } { 2 \pi } \sum _ { \omega = \pm \Lambda } \sum _ { k l } \sum _ { \sigma, \sigma ^ { \prime }, \sigma ^ { \prime \prime }, \sigma ^ { \prime \prime } } \gamma _ { 2 } ^ { \Lambda } ( j \sigma _ { 1 }, k \sigma ; j \sigma _ { 3 }, k \sigma ^ { \prime } ) \tilde { \mathcal { G } } _ { k l } ^ { \Lambda \sigma ^ { \prime } \sigma ^ { \prime \prime } } ( i \omega ) \tilde { \mathcal { G } } _ { l k } ^ { \Lambda ^ { \prime \prime \prime } \sigma ^ { \prime } } ( i \omega ) \gamma _ { 2 } ^ { \Lambda } ( l \sigma ^ { \prime \prime }, i \sigma _ { 2 } ; l \sigma ^ { \prime \prime \prime }, i \sigma _ { 4 } )$$
Observe that the summation goes over the second and fourth index of the first γ Λ 2 matrix and over the first and third index of the second γ Λ 2 matrix, while the other indices remain fixed. If we want to recast this expression as a matrix multiplication this indeed implies that the first and third spin index should be fixed along a row and the second and fourth index along one column. This justifies the matrix scheme (28). If we arrange the spin configurations according to this scheme we obtain the matrix equation
$$\frac { \text{d} } { \text{d} \Lambda } D _ { j i } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } = \sum _ { k l } \sum _ { \sigma, \sigma ^ { \prime }, \sigma ^ { \prime \prime \prime } } \gamma _ { 2 } ^ { \Lambda } ( j \sigma _ { 1 }, k \sigma ; j \sigma _ { 3 }, k \sigma ^ { \prime } ) \Pi _ { k l } ^ { d \Lambda \sigma \sigma ^ { \prime \prime } \sigma ^ { \prime } \sigma ^ { \prime \prime \prime } } \gamma _ { 2 } ^ { \Lambda } ( l \sigma ^ { \prime \prime }, i \sigma _ { 2 } ; l \sigma ^ { \prime \prime \prime }, i \sigma _ { 4 } )$$
where
$$\Pi _ { k l } ^ { d \Lambda \sigma \sigma ^ { \prime \prime } \sigma ^ { \prime } \sigma ^ { \prime \prime \prime } } \equiv - \frac { 1 } { 2 \pi } \sum _ { \omega = \pm \Lambda } \widetilde { \mathcal { G } } _ { k l } ^ { \Lambda \sigma ^ { \prime } \sigma ^ { \prime \prime } } ( i \omega ) \widetilde { \mathcal { G } } _ { l k } ^ { \Lambda \sigma ^ { \prime \prime \prime } \sigma } ( i \omega ).$$
Note that the order of the spin indices on Π is not the same as on the Green's functions. The symmetries from (28) remain valid. With our approximation we get
$$\frac { \text{d} } { \text{d} \Lambda } D _ { j i } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } = \tilde { D } _ { j k } ^ { \sigma _ { 1 } \sigma _ { 3 } \sigma ^ { \prime } } \Pi _ { k l } ^ { d \sigma ^ { \prime \prime } \sigma ^ { \prime } \sigma ^ { \prime \prime \prime } } \tilde { D } _ { l i } ^ { \sigma ^ { \prime \prime } \sigma _ { 2 } \sigma ^ { \prime \prime \prime } \sigma _ { 4 } }$$
where ˜ D equals D plus the diagonal contributions from the other channels which have the same symmetries as D . Explicitly we get
$$\tilde { D } _ { j k } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } & = D _ { j k } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } + \delta _ { j k } \left ( X _ { j j } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } + ( P _ { j j } + U _ { j } ) \delta _ { \sigma _ { 1 } \delta _ { 2 } } ( \delta _ { \sigma _ { 1 } \sigma _ { 3 } \sigma _ { 2 } \sigma _ { 4 } } - \delta _ { \sigma _ { 1 } \sigma _ { 4 } } \delta _ { \sigma _ { 2 } \sigma _ { 3 } } ) \right ) \\ & \quad - \ r v ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } + \ s + \ r v ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 4 } \sigma _ { 3 } } + \ r n + \ r r \, \subset \ r s - \ s - \ s - \ \ll$$
$$\begin{array} { l l } \cdot & \\ & = D _ { j k } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 3 } \sigma _ { 4 } } + \delta _ { j k } \left ( \stackrel { \cdot \cdot } { - D _ { j j } ^ { \sigma _ { 1 } \sigma _ { 2 } \sigma _ { 4 } \sigma _ { 3 } } } + ( P _ { j j } + U _ { j } ) \delta _ { \sigma _ { 1 } \bar { \sigma } _ { 2 } } ( \delta _ { \sigma _ { 1 } \sigma _ { 3 } } \delta _ { \sigma _ { 2 } \sigma _ { 4 } } - \delta _ { \sigma _ { 1 } \sigma _ { 4 } } \delta _ { \sigma _ { 2 } \sigma _ { 3 } } ) \right ) & \\ \end{array}$$
Just like D itself, the matrix Π d Λ σσ ′′ σ σ ′ ′′′ kl is symmetric, however in general not real. The structure of Π d Λ σσ ′′ σ σ ′ ′′′ kl in terms of its submatrices is the same as for D .
f. Flow equation for γ 1
For the self-energy equation
$$\frac { d } { d \Lambda } \gamma _ { 1 } ^ { \Lambda } ( k ^ { \prime } \sigma ^ { \prime }, k \sigma ) = \frac { 1 } { 2 \pi } \sum _ { \omega = \pm \Lambda } \sum _ { k _ { 1 }, k _ { 2 }, \sigma _ { 1 }, \sigma _ { 2 } } \widetilde { \mathcal { G } } _ { k _ { 2 } k _ { 1 } } ^ { \Lambda \sigma _ { 2 } \sigma _ { 1 } } ( i \omega ) \gamma _ { 2 } ^ { \Lambda } ( k _ { 1 } \sigma _ { 1 }, k ^ { \prime } \sigma ^ { \prime } ; k _ { 2 } \sigma _ { 2 }, k \sigma )$$
we need the full γ Λ 2 = v + γ Λ p + γ Λ x + γ Λ d . We abbreviate
$$\mathcal { S } _ { k _ { 2 } k _ { 1 } } ^ { \sigma _ { 2 } \sigma _ { 1 } } = \frac { 1 } { 2 \pi } \sum _ { \omega = \pm \Lambda } \widetilde { \mathcal { G } } _ { k _ { 2 } k _ { 1 } } ^ { \Lambda \sigma _ { 2 } \sigma _ { 1 } } ( i \omega ).$$
Taking into account the symmetry of each channel we obtain
$$\frac { \cdot } { d \Lambda } \gamma _ { 1 } ^ { \Lambda } ( k ^ { \prime } \sigma ^ { \prime }, k \sigma ) & = \delta _ { \sigma \sigma ^ { \prime } } \mathcal { S } _ { k k ^ { \prime } } \bar { \delta \sigma } ^ { \bar { \delta \sigma } } ( P _ { k ^ { \prime } k } + \delta _ { k k ^ { \prime } } U _ { k } ) - \delta _ { \sigma \bar { \sigma } ^ { \prime } } \mathcal { S } _ { k k ^ { \prime } } \bar { \delta \sigma } ^ { \sigma } ( P _ { k ^ { \prime } k } + \delta _ { k k ^ { \prime } } U _ { k } ) \\ & \quad - \sum _ { \sigma _ { 1 } \sigma _ { 2 } } \mathcal { S } _ { k ^ { \prime } k } ^ { \sigma _ { 2 } \sigma _ { 1 } } D _ { k k ^ { \prime } } ^ { \sigma _ { 1 } \sigma ^ { \prime } \sigma \sigma _ { 2 } } + \delta _ { k k ^ { \prime } } \sum _ { l, \sigma _ { 1 }, \sigma _ { 2 } } \mathcal { S } _ { l l } ^ { \sigma _ { 2 } \sigma _ { 1 } } D _ { l k } ^ { \sigma _ { 1 } \sigma ^ { \prime } \sigma _ { 2 } \sigma }.$$
The first line accounts for the bare vertex and the P/P ¯ -channel, while the second line contains the contribution from the X -channel and then the D -channel. Note that the D -channel only influences the diagonal elements of γ Λ 1 , due to its symmetry.
|
10.1103/PhysRevLett.113.266402
|
[
"Olga Goulko",
"Florian Bauer",
"Jan Heyder",
"Jan von Delft"
] |
2014-08-04T17:04:10+00:00
|
2015-01-05T21:26:46+00:00
|
[
"cond-mat.mes-hall",
"cond-mat.str-el"
] |
Effect of spin-orbit interactions on the 0.7 anomaly in quantum point contacts
|
We study how the conductance of a quantum point contact is affected by
spin-orbit interactions, for systems at zero temperature both with and without
electron-electron interactions. In the presence of spin-orbit coupling, tuning
the strength and direction of an external magnetic field can change the
dispersion relation and hence the local density of states in the point contact
region. This modifies the effect of electron-electron interactions, implying
striking changes in the shape of the 0.7-anomaly and introducing additional
distinctive features in the first conductance step.
|
1408.0747v1
|
## Fermilab Program and Plans
## Dmitri Denisov
Fermi National Accelerator Laboratory,
PO Box 500, Batavia IL 60510, USA
E-mail :
[email protected]
ABSTRACT: This article is a short summary of the talk presented at 2014 Instrumentation Conference in Novosibirsk about Fermilab's experimental program and future plans. It includes brief description of the P5 long term planning progressing in US as well as discussion of the future accelerators considered at Fermilab.
KEYWORDS: Top quark; Neutrinos; Accelerators.
## Contents
## 1. Planning Long Term Scientific Program in US
- 1.1 Big Qiestions for Particle Physics
## 2. Fermilab's Scientific Program
- 2.1 Tevatron
- 2.2 Fermilab Accelerator Complex
- 2.3 Nuetrino Program
- 2.4 Muon Experiments
- 2.5 Colliders
- 2.6 Dark Matter and Dark Energy Experiments
## 3. Concluding Remarks
## 1. Planning Long Term Scientific Program in US
In US high energy physics process of developing long term scientific program consists of the following five main steps:
- Groups of scientists develop proposals for future projects/experiments.
- 'Snowmass' community wide process discusses proposals, evaluates physics reach and costs and summarizes outcome in a written form.
- Organized by Division of Particles and Fields (DPF) professional organization, not Laboratories or NSF (National Science Foundation) or DOE (Department of Energy).
- P5 committee (Particle Physics Projects Prioritization Panel) is formed consisting of ~25 scientists representing all areas of particle physics.
- The committee recommends priorities for funding based on available funds and expected cost of the projects.
- HEPAP (High Energy Physics Advisory Panel) appointed by DOE reviews the P5 proposal and recommends it to be considered by DOE/NSF.
- DOE/NSF fund recommended projects based on the available funds.
The process described is initiated about every five years and most recent was initiated in 2013. The DPF charge for 'Snowmass 2013' reads: 'To develop the community's long term physics aspirations. Its narrative will communicate the opportunities for discovery in high energy physics to the broader scientific community and to the Government'.
Snowmass process was organized around so called "frontiers": energy, intensity, cosmic, instrumentation (which is the subject of Novosibirsk conference), facilities (mainly new
accelerators), education and outreach, and theory. Time scale for the proposals to be discussed is ~10 years, taking into account ~20 years time span. The process continued for about a year and culminated in ~10 days community meeting at the University of Minnesota late July 2013 where in depth discussion of all proposals, their scientific merit, facilities required as well as cost have been presented. The outcome of the Snowmass process is a large number of documents which are summarized on the Snowmass Web page [1].
## 1.1 Big Questions for Particle Physics
One of the main outcomes of the Snowmass process was the development of "big questions" for particle physics. They are
- · How do we understand the Higgs boson? Why does it condense and acquire a vacuum value throughout the universe? Is there one Higgs particle or many? Is the Higgs particle elementary or composite?
- · What principle determines the masses and mixings of quarks and leptons? Why is the mixing pattern apparently different for quarks and leptons? Why is the CKM CP phase nonzero? Is there CP violation in the lepton sector?
- · Why are neutrinos so light compared to other matter particles? Are neutrinos their own antiparticles? Are their small masses connected to the presence of a very high mass scale? Are there new interactions invisible except through their role in neutrino physics?
- · What mechanism produced the excess of matter over antimatter that we see in the universe? Why are the interactions of particles and antiparticles not exactly mirror opposites?
- · Dark matter is the dominant component of mass in the universe. What is the dark matter made of? Is it composed of one type of new particle or several? Are the dark matter particles connected to the particles of the Standard Model, or are they part of an entirely new dark sector of particles?
- · What is dark energy? Is it a static energy per unit volume of the vacuum, or is it dynamical and evolving with the universe? What principle determines its value?
- · What did the universe look like in its earliest moments, and how did it evolve to contain the structures we observe today?
Definition of what are the main topics for experimental and theoretical studies is an important first step to select most interesting and scientifically important projects for the next decade.
## 2. Fermilab's Scientific Program
## 2.1 Tevatron
The Tevatron accelerator complex presented in Fig. 1 was the center of the Laboratory scientific activities for almost 25 years. Two general-purpose collider detectors CDF and DZero published over 1000 papers cementing the Standard Models. Discovery of the top quark in 1995 was fundamental discovery as well as evidence for the Higgs boson production and decay to a pair of b-quarks in 2012. On the first day of Novosibirsk conference CDF and DZero jointly announced another discovery: observation of t-channel single top quark production with 6.3
sigma significance. More results, especially in very high precision measurements, are expected from 12 fb -1 data set of proton-antiproton collisions at 2 TeV delivered by the Tevatron to both collider experiments.
Fig. 1 Tevatron accelerator complex and two detectors CDF and DZero.
## 2.2 Fermilab Accelerator Complex
With Tevatron shutdown late 2011 the Fermilab accelerator complex is now concentrating on delivering high intensity proton beams at up to 150 GeV to a wide range of programs including neutrino experiments, high intensity muon beam experiments as well as test beams. The scheme of the current Fermilab's accelerator complex is presented I n Fig. 2.
Fig. 2. Current Fermilab's accelerator complex.
In the near term Fermilab is planning to construct a multi-MW proton linear accelerator with flexible beam structure based on SCRF technology. Power will reach 1 MW at 1 GeV and more at 3-8 GeV with over 2 MW of power to neutrino program at 120 GeV. Such upgrade could serve multiple experiments over a broad energy range and serve as a platform for future neutrino and muon facilities (including muon collider).
## 2.3 Neutrino Program
Neutrinos remian one of the most puzzling particles of the Standard Model. Studies of neutrino properties, including mixing matrix parameters, could shed light on many interesting topics, including matter-antimatter asymmetry of the Universe.
The MINOS experiment is a long-baseline neutrino oscillation experiment operating at Fermilab with two detectors, near and far in order to improve beam flux calculations and reduce neutrino interaction uncertainties. The L/E ratio for the MINOS experiment is ~500 km/GeV. Figure 3 presents one of the recent MINOS results which is indicative that the experiment is starting to approach sensitivity to the mass hierarchy and CP violation in neutrino mixing. MINOS was among experiments which rejected claims that neutrinos travel faster than the speed of light.
Fig. 3. Recent MINOS results are becoming sensitive to mass hierarchy and CP violation.
The MINERvA experiment is located just in front of MINOS in the neutrino beam with energies of a few GeV and is studying neutrino interactions in unprecedented detail on a variety of different nuclei - He, C, CH2, H2O, Fe, Pb. This experiment provides important information for all neutrino based experiments.
MicroBooNe is a new neutrino experiment with 170 ton of LAr based on time projection technology with a few meters of electrons drifts in LAr. This is a new technology development which could potentially improve neutrino detection substantially. This experiment is expected to start taking data in late 2014.
NOvA is yet another brand new neutrino "off axis" experiment to start data collection in 2014. The main idea of this experiment is to use an "off axis" beam of neutrino which has rather small energy spread with a peak at ~1 GeV. The detector total weight is 15 kton and it is located on surface close to the MINOS detector and uses liquid scintillator and WLS fibers with APD as readout.
Fermilab's future neutrino program flagship is so called Long Baseline Neutrino Experiment or LBNE. This experiment is under development now with mass of LAr detector of ~30 kton located deep underground in South Dakota mine with distance of 1300 km from Fermilab. This experiment has over 5 sigma sensitivity to resolve neutrino mass hierarchy for any value of CP violation parameter and an excellent detector for supernova and proton decay with sensitivity better than 10 35 years. Design and construction of such detector will take about a
decade with data taking starting by late 2020's. Figure 4 presents overall scheme of the LBNE experiment.
Fig. 4. LBNE experiment.
## 2.4 Muon Experiments
High intensity proton beam from Fermilab's 150 GeV accelerator of up to 2 10 . 17 protons per hour provides an opportunity to create very high intensity muon beams. Two experiments are planned to use such high intensity muon beams (see Fig. 2). One is g-2 experiment which is based on the BNL E-821 experiment and uses a superconducting coil moved from BNL to Fermilab. This experiment is expected to start in 2017 and have sensitivity ~4 times better than the BNL experiment. Such sensitivity is expected to address ~3 sigma difference between the g2 experimental measurement and theoretical predictions which currently exists.
A second experiment is the Mu2e experiment which is looking for forbidden in the Standard Model decay of muon into electron after muon capture on a nuclear target. This experiment is expected to improve sensitivity to such decay by ~4 orders of magnitude (in
comparison with the current limit) to ~10 -17 . The experiment uses an intense muon beam as well as interesting idea of absorbing a large number of muons in an aluminum target.
## 2.5 Colliders
Fermilab is actively involved in the LHC program: CMS experiment and LHC upgrades. LHC experiments are critical to study all properties of the Higgs boson: mass, width, spin, couplings and others. LHC is the key machine for the next ~10 years for searches for new particles and interactions and Fermilab will be an active part of this international program.
Japan expressed strong interest in hosting the International Linear Collider (ILC) - 500 GeV e e linear accelerator which will be a factory for precision studies of all known Standard + -Model particles as well as discovery machine for new particles and interactions. US accelerator community is capable and interested to contribute in many areas and there is strong interest in the development of detectors for the experiments at the ILC. ILC design is technically ready to go and incorporates US contributions to machine physics and technology including SCRF, high power targetry (e source), beam delivery, damping rings, and beam dynamics. +
Muons do not have high synchrotron radiation making a circular accelerator viable for multi TeV energies. The main challenge is that muons are unstable with life-time of 2 µ s. The accelerator challenge is to make a large number of muons quickly and then 'cool' them to focus into small diameter beams to collide. The 2x2 TeV muon collider at Fermilab layout is presented in Fig. 5. Another issue is decays and irradiation by electrons from muon decays and even neutrinos irradiation! This is an actively developing program in US with up to 10 TeV muon collider to fit on Fermilab's site.
Going to ever higher energies and smaller distances is the quest of particle physics for over 60 years. For a proton-proton collider next step is an about 100 TeV collider and active studies of options are performed at Fermilab. Figure 6 illustrates a design located in the vicinity of Fermilab with the total length of the tunnel 233 km and energy of 175 TeV. Such collider will be crucial for future progress especially if full energy LHC is not enough to reach new physics.
Fig. 5. Muon collider layout at Fermilab.
Fig. 6 175 TeV p-p collider design.
2.6 Dark Matter and Dark Energy
Fermilab is actively involved in studies of two astrophysics puzzles: dark matter and dark energy. In the dark matter searches key expertise of Fermilab is particle detectors. The Cold Dark Matter Search experiment or CDMS located in Sudan mine is already producing world best limits on mass and cross sections of WIMP particles. 200 kg "Super CDMS" upgrade is planned and will be located at SNOLAB. Sensitivity of Super CDMS will be below 10 -44 cm 2 for WIMP-nucleon cross sections with a mass of ~100 GeV.
In studies of dark energy Fermilab provides a 570 megapixels silicon camera based on experience with development of silicon vertex detectors for Tevatron and LHC. The Dark Energy Survey (DES) telescope is located in Chile and started operation in September 2013. DES major scientific areas are: studies of dark matter, dark energy, supernova, solar system survey and many others.
## 3. Concluding Remarks
Particle physics in US is undergoing changes after shutdowns of the SLAC B factory and Fermilab Tevatron over last 5 years. Snowmass process created well documented list of exciting proposals for new accelerators and experiments. The most probable accelerator projects for this decade are new Fermilab linac, ILC and LHC high luminosity upgrade. Among large new upgrades/experiments planned at Fermilab are g-2, Mu2e, LBNE, ATLAS/CMS upgrades, dark matter searches. P5 process is progressing to set priorities for the coming ~10 years. There is wealth of experience at Fermilab in both accelerators and detectors and interesting program ahead.
## Acknowledgments
I would like to thank Organizers of 2014 Novosibirsk Instrumentation conference for the invitation to give this review talk and their hospitality during my stay at Novosibirsk as well as to all my Fermilab colleagues who carry out interesting program of particle physics experiments and for their help with preparations of this presentation.
## References
- [1] Snowmass 2013 Web page http://www.snowmass2013.org/tiki-index.php
|
10.1088/1748-0221/9/07/C07021
|
[
"Dmitri Denisov"
] |
2014-08-04T17:19:40+00:00
|
2014-08-04T17:19:40+00:00
|
[
"physics.ins-det",
"hep-ex",
"physics.acc-ph"
] |
Fermilab Program and Plans
|
This article is a short summary of the talk presented at 2014 Instrumentation
Conference in Novosibirsk about Fermilab's experimental program and future
plans. It includes brief description of the P5 long term planning progressing
in US as well as discussion of the future accelerators considered at Fermilab.
|
1408.0748v1
|
## Elastocapillary coalescence of plates and pillars
Z. WEI , T.M. SCHNEIDER 1 1 4 , , J. KIM , H.-Y. KIM , J. AIZENBERG 5 5 1 3 , and L. MAHADEVAN 1 2 3 , , ∗
1 School of Engineering and Applied Sciences, 2 Department of Physics,
- 3 Kavli Institute for
Nano-Bio Science and Technology, Wyss Institute for Bio-inspired Engineering,
Harvard University, Cambridge, Massachusetts 02138, USA
4 Max Planck Institute for Dynamics and Self-Organization,
Am Fassberg 17, D-37077 Goettingen, Germany
5 Department of Mechanical and Aerospace Engineering,
Seoul National University, Seoul 151-744, Korea
## Abstract
When a fluid-immersed array of lamellae or filaments that is attached to a substrate is dried, evaporation leads to the formation of menisci on the tips of the plates or pillars that bring them together. Similarly, when hair dries it clumps together due to capillary forces induced by the liquid menisci between the flexible hairs. Building on prior experimental observations, we use a combination of theory and computation to understand the nature of this instability and its evolution in both the two-dimensional and three-dimensional setting of the problem. For the case of lamellae, we explicitly derive the interaction torques based on the relevant physical parameters. A Bloch-wave analysis for our periodic mechanical system captures the critical volume of the liquid and the 2-plate-collapse eigenmode at the onset of instability. We study the evolution of clusters and their arrest using numerical simulations to explain the hierarchical cluster formation and characterize the sensitive dependence of the final structures on the initial perturbations. We then generalize our analysis to treat the problem of pillar collapse in 3D, where the fluid domain is completely connected and the interface is a surface with the uniform mean curvature. Our theory and simulations capture the salient features of both previous experimental observations and our own in terms of the key parameters that can be used to control the kinetics of the process.
∗ Electronic address: [email protected]
## I. INTRODUCTION
While assembly of complex macromolecular structures on microscopic lengthscales [1] is driven by van der Waals interactions, dispersive forces and chemical interactions between constituents, on mesoscopic length scales of the order of microns to millimeters in the context of colloids and larger particles, other surface forces such as those due to capillarity play an important role [2-4]. When particles interact with each other via capillary forces, the resulting patterns are a function of the size and shape of the constituents, and any constraints on their movement. Capillary coalescence is a natural consequence of this and occurs when free particles aggregate at an interface [5, 6], and also when extended objects such as filaments and lamellae are brought together by interfacial forces which drive aggregation [7-10]. Sometimes, these systems coarsen indefinitely leading to a single cluster, while at other times elastic deformations eventually arrest the process leading to many finite sized clusters. In a typical experimental setting characterizing the latter, fibers, filaments or lamellae are fully immersed in a liquid which is subsequently evaporated so that capillary forces at the liquid-gas interfaces bring the constituents together. Fibers and lamellae are long and soft objects and can easily bend. Thus a competition between actuating capillarity and resisting elasticity selects the structures formed, which can be used to construct substrates with tunable wetting and adsorption properties [11, 12]. A variety of experimental systems that fall into this category include millimeter-scaled macroscopic brush hairs [8], micrometer-scaled mesoscopic polymeric surface mimicking gecko foot hairs [13], as well as nanometer-scaled carbon nanotube forests [14, 15].
Elastocapillary interaction have been well characterized in two-body systems [9, 16, 17]. However, for the collective behavior of many elastic fibers bundling together, typically static energy minimization arguments have been employed to estimate the expected finally assembled bundle size [8, 18, 19], although recent work [20] uses stability analysis of the unpatterned base state to predict the ordered hierarchical coalescence structure of an array of clamped parallel elastic sheets which are partially immersed in liquid. Complementing these discrete approaches, a phenomenological continuum mean-field approach has been used to model the arrested coarsening [21], while a recent continuum theory based on microscopic physics for the coupled dynamics of drying and coalescence explains the kinetics and refinement seen in experimental studies [22]. Unlike in ergodic systems where the state space is
stochastically sampled due to thermal excitations, the structures in elastocapillary systems are often not selected by energetics alone. Instead, they depend critically on the dynamics of the drying process. This leads to a path dependence caused by the strong coupling of the geometry of the air-liquid interface to the local evaporation when multiple unconnected liquid domains are formed. Additional aspects, such as pinning and contact angle hysteresis, as well as the permanent adhesion of contacts formed in intermediate structures, make purely energetic arguments unable to have any predicative power. Thus, to predict and control the assembled structures in capillarity-driven self-assembly experiments, we need to
(1) follow the dynamics over time as irreversible effects associated with evaporation, contact line motion and adhesion, and (2) account for non-linear effects due to large deviations from
- the background state and go beyond a linearized analysis.
Here we use a combination of theory and computation to understand the instability and the hierarchical evolution of cluster formation for both twoand three-dimensional elastocapillary systems driven by drying. Those systems represent generic situations, allow for a thorough theoretical treatment and can be validated by well controlled experiments. For the two dimensional case, we describe the drying induced collapse of an one-dimensional array of evenly spaced lamellae immersed in the evaporating liquid (Figure 1(a)(b)). For this system all relevant physical forces are considered, allowing us to derive the interaction potentials and forces, analyze the linear stability of the system, and compute the nonlinear dynamics associated with pattern coarsening. We demonstrate that different dynamical paths through the system's state space indeed lead to different final structures which are observed in experiments.
In the three-dimensional case, our theory focuses on explaining experiments associated with the bundling of a regular square grid of fluid-immersed elastic posts anchored to a substrate [7] (Figure 1(c)-(e)). We determine the constant mean curvature surface for the air-liquid interface subjected to the global liquid volume constraint on the multi-connected domain. This allows us to compute the interaction forces and thence the primary unstable mode. Finally we use a numerical method to compute the aggregation dynamics.
In § II we describe the experimental observations for the two-dimensional case, derive a discrete two-plate model for the deformation of the plates driven by capillary forces, and carry out a linear stability analysis of the base state. Wethen study the coalescence dynamics of a collection of plates in the nonlinear regime. In § III we describe the experimental
observations for the three-dimensional case, and introduce a dynamical model that allows us to simulate the morphology of clustered pillars, and compare these results with experiments. In § IV, we conclude with a description of open problems in this rich area.
## II. COLLECTIVE DYNAMICS OF ELASTIC LAMELLAE
## A. Experimental observations
For the two-dimensional case (Figure 2), we consider a one-dimensional array of elastic micro-lamellae with height L , thickness h and uniform spacing D , Young's modulus E and Poisson's ratio ν respectively. Each lamella is assumed to be free at one end and anchored at the other on a substrate [10]. The lamella array is wetted by a liquid of surface tension σ , density ρ and viscosity µ , which is confined between neighboring lamellae, defining a cell. The contact line slips from the tips as the liquid evaporates. When the system is completely immersed in the liquid, the stable configuration is a uniform array of non-interacting vertical lamellae. However, when the liquid evaporates, it is not necessarily locally stable any more: capillary forces associated with the liquid-air menisci between the free ends of the soft lamellae may cause them to deflect laterally and adhere together. In an experimental system with small imperfections, we observe a regular cascade of successive sticking events that leads to a hierarchical bundling pattern: every two neighboring lamellae incline towards each other to form a dimer first, which then collapses into quadrimers (Figure. 1(a)). The process repeats until the bending deformation induced elastic resistance eventually becomes large enough to prevent further coarsening. In the system with large imperfections, irregular bundles can and do arise but the 2-lamella-collapse mode still appears to be dominant right after the instability(Figure. 1(b)). After the liquid dries out, bundles separate if the adhesion in contact is not strong to counterbalance the elastic forces; else they persist. In our experiments, the liquid used is isopropyl alcohol (IPA), and the lamellae and the substrate are made of polydimethylsiloxane (PDMS). Throughout this entire section, we use the following experimental parameters σ = 0 022 . N/m ρ , = 786 Kg/m 3 , µ = 0 0196 . Pa · s , E = 1 5 . Mpa ν , = 0 5, . h = 10 µm L , = 40 µm D , = 10 µm , and gravity g = 9 8 . m/s 2 .
FIG. 1: Dynamics of coalescence of arrays of (a-b) lamellae and (c-e) filaments driven by drying. (a) The lamellae are of thickness h = 10 µm , depth W = 40 µm and height L = 40 µm , and spaced a distance D = 10 µm apart. As the liquid evaporates, a front of dimerization coalescence initiates and propagates from right to left, after which a front of quadrimerization moves through the system in the same direction. After the liquid dries out, the quad-bundles separate and the dimers persist. (b) The lamellae are of depth ∼ 1 mm , and other geometric parameters are the same as those in (a). Due to the imperfections in the system, e.g. the lamella geometry, roughness of the surface and et al., coalescence initiates simultaneously here. The 2-lamella mode still appears to be dominant at the onset of instability, and irregular bundles with a size distribution from 2-5 arise thereafter and persist even after the liquid dries out. (c) The final structures of the two-dimensional array of pillars of diameter after the liquid dries out form four-fold rhombic clusters. (d) Here the tips of the fourfold clusters form squares. (e) A larger domain shows that the four fold clusters are more asymmetric, and the tips form both rhombi and squares. The geometric parameters of the pillars and the material properties are the same in all three images; the pillars have a diameter 300 nm , height 4 5 . µm and spacing 2 µm . The only difference is the size of imperfections in the system.
## B. Mechanics: coupling plate bending to fluid interface shape
In our experiments, the lamellae are short and stiff, and remain almost straight as they are deflected by capillary forces, bending primarily in the neighborhood of the base. Therefore, we can approximate each lamella as a rigid plate and integrate all the bending response into an elastic hinge at the base [12]. This simplifies our analysis relative to the case that must account for the inhomogeneous bending and buckling (APPENDIX A) of the individual lamella [20, 23]. The hinge elastic constant can be approximately derived from the bending response of a short cantilever by a transverse force F at its free end, which is given by
$$k = \frac { d ( F \delta ) } { d \theta } = \left ( \frac { 3 L } { E h ^ { 3 } } + \frac { 6 } { 7 G h L } \right ) ^ { - 1 },$$
where δ = 4 FL 3 (1 -ν 2 ) /Eh 3 + α FL/Gh s is the deflection at the free end, θ is the tilting angle between the straight plate and the horizontal direction (Figure 2(a)), G is the shear modulus, and α s is the shear coefficient. The second term of the right hand side of Eq.(1) is due to shear deformations in the so-called Timoshenko beam theory [24], because the slender beam condition L glyph[greatermuch] h is violated in our experimental setup. We have taken ν = 0 5 and . α s = 10(1 + ν / ) (12 + 11 ν ) as an approximation for the rectangular cross section.
To explicitly derive the torques due to capillarity, we consider a unit cell consisting of two plates, with liquid confined in between, and air outside (Figure 2(a)). The pressure field inside the liquid is nonuniform due to the effects of gravity and flow. However, in our system, gravity can be neglected, because the Bond number Bo = ∆ ρgD /σ 2 ∼ 10 -5 glyph[lessmuch] 1. Comparing the viscous moment due to flow M µ ∼ µL θ/D 5 ˙ 3 (APPENDIX C) with that caused by the surface tension M σ ∼ σL /D 2 , we find that M /M µ σ ∼ 10 -3 glyph[lessmuch] 1 for ˙ θ ∼ O (1), which indicates that pressure effects due to fluid flow can be neglected. Therefore, the air-liquid interface is always a segment of a circle because of the uniform pressure in each cell. For each plate, the moment results from both the line tension at the contact line, and the pressure gradient across the plate caused by the curvature difference of two neighboring menisci. To calculate the moment we need to find the wetting length, the contact angle, and the pressure difference across the air-liquid interface due to its local curvature. We assume the contact angle can take any value equal to or larger than a critical value α , consistent with the fact that the meniscus can be pinned at the tip of the lamella while being concave-up (Figure 2(b1)), flat (Figure 2(b2)), or concave-down (Figure 2(b3)). We
L
Lwr n
2a n air
r n
β n
α
liquid
Vn
θ
n
θ n+1
D
(
a
)
FIG. 2: A unit cell confined by two adjacent lamellae/plates. (a) Each lamella is modeled as a rigid plate elastically hinged at the base. The plate is of height L and thickness h . θ n is the angle of n th plate with respect to the horizontal. The plate spacing is D and the distance between the two free tips is 2 a n . V n is the volume per unit depth of the liquid confined in the cell. The red curve represents the air-liquid interface, an arc of a circle of radius r n , with β n being the half angle subtended by the meniscus arc. α is the critical contact angle at which the meniscus slides down from the tip. L wln +1 is the wetting length on the left side of the ( n +1) th plate and L wrn is the wetting length on the right side of the n th plate. (b) 1-8 are the 8 possible cases of menisci, showing that the menisci can be pinned on both tips, or slip down from one or both tips. 7 and 8 show that when the two free tips are so close that no circular arcs exist to connect them, in which case the arc is replaced by a line. These situations are solely used to prevent simulation failure in rare cases.
assume that once the contact angle reaches the critical value α , it remains constant as the meniscus starts to slide down from the tip (Figure 2(b4)-(b6)). Consequently, the meniscus profile and the resulting moment can be calculated for any given tilting angles θ n , θ n +1 and the liquid volume per unit depth V n . Scaling lengths by L , volumes V n by L 2 and moments by σL leads to the following results for the different cases that we summarize here for our subsequent linear stability analysis (for details, see APPENDIX B).
- 1. The meniscus is pinned on both tips (Figure 2(b1)). The half angle β n subtended by
h
Lwl n+1
the meniscus arc is determined by solving
$$V _ { n } = \frac { 1 } { 2 } d ( \sin \theta _ { n } + \sin \theta _ { n + 1 } ) + \frac { 1 } { 2 } \sin ( \theta _ { n } - \theta _ { n + 1 } ) - a _ { n } ^ { 2 } ( \beta _ { n } \csc ^ { 2 } \beta _ { n } - \cot \beta _ { n } ), \quad ( 2 )$$
for given V n , θ n and θ n +1 , where d = D/L and tip separation is
$$2 a _ { n } = \sqrt { 2 - 2 \cos ( \theta _ { n + 1 } - \theta _ { n } ) + 2 d ( \cos \theta _ { n + 1 } - \cos \theta _ { n } ) + d ^ { 2 } }.$$
β n must satisfy β n ≤ ˆ , where β n ˆ β n is the critical angle at which the meniscus starts to slide down from at least one lamella. β n < 0 when the meniscus concaves down, β n = 0 when the meniscus is flat, and β n > 0 when the meniscus concaves up. The moments on the n th and n +1 th plates are given respectively by
$$M _ { n } = - \frac { 1 } { \underset { \omega } { 2 a _ { n } } } \left [ \sin ( \beta _ { n } + \theta _ { n } - \theta _ { n + 1 } ) + d \sin ( \beta _ { n } + \theta _ { n } ) \right ],$$
$$M _ { n + 1 } = \frac { 1 ^ { ^ { * } } } { 2 a _ { n } } \left [ \sin ( \beta _ { n } + \theta _ { n } - \theta _ { n + 1 } ) - d \sin ( \beta _ { n } - \theta _ { n + 1 } ) \right ],$$
- 2. The meniscus has slipped down from both tips (Figure 2(b3)). The contact angle is fixed at α . When θ n = θ n +1 = θ , the meniscus radius is independent of V n , and r n = d sin θ/ (2 cos α ). l n is determined by solving
$$V _ { n } = \frac { 2 l _ { n } - d \cos \theta } { 2 } d \sin \theta - \left ( \frac { d \sin \theta } { 2 } \right ) ^ { 2 } \tan \alpha - \frac { \pi - 2 \alpha } { 2 } \left ( \frac { d \sin \theta } { 2 \cos \alpha } \right ) ^ { 2 }.$$
The wetting length on the right side of the n th plate and that on the left side of the ( n +1) th plate are given respectively by
$$L _ { w r n } = l _ { n } - \frac { d } { 2 } \sin \theta \tan \alpha,$$
$$L _ { w l _ { n + 1 } } = l _ { n } - \frac { d } { 2 } \sin \theta \tan \alpha - d \cos \theta.$$
The moments on the n th and ( n +1) th plate are given respectively by
$$M _ { n } = - \frac { L _ { w r _ { n } } ^ { 2 } } { 2 r _ { n } } - L _ { w r _ { n } } \sin \alpha,$$
$$M _ { n + 1 } = \frac { L _ { w l _ { n + 1 } } ^ { \stackrel { n } { 2 } } } { 2 r _ { n } } + L _ { w l _ { n + 1 } } \sin \alpha.$$
To get the total moment on a plate, we must add the contributions from adjacent cells. For example, we can obtain the full expression of M n simply by replacing n by n -1 in Eq.
(5) and add up to Eq. (4) for the case when the meniscus is pinned at both tips, which readily yields M n = M θ n ( n -1 , θ n , θ n +1 , V n , V n -1 ).
The dynamics of the n th plate neglecting inertia (APPENDIX C) follows the overdamped first order equation of motion
$$C \frac { \partial \theta _ { n } } { \partial t } + k \left ( \theta _ { n } - \frac { \pi } { 2 } \right ) + \sigma L M _ { n } ( \theta _ { n - 1 }, \theta _ { n }, \theta _ { n + 1 }, V _ { n }, V _ { n - 1 } ) = 0,$$
where C is the damping coefficient, k is defined in Eq. (1) and the dimensionless moment M n is due to capillarity. (We use Eqs. (B3) and (B4), Eqs. (B11) and (B12), Eqs. (B17) and (B18), Eqs. (B23)-(B26), and Eqs. (B29) and (B30) for different situations to obtain the full expressions of M n as explained above. Please see APPENDIX B for more details.) To estimate C , we need to account for both the internal viscosity of the solid and the external viscosity of the fluid, and find that the former dominates, which gives C ≈ τ m k (APPENDIX C). Together with Eq. (11) and the dynamics of drying that will be discussed later, we can now understand the dynamics of the lamella array completely.
## C. Onset of bundling: linear stability of the uniform base state
For the base state with all lamella being vertical and a uniform meniscus associated with constant liquid volume in each cell is decreased, there is a potential for instability as the curvature of the menisci increase.
To understand this, we consider a periodic domain of 2 N plates and the same volume of liquid V confined in each cell, when all plates being vertical ( θ n = π/ , n 2 = 1 2 , , ..., 2 N ) is an equilibrium state. To determine the stability of this state, we study the perturbations in the moment as a function of variations in the angles θ n linearized around the current state,
$$d \overrightarrow { M } = ( \mathbb { K } _ { 1 } + \mathbb { K } _ { 2 } ) d \overrightarrow { \theta },$$
where K 1 and K 2 are the 2 N × 2 N stiffness matrices due to the elasticity of the plate and the geometrical change of menisci respectively. K 1 is a diagonal matrix with all elements being the dimensionless hinge constant k b = (3 σL /Eh 2 3 +6 σ/ Gh 7 ) -1 , and K 2 is a tridiagonal matrix with two additional elements of -k 1 on the upper-right and lower-left corners reflecting the periodic boundary condition. At the base state with θ n = π/ 2, K 2 is expressed
as
$$\mathbb { K } _ { 2 } = \left [ \ddots \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \end{right], \quad \left ( 1 3 \right ) \\ \mathbb { K } _ { 2 } = \left [ \ddots \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon + \infty$$
where k 1 = ∂M /∂θ n n +1 = ∂M n +1 /∂θ n and k 2 = -∂M /∂θ n n = -∂M n +1 /∂θ n +1 are the stiffness of the effective spring connecting two neighboring plates due to capillarity, where M n is defined in Eq. (4) or Eq. (9), and M n +1 is defined in Eq. (5) or Eq. (10). For the case when the meniscus is pinned at both tips,
$$k _ { 1 } & = k _ { 2 } + \sin \beta, \\ k _ { 2 } & = \frac { \cos \beta - d \sin \beta } { d } \left [ 1 + \frac { 1 - d ( \beta \csc ^ { 2 } \beta - \cot \beta ) } { d ^ { 2 } \csc ^ { 2 } \beta ( 1 - \beta \cot \beta ) } \right ] - \frac { \sin \beta + d \cos \beta } { d ^ { 2 } },$$
where β is determined by the volume
$$V = d - ( d / 2 ) ^ { 2 } ( \beta \csc ^ { 2 } \beta - \cot \beta ),$$
which follows from Eq. (2) and Eq. (3) by substituting θ n = θ n +1 = π/ 2. Here we have omitted the subscript n for the translationally invariant base state. Note that k 1 and k 2 in this case are not necessarily equal, because while the wetting length is constant, the contact angles change by different amount on the two neighboring plates as they are deflected except when the meniscus is flat. For the case when the meniscus is no longer at both tips,
$$k _ { 1 } = k _ { 2 } = - \left [ \frac { L _ { w } } { r } \sin \alpha + \frac { L _ { w } ^ { 2 } } { 2 r ^ { 2 } } + \sin ^ { 2 } \alpha \right ] \frac { 2 L _ { w } + d \tan \alpha } { 4 \cos \alpha },$$
with r = d/ (2 cos α ) is the radius of the meniscus, and the wetting length L w is determined by the volume
$$V = L _ { w } d + d ^ { 2 } \tan \left ( \frac { \alpha } { 4 } \right ) - d ^ { 2 } \frac { \pi - 2 \alpha } { 8 \cos ^ { 2 } \alpha },$$
which follows from Eqs. (6)-(8) by substituting θ = π/ 2. k 1 and k 2 in this case are always identical, because the contact angle keeps constant, and the change of wetting length is the same on both plates when they are deflected. From Eqs. (14)-(17), we see that for any given geometric parameter d and critical contact angle α k , i = k i ( V ) ( i = 1 2). The inset of , Figure 3(a) shows that k i switches sign as V decreases, corresponding to the case when the
V / d
FIG. 3: Stability analysis of a vertical array of plates. (a) The curve of 2 k 1 + 2 k 2 following Eqs. (14) and (16) as a function of V given by Eqs. (15) and (17) intersects with the line -k b . The interval between the two intersections indicates the region where the system is unstable as according to Eq. (18), λ π ( ) = 2 k 1 +2 k 2 + k b < 0. For typical experiments, any V/d ∈ (0 507 . , 0 939) . leads to coalesce given infinitesimal perturbations to the system. Inset figure shows the effective dimensionless spring constants k i ( V ) ( i = 1 2); , k i switches signs from positive to negative as V decreases, while the discontinuity corresponds to the meniscus sliding down from the tip. (b) shows the positive bifurcation branch using both simulations of Eq. (11) and the asymptotic result in Eq. (20). When V ≥ V ∗ = 0 9393 . d , φ = 0, and when V < V ∗ , φ ∼ √ V ∗ -V , which is a clear indication of supercritical bifurcation. The inset illustrates the dimer mode. We simulate 2 plates by solving Eq. (11) with periodic boundary conditions, and display them for visualization. The black dots correspond to stable dimers, while the red dots correspond to unstable dimers that will further coalescence.
system becomes unstable. We note that the discontinuity in k i corresponds to the transition from a static meniscus to a dynamic one that slides down from the tip.
To study the instability of the system, we investigate the eigenmodes of the stiffness matrix K = K 1 + K 2 which resembles a discrete Laplacian, with eigenvectors
$$\left [ 1, e ^ { i f }, e ^ { 2 i f }, \dots, e ^ { i ( 2 N - 3 ) f }, e ^ { i ( 2 N - 2 ) f }, e ^ { - i f } \right ],$$
and the corresponding eigenvalues
$$\lambda ( f ) = - 2 k _ { 1 } \cos ( f ) + 2 k _ { 2 } + k _ { b }, \text{ where } f = n \pi / N \text{ and } n = 1, 2, \dots, 2 N. \quad \ ( 1 9 )$$
K must be positive definite to ensure stability, which is equivalent to requiring the smallest eigenvalue positive. When k 1 ≥ 0, the meniscus is pinned at both tips, and the smallest
eigenvalue is λ (2 π ) = -2 k 1 +2 k 2 + k b = k b -2 sin β , which follows from Eqs. (14) and (19). In our experiments, k b ≈ 13 5 . > 2 sin β , so that the array of vertical plates is stable. When k 1 < 0, the smallest eigenvalue is λ π ( ) = 2 k 1 + 2 k 2 + k b , so that stability is controlled by the competition between elasticity and capillarity. As k i = k i ( V ) ( i = 1 2), the condition , λ π ( ) < 0 sets the range of V in which the system is unstable (Figure 3(a)). The primary eigenmode corresponds to f = π , and the eigenvector thus is [1 , -1 1 , , -1 , . . . , 1 , -1] from Eq. (18), corresponding to dimerization of the lamellae. This calculation shows that to avoid the lamella collapse, we need to keep k 1 > 0 or reduce | k 1 | when k 1 < 0 and increase k b . Practical approaches to implement this include the use of liquid with low surface tension and contact angle close to 90 , at which o k 1 is at its maximum positive value, the use of stiff solids, and/or a proper choice of geometric parameters, e.g. a large aspect ratio h/L .
To explain the nature of the instability transition, we make use of the fact that the fastest growing mode is the dimer mode and assume θ n -1 -π/ 2 = π/ 2 -θ n = θ n +1 -π/ 2 = φ . In the vicinity of the critical volume V ∗ at which the instability happens, the dynamics of a lamella is governed by the equation
$$c \frac { d \phi } { d t } = - ( 2 k _ { 1 } + 2 k _ { 2 } + k _ { b } ) \phi - g ( V ^ { * } ) \phi ^ { 3 } + O ( \phi ^ { 5 } ),$$
where c is the dimensionless damping coefficient, and g V ( ∗ ) > 0 is an algebraically lengthy coefficient of the cubic term. Eq. (20) is derived by expanding Eq. (11) in powers of φ ; even orders of φ do not appear in Eq. (20) due to the reflection symmetry φ →-φ inherent in the system. When the volume in a cell reaches the critical value V = V ∗ (equivalently β = β ∗ ), 2 k 1 +2 k 2 + k b = 0, while when 0 < V ∗ -V glyph[lessmuch] 1, a Taylor series expansion of Eq. (14) in the neighborhood of β ∗ gives us 2 k 1 +2 k 2 + k b ∼ β ∗ -β , while a similar expansion of Eq. (15) in the neighborhood of β ∗ yields V ∗ -V ∼ β -β ∗ , and hence 2 k 1 +2 k 2 + k b ∼ V -V ∗ . From Eq. (20), we see that the stable equilibrium state has two branches of solutions φ 1 = -φ 2 ∼ √ V ∗ -V for V ∗ > V and only one solution φ = 0 otherwise, suggesting that the bifurcation is supercritical. Figure 3(b) shows both the positive branch of the asymptotic solution φ 1 , and the results of simulation for 2 lamellae obtained by solving Eq. (11) with periodic boundary conditions. The results agree well, and confirm that the instability is supercritical and leads to lamellar dimerization, as shown in the inset.
FIG. 4: Elastocapillary coalescence of plates. (a) A periodic domain of 32 plates is simulated by solving Eq. (11) for different prescribed volumes V . The tilting angle θ of the first plate is perturbed by 0 1% from 90 . o as the initial condition. The left column shows a sequence of transient states and the right column shows the steady states. (b) The standard deviation of deflection angle φ as a function of time. t is time and τ m is the time scale for mechanical relaxation. (c) Complementing (b), these plots show the evolution of hierarchical formation of bundles for different prescribed volumes. The time resolution of the plots is 0 25 . τ m and plates are considered as a bundle if the tilting angle gradient is positive and larger than the perturbation amplitude.
## D. Nonlinear dynamics: drying, coarsening and refining
## 1. Controlled liquid volume
For an initially translationally invariant system, the two-plate-collapse mode is the fastest growing mode, yet the cluster of dimers is not necessarily the final stable state (red dots in Figure 3(b)). We notice that for a range of sufficient liquid volumes in each cell, both the deflection of the plates and the geometric change of the liquid menisci are large so that the linear approximation in Section II(II C) breaks down. Therefore, we numerically
integrate Eq. (11) directly to follow the hierarchical dynamics by which an array of vertical individual plates first forms dimers, then quadrimers, until eventually forming large bundles that are limited by elastic effects. Figure 4(a) shows snapshots of dynamical coarsening for different control parameters V , in a periodic domain of 32 plates, triggered by tilting one plate by 0 1% from 90 . o as the initial condition. As expected, for volumes outside the range V/d ∈ (0 5073 0 9393), e.g. . , . for the cases of V/d = 0 95 and 0 4, the array is stable to . . perturbations and remains uniformly vertical. In the unstable parameter range, the primary mode corresponds to two plates collapsing into dimers (Figure 3(a)(c)). These dimers may further collapse into quadrimers or stay as the final stable state with different amplitudes of deformation angles as shown in Figure 4(a)(b). For the same initial perturbation, the dynamical path of successive bundle aggregation depends on the control parameter V . As examples, we see that for V/d = 0 92, quadrimers sweep through the domain right after . the dimers form and the system reaches the stable equilibrium, for V/d = 0 80 the dimers . persist for a while before they eventually collapse to quadrimers, and for V/d = 0 70 the . dimers stay stable, as shown in Figure 4(c).
Having seen how the system coarsens when the volume is controlled, we now turn to the more realistic case when the volume itself evolves dynamically.
## 2. Coupling to dynamics of drying
For an evaporation dominated situation, the rate at which the liquid volume in each cell is reduced depends on the local surface area of the air-liquid interface and thus on deflection angles of the adjacent lamellae. Consequently, the drying dynamics is coupled to the evolution of the geometric configuration, and new instabilities associated with inhomogeneous cell volumes are expected, which cannot be captured by either energy minimization [8, 18, 19] or renormalization analysis [20].
A minimal evaporation model that is sufficient to capture the qualitative features of hierarchical bundle formation is given by
$$\frac { d V _ { n } } { d t } = \begin{cases} \begin{array} { c } 2 ^ { t _ { m } } _ { t _ { e } } c _ { e } r _ { n } \beta _ { n } \end{array} \text{ when} \quad V _ { n } > 0, \\ 0 \quad \text{otherwise}, \end{cases}$$
where c e is a constant, τ m and τ e are time scales for mechanical relaxation and evaporation respectively, r n is the radius of the meniscus and 2 β n is the angle subtended by the meniscus
t /
m
FIG. 5: Drying induced elastocapillary coalescence of plates assuming evaporation rates to depend on the surface area in each cell. A periodic domain of 100 plates is simulated by solving the coupled Eqs. (11) and (21), of which only 32 are displayed. (a) Snapshots of the array of plates and menisci at different times. The initial conditions are V /d n = 0 92 except that . V 99 is smaller by 2% and θ n = 90 . o (b) Number count of bundles of different size as a function of time for (a). (c) Same as (a) but with uniform random initial perturbations with the maximum amplitude of 5% on both V /d n = 0 92 and . θ n = 90 . (d) Number count of bundles of different size as a function of time for o (c). Note the 3- and 5-plate bundles.
arc. Eq. (11) and Eq. (21) coupled together determine the dynamics of lamella coalescence driven by evaporating liquid, with three regimes. When t m glyph[lessmuch] t e , V n decreases quasistatically so that Eq. (20) relaxes to a static state for a prescribed value of V n . When t m ∼ t e , the evolution of the cell volumes V n and the lamellar configuration θ n are coupled. When t m glyph[greatermuch] t e , the evaporation is so fast that the lamella array does not coarsen. In our typical experiments, τ m ∼ 22 5 . ms and τ e is of the order of seconds, so we choose c e correspondingly in Eq. (21) so that the liquid dries out in about 7 seconds. Figure 5 shows the simulation results for a periodic domain of 100 plates with 2 different initial perturbations.
In Figure 5(a), all plates are perfectly vertical, and the liquid volume in each cell is constant for all cells except that V 99 is smaller by 2% to mimic boundary effects in an experimental system. The evolution of the system indicates that a front of dimer coalescence propagates from the imperfection site and sweeps through the entire domain, followed by a
successive front of quadrimer coalescence. The largest transient bundles have 4 lamellae. If adhesion between contacting lamella is neglected, capillary forces are not sufficiently large to hold the plates together, and the bundles separate symmetrically when the liquid volume falls bellow a second threshold, and a perfectly vertical configuration is restored. Figure 5(b) shows the number of bundles of different sizes as a function of time corresponding to the configuration shown in Figure 5(a), and highlights the fact that bundle formation/separation is perfectly hierarchical and regular.
In Figure 5(c), uniform random perturbations with maximum relative amplitude of 5% are applied to all the tilting angles and liquid volume in all cells. Irregular bundles of size ranging from 2 to 5 form transiently, and separate as the liquid evaporates. Figure 5(d) presents the number count of bundles of different size for Figure 5(c), which shows that although the dimer is still the dominant mode in the early stages, bundle formation can be irregular. This is because the the jump in the location of the contact line when the contact angle reaches a critical value leads to a sudden decrease in the effective stiffness as shown in Figure 3(a), and the uniform random initial perturbation generates multiple sites from which the front of dimer coalescence starts propagating. Consequently the sites are not necessarily separated by an even number of plates. Therefore, trimers and pentamers also arise in addition to quadrimers. Bundles of larger size do not appear because of the large elastic energy associated with their formation.
The dynamics of coalescence in these two cases shows how the number of plates per cluster varies in a step-like manner, very similar to the experimental data reported by Pokroy [7] and Gat [20]. All these features also agree qualitatively well with our own experimental observations shown in Figure 1(a)(b).
## III. COLLECTIVE DYNAMICS OF A TWO-DIMENSIONAL ARRAY OF PILLARS
## A. Experimental observations
We now generalize our study of the one dimensional dynamics of plate or lamella aggregation driven by capillarity to the coalescence of a two-dimensional array of epoxy nano-pillars immersed in an evaporating wetting liquid as reported in detail in previous work [7, 25].
The dynamics of pillars is different from that of plates in 3 major ways. First, fluid can flow freely around the multi-connected domains associated with pillars, so that the interaction between them occurs over much longer ranges rather than being limited to just nearest neighbors. Secondly, the three-dimensional geometry allows the pillars to bend in two principal directions and also twist. For pillars with a circular cross-section, the twist must be a constant. If we neglect friction between adhering filaments, the twist must identically vanish, and here we will assume that this is the case. Thirdly, experimentally we see that a segmented, 'wormlike' geometry of the specially treated pillars increases pinning of the receding contact line by reentrant curvature [7]; here will neglect this effect for simplicity.
As in the case of lamellae, the uniform array of non-interacting straight pillars loses stability as the liquid evaporates. The dynamics of the ensuing structures is a result of the competition between elasticity and capillarity, and the morphology of the final assembly is determined by intrapillar elasticity and interpillar adhesion [25]. Figure 1(c)-(e) show the scanning electron microscopy (SEM) images of the assembly into uniform periodic fourfold clusters of nanopillars, in which the pillar height L = 4 5 . µm , the pillar radius R = 150 nm , the pillar spacing D = 2 µm , the Young's modulus E = 0 2 . GPa , the surface tension of the liquid σ = 0 022 . N/m , and the density of the liquid ρ = 786 Kg/m 3 . Unlike in the onedimensional array of lamellae, where the dimer is the primary unstable mode, for pillars the quadrimer is the primary unstable mode. As the liquid evaporates, this mode gives way to hierarchically grow into larger assemblies which eventually get arrested by the increase in the elastic resistance.
## B. Mechanics: coupling filament deformation to fluid interface shape
As in the lamellar case, inertial effects can be neglected here, so that the dynamics of each pillar tip can be characterized by its displacement vector relative to its base -→ X x,y ( ) (Figure 6(a)), and is given by
$$c \frac { d \overrightarrow { X } } { d t } + \overrightarrow { F } _ { b } ( \overrightarrow { X } ) + \overrightarrow { F } _ { \sigma } ( \overrightarrow { X }, V ) = 0,$$
where c is the drag coefficient, -→ F b is the elastic bending resistance force at the tip, -→ F σ is the capillary driving force due to surface tension σ , and V is the liquid volume in the system. The dominant contribution to c is from the internal damping of the viscoelastic solid similar
to the lamellar case, and c ≈ 3 t m πER / L 4 4 3 (APPENDIX D), where t m ∼ 10 -2 s is the time scale for the fiber to relax mechanically.
To compute the bending resistance force -→ F b in the horizontal direction, we use the theory of the elastica for the inextensional, unshearable deformation of thin filament. Letting ϑ be the angle of the pillar centerline tangent with the vertical direction, with s ∈ [0 , L ] is the arc length coordinate, and | -→ | F b the force amplitude at the tip, equilibrium implies that
$$\frac { \pi } { 4 } R ^ { 2 } E \vartheta _ { s s } + | \overrightarrow { F } _ { b } | \cos \vartheta = 0,$$
Geometry implies that | - | → X = ∫ L 0 cos[ ϑ s ( ; | -→ | F b )] ds , so given -→ | X , -→ | F b is uniquely determined, and -→ F b = | -→ | F b -→ | X/ X - | → .
To obtain -→ F σ , we need to determine the shape of the air-liquid interface. Since the Bond number Bo ∼ 10 -6 glyph[lessmuch] 1, gravity can be neglected. Moreover, the time scale for the fluid to equilibrate in the porous brush t f is much smaller than that for the pillars to relax mechanically t m , which is much smaller than that for the evaporation t e , i.e. t f ∼ 10 -3 s glyph[lessmuch] t m ∼ 10 -2 s glyph[lessmuch] t e ∼ 10 0 s (APPENDIX E). Therefore, the pressure throughout the liquid domain can be regarded as uniform and the air-liquid interface z = S x, y ( ) is thus a surface of uniform mean curvature and satisfies the equation
$$2 \sigma H = \sigma \frac { ( 1 + S _ { x } ^ { 2 } ) S _ { y y } - 2 S _ { x } S _ { y } S _ { x y } + ( 1 + S _ { y } ^ { 2 } ) S _ { x x } } { ( 1 + S _ { x } ^ { 2 } + S _ { y } ^ { 2 } ) ^ { 3 / 2 } } = p,$$
where H is the mean curvature of the interfacial surface. Without loss of generality, we set the ambient pressure to zero, and let p be the pressure inside the liquid. Volume conservation in the whole domain yields
$$V = \int _ { A } S ( x, y ) d x d y + \pi R ^ { 2 } \sum _ { i = 1 } ^ { N } ( h _ { i } - L ),$$
and serves to determine p . Here A is the projected domain of the air-liquid interface to the horizontal plane (meshed area in Figure 6(a)), and h i is the elevation of i th pillar tip. Since the menisci are always pinned on the pillar tips, we need to solve Eq. (24) on a multiply connected domain, in which pillar tips are regarded as solid circles with the identical radius (Figure 6(a)) and the height of the surface is fixed at the elevation of pillar tips (Figure 6(b)) h = ∫ L 0 sin ϑds calculated from Eq. (23). As the pillars are effectively immersed in liquid, the integration of pressure over the lateral surface of the cylinder does not contribute to -→ F σ , and the only active contribution is the line tension at the contact line. For a given air-liquid
interface (Figure 6(c)), the angle of the meniscus tangent with the horizontal direction on the circular boundary of the tip is known, which we denote as ϕ . To calculate the capillary driving force on each pillar, we integrate the interfacial force over the contact line contour at the tip and determine the component in the horizontal direction that contributes to the deflection, with -→ F σ = ∮ c σ n -→ cos ϕds , where the subscript c represents the tip circle and -→ n is its unit outward normal. Note that for very small values of V , the assumption of the surface being pinned at the pillar tips breaks down; however clusters form well before this assumption is violated, so that this is not a cause for concern.
To prevent penetration upon collision between pillars, we treat each pillar as a rod with finite radius with an artificial short-range repulsion force when the discs representing pillar tips come close enough (10% of the pillar diameter), but do not consider the elastic deformation of the cross section due to contact.
## C. Nonlinear dynamics: simulations and comparison with experiments
To complete the formulation of the problem, we need some boundary conditions. We consider a square array of pillars inside a domain with straight vertical walls and impose symmetry-related conditions on the walls for the fluid interface and the pillar deformations. On the pillars, the contact lines are assumed to be pinned on all pillar tips. For initial conditions, the pillar bases (open circles in Figure 6(g)) are assumed to form a perfect periodic square lattice, while the pillar tips (solid white circles in Figure 6(g)) are perturbed from the vertical configuration so that the layer of pillars closest to the boundaries inclines inwards to trigger the inward motion from the boundaries, and other pillar tips are uniformly and randomly perturbed. We numerically solve the coupled evolution equations (3.1)-(3.4) using a custom-coded finite element code. For a given pillar tip displacement - → X i , Eq. (23) is solved with the geometric condition | - → X i | = ∫ L 0 cos[ ϑ i ( s ; | -→ | F bi )] ds to determine the reaction force -→ F bi and tip elevation of each pillar h i . Given - → X i and h i , Eq. (24) is solved using the finite element method to determine the interface S x, y ( ) on the multiply-connected domain with the liquid volume constraint Eq. (25), which determines p and thence -→ F σi . Then Eq. (22) is integrated in time explicitly to update the pillar tip positions, and the domain is remeshed accordingly at every time step.
Figure 6(d)-(g) show the simulated evolution of an array of 14 by 14 pillars collapsing
FIG. 6: Elastocapillary coalescence of a square array of pillars. (a) Demonstration of triangular meshes on a domain that contains an array of 2 by 2 vertical pillars, where the white solid circles represents pillar tips. The mesh density used in the actual simulation is 4 times denser. (b) The three-dimensional air-liquid interface is obtained by solving Eq. (24) on the domain shown in (a) for a given liquid volume V/V flat = 0 85 in Eq. (25). (c) Given a slight perturbation to the vertical . state in (b), 4 pillars coalesce to form a bundle. (d)-(g) A domain of 14 by 14 pillars evolves to the steady state of fourfold clusters for a given liquid volume V/V flat = 0 90. . The dynamics follows the coupled evolution equations (22)-(25). τ is the dimensionless time scaled by τ m (see text). The gray scale shows the air-liquid interface height, scaled by the pillar spacing. White solid circles represent pillar tips, and the black open circles represent pillar bases. The dashed lines connecting bases and tips correspond to a projection of pillars viewed from the top.
FIG. 7: More examples of final structures of the fourfold clusters for different prescribed volumes and domain sizes with random initial perturbations and symmetric boundary conditions. (a) 14 by 14 pillars with a given liquid volume V/V flat = 0 90. . (b) 10 by 16 pillars with a given liquid volume V/V flat = 0 88. Simulations were done by solving equations (22)-(25). .
into fourfold bundles for a prescribed liquid volume of V/V flat = 0 90, where . V flat is the control volume inside the system when all the pillars are vertical and the air-liquid interface is flat, using the parameters from experiments in Figure 1(c)-(e). Due to the initial boundary perturbations, coalescence is initiated from the boundary and propagates towards the center of the domain. We observe that there is a critical liquid volume above which uniform vertical pillars are stable, and below which pillar coalescence occurs, similar to the case of lamellar collapse. However, here the primary eigenmode of instability has a fourfold symmetry (Figure 6(e)) independent of initial perturbations as observed in experiments (Figure 6(g) and Figure 7). An intuitive way to understand this is to recognize that the fourfold bundles have two principal directions, along which one sees dimers. Although the interaction potential between pillars in the three-dimensional case has a much longer range than in the two-dimensional case because of connectivity, it is still monotonic and decreases with distance. Provided that within the linear analysis the effective spring constants in the two principal directions are decoupled, the dimer is the primary mode in each direction as in the one-dimensional lamella array. Beyond the linear regime, our simulations capture the coarsening that is arrested and eventually leads to a maximum bundle size. We also note that the tips in a bundle can form either rhombi or squares depending on the initial perturbations, although rhombi are more likely as they are stable against shear deformations;
indeed as we neglect friction between the tips, the rhombus is a more energetically favorable configuration than the square. However, the energy difference between these two states is very small, so that contact and friction in real experiments leads to both square tips (Figure 1(d)) and rhombi tips (Figure 1(c)).
## IV. CONCLUSIONS
Our study has focused on understanding the onset and evolution of elasto-capillary coalescence of plates and pillars driven by evaporation. For the case of lamellar collapse, we explicitly derived the conditions for the primary dimerization instability in terms of the state variables - tilting angles θ n and liquid volumes V n , and the relevant geometrical and physical parameters. Complementing our analysis, full numerical simulations show that the final coalescent states sensitively depend on initial perturbations because of the discontinuous motion of the contact line when the contact angle reaches a critical value. This implies that the self-organization of clusters cannot be predicted by energy minimization arguments alone, but depend on the dynamics of the drying process - this is especially true when the coupling of geometry to local evaporation rates is taken into account. Our model accounts for this, and is in qualitative agreement with experimental observations of the intermittence of coalescent transitions. For the case of pillar collapse, our model correctly accounts for the multiply-connected nature of the fluid interface, and the large elastic deflections of the pillars. The analysis based on this model captures the primary fourfold eigenmode associated with the onset of collapse, consistent with experimental observations. Numerical simulations of the full dynamics allow us to follow the evolution of the clusters whose eventual size is determined by the competition between capillarity and elasticity. For both cases, our numerical results agree well with many of the salient experimental observations. In particular, we can explain the eigenmodes at the onset of instability, and the time scales on which clusters form, while providing explanations for both regular and irregular hierarchical bundling till the final state.
However, our analysis still leaves out some effects and thus cannot explain some observations. Neglecting adhesion and friction between pillars implies that we cannot explain the twisting of pillars that leads to the formation of chiral clusters often seen. This is a natural next step in the analysis. Furthermore, we have limited ourselves to a discrete theory in
both cases, but in the thermodynamic limit of a large number of pillars or lamellae, one might ask what the nature of a continuum theory might be. A recent continuum theory that addresses the explicit connection of the essential geometric and physical parameters to determine the maximal size and dynamics of the assembly has been carried out for lamellar coalescence [22], but the question for 3-dimensional coalescence remains an open question.
## Acknowledgment
We thank Sam Ocko for many discussions that helped to sharpen and clarify our arguments. We thank the Harvard-MRSEC DMR -0820484, the MacArthur Foundation (LM), and the NRF of Korea (Grant No. 2013034978, H.-Y.K.) for support.
- [1] G. R. Desiraju, Crystal Engineering: The Design of Organic Solids, Elsevier, New York, 1989.
- [2] N. Bowden, A. Terfort, J. Carbeck and G.M. Whitesides. Assembly of mesoscale bojects into ordered two-dimensional arrays. Science 1997 , 276, 233-235
- [3] P. W. K. Rothemund, Using lateral capillary forces to compute by self-assembly. Proc. Natl. Acad. Sci. U.S.A 2000 , 97, 984-989.
- [4] D. Vella and L. Mahadevan. The 'Cheerios effect'. Am. J. Phys. 2005 , 73, 817-825.
- [5] B. Grzybowski, H.A. Stone, and G.M. Whitesides. Dynamic self-assembly of magnetized, millimeter-sized objects rotating at the liquid-air interface. Nature 2000 , 405, 1033.
- [6] B.A. Grzybowski, X. Jiang, H.A. Stone and G.M. Whitesides. Dynamic, self-assembled aggregates of magnetized, millimeter-sized objects rotating at the liquid-air interface: macroscopic, two-dimensional classical artificial atoms and molecules. Phys. Rev. E 2001 , 64, 11603.
- [7] B. Pokroy, S. H. Kang, L. Mahadevan and J. Aizenberg. Self-organization of a mesoscale pillar into ordered, hierarchical helical assemblies. Science 2009 , 323, 237-240.
- [8] J. Bico, B. Roman, L. Moulin and A. Boudaoud. Elastocapillary coalescence in wet hair. Nature 2004 , 432, 690.
- [9] H. Y. Kim and L. Mahadevan. Capilliary rise between elastic sheets. J. Fluid Mech. 2006 , 548, 141-150.
- [10] T. Tanaka, M. Morigami and N. Atoda. Mechanism of resist pattern collapse during development process. Jpn. J. Appl. Phys. 1993 , 32, 6059-6059.
- [11] N. Bernardino, V. Blickle, and S. Dietrich. Wetting of surfaces covered by elastic hairs. Langmuir 2010 , 26, 7233-7241.
- [12] N. Bernardino and S. Dietrich. Complete wetting of elastically responsive substrates. Phys. Rev. E 2012 , 85, 051603.
- [13] A. K. Geim, S. V. Dubonos, I. V. Grigorieva, K. S. Novoselov, A. A. Zhukov and S. Y. Shapoval. Microfabricated adhesive mimicking gecko foot-hair. Nature Mater. 2003 , 2, 461463.
- [14] K. K. S. Lau, J. Bico, K. B. K. Teo, M. Chhowalla, G. A. J. Amaratunga, W. I. Milne, G. H. McKinley and K. K. Gleason. Superhydrophobic carbon nanotube forests. Nano Lett. 2003 , 3, 1701-1705.
- [15] N. Chakrapani, B. Wei, A. Carrillo, P. M. Ajayan, and R. S. Kane. Capillarity-driven assembly of two-dimensional cellular carbon nanotube foams. Proc. Natl. Acad. Sci. U.S.A 2004 , 101, 4009-4012.
- [16] C. Duprat, J. M. Aristoff and H. A. Stone. Dynamics of elastocapillary rise. J. Fluid Mech. 2011 , 679, 641-654.
- [17] M. Taroni and D. Vella. Multiple equlibria in a simple elastocapillary system. J. Fluid Mech. 2012 , 712, 273-294.
- [18] D. Chandra, S. Yang, A. A. Soshinsky and R. J. Gambogi. Biomimetic ultrathin whitening by capillary-force-induced random clustering of hydrogel micropillar arrays. ACS Appl. Mater. Interfaces 2009 , 1, 1698-1704.
- [19] Y. P. Zhao and J. G. Fan. Clusters of bundled nanorods in nanocarpet effect. Appl. Phys. Lett. 2006 , 88, 103123.
- [20] A. D. Gat and M. Gharib. Elasto-capilliary coalescence of multiple parallel sheets. J. Fluid Mech. 2013 , 723, 692-705.
- [21] A. Boudaoud, J. Bico and B. Roman. Elastocapillary coalescence: aggregation and fragmentation with a maximal size. Phys. Rev. E 2007 , 76, 060102.
- [22] Z. Wei and L. Mahadevan. Continuum dynamics of elastocapillary coalescence and arrest. Europhys. Lett. 2014 , 106, 14002.
- [23] S. Neukirch, B. Roman, B. Gaudemaris, J. Bico. Piercing a liquid surface with an elastic rod:
Buckling under capillary forces. J. Mech. Phys. Solids 2007 , 55, 1212-1235.
[24] S. Timoshenko and J. M. Gere. Mechanics of Materials, Van Nostrand Reinhold Co., 1972, p207.
[25] S. H. Kang, B. Pokroy, L. Mahadevan and J. Aizenberg. Control of shape and size of nanopillar assembly by adhesion-mediated elastocapillary interaction. ACS Nano 2010 , 4, 6323-6331.
## Appendix A: Critical buckling length of a plate
The critical buckling height of a thin plate under compression due to surface tension σ is
$$L _ { c r i t } \sim L _ { e c } = \sqrt { E h ^ { 3 } / \sigma },$$
where L ec is the typical elastocapillary length scale. For the case when the plate is clamped at one end and axially compressed at the other, the exact expression [1] for L crit is
$$L _ { c r i t } = \pi [ 4 8 ( 1 - \nu ^ { 2 } ) ] ^ { - 1 / 2 } L _ { e c } \approx 1 m m,$$
where we have substituted in the experimentally observed parameter values. As L crit is much larger than the lamella height L = 40 µm , buckling of the lamellae when they pierce the gas-liquid interface does not happen in our system.
## Appendix B: Moment calculation for the 2D case
The 8 possible meniscus configurations can be classified into 6 cases, where we explicitly express the moments in terms of the state variables - tilting angles θ n and liquid volumes V n . We have scaled all lengths by L and moment by σL , so all results below are dimensionless.
Case 1: The meniscus is pinned on both tips (Figure 2(b1)). The half angle β n subtended by the meniscus arc is determined by solving
$$V _ { n } = \frac { 1 } { 2 } d ( \sin \theta _ { n } + \sin \theta _ { n + 1 } ) + \frac { 1 } { 2 } \sin ( \theta _ { n } - \theta _ { n + 1 } ) - a _ { n } ^ { 2 } ( \beta _ { n } \csc ^ { 2 } \beta _ { n } - \cot \beta _ { n } ), \quad \text{(B1)}$$
for given V n , θ n and θ n +1 , where d = D/L and half of the tip distance a n is
$$a _ { n } = \frac { 1 } { 2 } \sqrt { 2 - 2 \cos ( \theta _ { n + 1 } - \theta _ { n } ) + 2 d ( \cos \theta _ { n + 1 } - \cos \theta _ { n } ) + d ^ { 2 } }.$$
β n must satisfy β n ≤ ˆ , where β n ˆ β n is the critical angle at which the meniscus starts to slide down from at least one lamella. β n < 0 when the meniscus concaves down, β n = 0 when the
meniscus is flat, and β n > 0 when the meniscus concaves up. The moments on the n th and ( n +1) th plates are given respectively by
$$M _ { n } = - \frac { 1 } { 2 a _ { n } } \left [ \sin ( \beta _ { n } + \theta _ { n } - \theta _ { n + 1 } ) + d \sin ( \beta _ { n } + \theta _ { n } ) \right ],$$
$$M _ { n + 1 } = \frac { 1 ^ { ^ { * } } } { 2 a _ { n } } \left [ \sin ( \beta _ { n } + \theta _ { n } - \theta _ { n + 1 } ) - d \sin ( \beta _ { n } - \theta _ { n + 1 } ) \right ],$$
glyph[negationslash]
Case 2: The meniscus is down from both tips (Figure 2(b3)). The contact angle is fixed at α . When θ n = θ n +1 , the radius of the meniscus r n is determined by solving
$$V _ { n } = & \frac { d ^ { 2 } } { 2 } \frac { \sin \theta _ { n } \sin \theta _ { n + 1 } } { \sin ( \theta _ { n + 1 } - \theta _ { n } ) } - r _ { n } ^ { 2 } \left [ \cos ^ { 2 } \alpha \cot \left ( \frac { \theta _ { n + 1 } - \theta _ { n } } { 2 } \right ) \\ & + \sin \alpha \cos \alpha + \frac { \pi + \theta _ { n + 1 } - \theta _ { n } - 2 \alpha } { 2 } \right ].$$
The wetting length on the right side of the n th plate and that on the left side of the ( n +1) th are given respectively by
$$L _ { w r _ { n } } = d \frac { \sin \theta _ { n + 1 } } { \sin ( \theta _ { n + 1 } - \theta _ { n } ) } - r _ { n } \left [ \sin \alpha + \cos \alpha \cot \left ( \frac { \theta _ { n + 1 } - \theta _ { n } } { 2 } \right ) \right ].$$
$$L _ { w l n + 1 } = d \frac { \sum _ { n } ^ { n } } { \sin ( \theta _ { n + 1 } - \theta _ { n } ) } - r _ { n } \, \tilde { \left [ \sin \alpha + \cos \alpha \cot \left ( \frac { \dot { \theta } _ { n + 1 } - \theta _ { n } } { 2 } \right ) } \tilde { \right ] } \,.$$
When θ n = θ n +1 = , the meniscus radius is independent of θ V n , and r n = d sin θ/ (2 cos α ). l n is determined by solving
$$V _ { n } = \frac { 2 l _ { n } - d \cos \theta } { 2 } d \sin \theta - \left ( \frac { d \sin \theta } { 2 } \right ) ^ { 2 } \tan \alpha - \frac { \pi - 2 \alpha } { 2 } \left ( \frac { d \sin \theta } { 2 \cos \alpha } \right ) ^ { 2 }.$$
The wetting length on the right side of the n th plate and that on the left side of the ( n +1) th plate are given respectively by
$$L _ { w r n } = l _ { n } - \frac { d } { 2 } \sin \theta \tan \alpha,$$
$$L _ { w l n + 1 } = l _ { n } - \frac { \bar { d } } { 2 } \sin \theta \tan \alpha - d \cos \theta.$$
In either case, the moment on the n th plate and that on the ( n + 1) th plate are given respectively by
$$M _ { n } = - \frac { L _ { w r _ { n } } ^ { 2 } } { 2 r _ { n } } - L _ { w r _ { n } } \sin \alpha.$$
$$M _ { n + 1 } = \frac { L _ { w l _ { n + 1 } } ^ { \stackrel { \nu } { 2 } ^ { \nu } } } { 2 r _ { n } } + L _ { w l _ { n + 1 } } \sin \alpha.$$
Case 3: when the meniscus slides down from the n th plate and is pinned on the ( n +1) th one. In this case L wrn = l n and L wln +1 = 1.
$$a _ { n } = \frac { 1 } { 2 } \sqrt { 1 + l _ { n } ^ { 2 } - 2 l _ { n } \cos ( \theta _ { n } - \theta _ { n + 1 } ) + 2 d ( \cos \theta _ { n + 1 } - l _ { n } \cos \theta _ { n } ) + d ^ { 2 } },$$
$$V _ { n } = \frac { 1 } { 2 } l _ { n } \sin ( \theta _ { n } - \theta _ { n + 1 } ) + \frac { 1 } { 2 } d ( l _ { n } \sin \theta _ { n } + \sin \theta _ { n + 1 } ) - a _ { n } ^ { 2 } ( \beta _ { n } \csc ^ { 2 } \beta _ { n } - \cot \beta _ { n } ).$$
From the condition that contact angle on the n th plate is α , we can get the following relations,
$$\cos ( \alpha + \theta _ { n } - \theta _ { n + 1 } ) + d \cos ( \alpha + \theta _ { n } ) - l _ { n } \cos \alpha = 2 a _ { n } \cos \beta _ { n },$$
and glyph[negationslash]
$$\i d \quad & \sin ( \alpha + \beta _ { n } + \theta _ { n } - \theta _ { n + 1 } ) + d \sin ( \alpha + \beta _ { n } + \theta _ { n } ) = l _ { n } \sin ( \alpha + \beta _ { n } ) \\ & \quad \text{when} \quad \theta _ { n } \neq \theta _ { n + 1 }, \\ & \quad d \cos \theta + 1 - l _ { n } = \cos ( \alpha + \beta _ { n } ) \\ & \quad \text{when} \quad \theta _ { n } = \theta _ { n + 1 } = \theta. \\ \dots \Delta \circ \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \alpha \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \beta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \dots \Delta \d$$
l n and β n are determined by solving either (B14) and (B15) or (B14) and (B16).
Moment on the n th plate and that on the ( n +1) th plate are given respectively by
$$M _ { n } = - \frac { \sin \beta _ { n } } { 2 a _ { n } } l _ { n } ^ { 2 } - l _ { n } \sin \alpha.$$
$$M _ { n + 1 } = \frac { \sin \beta _ { n } ^ { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel { \stackrel \stack { \stackrel { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \stack { \ \stack { \ \stack { \ \stack { \ \ } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } ) } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } { } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } \ } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } { \, \, \,$$
Case 4: when the meniscus slides down from the ( n + 1) th plate and is pinned on the n th one. In this case L wrn = 1 and L wln +1 = l n .
$$a _ { n } = \frac { 1 } { 2 } \sqrt { 1 + l _ { n } ^ { 2 } - 2 l _ { n } \cos ( \theta _ { n } - \theta _ { n + 1 } ) + 2 d ( l _ { n } \cos \theta _ { n + 1 } - \cos \theta _ { n } ) + d ^ { 2 } }, \quad \quad ( B 1 9 )$$
$$V _ { n } = \frac { 1 } { 2 } l _ { n } \sin ( \theta _ { n } - \theta _ { n + 1 } ) + \frac { 1 } { 2 } d ( \sin \theta _ { n } + l _ { n } \sin \theta _ { n + 1 } ) - a _ { n } ^ { 2 } ( \beta _ { n } \csc ^ { 2 } \beta _ { n } - \cot \beta _ { n } ).$$
From the condition that contact angle on the ( n +1) th plate is α , we can get the following relations,
$$\cos ( \alpha + \theta _ { n } - \theta _ { n + 1 } ) - d \cos ( \alpha - \theta _ { n + 1 } ) - l _ { n } \cos \alpha = - 2 a _ { n } \cos \beta _ { n },$$
and glyph[negationslash]
$$\text{and} \\ & \sin ( \alpha + \beta _ { n } + \theta _ { n } - \theta _ { n + 1 } ) - d \sin ( \alpha + \beta _ { n } - \theta _ { n + 1 } ) = l _ { n } \sin ( \alpha + \beta _ { n } ) \\ & \text{when} \quad \theta _ { n } \neq \theta _ { n + 1 }, \\ & d \cos \theta + l _ { n } - 1 = \cos ( \alpha + \beta _ { n } ) \\ & \text{when} \quad \theta _ { n } = \theta _ { n + 1 } = \theta.$$
Similarly, l n and β n are determined by solving either (B20) and (B21) or (B20) and (B22).
Moment on the n th plate and that on the ( n +1) th plate are given respectively by
$$M _ { n } = - \frac { \sin \beta _ { n } } { 2 a _ { n } } + \sin ( \theta _ { n + 1 } - \theta _ { n } - \alpha - 2 \beta _ { n } ).$$
$$M _ { n + 1 } = \frac { \sin \beta _ { n } } { 2 a _ { n } } l _ { n } ^ { 2 } + l _ { n } \sin \alpha.$$
For certain given V n , θ n and θ n +1 , although one end of the meniscus is depinned from the tip, there is no arc satisfying the enforced contact angle condition. In the following two cases, the menisci are approximated by straight lines. They are used to prevent the numerics from blowing up when two plates almost contact, yet very unlikely to happen.
Case 5: when the meniscus slides down from the tip of the n th plate and keeps flat. The moment on the n th plate and that on the ( n +1) th plate are given respectively by
$$M _ { n } = - \frac { l _ { n } } { b _ { n } } [ \sin ( \theta _ { n } - \theta _ { n + 1 } ) + d \sin \theta _ { n } ]$$
$$M _ { n + 1 } = \frac { 1 } { b _ { n } } ^ { \stackrel { \mu \nu } { \nu } } [ l _ { n } \sin ( \theta _ { n } - \theta _ { n + 1 } ) + d \sin \theta _ { n + 1 } ],$$
where l n is determined by solving from the given volume
$$V _ { n } = \frac { l _ { n } } { 2 } \sin ( \theta _ { n } - \theta _ { n + 1 } ) + \frac { d } { 2 } ( l _ { n } \sin \theta _ { n } + \sin \theta _ { n + 1 } ),$$
and b n is
$$b _ { n } = \sqrt { 1 + l _ { n } ^ { 2 } - 2 l _ { n } \cos ( \theta _ { n } - \theta _ { n + 1 } ) + 2 d ( \cos \theta _ { n + 1 } - l _ { n } \cos \theta _ { n } ) + d ^ { 2 } }.$$
Case 6: when the meniscus slides down from the tip of the n th plate and keeps flat. The moment on the n th plate and that on the ( n +1) th plate are given respectively by
$$M _ { n } = - \frac { 1 } { b _ { n } } [ l _ { n } \sin ( \theta _ { n } - \theta _ { n + 1 } ) + d \sin \theta _ { n } ]$$
$$M _ { n + 1 } = \frac { l _ { n } } { b _ { n } } [ \sin ( \theta _ { n } - \theta _ { n + 1 } ) + d \sin \theta _ { n + 1 } ],$$
where l n is determined by solving from the given volume
$$V _ { n } & = \frac { l _ { n } } { 2 } \sin ( \theta _ { n } - \theta _ { n + 1 } ) + \frac { d } { 2 } ( \sin \theta _ { n } + l _ { n } \sin \theta _ { n + 1 } ),$$
and b n is
$$b _ { n } = \sqrt { 1 + l _ { n } ^ { 2 } - 2 l _ { n } \cos ( \theta _ { n } - \theta _ { n + 1 } ) + 2 d ( l _ { n } \cos \theta _ { n + 1 } - \cos \theta _ { n } ) + d ^ { 2 } }.$$
## Appendix C: Damping coefficient for the 2D case of plates
In order to calculate the damping coefficient, we need to account for both the viscosity of the fluid and the viscoelasticity of the solid. First we consider the contribution from the fluid. The Reynolds number Re ∼ 10 -5 or less, so that inertia of the fluid is negligible. We use lubrication theory to calculate the moment acting on the plate caused by flow although the ratio of plate height L to the spacing D does not strictly satisfy that L/D glyph[greatermuch] 1. Figure 8 illustrates 3 rigid plates hinged at the base. The upper and lower one are perpendicular to the substrate, and the middle one is rotating clockwise at the angular velocity ˙ . θ The liquid is confined between two plates and the ambient pressure is set to be 0.
In the bottom chamber, the momentum conservation of the fluid in the x and y directions are
$$\frac { \partial p ^ { + } } { \partial x } = \mu \frac { \partial ^ { 2 } u } { \partial y ^ { 2 } } \quad \text{and} \quad \frac { \partial p ^ { + } } { \partial y } = 0,$$
where p + is the pressure and µ is the fluid viscosity. The boundary conditions are
$$u = 0 \text{ \ at \ } y = 0 \text{ \ and \ } y = H ( x, t ).$$
From (C1) and (C2), we can get
$$u ( x, y, t ) = \frac { 1 } { 2 \mu } \frac { \partial p ^ { + } } { \partial x } ( y ^ { 2 } - H y ).$$
The mass conservation is
$$\frac { d } { d t } \left [ \frac { 1 } { 2 } \rho x ( D + H ) \right ] = \rho [ Q ( 0, t ) - Q ( x, t ) ],$$
where Q x, t ( ) is the flux. As u = 0 at y = 0, Q (0 , t ) = 0. Substitute (C3) into (C4),
$$Q ( x, t ) & = \int _ { 0 } ^ { H } u ( x, y, t ) d y = \int _ { 0 } ^ { H } \frac { 1 } { 2 \mu } \frac { \partial p ^ { + } } { \partial x } ( y ^ { 2 } - H y ) d y \\ & = - \frac { H ^ { 3 } } { 1 2 \mu } \frac { \partial p ^ { + } } { \partial x } = - \frac { 1 } { 2 } x \frac { d H } { d t }.$$
From (C5) and H x,t ( ) = D -˙ θtx in the linear regime, we get
$$\frac { \partial p ^ { + } } { \partial x } = - \frac { 6 \mu } { H ^ { 3 } } x ^ { 2 } \dot { \theta }.$$
Integrate (C6) and use the boundary condition on pressure p + ( L, t ) = 0, and we see that
$$p ^ { + } ( s, 0 ) = \frac { 2 \mu L ^ { 3 } } { D ^ { 3 } } \dot { \theta } ( 1 - s ^ { 3 } ),$$
FIG. 8: Sketch of two neighboring cells. When the middle plate is rotating, the nonuniform pressure caused by flow acts on the plate from both sides and hinders the rotation.
where s = x/L ∈ [0 , 1].
So far we have only considered the pressure acting on the lower side of the middle pillar. The pressure difference between two sides is
$$p ^ { + } ( s, 0 ) - p ^ { - } ( s, 0 ) = 2 p ^ { + } ( s, 0 ) = \frac { 4 \mu L ^ { 3 } } { D ^ { 3 } } \dot { \theta } ( 1 - s ^ { 3 } ).$$
And the moment on the middle pillar due to fluid flow is
$$M _ { \mu } = L ^ { 2 } \int _ { 0 } ^ { 1 } ( p ^ { + } ( s, 0 ) - p ^ { - } ( s, 0 ) ) s d s = \frac { 6 \mu L ^ { 5 } \dot { \theta } } { 5 D ^ { 3 } },$$
where the last equality follows from (C8).
Therefore the damping coefficient from the viscous fluid is approximately
$$C _ { 1 } = \frac { M _ { \mu } } { \dot { \theta } } = \frac { 6 \mu L ^ { 5 } } { 5 D ^ { 3 } }.$$
Next we consider the damping coefficient resulted from viscoelasticity of the solid. We can estimate it from the vibration of a viscoelastic beam made of the Kelvin-Voigt material, which can be represented by a viscous damper and elastic spring connected in parallel. The constitutive law thus is
$$\Sigma = E \epsilon + \eta \dot { \epsilon },$$
where Σ is the stress, glyph[epsilon1] is the strain, E is the Young's modulus and η is the viscosity. The strain-displacement relation is
$$\epsilon = \frac { y } { r } = - y \frac { \partial ^ { 2 } u } { \partial x ^ { 2 } },$$
where r is the radius of curvature due to bending, and u is the transverse displacement of the beam. The force balance equation is
$$\frac { \partial ^ { 2 } M } { \partial x ^ { 2 } } + \rho _ { s } \ddot { u } = 0,$$
where the moment M = -∫ A Σ ydA , and ρ s is the density of the solid material.
Substitute (C12) into (C11), then substitute (C11) into (C13), and we can get the vibration equation of a viscoelastic beam, which is
$$E I \frac { \partial ^ { 4 } u } { \partial x ^ { 4 } } + \eta I \frac { \partial ^ { 4 } \dot { u } } { \partial x ^ { 4 } } + \rho _ { s } \ddot { u } = 0, \quad \text{where} \quad I = \int _ { A } y ^ { 2 } d A.$$
For the thin plate, E is replaced by E/ (1 -ν 2 ), ρ is replaced by ρh and I = h / 3 12, where h is the thickness of the plate. We can get two time scales τ m 1 ∼ √ 12 ρ (1 -ν 2 ) L /Eh 4 2 and τ m 2 ∼ η (1 -ν 2 ) /E by balancing the first and the third term, the first and the second term of the LHS of (C14) respectively. Substitute the parameters from the experiment L = 40 µm , h = 10 µm E , = 1 5 . Mpa ν , = 0 5 and . ρ s = 965 kg/m 3 , τ m 1 ∼ 10 µs . We roughly estimate η by using the indentation experimental result in the literature [2]. For the PDMS cured at the base to curing agent ratio of 10 : 1, the loss tangent at a vibration rate of 4 Hz is around 0 09, which is [ . η (1 -ν 2 ) ω/E ] | ω =4 = 0 09, so . τ m 2 ≈ 22 5 . ms . As τ m 2 glyph[greatermuch] τ m 1 , the inertia term is dominated over by the damping term in (C14) for the micro cantilever, and thus can be neglected. Then the damping coefficient due to internal viscosity of the solid is
$$C = C _ { 1 } + C _ { 2 } = \frac { 6 \mu L ^ { 5 } } { 5 D ^ { 3 } } + \tau _ { m } k \approx \tau _ { m } k.$$
where we have omitted the subscript '2' in τ m 2 and τ m ≈ 22 5 . ms is the time scale for the viscoelastic fiber to relax mechanically.
## Appendix D: Damping coefficient for the 3D case of pillars
First, To estimate the drag coefficient c and the time scale in Eq. (22), we consider both the internal contribution from viscoelasticity of the solid material and the external contribution from the viscous fluid, and find that the former one dominates.
AFM characterization of the epoxy microfibers [3] shows that the force vs displacement curve has a clear hysteresis loop, which reveals that the epoxy nanofibers are viscoelastic although primarily elastic. For a fiber of radius R = 1 5 . µm , length L = 9 µm , and bending
spring constant k = 3 πER / L 4 4 3 = 17 22 . N/m , where E = 1 GPa is the Young's modulus, the hysteresis is measured 32% of the total work done at max deflection x = 1 5 . µm and rate of deflection v = 6 µm/s . Thus, the effective damping coefficient c 1 from the internal viscoelasticity can be calculated from 2 c vx 1 = 0 3 . × 0 5 . × kx 2 , based on the Kelvin-Voigt model.
To calculate the damping coefficient c 2 from the external viscous fluid, we consider a circular cylinder of radius R moving with velocity U normal to its axis at small Reynolds number Re = 2 RρU/µ ∼ 10 -7 , in which case, the drag of magnitude 4 πµU/ ln(7 4 . /Re ) per unit length was first derived by Lamb [4]. As the center of the pillar is moving at a velocity of almost v/ 2 when the cantilever tip is deformed at a velocity of v in the AFM test, we take c 2 = 2 πµL/ ln(7 4 . /Re ) as an approximation.
It turns out that c /c 1 2 ∼ 10 7 for the experimental parameters listed above, which indicates that the external damping from viscous fluid can be neglected and c ≈ c 1 . Therefore, the time scale for the fiber to relax mechanically is τ m = c /k 1 ∼ 10 -2 s in Eq. (3.1), and τ m is a material property (as in the Prony series) and does not depend on the geometric dimensions of the fiber.
## Appendix E: The assumption of uniform liquid pressure for the 3D case
Next, we estimate the time scale for the fluid to equilibrate given nonuniform pressure distribution. To do so, we use Darcy's law
$$q = \frac { \kappa } { \mu } \nabla P$$
where q is the flux (discharge per unit area, with units of length per time), κ is the permeability of the two-dimensional pillar array, µ is the viscosity of the fluid, and ∇ P is the pressure gradient. ∇ P ∼ σ Dglyph[lscript] , where σ is the surface tension of the liquid, D is the pillar spacing, and glyph[lscript] is the system size. κ is approximated from a periodic square array of parallel cylinders, which is the initial configuration of our system. In the dilute solid volume fraction (porosity) regime, the asymptotic expression [5] for κ is
$$\kappa = R ^ { 2 } \frac { - 0. 5 \ln \phi - 0. 7 4 5 + \phi - 0. 2 5 \phi ^ { 2 } } { 4 \phi },$$
where R is the radius of the pillar, φ = πR /D 2 2 is the porosity. In our typical experiment, R = 150 nm L , = 4 5 . µm D , = 2 µm E , = 0 2 . GPa , and σ = 0 022 . N/m . For a domain of size
glyph[lscript] ∼ 100 µm , which contains thousands of pillars, the time scale for the fluid to relax so that the pressure inside the domain is uniform can be estimated as τ f = glyph[lscript]/q ∼ 10 -3 < τ b .
In conclusion, for a reasonably large patch of pillars, it is plausible to assume that the pressure throughout the liquid is uniform during the evaporation because of the well-separated time scales - the time scale for the fluid to relax ( ∼ 10 -3 s for 50 by 50 pillars) glyph[lessmuch] that for the pillars to respond ( ∼ 10 -2 s ) glyph[lessmuch] that for the evaporation ( ∼ 10 0 s ).
## Appendix F: Descriptions of movies
## Movie 1: 2D Hierarchical Bundles Experiment.mov
A one-dimensional array of elastic lamellae made of polydimethylsiloxane (PDMS) with pre-polymer and cross-linker weight ratio 10:1 is immersed in the isopropyl alcohol (IPA) at room temperature 25 o C . The lamellae have height 40 µm , thickness 10 µm , depth (into the screen) 40 µm , and uniform spacing 10 µm . The coalescence of lamellae initiates from the right and propagates to the left. Dimers appear first and further collapse into quadrimers. Bundles separate in the reverse order as the liquid evaporates. The surface adhesion force traps the final configuration in the dimer form. The movie is in real time and the duration is about 3 s .
## Movie 2: 2D Irregular Bundles Experiment.mov
The material and geometric parameters of the lamella array are the same as those in Movie 1 except that the depth of the lamella is of order millimeters. The geometric imperfections of the system is larger than that in Movie 1. The coalescence initiates from multiple sites simultaneously and irregular bundles with size varying from 2 to 5, although the primary mode at the onset is still the dimer. The movie is in real time and the duration is about 34 . s
## Movie 3: 2D Hierarchical Bundles Simulation.mov
A periodic array of 100 plates is simulated, and only 32 plates are displayed. The system starts with θ n = π/ 2, and V /V n flat = 0 92 except that . V 99 is 2% less, where V flat is the control volume inside the system when all the pillars are vertical and the air-liquid interface is flat. The coalescence initiates from the 99 th cell and propagates in both directions. Dimers arise first and further collapse into quadrimers right after. The bundles separate in the reverse
order as the liquid evaporates. The system recovers the configuration of uniform vertical lamellae in the end as no adhesion forces are included in the theoretical model. In order to see the onset of instability and hierarchical transitions clearly, the movie is not uniform in time, and the dimensionless time is included in the movie. The time scale is τ = 22 5 . ms , and the entire process completes in around 7 s in real time.
## Movie 4: 2D Irregular Bundles Simulation.mov
A periodic array of 100 plates is simulated, and only 31 plates are displayed. The initial condition is the uniform random perturbations with the maximum amplitude of 5% applied on both V /V n flat = 0 92 and . θ n = π/ 2, where V flat is the control volume inside the system when all the pillars are vertical and the air-liquid interface is flat. The coalescence initiates from multiple sites simultaneously. The dominant onset mode is still the dimer. During the process of evaporation, irregular bundles of size varying from 2 to 5 form and eventually separate. In order to see the onset of instability and early stages of coalescence clearly, the movie is not uniform in time, and the dimensionless time is included in the movie. The time scale is τ = 22 5 . ms , and the entire process completes in around 7 s in real time.
## Movie 5: 3D 14x14 V090.mov
It shows the simulated evolution of an array of 14 by 14 pillars collapsing into fourfold bundles for a prescribed liquid volume of V/V flat = 0 90, where . V flat is the control volume inside the system when all the pillars are vertical and the air-liquid interface is flat, using the parameters from experiments in Figure 1(c)-(e). The final stable configuration at the end of this movie corresponds to Figure 7(a).
## Movie 6: 3D 10x16 V088.mov
It shows the simulated evolution of an array of 10 by 16 pillars collapsing into fourfold bundles for a prescribed liquid volume of V/V flat = 0 88, where . V flat is the control volume inside the system when all the pillars are vertical and the air-liquid interface is flat, using the parameters from experiments in Figure 1(c)-(e). The final stable configuration at the end of this movie corresponds to Figure 7(b).
[1] S. P. Timoshenko and J. M. Gere. Theory of Elastic Stability, 2 ed., McGraw-Hill, 1961, p46.
- [2] A. Mata, A. J. Fleischman and S. Roy, Characterization of Polydimethylsiloxane (PDMS) Properties for Biomedical Micro/Nanosystems. Biomedical Microdevices 2005 , 7, 281-293.
- [3] A. Grinthal, S.H. Kang, A.K Epstein, M. Aizenberg, M. Khan, and J. Aizenberg. Steering nanofibers: An integrative approach to bio-inspired fiber fabrication and assembly. Nano Today 2011 , 7, 35-52.
- [4] H. Lamb. On the uniform motion of a sphere through a viscous fluid. Phil. Mag. 1911 , 21, 112.
- [5] A.S. Sangani and A. Acrivos. Slow flow past periodic arrays of cylinders with application to heat transfer. Int. J. Multiphase Flow 1982 , 8, 193.
|
10.1098/rspa.2014.0593
|
[
"Zhiyan Wei",
"Tobias Schneider",
"Jungchul Kim",
"Ho-Young Kim",
"Joanna Aizenberg",
"L. Mahadevan"
] |
2014-08-04T17:21:30+00:00
|
2014-08-04T17:21:30+00:00
|
[
"cond-mat.soft"
] |
Elastocapillary coalescence of plates and pillars
|
When a fluid-immersed array of lamellae or filaments that is attached to a
substrate is dried, evaporation leads to the formation of menisci on the tips
of the plates or pillars that bring them together. Similarly, when hair dries
it clumps together due to capillary forces induced by the liquid menisci
between the flexible hairs. Building on prior experimental observations, we use
a combination of theory and computation to understand the nature of this
instability and its evolution in both the two-dimensional and three-dimensional
setting of the problem. For the case of lamellae, we explicitly derive the
interaction torques based on the relevant physical parameters. A Bloch-wave
analysis for our periodic mechanical system captures the critical volume of the
liquid and the 2-plate-collapse eigenmode at the onset of instability. We study
the evolution of clusters and their arrest using numerical simulations to
explain the hierarchical cluster formation and characterize the sensitive
dependence of the final structures on the initial perturbations. We then
generalize our analysis to treat the problem of pillar collapse in 3D, where
the fluid domain is completely connected and the interface is a surface with
the uniform mean curvature. Our theory and simulations capture the salient
features of both previous experimental observations and our own in terms of the
key parameters that can be used to control the kinetics of the process.
|
1408.0749v1
|
## Long-period surface motion of the multi-patch Mw9.0 Tohoku-Oki earthquake
Panos A. Psimoulis 1,2* , Nicolas Houlié 2,3 , Clotaire Michel 4 , Michael Meindl and Markus 2 Rothacher 2
- 1 Nottingham Geospatial Institute, Dept. of Civil Engineering, The University of Nottingham,
Nottingham, NG7 2TU, UK
- 2 Geodesy and Geodynamics Lab., Institute of Geodesy and Photogrammetry, ETH Zurich, Zurich 8093, Switzerland
- 3 Seismology and Geodynamics, Institute of Geophysics, ETH Zurich, Zurich 8092, Switzerland
4 Swiss Seismological Service, ETH Zurich, Zurich 8092, Switzerland
## Abstract
We show that it is possible to capture the oscillatory ground motion induced by the TohokuOki event for periods ranging from 3 to 100s using Precise Point Positioning (PPP). We find that the ground motions of the sedimentary basins of Japan were large (respectively > 0.15m/s and >0.15m/s 2 for velocity and acceleration) even for periods larger than 3s. We compare geodetic observables with a Ground Motion Prediction Equation (GMPE) designed for Japan seismicity and find that the Spectral Acceleration (SA) is well estimated for periods larger than 3s and distances ranging from 100 to 500km. At last, through the analysis of the displacement attenuation plots, we show that the 2011 Tohoku-Oki event is likely composed of multiple rupture patches as suggested before by time-reversal inversions of seismic data. Keywords : Seismology, Geodesy, strong-motion, GPS, Tohoku-Oki, PPP, long-period ground motion, GMPE
* Corresponding author: Assistant Professor in Geospatial Eng., Tel: +44 (0) 115 951 4287, Fax: +44 (0) 115 951 3881, E-mail: [email protected]
## 1. Introduction
For solid Earth investigation, high-rate GPS is nowadays an essential tool in the study of transient deformation, seismic source at depth, or accurate determination of strain rate and surface velocity fields. In the field of engineering, high-rate GPS is used to determine the deformation or the motion of structures (bridges, high buildings, etc.) in time (Çelebi and Sanli, 2002; Psimoulis et al., 2008). Koketsu and Miyake (2008) showed that long-period ground motion is also of interest for the increasing number of large civil engineering structures such as bridges and tall buildings since far-distance resonance and short-distance directivity could occur. In the current study we investigate whether GPS is reliable and accurate in the determination of oscillatory ground motions (acceleration, velocity, displacement) for frequencies lower than or equal to 0.33 Hz.
Peak ground accelerations and velocities (PGA, PGV) are routinely computed using seismic data in near real-time (minutes or less) in the 0.5 to 25Hz frequency range (typically) in order to assess the possible consequences of an earthquake shortly after its occurrence. The relationship between magnitude and peak ground motions of an event is distorted by the interaction between seismic waves and the structures (sedimentary basins, fault systems, etc.) sampled during their travel from the source (e.g. Roten et al., 2011 or Denolle et al., 2013). Ground Motion Prediction Equations (GMPEs) predict that the peak motions representative of longer periods, such as the spectral acceleration (SA) at 3s depend proportionally more on the event's magnitudes than PGA or PGV (e.g. Zhao at al., 2006). Nevertheless, few GMPEs for peak motions representative of long-periods are available in the literature due to the limited amount of broadband data (Cauzzi and Faccioli, 2008). For this reason, peak ground displacements (PGDs) are generally not available and proxies like the Displacement Response Spectrum DRS (10s) are used (Cauzzi and Faccioli, 2008). We explore whether GPS can
contribute to GMPEs but also if the seismic source can be better characterized using longperiod surface oscillatory motions.
As seismic waveforms, GPS time-series provide insights into surface transients observed during seismic wave propagation or during a seismic rupture (Larson, 2001; Bock et al., 2004; Ji et al., 2004; Miyazaki et al., 2004; Blewitt et al., 2006; Kobayashi et al., 2006; Ohta et al., 2006; Emore et al., 2007; Wang et al., 2007; Yokota et al., 2009; Delouis et al., 2010; Feng et al., 2010; Houlié et al., 2011, Bock et al., 2011; Avallone et al., 2011; Ohta et al., 2012; Mitsui and Heki, 2012 Wright et al., 2012; Meng et al., 2013; Guo et al., 2013; Houlié et al., 2014, Psimoulis et al., 2015), tsunami generation for early warning (Blewitt et al., 2006; Sobolev et al., 2007; Blewitt et al., 2009; Crowell et al., 2009; Behrens et al., 2010; Ohta et al., 2012; Li et al., 2013), changes in ionosphere status (Ducic et al., 2003; Mai and Kiang, 2009; Karia et al., 2012; Galvan et al., 2012), volcanic plumes (Houlié et al., 2005; Houlié et al., 2006; Newman, et al., 2012), water vapour delays (Perler et al., 2011; Hurter et al., 2012), or moment tensor (O'Toole et al., 2012; O'Toole et al., 2013). However, the studies listed above did not make use of real-time or near real-time data processing strategies. Precise Point Positioning (PPP) allows to monitor the motion of standalone GPS sites (Blewitt, 2008; Ge et al., 2008; Jokinen et al., 2013). GPS, which is accurate up to a few millimetres for highfrequency oscillations in differential mode (Psimoulis and Stiros, 2008; Psimoulis and Stiros, 2012), is accurate at the centimetre level in PPP mode (Moschas et al., 2014), accuracy which is sufficient to determine the surface displacement (Figure 1) following large events (Mw > 7.0). The PPP GPS time-series have not been used until today as a tool to describe the surface oscillatory motion. In this study we propose to use PPP time-series processed for the GEONET continuous GPS (CGPS) sites to study the ground motions that followed the 2011 Tohoku-Oki event.
The Tohoku-Oki event (origin time T at 05:46:23 UTC on March 11, 2011) and the 0 subsequent tsunami stroke Japan with a heavy impact on structures. The rupture area is exceptionally compact for such magnitude (~Mw9.0 and L~500km, to be compared with the 1000 km of rupture of the 2004 Sumatra event with a similar magnitude, Vigny et al., 2005) with most of the slip contained above 30 km depth and a maximum slip on the fault of nearly 50m (Koketsu et al., 2011; Yagi et al., 2011; Suzuki et al., 2011; Yue and Lay, 2011). Because of the poor coverage near the epicentre, the slip distribution near the surface is not well constrained. Other studies (Maercklin et al., 2012; Roten et al., 2012) confirmed that the long-period (10-20s) maximum slip was up to 50m but also suggested that the Tohoku-Oki event was indeed composed of multiple ruptures of smaller magnitudes (<Mw8.8).
## 2. Data
## 2.1 GPS data
The Mw9.0 earthquake off the Pacific coast of Tohoku-Oki on March 11, 2011 was fully recorded by the GEONET network (Sagiya, 2004), operated by the Geospatial Information Authority of Japan (GSI). The GEONET network is composed of over 1200 continuously observing GPS receivers (with an average spacing of 20km). We processed 1Hz GPS records from 847 GEONET stations of 15-hour duration, fully covering the earthquake period. The data was analysed with the Bernese GPS Software (Dach et al., 2007). The data were postprocessed in a PPP mode using state-of-the-art models and a-priori information of highest quality from the Center for Orbit Determination in Europe (Dach et al. 2009, Bock et al., 2009). As an example for the results obtained, Figure 2 shows the displacement field at 300 seconds after the origin time (T ). The displacement time series in North, East, and vertical 0 components were established separately for each station. The GPS displacement time series
show an a-posteriori formal accuracy of about 1cm in the horizontal and 2cm in the vertical component, respectively (Psimoulis et al., 2015).
## 2.2 Strong motion data
The largest strong-motion networks of Japan called K-NET (Kyoshin network) and KiK-net (Kiban-Kyoshin network) consist of 1034 and 660 stations, respectively (Aoi et al., 2011). The main difference of the two networks is that the K-NET stations are located mainly on thick sedimentary sites, while the KiK-net stations are deployed on rock or thin sedimentary sites, as it is a sub-net of the Hi-net (High-sensitivity seismograph network) primarily designed for highly sensitive seismic observations. Consequently, the K-NET and KiK-net triggering thresholds are different (2 cm/s and 0.2 cm/s , Aoi et al., 2004). Furthermore, the 2 2 K-NET stations are installed on the ground surface, while each KiK-net site hosts two instruments, one installed at the surface and the second at the bottom of a borehole of 100200m depth.
The Mw9.0 earthquake of Tohoku-Oki 2011 was well recorded by the two strong-motion networks. Raw data of strong-motion records from 700 K-NET sites and 525 KiK-net sites were available. The records of three components, corresponding to the North, East and vertical directions, were sampled at 100Hz and up to 300s duration. The acceleration of each site was derived by correcting the raw data for the gain, converting the corresponding record time from UTC to GPS time and then by applying the time correction for 15 leap seconds. In order to be consistent with the recorded motion of the GPS and K-NET sites, only KiK-net surface records were investigated in this study.
## 3. Results
## 3.1 Consistency of GPS and strong-motion data
The evaluation of the consistency of GPS and strong-motion records at high frequencies must be completed for sites located very closely. In previous studies, the GPS and strong-motion sites with an inter-distance of less than 1km were assumed as collocated (Melgar et al., 2013). In our study, we used an even more conservative approach by using only the very closely located GPS and strong-motion sites, with an inter-distance less than 100m. In total, we identified 29 very closely located sites of the GPS and strong-motion networks (Figure 3), comprising 21 K-NET and 8 KiK-net sites (Table 1).
For these sites we assume that the GPS and strong-motion records are not shifted by a significant time difference (<1s), keeping in mind that the sampling rate (1Hz) of the GPS network is low compared to the sampling rate of the strong-motion network (100Hz).
The evaluation of the consistency of the GPS and the strong-motion sensors was examined by comparing the spectra of the two records and assessing them in their common frequency range (0-0.5Hz). The usual technique of integration of strong-motion records for the computation of velocities and displacements was not followed due to the significant accumulated error (Stiros, 2008), the large duration of the strong-motion records (up to 300s) and the significant coseismic displacement (up to 2-3m), which would make the estimation of the long-period displacement time series unreliable (Boore, 2003).
Thus, instead of integrating the strong-motion records, the GPS displacement series were differentiated twice resulting in acceleration time series. The derived accelerations are affected also by errors due to the differentiation, which, however, do not accumulate. These differentiation errors affect mainly the short-period acceleration (2-3s), due to the high shortperiod noise of GPS measurements relatively to the short time step of the differentiation. Furthermore, the error of the derived accelerations is expected to be higher for the vertical component due to the higher noise level of the corresponding PPP results.
In Figure 4, we present the spectra of the GPS and the K-NET acceleration time series of the East-West (EW), North-South (NS), and Up-Down (UD) components for two representative closely located sites, one close (179km; Figure 4a-c) and one far (771km; Figure 4d-f) from the epicentre. In addition, the spectra of the GPS acceleration time series, corresponding to the time interval before the seismic event, are shown in Figure 4 as black lines. These spectra provide information on the basic noise of the acceleration time series derived from GPS. For periods above 3-4 seconds, the GPS noise is lower than the seismic signal for the horizontal component. This supports the consistency of GPS and K-NET spectra (Figure 4a-c) mainly for periods above 3 seconds, while KiK-net strong-motion spectra prove to be slightly less consistent (Figure 4d-f) for the periods below 5 seconds. This is the case in general for sites far from the epicentre (>600 km) where the seismic signal is weak relative to the PPP noise and the S/N decreases, especially for short-periods (<5s) and for the vertical component (period <10s). However, such phenomena do not affect the current study, as it is focusing on the use of the horizontal components and long-periodic oscillations (>3 s), which usually are not included in the ground motion parameters computation.
## 3.2 Computation of MGA, MGV and MGD
The Maximum Ground Acceleration (MGA), Velocity (MGV), and Displacement (MGD) were computed from GPS and strong-motion records for short- (3-10s) and long-period (10100s) domains. The strong-motion records were integrated once and twice into velocities and displacements and the GPS records were differentiated once and twice to obtain velocities and accelerations. The derived strong-motion and GPS time series were filtered for a short- (310s) and a long-period (10-100s) band. Figure 5 shows the acceleration time series of the GPS and strong-motion instruments that are closely located for the two period bands. For each type of sensor, we compute the absolute maxima (corresponding to MGA, MGV, and MGD) of
acceleration, velocity, and displacement time series of every site of each network. Figures 6a, 6b and 6c show the MGA, MGV and MGD maps for the two period bands. As expected, by comparing the GPS and the strong-motion networks, GPS seems to be more sensitive to longperiod ground displacements and the strong-motion sensors to short-period accelerations.
The noise of the GPS acceleration, velocity, and displacement time series was estimated by analysing the time interval before the seismic event, where all the sites are assumed not to be affected by motions. Following the methodology of MGA, MGV and MGD, the GPS noise displacement time series were differentiated once and twice and subsequently filtered for the two period bands (3-10s and 10-100s). The maximum value of each resulting time series expresses the maximum noise level for each GPS site and for each quantity (acceleration, velocity, displacement). In Table 2 the mean average and the standard deviation of the estimated maximum noise values for the horizontal component and for the two examined period bands are listed. All the estimated MGA, MGV, and MGD values of the GPS sites are above their respective uncertainty maxima, with the latter computed as the mean maximum error+3 σ zone (2.0cm/s , 1.5cm/s, 1.0cm, Figure 7). The level of the estimated maximum 2 noise is smaller than that in other studies (e.g. Wright et al., 2012), due to the limitation of the noise level by filtering the GPS time series for specific period bands.
The qualitative examination of Figure 6 shows that strong-motion and GPS provide a consistent estimation of MGA, MGV and MGD for the two examined period domains. Some local differences for the short-period band might reflect weaknesses of GPS in estimating acceleration of low periods (<4-5s) or local effects, without disturbing though the overall compatibility of the MGA maps.
At long-periods, the maximum displacement map (Figure 6c, right) is smooth and reflects the distance to the largest slip patch on the fault: the maximum values occur close to the epicentre in the Sendai region with large values along the coast to the South. However, in the 3-10s
band (Figure 6b, c, left), the map is more complex and highlights large sedimentary basins. In addition to the Sendai basin, close to the epicentre, the largest values are found in the Kanto plain, known to exhibit resonance frequencies in this band (Yamanaka et al., 1989). On the western coast, large values can also be found at the Sakata and Niigata basins, in relation with the large amplification in these frequency band as shown by Mamula et al. (1984).
In Figure 8, we show the differences of the estimated GPS and strong-motion MGA, MGV and MGD for the very closely located sites, normalised to the maximum acceleration, velocity and displacement of the corresponding site. The derived normalised differences show the good agreement between the solutions of GPS and the strong-motion sensors (relative errors mostly below 10%) with the exception the short-period (3-10sec) MGA and MGD differences. The latter are observed mainly for large distances from the epicentre and they are caused by the relatively small oscillatory amplitude of the corresponding accelerations and displacements (up to 0.15m/s and 0.10m, respectively). This can be seen more clearly from 2 the absolute differences of the estimated MGA, MGV and MGD between the GPS and strongmotion sensors in Figure A1, which reveals the small values (<0.03cm/s , <0.03cm/s and 2 0.05cm, respectively) for the sites located mainly far from the epicentre. The relatively high differences in the displacement (>3cm) of some sites far from the epicentre (K-NET GIF006 GPS 0280, K-NET ISK015 - GPS 0972, K-NET HKD043 - GPS 0502 and K-NET HKD046 GPS 0504) may be attributed to local effects.
Since we find a general agreement of the GPS records with the seismic data in space and frequency, we compute the spectral acceleration (SA) at 3s for a 5% damping (Figure 9). In general the long-period signals are not of concern for the safety of typical building or small objects at the surface. However, a large object such as a bridge, a skyscraper or a deep basin might resonate at long-periods, particularly in the 3-10s range (Anderson et al., 1985; Singh et al. 1988; Lermo et al., 1994; Chávez-García and Bard, 1994). Even more rigid structures
might be affected by these long-periods due to site effects, soil-structure interaction and liquefaction (Margaris et al., 2010; Assimaki et al., 2012). In order to have a baseline in the current study, the attenuation of ground motion with distance (spectral acceleration and acceleration) is compared to the GMPE proposed for Japan by Zhao et al. (2006). Please note that we use the distance from the epicentre (as computed by USGS) whereas Zhao et al. (2006) assume the shortest distance between a rupture model and the station. We find that SA(3s) estimates from GPS and strong-motion data are compatible with maximum values predicted by the GMPE (Figure 9).
Figure 10 shows the strong-motion and GPS maximum accelerations, velocities and displacements of the two period bands (3-10s and 10-100s) with respect to the distance to the epicentre. The comparison of MGA with GMPE predictions is less relevant as the long-period signals were not used to build the GMPE. For large distances, several values of MGA exceed the estimated PGA predicted by the GMPE. We remind that this discrepancy is however related to small amplitude motions and more interestingly to the response of basins located at distance > 200km. Basins response is however not sufficient to explain the evolution of MGA, MGV and MGD with distance.
Like the accelerations, the maximum displacement estimates depend on the period range. We know that MGD for the short-periods are ~10 times smaller than for long-periods (at ~1m), and we might expect that PGD follows the same trend, so does the PGA. In the case of a single rupture patch, we expect a decay of the form 1/distance in the displacement/distance diagrams. In Figure 10f, displaying long-period oscillatory displacements, the displacement vs. distance curve does not constantly decrease, but also increases with a second peak at a distance of about 250 km from the epicentre. For better visibility, the averaged displacement for each distance is shown using a blue line. Such a pattern for long-periods (10-100s) suggests that the ground motion results from the rupture of two distinct fault segments. The
averaged velocity and acceleration is also computed (blue line, Figure 10d, e) but the feature of increment of the two variables is not so pronounced mainly due to the weak velocity and acceleration signal in long-periods.
Regarding the oscillatory acceleration and velocity, they appear to be constant for the shortperiod (3-10s) and distance up to 250-300km from the epicentre (Figure 10), and then decrease with distance linearly (in a log-log space). This pattern suggests that the strong shortperiod oscillatory component is generated in an extended rupture area, which seems to cover the location of the two different rupture patches, as it was revealed by the long-period displacement. The plateau cannot be seen clearly in the short-period oscillatory displacements due to the relatively small signal of the corresponding displacements.
Stewart et al. (2013) extensively compared GMPEs to the observed ground motion of the Tohoku-Oki event and could not detect the stability of MGA/MGV with increasing distance. Indeed, whereas high-frequency ground motion (intensity measures such as PGA or PGV) is generated by small-scale asperities all along the fault, low-frequency parameters of the ground motion such as displacement is caused by large-scale asperities that may be more localised on the fault as shown by source inversions. These phenomena impact the choice of an optimal distance measure: whereas distance to rupture is the most relevant for high frequencies as used by Stewart et al. (2013), distance to major asperities is better for low frequencies. The possibility of the presence of two sources is critical for correct ground motion estimation. Indeed, as the two sources would be shifted relatively to each other, expected ground motion would be larger at larger distances from the first epicentre.
## 4. Conclusions and outlook
The first conclusion of our work is that both GPS displacement histories and integrated accelerograms are in agreement for the frequency range they have in common (3 to 100s). We
find that long-period oscillatory ground motions are accurately determined when based on GPS data. The double integration of accelerograms does not impact the precise resolution of the derived displacement data for periods less than 100s for a large earthquake such as the Tohoku-Oki event.
In future, the PPP GPS data will be processed in near real-time or even real-time (Temporal Point Positioning; Li et al., 2013), allowing the computation of maximum ground motion to supplement other seismic monitoring systems (Liu et al., 2014; Tu, 2014). Considering that the ground acceleration is expected to be more than 5 times higher at a period of 1s than the signal at a period of 10s, the reliability of the GPS for higher sampling rates (>1Hz) will mostly depend on our capability to maintain the noise of the time series at a level close to cm as suggested by Moschas et al. (2014).
Cauzzi and Faccioli (2008) demonstrated the importance of accurate GMPEs at long-periods and the limited amount of available seismic data at these periods. We propose that the future generations of GMPEs should include more long-period data such as GPS displacement timeseries, in order to better estimate effects of large magnitude (Mw>7) events.
Furthermore, we highlight the response of major sedimentary basins of Japan to the shake that followed the Mw9.0 Tohoku-Oki 2011 earthquake, for periods ranging from 3 to 10 seconds. The increase in oscillatory displacement at ~250 km distance from the epicentre can be interpreted as a second source of displacement along the fault generated by a second asperity as proposed by the source models of Koketsu et al. (2011), Suzuki et al. (2011) and Maercklin et al. (2012). Long-period maximum oscillatory displacement can therefore be used as a constraint for source inversions, providing additional constraints on spatial pattern of rupture at depth.
## Acknowledgments
This study has been supported by Swiss National Fund (SNF) grants in the framework of the project 'High-rate GNSS for Seismology' (number 200021\_130061).
## 6. References
Anderson , J.G., P. Bodin, J.N. Brune, J. Prince, S.K. Singh, R. Quaas & Onãte, M., 1986. Strong ground motion from the Michoacan, Mexico, earthquake, Science, 233 , 1043-1049 Aoi, S., Kunugi, T. & Fujiwara, H., 2004. Strong-motion seismograph network operated by NIED: K-NET and KiK-net, Journal of Japan Association for Earthquake Engineering , 4(3), 65-74.
Aoi, S., Kunugi, T., Nakamura, H. & Fujiwara, H., 2011. Deployment of new strong motion seismographs of K-NET and KiK-net, in Earthquake Data in Engineering Seismology, Geotech. Geol. Earthquake Eng., 14, 167-186.
Assimaki, D., Ledezma, C., Montalva, G.A., Tassara, A., Mylonakis, G. & Boroschek, R., 2012. Site effects and damage patterns, Earthq. Sp ., 28 (1), S55-S74
Avallone, A., Marzario, M., Cirella, A., Piatanesi, A., Rovelli, A., Di Alessandro, C., D'Anastasio, E., D'Agostino, N., Giuliani, R. & Mattone, M., 2011. Very high rate (10 Hz) GPS seismology for moderate-magnitude earthquakes: The case of the Mw 6.3 L'Aquila (central Italy) event, J. Geophys. Res ., 116(B2), B02305.
Behrens, J., A. Androsov, A. Babeyko, S. Harig, F. Klaschka & L. Mentrup, 2010. A new multi-sensor approach to simulation assisted tsunami early warning, Nat. Hazards Earth Syst. Sci., 10 (6), 1085 Ð 1100.
Blewitt, G., Kreemer, C., Hammond, W.C., Plag, H.-P., Stein, S. & Okal, E., 2006. Rapid determination of Earthquake magnitude using GPS for Tsunami Warning Systems. Geophys. Res. Lett. 33, L11309.
Blewitt, G., 2008. Fixed point theorems of GPS carrier phase ambiguity resolution and their application to massive network processing: Ambizap, J. Geoph. Res. B: Solid Earth , 113(12), B12410.
Blewitt, G., Hammond, W.C., Kreemer, C., Plag, H.-P., Stein, S. & Okal, E., 2009. GPS for real-time earthquake source determination and tsunami warning systems, J. Geod ., 83(3-4), 335-343.
Bock, H., Dach, R., Jäggi, A. & Beutler, G., 2009. High-rate GPS clock corrections from CODE: Support of 1Hz applications, J. Geod ., 83(11), 1083-1094.
Bock, Y., Melgar, D. & Crowell, B., 2011. Real-Time Strong-Motion Broadband Displacements from Collocated GPS and Accelerometers, Bull. Seismol. Soc. Am ., 101, 2904-2925.
Bock, Y., Prawirodirdjo, L. & Melbourne, T., 2004. Detection of arbitrary large dynamic ground motions with a dense high-rate GPS network, Geophys. Res. Lett . 31(L06604).
Boore, D.M., 2003. Analog-to-digital conversion as a source of drifts in displacements derived from digital recording of ground acceleration, Bull. Seism. Soc. Am. , 93 (5), 20172024.
Cauzzi, C., & Faccioli, E., 2008. Broadband (0.05 to 20 s) prediction of displacement response spectra based on worldwide digital records. J. Seism. , 12 (4), 453-475.
Chávez-García, F.J. & Bard, P.-Y., 1994. Site effects in Mexico City eight years after the September 1985 Michoacan earthquakes, Soil Dynamics and Earthquake Engineering , 13 , 229-247.
Çelebi, M. & Sanli, A. 2002. GPS in pioneering dynamic monitoring of long-period structures, Earthq. Spec ., 18(1), 47-61.
Crowell, B.W., Bock, Y. & Squibb, M.B., 2009. Earthquake early-warning using total displacement waveforms from real-time GPS networks, Seism. Res. Lett ., 80(5), 772-782.
Dach, R., Hugentobler, U., Meindl, M. & Fridez, P., 2007. The Bernese GPS Software Version 5.0, 1 edn st , Astronomical Institute, University of Bern, Switzerland.
Dach, R., Brockmann, E., Schaer, S., Beutler, G., Meindl, M., Prange, L., Bock, H., Jäggi, A. & Ostini, L., 2009. GNSS processing at CODE: status report, J. Geod ., 83(3-4), 353-366.
Delouis, B., Nocquet, J.-M. & Vallée, M., 2010. Slip distribution of the February 27, 2010 Mw = 8.8 Maule Earthquake, central Chile, from static and high-rate GPS, InSAR, and broadband teleseismic data, Geophys. Res. Lett . 37, L17305.
Denolle, M. A., Dunham, E.M., Prieto, G.A. and Beroza, G.C. 2013. Ground motion prediction of realistic earthquake sources using the ambient seismic field, J. Geophys. Res. Solid Earth , 118 (5), 2102-2118, 10.1029/2012JB009603
Ducic, V., Artru, J. & Lognonné, P., 2003. Ionospheric remote sensing of the Deanli Earthquake Raleigh surface waves, Geoph. Res. Lett. , 30(18), 1951
Emore, G., Haase, J., Choi, K., Larson, K. & Yamagiwa, A. 2007. Recovering absolute seismic displacements through combined use of 1-Hz GPS and strong motion accelerometers, Bull. Seismol. Soc. Am . 97, 357-378.
Feng, L., Newman, A.V., Farmer, G.T., Psimoulis, P. & Stiros, S. 2010. Energetic rupture, coseismic and post-seismic response of the 2008 Mw6.4 Achaia-Elia Earthquake in northwestern Peloponnese, Greece: An indicator of an immature transform fault zone, Geoph. J. Int., 183(1), 103-110.
Galvan, D.A., Komjathy, A., Hickey, M.P., Stephens, P., Snively, J., Tony Song, Y., Butala, M.D. & Manucci, A.J., 2012. Ionospheric signatures of Tohoku-Oki tsunami of March 11, 2011: Model comparisons near the epicenter, Rad. Sc. , 47(4), RS4003
Ge, M., Gendt, G., Rothacher, M., Shi, C. & Liu, J., 2008. Resolution of GPS carrier-phase ambiguities in Precise Point Positioning (PPP) with daily observations, J. Geod ., 82(7): 389399.
Guo, A., Wang, Y., Li, Z., Ni, S., Wu, W., Liu, G., Zheng, Y. & Simons, M., 2013. Observation of core phase ScS from the mw 9.0 Tohoku-oki earthquake with highrate GPS, Seism. Res. Lett ., 84(4), 594-599.
HouliŽ, N., Briole, P., Nercessian, A. & Murakami, M., 2005, Volcanic plume above Mount St. Helens detected with GPS, EOS , 86(30), 277, 10.1029/2005EO300001.
HouliŽ, N., Briole, P., Bonforte, A. & Puglisi, G., 2006, Large scale ground deformation of Etna observed by GPS between 1994 and 2001, Geoph. Res. Lett ., 339(2), L02309, 10.1029/2005GL024414.
HouliŽ, N., Occhipinti, G., Shapiro, N., LognonnŽ, P. & Murakami, M., 2011, New approach to detect seismic surface waves in 1Hz-sampled GPS time series. Sci. Rep . 1(44), 10.1038/srep00044.
HouliŽ, N., Dreger, D. & Kim, A., 2014. GPS source solution of the 2004 Parkfield earthquake. Sci. Rep. 4 , 3646, 10.1038/srep03646 .
Hurter, F., Geiger, A., Perler, D. & Rothacher, M., 2012, GNSS water vapor monitoring in the Swiss Alp, Int. Geosc. Rem. Sens. Symp. (IGARSS) , 2012 IEEE International, 10.1109/IGARSS.2012.6351115, 1972-1975.
- Ji, C., Larson, K., Tan, Y., Hudnut, K., & Choi, K., (2004). Slip history of the 2003 San Simeon earthquake constrained by combining 1-Hz GPS, strong motion, and teleseismic data. Geophys. Res. Lett . 31, L17608.
Jokinen, A., Feng, S., Schuster, W., Ochieng, W., Hide, C., Moore, T. & Hill, C., 2013. Integrity monitoring of fixed ambiguity Precise Point Positioning (PPP) solutions, Geosp. Inf. Sc ., 16(3), 141-148.
Karia, S., Sarkar, S. & Pathak, K., 2012. Analysis of GPS-based TEC and electron density by the DEMETER satellite before the Sumatra earthquake on 30 September 2009, Int. J. Rem. Sens., 33(16), 5119-5134.
Kobayashi, R., Miyazaki, S. & Koketsu, K., 2006. Source processes of the 2005 West Off Fukuoka Prefecture earthquake and its largest aftershock inferred from strong motion and 1Hz GPS data. Ear. Pl. Sp ., 58, 57-62.
Koketsu, K. & Miyake, H., 2008. A seismological overview of long-period ground motion. J. Seism. , 12(2), 133-143.
Koketsu, K., Yokota, Y., Nishimura, N., Yagi, Y., Miyazaki, S., Satake, K., Fujii, Y., Miyake, H., Sakai, S., Yamanaka, Y. & Okada, T., 2011. A unified source model for the 2011 Tohoku earthquake, Ear. Plan. Sc. Lett., 310, 480-487.
Larson, K., et al., Cervelli, P., Lisowski, M., Miklius, A., Segall, P. & Owen, S., 2001. Volcano monitoring using the Global Positioning System: Filtering strategies. J. Geophys. Res . 106(B9), 19,453-19,464.
Lee, S.-J., Huang, B.-S., Ando, M., Chiu, H.-C. & Wang, J.-H., 2011. Evidence of large scale repeating slip during the 2011 Tohoku-Oki earthquake, Geophysical Research Letters , 38, 19, L19306
Lermo, J. & Chavez-Garcia, F.J. 1994. Site effect evaluation at Mexico City: dominant period and relative amplification from strong motion and microtremor records, Soil Dyn. Earthq. Eng ., 13, 413-423.
- Li, X., Ge, M., Zhang, Y., Wang, R., Xu, P., Wickert, J. & Shuh, H., 2013. New approach for earthquake/tsunami monitoring using dense GPS networks. Sci. Rep . 3.
- Li, X., Ge, M., Guo, B., Wickert, J. & Schuh, H. 2013. Temporal Point Positioning approach for real-time GNSS seismology using a single receiver, Geoph. Res. Lett ., 40(21), 56775682
- Liu, Z., Owen, S. & Moore, A., 2014. Rapid estimate modeling of permanent coseismic displacements for large earthquakes using high-rate global positioning system data, Seism. Res. Lett ., 85(2), 284-294
Maercklin, N., Festa, G., Colombelli, S. & Zollo, A., 2012. Twin ruptures grew to built up the giant 2011 Tohoku, Japan, earthquake, Sci. Rep ., 709(2), 1-7.
Margaris, B., Athanasopoulos, G., Mylonakis, G., Papaioannou, C., Klimis, N., Theodulidis, N., Savvaidis, A., Efthymiadou, V. & Stewart, J.P., 2010. The 8 June 2008 Mw6.5 AchaiaElia, Greece earthquake: Source characteristics, ground motions, and ground failure, 26(2), 399-424.
Mai, C.-L. & Kian, J.-F., 2009. Reconstruction of ionospheric perturbation induced by 2004 sumatra tsunami using computerized tomography technique, IEEE Trans. Geosc. Rem. Sens ., 47(10), 4, 3303-3312
Mamula, L., Kudo, K., & Shima, E., 1984. Distribution of Ground-Motion Amplification Factors as a Function of Period (3-15sec), in Japan. Bull Earthq Res Inst , 59, 467-500
Melgar, D., Brendan, W.C., Bock, Y, Haase, J.S. 2013. Rapid modelling of the 2011 Mw 9.0 Tohoku-oki earthquake with seismogeodesy, Geoph. Res. Lett ., 40, 2963-2968
Meng, G., Ren, J., Su, X., Yang, Y., Zhu, Z., Ge, L. & Li, X., 2013. Coseismic Deformation of the 2010 Mw 6.9 Yushu Earthquake Derived from GPS Data. Seis. Res. Lett . 84(1).
Mitsui, Y. & Heki, K., 2012. Observation of Earth's free oscillation by dense GPS array: After the 2011 Tohoku megathrust earthquake. Sci. Rep. 2(931).
Miyazaki, S., Larson, K.M., Choi, K.H., Hikina, K., Koketsu, K., Bodin, P., Haase, J., Emore, G. & Yamagiwa, A. 2004. Modeling the rupture process of the 2003 September 25 TokachiOki (Hokkaido) earthquake using 1-Hz GPS data. Geophys. Res. Lett. 31, L21603.
Moschas, F., Avallone, A., Saltogianni, N. & Stiros, S. 2014. Strong-motion displacement waveforms using 10-Hz precise point positioning GPS: an assessment based on free oscillation experiments, Earthq. Eng. Str. Dyn., doi: 10.1002/eqe.2426
Newman, A.V., Stiros, S., Feng, L., Psimoulis, P., Moschas, F., Saltogianni, V., Jiang, Y., Papazachos, C., Panagiotopoulos, D., Karagianni, E. & Vamvakaris, D., 2012. Recent
geodetic unrest at Santorini Caldera, Greece, Geoph. Res. Lett ., Vol. 39(6), L06309 Ohta, Y., Kobayashi, T., Tsushima, H., Miura, S., Hino, R., Takasu, T., Fujimoto, H., Iinuma, T., Tachibana, K., Demachi, T., Sato, T., Ohzono, Mako, Umino, N., 2012. Quasi real-time fault model estimation for near-field tsunami forecasting based on RTK-GPS analysis: Application to the 2011 Tohoku-Oki earthquake (M 9.0), w J. Geoph. Res ., 117, B02311 Ohta, Y., Meiano, I., Sagiya, T., Kimata, F. & Hirahara, K., 2006. Large surface wave of the 2004 Sumatra-Andaman earthquake captured by the very long baseline kinematic analysis of 1-Hz GPS data. Ear. Pl. Sp ., 58, 153-157.
Ohta, Y., Kobayashi, T., Tsushima, H., Satoshi, M., Ryota, H., Tomoji, T., Hiromi, F., Takeshi, I., Kenji, T., Tomotsugu, D., Sato, T., Ohzono, M. & Umino, N. 2012. Quasi realtime fault model estimation for near-field tsunami forecasting based on RTK-GPS analysis: Application to the 2011 Tohoku-Oki earthquake (Mw9.0), J. Geoph. Res. B: Sol. Ear ., 117, B02311
- O'Toole, T.B., Valentine, A.P. & Woodhouse, J.H., 2012. Centroid-moment tensor inversions using high0rate GPS waveforms, Geoph. J. Int., 187(3), 1516-1536
O'Toole, T.B., Valentine, A.P. & Woodhouse, J.H., 2013. Earthquake source parameters from GPS-measured static displacements with potential for real-time application, Geoph. Res. Lett., 40(1), 60-65
Perler, D., Geiger, A. & Hurter, F., (2011), GPS water vapour tomography, New parameterized approaches, J. Geod ., 85(8), 539-550
Psimoulis, P. & Stiros, S., 2008. Experimental assessment of the accuracy of GPS and RTS for the determination of the parameters of oscillation of major structures, Computer-Aided Civil and Infrastructure Engineering, 23(5), 389-403.
Psimoulis, P., Pytharouli, S., Karambali, D. & Stiros, S., 2008. Potential of Global Positioning System (GPS) to measure frequencies of oscillations of engineering structures, J. Sound and
Vibr ., 318(3), 606-623
Psimoulis, P. & Stiros, S., 2012. A supervised learning computer-based algorithm to derive the algorithm of oscillations of structures using noisy GPS and Robotic Theodolites (RTS) records, Comp. Str ., 92-93, 337-348.
Psimoulis, P., Houlié, N., Meindl, M. and Rothacher, M. (2015), Consistency of GPS and strong-motion records: case study of Mw9.0 Tohoku-Oki 2011 earthquake, Smart Structures and Systems, 26, 2 (accepted)
Rhie, J., Dreger, D.S., Murray, M., Houlié , 2009. Peak ground velocity ShakeMaps derived from geodetic slip models, Geoph. J. Int ., 179 (2), 1105-1112
Roten, D., Olsen, K.B., Pechmann, J.C., Cruz-Atienza, V.M., Magistrale, H. 2011. 3D simulations of M7 earthquakes on the Wasatch fault, Utah, Part I: Long-period (0-1Hz) groung motion, Bull. Seism. Soc. Am ., 101(5), 2045-2063
Roten, D., Miyake, H. & Koketsu, K., 2012. A Rayleigh wave back-projection method applied to the 2011 Tohoku earthquake, Geophys. Res. Lett., 39 , L02302, doi:10.1029/2011GL050183
Sagiya, T., 2004. A decade of GEONET: 1994-2003 - The continuous GPS observation in Japan and its impact on earthquake studies, Ear. Pl. Sp. , 56, xxix-xli.
Singh, S.K., Mena, E. & Castro, R., 1988. Some aspects of source characteristics of the 19 September 1985 Michoacan earthquake, and ground motion amplifications in and near Mexico City from strong motion data. Bull. Seism. Soc. Am ., 78 , 451-477
Sobolev, S. V., Babeyko, A.Y., Wang, R., Hoechner, A., Galas, R., Rothacher, M., Sein, D.V., Schröter, J., Lauterjung, J. & C. Subarya 2007. Tsunami early warning using GPS-Shield arrays, J. Geophys. Res., 112, B08415, doi:10.1029/2006JB004640.
Stewart, J. P., Midorikawa, S., Graves, R. W., Khodaverdi, K., Kishida, T., Miura, H. & Campbell, K. W. 2013. Implications of the M w 9.0 Tohoku-Oki Earthquake for Ground
Motion Scaling with Source, Path, and Site Parameters. Earthq. Sp. , 29 (S1), S1-S21. doi:10.1193/1.4000115
Stiros, S., 2008. Errors in velocities and displacements deduced from accelerographs: An approach based on the theory of error propagation. Soil Dyn. Earthq. Eng ., 28 , 415-420.
Suzuki, W., Aoi, S., Sekiguchi, H. & Kunugi, T., 2011. Rupture process of the 2011 TohokuOki megathrust earthquake (M9.0) inverted from strong-motion data, Geoph. Res. Let ., 38: L00G16.
Tu, R., 2014. Fast determination of displacement by PPP velocity estimation, Geoph. J. Int ., 196 (3), 1397-1401
Vigny, C., Simons, W. J. F., Abu, S., Bamphenyu, R., Satirapod, C., Choosakul, N., Subarya, C., Socquet, A., Omar, K., Abidin, H. Z. & Ambrosius, B. A. C., 2005. Insight into the 2004 Sumatra-Andaman earthquake from GPS measurements in southeast Asia, Nature, 436, 201206
Wang, G., Boore, D. M., Tang, G. & Zhou, X. 2007. Comparisons of ground motions from collocated and closely spaced one-sample-per-second Global Positioning System and accelerograph recordings of the 2003M6.5 San Simeon, California, earthquake in the Parkfield Region. Bull. Seismol. Soc. Am . 97(1B), 76-90.
Wright, T., Houlié, N., Hildyard, M. & Iwabuchi, T., 2012. Real-time, reliable magnitude for large earthquakes from 1Hz GPS precise point positioning: The 2011 Tohoku-Oki (Japan) earthquake, Geoph. Res. Let ., 39, L12302, 10.1029/2012GL051894.
Yagi, Y., Fukahata, Y., 2011. Rupture process of the 2011 Tohoku -oki earthquake and absolute elastic strain release. Geophys. Res. Lett. 38(19), L19307
Yamanaka, H., Seo., K. & Samano, T., 1989. Effects of sedimentary layers on surface-wae propagation, Bull Seism. Soc. Am ., 79(3), 631-644.
Yokota, Y., Koketsu, K., Hikima, K. & Miyazaki, S. 2009. Ability of 1-Hz GPS data to infer
the source process of a medium-sized earthquake: The case of the 2008 Iwate-Miyagi Nairiku, Japan, earthquake. Geophys. Res. Lett . 36, L12301. Yue, H. & Lay, T., 2011. Inversion of high-rate (1sps) GPS data for rupture process of the 11 March 2011 Tohoku earthquake (M 9.1), w Geophys. Res. Let ., 38: L00G09. Zhao, J.X., Zhang, J., Asano, A., Ohno, Y., Oouchi, T., Takahashi, T., Ogawa, H., Irikura, K., Thio, H.K., Somerville, P.G., Fukushima, Y. & Fukushima, Y., 2006. Attenuation Relations of Strong Ground Motion in Japan Using Site Classification Based on Predominant Period, Bull Seism. Soc. Am ., 96, 898-913.
Figure 1 : The EW displacement time series for the GPS site closest to the epicentre (GPS 0550; ~75km away) and for a remote site (GPS 0998; ~650km); time origin corresponds to the earthquake occurrence. The time difference between the two responses reflects the delay of the arrival of the seismic signal at the station far from the epicentre.
Figure 2 : Precise Point Positioning (PPP) coseismic horizontal displacement field for the Tohoku-Oki event measured by the GEONET GPS network at t = T +300s. 0
Figure 3 : KiK-net (a), K-NET (b) and GEONET GPS (c) sites used in this study. The very closely located sites of GPS and strong-motion networks are indicated with red triangles. The red star indicates the epicentre of the Tohoku-Oki earthquake in 2011.
Figure 4 : Spectra of derived acceleration time series using K-NET, KiK-net and GPS data for
two sites: (a-c) close (179km) and (d-f) far (771km) from the epicentre. The solid black line corresponds to the spectra of the GPS acceleration time series before the occurrence of the earthquake indicating the noise level.
Figure 5 : The initial acceleration time series of the very closely located sites of K-NET IWT019 (blue solid line) and GPS 0164 (dashed red line) and the derived time series after filtering for the 3-10s and 10-100s period bands. The time corresponds to seconds since the earthquake occurrence.
Figure 6a : Maximum ground acceleration maps for KiK-net (top), K-NET (middle), and GPS (bottom) for the two period bands (3-10s and 10-100s).
: Maximum ground velocity maps for KiK-net (top), K-NET (middle), and GPS
Figure 6b (bottom) for the two period bands (3-10s and 10-100s).
Figure 6c : Maximum ground displacement maps for KiK-net (top), K-NET (middle), and GPS (bottom) for the two period bands (3-10s and 10-100s). Note the different scale of the 10-100s band.
Figure 7: Maximum level of noise derived from the GPS acceleration, velocity and displacement records.
Figure 8 : Normalised differences of GPS and strong-motion MGA (top), MGV (middle) and MGD (bottom) at the very closely located sites. The positive and negative distances correspond to the position of the sites northward and southward from the epicentre.
Figure 9 : Spectral acceleration (SA) at 3s from GMPE (black line) of soft soil (a) and bedrock (b) conditions (Zhao et al., 2006) compared to the SA computed from GPS (red dots) and to the SA computed from the strong-motion records (green dots). The error bars indicate the 1 σ uncertainty zone of the GMPE.
Figure 10 : GPS (red) and strong-motion (grey) MGA, MGV and MGD for the 3-10s (left) and 10-100s (right) period bands. The blue curve shows the average GPS MGA/MGV/MGD values.
Figure A1 : The differences of the GPS and strong-motion MGA (top), MGV (middle) and MGD (bottom) values for each very closely located sites versus the distance from the epicentre. The positive and negative distances correspond to the position of the sites northward and southward from the epicentre.
Table 1 : The closely located GPS and strong-motion sites of K-NET and KiK-net with their location (latitude, longitude) and distance from the epicentre.
| GPS | K-NET | KiK-net | Latitude (degrees) | Longitude (degrees) | Sensor distance (m) | Epicentre distance (km) |
|-------|---------|-----------|----------------------|-----------------------|-----------------------|---------------------------|
| 0005 | HKD080 | - | 43.5079 | 144.449 | 45 | 604 |
| 0041 | FKS011 | - | 37.0907 | 140.903 | 97 | 186 |
| 0115 | HKD066 | - | 43.6618 | 145.131 | 20 | 639 |
| 0122 | HKD083 | - | 43.2329 | 144.325 | 20 | 572 |
| 0140 | HKD151 | - | 42.4944 | 140.354 | 25 | 496 |
| 0144 | HKD110 | - | 42.1309 | 142.935 | 39 | 428 |
| 0147 | HKD155 | - | 42.043 | 140.806 | 99 | 436 |
| 0164 | IWT019 | - | 39.8492 | 141.804 | 39 | 179 |
| 0183 | AKT006 | - | 40.2154 | 140.787 | 26 | 253 |
| 0280 | GIF006 | - | 36.0331 | 136.953 | 28 | 541 |
| 0285 | GIF012 | - | 35.6362 | 136.489 | 64 | 600 |
| 0502 | HKD043 | - | 44.582 | 142.964 | 6 | 700 |
| 0503 | HKD048 | - | 44.2197 | 143.616 | 24 | 666 |
| 0504 | - | ABSH04 | 44.1919 | 143.077 | 12 | 657 |
| 0511 | HKD056 | - | 43.6705 | 143.578 | 25 | 605 |
| 0519 | HKD072 | - | 43.1952 | 145.52 | 52 | 605 |
| 0535 | AOM027 | - | 41.1456 | 140.822 | 29 | 343 |
| 0779 | HKD006 | - | 45.1268 | 142.352 | 21 | 758 |
| 0783 | HKD020 | - | 44.1488 | 141.665 | 37 | 652 |
| 0792 | HKD131 | - | 42.4205 | 141.081 | 67 | 471 |
| 0793 | - | TKCH08 | 42.4864 | 143.152 | 10 | 470 |
| 0849 | - | SOYH04 | 45.2301 | 141.882 | 95 | 771 |
| 0864 | HKD065 | - | 43.794 | 145.057 | 39 | 650 |
| 0877 | - | IKRH02 | 43.2209 | 141.652 | 60 | 550 |
| 0972 | ISK015 | - | 37.2264 | 136.909 | 9 | 494 |
| 0983 | - | NGNH28 | 36.7067 | 138.097 | 99 | 416 |
| 0991 | - | GIFH23 | 35.7235 | 136.785 | 7 | 572 |
| 0998 | - | AICH21 | 34.74 | 136.938 | 6 | 625 |
| 1009 | - | NARH03 | 34.2929 | 136.003 | 63 | 723 |
Table 2 : Mean average and standard deviation of the estimated maximum noise values of the displacements, velocities and accelerations derived from the PPP time series for the two examined period bands.
| | period 3-10s | period 3-10s | period 10-100s | period 10-100s |
|------------------------|----------------|--------------------|------------------|--------------------|
| | mean average | standard deviation | mean average | standard deviation |
| Displacement (mm) | 4.5 | 2.3 | 5.1 | 2.1 |
| Velocity (mm/s) | 5.5 | 2.7 | 1.5 | 0.9 |
| Acceleration (mm/s 2 ) | 7.6 | 4.0 | 1.0 | 0.7 |
|
10.1093/gji/ggu302
|
[
"P. Psimoulis",
"N. Houlie",
"M. Meindl",
"M. Rothacher"
] |
2014-08-04T17:24:03+00:00
|
2014-08-04T17:24:03+00:00
|
[
"physics.geo-ph"
] |
Long-period surface motion of the multi-patch Mw9.0 Tohoku-Oki earthquake
|
We show that it is possible to capture the oscillatory ground motion induced
by the Tohoku-Oki event for periods ranging from 3 to 100s using Precise Point
Positioning (PPP). We find that the ground motions of the sedimentary basins of
Japan were large (respectively > 0.15m/s and >0.15m/s2 for velocity and
acceleration) even for periods larger than 3s. We compare geodetic observables
with a Ground Motion Prediction Equation (GMPE) designed for Japan seismicity
and find that the Spectral Acceleration (SA) is well estimated for periods
larger than 3s and distances ranging from 100 to 500km. At last, through the
analysis of the displacement attenuation plots, we show that the 2011
Tohoku-Oki event is likely composed of multiple rupture patches as suggested
before by time-reversal inversions of seismic data.
|
1408.0750v1
|
## INTERMEDIATE-AGE GLOBULAR CLUSTERS IN FOUR GALAXY MERGER REMNANTS
Gelys Trancho
Giant Magellan Telescope Organization, 251 S. Lake Avenue, Pasadena, CA 91101, USA, and Carnegie Observatories, 813 Santa Barbara Street, Pasadena, CA 91101, USA
Bryan W. Miller Gemini Observatory, Casilla 603, La Serena, Chile
Franc ¸ois Schweizer Carnegie Observatories, 813 Santa Barbara Street, Pasadena, CA 91101, USA
Daniel P. Burdett
The University of Adelaide, South Australia 5005, Australia
## and
## David Palamara
Monash University, Clayton, Victoria 3800, Australia Draft version August 5, 2014
## ABSTRACT
We present the results of combining Hubble Space Telescope optical photometry with ground-based K s -band photometry from the Gemini imagers NIRI and FLAMINGOS-I to study the globular-cluster populations in four early-type galaxies that are candidate remnants of recent mergers (NGC 1700, NGC 2865, NGC 4382, and NGC 7727). These galaxies were chosen based on their blue colors and fine structure, such as shells and ripples that are indicative of past interactions. We fit the combined VIK s globular-cluster data with simple toy models of mixed cluster populations that contain three subpopulations of different age and metallicity. The fits, done via Chi-square mapping of the parameter space, yield clear evidence for the presence of intermediate-age clusters in each galaxy. We find that the ages of ∼ 1-2 Gyr for these globular-cluster subpopulations are consistent with the previously estimated merger ages for the host galaxies.
Subject headings: galaxies: individual (NGC 1700, NGC 2865, NGC 4382, NGC 7727) - galaxies: interactions - galaxies: star clusters: general
## 1. INTRODUCTION
Studies of globular cluster (GC) systems play a critical role in our understanding of galaxy formation. Until two decades ago it was thought that all GCs were old, and this shaped our view of how galaxies, especially elliptical galaxies, formed (see Harris 1991). Imaging with the Hubble Space Telescope (HST) then revealed that young compact star clusters form copiously in galaxy mergers (Whitmore et al. 1993; Miller et al. 1997; Zepf et al. 1999), a fact that supports theories in which giant elliptical galaxies are formed through mergers of spirals (Schweizer 1987; Ashman & Zepf 1992).
to be a relatively accurate method for measuring cluster ages and metallicities. However, in practice spectroscopy is very time consuming, even for 8 m-class telescopes, leaving only the brightest GCs open to such analyses. Strader et al. (2003) spent 9.3 hours observing nine GC candidates in the E5 galaxy NGC 3610 with Keck, and found only one young GC.
However, the formation and evolution of GC systems is still not well understood. We should be able to observe how GC systems evolve from the very young systems with power-law luminosity functions into old systems with log-normal luminosity functions, such as are observed in typical elliptical galaxies. These observations would greatly clarify exactly which processes (shocking, evaporation, dynamical friction; e.g., Fall & Zhang 2001) are important in the evolution of star cluster systems.
Finding intermediate-age GC populations in elliptical galaxies has been difficult. Spectroscopy is often thought [email protected]
The majority of candidate GCs in galaxies identified via morphology and color to be intermediate-age merger remnants are too faint for spectroscopic study. Photometric studies have also had limited success, partly because most have been done only at optical wavelengths (specifically, in the V and I passbands). The problem is that the V -I colors of metal-rich GCs with ages of 1-3 Gyr overlap with the colors of ∼ 13 Gyr old metalpoor GCs, resulting in an age-metallicity degeneracy (Whitmore et al. 1997, esp. Fig. 15). Fortunately, the combination of optical and near-infrared (NIR) photometry can break this age-metallicity degeneracy. This is because optical-to-NIR indices are sensitive to metallicity, but only weakly dependent on age (e.g., Kissler-Patig 2000; Puzia et al. 2002). Physically this works because in older stellar populations the V -band predominantly samples the light of stars near the turn-off, while the K s -band is more sensitive to cooler stars on the giant branch.
The color of the giant branch is metallicity-dependent, so the index V -K s is primarily a metallicity indicator. Thus, we can lift the age-metallicity degeneracy by plotting the locations of Simple Stellar Populations (SSP) in a color-color diagram, specifically V -I vs. V -K s .
## 2. OBSERVATIONS AND REDUCTIONS
## 2.1. Sample Selection
The four early-type galaxies NGC 1700, NGC 2865, NGC 4382, and NGC 7727 were selected based on their tidal fine-structure content, blue UBV colors, and enhanced Balmer absorption lines (Schweizer & Seitzer 1992; Schweizer et al. 1990). The sample spans the morphology and age gap between on-going mergers and normal elliptical galaxies.
NGC 4382 is similar to the merger remnants NGC 7252, NGC 3921, and NGC 7727, but lacks tidal tails. Its rich fine structure and distortions (see Fig. 1 in Schweizer & Seitzer 1988 and Fig. 9 in Kormendy et al. 2009) suggest that it experienced a merger 1-2 Gyr ago, a hypothesis strengthened by its still bluish center (Fisher et al. 1996), double-peaked nucleus (Lauer et al. 2005), and counter-rotating core (McDermid et al. 2004). It was selected specifically to bridge the age gap between ongoing mergers like NGC 4038/39 or NGC 3256 and normal old ellipticals.
The elliptical galaxies NGC 1700 and NGC 2865 both show evidence for recent interactions or mergers and are therefore candidate 'young ellipticals.'
NGC 1700 features boxy isophotes, a strong isophotal twist, and two broad arms or tails that have long suggested it may be the remnant of a major merger of two disk galaxies (Seitzer & Schweizer 1990, esp. Fig. 1b; Brown et al. 2000). Its kinematics supports this hypothesis: it rotates fast (Statler et al. 1996), yet features a counterrotating core with a distinctly younger stellar population (Kleineberg et al. 2011). Whitmore et al. (1997) found that its GCs are consistent with an age of ∼ 4 Gyr, but do not necessarily form a population of obviously intermediate age.
NGC 2865 features bright irregular shells, a luminous plume of material, and a faint outer loop, all suggesting that it is the remnant of a relatively recent merger (see Fig. 5 of Malin & Carter 1983 and Fig. 1 of Hau et al. 1999). This hypothesis is supported by H I gas fragments associated with the optical fine structure (Schiminovich et al. 1995), a rapidly rotating small central core with a younger stellar population and gas (Hau et al. 1999; Serra & Oosterloo 2010), and a body with unusually blue UBV colors. Sikkema et al. (2006) note that in a color histogram the galaxy's GCs seem to contain a subpopulation bluer than normal metal-poor GCs and, hence, likely of much younger age ( ∼ 0.5 - 1 Gyr).
Finally, NGC 7727 (= VV67 = Arp 222) is so obviously tidally disturbed that it was included by both Vorontsov-Velyaminov (1959) and Arp (1966) in their catalogs of interacting galaxies. Because of its spiral-shaped tidal tails it was originally classified as a peculiar Sa galaxy (de Vaucouleurs & de Vaucouleurs 1964; Sandage & Tammann 1981), but was then reclassified as a recent merger remnant of age ∼ 1.3 Gyr by Georgakakis et al. (2000). The presence
Figure 1. K s -band images of NGC 1700, NGC 2865, and NGC 4382 obtained with NIRI, and of NGC 7727 obtained with FLAMINGOS-I. The integrated exposure times were 0.5, 0.2, 0.5, and 1.0 hours, respectively.
of bluish young star clusters in it was noted by Crabtree & Smecker-Hane (1994) and described in some more detail by Trancho et al. (2004).
## 2.2. Observations
We obtained K s -band images of the above four earlytype galaxies with the Near-InfraRed Imager and Spectrometer (NIRI) on Gemini North (Hodapp et al. 2003) and with FLAMINGOS-I on Gemini South (Elston et al. 2003), with the results shown in Figure 1.
The program IDs were GN-2006B-Q-97 and GS-2002ADD-4, respectively. We used NIRI in f/6 imaging mode, which yields an image scale of 0 117/pixel and a field . ′′ of view of 2 ′ × 2 , ′ and FLAMINGOS-I also in imaging mode, which yields a scale of 0 . ′′ 078/pixel and a field of view of 2 2 . ′ × 2 2. . ′ Both instrument have similar fields of view, comparable to those of the WFPC2 and ACS cameras on HST .
All four galaxies were also observed with HST in V and I (GO-7468, PI: Schweizer), and we used unpublished photometry from that project. In the case of NGC 4382, we used the most recent photometric data published by Jordan et al. (2009).
Table 1 summarizes our observations and the data from other authors used in this study.
## 2.3. Data Reduction
The data were reduced with the Gemini/NIRI package of tasks within IRAF. Normalized flats were constructed from images taken with the Gemini calibration unit. Flat-field images with the IR lamp on and off allowed separation of the instrumental thermal signature from the sensitivity response.
The near-IR sky level and structure varies on time scales of a few minutes. To account for such variations we used a dither pattern in our imaging sequences, which provided blank fields close enough in time to derive the sky levels reliably.
Table 1 Summary of galaxy observations and data
| Galaxy | Distance a | Filter | Telescope/Instrument | Author |
|----------|--------------|------------|------------------------------|-----------------------------------|
| NGC 1700 | 52 | V, I K s | HST /WFPC2 Gemini/NIRI | Whitmore et al. (1997) This Paper |
| NGC 2865 | 35 | V, I K s | HST /WFPC2 Gemini/NIRI | Schweizer (Cycle 7) b This Paper |
| NGC 4382 | 16 | V, I c K s | HST /ACS Gemini/NIRI | Jordan et al. (2009) This Paper |
| NGC 7727 | 26 | V, I K s | HST/WFPC2 Gemini/FLAMINGOS-I | Trancho et al. (2004) This Paper |
a Distance (in Mpc) based on recession velocity relative to the Local Group and H 0 = 73 km s -1 Mpc -1 (Freedman & Madore 2010).
b Unpublished data.
c The data were originally in the passbands g, z and were transformed to V, I using equations by Peng et al. (2006, Eqs. (1) and (2)).
Some of the NIRI frames contain an electronic pattern visible as vertical striping with a period of eight columns and not always present in all quadrants. Before including the affected frames in our image processing list we used the stand-alone python routine nirinoise.py which almost perfectly removed the striping. The same image reduction steps were applied to the standard star observations as well.
All reduced science and standard star exposures of the corresponding nights were registered to a common coordinate system with the task imcoadd , based on geometric solutions obtained with the task geomap for NIRI. The final K s science image for each galaxy was derived by averaging individual images, scaled to the galaxy image from the most photometric night, with the IRAF task imcombine .
## 3. PHOTOMETRY
## 3.1. Source Detection
To detect as many GCs as possible from our K s -band images and match them with existing HST photometry, we first created a model image of each galaxy with the tasks ellipse and bmodel of the isophote package in IRAF. This model image was then subtracted from the original combined galaxy image. This procedure facilitated the uniform detection of point sources at a detection threshold of 3 σ above the background, using the task daofind of the daophot package. Figure 2 shows an example of this technique.
## 3.2. Photometric Calibration
The photometry of the candidate GCs in our sample of merger remnants was calibrated to K s on the standard system via 14 ′′ -aperture photometry matching that used in the 2MASS survey. This method was chosen because the standard stars observed with Gemini indicated significant variability in the zero point during our observations. The conditions that these galaxies were observed under were IQ70 and, therefore, not guaranteed to be photometric. The FLAMINGOS-I observations, on the other hand, showed a high level of agreement with the calibration via the 2MASS stars; coincidently, this night had very good photometric conditions.
## 3.3. GC Selection
We matched the optical HST observations (already reduced and calibrated) with our infrared GC candidates using the IRAF tasks geomap and geotran . The brightest common objects in both images served as points of reference.
## 3.4. Completeness Correction
The initial selection of GC candidates was largely based on applying color limits to the objects identified in the V , I , and K s images. However, it was expected that this method alone would prove insufficient to remove contaminants such as background galaxies and foreground stars. Therefore, the photometric completeness of the data was determined, and a visual inspection of each GC candidate was performed.
Since this study utilized space-based V and I images, but ground-based K s images, it was expected that the K s images would set the completeness limit for the photometric data. Hence, a set of completeness curves were generated from the K s images obtained with NIRI and FLAMINGOS-I for each galaxy. These curves were created through the addition of artificial cluster images to each galaxy image, using the DAOPHOT task addstar . For each of the galaxies NGC 1700, NGC 2865, and NGC 7727, 100,000 artificial cluster images ranging from 15 to 25 magnitudes were added, with no more than 100 clusters added to any one image to prevent overcrowding. In the case of NGC 4382, which exhibits an already relatively crowded field of point sources, only 50 artificial clusters were added to each image, effectively halving the total number of artificial clusters studied. The DAOPHOT tasks daofind and phot were then used to attempt to recover these artificial clusters, and to perform photometry on them. This process was repeated for background values of 25, 65, 85, 125, and 250 counts per pixel.
This process allowed us to generate two kinds of completeness curves for the GCs of each galaxy. Curves of the first kind represented the fraction of artificial clusters that were successfully recovered by the task daofind . Completeness curves of the second kind represented the fraction of artificial clusters that were not only successfully recovered by daofind , but that also returned a photometric magnitude within acceptable limits via the task phot . There were two of these limits: the returned photometric magnitude was required to have an error of < 0.3 mag and was also required to be within 0.5 mag of the magnitude of the input artificial-cluster image.
A smoothed curve based on the completeness func-
Figure 2. From left to right: K s -band image of NGC 1700, galaxy model of NGC 1700, and model-subtracted image.
tion given in Fleming et al. (1995) was fit to each set of completeness curves generated. The two parameters returned by this fitting procedure were the 50% completeness limit, K s, lim (expressed in magnitudes), and a steepness parameter α . Table 2 presents for each galaxy and background value the two fit parameters derived, both for the first kind of completeness curves (subscript 'fnd' for 'found') and for the second kind (subscript 'acc' for 'accepted').
As mentioned above, several completeness curves were generated for each galaxy in order to determine the stability of the photometric completeness limit over a range of background values. Given that the background varied considerably over any one image, particularly across the center of the galaxy, such stability was essential. Indeed, the variance in background over each image considerably affected the point-spread functions (PSFs) that we attempted to construct. In order to negate this effect, the final PSFs that were used in the artificial cluster addition were generated from background-subtracted GC images, as illustrated in Figure 3. These background-subtracted GC images were created using a 15 × 15-pixel median subtraction.
¿From the results presented in Table 2 it is evident that-although the steepness of the completeness curves did vary-the 50% completeness limit was exceptionally stable over the range of background values investigated. Therefore, we were able to use a single completeness limit for each image, despite the fact that the background values varied dramatically across the cores of the galaxies.
## 3.4.1. NGC 1700
Figure 4 shows the completeness curve derived for GCs in NGC 1700 (shown for accepted artificial clusters only). As Table 2 details, the 50% K s completeness limits were K s, lim = 21 90- 21.91 for both the recovered and the ac-. cepted clusters at all background levels. Hence, it is evident that photometric errors do not play a significant role in setting the depth of completeness in this case. Rather, the completeness is limited primarily by our ability to detect the clusters.
Figure 3. PSFs generated from the median backgroundsubtracted images of NGC 1700, NGC 2865, NGC 4382, and NGC 7727. The field of view is 7 8 . ′′ × 7 8 . ′′ for each panel.
Figure 4. Completeness curve for NGC 1700 derived from artificial cluster addition. The histogram shows the fraction of clusters recovered to within acceptable error and magnitude limits, while the smooth curve indicates the fitted completeness function.
Figure 5 shows the completeness curve derived for GCs in NGC 2865. In this galaxy, the 50% completeness limits were K s, lim = 21 51-21.52 for both the recovered . and the accepted artificial clusters. As in the case of NGC 1700, the very small range of these values indicates that the completeness in K s was limited by our ability
Figure 5. Completeness curve for NGC 2865 derived from artificial cluster addition. The histogram shows the fraction of clusters recovered to within acceptable error and magnitude limits, while the smooth curve indicates the fitted completeness function.
Figure 6. Completeness curve for NGC 4382 derived from artificial cluster addition. The histogram shows the fraction of clusters recovered to within acceptable error and magnitude limits, while the smooth curve indicates the fitted completeness function.
Figure 7. Completeness curve for NGC 7727 derived from artificial cluster addition. The histogram shows the fraction of clusters recovered to within acceptable error and magnitude limits, while the smooth curve indicates the fitted completeness function.
to detect the clusters rather than by photometric errors.
## 3.4.3. NGC 4382
Figure 6 shows the completeness curve derived for GCs in NGC 4382. Again, the K s 50% completeness limits are similar for the recovered and accepted curves, though with wider scatter than in NGC 1700 and NGC 2865 and with a 0.06 mag difference between the two mean values (see Table 2).
## 3.4.4. NGC 7727
Figure 7 shows the completeness curve derived for GCs in NGC 7727.
As Table 2 details, the various K s 50% completeness limits scatter more and show a significant systematic difference of 0.09 mag between the recovered artificial clusters (mean K s, lim fnd , = 22 46) and the accepted clusters . (mean K s, lim acc , = 22 37). . Notice that these limits for the NGC 7727 GCs are more than 0.5 mag fainter than the limits for GCs in the other three galaxies, which is not
Table 2 Summary of the parameters for the completeness curves for each galaxy.
| Galaxy NGC | Sky Value (counts) | K s, lim , fnd a (mag) | α fnd b | K s, lim , acc a (mag) | α acc b |
|--------------|----------------------|--------------------------|-----------|--------------------------|-----------|
| 1700 | 25 | 21.91 | 22.05 | 21.9 | 19.27 |
| 1700 | 55 | 21.91 | 15.74 | 21.9 | 12.84 |
| 1700 | 85 | 21.91 | 21.19 | 21.9 | 9.17 |
| 1700 | 125 | 21.91 | 18.41 | 21.9 | 18.97 |
| 1700 | 250 | 21.91 | 17.62 | 21.9 | 16.33 |
| 2865 | 25 | 21.52 | 16 | 21.51 | 15.1 |
| 2865 | 55 | 21.52 | 11.81 | 21.51 | 12.81 |
| 2865 | 85 | 21.51 | 16.67 | 21.51 | 13.33 |
| 2865 | 125 | 21.52 | 15.39 | 21.51 | 7.09 |
| 2865 | 250 | 21.52 | 14.63 | 21.51 | 8.37 |
| 4382 | 25 | 21.55 | 11.57 | 21.51 | 10.55 |
| 4382 | 55 | 21.56 | 11.18 | 21.46 | 3.68 |
| 4382 | 85 | 21.55 | 11.87 | 21.51 | 14.11 |
| 4382 | 125 | 21.57 | 10.49 | 21.51 | 10.03 |
| 4382 | 250 | 21.54 | 13.03 | 21.47 | 4.01 |
| 7727 | 25 | 22.46 | 3.93 | 22.38 | 3.99 |
| 7727 | 55 | 22.45 | 3.76 | 22.37 | 4.44 |
| 7727 | 85 | 22.46 | 4.26 | 22.38 | 3.81 |
| 7727 | 125 | 22.46 | 3.94 | 22.37 | 3.55 |
| 7727 | 250 | 22.45 | 3.96 | 22.35 | 3.55 |
- a 50% K s completeness limit.
- b Curve fitting parameter (Fleming et al. 1995) governing the slope of the fit.
Table 3 Summary of globular-cluster candidates removed during each correction
| Galaxy NGC | Initial GC Candidates | Removed via Com- pleteness | Removed via Visual Inspection | Final GC Candidates |
|--------------|-------------------------|------------------------------|---------------------------------|-----------------------|
| 1700 | 45 | 6 | 13 | 26 |
| 2865 | 65 | 2 | 20 | 43 |
| 4382 | 65 | 3 | 2 | 60 |
| 7727 | 31 | 0 | 9 | 22 |
surprising given that the K s images of NGC 7727 were obtained with FLAMINGOS-I on a night of particularly good seeing. It is because of these fainter limits that the photometric errors are more influential in NGC 7727 than in the other three galaxies.
Nevertheless, we adopted the single value of K s, lim acc , = 22 37 as our acceptance criterion for . candidate GCs in NGC 7727.
## 3.5. Further Corrections
In addition to the completeness corrections we performed a visual examination of each GC candidate, using the IRAF task imexam to further remove any contaminants such as background galaxies. Candidates were dropped if their shape was particularly elliptical and/or if the counts per pixel within the object were well below those of the other candidates. Table 3 gives a summary of the GC candidate corrections applied for each GC system. Figures 8 and 9 show the resulting colorcolor and color-magnitude diagrams for the GC systems of NGC 2865 and NGC 7727, respectively.
## 4. FITTING TO SSP MODELS
In order to constrain the age of the merger event for each of the four galaxies, the ages of their GC subpopulations were determined. Specifically, we assumed that
Figure 8. Color-magnitude and color-color diagrams for GCs of NGC 2865.
the parameters of the youngest subpopulation may be associated with the most recent major star-formation episode-the galaxy-merger event itself.
Conventional studies of GC ages and metallicities are often based on optical-to-NIR color-magnitude diagrams such as, e.g., the V vs V -I diagram. However, such diagrams suffer from an inherent age-metallicity degeneracy (e.g., Worthey 1999), and it is often difficult to disentangle age and metallicity without additional detailed spectra.
In an effort to break-or at least partly lift-this degeneracy, we use VIK s color-color diagrams (specifically V -I vs V -K s ) and interpret the measured GC colors via the SSP models generated by Charlot & Bruzual (2007, private communication, hereafter CB07; see Bruzual & Charlot 2003 and Bruzual 2007 for published models and details).
The fitting algorithm used to match the models to our data is briefly described below and closely follows the method and algorithm used by Hempel & Kissler-Patig (2004). We also present the results of our efforts to verify the technique.
## 4.1. Methodology
Our principal goal was to identify the possible presence of multiple subpopulations among the GCs of our sample galaxies and to estimate their ages, based on simple assumptions about their metallicities.
We assume that there are three predominant cluster subpopulations of interest: one consisting of ∼ 13 Gyr old metal-poor GCs, a second one of also old ( > 10 Gyr) metal-rich GCs, and a third one of younger, intermediate-age metal-rich GCs. The universal old metal-poor subpopulation is present in all major galax-
Figure 9. Color-magnitude and color-color diagrams for GCs of NGC 7727.
ies and is represented in our model simulations by clusters of mean logarithmic metallicity [ Z/Z /circledot ] = -1 7, with . VIK s colors computed by CB07; the size and age of this subpopulation is being held constant during the simulations for each galaxy. The two metal-rich subpopulations are assumed to both have logarithmic metallicities in the range of -0 7 . < Z/Z [ /circledot ] < 0 4. . The number ratios of metal-rich to metal-poor clusters, and of intermediateage to old clusters within the metal-rich subpopulation, are then varied systematically.
With these assumptions made, the goal is to simulate the appearance of the mixed GC populations in V -I vs V -K s diagrams with BC07 model clusters and to then compare these model diagrams with the observed diagrams for the GCs of each galaxy. The measure used for the comparison is provided by the cumulative age distributions of the artificial clusters and the observed GCs, both derived from the positions of the clusters in their respective V -I vs V -K s diagrams and compared via a chi-squared (hereafter χ 2 ) test described below.
galaxy, a set of model cluster V -K s colors was generated using a Monte Carlo technique. For each of the many simulations per galaxy, the total number of model clusters was made to equal the number of GC candidates in the galaxy (Table 3, last column). Corresponding model cluster V -I colors were then computed based on the SSP models of CB07. To simulate observational errors, each model cluster population was furthermore convolved with photometric errors based on the indicative errors of our data. Finally, cumulative age distributions were calculated for each model cluster population, based on the same SSP models by CB07.
To simulate the observed mixed GC population of each
For each of our four galaxies, we generated a grid of 121 model cluster populations as follows. The age of the old metal-rich subpopulation was held fixed at 13 Gyr, while the age of the younger metal-rich subpopulation was set to 0.2, 0.5, 0.75, 1, 1.5, 2, 3, 5, 7, 10, and 13 Gyr. For each of these 11 ages, the ratio of young to old clusters was varied from 0% to 100% in 10% steps, thus yielding a total of 121 model populations for comparison with the observed GC population of each galaxy.
For each of the 121 age and number-ratio combinations, 1000 model samples were generated via the Monte Carlo technique, and the average cumulative age distribution (based on the SSP models) of the simulated clusters was calculated. The cumulative age distribution of the observed GCs was also calculated based on the same SSP models, via a transformation from the V -I vs V -K s plane to the abundance vs log-age plane based on these models (for details, see Hempel & Kissler-Patig 2004).
We then performed χ 2 fitting of the model cumulative age distributions to the observed cumulative age distribution to determine which of the age and number-ratio combinations best fit the data. For each galaxy, a contour plot of the χ 2 values over the age vs. number-ratio configuration plane was drawn and used to infer the age and number ratio of the young metal-rich subpopulation as compared to the old subpopulation. These contour plots are shown and discussed in § 4.3 below.
## 4.2. Verification of Technique
In order to validate the use of this technique, we performed a series of verification tests independent from those performed by Hempel & Kissler-Patig (2004). Our verification method involved replacing the observed GC population with a simulated population of model clusters. The simulation of this population followed the same procedure as the simulations of our model populations described above.
First, we tested the ability of our adopted method to return an accurate value for the age of the younger metalrich input population. As Figure 10 shows, while the age returned by the χ 2 -fitting did not converge exactly to the simulated input age, it was largely consistent (to within ∼ 10%). Hence, we can confirm that, over the range of ages ( ∼ 1 - 5 Gyr) that we are investigating, the method returns a reasonably accurate age for a given input age.
Second, we also tested our method's stability with respect to the photometric errors of the input data. For a given age and ratio configuration, we varied the photometric errors from 0.1 mag to 0.3 mag. Note that (i) the indicative photometric errors in our observed data were below 0.3 mag, and (ii) while there were some clusters with higher errors, they were cut before the χ 2 -fitting was performed.
Figure 11 highlights that, as expected, increasing the errors decreased the convergence of the χ 2 fit. Even so, however, the age returned by the fitting contours was still largely consistent with the simulated input age. Thus, we confirm the validity of our method for both the ages and the indicative photometric errors of the input data.
¿From the spread in the χ 2 contours and the deviation of the converged age from the input age we infer an upper limit for the age uncertainty of about ± 0 75 Gyr. . Similarly, for the young-to-old ratio we find an upper limit to the uncertainty of about ± 20%. These uncertainties will be used for the GC subpopulations of each galaxy in the following sections.
## 4.3. Cluster Ages
Figure 12 shows contour plots of χ 2 values over the modeled configuration space for each GC system of our four sample galaxies. In each contour plot, the convergence of χ 2 to a minimum value marks the most likely
Figure 10. χ 2 contour plots using a simulated cluster population as input. The age of the younger subpopulation increases from 1 Gyr (top) to 2 Gyr (middle) and 5 Gyr (bottom). The input age, number ratio, and photometric error are given for each.
combination of age and number-ratio for the younger GC subpopulation of that galaxy, based on the SSP models by CB07. Table 4 summarizes these resulting combinations, which we now briefly discuss for each galaxy.
## 4.3.1. NGC 1700
For this galaxy, Figure 12 shows that the χ 2 values converge to a minimum at a GC population mix consisting of about 5% of an old subpopulation and 95% of a younger subpopulation with an age of 1 7 . ± 0 8 Gyr. .
Note that this age is consistent with the previously published age estimate of 3 ± 1 Gyr (Brown et al. 2000) to within the combined errors. Note also that the quoted percentages apply only to the observed GC candidates, and not to the entire GC population, whose total number and subpopulation percentages are unknown.
Figure 11. χ 2 contour plots using a simulated cluster population as input. The size of the photometric errors increases from 0.1 mag (top) to 0.2 mag (middle) and 0.3 mag (bottom). The input age, number ratio, and photometric error are given for each.
## 4.3.2. NGC 2865
For NGC 2865, the χ 2 contours again show a high level of convergence, as can be seen in the top-right panel of Figure 12. From the contour minimum we infer a younger metal-rich GC subpopulation of age 1 8 . ± 0 8 Gyr, com-. prising about 90% of the observed GC candidates.
Again, this newly determined age for the younger GCs compares well with the previously published age of 1 8 . ± 0 5 Gyr for the young . stellar subpopulation in this galaxy (Rampazzo et al. 2007).
## 4.3.3. NGC 4382
For NGC 4382, the χ 2 contours shown in Figure 12 indicate a younger metal-rich GC subpopulation of age 1 8 . ± 0 8 Gyr, comprising between about 75% and 95% .
of the observed GC candidates.
Note that the χ 2 values for the GCs in NGC 4382 are much smaller than those for the GCs in the other three galaxies, suggesting a better agreement between the observed GC data and the simulated cluster populations.
As can be seen, there is also a significantly wider spread for the ratio of young-to-old GCs. Recall, however, that our primary interest is in the age of the younger subpopulation, and note that the convergence in age is very high. This newly determined age for the younger metalrich GC subpopulation agrees well with the previously published age of 1 6 . ± 0 3 Gyr for the young . stellar subpopulation of this galaxy (Sansom et al. 2006).
## 4.3.4. NGC 7727
Finally, for NGC 7727 the χ 2 contours of Figure 12 indicate a purely young metal-rich GC population of age 0 9 . ± 0 8 Gyr. The observed GC candidates of this galaxy . are noteworthy both for their particularly young ages and for the apparent absence of an older subpopulation. Both of these characteristics were previously reported by Trancho et al. (2004), who obtained a GC population age of 1 - 2 Gyrs. Note, however, that the population of old metal-poor GCs universally present in all major galaxies may be too faint to be detected on our K s -band images. Hence, its apparent absence may simply reflect a selection effect and does not imply that this old metalpoor GC population doesn't exist.
## 5. SUMMARY AND CONCLUSIONS
We have presented an age analysis of GC systems in four early-type galaxies that are considered candidate remnants of recent mergers. New K s -band photometry of the GCs in these galaxies has been obtained with two near-infrared imagers on Gemini and combined with existing HST photometry in V and I to produce colorcolor and color-magnitude diagrams of the GC systems in VIK s . These data have been fitted to simple toy models of mixed GC populations containing three subpopulations that differ in age and metallicity. The mixed GC populations have been simulated with SSP models by Charlot & Bruzual (2007, private communication) and compared to the observations. Various χ 2 fits to the cumulative age distributions derived from our data and from the toy models yield clear evidence for the presence of intermediate-age, metal-rich subpopulations of GCs in each of the four early-type galaxies. At the observed, relatively bright K s magnitudes these intermediate-age subpopulations exceed the old metal-rich and metal-poor subpopulations in cluster numbers, and their estimated ages of ∼ 1 - 2 Gyr are consistent with published ages for the same galaxies based on the integrated light. Thus, VIK s photometry of GC populations in early-type galaxies can be used to date epochs of major star and cluster formation presumably due to relatively recent merger events.
We thank St´ ephane Charlot and Gustavo Bruzual for sending us an early version of their new clusterpopulation models ('CB07'), presently still under development, and Maren Hempel and Markus Kissler-Patig for helpful information and a copy of their simulation code. We also thank the anonymous referee, whose
Figure 12. Contour plots of χ 2 values from fits of SSP model colors (CB07) to our VIK s colors for GCs in NGC 1700, 2865, 4382, and 7727. Notice that the percentages of 13 Gyr old GCs among those observed are small. Most observed GCs in the four sample galaxies belong to younger, metal-rich subpopulations of intermediate age (see Table 4).
Table 4 Results of χ 2 fitting: Ages and percentages of intermediate-age GC subpopulations
| Galaxy NGC | Age a (Gyr) | Percentage Young b | Published Age (Gyr) | Source |
|--------------|---------------|----------------------|-----------------------|------------------------|
| 1700 | 1.7 ± 0.8 | 95% | 3 ± 1 . 0 | Brown et al. (2000) |
| 2865 | 1.8 ± 0.8 | 90% | 1 . 8 ± 0 . 5 | Rampazzo et al. (2007) |
| 4382 | 1.8 ± 0.8 | 85% | 1 . 6 ± 0 . 3 | Sansom et al. (2006) |
| 7727 | 0.9 ± 0.8 | 100% | 1 - 2 | Trancho et al. (2004) |
- a Age of young metal-rich GC subpopulation as determined from χ 2 contours.
- b Percentage of observed GC candidates belonging to the younger subpopulation
- as determined from χ 2 contours.
constructive comments helped us improve the presentation of our results. This research is based on observations obtained at the Gemini Observatory, which is operated by the Association of Universities for Research in Astronomy, Inc., under a cooperative agreement with the NSF on behalf of the Gemini partnership: the National Science Foundation (United States), the Particle Physics and Astronomy Research Council (United Kingdom), the National Research Council (Canada), CONICYT (Chile), the Australian Research Council (Australia), CNPq (Brazil) and CONICET (Argentina).
## REFERENCES
Arp, H. 1966, Atlas of Peculiar Galaxies (Pasadena: California Institute of Technology); also: ApJS, 14, 1 Ashman, K. M., & Zepf, S. E. 1992, ApJ, 384, 50 Brown, R. J. N., Forbes, D. A., Kissler-Patig, M., & Brodie, J. P. 2000, MNRAS, 317, 406 Bruzual A., G. 2007, in ASP Conf. Proc. 374, From Stars to Galaxies: Building the Pieces to Build Up the Universe, ed. A. Vallenari et al. (San Francisco, CA: ASP), 303 Bruzual, G., & Charlot, S. 2003, MNRAS, 344, 1000 Crabtree, D. R., & Smecker-Hane, T. 1994, BAAS, 26, 1499 de Vaucouleurs, G., & de Vaucouleurs, A. 1964, Reference Catalogue of Bright Galaxies (Austin, TX: Univ. of Texas Press) Elston, R., Raines, S. N., Hanna, K. T., et al. 2003, Proc. SPIEE, 4841, 1611 Fall, S. M., & Zhang, Q. 2001, ApJ, 561, 751
Fisher, D., Franx, M., & Illingworth, G. 1996, ApJ, 459, 110
- Fleming, D. E. B., Harris, W. E., Pritchet, C. J., & Hanes, D. A. 1995, AJ, 109, 1044
- Freedman, W. L., & Madore, B. F. 2010, ARA&A, 48, 673 Georgakakis, A., Forbes, D. A., & Norris, R. P. 2000, MNRAS,
- 318, 124
- Harris, W. E. 1991, ARA&A, 29, 543
- Hau, G. K. T., Carter, D., & Balcells, M. 1999, MNRAS, 306, 437 Hempel, M., & Kissler-Patig, M. 2004, A&A, 419, 863 Hodapp, K. W., Jensen, J. B., Irwin, E. M., et al. 2003, PASP,
- 115, 1388
- Jordan, A., Peng, E. W., Blakeslee, J. P., et al. 2009, VizieR On-line Data Catalog; also: ApJS, 180, 54
- Kissler-Patig, M. 2000, in ASP Conf. Ser. 211, Massive Stellar Clusters, ed. A. Lan¸on & C. Boily (San Francisco: ASP), 157 c Kleineberg, K., S´nchez-Bl´zquez, P., & Vazdekis, A. 2011, ApJ, a a 732, L33
- Kormendy, J., Fisher, D. B., Cornell, M. E., & Bender, R. 2009, ApJS, 182, 216
- Lauer, T. R., Faber, S. M., Gebhardt, K., et al. 2005, AJ, 129, 2138
- Malin, D. F., & Carter, D. 1983, ApJ, 274, 534
- McDermid, R., Emsellem, E., Cappellari, M., et al. 2004, Astron. Nachr., 325, 100
- Miller, B., Whitmore, B. C., Schweizer, F., & Fall, S. M. 1997, AJ, 114, 2381
- Peng, E. W., Cˆt´ e, P., Jord´n, A., et al. 2006, ApJ, 639, 838 o a Puzia, T. H., Zepf, S. E., Kissler-Patig, M., et al. 2002, A&A, 391, 453
- Rampazzo, R., Marino, A., Tantalo, R., et al. 2007, MNRAS, 381, 245
- Sandage, A., & Tammann, G. A. 1981, A Revised Shapley-Ames Catalog of Bright Galaxies (Washington, DC: Carnegie Institution of Washington)
- Sansom, A. E., O'Sullivan, E., Forbes, D. A., Proctor, R. N., & Davis, D. S. 2006, MNRAS, 370, 1541
- Schiminovich, D., van Gorkom, J. H., van der Hulst, J. M., & Malin, D. F. 1995, ApJ, 444, L77
- Schweizer, F. 1987, in Nearly Normal Galaxies, ed. S. M. Faber (Springer: New York), 18
- Schweizer, F., & Seitzer, P. 1988, ApJ, 328, 88
- Schweizer, F., Seitzer, P., Faber, S. M., et al. 1990, ApJ, 364, L33
- Schweizer, F., & Seitzer, P. 1992, AJ, 104, 1039
- Seitzer, P., & Schweizer, F. 1990, in Dynamics and Interactions of Galaxies, ed. R. Wielen (Berlin: Springer), 60
- Serra, P., & Oosterloo, T. A. 2010, MNRAS, 401, L29
- Sikkema, G., Peletier, R. F., Carter, D., Valentijn, E. A., & Balcells, M. 2006, A&A, 458, 53
- Statler, T. S., Smecker-Hane, T., & Cecil, G. N. 1996, AJ, 111, 1512
- Strader, J., Brodie, J. P., Schweizer, F., Larsen, S. S., & Seitzer, P. 2003, AJ, 125, 626
- Trancho, G., Miller, B. W., Schweizer, F., & Whitmore, B. 2004, in ASP Conf. Ser. 322, The Formation and Evolution of Massive Young Star Clusters, ed. H. J. G. L. M. Lamers, L. J. Smith, & A. Nota (San Francisco: ASP), 219
- Vorontsov-Velyaminov, B. A. 1959, Atlas and Catalog of Interacting Galaxies (Moscow: Sternberg Institute, Moscow State Univ.)
- Whitmore, B. C., Schweizer, F., Leitherer, C., Borne, K., & Robert, C. 1993, AJ, 106, 1354
- Whitmore, B. C., Miller, B. W., Schweizer, F., & Fall, S. M. 1997, AJ, 114, 1797
- Worthey, G. 1999, in ASP Conf. Ser. 192, Spectrophotometric Dating of Stars and Galaxies, ed. I. Hubany, S. R. Heap, & R. H. Cornett (San Francisco: ASP), 283
- Zepf, S. E., Ashman, K. M., English, J., Freeman, K. C., & Sharples, R. M. 1999, AJ, 118, 752
|
10.1088/0004-637X/790/2/122
|
[
"Gelys Trancho",
"Bryan W. Miller",
"Francois Schweizer",
"Daniel P. Burdett",
"David Palamara"
] |
2014-08-04T17:34:54+00:00
|
2014-08-04T17:34:54+00:00
|
[
"astro-ph.GA"
] |
Intermediate-age Globular Clusters in Four Galaxy Merger Remnants
|
We present the results of combining Hubble Space Telescope optical photometry
with ground-based Ks-band photometry from the Gemini imagers NIRI and
FLAMINGOS-I to study the globular-cluster populations in four early-type
galaxies that are candidate remnants of recent mergers (NGC1700, NGC2865,
NGC4382, and NGC7727). These galaxies were chosen based on their blue colors
and fine structure, such as shells and ripples that are indicative of past
interactions. We fit the combined VIKs globular-cluster data with simple toy
models of mixed cluster populations that contain three subpopulations of
different age and metallicity. The fits, done via Chi-square mapping of the
parameter space, yield clear evidence for the presence of intermediate-age
clusters in each galaxy. We find that the ages of 1-2 Gyr for these
globular-cluster subpopulations are consistent with the previously estimated
merger ages for the host galaxies.
|
1408.0753v1
|
## Fundamental Neutron Physics Beamline at the Spallation Neutron Source at ORNL
N. Fomin 1 , G. L. Greene 1 2 , , R. Allen 2 , V. Cianciolo 2 , C. Crawford 3 , T. Ito 7 , P. R. Hu ff man 2 4 , , E. B. Iverson 2 , R. Mahurin 5 , W. M. Snow 6
1 University of Tennessee, Knoxville, TN, USA
2 Oak Ridge National Laboratory, Oak Ridge, TN, USA
3 University of Kentucky, Lexington, KY, USA
4 North Carolina State University, Raleigh, NC, USA
5 University of Manitoba, Winnipeg, Manitoba, Canada
6 Indiana University and Center for the Exploration of Energy and Matter, Bloomington, IN, USA
7 Los Alamos National Laboratory, Los Alamos, NM, USA
## Abstract
We describe the Fundamental Neutron Physics Beamline (FnPB) facility located at the Spallation Neutron Source at Oak Ridge National Laboratory. The FnPB was designed for the conduct of experiments that investigate scientific issues in nuclear physics, particle physics, and astrophysics and cosmology using a pulsed slow neutron beam. We present a detailed description of the design philosophy, beamline components, and measured fluxes of the polychromatic and monochromatic beams.
## 1. Introduction
Cold neutrons and ultracold neutrons (UCN) have been employed in a wide variety of experiments that shed light on important issues in nuclear, particle, and astrophysics. These include the determination of fundamental constants, the study of fundamental symmetry violation, searches for new interactions in nature and tests of the fundamental laws of quantum mechanics. Their special combination of properties make them a good choice to address the expanded list of fundamental scientific mysteries which now confront us in the wake of the discoveries of dark matter and dark energy, which shows that 95% of the energy content of the universe resides in some unknown form. In many cases, experiments with slow neutrons provide information not available from existing accelerator-based nuclear physics facilities or high-energy accelerators [1, 2, 3].
For this reason, most major neutron user facilities have included, as a component of their scientific program, the investigation of fundamental interactions. The great majority of these facilities have been located at continuous-wave (CW) spallation sources or intense research reactors to take advantage of the highest neutron intensities available. However, many precision experiments could benefit from the special characteristics of pulsed mode spallation neutron sources, which enables experiments to use the time and energy structure of the slow neutron beam to advantage. The peak and average cold neutron fluxes from pulsed reactors like the
Frank Laboratory of Neutron Physics and spallation sources like the Los Alamos Neutron Science Center (LANSCE), the Spallation Neutron Source (SNS), the Japanese Spallation Neutron Source (JSNS), the ISIS spallation source at the Rutherford Appleton Lab, and the future European Spallation Source (ESS) are now high enough that a relatively broad class of fundamental neutron physics experiments can be best performed at such facilities.
In this paper we describe the construction and commissioning of a pulsed slow neutron beamline and facility at the Spallation Neutron Source (SNS) at Oak Ridge National Laboratory which is optimized to conduct certain experiments in this field. The design of the facility attempted to preserve the advantages of pulsed spallation sources in reducing systematic errors in a broad class of fundamental neutron physics experiments. One such feature is the well-known time structure of the beam. It can be used to identify and analyze the neutron energy dependence of background signals which are invisible at a CW source, to shape the beam phase space in special ways using time-dependent neutron optical components, or to help perform absolute neutron polarization measurements. Another feature of a spallation neutron source is the fact that the neutron source is o ff by the time the slow neutrons arrive at the apparatus, which generally reduces background signals in detectors. With improved neutron optics technology one can bend the slow neutron beam enough to eliminate line-of-sight to the production target and to guide it far from the spallation source so that the very high energy neutrons and gamma rays generated in spallation do not adversely a ff ect the physics experiments.
Furthermore, since this beamline is operated as a user facility with all beam time allocated on the basis of independent peer reviews, it was essential to design the facility to accommodate a wide variety of di ff erent types of slow neutron experiments and especially to avoid precluding to the extent possible new unforeseen ideas for future experiments. An Instrument Development Team (IDT) for the beamline facility reached a consensus based on experience at other facilities combined with feedback and discussions from potential future users in the form of the following design principles as follows:
- 1. Total intensity was to be maximized. Essentially all of the experiments in this field are limited by statistics. The facility design must not include features that lead to a permanent reduction in the neutron fluence (total number per second).
- 2. Cold neutron intensity is of the highest priority. Almost all of the experiments of relevance for the areas of nuclear and particle physics prefer low energy neutrons. Slower neutron beams raise the signal background ratio in neutron decay experiments. / The opportunities for creative manipulation of the beam properties (phase space, polarization, etc.) are greater for cold neutrons.
- 3. The unique properties of a spallation neutron source should not be compromised in the beamline design. The advantages that a spallation source o ff ers experiments in this area relative to a steady-state source are directly or indirectly rooted in the built-in use of neutron time-of-flight and a corresponding potential for increase in the signal background / ratio. Everything must be done to preserve this advantage while at the same time being consistent with (1) and (2).
- 4. Accommodate the di ff erent demands of di ff erent classes of experiments consistent with (1)-(3).
- 5. Leave as much floor space as possible. This allows for flexibility in the design of future experiments.
In the remainder of this paper we describe the details of the design of the facility which we have realized consistent with these principles and the physical constraints and properties specific to
the Spallation Neutron Source at ORNL. We hope that this detailed description of the beamline design and properties will be useful both to potential scientific users and to the design of possible future facilities at pulsed spallation sources such as the ESS [4]. The specific design choices made for the SNS facility are by no means unique: we encourage the reader to contrast the SNS facility with the other fundamental neutron beamlines constructed at pulsed spallation neutron source, namely the pioneering beamline at LANSCE [5] and the fundamental neutron physics beamline at the JSNS, which supplies a collection of three slow neutron beamlines with di ff erent phase space profiles optimized for di ff erent subclasses of experiments [6, 7]. We also encourage a comparison with the intense CW slow neutron beam facilities for fundamental neutron physics already in operation or under development at PSI [8, 9, 10, 11], ILL [12], and NIST [13, 14].
## 2. The Spallation Neutron Source
The Spallation Neutron Source (SNS) is the most intense pulsed neutron source in the world, designed to deliver a sub-microsecond proton pulse onto a mercury target at a repetition rate of 60 Hz with time-averaged proton power of 1.4 MW. The released spallation neutrons are moderated by supercritical hydrogen and water moderators. The resulting slow neutrons are used for a variety of experiments. The SNS supports 24 instruments which can conduct experiments simultaneously.
The SNS uses a cesium-enhanced, RF-driven multi-cusp ion source [15] to provide a 65 keV H -beam at 60 Hz with a pulse length of up to 1 ms. The normal conducting linac consists of six drift tube linac (DTL) tanks, which bring the H -beam energy up to 86.8 MeV, and four coupled cavity linac (CCL) structures, which supply additional acceleration to bring the energy to 185.6 MeV. This is followed by a superconducting linac consisting of 11 medium-beta ( β = 0.61) cryomodules and 12 high-beta ( β = 0.81) cryomodules. The medium-beta cryomodules each contain 3 cavities and provide 10.1 MV m, bringing the H / -ions upto an energy of 379 MeV. The high-beta cryomodules (consisting of 4 cavities each) are designed to provide upto 15.9 MV m / resulting in a maximum proton energy of 1.3 GeV. In June of 2014, the SNS was operating at 940 MeV of proton energy.
The H -pulses arrive at the ring injection point via a 150-m long beam line which is used for energy collimation (bending magnets) and transverse halo collimation (straight sections). The ring consists of four straight sections as well as four arcs, with a total flight path corresponding to 1 us of accumulated proton pulse length. The incoming H -beam is stripped of its two electrons by passing through a thin carbon or diamond foil, and is merged in phase space with the circulating proton beam. Currently, the ring design allows operation at up to 1.0 GeV, but RF systems and injection kickers have been designed to support 1.3 GeV operation with minimal upgrades.
The protons are incident on a target of circulating mercury, producing spallation reactions as they deposit their energy. The mercury is contained inside the stainless steel target module, one of the components of the target system. The target module is made of two concentric vessels, where the inner vessel contains the target mercury during normal operation, and the outer vessel contains any mercury that may leak from the inner vessel. The main process loop contains ≈ 1.4 m 3 of mercury, circulating at ≈ 325 kg s, at a pressure of / ≈ 0.3 MPa. The inner vessel is cooled by flowing mercury, whereas the outer vessel is cooled by water. The two vessels are separated by a helium-filled interstitial region, where there are two instruments present. A heated resistance temperature detector (RTD) is able to detect leakage as well as distinguish between
Figure 1: A cutaway view of the mercury target, surrounded by four moderators enclosed in a beryllium reflector. Each moderator is viewed by several beamlines. See text for details.
mercury and water; an electrical conductivity probe detects the presence of mercury between the contacts [16, 17].
Spallation neutrons are moderated by undergoing repeated scattering primarily with hydrogen and beryllium atoms. The configuration at the SNS includes four moderators, two of which are viewed from both sides of the target. Three moderators are supercritical hydrogen at ≈ 20 K, and one is liquid water at ≈ 320 K. The moderators are surrounded by a water-cooled beryllium inner reflector, which is then surrounded by water-cooled stainless steel [16, 18]. The configuration can be seen in Fig. 2. The viewed faces of all the moderators are 10 cm (horizontal) by 12 cm (vertical).
The FnPB views the bottom downstream moderator, where the hydrogen is delivered through the bottom of the vessel via a jet, which forces the hydrogen to circulate. This moderator is fully coupled unpoisoned hydrogen (nominally parahydrogen) at 20 K (viewed from one side only), with a curved viewing surface and maximum moderator thickness of 60 mm (average of 55 mm). The moderator is surrounded by ≈ 20 mm of light water acting as a premoderator.
Figure 2: The mercury target vessel surrounded by 4 moderators inside of a beryllium reflector. FnPB views the bottom downstream moderator (bottom right in the figure). See text for details.
Figure 3: A top view of the Fundamental neutron Physics Beamline(s). 13A is the monochromatic line (left) and will extend into the external building to serve the nEDM experiment. 13B is the polychromatic beamline (right), shown with the NPDGamma experiment [19, 20, 21] installed in the experimental area.
## 3. Fundamental Neutron Physics Beamline - Overview
The Fundamental Neutron Physics Beamline (FnPB) is one of three experimental areas which view this coupled liquid hydrogen moderator. These areas are 'pie-like' slices, surrounding the SNS target monolith. An overhead view of the FnPB is presented in Fig. 3. FnPB delivers
Figure 4: Schematic view of the Beamline 13 components, including bender guide, choppers, monochromator assembly, secondary shutters (cold and UCN lines), as well as straight (A cold) and expanding (B UCN) guides. See text for details. / /
neutrons to two smaller experimental areas via a beamline which is split shortly after it exits the target shielding monolith. Beamline 13A, or the 'UCN' line, delivers 8.9 Å neutrons to an external building expected to host an experiment to search for the neutron electric dipole moment [22]. These neutrons will be converted to UCN via the super-thermal processes in 4 He [23]. The external FnPB building begins at the outside wall of the SNS target building. Its volume is 2789.2 m 3 with a useful floor area of 220.18 m 2 and height of the crane hook 7.931 m. The bridge of the crane is parallel to the wall of the SNS target building. The nEDM experiment [22] requires an isolation pad, which is made of 1.22 m thick concrete, is 4.62 m wide and 7.28 m long, begins 21.90 m from the end of the UCN guide and is oriented so that its center is along the line of sight to the UCN guide. There is a second isolation pad to support the neutron guide (not yet installed) that begins at the external building boundary and extends to the main isolation pad. The guide isolation pad is made of 0.61 m thick concrete and is 1.30 m wide and 4.44 m long. The boundaries between the isolation pads and the external building floor are filled in with styrofoam.
Beamline 13B, or the 'cold' beamline, is polychromatic and completely enclosed in a cave in the SNS target building. The broad distribution of slow neutron energies in this beam can serve a suite of experiments. The boundaries of the experimental area are shown in yellow in Fig. 3, with two labyrinth walls defining the entrance. The useful space is 2.84 m across on the upstream end of the cave and 5.38 m on the upstream end of the first labyrinth wall, with 12.65 m along the direction of the beam. The experimental area has a pit in the floor to accommodate a large magnetic spectrometer for the Nab experiment [24, 25]. The pit is 4.88 m long, 2.13 m wide, 2.44 m deep and begins 16.02 m downstream of the moderator face.
## 3.1. Shielding
Anextensive shielding package that satisfies the SNS radiological requirement of < 0.25 mrem hr / combined gamma and neutron radiation at the boundary between adjacent beamlines and in the instrument hall at 2 MW power extends from the moderator outwards to the first labyrinth wall. The individual shield blocks were designed so that there is no direct line of sight through joints extending from the source to the experimental area. The shielding configuration starts with 2.5-5.1 cm of steel around the guide, which staggers periodically along the beamline to prevent neutrons and gammas streaming through gaps in the shielding. The steel is followed by high density (HD) concrete up to 9 m from the moderator. The height of the beamline shielding is 4.45 m from the instrument floor and extends from the source out to a distance of 7.4 meters. Here, the shielding height drops to 3.96 m and continues to the experimental cave, ≈ 15.2 m downstream of the moderator. The cave is covered by a 45 cm thick regular concrete roof. A 4.5 m 3 heavy concrete cube beamstop begins at 6.91 m from the upstream end of the cave. Its support stand is 1.78 m wide, 1.65 m deep, 96.5 m tall and made of regular concrete. On the upstream face of the concrete beamstop, a 0.30 × 0.3 × 0.3 m 3 cutout has been made to accommodate a lithium carbonate neutron beam stop at the rear. This cutout is made at beam height, meaning its center is 1.8 m from the cave floor.
In addition to the radiological shielding requirements, SNS guidelines also dictate that the magnetic field should be < 50 mG at the boundaries between adjacent beamlines. The FnPB itself requires no modifications to comply with this requirement. However, before initial commissioning of 13B, magnetic shielding was installed for the NPDGamma experiment [19, 20, 21]. The walls and ceiling of the experimental cave are lined with 0.635 cm plates of A1010 steel, whereas the walls are lined with multiple plates to give 2.54 cm on the beam-left side and 2.54-5 cm on the beam-right side.
## 3.2. Neutron Guide
Slow neutrons can be reflected from polished surfaces of various materials. For most materials, which possess neutron refractive indices slightly smaller than 1, neutrons will undergo total external reflection from the optical potential Vopt of the medium if their angle of incidence is below a critical angle θ c , determined by the material density, the neutron coherent scattering length, and the neutron wavelength. This critical angle is set by the condition q 2 / 2 m < Vopt where q is the momentum transfer perpendicular to the surface. In the case of natural nickel, the critical angle is θ c = 1 73mrad . ∗ λ , where λ is given in Å. The critical momentum transfer is then qc Ni , = 4 π λ sin θ c = 0.0217 Å -1 . Supermirrors are multilayers of reflective materials arranged to increase the reflectivity above the critical angle of a uniform medium through interference scattering. The e ff ective critical angle for a supermirror is commonly specified by a parameter m given in multiples of qc Ni , .
The upstream portion of the neutron guide is curved in order to minimize background from fast neutrons and gammas inside the experimental area. The beamline begins 1 m from the face of the hydrogen moderator with a straight, rectangular shaped 'core' guide that is 10 cm in width and 12 cm in height. This first guide element is a 1.275 m long, m = 3 6 supermirror guide. .
Following the core guide is a ∼ 4 5 m long, five channel bender with a 117 m radius bend . towards the beam-left direction. The bender is composed of a series of straight, ∼ 0 5 m long . sections of guide, each rotated about 0.22 ◦ relative to the previous section. Each segment consists of 5 vertical channels, separated by partitioning septa of thickness 0.55 mm. The top, bottom, and beam-right faces of the guide are coated with m = 3 8 supermirror. The beam-left faces are .
coated with m = 2 3 supermirror. The first 1.8 m of this bender (four sections) is located in the . primary shutter housing and is translated out of the beam path when the shutter is closed. The next 1.2 m of bender guide contains three sections and extends from the shutter housing to the first chopper housing. An additional 1.2 m (three sections) of bender follows from the chopper to the monochromator. Finally, 0.3 m of bender extends from the monochromator housing to the straight guide. Loss of direct line-of-sight to the moderator occurs at a distance of 7.5 m from the moderator.
The remaining guide is straight and extends from the end of the bender guide to the experimental area, terminating at 15 m from the face of the moderator. There are small gaps in this guide for the remaining three choppers as well as the secondary shutter. All surfaces of the straight guide are m = 3 6 supermirror and there are no partitioning septa. .
All of the guide sections are enclosed in a vacuum tube evacuated with a dry roughing pump. All of the guide substrates are glass except the core guide, which is polished aluminum cooled with helium gas to lower the temperature and suppress possible interdi ff usion of the supermirror layers which could degrade their reflectivity.
## 3.3. Choppers
The neutron choppers are rotating disks (axis of rotation parallel to the neutron beam) coated with a neutron-absorbent material, containing one or more apertures to allow neutrons to pass through. By adjusting the chopper phase relative to the proton pulse such that the aperture is aligned with the beam when neutrons of a desired energy (corresponding to a particular time of flight to the chopper) arrive at the chopper - a neutron energy band can be selected.
The FnPB includes two choppers located 5.5 m and 7.5 m downstream of the moderator, as well as housing to accommodate two more choppers at 9 m and 10.5 m if needed for specific experiments. The gaps in the guide are 58 mm for the first chopper position, and 56 mm for the remaining 3 positions. The FnPB choppers spin at 3600 RPM as the neutron pulses are delivered at 60 Hz. The chopper disks are made of carbon fiber composite with a coating containing 10 B (minimum of 0.13 g cm 2 ) to absorb neutrons. The physical parameters of the choppers are listed / in Table. 1.
Table 1: Parameters of the FnPB choppers
| | Chopper 1 | Chopper 2 |
|---------------------|-------------|-------------|
| Axis to beam center | 25.0 cm | 25.0 cm |
| Outer diameter | 63.7 cm | 63.7 cm |
| Cutout Angle | 131 ◦ | 167 ◦ |
| Cutout inner radius | 18.6 cm | 18.6 cm |
An example of a chopped spectrum is shown in Fig. 5. Here, the first chopper was parked open, and the phase delay of the second chopper relative to the proton pulse was scanned. These data were taken during a special 5 Hz operating mode of the SNS, which is normally operated at 60 Hz.
Figure 5: Unchopped spectrum (both choppers parked open) is shown in blue. The other spectra correspond to chopper 1 parked open and chopper 2 running with the phase delays indicated.
## 3.4. Monochromator Crystals
Adouble-crystal monochromatorsystem based on the principles outlined in Ref. [26] was designed to direct the 8.9 Å beam towards the external building . If the two reflections in the doublecrystal monochromatorwere from lattices with the same d -spacing, the outgoing monochromated beam would emerge parallel to the cold beamline at its intersection with the first monochromating crystal. Given the location of the external building and the desire to maximize the floor space for cold neutron experiments, one would ideally like for the monochromatic beam to be directed along the side wall of the 13B enclosure, which is at an angle of ∼ 10 ◦ relative to the cold beamline. By using a combination of two di ff erent intercalant atoms [27], and thus crystal spacings, we direct the monochromated beam at an angle of 9 ◦ relative to the cold beamline, along the said side wall, maximizing the separation of the two beamlines.
The first monochromator is stage-1 potassium intercalated graphite, and consists of an array of 24 crystals, each having dimensions 20 mm x 45 mm for a total area of 120 mm x 180 mm. It intersects the full beam (100 mm x 120 mm) and reflects neutrons of 8.90 Å with high probability, as well as λ/ n wavelengths with lower probability. A mosaic of ∼ ◦ 3 was chosen to match the divergence of the neutron guide.
Between the two monochromators, pyrolytic graphite crystals were inserted to remove the unwanted λ/ n wavelength neutrons, one oriented to reflect ≈ 4.45Å ( λ/ 2) neutrons and the other oriented to reflect 2.97Å ( λ/ 3) neutrons. The mosaic widths (5 ) of these crystals are broad ◦ enough that precise wavelength tuning is not required. Note that neutrons of 2.23Å ( λ/ 4) are also reflected in second order by the λ/ 2 filter.
The second monochromator, stage-1 rubidium intercalated graphite, consists of an array of 35 crystals, each having dimensions 20 mm x 45 mm for a total area of 140 mm x 225 mm. This monochromator reflects the beam of 8.90Å neutrons (as well as any remaining λ/ n neutrons)
Figure 6: Top view of the monochromator. First monochromator crystals are potassium intercalated in graphite, second monochromator crystals are rubidium intercalated in graphite, with pyrolytic graphite crystals in between. See text for details.
into the UCN ballistic guide. The orientations of the crystals as well as the primary and reflected beams are shown in Fig. 6. Detailed characterization measurements were carried out at the FRM2 reactor in Munich with results presented in Ref. [27].
The monochromator housing begins 6.5 m downstream of the moderator and is accommodated by a 29.6 cm gap in the guide. Fig. 7 shows the arrangement of the crystals inside the monochromator housing. All components are mounted to the top flange of the chamber for ease of service and in addition, the chamber itself is designed to allow for the first crystal assembly to be retracted from the cold beam when not in use to avoid neutron damage as well as neutron loss in the cold beam.
## 3.5. Cold Beamline
Asecondary shutter on the cold beamline is located beyond the end of the loss of sight to the moderator, starting 10.5 m downstream of the moderator face. It consists of a steel drum with a rotating cylinder inside containing 0.5 m of neutron guide in the open position and 0.48 m of steel in the closed position. The upstream end of the steel includes a a beamstop of 2 cm think 6 Li-phosphate tile weighing 218 g and containing 33.13 g of 6 Li.
The spectral flux was measured at the end of the cold beamline 15.15 m from the moderator face using an e ffi ciency calibrated 3 He proportional counter [28]. As the neutrons make several bounces o ff the walls of the guide, the beam phase space is presumed to have little position dependence over the area of the guide exit. The energy of the neutrons was calculated based on time of flight from the moderator to the 3 He detector and calibrated using aluminum Bragg edges from windows along the beamline. This measured spectrum is compared to the flux spectrum calculated using a McStas model of the beamline [29] in Figure 8.
Figure 7: Monochromator crystal assemblies: (A) potassium intercalated in graphite, (B) rubidium intercalated in graphite, and (C) graphite λ/ n filters, along with their housing. On the right, crystal assembly A can be retracted into the top of the housing when BL13A is not taking data. See text for details.
## 3.6. UCN beamline
The UCN beamline continues downstream of the monochromator assembly with an 8 m section of 'ballistic' guide [30] in order to optimize neutron transport over long distances. The guide starts with a rectangular cross section 12 cm wide and 14 cm tall and expands to 20 cm by 30 cm. The top and bottom of the guide have reflectivity of m = 3 6 and the sides have . m = 2 2. .
The transverse distribution of neutrons passing through the monochromator assembly was measured at the end of the 8 m long ballistic section of the UCN guide. Several techniques were employed, including using the previously mentioned absolutely calibrated 3 He detector in combination with neutron-sensitive image plates, as well as a CCD camera [31]. The time-offlight spectrum was measured with a 3 He detector behind a 1.27 cm radius pinhole, to keep the rates down. The image plates and CCD camera were used to establish and confirm the fraction of the total flux seen through the pinhole.
The spectrum of neutrons from the monochromatic beam is shown in Fig. 9 along with the calculated flux from McStas. In the model, the peak reflectivities of the first and second monochromators were taken to be 0.675 and 0.63, respectively. In Ref. [27], the ranges of the crystal reflectivites are given as 70-80% for the potassium and 65-75% for the rubidium monochromators. The middle of the range was used, with a 10% reduction due to neutron attenuation. Additionally, in the Ref. [27] measurement, each crystal was tested individually, whereas
Figure 8: Measured and calculated neutron brightness per MW of proton power on the spallation target for the cold guide at the FnPB as a function of neutron wavelength.
in the FnPB measurement, the crystals were arranged in 2D monochromator arrays. To allow for possible misalignments, the peak reflectivities were scaled down by another 5%. The mosaics for each individual crystal were specified in the model. The disagreement between the modelled and measured spectra at 8.9Å (the wavelength of interest) is almost a factor of two. This is not currently understood. The measurements reported in Ref. [27] yield peak reflectivities that are down 15% (K-intercalated crystals) and 25% (Rb-intercalated crystals) from ideal values. The reflectivities are a function of wavelength, possibly explaining a small fraction of the di ff erence. It's also possible that some of the disagreement is due to the imperfect modeling of guide. The decrease in measured flux from what was expected based on the McStas model has caused a modification in the planning of the nEDM experiment [32, 33], which is now expected to be using the cold beamline, BL13B.
## 4. Summary
The FnPB beamline at the Spallation Neutron Source has been comissioned and is now in operation for science experiments. Its measured performance is in reasonable agreement with simulations conducted in the design phase of the facility. Physics proposals are reviewed by the Fundamental Neutron Physics Proposal and Advisory Committee. NPDGamma [19, 20, 21], the first of the approved peer-reviewed experiments, has recently been completed. It will be followed by the n -3 He hadronic parity violation experiment [34], the Nab beta decay experiment [24, 25], and the nEDM experiment [32, 33].
Figure 9: Measured spectrum for the UCN guide at the FnPB. A chopper upstream of the monochromator was used in order to separate neutrons of di ff erent wavelengths and several measurements at di ff erent chopper settings were combined to produce the final spectrum. Data were taken at a reduced repetition rate of 10 Hz.
## 5. Acknowledgements
W. M. Snow acknowledges support from NSF grants # NSF PHYS-0457219 and NSF PHYS0758018, as well as the Indiana University Center for Spacetime Symmetries. The material is based upon work supported by the U.S. Departmet of Energy, O ffi ce of Science, O ffi ce of Nuclear Physics, under award numbers DE-FG02-03ER41258 and DE-FG02-97ER41042.
## References
- [1] J. S. Nico, W. M. Snow, Experiments in Fundamental Neutron Physics, Annual Reviews of Nuclear and Particle Science 55 (2005) 27.
- [2] H. Abele, The neutron. Its properties and basic interactions, Progress in Particle and Nuclear Physics 60 (2008) 1.
- [3] D. Dubbers, M. G. Schmidt, The Neutron and Its Role in Cosmology and Particle Physics, Reviews of Modern Physics 83 (2011) 1111.
[4] W. M. Snow, Fundamental Neutron Physics with Long Pulsed Spallation Sources, Physics Procedia 51 (2014) 31.
- [5] P. N. Seo, et al., A measurement of the Flight Path 12 cold H2 moderator brightness at LANSCE, Nuclear Instruments and Methods in Physics Research Section A 517 (2004) 285.
- [6] Y. Arimoto, H. Funahashi, N. Higashi, M. Hino, K. Hirota, et al., Present status of neutron fundamental physics at J-PARC, Prog. Theo. Exp. Phys. 2012 (2012) 02B007, doi:10.1093 ptep pts075. / /
[7] K. Mishima, et al., Design of neutron beamline for fundamental physocs at J-PARC BL05, Nucl. Instr. and Meth. A 600 (2009) 342-345.
[8] W. E. Fischer, et al., SINQ - The spallation neutrons source, a new research facility at PSI, Physica B 234 (1997) 1202-1208.
[9] G. Bauer, Operation and development of the new spallation neutron source SINQ at the Paul Sherrer Institute, Nuclear Instruments and Methods in Physics Research Section B 139 (1998) 65.
[10] A. Schebetov, et al., New Facility for fundamental research in nuclear physics with polarized cold neutrons at PSI, Nuclear Instruments and Methods in Physics Research Section A 497 (2003) 479.
- [11] J. Zejma, et al., FUNSPIN polarized cold-neutron beam at PSI, Nuclear Instruments and Methods in Physics Research Section A 539 (2005) 622.
- [12] H. Abele, et al., Characterization of a ballisic supermirror neutron guide, Nuclear Instruments and Methods in Physics Research Section A 562 (2006) 407.
- [13] J. S. Nico, et al., The Fundamental Neutron Physics Facilities at NIST, J. Res. Natl. Inst. Stand. Technol 110 (2005) 137.
- [14] J. C. Cook, Design and estimated performance of a new neutron guide system for the NCNR expansion project, J. Res. Natl. Inst. Stand. Technol 80 (2009) 023101.
- [15] M. P. Stockli, B. Han, S. N. Murray, T. R. Pennisi, M. Santana, R. F. Welton, Ramping up the Spallation Neutron Source beam power with the H- source using 0 mg Cs day, Rev. Sci. Instrum. / 81 (2010) 02A729.
- [16] Spallation Neutron Source Project Completion Report, 2006.
- [17] S. Henderson, et al., The Spallation Neutron Source accelerator system design, Nuclear Instruments and Methods in Physics Research Section A, In Press.
- [18] E. B. Iverson, P. D. Ferguson, F. X. Gallmeier, I. I. Popova, Detailed SNS neutronics calculations for scattering instrument design: SCT configuration, Tech. Rep. SNS 110040300-DA0001-R00, Oak Ridge National Laboratory, 2002.
- [19] J. D. Bowman, Precision Measurement of A γ in n + p → d + γ , SNS proposal, URL http://npdgamma.com , 2005.
- [20] W. M. Snow, et al., Measurement of the parity Violating Asymmery in n + p → D + γ , Nucl. Instr. and Meth. A 440 (2000) 729-735.
- [21] N. Fomin, First results from the NPDGamma experiment at the spallation neutron source, AIP Conf.Proc. 1560 (2013) 145-148, doi:10.1063 1.4826740. /
- [22] S. Lamoreaux, R. Golub, Recent progress in the development of a new technique to measure the neutron electric dipole moment (1995) 597-604.
- [23] R. Golub, J. M. Pendlebury, Super-thermal sources of ultra-cold neutrons, Phys. Lett. A (1975) 133-135.
- [24] J. D. Bowman, D. Poˇ cani´ c, Precise Measurement of the Neutron Beta Decay Parameters, 'a' and 'b', SNS proposal, URL http://nab.phys.virginia.edu , 2007.
- [25] D. Poˇ cani´, c et al., Nab: Measurement Principles, Apparatus and Uncertainties, Nucl. Instr. Meth. 611 (2009) 211-215.
- [26] C. E. H. Mattoni, C. P. Adams, K. J. Alvine, J. M. Doyle, S. N. Dzhosyuk, R. Golub, E. Korobkina, D. N. McKinsey, A. K. Thompson, L. Yang, H. Zabel, P. R. Hu ff man, A long wavelength neutron monochromator for superthermal production of ultracold neutrons, Physica B: Condensed Matter 344 (14) (2004) 343 - 357.
- [27] P. Courtois, C. Menthonnex, R. Hehn, K. Andersen, V. Nesvizhevsky, O. Zimmer, F. Piegsa, P. Geltenbort, G. Greene, R. Allen, P. Hu ff man, P. Schmidt-Wellenburg, M. Fertl, S. Mayer, Production and characterization of intercalated graphite crystals for cold neutron monochromators, Nuclear Instruments and Methods in Physics Research Section A 634 (1, Supplement) (2011) S37 - S40.
- [28] E. B. Iverson, B. J. Micklich, D. V . Baxter, R. G. Cooper, P. D. Ferguson, D. W. Freeman, F. X. Gallmeier, S. E. Hammons, C. M. Lavelle, I. Popova, Neutronic measurements to commission the SNS, in: Proceedings of ICANS XVII, the Seventeenth Meeting of the International Collaboration on Advanced Neutron Sources, 2005.
- [29] P. R. Hu ff man, L. Greene, G., R. Allen, V. Cianciolo, P. Koehler, D. Desai, R. Mahurin, A. Yue, G. R. Palmquist, M. Snow, W., Beamline Performance Simulations for the Fundamental Neutron Physics Beamline at the Spallation Neutron Source, Journal of Research of the National Institute of Standards and Technology 110 (2005) 161-168.
- [30] T. M. Ito, C. B. Crawford, G. L. Greene, Optimization of the Ballistic Guide Design for the SNS FNPB 8.9 A Neutron Line, Nucl. Instr. Meth. A564 (2006) 414-423.
- [31] A. C. Hamilton, E. B. Iverson, Diagnostic Use of Neutron-Sensitive Image Plates at ORNL Neutron Facilities, in: Proceedings of the Tenth International Meeting on Nuclear Applications of Accelerators-AccApp'11, 235, 2012.
- [32] M. Cooper, S. Lamoreaux, A New Search for the Neutron Electric Dipole Moment, DOE proposal, URL http://www.phy.ornl.gov/nedm/ , 2002.
- [33] R. Golub, K. Lamoreaux, Neutron electric dipole moment, ultracold neutrons and polarized He-3, Phys.Rept. 237 (1994) 1-62.
- [34] J. D. Bowman, C. Crawford, M. Gericke, A Measurement of the Parity Violating Proton Asymmetry in the Capture of Polarized Cold Neutrons on 3 He, SNS proposal, 2007.
|
10.1016/j.nima.2014.10.042
|
[
"N. Fomin",
"G. L. Greene",
"R. Allen",
"V. Cianciolo",
"C. Crawford",
"T. Ito",
"P. R. Huffman",
"E. B. Iverson",
"R. Mahurin",
"W. M. Snow"
] |
2014-08-04T18:03:16+00:00
|
2014-08-04T18:03:16+00:00
|
[
"physics.ins-det",
"nucl-ex",
"physics.acc-ph"
] |
Fundamental Neutron Physics Beamline at the Spallation Neutron Source at ORNL
|
We describe the Fundamental Neutron Physics Beamline (FnPB) facility located
at the Spallation Neutron Source at Oak Ridge National Laboratory. The FnPB was
designed for the conduct of experiments that investigate scientific issues in
nuclear physics, particle physics, astrophysics and cosmology using a pulsed
slow neutron beam. We present a detailed description of the design philosophy,
beamline components, and measured fluxes of the polychromatic and monochromatic
beams.
|
1408.0754v1
|
## Lepton Flavor Violation in Flavored Gauge Mediation
## Lorenzo Calibbi , Paride Paradisi a b,c,d , Robert Ziegler e,f
a Service de Physique Th´orique, Universit´ Libre de Bruxelles, B-1050 Brussels, Belgium e e b Dipartimento di Fisica e Astronomia, Universit´ di Padova, Via Marzolo 8, I-35131 Padova, Italy a c INFN Sezione di Padova, Via Marzolo 8, I-35131 Padova, Italy d SISSA, Via Bonomea 265, I-34136 Trieste, Italy e Sorbonne Universit´s, UPMC Univ Paris 06, UMR 7589, LPTHE, F-75005, Paris, France e f CNRS, UMR 7589, LPTHE, F-75005, Paris, France
## Abstract
We study the anatomy and phenomenology of Lepton Flavor Violation (LFV) in the context of Flavored Gauge Mediation (FGM). Within FGM, the messenger sector couples directly to the MSSM matter fields with couplings controlled by the same dynamics that explains the hierarchies in the SM Yukawas. Although the pattern of flavor violation depends on the particular underlying flavor model, FGM provides a built-in flavor suppression similar to wave function renormalization or SUSY Partial Compositeness. Moreover, in contrast to these models, there is an additional suppression of left-right (LR) flavor transitions by third-generation Yukawas that in particular provides an extra protection against flavor-blind phases. We exploit the consequences of this setup for lepton flavor phenomenology, assuming that the new couplings are controlled by simple U (1) flavor models that have been proposed to accommodate large neutrino mixing angles. Remarkably, it turns out that in the context of FGM these models can pass the impressive constraints from LFV processes and leptonic EDMs even for light superpartners, therefore offering the possibility of resolving the longstanding muon g -2 anomaly.
## 1 Introduction
One of the longstanding problems in particle physics is the origin of flavor hierarchies in the Standard Model (SM). The most popular attempt to address this problem is in terms of flavor symmetries in which the flavor hierarchies arise from a suitable symmetry breaking pattern. Among the numerous possibilities, the simplest models are based on a single U (1) flavor symmetry [1-4]. In the quark sector this ansatz works pretty well and can account for all hierarchies in quark masses and mixing, with an order-of-magnitude prediction V ub ∼ V us V cb that is in good agreement with data. Also in the lepton sector a single U (1) works very well, since charge lepton mass hierarchies can arise from large charge differences of right-handed leptons, while large mixing angles are due to small charge differences of left-handed leptons. In this way U (1) models can naturally realize the paradigm of an 'anarchical' structure [5-8] in lepton mixing, which has recently received renewed attention [9-11] after the reactor neutrino angle θ 13 turned out to be sizable.
Independently of the nature of the underlying flavor symmetry, the crucial question about these kind of models regard their predictivity. Since flavor models aim at explaining the origin of dimensionless Yukawa couplings, there is no preferred mass scale of the new degrees of freedom. As new effects in the SM flavor sector are suppressed by this mass scale, there are no observable deviations from the SM flavor predictions, unless this scale is unexpectedly light [12]. Therefore the only way to test these models in laboratory experiments for a highscale flavor sector is the presence of new physics around the TeV scale, as suggested by the hierarchy problem. If such physics comes with a flavor structure, it can possibly carry down the information of the high-scale flavor sector to the TeV scale and lead to testable predictions for precision flavor observables. The prime example is Supersymmetry (SUSY), which in the case of high-scale mediation of SUSY breaking around or above the flavor sector scale directly contains the imprint of the flavor symmetry in the soft-breaking sfermion masses.
However, within the context of Gravity Mediation simple U (1) models are in big trouble as the suppression of flavor violation is too weak. The reason is that off-diagonal entries in the left-handed and right-handed sfermion mass matrices are suppressed by the differences of the corresponding charges due to their non-holomorphic nature. In the left-handed sector these charge differences are directly related to mixing angles, which for the first two generations are sizable both in the lepton and quark sector. Since the strongest constraint precisely arise from observables involving light families, like glyph[epsilon1] K and µ → eγ , such U (1) flavor models in the context of Gravity Mediation are essentially incompatible with SUSY around the TeV scale. 1
The situation is completely different in Gauge Mediation (see Ref. [14]), where the SUSY breaking and the flavor sector can be decoupled. Indeed, if the flavor scale is much higher than the SUSY messenger scale then soft masses are screened from the high-energy flavor sector and have a flavor structure determined only by SM Yukawas, thus realizing the paradigm
1 This conclusion holds even under the assumption of a mechanism inducing degenerate sfermion masses at the SUSY breaking mediation scale. In fact, the assumed flavor universality is broken by the renormalization group evolution of the soft masses down to the flavor breaking scale, so that large flavor mixing is anyway generated at the level predicted by the U (1) symmetry [13].
of Minimal Flavor Violation (MFV) [15]. While this scenario provides a very appealing mechanism to solve the SUSY flavor problem, the imprint of the flavor sector in low-energy physics and thus the possibility to test flavor symmetry models is completely lost.
It is therefore interesting to construct extensions of Minimal Gauge Mediation (MGM) that re-introduce the dependence on the underlying flavor sector, and thus lead to a broad variety of sfermion flavor structures beyond MFV. An example for such extensions is provided by a class of models that has been dubbed 'Flavored Gauge Mediation' (FGM) [16]. In these scenarios, new direct couplings between the messengers and the MSSM matter fields are introduced with a flavor structure that is assumed to be controlled by the same underlying flavor symmetry that explains the smallness of the Yukawas. 2 These couplings generate new contributions to sfermion masses (on top of of the flavor-universal MGM ones) that are controlled by the underlying flavor symmetry. Interestingly, due to the loop origin of the soft terms, there is a built-in suppression of flavor violation that is independent of the underlying flavor model [28]. This implies that even single U (1) flavor models are perfectly viable (in contrast to Gravity Mediation), as the flavor pattern of the resulting sfermion masses resembles the suppression in wave function renormalization [29,30] or SUSY Partial Compositeness [31,32]. Moreover, in contrast to those scenarios, there is also a built-in suppression of LR flavor transitions and in particular flavor-blind phases by third-generation Yukawas, which becomes very efficient in the down and charged-lepton sector provided tan β is not very large.
While in Ref. [28] we have focused on the quark sector, in this paper we analyze the impact of FGM models with underlying U (1) flavor models on the lepton sector. There are good arguments that motivate this study: i) in contrast to the quark sector the large neutrino mixing angles require milder hierarchies in left-handed charges, leading in turn to weaker suppression in the left-handed slepton sector and therefore potentially large effects in LFV processes, ii) the experimental bounds on LFV channels with an underlying µ → e transition as well as the electron EDM underwent recently a very significant improvement challenging many models with New Physics (NP) at the TeV scale, even with modest sources of flavor violation. Therefore the major aim of this work is to analyze whether and to which extent single U (1) flavor models for the lepton sector are viable in the context of FGM. A related question is whether we can account for the current muon g -2 anomaly, that is if light sleptons are still allowed by the LFV and EDM bounds (for a general discussion on the interrelationship of leptonic dipoles see Ref. [33]).
The rest of the paper is organized as follows: in Section 2 we recall the main ingredients of FGM models providing explicit expressions for the soft masses in the slepton sector. Concrete examples of U (1) leptonic flavor models and their imprint in the soft sector are presented in Section 3. The low-energy phenomenology of FGM models supplemented by the above U (1) flavor models is studied in Section 4. In Section 5, we compare the flavor structure of the soft terms and related phenomenological implications of FGM models to U (1) models with Gravity Mediation and models with SUSY Partial Compositeness. We conclude in Section 6.
2 Such matter-messenger couplings have recently received new interest, as they allow to obtain a large Higgs mass with light stops by generating non-vanishing A-terms at the messenger scale [17-27].
In an appendix we collect the formulae for LFV branching ratios, lepton anomalous magnetic moments and lepton electric dipole moments using a generalized mass insertion approximation without assuming large tan β , thus improving on existing results that take into account only the tan β enhanced terms.
## 2 Flavored Gauge Mediation
We begin with a brief review of Minimal Gauge Mediation (see Ref. [14]). In this scenario N copies of heavy chiral superfields Φ i +Φ in i 5 + 5 of SU(5) are introduced. These messenger fields couple directly to the SUSY breaking sector, which is effectively parameterized by a single spurion field X that gets a vev 〈 X 〉 = M + Fθ 2 . Through the following superpotential coupling
$$W \supset X \overline { \Phi } _ { i } \Phi _ { i }, \quad i = 1 \dots N \,,$$
the messengers acquire large supersymmetric mass terms M and SUSY breaking masses proportional to F . By integrating out the messengers at loop-level, soft terms are generated. At the messenger scale, A-terms vanish and gaugino masses and sfermion masses are given by
$$M _ { i } ( M ) = N \frac { \alpha _ { i } ( M ) } { 4 \pi } \ \Lambda, \quad \quad \ \Lambda = \frac { F } { M },$$
$$m _ { \tilde { f } } ^ { 2 } ( M ) = 2 N \sum _ { i = 1 } ^ { 3 } C _ { i } ( f ) \stackrel { \dots } { \frac { \alpha _ { i } ^ { 2 } ( M ) } { ( 4 \pi ) ^ { 2 } } } \ \Lambda ^ { 2 }, \quad \ f = q, \, u, \, d, \dots,$$
where C f i ( ), i = 1 2 3 is the quadratic Casimir of the representation of the field , , f under the gauge group SU(3) × SU(2) × U(1).
Since the messengers have the same gauge quantum numbers as the MSSM Higgs fields, in addition to the Yukawa couplings
$$W \supset ( y _ { U } ) _ { i j } Q _ { i } U _ { j } H _ { u } + ( y _ { D } ) _ { i j } Q _ { i } D _ { j } H _ { d } + ( y _ { E } ) _ { i j } L _ { i } E _ { j } H _ { d } \,,$$
also direct couplings of messengers to MSSM fields are allowed by the gauge symmetries. If we restrict to R-parity even messenger fields, 3 the messengers can couple only to the MSSM matter fields. For the messenger doublets these couplings read in general
$$\Delta W = ( \lambda _ { U } ) _ { i j } Q _ { i } U _ { j } \Phi _ { H _ { u } } + ( \lambda _ { D } ) _ { i j } Q _ { i } D _ { j } \overline { \Phi } _ { H _ { d } } + ( \lambda _ { E } ) _ { i j } L _ { i } E _ { j } \overline { \Phi } _ { H _ { d } },$$
where Φ H u , Φ H d denote the SU(2) doublet components of the 5 5 , messengers, and we restricted to the case of one messenger pair for simplicity.
The presence of direct messenger-matter couplings gives rise to new contributions to sfermion masses and A-terms with a flavor structure that depends on the new parameters
3 For R-parity odd messengers similar couplings to Higgs fields are allowed and have been discussed in the literature, see e.g. [25]. Note that in this way one preserves the MFV structure of MGM. Here we are instead interested in non-trivial flavor structures.
λ ij . If these couplings were flavor-anarchic O (1) numbers, the elegant solution of Gauge Mediation to the SUSY flavor problem would be completely spoiled. Therefore it is usually assumed that all direct couplings of the messengers to matter fields vanish, which can be enforced for example by introducing a new Z 2 symmetry under which MSSM fields are even and messengers are odd. Note that this symmetry extends to a full accidental U (1) symmetry in the case of one messenger pair
$$\Phi \to e ^ { i \alpha } \Phi, & & \overline { \Phi } \to e ^ { - i \alpha } \overline { \Phi }.$$
However, in order to preserve the neat solution of the SUSY flavor problem in MGM, it is enough that the new couplings in Eq. (5) are just sufficiently small. Such small couplings can be easily motivated in the context of flavor models, since they break the global flavor symmetries of MSSM kinetic terms exactly as the Yukawas, and therefore they can naturally have a similar hierarchical structure. This can be realized in explicit flavor models in which the messenger fields transform like the Higgs fields (in particular one can choose that they do not transform at all under the flavor sector), which implies that the new couplings have the same parametric suppression as the Yukawas,
$$\lambda _ { U } & \sim y _ { U }, & \lambda _ { D } & \sim y _ { D }, & \lambda _ { E } & \sim y _ { E }.$$
Following Ref. [16], we refer to these kind of models as 'Flavored Gauge Mediation' (FGM).
The new contributions to soft terms induced by the couplings in Eq. (5) can be calculated using the general expressions in Ref. [25]. At leading order in SUSY breaking one finds new contributions to sfermion masses at 2-loop and non-vanishing A-terms at 1-loop. While these new effects can have interesting consequences for the low-energy spectrum [34, 35], here we are mainly interested in the flavor structure of the new contributions to sfermion masses, in particular in the slepton sector. Therefore we will now take a bottom-up point of view and restrict the analysis to the consequences of the presence of the λ E coupling for the lepton sector. We will not discuss the impact of other possible messenger-matter couplings on the low-energy spectrum, in particular the mass of the lightest Higgs boson. We just note that the Higgs mass does not represent a serious constraint in these kind of models, and can be due to large A-terms or an implementation in the NMSSM. The latter also represents a natural possibility to generate the µ -term and to elegantly solve the µ -B µ problem of Gauge Mediation, since in the NMSSM the general structure of FGM motivates a direct coupling of the NMSSM singlet to the messengers which can easily allow for correct EWSB [21,36].
Furthermore, let us notice that the couplings λ E do not deform the spectrum predicted by the underlying gauge mediation scheme, at least for low to moderate values of tan β , as we are going to consider in the next sections. In particular, if m h ≈ 126 GeV is accounted for by a large top A-term, induced by an O (1) coupling ( λ U ) 33 in Eq. (5), the spectrum would resemble the one discussed e.g. in Ref. [28]. This would have interesting consequences for the leptonic sector we consider here, since ( λ U ) 33 = O (1) also suppresses the masses of the left-handed sleptons, through an induced Fayet-Iliopoulos term, thus naturally accomodating the Higgs mass with a light slepton spectrum that can give a sizable contribution to the muon g -2 [19,28].
For soft terms in the slepton sector we use the conventions
$$\mathcal { L } & \supset - \left ( ( \tilde { m } _ { L } ^ { 2 } ) _ { i j } L _ { i } L _ { j } ^ { \dagger } + ( \tilde { m } _ { E } ^ { 2 } ) _ { i j } E _ { i } ^ { \dagger } E _ { j } + ( A _ { e } ) _ { i j } L _ { i } E _ { j } H _ { d } \right ) | _ { s c a l a r } \\ & = - \left ( \tilde { l } _ { L } ^ { T } \tilde { m } _ { L } ^ { 2 } \tilde { l } _ { L } ^ { * } + \tilde { e } _ { R } ^ { T } ( \tilde { m } _ { E } ^ { 2 } ) _ { i j } \tilde { e } _ { R } ^ { * } + \tilde { l } _ { L } ^ { T } A _ { e } \tilde { e } _ { R } ^ { * } H _ { d } \right ),$$
where the first line denotes the scalar components of superfields. Using the results of Ref. [25], the presence of λ E gives rise to the following expressions for the non-holomorphic masses 4
$$\mu _ { L } ^ { 2 } \ & = \ \frac { \Lambda ^ { 2 } } { 2 5 6 \pi ^ { 4 } } \left [ N \left ( \frac { 3 } { 2 } g _ { 2 } ^ { 4 } + \frac { 3 } { 1 0 } g _ { 1 } ^ { 4 } \right ) - \left ( \frac { 9 } { 5 } g _ { 1 } ^ { 2 } + 3 g _ { 2 } ^ { 2 } \right ) \lambda _ { E } \lambda _ { E } ^ { \dagger } \\ & \quad + \ 3 \lambda _ { E } \lambda _ { E } ^ { \dagger } \lambda _ { E } \lambda _ { E } ^ { \dagger } + 2 \lambda _ { E } y _ { E } ^ { \dagger } y _ { E } \lambda _ { E } ^ { \dagger } - 2 y _ { E } \lambda _ { E } ^ { \dagger } \lambda _ { E } y _ { E } ^ { \dagger } \\ & \quad + \ \lambda _ { E } \lambda _ { E } ^ { \dagger } \text{Tr} \left ( \lambda _ { E } \lambda _ { E } ^ { \dagger } \right ) + y _ { E } \lambda _ { E } y _ { E } ^ { \dagger } \text{Tr} \left ( y _ { E } \lambda _ { E } ^ { \dagger } \right ) \right ],$$
and
$$\tilde { m } _ { E } ^ { 2 } \ & = \ \frac { \Lambda ^ { 2 } } { 2 5 6 \pi ^ { 4 } } \left [ \frac { 6 } { 5 } g _ { 1 } ^ { 4 } N - \left ( \frac { 1 8 } { 5 } g _ { 1 } ^ { 2 } + 6 g _ { 2 } ^ { 2 } \right ) \lambda _ { E } ^ { \dagger } \lambda _ { E } + 6 \lambda _ { E } ^ { \dagger } \lambda _ { E } \lambda _ { E } ^ { \dagger } \lambda _ { E } \\ & + \ 2 \lambda _ { E } ^ { \dagger } y _ { E } y _ { E } ^ { \dagger } \lambda _ { E } - 2 y _ { E } ^ { \dagger } \lambda _ { E } \lambda _ { E } ^ { \dagger } y _ { E } + 2 \lambda _ { E } ^ { \dagger } \lambda _ { E } \text{Tr} \left ( \lambda _ { E } \lambda _ { E } ^ { \dagger } \right ) \\ & + \ 2 \lambda _ { E } ^ { \dagger } y _ { E } \text{Tr} \left ( \lambda _ { E } y _ { E } ^ { \dagger } \right ) + 2 y _ { E } ^ { \dagger } \lambda _ { E } \text{Tr} \left ( y _ { E } \lambda _ { E } ^ { \dagger } \right ) \right ],$$
while the A-terms are given by
$$A _ { E } = - \frac { \Lambda } { 1 6 \pi ^ { 2 } } \left ( \lambda _ { E } \lambda _ { E } ^ { \dagger } y _ { E } + 2 y _ { E } \lambda _ { E } ^ { \dagger } \lambda _ { E } \right ) \,.$$
Note that the flavor dependence of the above expressions can be obtained using a simple spurion analysis, taking into account also the U (1) M 'messenger number' in Eq. (6) as a spurious symmetry under which the new couplings are charged. The U (1) M symmetry prevents terms like λ E y † E and λ † E y E for the non-holomorphic masses ˜ m 2 L and ˜ m 2 E , respectively, and terms like λ E , λ E y † E λ E and λ † E y E λ † E for the A-terms. As a result, the A-terms are partially aligned to the Yukawa couplings and their diagonal components are necessarily real and therefore do not induce contributions to the EDMs.
For future convenience, we define the flavor violating mass insertions (MIs) as usual
$$( \delta _ { L L } ^ { e } ) _ { i j } = \frac { ( \tilde { m } _ { L } ^ { 2 } ) _ { i j } } { \tilde { m } _ { L } ^ { 2 } } \, \quad \ ( \delta _ { R R } ^ { e } ) _ { i j } = \frac { ( \tilde { m } _ { E } ^ { 2 } ) _ { i j } } { \tilde { m } _ { E } ^ { 2 } } \, \quad \ ( \delta _ { L R } ^ { e } ) _ { i j } = \frac { v _ { d } ( A _ { E } ) _ { i j } } { \tilde { m } _ { L } \tilde { m } _ { E } } \. \quad ( 1 2 )$$
In the limit of y E , λ E glyph[lessmuch] 1, i.e. for moderate/low tan β values, we obtain the following approximate expressions at the messenger scale:
$$( \delta ^ { e } _ { L L } ) _ { i j } \simeq - \left ( \frac { 1 0 g _ { 2 } ^ { 2 } + 6 g _ { 1 } ^ { 2 } } { 5 g _ { 2 } ^ { 4 } + g _ { 1 } ^ { 4 } } \right ) \frac { ( \lambda _ { E } \lambda _ { E } ^ { \dagger } ) _ { i j } } { N } \,, \quad ( \delta ^ { e } _ { R R } ) _ { i j } \simeq - \left ( \frac { 5 g _ { 2 } ^ { 2 } + 3 g _ { 1 } ^ { 2 } } { g _ { 1 } ^ { 4 } } \right ) \frac { ( \lambda _ { E } ^ { \dagger } \lambda _ { E } ) _ { i j } } { N } \,, \quad ( 1 3 )$$
4 We only consider terms in leading order in Λ 2 /M 2 , i.e. a messenger scale that is not particularly low.
$$( \delta _ { L R } ^ { e } ) _ { i j } \simeq - \frac { 1 } { \sqrt { \frac { 3 N } { \sqrt { 5 } } } g _ { 1 } g _ { 2 } } \ \frac { m _ { j } ^ { e } ( \lambda _ { E } \lambda _ { E } ^ { \dagger } ) _ { i j } + 2 m _ { i } ^ { e } ( \lambda _ { E } ^ { \dagger } \lambda _ { E } ) _ { i j } } { \sqrt { \tilde { m } _ { L } \tilde { m } _ { E } } } \,.$$
Few comments are in order:
- · The above MIs, as well as all superpotential couplings, are defined in the basis where we define the flavor model. In order to study their phenomenological consequences, we go to the mass basis for the charged lepton Yukawas by means of the rotation y E → V T EL E y V ER = y diag E . Under this change of basis, the spurion λ E transforms accordingly. However, one can easily check that in U (1) models with non-negative charges the parametric flavor suppression remains the same and only the O (1) coefficients change. We therefore simply ignore these differences, that is we take V T EL λ E V ER ∼ λ E .
- · Interestingly, the diagonal A-terms are real. As a result, the leading CP violating phases generating the EDMs can only arise at higher order in the MIs, through the combination ( δ e LL ik ) ( δ e LR kj ) , ( δ e LR ik ) ( δ e RR kj ) , and ( δ e LL ik ) ( δ e LR kk ) ( δ e RR kj ) when ij = 11. This however leads to an additional suppression by powers of ( λ E ) 33 ∼ y τ . 5
- · As a consequence of Eqs. (13),(14), the naive expectations for the MIs are enhanced, for a given number of messengers N , by large (mediation scale dependent) gauge factors. This is especially true in the case of ( δ e RR ij ) and, to less extent, also in the cases of ( δ e LL ij ) and ( δ e LR ij ) .
In the following, we will analyze the impact of our FGM model on the branching ratio of µ → eγ and the electron EDM which are the most powerful probes of new physics in the leptonic sector. To do so, we need to specify the underlying flavor model that controls the flavor structure of the new couplings.
## 3 Flavored Gauge Mediation and U (1) Flavor Models
While the results of the last section can be applied to any flavor model that predicts the flavor structure of y E and therefore λ E , in this section we concentrate on simple U (1) flavor models. We first recall the basic structure of these models, then we analyze their predictions for the soft terms in the lepton sector in the context of our FGM model.
## 3.1 U (1) Flavor Models
In the simplest realization of these models the flavor symmetry is spontaneously broken by the vev of a single 'flavon' field with negative unit charge. Yukawa couplings then arise from higher-dimensional operators that involve suitable powers of the flavon to make the operator invariant under the U (1) symmetry, with some undetermined coefficients that are assumed to
5 Notice that the µ -and B µ -terms, which are not controlled by GMSB, could still introduce CP violating phases depending on the underlying mechanism that generates them. However, if this mechanism is such that the phases of µ and B µ are correlated to the phase of Λ, then no phases arise from this sector [14].
be O (1). The suppression scale is the typical scale of the flavor sector that could correspond to the mass scale of Froggatt-Nielsen messengers in explicit UV completions. The Yukawas then depend only on powers of the ratio glyph[epsilon1] of flavon vev and flavor scale, which typically is taken to be of the order of the Cabibbo angle glyph[epsilon1] ∼ 0 2. . If we restrict to models where only the matter fields are charged, i.e. H u = H d = 0, we get for the lepton Yukawa couplings
$$( y _ { E } ) _ { i j } & \sim \epsilon ^ { L _ { i } + E _ { j } },$$
where L i and E i stand for the U (1) charges of the left-handed and right-handed leptons, respectively. The neutrino sector depends on the origin of neutrino masses. If neutrinos are Dirac, then the Yukawa coupling takes the same form as the charged lepton Yukawa above with E j → N j . In this case, the left-handed rotations V EL , V NL for the charged lepton and neutrino sectors, respectively, and therefore the PMNS matrix V PMNS , have the same parametric structure
$$( V _ { P M N S } ) _ { i j } & \sim ( V _ { E L } ) _ { i j } \sim ( V _ { N L } ) _ { i j } \sim \epsilon ^ { | L _ { i } - L _ { j } | }.$$
Large neutrino mixing angles can therefore be reproduced by taking small left-handed charge differences L i -L j . Instead small neutrino masses can be accommodated by taking sufficiently large charges N i of right-handed neutrinos. A more plausible explanation of light neutrinos can be achieved if they originate from the Weinberg operator
$$\Delta W = \frac { ( y _ { u } ) _ { i j } } { \Lambda } L _ { i } L _ { j } H _ { u } H _ { u },$$
with a flavor structure determined by the U (1) symmetry
$$( y _ { l l } ) _ { i j } & \sim \epsilon ^ { L _ { i } + L _ { j } }.$$
In this way the smallness of neutrino masses can be elegantly explained by assuming a large UVscale v u / Λ glyph[lessmuch] 1, but the prediction for the parametric structure of the left-handed neutrino rotations and therefore for the PMNS matrix does not change, and we still get the result of Eq. (16). One possibility for an explicit UV completion is the type-I seesaw mechanism. In this scenario, one adds three heavy right-handed neutrinos and Dirac Yukawa couplings
$$\Delta W = ( y _ { \nu } ) _ { i j } L _ { i } N _ { j } H _ { u } + \frac { 1 } { 2 } ( M _ { N } ) _ { i j } N _ { i } N _ { j },$$
with their flavor structure given by
$$( y _ { \nu } ) _ { i j } & \sim \epsilon ^ { L _ { i } + N _ { j } } & ( M _ { N } ) _ { i j } & \sim M _ { N } \epsilon ^ { N _ { i } + N _ { j } }.$$
Integrating out the right-handed neutrinos generates the Weinberg operator with a coefficient given by
$$\frac { ( y _ { l l } ) _ { i j } } { \Lambda } = - \frac { 1 } { 2 } ( y _ { \nu } M _ { N } ^ { - 1 } y _ { \nu } ^ { T } ) _ { i j }.$$
Note that in the simple U (1) models that we will consider here, the parametric flavor structure of the coefficient of the Weinberg operator is the same as in the effective theory
$$( y _ { l l } ) _ { i j } & \sim \epsilon ^ { L _ { i } + L _ { j } },$$
and therefore we recover the same estimate for the PMNS matrix as in Eq. (16).
Various U (1) models have been discussed in the literature, see e.g. [6, 10, 37, 38]. There is some ambiguity in the choice of charge assignments, since glyph[epsilon1] is typically not a very small parameter (one has glyph[epsilon1] ≈ 0 2 . ÷ 0 5) so that the unknown . O (1) parameters can account for one or two units of charge differences. Here we choose to consider just two representative models that have been presented in Ref. [10] and more carefully analyzed in Ref. [11]. The first one, 'Anarchy', features degenerate charges of left-handed lepton doublets, so that all mixing angles are predicted to be O (1). The second one, 'Hierarchy', has non-degenerate charges in order to account for the relative smallness of θ 13 and ∆ m 2 solar / ∆ m 2 atm . Other models that have been considered in Ref. [10,37] fall in between these two models for what regards their phenomenological consequences in FGM. The charge assignments of the two models are given by
- · Anarchy
$$E _ { i } = ( 3, 2, 0 ) \quad \ \ L _ { i } = ( L _ { 3 }, L _ { 3 }, L _ { 3 } ), \quad \ \ N _ { i } = ( 0, 0, 0 ), \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
- · Hierarchy
$$E _ { i } = ( 5, 3, 0 ) \quad L _ { i } = ( 2 + L _ { 3 }, 1 + L _ { 3 }, L _ { 3 } ), \quad N _ { i } = ( 2, 1, 0 ), \quad \epsilon _ { H } \approx 0. 3 \,. \quad ( 2 4 )$$
For simplicity the expansion parameters are taken here as the central values of the accurate fit in Ref. [11], although there is of course some range due to the unknown order one coefficients. We will use these values in the numerical analysis of Section 4. Note the dependence on an overall charge shift L 3 that essentially corresponds to tan β .
## 3.2 Application to FGM
We now discuss the implementation of the above U (1) models in FGM. For this we only have to specify the charges of the messengers. While in principle they can be arbitrary, we only consider the simplest choice in which they have the same charges as the Higgs fields, i.e. they transform trivially under the flavor symmetry Φ = Φ = 0. This immediately implies that the new couplings of matter fields to the messengers have exactly the same parametric suppression as the corresponding matter-Higgs couplings, but with different O (1) coefficients. In order to see this point more explicitly, we go to the mass basis for the charged leptons by means of the superfield transformations
$$L = \begin{pmatrix} L _ { N } \\ L _ { E } \end{pmatrix} \rightarrow \begin{pmatrix} V _ { N L } L _ { N } \\ V _ { E L } L _ { E } \end{pmatrix}, & E \rightarrow V _ { E R } E,$$
so that
$$y _ { E } \rightarrow V _ { E L } ^ { T } y _ { E } V _ { E R } = y _ { E } ^ { d i a g }.$$
Note that in U (1) models the rotations have the simple parametric structure
$$( V _ { E L } ) _ { i j } & \sim \epsilon ^ { | L _ { i } - L _ { j } | } & ( V _ { E R } ) _ { i j } & \sim \epsilon ^ { | E _ { i } - E _ { j } | }.$$
The spurion λ E transforms accordingly under the above rotations. However, it is straightforward to check that its parametric flavor suppression remains unchanged and only the O (1) coefficients do change so that V T EL λ E V ER ∼ λ E . As a result, since the flavor suppression of λ E is the same of y E , one gets in the mass basis
$$( y _ { E } ) _ { i i } = a _ { i i } \epsilon ^ { L _ { i } + E _ { i } } \,, \quad ( \lambda _ { E } ) _ { i j } = \kappa _ { i j } \epsilon ^ { L _ { i } + E _ { j } } \,,$$
where a ii and κ ij account for unknown, flavor dependent, O (1) coefficients. Assuming hierarchical charges ( E 3 ≤ E k etc.), we finally find the following MIs:
$$( \delta _ { L L } ^ { e } ) _ { i j } \sim \epsilon ^ { L _ { i } + L _ { j } + 2 E _ { 3 } } \sim y _ { \tau } ^ { 2 } \epsilon ^ { L _ { i } + L _ { j } - 2 L _ { 3 } },$$
$$( \delta _ { R R } ^ { e } ) _ { i j } \sim \epsilon ^ { E _ { i } + E _ { j } + 2 L _ { 3 } } \sim y _ { \tau } ^ { 2 } \epsilon ^ { E _ { i } + E _ { j } - 2 E _ { 3 } },$$
and similarly
$$( \delta _ { L R } ^ { e } ) _ { i j } \sim y _ { \tau } ^ { 2 } \ \frac { m _ { j } ^ { e } \ \epsilon ^ { L _ { i } + L _ { j } - 2 L _ { 3 } } + 2 m _ { i } ^ { e } \ \epsilon ^ { E _ { i } + E _ { j } - 2 E _ { 3 } } } { \sqrt { \tilde { m } _ { L } \tilde { m } _ { E } } },$$
where the overall coefficients are given by a calculable part that can be read off from Eqs. (13),(14) and an unknown O (1) coefficient coming from Eq. (28).
As already discussed, since the diagonal A-terms are real, the EDMs can be only generated by means of the combination of MIs ( δ e LL ik ) ( δ e LR kj ) , ( δ e LR ik ) ( δ e RR kj ) , and ( δ e LL ik ) ( δ e LR kk ) ( δ e RR kj ) when ij = 11. In particular, it turns out that the leading effect is captured by
$$( \delta ^ { e } _ { L L } ) _ { i 3 } ( \delta ^ { e } _ { L R } ) _ { 3 3 } ( \delta ^ { e } _ { R R } ) _ { 3 j } \sim \frac { \mu \tan \beta \, m _ { \tau } } { \tilde { m } _ { L } \tilde { m } _ { E } } y _ { \tau } ^ { 3 } \epsilon ^ { L _ { i } + E _ { j } },$$
which involves additional powers of y τ . In principle, the effective MI of Eq. (32) also contributes to µ → eγ when ij = 12 21 however, within our models, single MI contributions , always dominate.
We conclude this section with a discussion of the general structure of the flavor suppression in these terms. First of all, note that LL and RR mass insertions are suppressed by powers of the spurion that are the sum of U(1) charges, in contrast to the leading order terms allowed by the symmetry that have powers given by charge differences. The origin of this suppression is due to the fact that the U (1) controls soft terms only indirectly via the messenger sector, which in turn generates soft terms only at loop level, thus leading to a double suppression by small couplings. As can be seen e.g. in Eq. (30) and the corresponding 2-loop diagram for δ e RR in Fig. 1, this suppression can be split into two parts, one given by the sum of charges of
◦
Figure 1: Example diagram for the 2-loop generation of ( δ e RR ij ) . ˜ L , E i ˜ i denotes the scalar components of the superfields L , E i i and Φ ˜ ± denotes the scalar mass eigenstates of the messengers.
◦
Figure 2: Example diagram for the 1-loop generation of ( δ e LR ij ) in the fermion mass basis.
the external sfermions and the second by (twice) the charge of the field that runs in the loop together with the messenger. As we will discuss later on, the first suppression is exactly the same as in SUSY Partial Compositeness, while the second can lead to a further suppression by powers of y τ .
Turning to LR mass insertions, again the loop origin implies a much stronger suppression than the leading order term glyph[epsilon1] L i + E j respecting the U (1) symmetry, see Fig. 2. This suppression is partially due to the alignment to Yukawas in the pure LR term, which potentially can be avoided in the effective LR terms at the price of an additional suppression by powers of y τ . This is also the only way in which phases can arise in the diagonal elements, since the pure LR term is always the product of a hermitian and a real diagonal matrix.
## 4 Flavor Phenomenology
We are now ready to discuss the lepton flavor phenomenology of the FGM model, which includes LFV processes with an underlying µ → e transition, the electron EDM d e and the anomalous magnetic moment of the muon a µ ≡ ( g -2) / 2. The current experimental bounds and future sensitivities for some of the most relevant LFV channels and for d e are reported in Table 1. On the other hand, a µ currently shows a discrepancy between the SM prediction and the experimental value [48-51]
$$\Delta a _ { \mu } = a _ { \mu } ^ { \text{EXP} } - a _ { \mu } ^ { \text{SM} } = 2. 9 0 ( 9 0 ) \times 1 0 ^ { - 9 } \,.$$
Table 1: Current experimental bounds and future sensitivities for some low-energy LFV observables and the electron EDM.
| LFV Process | Present Bound | Future Sensitivity |
|----------------------------------------------------------------|---------------------------------------------------------------------------------------|----------------------------------------------------------|
| µ → eγ µ → 3 e µ - Au → e - Au µ - Ti → e - Ti µ - Al → e - Al | 5 . 7 × 10 - 13 [39] 1 . 0 × 10 - 12 [41] 7 . 0 × 10 - 13 [43] 4 . 3 × 10 - 12 [44] - | ≈ 6 × 10 - 14 [40] ≈ 10 - 16 [42] ? ? ≈ 10 - 16 [45, 46] |
| Electron EDM | Present Bound | Future Sensitivity |
| d e (e cm) | 8 . 7 × 10 - 29 [47] | ? |
One of the goals of the present paper is to investigate whether it is possible to explain this anomaly in our model while being compatible with LFV and EDM bounds.
Concerning LFV processes, hereafter we focus only on µ → eγ since it represents the best probe of our scenario. The branching ratio of µ → eγ is defined as
$$\text{BR} ( \mu \rightarrow e \gamma ) = \frac { 4 8 \pi ^ { 3 } \alpha } { G _ { F } ^ { 2 } } \left ( \left | A _ { L } ^ { 2 1 } \right | ^ { 2 } + \left | A _ { R } ^ { 2 1 } \right | ^ { 2 } \right ),$$
where the amplitudes A 21 L and A 21 R , in the limit of M 1 = M 2 = µ = ˜ m L = ˜ m R = ˜ and m keeping only tan β enhanced terms, 6 read
$$A _ { L } ^ { 2 1 } \ = \ \frac { 4 \alpha _ { 2 } + 5 \alpha _ { Y } } { 2 4 0 \pi } \ \frac { \tan \beta } { \tilde { m } ^ { 2 } } \ ( \delta _ { L L } ^ { e } ) _ { 2 1 } + \frac { \alpha _ { Y } } { 4 8 \pi } \left ( \frac { \tilde { m } } { m _ { \mu } } \right ) \ \frac { 1 } { \tilde { m } ^ { 2 } } \ ( \delta _ { L R } ^ { e } ) _ { 1 2 } ^ { * },$$
$$A _ { R } ^ { 2 1 } \ = \ - \frac { \alpha _ { Y } } { 2 4 0 \pi } \ \frac { \tan \beta } { \tilde { m } ^ { 2 } } \ ( \delta _ { R R } ^ { e } ) _ { 2 1 } + \frac { \alpha _ { Y } } { 4 8 \pi } \left ( \frac { \tilde { m } } { m _ { \mu } } \right ) ^ { \cdot \cdot \cdot } \frac { 1 } { \tilde { m } ^ { 2 } } \ ( \delta _ { L R } ^ { e } ) _ { 2 1 } \,.$$
Notice that in the above amplitudes we have kept only single MI effects since they are dominant in our scenarios. The expressions for ∆ a µ and d e are well approximated by
$$\Delta a _ { \mu } \ = \ \frac { 5 \alpha _ { 2 } + \alpha _ { Y } } { 4 8 \pi } \frac { m _ { \mu } ^ { 2 } } { \tilde { m } ^ { 2 } } \tan \beta \,,$$
$$\frac { d _ { e } } { e } \ = \ \frac { \alpha _ { Y } } { 1 2 0 \pi } \frac { m _ { \tau } } { \tilde { m } ^ { 2 } } \tan \beta \, \text{Im} [ ( \delta _ { L L } ^ { e } ) _ { 1 3 } ( \delta _ { R R } ^ { e } ) _ { 3 1 } ] \,.$$
In order to highlight the relevant effects, we now provide some numerical estimates for the above observables outlining also their possible correlations. We find that
$$\text{BR} ( \mu \to e \gamma ) \ \approx \ 3 \times 1 0 ^ { - 1 4 } \left ( \frac { 2 0 0 \, \text{GeV} } { \tilde { m } } \right ) ^ { 4 } \tan ^ { 2 } \beta \ \left ( \frac { | ( \delta _ { L L } ^ { e } ) _ { 2 1 } | } { 1 0 ^ { - 4 } } \right ) ^ { 2 } \,,$$
$$\Delta a _ { \mu } \ \approx \ 3 \times 1 0 ^ { - 1 0 } \left ( \frac { 2 0 0 \, \text{GeV} } { \tilde { m } } \right ) ^ { 2 } \tan \beta \,,$$
$$| d _ { e } | \ \approx \ 2 \times 1 0 ^ { - 2 9 } \left ( \frac { 2 0 0 \, \text{GeV} } { \tilde { m } } \right ) ^ { 2 } \tan \beta \ \left | \frac { \text{Im} [ ( \delta _ { L L } ^ { e } ) _ { 1 3 } ( \delta _ { R R } ^ { e } ) _ { 3 1 } ] } { 1 0 ^ { - 6 } } \right | e \, \text{cm} \,. \quad ( 4 1 )$$
6 We will eventually prefer low values for tan β ( glyph[lessorsimilar] 5). For the numerical analysis later on one has therefore to take into account all contributions, which are collected in the Appendix. At this point we are rather interested in keeping the formulae simple and just give order-of-magnitude estimates.
Making the correlations among BR( µ → eγ ), ∆ a µ and d e more explicit, it turns out that
$$\text{BR} ( \mu \to e \gamma ) \ \approx \ 3 \times 1 0 ^ { - 1 3 } \left ( \frac { \Delta a _ { \mu } } { 1 0 ^ { - 9 } } \right ) ^ { 2 } \left ( \frac { | ( \delta _ { L L } ^ { e } ) _ { 2 1 } | } { 1 0 ^ { - 4 } } \right ) ^ { 2 } \,,$$
$$| d _ { e } | \ \approx \ 7 \times 1 0 ^ { - 2 9 } \left ( \frac { \Delta a _ { \mu } } { 1 0 ^ { - 9 } } \right ) \left | \frac { \L ^ { \prime } _ { \text{Im} } [ ( \delta ^ { e } _ { L L } ) _ { 1 3 } ( \delta ^ { e } _ { R R } ) _ { 3 1 } ] } { 1 0 ^ { - 6 } } \right | e \, \text{cm} \,.$$
Eqs. (39-41) deserve few comments:
- · In both flavor models we have considered, the dominant contribution to BR( µ → eγ ) stems from A 21 L , in particular from the tan β -enhanced term proportional to ( δ e LL ) 21 , due to smaller flavor hierarchies in the left-handed lepton sector.
- · The dominant µ → eγ amplitude grows with tan β as A 21 L ∼ ( δ e LL ) 21 tan β ∼ tan 3 β , since ( δ e LL ) 21 ∼ y 2 τ ≈ 10 -4 tan 2 β , which implies that A 21 L is very efficiently suppressed for relatively low tan β . As a result, the very stringent experimental bound on BR( µ → eγ ) might be fulfilled even for a light spectrum ˜ m ∼ 200 GeV provided tan β ∼ 1.
- · The electron EDM can be induced at the leading order only through the effective MI of Eq. (32) and it turns out that d e ∼ tan 5 β . Therefore d e is well under control for low tan β values, analogously to BR( µ → eγ ).
- · The a µ anomaly can be accounted for while satisfying the stringent bounds from BR( µ → eγ ) and d e , only provided that the relevant flavor mixing angles are suppressed at the level of ( δ e LL ) 21 glyph[lessorsimilar] 10 -4 and ( δ e LL ) 13 ( δ e RR ) 31 glyph[lessorsimilar] 10 -6 .
In order to quantify the above considerations, we specialize now to the U (1) flavor models that have been introduced in the previous section: the anarchical and the hierarchical models. The predictions of other scenarios discussed in Refs. [10, 11], fall in between the ones we discuss here. In these two models, the relevant MIs entering the predictions of BR( µ → eγ ) and d e are estimated as:
- · Anarchy
$$( \delta _ { L L } ^ { e } ) _ { 2 1 } & \approx \kappa \frac { 6 } { N } \ y _ { \tau } ^ { 2 } \approx \kappa \frac { 6 \times 1 0 ^ { - 4 } } { N } \tan ^ { 2 } \beta, \\ ( \delta _ { L L } ^ { e } ) _ { 1 3 } ( \delta _ { R R } ^ { e } ) _ { 3 1 } & \approx \kappa ^ { \prime } \frac { 2 0 0 } { N ^ { 2 } } \ y _ { \tau } ^ { 4 } \epsilon _ { A } ^ { 3 } \approx \kappa ^ { \prime } \frac { 2 \times 1 0 ^ { - 8 } } { N ^ { 2 } } \left ( \frac { \epsilon _ { A } } { 0. 2 } \right ) ^ { 3 } \tan ^ { 4 } \beta \,.$$
$$( \delta _ { L L } ^ { e } ) _ { 2 1 } & \approx \kappa \frac { 6 } { N } \ y _ { \tau } ^ { 2 } \varepsilon _ { H } ^ { 3 } \approx \kappa \frac { 2 \times 1 0 ^ { - 5 } } { N } \tan ^ { 2 } \beta \left ( \frac { \epsilon _ { H } } { 0. 3 } \right ) ^ { 3 }, \\ ( \delta _ { L L } ^ { e } ) _ { 1 3 } ( \delta _ { R R } ^ { e } ) _ { 3 1 } & \approx \kappa ^ { \prime } \frac { 2 0 0 } { N ^ { 2 } } \ y _ { \tau } ^ { 4 } \epsilon _ { H } ^ { 7 } \approx \kappa ^ { \prime } \frac { 4 \times 1 0 ^ { - 1 0 } } { N ^ { 2 } } \left ( \frac { \epsilon _ { H } } { 0. 3 } \right ) ^ { 7 } \tan ^ { 4 } \beta \,.$$
- · Hierarchy
where we have used Eqs. (13), (28) assuming an intermediate mediation scale M ∼ 10 10 GeV. Moreover, we have explicitly included the dependence on the unknown O (1) coefficients parameterized through κ and κ ′ that are defined as
$$\kappa \equiv \frac { \kappa _ { 2 3 } \kappa _ { 1 3 } ^ { * } } { a _ { 3 3 } ^ { 2 } } \,, & \kappa ^ { \prime } \equiv \frac { \kappa _ { 1 3 } \kappa _ { 3 1 } \kappa _ { 3 3 } ^ { * 2 } } { a _ { 3 3 } ^ { 4 } }.$$
A prominent feature emerging from Eqs. (44)-(45) is the sensitivity of the MIs to the number of messenger N , since the diagonal sfermion masses are dominated by the MGM contribution proportional to N . We finally get for the µ → eγ branching ratio
$$\text{BR} ( \mu \to e \gamma ) \approx \tan ^ { 6 } \beta \left ( \frac { \kappa } { N } \right ) ^ { 2 } \left ( \frac { 2 0 0 \, \text{GeV} } { \tilde { m } } \right ) ^ { 4 } \times \begin{cases} 1 \times 1 0 ^ { - 1 2 } & \text{Anarchy} \\ 9 \times 1 0 ^ { - 1 6 } & \text{Hierarchy} \end{cases} \quad ( 4 7 )$$
$$\approx \tan ^ { 4 } \beta \left ( \frac { \kappa } { N } \right ) ^ { 2 } \left ( \frac { \Delta a _ { \mu } } { 1 0 ^ { - 9 } } \right ) ^ { 2 } \times \begin{cases} 1 \times 1 0 ^ { - 1 1 } & \text{Anarchy} \\ 9 \times 1 0 ^ { - 1 5 } & \text{Hierarchy} \end{cases}$$
and the eEDM
$$| d _ { e } | \approx \tan ^ { 5 } \beta \left ( \frac { \kappa ^ { \prime } } { N ^ { 2 } } \right ) \left ( \frac { 2 0 0 \, \text{GeV} } { \tilde { m } } \right ) ^ { 2 } e \, \text{cm} \, \times \, \begin{cases} 4 \times 1 0 ^ { - 3 1 } & \text{Anarchy} \\ 1 \times 1 0 ^ { - 3 2 } & \text{Hierarchy} \end{cases}$$
$$\approx \tan ^ { 4 } \beta \left ( \frac { \kappa ^ { \prime } } { N ^ { 2 } } \right ) \left ( \frac { \Delta a _ { \mu } } { 1 0 ^ { - 9 } } \right ) e \, \text{em} \times \begin{cases} 1 \times 1 0 ^ { - 3 0 } & \text{Anarchy} \\ 3 \times 1 0 ^ { - 3 2 } & \text{Hierarchy} \,. \end{cases}$$
Reformulating the constraint from µ → eγ as a bound on the SUSY scale gives approximately
$$\frac { \tilde { m } } { 2 0 0 \, \text{GeV} } \gtrsim \tan ^ { 3 / 2 } \beta \sqrt { \frac { \kappa } { N } } \left ( \frac { \text{BR} ( \mu \to e \gamma ) } { 5. 7 \times 1 0 ^ { - 1 3 } } \right ) ^ { 1 / 4 } \times \begin{cases} 1 & \text{Anarchy} \\ 0. 2 & \text{Hierarchy} \,. \end{cases}$$
Having outlined the expected behaviors and main features of flavor observables within our FGM setup supplemented by U (1) flavor models, we are ready now to perform a complete numerical analysis. In Fig. 3, we show the predictions for BR( µ → eγ ) vs. ∆ a µ for different values of tan β : purple, blue and orange dots correspond to tan β = 1 5 3 5, respectively. . , , The plots on the left (right) refer to the anarchical (hierarchical) case. For the upper (lower) plots the number of messenger is set to N = 1 (5). In Fig. 4, we show the analogous plots for d e vs. ∆ a µ . In the scan we have varied the unknown O (1) coefficients κ, κ ′ for the MIs in the range (0 3 . , 1 5). . The other parameters were varied in the following ranges:
$$1 0 ^ { 6 } \, \text{GeV} \leq M \leq 1 0 ^ { 1 5 } \, \text{GeV}, \, \ 1 0 0 \, \text{GeV} \leq \tilde { m } _ { E } ( M ) \leq 1 \, \text{TeV}, \, \ 1 0 0 \, \text{GeV} \leq \mu \leq \mu _ { \max }, \quad ( 5 2 )$$
where µ max ≡ ˜ m τ L ˜ m τ R / m ( τ tan β ) is the maximal value giving a non-tachyonic stau. In the plots, the gray shaded regions are excluded by the current bounds from µ → eγ or d e reported in Table 1, while the green (dark green) area approximately corresponds to values of ∆ a µ lowering the discrepancy below the 2 σ (1 σ ) level.
Figure 3: Predictions for BR( µ → eγ ) vs. ∆ a µ for different values of tan β : purple, blue and orange dots correspond to tan β = 1 5 3 5, respectively. . , , The plots on the left (right) refer to the anarchical (hierarchical) case. For the upper (lower) plots the number of messenger is set to N = 1 (5).
A direct comparison of the bounds and the discovery potential of µ → eγ and d e is shown in Fig. 5, where we plot the result of a random variation of the full set of parameters for the anarchical (left) and hierarchical (right) cases:
$$1 0 ^ { 6 } \, \text{GeV} \leq M \leq 1 0 ^ { 1 5 } \, \text{GeV}, \, \ 1 0 0 \, \text{GeV} \leq \tilde { m } _ { E } ( M ) \leq 2 \, \text{TeV}, \, \ 1 0 0 \, \text{GeV} \leq \mu \leq \mu _ { \max }, \\ 1 \leq N \leq 5, \quad 1. 5 \leq \tan \beta \leq 5, \quad 0. 3 \leq \kappa, \kappa ^ { \prime } \leq 1. 5.$$
In addition, the yellow (green) points correspond to ∆ a µ ≥ 10 -9 (2 × 10 -9 ).
The main results emerging from our numerical analysis can be summarizes as follows:
- · In the anarchical scenario, it is very difficult if not impossible to explain the ∆ a µ
Figure 4: The same scenarios as in Fig. 3 for d e vs. ∆ a µ .
anomaly because of the strong bounds from both BR( µ → eγ ) and d e (see left panels of Figs. 3-5). The latter observables have a comparable sensitivity to the scenario in question and might reach experimentally visible values even for SUSY masses far beyond the LHC reach in the multi-TeV regime. As already discussed, BR( µ → eγ ) and d e grow fast with tan β (like tan 6 β and tan 5 β , respectively) and are both suppressed by increasing N . As a result, the scenario with low tan β and N = 5 (we remind that for low mediation scales perturbativity requires N glyph[lessorsimilar] 5) is the most favorable scenario, as clearly shown in Figs. 3 and 4.
- · The hierarchical scenario easily offers the possibility to explain the ∆ a µ anomaly while satisfying the limits on BR( µ → eγ ) and d e (right panels of Figs. 3-5). This happens thanks to the stronger suppression of the flavor mixing angles compared to the anarchical case. On the other hand, all the others considerations made above for the anarchical
Figure 5: Predictions for BR( µ → eγ ) vs. d e for the anarchical (left) and hierarchical (right) cases. Yellow (green) points correspond to ∆ a µ ≥ 10 -9 (2 × 10 -9 ).
case apply here as well.
- · µ → eγ and d e have comparable sensitivities, but µ → eγ is currently more constraining, as we can see from Fig. 5. Interestingly, an improvement of the sensitivity by one or two orders of magnitude would make the electron EDM the most powerful probe of FGM scenarios especially in case of heavy superpartners, corresponding to the red points in Fig. 5. This is a consequence of the slower decoupling of d e with respect to the NP scale: d e ∼ ˜ m -2 , while BR( µ → eγ ) ∼ ˜ m -4 .
- · Given the expected future sensitivities to the µ → e transitions reported in Table 1 and the following approximate relations among different decay modes:
$$\text{BR} ( \mu \to e e e ) \ \ & \simeq \ \ \frac { \alpha } { 3 \pi } \left ( \log \frac { m _ { \mu } ^ { 2 } } { m _ { 2 } ^ { 2 } } - 3 \right ) \text{BR} ( \mu \to e \gamma ) \, \\ \text{CR} ( \mu \to e \text{ in } N ) \ \ & \simeq \ \ \alpha \times \text{BR} ( \mu \to e \gamma )$$
we see that there are good prospects for a full test of the parameter space favored by ∆ a µ at future experiments.
Let us now also show how the µ → eγ and d e constraints appear in terms of the gaugino and slepton masses. For illustration purposes, we adopt a more general low-energy spectrum than the one predicted by MGM, which allows us to parameterize in a model-independent way possible distortions of the spectrum due to the full set of matter-messenger couplings studied in [25], including the other couplings in Eq. (4), as well as more generic SUSY breaking sectors, in the spirit of General Gauge Mediation [52]. In practice, we still use Eqs. (9-11) to set the off-diagonal entries but we treat slepton and gaugino masses as free parameters at low energy.
Figure 6: Bounds on the O (1) coefficient κ , see Eq. (46), from µ → eγ in the hierarchical scenario. For definiteness, we have assumed M 2 = 2 M 1 , ˜ m L = 2 ˜ m E and different choices of µ and tan β . The yellow (green) areas give ∆ a µ ≥ 10 -9 (2 × 10 -9 ).
In Fig. 6, we show the current bounds on the O (1) coefficient κ , as defined in Eq. (46), from µ → eγ for different choices of µ and tan β in the hierarchical case. For definiteness, we fixed the relation among gauginos and slepton masses as follows: M 2 = 2 × M 1 , ˜ m L = 2 × ˜ m E . The yellow (green) areas give ∆ a µ ≥ 10 -9 (2 × 10 -9 ). As we can see, it is not necessary that the unknown coefficients conspire to provide an unnaturally small suppression, in order to take µ → eγ under control in the region favored by ( g -2) µ .
In Fig. 7, we show contours of BR( µ → eγ ) and d e for the same choice of the parameters as above and the specific values κ = 0 5, . κ ′ = 1, in order to illustrate the present bounds and the possible impact of the future experiments in terms of the masses of the SUSY particles in the game. In particular, the gray shaded regions are presently excluded by µ → eγ or d e . Again, we see that a large contribution to ( g -2) µ is perfectly compatible with the present bounds, but there are good prospects for a full test of the relevant parameter space in the future.
Besides the LEP constraints (corresponding to ˜ m , m ,M E ˜ L 2 glyph[greaterorsimilar] 100 GeV), the mass plane shown in the above plots is now challenged by searches for the electroweak production of SUSY particles performed by the LHC experiments, based on events with two or more leptons plus missing transverse momentum. The exact bounds are model dependent and their precise derivation is beyond the scope of the present study. Nevertheless, we briefly summarize here their possible impact.
Since we are considering scenarios with gauge-mediated SUSY breaking, the LSP is always a practically massless gravitino. The limits set by LHC searches then strongly depend on the nature and the life-time of the next to LSP (NLSP). In case the sleptons are lighter than the Bino, as it can occur even in MGM for large values of N , each decay chain would end with the degenerate sleptons NLSP decaying into leptons and gravitino. If this decay occurs
Figure 7: Contours of BR( µ → eγ ) and d e for the same choice of the parameters as in Fig. 6 and κ = 0 5, . κ ′ = 1. The gray shaded regions are presently excluded by µ → eγ or d e .
promptly (which requires low mediation scales, M glyph[lessorsimilar] 10 6 GeV), then the bounds from direct (Drell-Yan) slepton production translate to a limit on the mass of the right-handed (lefthanded) sleptons at about 250 (300) GeV [53]. In case sleptons are long-lived compared to the detector scale, searches for charged tracks set a bound on degenerate slepton NLSP mass at about 400 GeV [54]. Interestingly, no searches performed so far constrain the intermediate case, featuring a disappearing track with a displaced vertex inside the detector, occurring for a wide range of the messenger scale, 10 6 GeV glyph[lessorsimilar] M glyph[lessorsimilar] 10 9 GeV [55]. The above limits can be substantially relaxed if the Bino is lighter than the sleptons and escapes the detector, thus resembling searches within gravity mediation. In particular for a neutralino NLSP heavier than about 150 GeV, there is no constraint from direct slepton production [53].
The most stringent constraint would occur in the case of the hierarchy M < m 1 ˜ L <
M 2 from Wino-like chargino/neutralino production followed by decays into on-shell sleptons/sneutrinos, with bounds up to 700 GeV on the Wino mass from multi-lepton plus missing energy searches [56]. However, such searches loose sensitivity if the mass splitting of the sleptons with either the Bino or the Wino gets small.
Comparing the limits reported above, with our plots in Fig. 6 and 7, we see that there is still room for a large SUSY contribution to a µ , at least at the 10 -9 level (yellow regions), especially if µ approaches the maximal value µ max ≡ ˜ m τ L ˜ m τ R / m ( τ tan β ). This conclusion is supported by the results of Refs. [57,58] where a systematic study of the LHC constraints on the parameter space favored by ( g -2) µ has been presented.
## 5 Comparison with other models
In this section, we compare the peculiar flavor structure of FGM to other models that predict the parametric flavor suppression of soft terms. In particular, we consider U (1) flavor models within SUGRA scenarios and models with SUSY Partial Compositeness (PC).
In those models the SUSY mediation scale Λ S is assumed to be above the scale of flavor messengers Λ F , so that the flavor structure of soft terms at the scale Λ F is controlled entirely by the flavor dynamics at this scale, irrespectively of their structure at the scale Λ S . In FGM the situation is reversed as the SUSY messenger scale Λ S = M is below Λ F . We stress that this setup is therefore complementary to the other scenarios, allowing also for very low SUSY mediation scales. All the unspecified dynamics of the flavor sector is imprinted in the mattermessenger couplings, just like Yukawas, and the full SUSY spectrum is totally calculable in terms of these couplings.
## 5.1 U (1) and Gravity Mediation
In Gravity Mediation the natural expectation for soft terms at the flavor scale is given by the most general terms invariant under the symmetry using the flavon as a spurion. This gives for slepton mass insertions
$$( \delta _ { L R } ^ { e } ) _ { i j } \sim \frac { A v _ { d } } { \tilde { m } _ { L } \tilde { m } _ { E } } \epsilon ^ { L _ { i } + E _ { j } } \,,$$
$$( \delta _ { L L } ^ { e } ) _ { i j } & \sim \epsilon ^ { | L _ { i } - L _ { j } | } \,, & ( \delta _ { R R } ^ { e } ) _ { i j } & \sim \epsilon ^ { | E _ { i } - E _ { j } | } \,.$$
Focusing on the anarchical and hierarchical models of Section 3.1, the relevant MIs for BR( µ → eγ ) and d e are again ( δ e LL ) 21 and ( δ e LL ) 13 ( δ e RR ) 31 , respectively. In the anarchic case we find:
- · Anarchy
$$( \delta _ { L L } ^ { e } ) _ { 2 1 } \sim 1 \,, \quad ( \delta _ { L L } ^ { e } ) _ { 1 3 } ( \delta _ { R R } ^ { e } ) _ { 3 1 } \sim \epsilon _ { A } ^ { 3 } \,,$$
Table 2: Predictions for the mass insertions in various SUSY models with an underlying U (1) flavor model where L i ( E i ) stands for the charges of SU (2) doublets (singlets). Note that for the sake of simplicity we compare only single mass insertions, for large tan β triple mass insertions can possibly give the dominant contributions to LR transitions.
| | SUGRA U (1) | PC | FGM U (1) |
|-----------------------|---------------------------|---------------------------|------------------------------------------------|
| ˜ m m e Im( δ LR ) 11 | 1 | 1 | y 4 τ t β |
| ˜ m m µ ( δ LR ) 12 | glyph[epsilon1] L 1 - L 2 | glyph[epsilon1] L 1 - L 2 | y 2 τ glyph[epsilon1] L 1 + L 2 - 2 L 3 |
| ˜ m m µ ( δ LR ) 21 | glyph[epsilon1] E 1 - E 2 | glyph[epsilon1] E 1 - E 2 | y 2 τ glyph[epsilon1] E 1 - E 2 +2 L 1 - 2 L 3 |
| ( δ LL ) 12 | glyph[epsilon1] L 1 - L 2 | glyph[epsilon1] L 1 + L 2 | y 2 τ glyph[epsilon1] L 1 + L 2 - 2 L 3 |
| ( δ RR ) 12 | glyph[epsilon1] E 1 - E 2 | glyph[epsilon1] E 1 + E 2 | y 2 τ glyph[epsilon1] E 1 + E 2 - 2 E 3 |
leading to the following predictions:
$$\text{BR} ( \mu \to e \gamma ) \ \sim \ 5 \times 1 0 ^ { - 1 3 } \left ( \frac { 1 0 \, \text{TeV} } { \tilde { m } } \right ) ^ { 4 } \tan ^ { 2 } \beta \,,$$
$$| d _ { e } | \ \sim \ \ 7 \times 1 0 ^ { - 2 9 } \left ( \frac { 1 0 \, \text{TeV} } { \tilde { m } } \right ) ^ { 2 } \tan \beta \, e \, \text{cm} \,.$$
## · Hierarchy
$$( \delta ^ { e } _ { L L } ) _ { 2 1 } \sim \epsilon _ { H } \sim 0. 3 \,, \quad ( \delta ^ { e } _ { L L } ) _ { 1 2 } ( \delta ^ { e } _ { R R } ) _ { 2 1 } \sim \epsilon ^ { 3 } _ { H } \,,$$
where we took ( δ e LL ) 12 ( δ e RR ) 21 instead of ( δ e LL ) 13 ( δ e RR ) 31 since the contribution of the latter to the eEDM is smaller by a factor of ( y /y τ µ ) × glyph[epsilon1] 4 H ∼ 0 1 compared to that induced . by ( δ e LL ) 12 ( δ e RR ) 21 . We therefore have the following predictions:
$$\text{BR} ( \mu \to e \gamma ) \ \sim \ 7 \times 1 0 ^ { - 1 3 } \left ( \frac { 5 \text{ TeV} } { \tilde { m } } \right ) ^ { 4 } \tan ^ { 2 } \beta \ \left ( \frac { \epsilon _ { H } } { 0. 3 } \right ) ^ { 2 },$$
$$| d _ { e } | \ \sim \ 6 \times 1 0 ^ { - 2 9 } \left ( \frac { 5 \ \text{TeV} } { \tilde { m } } \right ) ^ { 2 } \tan \beta \, e \, \text{cm} \,.$$
As a result, single U (1) flavor models with Gravity Mediation need sleptons well above the TeV scale, ˜ m glyph[greaterorsimilar] 10 TeV × √ tan β (5 TeV × √ tan β ) in the anarchical (hierarchical) scenario. Note that the bounds on the SUSY spectrum in the quark sector, in particular from glyph[epsilon1] K , are much stronger [28].
## 5.2 SUSY Partial Compositeness
According to the paradigm of Partial Compositeness, the lepton Yukawa matrices have the form
$$( y _ { E } ) _ { i j } & \sim g _ { \rho } \epsilon _ { i } ^ { \ell } \epsilon _ { j } ^ { e },$$
where g ρ is a strong coupling and glyph[epsilon1] glyph[lscript],e i glyph[lessorsimilar] 1 measures the amount of compositeness for the leptons. Such a scheme closely resembles the case of a single U (1) flavor model, with the correspondence (in the limit of g ρ = 1)
$$\epsilon _ { i } ^ { \ell, e } \longleftrightarrow \epsilon ^ { L _ { i }, E _ { i } } \,.$$
As a result, the MIs are expected to take the following form [32]
$$( \delta ^ { e } _ { L L } ) _ { i j } \sim \epsilon ^ { \ell } _ { i } \epsilon ^ { \ell } _ { j } \sim \epsilon ^ { L _ { i } + L _ { j } }, \quad ( \delta ^ { e } _ { R R } ) _ { i j } \sim \epsilon ^ { e } _ { i } \epsilon ^ { e } _ { j } \sim \epsilon ^ { E _ { i } + E _ { j } }.$$
$$( \delta _ { L R } ^ { e } ) _ { i j } \sim \frac { v _ { d } A g _ { \rho } } { \tilde { m } _ { L } \tilde { m } _ { E } } \epsilon _ { i } ^ { \ell } \epsilon _ { j } ^ { e } \sim \frac { m _ { i } ^ { e } A } { \tilde { m } _ { L } \tilde { m } _ { E } } \epsilon ^ { E _ { j } - E _ { i } } \sim \frac { m _ { j } ^ { e } A } { \tilde { m } _ { L } \tilde { m } _ { E } } \epsilon ^ { L _ { i } - L _ { j } } \,.$$
In PC, the leading contributions to BR( µ → eγ ) typically arise from ( δ e LR ) 12 . In particular, in the anarchical scenario we find
$$\text{BR} ( \mu \to e \gamma ) \ \sim \ 6 \times 1 0 ^ { - 1 3 } \left ( \frac { 5 \text{ TeV} } { \tilde { m } } \right ) ^ { 4 } \,,$$
while in the hierarchical case we have a mild additional suppression by a factor of glyph[epsilon1] 2 H ≈ 0 1. . Note however that in PC the left-handed 'charges' L i are determined from the PMNS matrix analogously to U (1) models only in the case of light Dirac Neutrinos. If instead light neutrinos are Majorana, then the Weinberg operator can arise from a bilinear coupling to the composite sector (instead of linear couplings that resemble the U (1) structure). In this case only the combination L i + E j is determined by charged lepton Yukawa couplings, and the constraints from LFV can be significantly relaxed by choosing symmetric charges [32]
$$\epsilon ^ { L _ { i } } \sim \epsilon ^ { E _ { i } } \sim \sqrt { \frac { y _ { i } ^ { e } } { g _ { \rho } } }.$$
This implies and thus
$$\text{BR} ( \mu \to e \gamma ) \sim 7 \times 1 0 ^ { - 1 3 } \left ( \frac { 1. 5 \, \text{TeV} } { \tilde { m } } \right ) ^ { 4 } \,.$$
On the other hand, the predictions for the electron EDM are completely independent of any charge assignments since in PC the diagonal elements of the A-terms are generally complex. As a result, we find
$$| d _ { e } | \ \sim \ 7 \times 1 0 ^ { - 2 9 } \left ( \frac { 3 \ T e V } { \tilde { m } } \right ) ^ { 2 } \text{Im} \left ( \frac { M _ { 1 } A } { \tilde { m } ^ { 2 } } \right ) \text{ e } \text{cm} \,,$$
$$\frac { \tilde { m } } { m _ { \mu } } ( \delta _ { L R } ^ { e } ) _ { 1 2 } \sim \epsilon ^ { L _ { 1 } - L _ { 2 } } \sim \sqrt { \frac { m _ { e } } { m _ { \mu } } } \,,$$
and therefore the eEDM now provides the strongest constraint on the PC scenario. Notice that the electron EDM has a similar sensitivity to NP effects in PC scenarios and U (1) flavor models, independently of the particular charge assignments, pushing the SUSY scale to ˜ m glyph[greaterorsimilar] 3 ÷ 5 TeV. Needless to say, neither PC scenarios nor SUGRA with an underlying U (1) flavor model can explain the muon g -2 anomaly.
In Table 2, we summarize the predictions for the MIs most relevant for phenomenology in various models: SUGRA (first column), PC (second column) and FGM (last column). On general ground, comparing the flavor structure of the soft sector of SUGRA and PC/FGM scenarios, the most prominent feature is the higher suppression for off-diagonal sfermion masses in the LL and RR sectors in the PC/FGM case. The LR sector has the same parametric structure in PC and SUGRA, since in both scenarios the A-terms are proportional to the SM Yukawas, while in FGM we have a much stronger suppression arising from a partial alignment among SM Yukawas and A-terms. Finally, PC and SUGRA share also the same SUSY CP problem as they allow complex diagonal elements for the A-terms. In contrast, within FGM, the leading CPV phases arise only at higher order in MI expansions and therefore are very suppressed.
## 6 Conclusions
Now that the Higgs boson has been discovered, naturalness becomes a pressing question waiting for the final answer of LHC14. If new dynamics is present around the TeV scale, as needed to explain the smallness of the electro-weak scale, one would expect too large contributions to flavor transitions mediated by the new physics states unless some protection mechanism is at work. Therefore, the possibility of finding new physics at the LHC is closely related to the existence of a suppression of flavor violating processes.
In this respect, Minimal Gauge Mediation (MGM) provides an ideal framework to accomplish this job. Indeed, if the flavor scale is much higher than the SUSY messenger scale then the flavor structure of the soft terms is entirely determined by the SM Yukawas, thus realizing the paradigm of Minimal Flavor Violation (MFV) [15]. The drawback of this scenario is that any imprint of the flavor sector in low-energy physics and thus the possibility to test the flavor dynamics is completely lost.
On the other hand, minimal realizations of GMSB are now seriously challenged by the Higgs boson discovery at the LHC, since they can account for m h ≈ 126 GeV only at the price of a SUSY spectrum that is beyond the reach of the LHC. This has motivated extensions of minimal GMSB models by introducing new direct couplings between the messengers and the MSSM matter fields in order to obtain a large Higgs mass for light stops by generating non-vanishing A-terms at the messenger scale [17-27].
Among these scenarios, 'Flavored Gauge Mediation' (FGM) [16] assumes that these new couplings have a flavor structure which is controlled by the same underlying flavor symmetry that explains the smallness of Yukawa couplings. As a result, FGM allows to generate soft masses which still carry information about the high scale flavor sector. Interestingly, due to
the loop origin of soft terms, sfermion masses exhibit a flavor pattern that is much stronger suppressed than in Gravity Mediation. This strong suppression, arising even in the context of single U (1) flavor models, is reminiscent of what happens in the case of wave function renormalization [29,30] or Partial Compositeness [31,32]. In addition there is a strong suppression of LR flavor transitions that is particularly effective for accompanying flavor-blind phases, thus rendering the strong bounds from EDMs under control. Therefore FGM does not only modify the SUSY spectrum of MGM in a way interesting for collider phenomenology, but it also allows to obtain a rich flavor phenomenology beyond MFV. In particular it offers a viable SUSY implementation of simple U(1) flavor symmetry models that in the context of gravity mediation have huge difficulties in passing the bounds from precision observables, and in the context of MGM are not testable at all.
While in Ref. [28] we concentrated on the quark sector, in this work we have focused on the lepton sector analyzing the implications of FGM with underlying U (1) flavor models. In particular, we have studied the predictions of two models (the anarchical and hierarchical scenarios of Ref. [10]) that are representative for a whole class of U (1) models that accommodate lepton masses and mixing angles. We have analyzed µ → eγ (which turned out to be the most constraining LFV channel), the electron EDM d e and the muon anomalous magnetic moment ∆ a µ . Since the experimental bounds on both µ → eγ and d e underwent recently an impressive improvement, an important question of this work was to establish whether and to which extent single U (1) flavor models were viable in the context of FGM models. A related relevant question was to establish whether the current muon g -2 anomaly could be resolved while being compatible with the LFV and EDM bounds.
In the following, we summarize our main findings:
- · The non-holomorphic soft masses (LL and RR mass insertions) are suppressed by powers of the spurion that are the sum of U(1) charges, in contrast to the corresponding gravitymediated case where charge differences enter. The origin of this suppression is due to the fact that the U (1) controls soft terms only indirectly via the messenger sector, which in turn generates soft terms only at loop level, thus leading to a double suppression by small couplings. This strong suppression is similar to the cases of wave function renormalization or SUSY Partial Compositeness.
- · The A-terms are much more suppressed than in PC and gravity-mediated scenarios where they are proportional to the leading order term glyph[epsilon1] L i + E j allowed by the U (1) symmetry. Moreover, the A-terms are partially aligned to the Yukawa couplings and their diagonal components are real therefore not inducing contributions for the EDM. The first non-vanishing CP violating phase in the diagonal A-terms can arise only through higher order expansions in the MIs, which leads to am additional suppression by powers of y τ that becomes particularly effective for low tan β .
- · LFV processes and the electron EDM can be kept under control even for a light spectrum well below the TeV scale, provided that tan β is small (the smaller the better). This is true both in the anarchical and especially in the hierarchical scenarios. In contrast,
PC and gravity-mediated scenarios require a very heavy spectrum above the TeV scale in order to fulfill the experimental bounds on BR( µ → eγ ) and d e . Very low values for tan β also perfectly fit a complete realization of this setup within the NMSSM, which is the natural choice to generate the µ -term in FGM scenarios, solving also the µ -B µ problem of MGM [21,36].
- · In spite of the tremendous experimental bounds on the electron EDM and LFV processes, we have found that is still possible to account for the muon g -2 anomaly within FGM models in the hierarchical but not in the anarchical scenarios. The same conclusion is not true in PC and gravity-mediated models where both µ → eγ and d e prevent any sizable effect in ∆ a µ .
- · Although BR( µ → eγ ) and d e have comparable sensitivities to FGM scenarios, µ → eγ is currently more constraining. However, considering the slower decoupling of NP effects in d e ∼ ˜ m -2 with respect to BR( µ → eγ ) ∼ ˜ m -4 , the electron EDM might become the most powerful probe of the scenarios in question with improved experimental data especially in case of heavy superpartners.
In conclusion, we have analyzed in detail the anatomy and phenomenology in the lepton sector of FGM models with underlying U (1) models. Remarkably, it turned out that these models can pass the impressive bounds on LFV processes and leptonic EDMs even for light superpartners, potentially observable at LHC14, leaving open the possibility of accommodating the longstanding muon ( g -2) anomaly and testing U (1) flavor models in upcoming experiments.
## Acknowledgments
The work of P.P. was supported by the ERC Advanced Grant No.267985 Electroweak Symmetry Breaking, Flavour and Dark Matter: One Solution for Three Mysteries (DaMeSyFla) . This work made in the ILP LABEX (under reference ANR-10-LABX-63) was partially supported by French state funds managed by the ANR within the Investissements d'Avenir programme under reference ANR-11-IDEX-0004-02.
## A Formulae for the leptonic dipoles
In this appendix we collect the formulae for LFV branching ratios BR( l i → l j γ ), lepton anomalous magnetic moments ∆ a i , and lepton electric dipole moments d i . The results have been obtained from the exact results of Refs. [59, 60] using a generalized mass insertion approximation (see e.g. Ref. [61]) without assuming large tan β .
## A.1 Lepton Flavor Violation: l i → l j γ
The branching ratio BR( l i → l j γ ) is given by
$$\frac { \text{BR} ( l _ { i } \rightarrow l _ { j } \gamma ) } { \text{BR} ( l _ { i } \rightarrow l _ { j } \nu _ { i } \bar { \nu } _ { j } ) } = \frac { 4 8 \pi ^ { 3 } \alpha } { G _ { F } ^ { 2 } } \left ( | A _ { L } ^ { i j } | ^ { 2 } + | A _ { R } ^ { i j } | ^ { 2 } \right ).$$
The amplitude A ij L receives a neutralino and chargino contribution A ij L = A ij ( n ) L + A ij ( c ) L , whereas A ij R receives only a neutralino contribution A ij R = A ij ( n ) R . These contributions are given by:
$$\text{where} \quad A _ { R } ^ { i j } \text{ receives only a neutralino contribution } A _ { R } ^ { i j } = A _ { R } ^ { i j ( n ) } \text{. These contributions are} \\ \text{given by:} \\ A _ { i j } ( n ) & = \frac { \alpha _ { 2 } } { 4 } \frac { ( m _ { 2 } ^ { 2 } L ) i j } { m _ { 4 } ^ { 4 } } \left [ f _ { 1 n } ( x _ { 2 L } ) + \left ( | M _ { 2 } | ^ { 2 } + \mu M _ { 2 } t _ { \beta } \right ) } { | M _ { 2 } | ^ { 2 } - | \mu | ^ { 2 } } \frac { ( | \mu | ^ { 2 } + \mu M _ { 2 } t _ { \beta } ) } { | M _ { 2 } | ^ { 2 } - | \mu | ^ { 2 } } \frac { f _ { 2 n } ( x _ { \mu L } ) } { | M _ { 2 } | ^ { 2 } - | \mu | ^ { 2 } } \frac { f _ { 2 n } ( x _ { \mu L } ) } { \right ] } \\ & + \frac { \alpha _ { Y } } { 4 \pi } \frac { ( m _ { 2 } ^ { 2 } L ) i j } { m _ { 4 } ^ { 4 } } \left [ f _ { 1 n } ( x _ { 1 L } ) - \left ( | M _ { 1 } | ^ { 2 } + \mu M _ { 1 } t _ { \beta } \right ) } { | M _ { 1 } | ^ { 2 } - | \mu | ^ { 2 } } \frac { f _ { 2 n } ( x _ { 1 L } ) + \left ( | \mu | ^ { 2 } + \mu M _ { 1 } t _ { \beta } \right ) } { | M _ { 1 } | ^ { 2 } - | \mu | ^ { 2 } } \frac { f _ { 2 n } ( x _ { \mu L } ) } { \right ] } \\ & - \frac { \alpha _ { Y } } { 4 \pi } \frac { M _ { 1 } } { m _ { \mu } ^ { 2 } L _ { - } m _ { R } ^ { 2 } } \left [ \frac { 1 } { m _ { 2 } ^ { 2 } L _ { R } ^ { 2 } } { f _ { 3 n } ( x _ { 1 L } ) - \frac { 1 } { m _ { R } ^ { 2 } } } { f _ { 3 n } ( x _ { 1 R } ) } \right ] \\ & - \frac { \alpha _ { Y } } { 4 \pi } \frac { M _ { 1 } } { m _ { \mu } ^ { 2 } L _ { - } m _ { R } ^ { 2 } } { ( m _ { 2 } ^ { 2 } L _ { - } m _ { R } ^ { 2 } ) m _ { 2 } ^ { 2 } } \left [ \frac { 1 } { m _ { 2 } ^ { 2 } L _ { - } m _ { R } ^ { 2 } } { f _ { 3 n } ( x _ { 1 L } ) - \frac { 1 } { m _ { R } ^ { 2 } } } { f _ { 3 n } ( x _ { 1 R } ) } \right ) + \frac { 2 } { m _ { 2 } ^ { 2 } } { f _ { 2 n } ( x _ { 1 R } ) } \right ] \\ & + \frac { \alpha _ { Y } } { 4 \pi } \frac { M _ { 1 } } { m _ { \mu } ^ { 2 } L _ { - } m _ { R } ^ { 2 } } { ( m _ { 2 } ^ { 2 } L _ { - } m _ { R } ^ { 2 } ) m _ { 2 } ^ { 2 } } \left [ \frac { 1 } { m _ { 2 } ^ { 2 } L _ { - } m _ { R } ^ { 2 } } { f _ { 3 n } ( x _ { 1 L } ) - \frac { 1 } { m _ { R } ^ { 2 } } } { f _ { 3 n } ( x _ { 1 R } ) } \right ) + \frac { 2 } { m _ { 2 } ^ { 2 } } { f _ { 2 n } ( x _ {1 L } ) } \right ] \\ & + \frac { \alpha _ { Y } } { 2 \pi } \frac { M _ { 1 } } { m _ { \mu } ^ { 2 } L _ { - } m _ { R } ^ { 2 } } { ( m _ { 2 } ^ { 2 } L _ { - } m _ { R } ^ { 2 } ) ^ { 2 } } { m _ { 2 } ^ { 2 } L _ { - } m _ { R } ^ { 2 } } \times \\ & \left [ \frac { m _ { 2 } ^ { 2 } m _ { 2 } ^ { 2 } R _ { 2 } } { \left ( \frac { 1 } { m _ { 2 } ^ { 2 } } { f _ { 3 n } ( x _ { 1 L } ) - \frac { 1 } { m _ { R } ^ { 2 } } } { f _ { 3 n } ( x _ { 1 R } ) } \right ) + \frac { m _ { 2 } ^ { 2 } } { m _ { 2 } ^ { 2 } } { f _ { 2 n } ( x _ { 1 L } ) + \frac { m _ { 2 } ^ { 2 } } { m _ { 2 } ^ { 2 } } { f _ { 2 n } ( x _ { 1 R } ) } \right ], \quad \text{(73)} \\ A _ { L } ^ { i j ( c ) } & = \frac { \alpha _ { 2 } } { 2 } \frac { ( m _ { 2 } ^ { 2 } L ) i j } { \left [ f _ { 1 c } ( x _ { 2 L } ) + \left ( | M _ { 2 } | ^ { 2 } + \mu M _ { 2 } t _ { \beta } \right ) } { f _ { 2 c } ( x _ { 2 L } ) - \left ( | \mu | ^ { 2 } + \mu M _ { 2 } t _ { \beta } \right ) } { f _ { 2 c } ( x _ { 2 L } ) } \right ],$$
$$A _ { L } ^ { i j ( c ) } = \frac { \alpha _ { 2 } } { 4 \pi } \frac { ( m _ { L L } ^ { 2 } ) _ { i j } } { m _ { L } ^ { 4 } } \left [ f _ { 1 c } ( x _ { 2 L } ) + \frac { ( | M _ { 2 } | ^ { 2 } + \mu M _ { 2 } t _ { \beta } ) } { | M _ { 2 } | ^ { 2 } - | \mu | ^ { 2 } } f _ { 2 c } ( x _ { 2 L } ) - \frac { ( | \mu | ^ { 2 } + \mu M _ { 2 } \tan \beta ) } { | M _ { 2 } | ^ { 2 } - | \mu | ^ { 2 } } f _ { 2 c } ( x _ { \mu L } ) \right ] \,,$$
$$A _ { R } ^ { i j ( n ) } & = \alpha _ { \overline { \frac { m } { 2 } } } \left ( m _ { R } ^ { 2 } \overline { \frac { R } { 2 } } \overline { \frac { m } { 2 } } _ { \mu } m _ { R } ^ { 4 } \left [ 2 f _ { 1 n } ( x _ { 1 R } ) + \left ( | M _ { 1 } | ^ { 2 } + \mu M _ { 1 } t _ { \beta } \right ) f _ { 2 n } ( x _ { 1 R } ) - \left ( | \mu | ^ { 2 } + \mu M _ { 1 } t _ { \beta } \right ) f _ { 2 n } ( x _ { \mu R } ) \right ] \\ & - \frac { \alpha _ { \overline { \frac { M } { 2 } } } } \overline { \frac { M _ { 1 } } { m _ { \mu } } } \overline { \frac { m _ { 2 } } { m _ { L } } - m _ { R } ^ { 2 } \right ) m _ { R } ^ { 2 } \left [ \frac { 1 } { m _ { L } } } f _ { 3 n } ( x _ { 1 L } ) - \frac { 1 } { m _ { R } ^ { 2 } } f _ { 3 n } ( x _ { 1 R } ) \right ] \\ & - \frac { \alpha _ { \overline { \frac { M } { 2 } } } } \overline { \frac { M _ { 1 } } { m _ { \mu } } } \overline { \frac { m _ { 2 } } { m _ { L } } - m _ { R } ^ { 2 } \right ) m _ { R } ^ { 2 } \left [ \frac { 1 } { m _ { L } } } f _ { 3 n } ( x _ { 1 L } ) - \frac { 1 } { m _ { R } ^ { 2 } } f _ { 3 n } ( x _ { 1 R } ) \right ) + \frac { 2 } { m _ { R } ^ { 2 } } f _ { 2 n } ( x _ { 1 R } ) \right ] \\ & + \frac { \alpha _ { \overline { \frac { M } { 2 } } } } \overline { \frac { M _ { 1 } } { m _ { \mu } } } \overline { \frac { m _ { 2 } } { m _ { L } } - m _ { R } ^ { 2 } \right ) m _ { L } ^ { 2 } \left [ \frac { m _ { 2 } } { m _ { L } } } f _ { 3 n } ( x _ { 1 L } ) - \frac { 1 } { m _ { R } ^ { 2 } } f _ { 3 n } ( x _ { 1 R } ) \right ) + \frac { 2 } { m _ { L } ^ { 2 } } f _ { 2 n } ( x _ { 1 L } ) \right ] \\ & + \frac { \alpha _ { \overline { \frac { M } { 2 } } } } \overline { \frac { M _ { 1 } } { m _ { \mu } } } \overline { \frac { m _ { 2 } } { m _ { L } } - m _ { R } ^ { 2 } \right ) m _ { L } ^ { 2 } \left [ \frac { 1 } { m _ { L } } } f _ { 3 n } ( x _ { 1 L } ) - \frac { 1 } { m _ { R } ^ { 2 } } f _ { 3 n } ( x _ { 1 R } ) \right ) + \frac { m _ { 2 } ^ { 2 } } { m _ { L } ^ { 2 } } f _ { 2 n } ( x _ { 1 R } ) \right ].$$
Here ( m 2 LR ii ) = m l i ( A i -µ t ∗ β ), x iA = | M /m 2 i | 2 A , x µA = | µ | 2 /m 2 A with i = 1 2 and , A = L, R . The explicit expressions for the loop functions are:
$$f _ { 1 n } ( x ) \ = \ \frac { - 1 7 x ^ { 3 } + 9 x ^ { 2 } + 9 x - 1 + 6 x ^ { 2 } ( x + 3 ) \ln x } { 2 4 ( 1 - x ) ^ { 5 } } \,,$$
$$f _ { 2 n } ( x ) \ = \ \frac { - 5 x ^ { 2 } + 4 x + 1 + 2 x ( x + 2 ) \ln x } { 4 ( 1 - x ) ^ { 4 } } \,,$$
$$f _ { 3 n } ( x ) \ = \ \frac { 1 + 2 x \ln x - x ^ { 2 } } { 2 ( 1 - x ) ^ { 3 } } \,,$$
$$f _ { 1 c } ( x ) \ = \ \frac { - x ^ { 3 } - 9 x ^ { 2 } + 9 x + 1 + 6 x ( x + 1 ) \ln x } { 6 ( 1 - x ) ^ { 5 } } \,,$$
$$f _ { 2 c } ( x ) \ = \ \frac { - x ^ { 2 } - 4 x + 5 + 2 ( 2 x + 1 ) \ln x } { 2 ( 1 - x ) ^ { 4 } } \,.$$
In the degenerate SUSY limit M 1 = M 2 = µ = m L = m R = ˜ one obtains m
$$A _ { L } ^ { i j ( n ) } & = - \frac { \alpha _ { 2 } } { 1 2 0 \pi } \frac { ( m _ { L L } ^ { 2 } ) _ { i j } } { \tilde { m } ^ { 4 } } \left [ \frac { 1 } { 8 } + t _ { \beta } \right ] + \frac { \alpha _ { Y } } { 1 2 0 \pi } \frac { ( m _ { L L } ^ { 2 } ) _ { i j } } { \tilde { m } ^ { 4 } } \left [ - \frac { 5 } { 8 } + t _ { \beta } \right ] + \frac { \alpha _ { Y } } { 4 8 \pi } \frac { \tilde { m } } { m _ { l _ { i } } } \frac { ( m _ { L R } ^ { 2 } ) _ { j i } } { \tilde { m } ^ { 4 } } \\ & - \frac { \alpha _ { Y } } { 8 0 \pi } \frac { \tilde { m } } { m _ { l _ { i } } } \left [ \frac { ( m _ { L R } ^ { 2 } m _ { R R } ^ { 2 } ) _ { j i } ^ { * } } { \tilde { m } ^ { 6 } } + \frac { ( m _ { L L } ^ { 2 } m _ { L R } ^ { 2 } ) _ { j i } ^ { * } } { \tilde { m } ^ { 6 } } - \frac { 2 } { 3 } \frac { ( m _ { L L } ^ { 2 } m _ { L R } ^ { 2 } m _ { R R } ^ { 2 } ) _ { j i } ^ { * } } { \tilde { m } ^ { 8 } } \right ],$$
$$A _ { L } ^ { i j ( c ) } = \frac { \alpha _ { 2 } } { 4 0 \pi } \frac { ( m _ { L L } ^ { 2 } ) _ { i j } } { \tilde { m } ^ { 4 } } \left [ \frac { 1 } { 3 } + t _ { \beta } \right ] \,,$$
$$A _ { R } ^ { ( n ) } & = - \frac { \alpha _ { Y } } { 6 0 \pi } \frac { ( m _ { R R } ^ { 2 } ) _ { i j } } { \tilde { m } ^ { 4 } } \left [ \frac { 1 } { 2 } + t _ { \beta } \right ] + \frac { \alpha _ { Y } } { 4 8 \pi } \frac { \tilde { m } } { m _ { l _ { i } } } \frac { ( m _ { L R } ^ { 2 } ) _ { i j } } { \tilde { m } ^ { 4 } } - \frac { \alpha _ { Y } } { 8 0 \pi } \frac { \tilde { m } } { m _ { l _ { i } } } \frac { ( m _ { L R } ^ { 2 } m _ { R R } ^ { 2 } ) _ { i j } } { \tilde { m } ^ { 6 } } \\ & - \frac { \alpha _ { Y } } { 8 0 \pi } \frac { \tilde { m } } { m _ { l _ { i } } } \frac { ( m _ { L L } ^ { 2 } m _ { L R } ^ { 2 } ) _ { i j } } { \tilde { m } ^ { 6 } } + \frac { \alpha _ { Y } } { 1 2 0 \pi } \frac { \tilde { m } } { m _ { l _ { i } } } \frac { ( m _ { L L } ^ { 2 } m _ { L R } ^ { 2 } m _ { R R } ^ { 2 } ) _ { i j } } { \tilde { m } ^ { 8 } } \,.$$
## A.2 Anomalous Magnetic Moments
The supersymmetric contributions to the anomalous magnetic moment ∆ a l i come from neutralino and chargino loops ∆ a l i = ∆ a ( n ) l i +∆ a ( c ) l i . They read:
$$\cdot \Big |$$
$$\begin{array} { c } \text{The supersymmetric contributions to the anomalous magnetic moment } \Delta a _ { i } = \Delta a _ { i } ^ { ( n ) } + \Delta a _ { i } ^ { ( c ) }. \text{They read} \colon \\ \Delta a _ { i } ^ { ( n ) } = \frac { \alpha _ { 2 } } { 8 \pi } \frac { m _ { 2 } ^ { 2 } } { m _ { 2 } ^ { 2 } } \left [ - f _ { n } ^ { L } ( x _ { 2 L } ) + 2 ^ { | M _ { 2 } | ^ { 2 } } + R e ( \mu M _ { 2 } ) ^ { t _ { \beta } } f _ { 3 n } ( x _ { 2 L } ) - 2 ^ { | \mu | ^ { 2 } } + R e ( \mu M _ { 2 } ) ^ { t _ { \beta } } f _ { 3 n } ( x _ { \mu } | ) \right ] \\ \Delta a _ { i } ^ { ( n ) } = \frac { \alpha _ { 2 } } { 8 \pi } \frac { m _ { 2 } ^ { 2 } } { m _ { 2 } ^ { 2 } } \left [ L _ { n } ^ { L } ( x _ { 1 R } ) - \frac { | M _ { 1 } | ^ { 2 } + R e ( \mu M _ { 1 } ) ^ { t _ { \beta } } } { | M _ { 1 } | ^ { 2 } - | \mu | ^ { 2 } } f _ { 3 n } ( x _ { 1 R } ) \right ] \\ \Delta a _ { i } ^ { ( n ) } = \frac { m _ { 2 } ^ { 2 } } { m _ { 2 } ^ { 2 } } \left [ - f _ { n } ^ { L } ( x _ { 1 L } ) - 2 ^ { | M _ { 1 } | ^ { 2 } } + R e ( \mu M _ { 1 } ) ^ { t _ { \beta } } f _ { 3 n } ( x _ { 1 L } ) + 2 ^ { | \mu | ^ { 2 } } + R e ( \mu M _ { 1 } ) ^ { t _ { \beta } } f _ { 3 n } ( x _ { \mu } | ) \right ] \\ \Delta a _ { i } ^ { ( n ) } = \frac { m _ { 2 } ^ { 2 } } { m _ { 2 } ^ { 2 } } \left [ - f _ { n } ^ { L } ( x _ { 1 L } ) - 2 ^ { | M _ { 1 } | ^ { 2 } } + R e ( \mu M _ { 1 } ) ^ { t _ { \beta } } f _ { 3 n } ( x _ { 1 L } ) + 2 ^ { | \mu | ^ { 2 } } + R e ( \mu M _ { 1 } ) ^ { t _ { \beta } } f _ { 3 n } ( x _ { \mu } | ) \right ] \\ \Delta a _ { i } ^ { ( n ) } = \frac { m _ { 2 } ^ { 2 } } { m _ { 1 } ^ { 2 } } \left [ - m _ { 2 } ^ { R } R e ( M _ { 1 } m _ { 2 } ^ { 2 } R ) i i \left [ \frac { 1 \frac { m _ { 2 } ^ { 2 } } { m _ { 1 } ^ { 2 } } f _ { 3 n } ( x _ { 1 L } ) - \frac { 1 \frac { m _ { 2 } ^ { 2 } } { m _ { 2 } ^ { 2 } } f _ { 3 n } ( x _ { 1 R } ) } { | M _ { 1 } | ^ { 2 } - | \mu | ^ { 2 } } f _ { 3 n } ( x _ { 1 R } ) \right ] \\ \Delta a _ { i } ^ { ( n ) } = \frac { \alpha _ { 2 } } { m _ { 1 } ^ { 2 } } \left [ L _ { n } ^ { L } ( x _ { 2 L } ) - \frac { | M _ { 2 } | ^ { 2 } + R e ( \mu M _ { 2 } ) ^ { t _ { \beta } } } { | M _ { 2 } | ^ { 2 } - | \mu | ^ { 2 } } f _ { 3 n } ( x _ { 1 R } ) } + 2 ^ { 1 } \frac { 1 } { m _ { 4 } ^ { 2 } } f _ { 3 n } ( x _ { 1 R } ) \right ] \\ \Delta a _ { i } ^ { ( c ) } = \frac { \alpha _ { 2 } } { 4 \pi } \frac { m _ { 2 } ^ { 2 } } { m _ { 2 } ^ { 2 } } \left [ l _ { c } ^ { L } ( x _ { 2 L } ) - \frac { | M _ { 2 } | ^ { 2 } + R e ( \mu M _ { 2 } ) ^ { t _ { \beta } } } { | M _ { 2 } | ^ { 2 } - | \mu | ^ { 2 } } f _ { c } ^ { L } ( x _ { 2 L } ) + \frac { 1 \mu | ^ { 2 } + R e ( \mu M _ { 2 } ) ^ { t _ { \beta } } } { | M _ { 2 } | ^ { 2 } - | \mu | ^ { 2 } } f _ { c } ^ { L } ( x _ { \mu } | ) \right ]. \end{array}$$
$$\Delta _ { l _ { i } } ^ { ( c ) } = \frac { \alpha _ { 2 } } { 4 \pi } \frac { m _ { L _ { i } } ^ { 2 } } { m _ { L } ^ { 2 } } \left [ f _ { c } ^ { L } ( x _ { 2 L } ) - \frac { | M _ { 2 } | ^ { 2 } + \text{Re} ( \mu M _ { 2 } ) t _ { \beta } } { | M _ { 2 } | ^ { 2 } - | \mu | ^ { 2 } } f _ { c } ^ { L R } ( x _ { 2 L } ) + \frac { | \mu | ^ { 2 } + \text{Re} ( \mu M _ { 2 } ) t _ { \beta } } { | M _ { 2 } | ^ { 2 } - | \mu | ^ { 2 } } f _ { c } ^ { L R } ( x _ { \mu L } ) \right ] \,.$$
Here we introduced the additional loop functions:
$$f _ { n } ^ { L } ( x ) \ = \ \frac { 1 - 6 x + 3 x ^ { 2 } + 2 x ^ { 3 } - 6 x ^ { 2 } \log x } { 6 ( 1 - x ) ^ { 4 } } \,,$$
$$f _ { c } ^ { L } ( x ) \ = \ \frac { 2 + 3 x - 6 x ^ { 2 } + x ^ { 3 } + 6 x \log x } { 6 ( 1 - x ) ^ { 4 } } \,,$$
$$f _ { c } ^ { L R } ( x ) \ = \ \frac { - 3 + 4 x - \overset { \cdot } { x } ^ { 2 } - 2 \log x } { ( 1 - x ) ^ { 3 } } \,.$$
In the degenerate SUSY limit M 1 = M 2 = µ = m L = m R = ˜ one obtains m
$$\Delta a _ { l _ { i } } ^ { ( n ) } & = - \frac { \alpha _ { 2 } } { 4 8 \pi } \frac { m _ { l _ { i } } ^ { 2 } } { \tilde { m } ^ { 2 } } \left [ - \frac { 1 } { 2 } + \frac { \text{Re} ( \mu M _ { 2 } ) } { \tilde { m } ^ { 2 } } t _ { \beta } \right ] - \frac { \alpha _ { Y } } { 4 8 \pi } \frac { m _ { l _ { i } } ^ { 2 } } { \tilde { m } ^ { 2 } } \left [ \left ( \frac { 3 } { 2 } + \frac { \text{Re} ( \mu M _ { 1 } ) } { \tilde { m } ^ { 2 } } t _ { \beta } \right ) + \frac { 2 \text{Re} ( M _ { 1 } m _ { L _ { R } } ^ { 2 } ) i i } { m _ { l _ { i } } \tilde { m } ^ { 2 } } \right ] \\ & + \frac { \alpha _ { Y } } { 4 0 \pi } \frac { m _ { l _ { i } } } { \tilde { m } ^ { 2 } } \left [ \frac { \text{Re} ( M _ { 1 } m _ { L _ { R } } ^ { 2 } m _ { R R } ^ { 2 } ) i i } { \tilde { m } ^ { 4 } } + \frac { \text{Re} ( M _ { 1 } m _ { L L } ^ { 2 } m _ { L R } ^ { 2 } ) i i } { \tilde { m } ^ { 4 } } - \frac { 2 } { 3 } \frac { \text{Re} ( M _ { 1 } m _ { L L } ^ { 2 } m _ { L R } ^ { 2 } m _ { R R } ^ { 2 } ) i i } { \tilde { m } ^ { 6 } } \right ], \\ \cdots \quad \, _ { \quad \, \quad m ^ { 2 } } \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{$\Gamma$} \quad \, \text{\quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \,$$
$$\Delta a _ { l _ { i } } ^ { ( c ) } = \frac { \alpha _ { 2 } } { 8 \pi } \frac { m _ { l _ { i } } ^ { 2 } } { \tilde { m } ^ { 2 } } \left [ - \frac { 1 } { 6 } + \frac { \text{Re} ( \mu M _ { 2 } ) } { \tilde { m } ^ { 2 } } t _ { \beta } \right ].$$
## A.3 Electric Dipole Moments
The supersymmetric contributions to the Electric Dipole Moment d i come from neutralino and chargino loops d i = d ( n ) i + d ( c ) i . They read:
$$A. \, \text{Receive Dipode Volume} \\ \begin{array} { c c c } \frac { d _ { i } ^ { ( n ) } } { e } & = \frac { \alpha _ { 2 } } { 8 \pi } \, m _ { 1 } ^ { 2 } \, | M _ { 2 } | ^ { 2 } - | \mu | ^ { 2 } \beta \left [ f _ { 3 n } ( x _ { 1 } ) - f _ { 3 n } ( x _ { \mu } ) \right ] \\ & + \frac { \alpha _ { Y } } { 4 \pi } \, m _ { 1 } ^ { 2 } \, | M _ { 1 } | ^ { 2 } - | \mu | ^ { 2 } \, ^ { 2 } \beta \left [ f _ { 3 n } ( x _ { 1 } R ) - f _ { 3 n } ( x _ { \mu } R ) \right ] \\ & - \frac { \alpha _ { Y } } { 8 \pi } \, m _ { 1 } ^ { 2 } \, | M _ { 1 } | ^ { 2 } - | \mu | ^ { 2 } \beta \left [ f _ { 3 n } ( x _ { 1 } L ) - f _ { 3 n } ( x _ { \mu } L ) \right ] \\ & + \frac { \alpha _ { Y } } { 4 \pi } \, | M _ { 1 } | ^ { 2 } - | m _ { 2 } ^ { 2 } \, | \frac { 1 } { m _ { 1 } ^ { 2 } } \, f _ { 3 n } ( x _ { 1 } L ) - \frac { 1 } { m _ { 2 } ^ { 2 } } \, f _ { 3 n } ( x _ { 1 } R ) \right ] \\ & + \frac { \alpha _ { Y } } { 4 \pi } \, | M _ { 1 } | ^ { 2 } - | m _ { 2 } ^ { 2 } \, R _ { R } \right ) _ { i i } \left [ \frac { 1 } { m _ { 1 } ^ { 2 } } \, f _ { 3 n } ( x _ { 1 } L ) - \frac { 1 } { m _ { 2 } ^ { 2 } } \, f _ { 3 n } ( x _ { 1 } R ) \right ] \\ & + \frac { \alpha _ { Y } } { 4 \pi } \, | M _ { 1 } | ^ { 2 } - | m _ { 2 } ^ { 2 } \, R _ { R } \right ) _ { i i } \left [ \frac { 1 } { m _ { 1 } ^ { 2 } } \, f _ { 3 n } ( x _ { 1 } L ) - \frac { 1 } { m _ { 2 } ^ { 2 } } \, f _ { 3 n } ( x _ { 1 } R ) \right ] \\ & - \frac { \alpha _ { Y } } { M _ { 1 } | ^ { 2 } } \, | M _ { 1 } | ^ { 2 } - | m _ { 2 } ^ { 2 } \, R _ { R } \right ) _ { i i } \left [ \frac { 1 } { m _ { 1 } ^ { 2 } } \, f _ { 3 n } ( x _ { 1 } L ) - \frac { 1 } { m _ { 2 } ^ { 2 } } \, f _ { 3 n } ( x _ { 1 } R ) \right ] \\ & - \frac { \alpha _ { Y } } { M _ { 1 } | ^ { 2 } } \, | M _ { 1 } | ^ { 2 } - | m _ { 2 } ^ { 2 } \, R _ { R } \right ) _ { i i } \times \\ & \frac { \left [ \frac { m _ { 1 } } { m _ { 1 } ^ { 2 } } \, f _ { 3 n } ( x _ { 1 } L ) - \frac { 1 } { m _ { 2 } ^ { 2 } } \, f _ { 3 n } ( x _ { 1 } R ) \right ] \\ & - \frac { m _ { 1 } | ^ { 2 } } { m _ { 1 } | ^ { 2 } - | m _ { 2 } ^ { 2 } } \, + \frac { m _ { 1 } ^ { 2 } } { m _ { 1 } ^ { 2 } } \, f _ { 3 n } ( x _ { 1 } L ) \, + \frac { 1 } { m _ { 1 } ^ { 2 } } \, f _ { 3 n } ( x _ { 1 } R ) \right ], \\ \frac { d _ { i } ^ { ( c ) } } { e } & = - \frac { \alpha _ { 2 } } { 8 \pi } \, m _ { 1 } ^ { 2 } \, | M _ { 2 } | ^ { 2 } - | \mu | ^ { 2 } \, ^ { 2 } \beta \left [ f _ { c } ^ { L } ( x _ { 2 } L ) - f _ { c } ^ { L } ( x _ { \mu } L ) \right ]. \\ \end{array}$$
In the degenerate SUSY limit M 1 = M 2 = µ = m L = m R = ˜ one obtains m
$$\frac { d _ { i } ^ { ( n ) } } { e } & = - \frac { \alpha _ { 2 } } { 9 6 \pi } \frac { m _ { l _ { i } } } { \tilde { m } ^ { 2 } } \frac { \text{Im} ( \mu M _ { 2 } ) } { \tilde { m } ^ { 2 } } t _ { \beta } - \frac { \alpha _ { Y } } { 9 6 \pi } \frac { m _ { l _ { i } } } { \tilde { m } ^ { 2 } } \frac { \text{Im} ( \mu M _ { 1 } ) } { \tilde { m } ^ { 2 } } t _ { \beta } - \frac { \alpha _ { Y } } { 1 2 0 \pi } \frac { \text{Im} ( M _ { 1 } m _ { L L } ^ { 2 } m _ { L R } ^ { 2 } m _ { R R } ^ { 2 } ) _ { i i } } { \tilde { m } ^ { 8 } } \\ & - \frac { \alpha _ { Y } } { 4 8 \pi } \frac { \text{Im} ( M _ { 1 } m _ { L R } ^ { 2 } ) _ { i i } } { \tilde { m } ^ { 4 } } + \frac { \alpha _ { Y } } { 8 0 \pi } \frac { \text{Im} ( M _ { 1 } m _ { L R } ^ { 2 } m _ { R R } ^ { 2 } ) _ { i i } } { \tilde { m } ^ { 6 } } + \frac { \alpha _ { Y } } { 8 0 \pi } \frac { \text{Im} ( M _ { 1 } m _ { L L } ^ { 2 } m _ { L R } ^ { 2 } ) _ { i i } } { \tilde { m } ^ { 6 } } \,,$$
$$\frac { d _ { i } ^ { ( c ) } } { e } = \frac { \alpha _ { 2 } } { 1 6 \pi } \frac { m _ { l _ { i } } } { \tilde { m } ^ { 2 } } \frac { \text{Im} ( \mu M _ { 2 } ) } { \tilde { m } ^ { 2 } } t _ { \beta } \,.$$
## References
- [1] C. D. Froggatt and H. B. Nielsen, Nucl. Phys. B 147 (1979) 277.
- [2] M. Leurer, Y. Nir and N. Seiberg, Nucl. Phys. B 398 (1993) 319 [hep-ph/9212278].
- [3] M. Leurer, Y. Nir and N. Seiberg, Nucl. Phys. B 420 (1994) 468 [hep-ph/9310320].
- [4] P. Binetruy and P. Ramond, Phys. Lett. B 350 (1995) 49 [hep-ph/9412385].
- [5] L. J. Hall, H. Murayama and N. Weiner, Phys. Rev. Lett. 84 (2000) 2572 [hepph/9911341].
- [6] N. Haba and H. Murayama, Phys. Rev. D 63 (2001) 053010 [hep-ph/0009174].
- [7] G. Altarelli, F. Feruglio and I. Masina, JHEP 0301 (2003) 035 [hep-ph/0210342].
- [8] A. de Gouvea and H. Murayama, Phys. Lett. B 573 (2003) 94 [hep-ph/0301050].
- [9] A. de Gouvea and H. Murayama, arXiv:1204.1249 [hep-ph].
- [10] G. Altarelli, F. Feruglio, I. Masina and L. Merlo, JHEP 1211 (2012) 139 [arXiv:1207.0587 [hep-ph]].
- [11] J. Bergstrom, D. Meloni and L. Merlo, arXiv:1403.4528 [hep-ph].
- [12] L. Calibbi, Z. Lalak, S. Pokorski and R. Ziegler, JHEP 1207 (2012) 004 [arXiv:1204.1275 [hep-ph]].
- [13] L. Calibbi, Z. Lalak, S. Pokorski and R. Ziegler, JHEP 1206 , 018 (2012) [arXiv:1203.1489 [hep-ph]].
[14]
G. F. Giudice and R. Rattazzi, Phys. Rept.
322
(1999) 419 [hep-ph/9801271].
- [15] G. D'Ambrosio, G. F. Giudice, G. Isidori and A. Strumia, Nucl. Phys. B 645 (2002) 155 [hep-ph/0207036].
- [16] Y. Shadmi and P. Z. Szabo, JHEP 1206 (2012) 124 [arXiv:1103.0292 [hep-ph]].
- [17] J. L. Evans, M. Ibe and T. T. Yanagida, Phys. Lett. B 705 (2011) 342 [arXiv:1107.3006 [hep-ph]].
- [18] P. Draper, P. Meade, M. Reece and D. Shih, Phys. Rev. D 85 (2012) 095007 [arXiv:1112.3068 [hep-ph]].
- [19] J. L. Evans, M. Ibe, S. Shirai and T. T. Yanagida, Phys. Rev. D 85 (2012) 095004 [arXiv:1201.2611 [hep-ph]].
- [20] Z. Kang, T. Li, T. Liu, C. Tong and J. M. Yang, Phys. Rev. D 86 (2012) 095020 [arXiv:1203.2336 [hep-ph]].
- [21] N. Craig, S. Knapen, D. Shih and Y. Zhao, JHEP 1303 (2013) 154 [arXiv:1206.4086 [hep-ph]].
- [22] A. Albaid and K. S. Babu, Phys. Rev. D 88 (2013) 055007 [arXiv:1207.1014 [hep-ph]].
- [23] M. Abdullah, I. Galon, Y. Shadmi and Y. Shirman, JHEP 1306 , 057 (2013) [arXiv:1209.4904 [hep-ph]].
- [24] P. Byakti and T. S. Ray, JHEP 1305 , 055 (2013) [arXiv:1301.7605 [hep-ph]].
- [25] J. A. Evans and D. Shih, JHEP 1308 (2013) 093 [arXiv:1303.0228 [hep-ph]].
- [26] T. Jelinski, JHEP 1309 (2013) 107 [arXiv:1305.6277 [hep-ph]].
- [27] T. Jelinski and J. Pawelczyk, arXiv:1406.4001 [hep-ph].
- [28] L. Calibbi, P. Paradisi and R. Ziegler, JHEP 1306 (2013) 052 [arXiv:1304.1453 [hep-ph]].
- [29] S. Davidson, G. Isidori and S. Uhlig, Phys. Lett. B 663 (2008) 73 [arXiv:0711.3376 [hep-ph]].
- [30] E. Dudas, G. von Gersdorff, J. Parmentier and S. Pokorski, JHEP 1012 (2010) 015 [arXiv:1007.5208 [hep-ph]].
- [31] Y. Nomura, M. Papucci and D. Stolarski, Phys. Rev. D 77 (2008) 075006 [arXiv:0712.2074 [hep-ph]]; Y. Nomura, M. Papucci and D. Stolarski, JHEP 0807 (2008) 055 [arXiv:0802.2582 [hep-ph]]; Y. Nomura and D. Stolarski, Phys. Rev. D 78 (2008) 095011 [arXiv:0808.1380 [hep-ph]].
- [32] B. Keren-Zur, P. Lodone, M. Nardecchia, D. Pappadopulo, R. Rattazzi and L. Vecchi, Nucl. Phys. B 867 , 429 (2013) [arXiv:1205.5803 [hep-ph]].
- [33] G. F. Giudice, P. Paradisi and M. Passera, JHEP 1211 (2012) 113 [arXiv:1208.6583 [hep-ph]].
- [34] M. Abdullah, I. Galon, Y. Shadmi and Y. Shirman, arXiv:1209.4904 [hep-ph].
- [35] I. Galon, G. Perez and Y. Shadmi, JHEP 1309 (2013) 117 [arXiv:1306.6631 [hep-ph]].
- [36] Work in progress.
- [37] W. Buchmuller, V. Domcke and K. Schmitz, JHEP 1203 (2012) 008 [arXiv:1111.3872 [hep-ph]].
- [38] G. Altarelli and F. Feruglio, New J. Phys. 6 (2004) 106 [hep-ph/0405048].
- [39] J. Adam et al. [MEG Collaboration], Phys. Rev. Lett. 110 (2013) 20, 201801 [arXiv:1303.0754 [hep-ex]].
- [40] A. M. Baldini, F. Cei, C. Cerri, S. Dussoni, L. Galli, M. Grassi, D. Nicolo and F. Raffaelli et al. , arXiv:1301.7225 [physics.ins-det].
- [41] U. Bellgardt et al. [SINDRUM Collaboration], Nucl. Phys. B 299 (1988) 1.
- [42] A. Blondel, A. Bravar, M. Pohl, S. Bachmann, N. Berger, M. Kiehn, A. Schoning and D. Wiedner et al. , arXiv:1301.6113 [physics.ins-det].
- [43] W. H. Bertl et al. [SINDRUM II Collaboration], Eur. Phys. J. C 47 (2006) 337.
- [44] C. Dohmen et al. [SINDRUM II. Collaboration], Phys. Lett. B 317 (1993) 631.
- [45] COMET Collaboration, http://comet.kek.jp/Documents.html
- [46] R. J. Abrams et al. [The mu2e Project Collaboration], arXiv:1211.7019 [physics.ins-det].
- [47] J. Baron et al. [ACME Collaboration], Science 343 (2014) 6168, 269 [arXiv:1310.7534 [physics.atom-ph]].
- [48] G.W. Bennett et al. , Phys. Rev. D 73 (2006) 072003; Phys. Rev. Lett. 92 (2004) 161802; 89 (2002) 101804; 89 (2002) 129903(E); H.N. Brown et al. , Phys. Rev. Lett. 86 (2001) 2227.
- [49] F. Jegerlehner and A. Nyffeler, Phys. Rept. 477 (2009) 1.
- [50] K. Hagiwara, R. Liao, A.D. Martin, D. Nomura and T. Teubner, J. Phys. G 38 (2011) 085003.
- [51] M. Davier, A. Hoecker, B. Malaescu and Z. Zhang, Eur. Phys. J. C 71 (2011) 1515.
- [52] P. Meade, N. Seiberg and D. Shih, Prog. Theor. Phys. Suppl. 177 (2009) 143 [arXiv:0801.3278 [hep-ph]].
- [53] G. Aad et al. [ATLAS Collaboration], JHEP 1405 (2014) 071 [arXiv:1403.5294 [hep-ex]].
- [54] S. Chatrchyan et al. [CMS Collaboration], JHEP 1307 (2013) 122 [arXiv:1305.0491 [hepex]].
- [55] L. Calibbi, A. Mariotti, C. Petersson and D. Redigolo, arXiv:1405.4859 [hep-ph].
- [56] V. Khachatryan et al. [CMS Collaboration], arXiv:1405.7570 [hep-ex].
- [57] M. Endo, K. Hamaguchi, S. Iwamoto and T. Yoshinaga, JHEP 1401 (2014) 123 [arXiv:1303.4256 [hep-ph]].
- [58] M. Endo, K. Hamaguchi, T. Kitahara and T. Yoshinaga, JHEP 1311 (2013) 013 [arXiv:1309.3065 [hep-ph]].
- [59] J. Hisano, T. Moroi, K. Tobe and M. Yamaguchi, Phys. Rev. D 53 (1996) 2442 [hepph/9510309].
- [60] T. Moroi, Phys. Rev. D 53 (1996) 6565 [Erratum-ibid. D 56 (1997) 4424] [hepph/9512396].
- [61] P. Paradisi, JHEP 0510 (2005) 006 [hep-ph/0505046].
|
10.1140/epjc/s10052-014-3211-x
|
[
"Lorenzo Calibbi",
"Paride Paradisi",
"Robert Ziegler"
] |
2014-08-04T18:04:06+00:00
|
2014-08-04T18:04:06+00:00
|
[
"hep-ph"
] |
Lepton Flavor Violation in Flavored Gauge Mediation
|
We study the anatomy and phenomenology of Lepton Flavor Violation (LFV) in
the context of Flavored Gauge Mediation (FGM). Within FGM, the messenger sector
couples directly to the MSSM matter fields with couplings controlled by the
same dynamics that explains the hierarchies in the SM Yukawas. Although the
pattern of flavor violation depends on the particular underlying flavor model,
FGM provides a built-in flavor suppression similar to wave function
renormalization or SUSY Partial Compositeness. Moreover, in contrast to these
models, there is an additional suppression of left-right (LR) flavor
transitions by third-generation Yukawas that in particular provides an extra
protection against flavor-blind phases. We exploit the consequences of this
setup for lepton flavor phenomenology, assuming that the new couplings are
controlled by simple U(1) flavor models that have been proposed to accommodate
large neutrino mixing angles. Remarkably, it turns out that in the context of
FGM these models can pass the impressive constraints from LFV processes and
leptonic EDMs even for light superpartners, therefore offering the possibility
of resolving the longstanding muon g-2 anomaly.
|
1408.0755v1
|
## Immune Evasion through Competitive Inhibition: the Shielding Effect of non-Stem
## Cancer Cells
## Irina Kareva 1
1 Newman Lakka Institute, Floating Hospital for Children, Tufts Medical Center,
75 Kneeland St., Boston, MA, 02111
Email: [email protected]
## Abstract
It has been recently proposed that the two 'emerging' hallmarks of cancer, namely altered glucose metabolism and immune evasion, may in fact be fundamentally linked (1). This connection comes from up-regulation of glycolysis by tumor cells, which can lead to active competition for resources in the tumor microenvironment between tumor and immune cells. Here it is further proposed that cancer stem cells (CSCs) can circumvent the anti-tumor immune response by creating a 'protective shield' of non-stem cancer cells around them. This shield can protect the CSCs both by creating a physical barrier between them and cytotoxic lymphocytes (CTLs), and by promoting competition for the common resources, such as glucose, between non-stem cancer cells and CTLs. The implications of this hypothesis are investigate using an agent-based model, leading to a prediction that relative CSC to non-CSC ratio will vary with the strength of the host immune response, with the highest occurring at an intermediate state of immune
activation. A discussion of possible therapeutic approaches concludes the paper, suggesting that a chemotherapeutic regimen consisting of regular pulsed doses, i.e., metronomic chemotherapy, would yield the best clinical outcome by allowing CTLs to most effectively reach and eliminate CSCs.
## Introduction
The immune system is generally very effective in preventing budding tumors from progressing to a fully malignant disease (2-4). However, as the cytotoxic cells of the immune system attempt to eliminate the tumor, they effectively select for nonimmunogenic clones, thus delaying but not arresting tumor progression, a process that has become known as tumor immunoediting (2; 5-7). Moreover, tumors that have been 'edited' in this fashion are typically more aggressive than their 'unedited' counterparts due to selection for more immunoresistant clones; a property first demonstrated by Shankaran and colleagues (8).
It has been recently proposed that the process of immunoediting may additionally be driven by competition for resources between cancer and immune cells in the tumor microenvironment (1). Specifically, the following microenvironmentally-based immunoediting scenario has been proposed:
- 1) As the tumor increases in size while not yet recognized by the cytotoxic cells of the immune system, it begins to outgrow its blood supply, causing the formation of oxygen-deprived regions in the tumor interior.
- 2) Intra-tumoral hypoxia causes the affected tumor cells to turn to glycolysis as a primary mode of glucose metabolism; an event known to be accompanied by an up to 30-fold up-regulation of the activity of glucose transporters (9).
- 3) Once activated, the CTLs first attack the topologically accessible outer rim of the tumor, what we term the 'protective shield', thus effectively exposing the tumor's 'glycolytic core'.
- 4) The exposed glycolytic tumor cells, which have been forced to up-regulate their glucose transporters, now actively compete with CTLs for glucose.
- 5) In a nutrient-deprived state, CTLs are less able to perform their tumoricidal function, allowing the tumor as a whole to evade immune attack.
In this way, the process of immune evasion may be the result of a competition for resources in the tumor microenvironment.
Here, this construct is expanded to incorporate cancer stem cells (CSCs). Because CSCs comprise a relatively small fraction of the tumor population on account of their high probability of asymmetric division (producing one CSC and one non-stem cell) (10), they will tend to surround themselves with non-stem progeny, which then could follow the aforementioned immunoediting scenario . The non-CSCs will then protect CSCs from cytotoxic lymphocytes both by creating a physical barrier and through outcompeting CTLs for essential nutrients. These considerations are summarized in Figure 1.
Figure 1. Schematic representation of the proposed immunoediting construct, which incorporates CSCs using aerobic and particularly glycolytic non-CSCs as a protective 'shield' from cytotoxic lymphocytes.
The proposed theoretical framework allows making predictions about CSC content in different tumors with regards to the immune response, which can be of vital importance, as CSCs, the driving cells in cancer progression, are believed to be more resistant to cytotoxic therapy than non-CSCs (11-14). In this case, if CSCs produced too few non-CSCs, protection from the immune system could be insufficient. If CSCs produced too many non-CSCs, then the progeny may themselves become limiting to CSC expansion, offsetting their benefit as an immune shield. Successful tumors might therefore be predicted to be those that have some optimal intermediate CSC-to-progeny
proportion. In addition, this proportion may well be different for different initial states of the immune system.
To test these ideas, an agent-based model is implemented, incorporating the foregoing and other basic assumptions about the properties of CSCs, non-CSCs, immune cells and their interactions with each other and with the resources in the tumor microenvironment. We use this model to illustrate the proposed immunoediting construct and in particular, to evaluate the proposition that varying proportions of CSCs to nonCSCs might be expected in successful tumors under different immune conditions.
## Model description
Here we build upon a previously introduced agent-based model (15), adapting it to describe the interactions between cancer stem cells (CSCs), aerobic and glycolytic nonstem cancer cells (non-CSCs), and cytotoxic lymphocytes (CTLs) with each other and their microenvironment. The model is implemented in Netlogo 5.0.2, a freely available agent-based modeling platform (16). Space in this simulation is represented by a 2dimensional 51x51 lattice. Each patch (square) in the lattice represents a discrete microenvironment, the properties of which are detailed below, for a total of 2601 microenvironments. Cells occupy coordinates in continuous space, and more than one cell can occupy a single microenvironment. Time is modeled in discrete steps.
## Parameter estimation
Parameter estimation is a common challenge in analyzing results from computational models. One particular difficulty lies in the fact that the estimates obtained in an experimental setting are typically a result of a multitude of complex
interconnected processes, which cannot always be separated out for the purposes of a conceptual proof-of-concept model. We address this issue in the following ways. A number of parameters, such as cell growth rates, were previously estimated in a model introduced and analyzed in (15), where the authors utilize an agent-based model to evaluate the impacts of disregulated glucose metabolism on overall tumor dynamics. We modify their proposed model, shifting the focus of our investigation away from evolution of cell motility and instead introducing immune cells and cancer stem cells as active competitors for glucose in the tumor microenvironment. Nevertheless, parameters that pertain to tumor growth rates and glucose consumption rates can be well translated into our model.
For a number of other parameters, while it may not always be possible to obtain exact values for reasons outlined above, we attempt to make estimations based on the parameters' relative values to each other. That is, while we may not be able to estimate exact rates of glucose consumption by cytotoxic immune cells, for instance, we can nevertheless support the notion that highly proliferative CTLs, just as glycolytic tumor cells, consume more energy than do aerobic tumor cells (17). Morover, Buttgereit and Brand (18) reported that in a model of concanavlin Astimulated thymocytes, where it is possible to account for over 80% of ATP consumption, protein synthesis was the most sensitive to limitations in glucose supply, followed by DNA and RNA synthesis and then by ion transport across plasma membrane. Motility also has been reported to incur significant ATP costs for immune cells due to energy requirements for cytoskeleton rearrangement during movement (19). Therefore, while we may not be able to obtain the exact values for
the costs of motility and cytotoxicity for CTLs, we can nevertheless assert that these costs exist, and are incorporated as such. Moreover, since we do not have exact estimates for these costs, they are taken to be a unit of energy each, an assumption which can be relaxed or modified pending data availability.
The specific assumptions for each of the cell types and for the microenvironments are as follows:
Properties of aerobic non-CSCs : Aerobic non-CSCs require glucose and oxygen for survival and glucose for proliferation, although they have lower glucose requirements than glycolytic non-CSCs. They die at some natural rate, as well as after a certain number of divisions, and can also be killed by CTLs. They can 'switch' to the glycolytic mode of metabolism under conditions of oxygen deprivation. When a cell divides, the daughter cell moves to a random neighboring patch.
Properties of glycolytic non-CSCs : Glycolytic non-CSCs do not require oxygen, but require glucose for survival and proliferation; they consume more glucose to support these processes than aerobic non-CSCs. They die at some natural rate, as well as after a certain number of divisions, and can also be killed by CTLs. They can 'switch' to the aerobic mode of glucose metabolism under conditions of sufficient oxygen availability. When a cell divides, the daughter cell moves to a random neighboring patch.
Properties of cancer stem cells : Depending on oxygen availability, CSCs consume oxygen and glucose like aerobic or glycolytic cells. Each CSC division occurs symmetrically with a certain probability. If a CSC divides asymmetrically, the daughter non-CSC cell is initialized to be aerobic, a state which may change. Each cell can die due
to nutrient deprivation or be killed by CTLs. CSCs can divide an unlimited number of times. When a CSC divides, the daughter cell moves to a random neighboring patch.
Properties of cytotoxic immune cells : It is assumed that all cytotoxic CTLs rely on glycolysis for glucose metabolism, consuming nutrients at the same rate as glycolytic tumor cells. CTLs can proliferate only under conditions of sufficient glucose supply. They can die from nutrient deprivation. Unlike tumor cells, CTLs are motile and move randomly throughout the microenvironment. As motility is costly, each step to a random neighboring patch costs CTLs a unit of energy. CTLs can kill tumor cells once they encounter them; however, killing a tumor cell also costs a unit of energy. CTLs are replenished at some constant rate when the number of immune cells in the tumor microenvironment becomes too low.
Properties of the microenvironment : On each patch there is a certain amount of glucose and oxygen. Both glucose and oxygen are replenished differentially, depending on the number of cells on the patch. Too many cells on the patch lowers the rate of replenishment, which simulates the crowding effect. Glucose and oxygen are taken up by different cell types at different rates, depending on whether the cell type is aerobic, glycolytic, stem or immune.
It is worth noting that while alternative energy sources, such as glutamine, are available to tumor cells, it has been shown by MacIver and colleagues (20) that activated T cells are unable to perform their function in the absence of glucose, even in the presence of amply available glutamine. Therefore, glutamine was not incorporated as an
energy source in the proposed model as it is not the driving force for competition between tumor and immune cells.
The model is initialized with a single CSC and a pre-set initial number of cytotoxic immune cells placed randomly throughout the lattice. At each time step, the following sequence of steps is realized:
- 1) Each tumor cell evaluates oxygen availability in its microenvironment, which determines the cell's subsequent mode of glucose metabolism.
- 2) Each cell consumes glucose, subtracting differentially from the glucose available on the patch and adding a number of energy units to the cell's storage, according to its 'metabolic phenotype', i.e., aerobic or glycolytic; glycolytic cells consume glucose, while aerobic cells consume both glucose and oxygen.
- 3) Cytotoxic lymphocytes move randomly throughout the microenvironment, killing the tumor cells that they encounter; both motility and the process of cell killing have a cost of a unit of energy each; CTLs die if their energy stores are depleted and cannot be replenished.
- 4) Each cell reproduces with a certain fixed probability, given adequate resources in the microenvironment; the daughter cell moves to a random neighboring patch; if the cell is a CSC, it divides symmetrically or asymmetrically with a certain probability; in case of asymmetric division, the daughter cell is initialized as aerobic non-CSC.
- 5) Glucose and oxygen are replenished in the microenvironment depending on the number of cells on each patch, with a higher rate of replenishment in the areas of lower cell density.
- 6) CTLs are replenished by the value of parameter i0 if the number of immune cells on the lattice falls below a critical threshold.
Since the purpose of the proposed model is to offer a 'proof-of-concept' as to whether tumor composition can indeed be affected by the initial state of the immune system of the host, parameter values were chosen as relative to each other rather than matching them to fit a particular data set.
It is hypothesized that in successfully formed tumors, the relative proportion of non-CSCs will increase proportionally to the strength of the immune response because CSCs will require the shielding effect of non-CSCs to protect them from CTLs. Specifically, an increase in the proportion of glycolytic non-CSCs is predicted, since they are capable of competing most effectively with CTLs for glucose in the tumor microenvironment.
## Results
The numerical experiments were designed in such a way as to enable evaluation of tumor composition at each time point with respect to the state of the immune system. The initial state of the immune response was modeled with parameter i0=0, 10, 20, 30, 35, 40. For each i0, 20 simulations were run for 1000 time steps, and the mean number of CSCs, aerobic and glycolytic non-CSCs were reported; only the data for tumors that had at least one CSC at t=1000 were collected. The data is represented in two types of plots: the
change over time of tumor composition for each respective i0 and the final ratio of CSC to non-CSC at t=1000 for each i0.
## Proportion of glycolytic cells increases with the strength of the immune response
Our simulations show that indeed, stronger immune response predicts larger overall proportion of glycolytic non-CSCs in the tumor (see Figure 2), which within the proposed theoretical framework is explained by the increased 'need' of CSCs to protect themselves from the immune system. However, the largest proportion of CSCs in the tumor at t=1000 was observed at the intermediate values of i0 (see Figure 3).
Figure 2. Change in tumor composition of CSCs (green) to glycolytic nCSCs (red) to aerobic nCSCs (blue) over time for varying initial states of the host immune system, represented by parameter i0. As one can see, the proportion of glycolytic nCSCs tends to increase with i0 as CSCs require increased protection from CTLs. Threshold of sensitivity to oxygen for all cells is set to ω =5.
## CSC:nCSC at t=1000
Figure 3. Proportion of CSCs and nCSCs at t=1000 for all varying initial states of the host immune system, represented by parameter i0. As one can see, the highest proportion of CSCs is observed not for immunodeficient or extremely immunocompetent host, but for intermediate values of i0. Threshold of sensitivity to oxygen for all cells is set to ω =5.
The following explanation for this phenomenon is proposed:
At low values of i0, aerobic non-CSCs that are located on the tumor rim have the greatest competitive advantage compared to other tumor cells, as they have the most access to oxygen and glucose compared to both CSCs and glycolytic non-CSCs inside the tumor. Therefore, since they are not 'sacrificed' to the immune system, aerobic non-CSCs outcompete CSCs.
At high values of i0, the tumor is not able to grow to a large enough size to create sufficiently hypoxic regions that would allow for the appearance of large clusters of
glycolytic non-CSCs that would present sufficient competition to cytotoxic lymphocytes. In this case, the non-CSC 'shield' would not be strong enough to contain CTLs. At 'intermediate' values of i0, the tumor can grow to be large enough in size to have sufficiently large hypoxic regions to require up-regulation of purely glycolytic mode of glucose metabolism in a sufficiently large number of non-CSCs, while the immune system removes enough of the aerobic non-CSCs at the outer rim of the tumor to prevent them from outcompeting the rest of the tumor cells.
Therefore, one should expect the largest proportions of CSC in hosts with 'intermediate' states of immune system. (Noticeably, these observations can be made based on assumptions of relative values of parameters, without requiring exact parameter estimation from specific data sets).
It has been reported by some groups that CSCs may have lower immunogenicity compared to non-CSCs (21,22). Intriguingly, while the same pattern of results was observed for various modifications in parameter values for costs of CTL motility and cytotoxicity (data not reported), differences were observed when the probability of recognition of CSCs by CTLs was decreased by 50% (Figures 4 and 5); this value was chosen arbitrarily due to lack of specific data. It appears that the proportion of CSCs varies with respect to the values of i0 only if CSCs are as likely to be killed by CTLs upon encounter as are non-CSCs. Otherwise, proportion of CSC:non-CSC remains the same regardless of the strength of the immune response.
Figure 4. Change in tumor composition of CSCs (green) to glycolytic nCSCs (red) to aerobic nCSCs (blue) over time for varying initial states of the host immune system, represented by parameter i0. The probability of recognition of CSC by a CTL is 50% lower than probability of recognition of nCSC. Threshold of sensitivity to oxygen for all cells is set to ω =5.
## CSC:nCSC at t-1000, probability of CSC kill 50% lower compared to nCSC
Figure 5. Proportion of CSCs and nCSCs at t=1000 for all varying initial states of the host immune system, represented by parameter i0. The probability of recognition of CSC by a CTL is 50% lower than probability of recognition of nCSC. Threshold of sensitivity to oxygen for all cells is set to ω =5.
## Variability in sensitivity to oxygen does not affect final tumor composition regardless of the state of the immune system
In the previous set of in silico experiments, it was assumed that all cells have a fixed threshold of sensitivity to oxygen and therefore revert to either aerobic or glycolytic mode of glucose metabolism based on a criterion that is uniform for all cells. It is however possible that if there were heterogeneity among cells based on this metric, one could observe selection for cells that are either more or less sensitive to oxygen concentrations in their microenvironment depending on the state of the immune response. That is, one could observe selection towards lower oxygen sensitivity thresholds (described by the value of parameter ω) under the conditions of increased selective pressure from the immune system.
In order to evaluate this hypothesis, the above-described protocol was modified to introduce mutations in parameter ω every time a tumor cell divides. The changes in the mean value of ω were recorded for all three cell types for all i0. The overall tumor composition changed in the same way with respect to different i0 as it did for the case of constant ω (Figure 6), and the largest proportion of CSCs was also observed for the intermediate values of i0 (Figure 7). The mean value ω for glycolytic cancer cells did oscillate increasingly with stronger immune response (Figure 8). However, a 2-tailed type 1 t-test did not reveal any statistically significant changes either in overall tumor composition or in the values of ω at t=1000 for either of the three tumor cell types (Figure 9), These results suggest that it is not oxygen availability but competition for
glucose, both within the tumor and between cancer and immune cells, that is the driving force behind the observed dynamics of immune evasion.
## io = 0 for mutating 0
## io 10 for mutating 0
## io 20 for mutating @
% CSC
## io = 30 for mutating @
% aerobic
%
glycolytic
% CSC
% aerobic
% glycolytic
% CSC
io = 35 for mutating
0
% aerobic
%
glycolytic
% CSC
% aerobic
% glycolytic
% CSC
## io 40 for mutating 0
% aerobic
%
glycolytic
% CSC
Figure 6. Change in tumor composition of CSCs (green) to glycolytic nCSCs (red) to aerobic nCSCs (blue) over time for varying initial states of the host immune system, represented by parameter i0. As one can see, the proportion of glycolytic nCSCs tends to increase with i0 as CSCs require increased protection from CTLs. Threshold of sensitivity to oxygen for all cells is initiated at ω =5 but is allowed to mutate with each cell division for each cell.
Figure 7. Proportion of CSCs and nCSCs at t=1000 for all varying initial states of the host immune system, represented by parameter i0. As one can see, the highest proportion of CSCs is observed not for immunodeficient or extremely immunocompetent host, but for intermediate values of i0. Threshold of sensitivity to oxygen for all cells is initiated at ω =5 but is allowed to mutate with each cell division for each cell.
Figure 8. Change in the mutating threshold of sensitivity to oxygen (parameter ω ) over time for varying initial states of the host immune system, represented by parameter i0. As one can see, while ω oscillates increasingly for larger i0, the mean of ω remains largely the same for all three cell types.
Figure 9. Mean value of the threshold of sensitivity to oxygen (parameter ω ) at t=1000 for varying initial states of the host immune system, represented by parameter i0, compared to fixed value of ω =5 for the first set of simulations. While there are slight changes in ω for all three cell types, none of them were statistically significant as evaluated by a 2-tailed type 1 t-test, suggesting that it is not oxygen availability.
## Discussion
Here a theoretical framework is proposed, where CSCs can escape recognition by cytotoxic lymphocytes by creating a 'shield' of differentiated non-CSCs around them. These shielding cells not only provide a physical barrier to the immune cells but also create a local nutrient flux, thereby starving CTLs out of functionality. Specifically, we proposed that if a CSC divided asymmetrically before it was recognized by the CTLs, and if the progeny grew to form a sufficiently large tumor to cause the appearance of hypoxic regions and consequent reversion by many of the tumor cells to purely glycolytic mode of glucose metabolism, then, once the immune cells eventually clear the aerobic cells on the outer rim of the tumor, they will encounter competition for glucose from
glycolytic cancer cells. Successful acquisition of the common glucose, made possible by up-regulation of nutrient transporters of glycolytic cancer cells while in the hypoxic regions of the tumor, could cause incapacitation of CTLs due to starvation, thereby allowing for continued unrestrained tumor growth (see Figure 1). Within this construct one would also expect CSCs to be found primarily in hypoxic regions of the tumors, where they would be most protected from the immune response.
In order to evaluate the proposed hypothesis,.an agent-based model was created, focusing on interactions between CSCs, aerobic and glycolytic non-CSCs with varying glucose demands, , and motile cytotoxic immune cells, whose functionality is nutrient dependent. Simulations were run for 1000 time steps, for various initial states of the immune system of the host, from complete immunodeficiency to extremely strong immune response. Tumor composition was then evaluated for all of the initial immune states.
The proportion of glycolytic non-CSCs increased dramatically with increased immune response (Figure 2), supporting the initial hypothesis. Moreover, the proportion of CSCs was largest not for the strongest or weakest states of the immune response but for intermediate ones (Figure 3), supposedly because in this case the tumor can grow large enough to create hypoxic regions but it small enough that the non-stem progeny do not outcompete CSCs.
Next, we hypothesized that if each cell had an individual threshold of sensitivity to oxygen, which would determine oxygen-based threshold for a 'glycolytic switch', then over time selection towards cells with lower threshold could be observed. However,
while the threshold of sensitivity to oxygen ω did change slightly over time, the changes both in tumor composition, as well as in the mean values of ω for all three cell types was not statistically significant (Figures 8 and 9), suggesting that oxygen availability is less of a crucial determinant of the outcome of tumor progression compared to glucose.
It is of course not possible to claim that competition for common nutrients is the sole mechanism whereby CSCs can avoid recognition by the immune system. Instead, we propose that competition for resources can provide an additional mechanism whereby CTLs are incapacitated, contributing to immune evasion. Noticeably, modifying the model to increase or decrease the actual costs of motility and cytotoxicity of CTLs does not affect the predicted pattern of behavior (data not reported). Similarly, modification of the level of possible infiltration of the tumor by CTLs based on crowding effects did not result in qualitative changes in dynamics (data not reported). This can be explained by the fact that even when CTLs are physically able to infiltrate the simulated tumor, they are quickly starved by glycolytic tumor cells in the hypoxic tumor core, thus achieving the same qualitative effects as artificially imposed restrictions on CTL infiltration.
The predicted pattern dynamics were robust to changes in all parameters evaluated with the exception of sensitivity of CSCs to CTLs. Decreasing the probability of cell kill for CSCs upon encounter with CTLs by 50% compared to non-CSCs resulted in a similar pattern with respect to changes in tumor composition over time for various i0 (Figure 4) but the prediction about proportions of CSC:non-CSC at t=1000 was not preserved (Figure 5). Literature about sensitivity of CSCs to CTLs is inconclusive: while a number of authors report lower sensitivity of CSCs to being recognized by CTLs compared to non-CSCs (21, 22), it is unclear whether this is solely an intrinsic property of CSCs, such
as lower expression of MHC-II molecules (22), or a by-product of the 'shielding' effect of non-CSCs. Other groups report efficient recognition of CSCs by CTLs (23), which suggests that this is a question that in fact pends further investigation.
One possible way to understand the interactions of CSCs with their environment is from the point of view of normal processes that occur during embryogenesis or tissue regeneration. Tissue injury often results in locally created hypoxia due to damage to blood vessels, which results in local acidosis as the cells downstream from the damaged vessels become oxygen deprived. Lower pH appears to create a locally immunosuppressive environment (1), which would allow tissues to rebuild themselves. Normalized vascularization after tissue regeneration would allow for exposure of adult stem cells to immune cells, thus providing an additional regulatory mechanism to stop generation of new tissues. Tumors in turn could harness this mechanism, continuing to grow through creating a hypoxic environment, inadvertently protecting CSCs. This hypothesis will require further investigation.
## Game theory and cancer
It has become increasingly recognized and accepted that tumor dynamics is governed largely by the process of Darwinian selection, where tumor cells experience and respond to various selective pressures, such as competition for resources with each other and somatic cells, interaction with predators (the immune system), migration (metastases), all within the 'ecosystem' of the human body (24-31). One of the well-developed analytical approaches to understanding evolving populations is evolutionary game theory, which has already been partially applied to cancer (33-39). The main premise of this approach is
the assumption that individuals, such as cells, operate for immediate optimization of their payoff, which in the context of evolutionary game theory is fitness, i.e., survival and the number of progeny. Individuals may use different strategies to achieve this goal, and understanding of these strategies and their immediate consequences can provide understanding about why a system is evolving in a particular direction.
In the context of the proposed model, we assume that it is CSCs that want to maximize their fitness and produce a maximal number of other CSCs. They can achieve this goal using two strategies: symmetric or asymmetric division, yielding either two CSCs or one CSC and one non-CSC, respectively. Symmetric division allows yields more CSCs immediately. Asymmetric division may allow for survival of the progeny in the long term.
From this point of view, in the absence of the immune response, or if it did not affect the growth dynamics CSCs, one would expect eventual CSC enrichment, as is predicted by a number of previously published theoretical models(40, 41), since it is symmetric division that would allow CSCs to maximize their fitness. However, in the presence of the immune response, the 'optimal strategy' becomes mixed, since in order to maximize their fitness CSCs now have to divide asymmetrically to protect themselves from the immune cells, and the position of the 'equilibrium strategy', i.e., the number of times the CSC would need to divide symmetrically or asymmetrically in order to maximize its fitness, would be different depending on the initial state of the immune response. Perhaps, manipulating the tumor microenvironment in such a way as to select for the increased asymmetric CSC division could be a potentially fruitful strategy to make the tumor more vulnerable to anti-cancer therapies.
## Therapeutic implications
Cancer stem cells are hypothesized to be the driving force behind unrestrained tumor growth, as they are also resistant to a number of cytotoxic therapies (13; 42-45). One possible way to circumvent this problem could involve taking advantage of the immune cells, which might be able to eliminate CSCs, even if less efficiently compared to nonCSCs (23,46-48). However, if the proposed construct is correct, CTLs are unable to do so because of the 'shield' of differentiated tumor cells that restricts their access to CSCs. Therefore, the optimal therapeutic strategy in this case would be not the one to cause the most extensive cell mortality but the one that will facilitate the access of cytotoxic immune cells to CSCs.
Two main considerations CSCs and CTLs need to be taken into account. Firstly, both theoretical considerations, outlined here, and experimental evidence suggest that CSCs are to be primarily found in the hypoxic areas of the tumor (49, 50), where they are the most 'protected' from the immune cells. Secondly, extensive cytotoxic therapies are highly damaging to the immune cells themselves (51), and consequently gained access the tumor core and the CSCs would be ineffective if there remain no immune cells that could then attack the CSCs.
The three major approaches that can be taken in chemotherapy administration are MTD (maximum tolerated dose), metronomic, and continuous therapy, which is a particular case of the metronomic therapy.
As the name suggests, MTD involves administering the highest possible dose of cytotoxic drugs that a patient can tolerate, with the intention of damaging a maximal
number of cells and expectation that normal tissues would be able to recover better than cancerous ones. The major drawback of MTD from the point of view of the considerations, outlined above, is the damage that it causes to the immune cells (51). Therefore, even if this could be a good way to reach the CSCs, it is very likely that the time it would take for the anti-tumor immune cells to recover would be greater than the time it would take for the CSCs to form another 'shield'. Noticeably, these predictions should hold as long as CSCs can be recognized by CTLs.
Continuous therapy involves administering very low doses of chemotherapy on a daily basis, ensuring constant presence of the cytotoxic drug in the patient's body. While this approach is much less toxic to the patient, it still has a problem of giving no time for the anti-tumor immune cells to recover, since they would be constantly eliminated by continuous inflow of cytotoxic chemical agents.
Metronomic therapy provides an intermediate case between MTD and continuous therapy, and involves administering lower doses of chemotherapy at more frequent intervals, but not on a daily basis. This approach could ensure continual 'peeling off' of the outer layers of the tumor, which protect the CSCs, thus giving anti-tumor immune cells access to the tumor core, while allowing sufficient time for CTLs to recover from the damaging effects of the therapy and attack the CSCs, thereby giving the best chance for more successful long-term tumor elimination.
## Conclusions
Limited success of some therapies may be rooted not in the lack of effectiveness of the cytotoxic drugs themselves but in the lack of understanding of tumor topology. If the
differentiated non-CSC cells do indeed provide a barrier to both chemotherapeutic agents and the body's natural defenses, then the problem of halting cancer progression might be overcome with appropriate mode of therapy administration. Specifically, if therapyresistant CSCs do shield themselves from anti-tumor immune cells by 'hiding' in the tumor's hypoxic regions, then slowly exposing these areas using moderate but relatively frequently administered levels of chemotherapy might give a better chance for cytotoxic lymphocytes to eliminate the tumor.
## Acknowledgments
The author would like to thank Philip Hahnfeldt, Lynn Hlatky and Heiko Enderling for extremely valuable discussions and suggestions during manuscript preparation. This research was supported by the Office of Science (BER), U.S. Department of Energy, under Award Number DE-SC0001434 (to Philip Hahnfeldt).
## References
- 1. Kareva I, Hahnfeldt P. The Emerging "Hallmarks" of Metabolic Reprogramming and Immune Evasion: Distinct or Linked?. Cancer Res. 2013 Feb 19.
- 2. Dunn GP, Bruce AT, Ikeda H, Old LJ, Schreiber RD. Cancer immunoediting: from immunosurveillance to tumor escape. Nat Immunol. 2002 Nov;3(11):991-8.
- 3. Teng MW, Swann JB, Koebel CM, Schreiber RD, Smyth MJ. Immunemediated dormancy: an equilibrium with cancer. J Leukoc Biol. 2008 Oct;84(4):988-93.
- 4. Ruffell B, DeNardo DG, Affara NI, Coussens LM. Lymphocytes in cancer development: polarization towards pro-tumor immunity. Cytokine Growth Factor Rev. 2010 Feb;21(1):3-10.
- 5. Kim R, Emi M, Tanabe K. Cancer immunoediting from immune surveillance to immune escape. Immunology. 2007 May;121(1):1-14.
- 6. Dunn GP, Old LJ, Schreiber RD. The three Es of cancer immunoediting. Annu Rev Immunol. 2004;22:329-60.
- 7. Schreiber RD, Old LJ, Smyth MJ. Cancer immunoediting: integrating immunity's roles in cancer suppression and promotion. Science. 2011 Mar 25;331(6024):1565-70.
- 8. Shankaran V, Ikeda H, Bruce AT, White JM, Swanson PE, Old LJ, Schreiber RD. IFNgamma and lymphocytes prevent primary tumour development and shape tumour immunogenicity. Nature. 2001 Apr 26;410(6832):1107-11.
- 9. Ganapathy V, Thangaraju M, Prasad PD. Nutrient transporters in cancer: relevance to Warburg hypothesis and beyond. Pharmacol Ther. 2009 Jan;121(1):29-40.
- 10. Morrison SJ, Kimble J. Asymmetric and symmetric stem-cell divisions in development and cancer. Nature. 2006 Jun 29;441(7097):1068-74.
- 11. O'Hare T, Corbin AS, Druker BJ. Targeted CML therapy: controlling drug resistance, seeking cure. Curr Opin Genet Dev. 2006 Feb;16(1):92-9.
- 12. Oravecz-Wilson KI, Philips ST, Yilmaz OH, Ames HM, Li L, Crawford BD, Gauvin AM, Lucas PC, Sitwala K, Downing JR, Morrison SJ, Ross TS. Persistence of leukemia-initiating cells in a conditional knockin model of an imatinib-responsive myeloproliferative disorder. Cancer Cell. 2009 Aug 4;16(2):137-48.
- 13. Li X, Lewis MT, Huang J, Gutierrez C, Osborne CK, Wu MF, Hilsenbeck SG, Pavlick A, Zhang X, Chamness GC, Wong H, Rosen J, Chang JC. Intrinsic resistance of tumorigenic breast cancer cells to chemotherapy. J Natl Cancer Inst. 2008 May 7;100(9):672-9.
- 14. Diehn M, Cho RW, Lobo NA, Kalisky T, Dorie MJ, Kulp AN, Qian D, Lam JS, Ailles LE, Wong M, Joshua B, Kaplan MJ, Wapnir I, Dirbas FM, Somlo G, Garberoglio C, Paz B, Shen J, Lau SK, Quake SR, Brown JM, Weissman IL, Clarke MF. Association of reactive oxygen species levels and radioresistance in cancer stem cells. Nature. 2009 Apr 9;458(7239):780-3.
- 15. Aktipis CA, Maley CC, Pepper JW. Dispersal evolution in neoplasms: the role of disregulated metabolism in the evolution of cell motility. Cancer Prev Res (Phila). 2012 Feb;5(2):266-75.
- 16. Wilenski U, Rand W. An introduction to agent-based modeling: Modeling natural, social and engineered complex systems with NetLogo. Cambridge, MA: MIT Press; 2009.
- 17. Fox CJ, Hammerman PS, Thompson CB. Fuel feeds function: energy metabolism and the T-cell response. Nat Rev Immunol. 2005 Nov;5(11):84452.
- 18. Buttgereit F, Brand MD. A hierarchy of ATP-consuming processes in mammalian cells. Biochem J. 1995 Nov 15;312 ( Pt 1):163-7.
- 19. Buttgereit F, Burmester GR, Brand MD. Bioenergetics of immune functions: fundamental and therapeutic aspects. Immunol Today. 2000 Apr;21(4):192-9.
- 20. Maciver NJ, Jacobs SR, Wieman HL, Wofford JA, Coloff JL, Rathmell JC. Glucose metabolism in lymphocytes is a regulated process with significant effects on immune cell function and survival. J Leukoc Biol. 2008 Oct;84(4):949-57.
- 21. Di Tomaso T, Mazzoleni S, Wang E, Sovena G, Clavenna D, Franzin A, Mortini P, Ferrone S, Doglioni C, Marincola FM, Galli R, Parmiani G, Maccalli C. Immunobiological characterization of cancer stem cells isolated from glioblastoma patients. Clin Cancer Res. 2010 Feb 1;16(3):800-13.
- 22. Chikamatsu K, Takahashi G, Sakakura K, Ferrone S, Masuyama K. Immunoregulatory properties of CD44+ cancer stem-like cells in squamous cell carcinoma of the head and neck. Head Neck. 2011 Feb;33(2):208-15.
- 23. Inoda S, Hirohashi Y, Torigoe T, Morita R, Takahashi A, Asanuma H, Nakatsugawa M, Nishizawa S, Tamura Y, Tsuruma T, Terui T, Kondo T, Ishitani K, Hasegawa T, Hirata K, Sato N. Cytotoxic T lymphocytes efficiently recognize human colon cancer stem-like cells. Am J Pathol. 2011 Apr;178(4):1805-13.
- 24. Pienta KJ, McGregor N, Axelrod R, Axelrod DE. Ecological therapy for cancer: defining tumors using an ecosystem paradigm suggests new opportunities for novel cancer treatments. Transl Oncol. 2008 Dec;1(4):15864.
- 25. Breivik J. Don't stop for repairs in a war zone: Darwinian evolution unites genes and environment in cancer development. Proc Natl Acad Sci U S A. 2001 May 8;98(10):5379-81.
- 26. Greaves M, Maley CC. Clonal evolution in cancer. Nature. 2012 Jan 18;481(7381):306-13.
- 27. Nowell PC. The clonal evolution of tumor cell populations. Science. 1976 Oct 1;194(4260):23-8.
- 28. Calabrese P, Tavaré S, Shibata D. Pretumor progression: clonal evolution of human stem cell populations. Am J Pathol. 2004 Apr;164(4):1337-46.
- 29. Greaves M. Cancer stem cells: back to Darwin?. Semin Cancer Biol. 2010 Apr;20(2):65-70.
- 30. Kareva I. What can ecology teach us about cancer?. Transl Oncol. 2011 Oct;4(5):266-70.
- 31. Merlo LM, Pepper JW, Reid BJ, Maley CC. Cancer as an evolutionary and ecological process. Nat Rev Cancer. 2006 Dec;6(12):924-35.
- 32. Tomlinson IP. Game-theory models of interactions between tumour cells. Eur J Cancer. 1997 Aug;33(9):1495-500.
- 33. Basanta D, Deutsch A. A game theoretical perspective on the somatic evolution of cancer. In: Bellomo N, editor. Selected topics on cancer modeling: genesis, evolution, immune competition, therapy. Boston: Birkhauser; 2008.
- 34. Basanta D, Simon M, Hatzikirou H, Deutsch A. Evolutionary game theory elucidates the role of glycolysis in glioma progression and invasion. Cell Prolif. 2008 Dec;41(6):980-7.
- 35. Mansury Y, Diggory M, Deisboeck TS. Evolutionary game theory in an agent-based brain tumor model: exploring the 'Genotype-Phenotype' link. J Theor Biol. 2006 Jan 7;238(1):146-56.
- 36. Thomas VL, Joel BS. Evolutionary game theory, natural selection, and Darwinian dynamics. Cambridge: Cambridge University Press; 2005.
- 37. Gatenby RA, Vincent TL. An evolutionary model of carcinogenesis. Cancer Res. 2003 Oct 1;63(19):6212-20.
- 38. Gatenby RA, Maini PK. Mathematical oncology: cancer summed up. Nature. 2003 Jan 23;421(6921):321.
- 39. Kareva I. Prisoner's dilemma in cancer metabolism. PLoS One. 2011;6(12):e28576.
- 40. Enderling H, Hlatky L, Hahnfeldt P. Immunoediting: evidence of the multifaceted role of the immune system in self-metastatic tumor growth. Theor Biol Med Model. 2012 Jul 28;9:31.
- 41. Hillen T, Enderling H, Hahnfeldt P. The tumor growth paradox and immune system-mediated selection for cancer stem cells. Bull Math Biol. 2013 Jan;75(1):161-84.
- 42. Moncharmont C, Levy A, Gilormini M, Bertrand G, Chargari C, Alphonse G, Ardail D, Rodriguez-Lafrasse C, Magné N. Targeting a cornerstone of radiation resistance: cancer stem cell. Cancer Lett. 2012 Sep 28;322(2):13947.
- 43. McCord AM, Jamal M, Williams ES, Camphausen K, Tofilon PJ. CD133+ glioblastoma stem-like cells are radiosensitive with a defective DNA damage response compared with established cell lines. Clin Cancer Res. 2009 Aug 15;15(16):5145-53.
- 44. Marcato P, Dean CA, Giacomantonio CA, Lee PW. If cancer stem cells are resistant to current therapies, what's next?. Future Oncol. 2009 Aug;5(6):74750.
- 45. Rich JN. Cancer stem cells in radiation resistance. Cancer Res. 2007 Oct 1;67(19):8980-4.
- 46. Schatton T, Frank MH. Antitumor immunity and cancer stem cells. Ann N Y Acad Sci. 2009 Sep;1176:154-69.
- 47. Hirohashi Y, Torigoe T, Inoda S, Takahashi A, Morita R, Nishizawa S, Tamura Y, Suzuki H, Toyota M, Sato N. Immune response against tumor antigens expressed on human cancer stem-like cells/tumor-initiating cells. Immunotherapy. 2010 Mar;2(2):201-11.
- 48. Todaro M, D'Asaro M, Caccamo N, Iovino F, Francipane MG, Meraviglia S, Orlando V, La Mendola C, Gulotta G, Salerno A, Dieli F, Stassi G. Efficient killing of human colon cancer stem cells by gammadelta T lymphocytes. J Immunol. 2009 Jun 1;182(11):7287-96.
- 49. Heddleston JM, Li Z, McLendon RE, Hjelmeland AB, Rich JN. The hypoxic microenvironment maintains glioblastoma stem cells and promotes reprogramming towards a cancer stem cell phenotype. Cell Cycle. 2009 Oct 15;8(20):3274-84.
- 50. Lin Q, Yun Z. Impact of the hypoxic tumor microenvironment on the regulation of cancer stem cell characteristics. Cancer Biol Ther. 2010 Jun 15;9(12):949-56.
- 51. Doloff JC, Waxman DJ. VEGF receptor inhibitors block the ability of metronomically dosed cyclophosphamide to activate innate immunity-induced tumor regression. Cancer Res. 2012 Mar 1;72(5):1103-15.
| null |
[
"Irina Kareva"
] |
2014-08-04T18:06:43+00:00
|
2014-08-04T18:06:43+00:00
|
[
"q-bio.TO",
"q-bio.PE"
] |
Immune Evasion through Competitive Inhibition: the Shielding Effect of non-Stem Cancer Cells
|
It has been recently proposed that the two "emerging" hallmarks of cancer,
namely altered glucose metabolism and immune evasion, may in fact be
fundamentally linked (Kareva and Hahnfeldt, 2013). This connection comes from
up-regulation of glycolysis by tumor cells, which can lead to active
competition for resources in the tumor microenvironment between tumor and
immune cells. Here it is further proposed that cancer stem cells (CSCs) can
circumvent the anti-tumor immune response by creating a 'protective shield' of
non-stem cancer cells around them. This shield can protect the CSCs both by
creating a physical barrier between them and cytotoxic lymphocytes (CTLs), and
by promoting competition for the common resources, such as glucose, between
non-stem cancer cells and CTLs. The implications of this hypothesis are
investigate using an agent-based model, leading to a prediction that relative
CSC to non-CSC ratio will vary with the strength of the host immune response,
with the highest occurring at an intermediate state of immune activation. A
discussion of possible therapeutic approaches concludes the paper, suggesting
that a chemotherapeutic regimen consisting of regular pulsed doses, i.e.,
metronomic chemotherapy, would yield the best clinical outcome by allowing CTLs
to most effectively reach and eliminate CSCs.
|
1408.0758v1
|
## 55 Tesla coercive magnetic field in frustrated Sr NiIrO 3 6
John Singleton , Jae Wook Kim , Craig V. Topping 1 1 1 2 3 , , , Anders
Hansen , Eun-Deok Mun 1 1 4 , , Saman Ghannadzadeh , Paul Goddard , 3 5
Xuan Luo 6 7 , , Yoon Seok Oh , Sang-Wook Cheong , Vivien S. Zapf 6 6 1
1 National High Magnetic Field Laboratory (NHMFL), MS E536,
Los Alamos National Laboratory, Los Alamos, NM 87545, USA
2 Department of Chemistry, University of Edinburgh,
Edinburgh, Midlothian EH8 9YL, United Kingdom
3 University of Oxford, Department of Physics,
The Clarendon Laboratory, Parks Road, Oxford, OX1 3PU, United Kingdom
4 Department of Physics, Simon Fraser University, Burnaby, BC, V5A 1S6, Canada
5 Department of Physics, University of Warwick,
Gibbet Hill Road, Coventry, CV4 7AL, United Kingdom
6 RCEM & Dept. of Physics and Astronomy,
Rutgers University, Piscataway, NJ 08854, USA and
7 POSTECH, Pohang University of Science and Technology,
San 31 Hyoja-dong, Nam-gu, Pohang-si,
Gyungbuk, 790-784, Republic of Korea
## Abstract
We have measured extremely large coercive magnetic fields of up to 55 T in Sr NiIrO , with 3 6 a switched magnetic moment ≈ 0 8 . µ B per formula unit. As far as we are aware, this is the largest coercive field observed thus far. This extraordinarily hard magnetism has a completely different origin from that found in conventional ferromagnets. Instead, it is due to the evolution of a frustrated antiferromagnetic state in the presence of strong magnetocrystalline anisotropy due to the overlap of spatially-extended Ir 4+ 5 d orbitals with oxygen 2 p and Ni 2+ 3 d orbitals. This work highlights the unusual physics that can result from combining the extended 5 d orbitals in Ir 4+ with the frustrated behaviour of triangular lattice antiferromagnets.
PACS numbers: 71.70.Ej, 75.30.Gw, 75.50.Vv
## I. INTRODUCTION
Oxides containing iridium in the Ir 4+ (5 d 5 ) state have received much recent attention because the energy scales for spin-orbit interactions (SOIs), Coulomb repulsion, and crystallineelectric fields are all very similar [1-11]. This unusual situation, in comparison to analogous 3 d systems, results from a decrease in the strength of correlation effects and an increase in SOIs as one descends the periodic table. Usually, the Coulomb repulsion and SOIs are responsible for Hund's rules that determine the groundstates of magnetic ions; however, in these 5 d (and some 4 d ) systems, the competition between the three similarly-sized energy scales can result in exotic magnetic states [1-11], leading to proposals for the use of Ir 4+ -based systems in quantum information processing and other novel applications (see Refs. [2, 9] and references therein). In this paper, we report another manifestation of unusual 5 d physics: extremely large coercive magnetic fields of up to 55 T in Sr NiIrO , with 3 6 a switched magnetic moment of about 0 8 . µ B per formula unit. As far as we are aware, this is the largest coercive field measured thus far, and a factor ∼ 5 times those reported recently in novel, ultra-hard ferromagnets [12]. This exemplary hard magnetism is, however, very different from that found in conventional ferromagnets in that it results from an unusual, frustrated, antiferromagnetic ground state that incorporates a relatively large magnetocrystalline anisotropy.
2+ 4+
The salient structural details [13] are shown in Fig. 1. The Ni and Ir magnetic ions in Sr NiIrO 3 6 occupy oxygen cages that alternate in chains parallel to the c -axis (Fig. 1(a)). These chains are in turn arranged in a hexagonal pattern in the ab -plane (Fig. 1(b)) [13]. The Ni 2+ is surrounded by a trigonal bipyramid of oxygen atoms, while the Ir 4+ ion sits in an octahedral oxygen cage. Magnetic frustration is intrinsic to this structure, and can result from antiferromagnetic interactions within the triangular lattice in the ab -plane, and from frustration between nearest-neighbour and next-nearest-neighbour interactions along the c -axis chains. Electronic structure calculations have suggested both possibilities for Sr NiIrO 3 6 [14-16]; the same calculations indicate that the overlap of spatially-extended Ir 4+ 5 d orbitals with oxygen 2 p and Ni 2+ 3 d orbitals leads to a magnetocrystalline anisotropy energy ≈ 13 5 meV. Schematic energy level diagrams for Ir . 4+ and Ni 2+ are shown in Fig. 1(c); the effective spin of Ni 2+ is S = 1 and that of Ir 4+ is S = 1 2 [14-16].
FIG. 1. Structure and energy levels. The crystal structure of Sr NiIrO 3 6 as viewed from the (a) [110] and (b) [001] directions. The Ni 2+ and Ir 4+ ions are within oxygen trigonal bipyramids (blue) and oxygen octahedra (orange), respectively. For clarity, Sr ions are not shown in (a). (c) Schematic level diagrams for Ir 4+ and Ni 2+ (based on Refs [14-16]). Unperturbed Ir 4+ levels are shown in blue; a combination of antiferromagnetic interactions and spin-orbit coupling splits the e ′ g doublet to give the configuration shown in red. Electron spins are shown as arrows: the effective spin of Ni 2+ is S = 1 and that of Ir 4+ is S = 1 2 in the opposite direction.
## II. RESULTS
## A. Low-field magnetization and dynamics
Our study involves 10 different samples of Sr NiIrO 3 6 from several different batches encompassing minor variations in growth conditions. The results for all samples are rather consistent; typical data from four single crystals (labelled S1 to S4) and three polycrystalline samples (P1, P2, P3) are shown in the figures that follow. Note that sample S2 is a very small crystal with exceptionally clean hexagonal faces containing virtually no visible defects; as will be seen below, this apparent cleanliness is of relevance to the timescales over which the remanent magnetization persists.
Typical examples of low-magnetic-field DC magnetization data for Sr NiIrO 3 6 (samples S1, S2 and P1) are shown in Fig. 2. Magnetic order sets in below 85 K, being marked by a broad feature in magnetization ( M ) vs. temperature ( T ) data (Fig. 2(a)), and additional peaks in elastic neutron scattering [17-19]. The neutron diffraction data are consistent with either
FIG. 2. Frustration-induced slow dynamics. (a) DC magnetic susceptibility ( χ = M/H ) as a function of temperature T for Sr 3 NiIrO 6 single-crystal samples S1 and S2 and polycrystalline sample P1. The onset of magnetic order is marked by a broad hump superimposed on the background slope and centered on T ≈ 85 K. (b) T -dependent χ data for Sr NiIrO 3 6 single crystal sample S1 measured DC (black) and using an AC susceptometer operated at frequencies of 10 Hz (red) and 10 kHz (blue). The sharp fall in χ shifts to higher T as the frequency increases. The contour plots show d M/ H d versus µ H 0 and T for polycrystalline sample P2 measured (c) in a pulsed magnetic field with a sweep rate ≈ 3500 T/s and (d) in a SQUID (DC) magnetometer with a magnetic field sweep rate of 0.008 T/s. Corresponding plots of the hysteresis between falling and rising magnetic field sweeps, defined as hysteresis = [ M H,T ( ) falling -M H,T ( ) rising ], are shown in (e) (pulsed field) and (f) (SQUID). Note that as the field sweep rate increases, features in d M/ H d and regions of pronounced hysteresis are pushed to higher T .
a partially-disordered antiferromagnet [20] or an amplitude-modulated antiferromagnetic state at low T . Despite the onset of long-range order at 85 K, the heat capacity is nearly featureless between 4 and 250 K [18].
Fig. 2(b) shows the DC magnetic susceptibility of Sr NiIrO 3 6 single crystal sample S1, in an applied field of µ H 0 = 0 2 T, and the AC susceptibility for an oscillating field amplitude . of 10 -3 T at 10 Hz and 10 kHz. In all three data sets, the sharp drop in M already seen in
Fig. 2(a) is observed as T is lowered. However, the temperature at which the drop occurs is frequency dependent; the half-height point occurs at 28 K for 10 kHz, 20 K for 10 Hz, and 13 K for DC measurements. Further evidence of slow dynamics over a similar temperature range is seen in the contour plot in Figs. 2 (c) to (f). Here we show d M H,T / H ( ) d on the initial field upsweep following zero-field cooling for polycrystalline sample P2. These data were measured in a 7 T SQUID magnetometer with a field-sweep rate ≈ 0 008 T/s (Fig. 2(d)) . and in a pulsed magnetic field with a sweep rate ≈ 3500 T/s (Fig. 2(c)), where the pulse shape is shown in Fig. 6(a). Corresponding plots of the hysteresis in M between rising and falling magnetic field sweeps are shown in (e) (pulsed field) and (f) (SQUID). As the field sweep rate increases (and therefore the timescale of the measurement decreases), features in d M/ H d and regions of pronounced hysteresis are pushed to higher T . Thus, this dependence of M on sweep rate is analogous to the frequency dependence seen in the AC susceptibility data (Fig. 2(b)), where decreasing the measurement timescale forces features in M to higher T . Similar phenomena have been seen in previous experiments within the ordered phase; as T was lowered below 30 K, a frequency-dependent crossover occurred to hysteretic magnetic behaviour as a function of T and H , as well as a strong frequency dependence of the AC susceptibility [17-19]. A number of other frustrated antiferromagnets isostructural with Sr NiIrO 3 6 show qualitatively similar frequency-dependent magnetic behaviour [21-33].
## B. Large coercive fields
Large coercive fields are demonstrated in Fig. 3, which shows pulsed-field M H ( ) data for four representative samples of Sr NiIrO 3 6 (the field-pulse profile is shown in Fig. 6(a)). Data for Figs. 3(a), (b), and (d) are calibrated against absolute measurements on the same samples in quasistatic magnetic fields of up to 13 T.
Though the behaviour of all samples is similar, showing that the large coercive fields are intrinsic to Sr 3 NiIrO , there are some detail differences from sample to sample. 6 In Fig. 3(a) (single crystal S3), the magnetic hysteresis for H ‖ c is superimposed on a linear M H ( ) background, and shows a large µ H 0 c = 34 1 T and a remanent moment of 0 45 . . µ B per formula unit at zero field. The magnetization jump at 34.1 T is ≈ 0 8 . µ B per formula unit. The data in Fig. 3(e) (Sample S4) show an even larger µ H 0 c = 42 7 T. As mentioned above, . sample S2 is a very small single crystal with exceptionally clean faces, and Fig. 3(c)) shows
FIG. 3. Hysteresis loops and large coercive fields. Magnetization M as a function of magnetic field µ H 0 measured in a series of pulses using a capacitor-bank-driven 65 T pulsed magnet. Sample numbers, field directions and measurement temperatures T are given in each section of the figure; the vertical jumps in M occur at the coercive field, µ H 0 c . In (b), (f) no hysteresis is observed in the H ⊥ c configurations. In (a), (b), and (d), absolute magnetization values, measured in commercial SQUID and vibrating-sample magnetometers, were used to calibrate the pulsed field data.
that it exhibits qualitatively different behaviour: the coercive field is very large (38.4 T), as in other samples, but M relaxes during the millisecond timescales (see Fig. 6(a)) of the downsweep of the magnetic field. All subsequent field pulses after the first one show jumps of similar height, consistent with relaxation of M back to zero between field pulses. On the other hand, in Fig. 3(a), (b), (d), (e), and (f) the first upsweeps following a zero-field cool result in an M jump about half the height of subsequent jumps, showing that the latter represent a complete reversal of M . For those samples, two consecutive magnetic fields sweeps in the same direction produce no jump (not shown). In polycrystalline sample P2 (Fig. 3(d)), the M H ( ) data also show some curvature at low fields, but M does not relax back to zero between field pulses. (As will be shown in Fig. 4(a) below, polycrystalline sample P3 shows similar behaviour, but superimposed on a much less curved background.) In the initial up-sweep, M jumps at a record-high magnetic coercive field of µ H 0 c = 55 T after zero-field cooling; in subsequent pulses, the M jump doubles in height, but is observed at lower fields (40 T). Note that all samples exhibit a higher H c value in the initial field sweep after zero-field cooling, compared to those seen in subsequent field sweeps; we shall return to this behaviour below. Finally, we remark that the jumps in M are sharp, occurring over a small range of field ( < 1 T at 4 K). In particular, the transition width in Fig. 3(c) is glyph[lessorsimilar] 0 03 T. .
There is no systematic dependence of the magnetic hysteresis on details of the growth method in the samples shown in Fig. 3 or in the three other samples measured, suggesting that all samples have the same Ni/Ir ratio. There does, however, appear to be a pinning effect that stabilizes M in most of the samples; in such cases, the remanent magnetization was found to persist unchanged for at least 24 hours at T = 4 K. By contrast, sample S2 in Fig. 3(c) shows magnetic relaxation on millisecond timescales ( i.e. similar timescales to the downsweep of the pulsed field; see Fig. 6(a)). In addition, none of the samples studied appear to show signs of saturation of the magnetization even in fields of 65 T. The anisotropy of the hysteresis is illustrated in Figs. 3(b) and (f); here M H ( ) of single crystals S3 and S4 is shown for H ⊥ c . For this field direction, no jump in M or hysteresis is observed. In both single crystals, M is comparable in size at 65 T for H ⊥ c and H ‖ c .
FIG. 4. Sweep-rate dependence of the coercive field. (a) M H ( ) data for polycrystalline sample P3 measured in a capacitor bank-driven pulsed magnet. Black dotted lines denote the initial pulse after zero-field cooling whilst the green solid line indicates data for subsequent pulses at 4 K. (b) Sweep-rate dependence of coercive magnetic field ( µ H 0 c ). (c) History dependence of coercive magnetic field ( µ H 0 c ) under various sweep rates (plotted as a function of pulse number). For sweep rates larger than 360 T/s, a capacitor-bank-driven pulsed magnet was used; other data were taken using the generator-driven 60 T Long-Pulse Magnet (see Fig. 6 for magnetic field profiles). Twice during this experiment, the sample was warmed to room temperature (denoted by red vertical dashed lines).
## C. Dynamics, and history- and temperature dependences of the coercive field
In view of the field-sweep-rate dependence of the low-field magnetization noted earlier (Fig. 2(b)-(f)), it is interesting to see if the hysteresis loops are affected by similar dynamical issues, and so Fig. 4 features M H ( ) measurements of polycrystalline sample P3 in the Los Alamos generator-driven 60 T Long-Pulse Magnet, which provides sweep rates between 25 and 360 T/s, and in a capacitor-driven 65 T magnet with faster sweep rates (pulse shapes are shown in Fig. 6). Fig. 4(a) once again demonstrates that the coercive field is higher on the initial field sweep after zero-field cooling (c.f. Figs. 3(c) and (d)), whilst (b) shows that this trend also occurs at lower field-sweep rates. The history dependence of the coercive field throughout a sequence of field pulses of varying sweep rate and a couple of thermal cycles to room temperature and back is illustrated in Fig. 4(c). All of these data show that the magnetic hysteresis loop is robust at least down to 25 T/s and illustrate that on magnet pulses subsequent to the initial pulse following zero-field cooling, H c is slightly smaller in lower field sweep rates; this was observed in all measured samples.
The T -dependences of M H ( ) for H ‖ c and H ⊥ c are shown in Figs. 3(a) and (b) for single crystal S3. For H ⊥ c , M for T < 25 K is suppressed compared to the value at 25 K, consistent with antiferromagnetic exchange interactions [36]. For H ‖ c , hysteresis is observed at 10 K but has disappeared by 25 K. The T -dependence of H c is given in more detail in Fig. 5(a), which shows d M/ H d data for H ‖ c (sample S2); peaks occur at the coercive field, H c . Fig. 5(b) shows that H c decreases linearly with increasing T (data from samples S2 and S3), extrapolating to zero at ≈ 25 K. This is close to the onset of hysteretic behaviour and the frequency-dependent fall in M (see Fig. 2(b-f)), linking the observation of finite coercive field with the low-temperature phase of the antiferromagnetic groundstate. The almost identical variations of H c with temperature in the two samples show that the value of this field and its temperature dependence are intrinsic to Sr NiIrO 3 6 single crystals, whereas the persistence (or not) of the remanent magnetization depends on extrinsic factors such as defects or disorder (the same samples are used in Figs. 3(a),(c)).
Finally, we note that a coercive field µ H 0 c ≈ 22 T was measured using polycrystalline samples of Sr NiIrO 3 6 in Ref. [17]. All of our polycrystalline samples exhibit low-temperature coercive fields µ H 0 c ≈ 40 T, very similar to those seen in the single crystals (see Figs. 3 and 4). Therefore, the difference between polycrystalline and single-crystal samples cannot
dM
0.5
0.0
-40
-30
-20
-10
0
10
20
30
40
Magnetic field (T)
0 5 10 15 20 25 30 -40 -30 -20 -10 0 10 20 30 40 SNIO-I SNIO-N H || c Coercive magnetic field (T) (b) S2 S3 FIG. 5. Delineating the region of large coercive field. (a) d M/ H d for single-crystal sample S2 at several T for H ‖ c (c.f. Fig. 3(a,c)). The peaks denote the sharp change in M at the coercive field; note that multiplication factors of 5 and 50 are applied for 6, 8, 10 K and 15, 19 K data, respectively as the peak broadens with increasing T . (b) T -dependence of the coercive field for single crystals S2 and S3. Filled symbols indicate the result taken in the initial pulse after zero-field cooling from room T ; these show slightly larger H c values compared those measured in subsequent pulses (see Fig. 4). Lines are linear fits to the data.
Temperature (K)
account for the much lower values of H c observed in Ref. [17] compared to those in this work.
## III. DISCUSSION
## A. Summary of results
Combining the H -and T -dependent magnetization studies of Sr NiIrO , 3 6 a picture emerges of an antiferromagnetic order [19] for T glyph[lessorsimilar] 30 K and H = 0 that evolves in applied magnetic fields. This evolution is manifested most markedly in the sharp jump in magnetization that occurs on rising fields. The field position µ H 0 c of the jump shows a similar history and field-sweep-rate dependence for all samples (Figs. 3, 4, 5), with slightly smaller values of H c occurring at lower sweep rates. This strongly suggests that the magnetization jump at H c is an intrinsic property of Sr NiIrO , with the sweep-rate dependence 3 6 being caused by sluggish kinetics associated with the magnetic frustration intrinsic to this
structural family [11]; the same frustration is responsible for the slow magnetic relaxation (Figs. 2(c-f)) and the strong variation (13 -28 K) of the fall in M with frequency (Fig. 2(b)). By contrast, in systems such as (Sm,Sr)MnO 3 [34], where the magnetization jumps are not intrinsic, but associated with quenched disorder, the smaller the field-sweep rate, the larger the field needed to realize the transition, the opposite of what we observe in Sr NiIrO 3 6 (Fig. 4).
It is likely that the millisecond-timescale magnetic relaxation observed in the hysteresis loops for sample S2 (Fig. 3(c)) is also intrinsic behaviour due to frustration, whilst samples S3, S4, P2 (Figs. 3(a), (d), (e)) and P3 (Fig. 4(a)) and all other samples studied exhibit pinning or freezing of the magnetization or an order-by-disorder mechanism, as seen at lower fields in isostructural Ca CoMnO 3 6 [35]. In the latter mechanism, magnetic-site disorder disrupts longer-range interactions, leading to a reduction of frustration [35].
## B. Comparison with ferromagnets and other systems
Magnets with high coercivity, i.e. , 'hard' magnets, are important for a wide range of applications involving magnetic actuation and induction, such as loud speakers, wind turbines and electric motors [36]. In rare-earth iron boride magnets, among the hardest commercial magnets, µ H 0 c can be as large as 1.5 T at room temperature [37]. At cryogenic temperatures, magnetic hysteresis effects can extend to higher fields, e.g. , up to 10 T in the colossal magnetoresistance manganites, and in Li (Li 2 1 -x Fe )N, Gd Ge , Ga-doped CeFe , LuFe O , x 5 4 2 2 4 and Fe 1 4 / TaSe 2 [12, 28, 38-42]. Large coercive fields in such magnets are typically caused by magnetocrystalline anisotropy due to SOI [37], whilst magnetic hysteresis results from a combination of domain dynamics and anisotropy [36]. In traditional ferromagnets, domains result from competition between the short-range exchange interactions that prefer parallel spin alignment, and the free-energy penalty of maintaining a magnetic field in an extended region of space around the sample [36]. The effect of this competition is that the energy scale for switching magnetic domains can be orders of magnitude smaller than those of the nearest-neighbor ferromagnetic exchange interactions. Generally, the microscopic order in a traditional ferromagnet does not change significantly around the hysteresis loop as the domains change direction and/or the domain walls move [36].
By contrast, the conventional phenomenology of ferromagnetic domains is not involved
in the notable H c values observed for Sr NiIrO 3 3 (35 -55 T, this work) and in the lower, but nevertheless impressive, values measured in the isostructural family of frustrated triangularlattice antiferromagnets, Ca Co O 3 2 6 (7 T) [28], and Ca CoMnO 3 6 (10 T) [33, 43], Sr CoIrO 3 6 ( ≈ 20 T) [18], and Ca CoRhO 3 6 ( ≈ 30 T) [23]. Sr NiIrO 3 6 is initially antiferromagnetic after zero-field cooling and its magnetic groundstate must evolve to produce a net magnetization [19]. The energy scale for the coercive field ( µ H 0 c = 34 -55 T) is roughly of the same order of magnitude as the temperature of the onset of magnetic hysteresis (13 -28 K), and longrange order (85 K), consistent with evolution of the magnetic order driven by large magnetic fields. Indeed, in compounds isostructural with Sr NiIrO , the microscopic order has been 3 6 shown to change around the hysteresis loop using elastic neutron diffraction measurements [43, 44]. We can therefore infer that the magnetic hysteresis in Sr NiIrO 3 6 can be attributed to the evolution of a microscopic frustrated order with magnetic field, rather than domain effects ( c.f. Ref. [45]).
As mentioned previously, electronic structure calculations suggest that the magnetocrystalline anisotropy in Sr NiIrO 3 6 results from the configuration of overlapping orbitals in Ir 4+ -O-Ni 2+ chain [14-16]. This picture is supported by the prevalence of relatively large coercive magnetic fields in the isostructural family members that have 4 d or 5 d ions (with relatively extended orbitals) on the octahedral site and 3 d ions on the bipyramidal site: Ca CoRhO , Sr NiIrO 3 6 3 6 and Sr CoIrO 3 6 [17, 18, 23]. The exceptional coercive magnetic field in the title compound serves as another example of notable physical effects resulting from the unusual 5 d orbital physics of Ir 4+ .
## IV. CONCLUSION
In conclusion, we observed coercive magnetic fields of 34 -55 T for H ‖ c in flux-grown single crystals and polycrystalline Sr NiIrO , to our knowledge, a record high coercive magnetic 3 6 field for any material. Sr NiIrO 3 6 shows signatures of a type of frustrated antiferromagnetic order in zero field [19], and the hysteresis loop is likely due to evolution of this microscopic order. The high coercive magnetic field is consistent with the large magnetocrystalline anisotropy predicted in ab-initio calculations for this compound, due to the Ir 4+ 5 d orbitals overlapping via intermediate oxygen with Ni 2+ 3 d orbitals.
## V. METHODS
## A. Sample growth
Polycrystalline Sr NiIrO 3 6 was prepared through solid-state reaction at 1300 C. Single o crystals of Sr 3 NiIrO 6 were grown from either a stoichiometric or 20 % Ni-rich or 20 % Ir-rich (in molar ratio) composition using K CO 2 3 as flux. The single crystals are hexagonal plates with typical dimensions 2 × 2 × 0 5 mm . . 3
## B. Magnetization measurements
Quasistatic-field magnetization ( M H ( )) data were measured in a vibrating sample magnetometer in a superconducting magnet (PPMS-14, Quantum Design), or in a SQUID (MPMS7, Quantum Design). AC susceptibility data were measured in a 7 T AC SQUID and an ac susceptometer in a 14 T PPMS (Quantum Design).
The pulsed-field magnetization experiments used a 1.5 mm bore, 1.5 mm long, 1500-turn compensated-coil susceptometer, constructed from 50 gauge high-purity copper wire [46]. When a sample is within the coil, the signal is V ∝ (d M/ t d ), where t is the time. Numerical integration is used to evaluate M . Samples were mounted within a 1.3 mm diameter ampoule that can be moved in and out of the coil. Accurate values of M are obtained by subtracting empty-coil data from that measured under identical conditions with the sample present. The susceptometer is calibrated by scaling low-field M values to match those recorded with a sample of known mass measured in a commercial SQUID or vibrating-sample magnetometer.
Fields were provided by a 65 T short-pulse magnet energized by a 4 MJ capacitor bank, or the generator-driven 60 T Long-Pulse Magnet at NHMFL Los Alamos; the field versus time profiles for these two magnets are shown in Fig. 6. The susceptometer was placed within a 3 He cryostat providing T s down to 0.4 K. µ H 0 was measured by integrating the voltage induced in a ten-turn coil calibrated by observing the de Haas-van Alphen oscillations of the belly orbits of the copper coils of the susceptometer [46].
In measuring hysteresis loops, the initial field sweep (up and down) is performed after zero-field cooling from room temperature. Subsequent pulses are delivered after 40 to 60 minutes (the cooling time of the magnet in question) while maintaining constant sample temperature.
70
FIG. 6. Time-dependence of magnetic fields for (a) a capacitor-bank-driven 65 T short-pulse magnet and (b) three examples of the controlled sweep patterns possible with the generator-driven 60 T Long Pulse Magnet. (a) µ 0 d H/ t d for the short-pulse magnet is shown in blue (right axis). In (b), the three stages in each of the field-sweep patterns are due to three separate coils that are energized in sequence by the generator.
## VI. REFERENCES
- [1] Kim, B.J., et al. . Phase-Sensitive Observation of a Spin-Orbital Mott State in Sr 2 IrO . 4 Science 323 , 3129 (2009).
- [2] Lovesy, S.W., Khalyavin, D.D., Manuel. P., Chapon, L.C. & Qi, T.F.. Magnetic symmetries in neutron and resonant x-ray Bragg diffraction of four iridium oxides, J. Phys.: Condens. Matter 24 , 496003 (2012).
- [3] Chapon, L.C., and Lovesy, S.W.. The magnetic motif and the wavefuntion of Kramers ions in strontium irridate. J. Phys.: Condens. Matter 23 , 252201 (2011)
- [4] Shitade, Atsuo et al. . Quantum Spin Hall Effect in a Transition Metal Oxide Na IrO . 2 3 Phys.
Rev. Lett. 102 , 256403 (2009)
- [5] Laguna-Marco, M. A. et al. . Orbital Magnetism and Spin-Orbit Effects in the Electronic Structure of BaIrO . 3 Phys. Rev. Lett. 105 , 216407 (2010)
- [6] Chikara, S., Korneta, O. Crummett, W. P., DeLong, L. E., Schlottmann, P., & Cao, G.. Giant magnetoelectric effect in the J eft = 1 2 Mott insulator Sr IrO . / 2 4 Phys. Rev. B 80 , 140407 (2009)
- [7] Chaloupka, J., Jackeli, George & Khaliullin, Giniyat. Kitaev-Heisenberg Model on a Honeycomb Lattice: Possible Exotic Phases in Iridium Oxides A IrO . 2 3 Phys. Rev. Lett. 105 , 027204 (2010)
- [8] Witczak-Krempa, William and Kim, Yong Baek. Topological and magnetic phases of interacting electrons in the pyrochlore iridates. Phys. Rev. B 85 , 045124 (2012)
- [9] Choi, S.K., et al. Spin Waves and Revised Crystal Structure of Honeycomb Iridate Na IrO . 2 3 Phys. Rev. Lett. 108 , 127204 (2012)
- [10] Kim, B. J., et al. . Novel J eff = 1 2 Mott state induced by relativistic spin-orbit coupling in / Sr IrO , 2 4 Phys. Rev. Lett. , 101 076402 (2008)
- [11] Kim, Heung-Sik, Kim, Choong H., Jeong, Hogyun, Jin, Hosub & Yu, Jaejun. Strain-induced topological insulator phase and effective magnetic interactions in Li IrO . 2 3 Phys. Rev. B 87 , 165117 (2013)
- [12] Jesche, A., et al. . Giant magnetic anisotropy and tunneling of the magnetization in Li (Li 2 1 -x Fe )N. x Nature Communications 5 , 3333 (2014)
- [13] Nguyen, T.N., & Zur Loye, H.C.. A family of one-dimensional oxides: Sr MIrO 3 6 (M = Ni, Cu, Zn): Structure and Magnetic Properties. J. Solid State Chem. 117 300 (1995).
- [14] Zhang, G.R., Zhang, X.L., Jia, T., Zeng, Z. & Lin, H.Q.. Intrachain antiferromagnetic interaction and Mott state induced by spin-orbit coupling in Sr NiIrO . 3 6 J. Appl. Phys. 107 09E120 (2010)
- [15] Sarkar, Soumyajit, Kanungo, Sudipta & Saha-Dasgupta, T.. Ab initio study of lowdimensional quantum spin systems Sr NiPtO 3 6 , Sr CuPtO 3 6 , and Sr NiIrO 3 6 Phys. Rev. B 82 , 235122 (2010)
- [16] Ou, Xuedong, & Wu, Hua. Impact of spin-orbit coupling on the magnetism of Sr 3 M IrO 6 ( M = Ni, Co). Preprint arXiv:1312.7411
- [17] Flahaut, D., et al. . A magnetic study of the one dimensional Sr NiIrO 3 6 compound Eur. Phys.
- J. B 35 , 317 (2003)
- [18] Mikhailova, D., et al. . Magnetic properties and crystal structure of Sr CoIrO 3 6 and Sr NiIrO . 3 6 Phys. Rev. B 86 134409 (2012)
- [19] Lefrancois, E., et al. Magnetic order in the frustrated Ising-like chain compound Sr NiIrO . 3 6 Preprint arXiv: 1403.4815v1
- [20] Mekata, M.. Antiferro-Ferrimagnetic Transition in Triangular Ising Lattice. J. Phys. Soc. Jpn. 42 , 76 (1977)
- [21] Kageyama, H., Yoshimura, K., & Kosuge, K.. Synthesis and Magnetic Properties of New Series of One-Dimensional Oxides Ca Co 3 1+ x B 1 -x O ( 6 B =Ir, Ru). J. Solid State Chem. 140 , 14 (1998).
- [22] Kawasaki, S., Takano, M., &. Inami, T.. Synthesis, Structure, and Magnetic Properties of Ca BMnO 3 6 (B = Ni, Zn) and Ca ZnCoO 3 6 Crystallizing in the K CdCl 4 6 Structure. J. Solid State Chem. 145 , 302 (1999).
- [23] Niitaka, S., Yoshimura, K., Kosuge, K., Nishi, M., & Kakurai, K.. High-Field Magnetization and Neutron Diffraction Studies of One-Dimensional Compound Ca CoRhO . 3 6 Phys. Rev. Lett. 87 , 177202 (2001).
- [24] Stitzer, K.E., Henley, W.H., Claridge, J.B., zur Loye, H.-C., & Layland, R.C.. Sr NiRhO 3 6 and Sr CuRhO : 3 6 Two New One-Dimensional Oxides. Magnetic behaviour as a Function of Structure: Commensurate vs Incommensurate. J. Solid State Chem. 164 , 220 (2002).
- [25] Rayaprol, S., Sengupta, K., & Sampathkumaran, E.V.. Magnetic frustration in the stoichiometric spin-chain compound Ca CoIrO . 3 6 Phys. Rev. B 67 , 180404 (2003).
- [26] Loewenhaupt, M., Schlafer, W., Niazi, A. & Sampathkumaran, E.V.. Evidence for the coexistence of low-dimensional magnetism and long-range order in Ca CoRhO . 3 6 Europhys. Lett. 63 , 374 (2003).
- [27] Hardy, V., et al. . Specific heat investigation of the magnetic ordering in two frustrated spinchain oxides: Ca Co O 3 2 6 and Ca CoRhO . 3 6 J. Phys.: Cond. Matter 15 , 5737 (2003).
- [28] Hardy, V., Flahaut, D., Lees, M.R., & Petrenko, O.A.. Magnetic quantum tunneling in Ca Co O 3 2 6 studied by ac susceptibility: Temperature and magnetic-field dependence of the spin-relaxation time. Phys. Rev. B 70 , 214439 (2004).
- [29] Mohapatra, N., Iyer, K.K., Rayaprol, S., & Sampathkumaran, E.V.. Geometrically frustrated magnetic behaviour of Sr NiRhO 3 6 and Sr NiPtO . 3 6 Phys. Rev. B 75 , 214422 (2007).
- [30] Choi, Y.J., Yi, H.T., Lee, S., Huang, Q., Kiryukhin, V., & Cheong, S.-W.. Ferroelectricity in an Ising Chain Magnet. Phys. Rev. Lett. 100 , 047601 (2008).
- [31] Agrestini, S., Mazzoli, C., Bombardi, A., & Lees, M.R.. Incommensurate magnetic ground state revealed by resonant x-ray scattering in the frustrated spin system Ca Co O . 3 2 6 Phys. Rev. B 77 , 140403 (2008).
- [32] Agrestini, S., et al. . Slow Magnetic Order-Order Transition in the Spin Chain Antiferromagnet Ca Co O . 3 2 6 Phys. Rev. Lett. 106 , 197204 (2011).
- [33] Kim, J.W., et al. . Multiferroicity with coexisting isotropic and anisotropic spins in Ca Co 3 2 -x Mn O . x 6 Phys. Rev. B 89 , 060404 (2014).
- [34] Fisher, L.M., et al. . Quenched-disorder-induced magnetization jumps in (Sm,Sr)MnO . 3 Phys. Rev. B 70 , 212411 (2004).
- [35] Kiryukhin, V., et al. . Order by Static Disorder in the Ising Chain Magnet Ca Co 3 2 -x Mn O . x 6 Phys. Rev. Lett. 102 , 187202 (2009)
- [36] Cullity B.D., & Graham, C.D.. Introduction to Magnetic Materials. 2nd edition (Wiley, New York, 2009)
- [37] Coey, J.M.D.. Advances in Magnetism - Hard Magnetic Materials: A Perspective. IEEE Trans. Magn. 47 , 4671 (2011).
- [38] Autret, C., et al. . Magnetization steps in a noncharge-ordered manganite, Pr 0 5 . Ba 0 5 . MnO . 3 Appl. Phys. Lett. 82 , 4746 (2003)
- [39] Morosan, E., et al. . Sharp switching of the magnetization in Fe 4 TaS . 2 Phys. Rev. B 75 , 104401 (2007)
- [40] Haldar, Arabinda, Suresh, K. G., & Nigam, A. K.. Magnetism in gallium-doped CeFe : 2 Martensitic scenario. Phys. Rev. B 78 144429 (2008)
- [41] Wu, Weida, et al. . Formation of Pancakelike Ising Domains and Giant Magnetic Coercivity in Ferrimagnetic LuFe . 2 Phys. Rev. Lett. 101 , 137203 (2008).
- [42] Ko, K.-T., et al. . RKKY Ferromagnetism with Ising-Like Spin States in Intercalated Fe TaS . 4 2 Phys. Rev. Lett. 107 . 247201 (2011).
- [43] Jo, Y.J., et al. . 3:1 magnetization plateau and suppression of ferroelectric polarization in an Ising chain multiferroic. Phys. Rev. B 79 , 012407 (2009)
- [44] Fleck, C.L., Lees, M.R., Agrestini, S., McIntyre, G.J., & Petrenko, O.A.. Field-driven magnetisation steps in Ca 3 Co O : A single-crystal neutron-diffraction study. 2 6 Europhys. Lett. 90 ,
67006 (2010).
- [45] Yanez-Vilar, S., et al. Multiferroic behavior in the double perovskite Lu MnCoO . 2 6 Phys. Rev. B 84 , 134427 (2011).
- [46] Goddard, P.A., et al. . Experimentally determining the exchange parameters of quasi-twodimensional Heisenberg magnets. New. J. Phys. 10 , 083025 (2008).
## VII. ACKNOWLEDGMENTS
This work is funded by the U.S. Department of Energy, Basic Energy Sciences program 'Science at 100 Tesla'. The NHMFL Pulsed Field Facility is funded by the US National Science Foundation through Cooperative Grant No. DMR-1157490, the State of Florida, and the US Department of Energy. Work in the UK is supported by the EPSRC. PAG and JS would like to thank the University of Oxford for the provision of visiting fellowships. The work at Rutgers was supported by the NSF under Grant number NSF-DMREF-1233349, and the work at Postech was supported by the Max Planck POSTECH/KOREA Research Initiative Program (Grant number 2011-031558) through NRF of Korea funded by MEST.
## VIII. AUTHOR CONTRIBUTIONS
J.S., J-W.K. and C.T. carried out the pulsed-field magnetization measurements, analyzed the data and prepared the figures in this paper. C.T., P.A.G. and J.S. constructed the pulsed-field magnetometer. E.D.M. assisted with pulsed-field experiments. The SQUID and AC and DC magnetometry data in this paper were measured by S.G., P.A.G. and A.H.. Samples were grown, characterized and prepared by X.L, Y.S.O. and S-W.C.. The text was written by J.S. and V.Z. with input from all authors. This study is part of a research program on multiferroics directed by V.Z..
## IX. ADDITIONAL INFORMATION
The authors declare no competing financial interests.
| null |
[
"John Singleton",
"Jae Wook Kim",
"Craig V. Topping",
"Anders Hansen",
"Eun-Deok Mun",
"Saman Ghannadzadeh",
"Paul Goddard",
"Xuan Luo",
"Yoon Seok Oh",
"Sang-Wook Cheong",
"Vivien S. Zapf"
] |
2014-08-04T18:21:01+00:00
|
2014-08-04T18:21:01+00:00
|
[
"cond-mat.mtrl-sci",
"cond-mat.str-el"
] |
55~Tesla coercive magnetic field in frustrated Sr$_3$NiIrO$_6$
|
We have measured extremely large coercive magnetic fields of up to 55~T in
Sr$_3$NiIrO$_6$, with a switched magnetic moment $\approx 0.8~\mu_{\rm B}$ per
formula unit. As far as we are aware, this is the largest coercive field
observed thus far. This extraordinarily hard magnetism has a completely
different origin from that found in conventional ferromagnets. Instead, it is
due to the evolution of a frustrated antiferromagnetic state in the presence of
strong magnetocrystalline anisotropy due to the overlap of spatially-extended
Ir$^{4+}$ 5$d$ orbitals with oxygen 2$p$ and Ni$^{2+}$ 3$d$ orbitals. This work
highlights the unusual physics that can result from combining the extended $5d$
orbitals in Ir$^{4+}$ with the frustrated behaviour of triangular lattice
antiferromagnets.
|
1408.0759v1
|
## Fermilab Antiproton Source, Recycler Ring, and Main Injector
Sergei Nagaitsev, Fermilab, Batavia, IL 60510
## Introduction
The antiproton source for a proton-antiproton collider at Fermilab was proposed in 1976 [1]. The proposal argued that the requisite luminosity (~10 29 cm -2 sec -1 ) could be achieved with a facility that would produce and cool approximately 10 11 antiprotons per day. Funding for the Tevatron I project (to construct the Antiproton source) was initiated in 1981 and the Tevatron ring itself was completed, as a fixed target accelerator, in the summer of 1983 and the Antiproton Source was completed in 1985. At the end of its operations in 2011, the Fermilab antiproton production complex consisted of a sophisticated target system, three 8-GeV storage rings (namely the Debuncher, Accumulator and Recycler), 25 independent multi-GHz stochastic cooling systems, the world's only relativistic electron cooling system and a team of technical experts equal to none. Sustained accumulation of antiprotons was possible at the rate of greater than 2.5×10 11 per hour. Record-size stacks of antiprotons in excess of 3×10 12 were accumulated in the Accumulator ring and 6×10 12 in the Recycler. In some special cases, the antiprotons were stored in rings for more than 50 days. Note, that over the years, some 10 16 antiprotons were produced and accumulated at Fermilab, which is about 17 nanograms and more than 90% of the world's total man-made quantity of nuclear antimatter.
The accelerator complex at Fermilab supported a broad physics program including the Tevatron Collider Run II [2], neutrino experiments using 8 GeV and 120 GeV proton beams, as well as a test beam facility and other fixed target experiments using 120 GeV primary proton beams. The following sections provide a brief description of Fermilab accelerators as they operated at the end of the Collider Run II (2011).
## 1. The Proton Source
The Proton Source consists of the Pre-Accelerator (Pre-Acc), the Linac, and the Booster. For operational redundancy, there were two independent 750-kV Pre-Acc systems which provided H ions for acceleration through the Linac. -Each Pre-Acc was a Cockcroft-Walton accelerator having its own magnetrontype H source running at a 15 Hz repetition rate, a voltage multiplier to generate -the 750 kV accelerating voltage, and a chopper to set the beam pulse length going into the Linac. The typical H source output current is 40-60 mA. -
The Linac accelerated H ions from 750 keV to 400 MeV. Originally, the -Linac was a 200 MeV machine made entirely of Alvarez-style drift tube tanks [3], but a 1991 upgrade replaced four of the drift tubes with side coupled cavities to allow acceleration up to 400 MeV [4]. Today, the low energy section (up to 116 MeV) is made of drift tube tanks operating with 201.25 MHz RF fed from triod-based 5 MW power amplifier tubes. The high energy section (116 - 400 MeV) consists of 7 side-coupled cavity girders powered by 805-MHz, 12-MW Klystrons providing a gradient of ≈ 7 MV/m. A transition section between the two linac sections provides the optics matching and rebunching into the higher frequency RF system. The nominal beam current in the Linac is 30-35 mA.
The Booster is a 474 meter circumference, rapid-cycling synchrotron ramping from 400 MeV to 8 GeV at a 15 Hz repetition rate. (Note that while the magnets ramp at 15 Hz, beam is not present on every cycle.) Multi-turn injection is achieved by passing the incoming H ions through 1.5 - m thick (300 g/cm ) 2 carbon stripping foils as they merge with the circulating proton beam on a common orbit. The 96 10-foot long combined-function Booster gradient magnets are grouped into 24 identical periods in a FOFDOOD lattice [5]. The Booster RF system (harmonic number = 84) consists of 19 cavities (18 operational + 1 spare) that must sweep from 37.9-52.8 MHz as the beam velocity increases during acceleration. The ferrite tuners and power amplifiers are mounted on the cavities in the tunnel. The cavities provide a total of ≈750 kV per turn for acceleration. The Booster transition energy (4.2 GeV) occurs at 17 ms in the cycle. The Booster throughput efficiency is 85-90% for typical beam intensities of 4.5-5.0×10 12 protons per pulse. The majority of the proton flux through the Booster is delivered to the 8 GeV and 120 GeV neutrino production targets.
## 2. The Main Injector
The Main Injector [6] (MI) is a 3319.4-m circumference synchrotron, which can accelerate both proton and antiproton beams from 8 GeV up to 150 GeV. It has a FODO lattice using conventional, separated function dipole and quadrupole magnets. There are also trim dipole and quadrupoles, skew quadrupole, sextupole and octupole magnets in the lattice. Since the Main Injector circumference is seven times the Booster circumference, beam from multiple consecutive Booster cycles, called batches, can be injected around the Main Injector. In addition, even higher beam intensity can be accelerated by injecting more than seven Booster batches through the process of slip-stacking: capturing one set of injected proton batches with one RF system, decelerating them slightly, then capturing another set of proton injections with another independent RF system, and merging them prior to acceleration. There are 18 53-MHz RF cavities, grouped into 2 independently controlled systems to allow slip-stacking and the flexibility for maintenance. Beam-loading compensation and active damping systems have been implemented to help maintain beam stability. For beam injections into the Tevatron, coalescing of several 53 MHz bunches of protons and antiprotons into single, high intensity bunches also requires a 2.5 MHz system for bunch rotations and a 106 MHz cavity to flatten the potential when recapturing beam into the single 53 MHz bunch to be injected into the Tevatron ring. A set of collimators was installed in the Main Injector to help localize beam losses to reduce widespread activation of ring components in the tunnel.
The Main Injector supports various operational modes for delivering beam across the complex. For the antiproton and neutrino production, up to 11 proton batches from the Booster were injected and slip-stacked prior to acceleration. After reaching 120 GeV, 2 batches were extracted to the antiproton production target while the remaining 9 batches were extracted to the NuMI neutrino production target (Figure 1). At its peak performance, the Main Injector can sustain 400 kW delivery of 120 GeV proton beam power at 2.2 sec cycle times. The Main Injector also provides 120 GeV protons in a 4 sec long slow-spill extracted to the Switchyard as a primary beam or for production of secondary and tertiary beams for the Meson Test Beam Facility and other fixed-target experiments. In addition, the Main Injector served as an 'effective' transport line for 8 GeV antiprotons being transferred from the Accumulator to the Recycler for later use in the Tevatron. Protons from the Booster and antiprotons from the Recycler were accelerated to 150 GeV in the Main Injector and
coalesced into higher intensity bunches for injection into the Tevatron for a colliding beam store.
Figure 1: The Main Injector cycle illustrating an 11-batch proton injection, acceleration and extraction.
## 3. The Antiproton Source
The Antiproton Source [7] had 3 main parts: the Target Station, the Debuncher, and the Accumulator. Each of these is described briefly below while outlining the steps of an antiproton production cycle. In the Target Station, batches of 120-GeV protons (~8×10 12 per batch), delivered from the Main Injector, strike the Inconel (a nickel-iron alloy) target every 2.2 sec. The beam spot on the target can be controlled by a set of quadrupole magnets. The target is rotated between beam pulses to spread the depletion and damage uniformly around its circumference. The shower of secondary particles, emanating from the target, was focused both horizontally and vertically by a pulsed, high current lithium lens that can provide up 1000 T/m gradient.
Figure 2: The Antiproton Source consists of the Debuncher ring (outer circumference, left) and the Accumulator ring (inner circumference, right).
Downstream of the Li lens was a pulsed dipole magnet, which steered negatively-charged particles at 8.9-GeV/c momentum into the transport line toward the Debuncher. A collimator between the lens and pulsed magnet was installed to help protect the pulsed magnet from radiation damage as the incoming primary proton beam intensity increased with proton slip-stacking in Main Injector.
The Debuncher and Accumulator (Figure 2) were both triangular-shaped rings of conventional magnets sharing the same tunnel. While the Debuncher had a FODO lattice, the Accumulator lattice had particular features needed for cooling and accumulating antiprotons with stochastic cooling systems. A total of 21 independent stochastic cooling systems were implemented in the Accumulator and Debuncher [8]. Such a variety of cooling systems was possible after a series of development efforts [9, 10] allowing for more robust and less expensive pickup arrays.
The ~2×10 bunches of antiprotons entering the 8 Debuncher from the transport line retained the 53-MHz bunch structure from the primary proton beam on the target. A 53-MHz RF system (harmonic number = 90) was used for the bunch rotation and debunching of the antiprotons into a continuous beam with a low momentum spread. An independent 2.4 MHz RF system provided a barrier bucket to allow a gap for extraction to the Accumulator. Stochastic cooling
systems reduced the transverse emittance from 300 to 30 π mm-mrad (rms, normalized) and momentum spread from 0.30% to <0.14% prior to injection into the Accumulator.
Figure 3: The frequency (energy) distribution of antiprotons in the Accumulator highlighting incoming antiprotons (left), the stacktail beam (middle) being cooled and decelerated toward the core (right). Higher beam energy is to the left, lower energy is to the right.
In the Accumulator, antiprotons were momentum-stacked and cooled by a series of RF manipulations and stochastic cooling. The incoming antiprotons were captured and decelerated by 60 MeV with a 53-MHz RF system (harmonic number = 84) to the central orbit where the beam was adiabatically debunched. Before the next pulse of antiprotons arrived (every 2.2 sec), the so-called stacktail momentum stochastic cooling system [11] decelerated the antiprotons another 150 MeV toward the core orbit where another set of independent betatron and momentum stochastic cooling systems provided additional cooling while building a 'stack' of antiprotons. Figure 3 illustrates the frequency (energy) distribution of antiprotons in the Accumulator. Figure 4 shows the average antiproton accumulation rates since 1994; typical values at the end of Run II were in the range 24-26 ×10 10 /hr, with a maximum recorder rate of 28.5 ×10 10 /hr.
Figure 4: The average antiproton accumulation rate since 1994 and during all of Collider Run II (including the production in the Antiproton Source and storage in the Recycler). ). Note that some data points at the highest rates, in particular in the early years, are merely artifacts of the data acquisition and logging system.
## 3. The Recycler
The Recycler [12] is a permanent-magnet, fixed momentum (8.9-GeV/c) storage ring located in the Main Injector tunnel (Figure 5). As conceived the Recycler would provide storage for very large numbers of antiprotons (up to 6×10 12 ) and would increase the effective production rate by recapturing unused antiprotons at the end of collider stores (hence the name Recycler). Recycling of antiprotons was determined to be ineffective and was never implemented. However, the Recycler was used as a final antiproton cooling and storage ring for accumulating significantly larger stashes (so called to differentiate from Accumulator 'stacks') of antiprotons than can be accumulated in the Antiproton Accumulator. The main Recycler magnets are combined-function strontium ferrite permanent magnets arranged in a FODO lattice. Trim electro-magnets are used to provide orbit and lattice corrections.
Figure 5: The Recycler (top) and the Main Injector (bottom) rings installed in a common tunnel.
An important feature of the Recycler was an electron cooling system [13]. It augmented the Recycler's cooling capability and complemented the stochastic cooling system and its inherent limitations. The Pelletron (an electrostatic accelerator manufactured by the National Electrostatics Corp.) provided a 4.3 MeV electron beam (up to 500 mA) which overlaped with the 8-GeV antiprotons, circulating in the Recycler, in a 20-m long section and cooled the antiprotons both transversely and longitudinally. Figure 6 shows the schematic layout of the Fermilab electron cooling system. The dc electron beam was generated by a thermionic gun, located in the high-voltage terminal of the electrostatic accelerator. This accelerator was incapable of sustaining dc beam currents to ground in excess of about 100 µA. Hence, to attain the electron dc current of 500 mA, a recirculation scheme was employed. A typical relative beam current loss in such a scheme was 2×10 -5 .
Figure 6: Schematic layout of the Recycler electron cooling system and accelerator cross-section (inset).
The Fermilab system employed a unique beam transport scheme [14]. The electron gun was immersed in a solenoidal magnetic field, which created a beam with large angular momentum. After the beam was extracted from the magnetic field and accelerated to 4.3 MeV, it was transported to the 20-m long cooling section solenoid using conventional focusing elements (as opposed to low-energy electron coolers where the beam remains immersed in a strong magnetic field at all times). The cooling section solenoid removed this angular momentum and the beam was made round and parallel such that the beam radius, a , produced the same magnetic flux, Ba 2 , as at the cathode. The magnetic field in the cooling section was quite weak (100 G), therefore the kinetics of the electron-antiproton scattering was weakly affected by the magnetic field.
After becoming operational in September 2005, electron cooling in the Recycler allowed for significant improvements in Tevatron luminosity by providing higher intensity antiprotons with smaller emittances. With electron cooling, the Recycler has been able to store up to 6×10 12 antiprotons. In routine operations, the Recycler accumulated 3.5-4.0×10 12 antiprotons with a ~200-hr lifetime for injection to the Tevatron [15].
Among other unique features of the Recycler was the so-called barrierbucket rf system [16] which allowed for crucial longitudinal beam manipulations of the accumulated antiproton beam.
## 5. Antiproton Flow and Tevatron Shot-Setup
As mentioned previously, stacks of freshly produced antiprotons were stored temporarily in the Accumulator. The Accumulator antiproton stack was periodically transferred to the Recycler where electron cooling allowed for a much larger antiproton intensity to be accumulated with smaller emittances. Typically 22-25×10 10 antiprotons were transferred to the Recycler every ~60 minutes. Prior to electron cooling in the Recycler, antiprotons destined for the Tevatron were extracted from the Accumulator only. Since late 2005, all Tevatron antiprotons were extracted from the Recycler only. Figure 7 illustrates the flow of antiprotons between the Accumulator, Recycler and Tevatron over a 1 week period.
100
Figure 7: Production and transfers of antiprotons between the Accumulator and Recycler over 1 week of operation. While the Tevatron has a colliding beam store, small stacks of antiprotons are produced and stored in the Accumulator, and then periodically transferred to the Recycler in preparation for the subsequent Tevatron fill.
A typical Tevatron collider fill cycle is shown in Figure 8 [17]. First, proton bunches were injected two at a time on the central orbit. Then, electrostatic separators were powered to put the protons onto a helical orbit. Antiproton bunches were then injected (four bunches at a time) into gaps between the three proton bunch trains. After each group of 3 antiproton transfers, the gaps were
cleared for the subsequent set of transfers by 'cogging' the antiprotons changing the antiproton RF cavity frequency to let them slip longitudinally relative to the protons. Once the beam loading was complete, the beams were accelerated to the top energy (in 86 seconds) and the machine optics was changed to the collision configuration in 25 steps over 125 sec (low-beta squeeze). The last two stages included initiating collisions at the two collision points and removing halo by moving in the collimators. The experiments then commenced data acquisition for the duration of the high-energy physics (HEP) store.
Figure 8: The collider fill cycle for store #8709 (May 2011).
## Summary
For more than 25 years (1985-2011) the Fermilab antiproton complex was the centerpiece of the Tevatron collider program [18] and provided antiprotons for other particle physics experiments [19]. The continued Tevatron luminosity increase was mainly due to a larger number of antiprotons being available, which in turn was the result of a continuous and dedicated effort of hundreds of experts to optimize and improve antiproton accumulation and cooling. The antiproton stochastic and electron cooling methods were not invented at Fermilab, but they were perfected to a degree not achieved anywhere else in the world.
## References
- [1] D. Cline et al ., Proposal to construct an antiproton source for the Fermilab accelerators , proposal 492, in Proceedings of 1976 NAL Summer Study on Utilization of the Energy Doubler/Saver, Fermilab, Batavia U.S.A. (1976), pg. 309.
- [2] Tevatron Collider Run II Handbook , Ed. S.D. Holmes, Fermilab Preprint TM-2484 (1998).
- [3] C. D. Curtis et al., Linac H- Beam Operations and uses at Fermilab, IEEE Transactions on Nuclear Science, NS-26, No. 3 (1979), 3760-3762.
- [4] C.W. Schmidt, The Fermilab 400-MeV Linac Upgrade , in Proceedings of 1993 Particle Accelerator Conf., Washington, DC 1993, pp.1655-1659.
- [5] E. Hubbard et al., Booster Synchrotron, Fermilab Technical Memo TM-405 (1973).
- [6] S.D. Holmes, Design Criteria and Performance Goals for the Fermilab Main Injector , Particle Accelerators, 58 (1997), 39.
- [7] M. Church and J. Marriner, The Antiproton Sources: Design and Operation, Annual Review of Nuclear and Particle Science, 43(1993), 253-295.
- [8] R. J. Pasquinelli, Implementation of stochastic cooling hardware at Fermilab's Tevatron collider , JINST 6 T08002 (2011).
- [9] D. McGinnis, Theory and design of microwave planar electrodes for stochastic cooling of particle beams , Microw. Opt. Technol. Lett. 4 (1991) 439.
- [10] D. McGinnis, Slotted waveguide slow-wave stochastic cooling arrays , Proceedings of 1999 PAC, IEEE Conference, New York U.S.A. (1999), pg. 1713.
- [11] V. Lebedev, Improvements to the stacktail and Debuncher momentum cooling systems , in Proceedings of COOL 09, MOA1MCCO02, Lanzhou China (2009).
- [12] Fermilab Recycler Ring Technical Design Report , Ed. G. Jackson, Fermilab Preprint TM-1991 (1997).
- [13] S. Nagaitsev et al., Experimental Demonstration of Relativistic Electron Cooling , Phys. Rev. Lett. 96, 044801 (2006).
- [14] A. Burov et al., Optical principles of beam transport for relativistic electron cooling , Phys. Rev. Spec. Top., Accel. Beams 3, 094002 (2000).
- [15] A. Shemyakin and L.R. Prost, Ultimate performance of relativistic electron cooling at Fermilab , in Proceedings of COOL11, THIOA01, Alushta, Ukraine (2011).
- [16] C.M. Bhat, Longitudinal momentum mining of beam particles in a storage ring , Physics Letters A 330 (2004), p.481
- [17] C. Gattuso et al., Optimization of Integrated Luminosity of the Tevatron , in proceedings of PAC 2009, Vancouver, B. C., Canada.
- [18] S. Holmes et al., Overview of the Tevatron collider complex: goals, operations and performance , JINST 6 T08001 (2011).
- [19] For example: Charmonium [E760/E835], Antihydrogen [E862], and APEX [E868].
| null |
[
"Sergei Nagaitsev"
] |
2014-08-04T18:23:19+00:00
|
2014-08-04T18:23:19+00:00
|
[
"physics.acc-ph",
"hep-ex"
] |
Fermilab Antiproton Source, Recycler Ring, and Main Injector
|
At the end of its operations in 2011, the Fermilab antiproton production
complex consisted of a sophisticated target system, three 8-GeV storage rings
(namely the Debuncher, the Accumulator and the Recycler), 25 independent
multi-GHz stochastic cooling systems, the world's only relativistic electron
cooling system and a team of technical experts equal to none. The accelerator
complex at Fermilab supported a broad physics program including the Tevatron
Collider Run II, neutrino experiments using 8-GeV and 120-GeV proton beams, as
well as a test beam facility and other fixed target experiments using 120-GeV
primary proton beams. This paper provides a brief description of Fermilab
accelerators as they operated at the end of the Collider Run II (2011).
|
1408.0760v1
|
## BLOCH'S CONJECTURE FOR GENERALIZED BURNIAT TYPE SURFACES WITH p g = 0
## INGRID BAUER, DAVIDE FRAPPORTI
Abstract. The aim of this article is to prove Bloch's conjecture, asserting that the group of rational equivalence classes of zero cycles of degree 0 is trivial for surfaces with geometric genus zero, for regular generalized Burniat type surfaces. The technique is the method of 'enough automorphisms' introduced by InoseMizukami in a simplified version due to the first author.
## Introduction
Let S be a smooth projective surface and let
$$\begin{array} {$$
be the group of rational equivalence classes of zero cycles on S . Then Bloch's conjecture asserts the following:
Conjecture ([Blo75]) . Let S be a smooth surface with p g ( S ) = 0 . Then the kernel T S ( ) of the natural morphism:
is trivial.
$$A _ { 0 } ^ { 0 } ( S ) \longrightarrow \text{Alb}$$
The conjecture has been proven for surfaces S with Kodaira dimension kod( S ) ≤ 1 by Bloch, Kas and Liebermann (cf. [BKL76]) and has been verified for several examples, see e.g. [Bar85], [Bau14], [CC13], [IM79], [Keu88], [Voi92]. Thanks to a result of S. Kimura (cf. [Kim05]), all product quotient surfaces (i.e. minimal models of ( C 1 × C 2 ) /G , where G is a finite group acting on the product of two curves of genus at least 2) with p g = 0 satisfy Bloch's conjecture (cf. [BCGP12]).
Burniat surfaces are surfaces of general type with invariants p g = 0 and K 2 = 6 5 4 3 2 whose birational models were constructed by , , , , P. Burniat in 1966 (cf. [Bur66]) as singular bidouble covers of the projective plane. In 1994, M. Inoue (cf. [Ino94]) reconstructed them as quotients of a divisor in a product of three elliptic curves by a finite group acting freely (see also [BC12] and [BC13]).
Date : August 5, 2014.
2000 Mathematics Subject Classification. 14C25, 14J29, 14J50.
The present work took place in the realm of the DFG Forschergruppe 790 'Classification of algebraic surfaces and compact complex manifolds'.
Following and generalizing Inoue's approach, in the recent paper [BCF], we construct and classify a new class of surfaces of general type 'generalized Burniat type surfaces' . These surfaces have invariants K 2 = 6 and 0 ≤ p g = q ≤ 3 and have been constructed as quotient of a divisor of multi-degree (2 , 2 2) in a product of three elliptic curves by a , free ( Z / 2 Z ) 3 -action (see Section 3). Generalized Burniat type surfaces form 16 irreducible families; four families have p g = 0 and form four connected components N i of the Gieseker moduli space M can 1 6 , . Each component is irreducible, generically smooth, normal and unirational of dimension 4 in two cases and 3 in the others.
The main result of this note is to show that Bloch's conjecture holds for generalized Burniat type surfaces with p g = 0. The proof uses the method of 'enough automorphisms' introduced by Inose and Mizukami (cf. [IM79]) and refined by Barlow (cf. [Bar85]), but in a simpler way (cf. [Bau14]).
Theorem. Let S be a generalized Burniat type surface with p g ( S ) = 0 . Then S verifies Bloch's conjecture: T S ( ) = A 0 0 ( S ) = 0 .
In [BCF] the authors show among others the following result:
Theorem. Let S be any surface whose moduli point lies in the connected component of the Gieseker moduli space of surfaces of general type of a regular generalized Burniat type surface, then S is a generalized Burniat type surface.
More precisely, in [BCF] it is proven that generalized Burniat type surfaces form exactly four irreducible connected components in their moduli space (cf. Theorem 4.6). Therefore our result proves Bloch's conjecture for each surface in the connected component of any regular generalized Burniat type surface.
## 1. Bloch's conjecture for surfaces with a ( Z / 2 Z ) 2 -action
The aim of this note is to prove Bloch's conjecture for generalized Burniat type surfaces with p g = q = 0. The proof uses the method of 'enough automorphisms' introduced by Inose and Mizukami (cf. [IM79]) and refined by Barlow (cf. [Bar85]).
Definition 1.1. Let G be a finite group and H ≤ G be a subgroup. Then we set
$$z ( H ) \coloneqq \sum _ { h \in H } h \in \math$$
Lemma 1.2 ([Bar85]) . Let S be a nonsingular surface and G a finite subgroup of Aut( S ) . Let H,H ,..., H 1 r be subgroups of G . We denote by I the two-sided ideal of C G generated by z H ( 1 ) , . . . , z ( H r ) . Assume that
$$i \right ) \, z ( H ) \in \mathcal { I },$$
ii) T S/H ( i ) = 0 , for every i ∈ { 1 , . . . , r } . Then T S/H ( ) = 0 .
Using this result, the first author proved the following
Proposition 1.3 ([Bau14, Proposition 1.3]) . Let S be a surface of general type with p g ( S ) = 0 . Assume that G = ( Z / 2 Z ) 2 /triangleleft Aut( S ) . Then S satisfies Bloch's conjecture if and only if for each σ ∈ G \ { 0 } the quotient S/σ satisfies Bloch's conjecture.
Remark 1.4 . Note that S/σ is a surface with at most nodes as singularities and denoting by X σ → S/σ the resolution of its singularities, X σ is minimal and has p g = 0. Moreover, since nodes are rational singularities, T S/σ ( ) = T X ( σ ).
Using the fact that by the result of Bloch, Kas and Liebermann (cf. [BKL76]) Bloch's conjecture is true for surfaces S with kod( S ) ≤ 1, we obtain the following:
Corollary 1.5 ([Bau14, Corollary 1.5]) . Let S be a surface of general type with p g ( S ) = 0 and assume that G = ( Z / 2 Z ) 2 /triangleleft Aut( S ) . Assume that for each σ ∈ G \ { 0 } the quotient S/σ has kod( S/σ ) ≤ 1 , then S satisfies Bloch's conjecture.
## 2. Involutions on surfaces of general type
In this section we collect some results regarding involutions on surfaces of general type, that we need in Section 4. We start fixing some notation.
Let S be a minimal regular surface of general type with an involution σ . Then σ is biregular and its fixed locus is the union of k isolated points P . . . , P 1 k and a smooth (not necessarily connected) curve R . We denote by p : S → Σ := S/ σ 〈 〉 the projection onto the quotient, by B the image of R and by Q j the image of P j , j = 1 , . . . , k . The surface Σ is normal, Q , . . . , Q 1 k are nodes and they are the only singularities of Σ. Let h : V → S be the blow-up of S at P , . . . , P 1 k and E j be the exceptional curve over P j , j = 1 , . . . , k .
The involution σ induces an involution ˜ on σ V whose fixed locus is the union of R ′ = h -1 R and of E , . . . , E 1 k . Let π : V → W := V/ σ 〈 ˜ 〉 be the projection onto the quotient and set B ′ := π R ( ′ ), A j := π E ( j ), j = 1 , . . . , k . The surface W is smooth and the A j are disjoint ( -2)-curves. Let g be the morphism induced by h g , is the minimal resolution of the singularities of Σ and we have the following commutative diagram:
/d47
/d47
/d15
/d15
/d15
/d15
/d47
/d47
Let B ′′ := B ′ + ∑ k j =1 A j be the branch divisor of π , one has π ∗ O V = O W ⊕O W ( -∆) with ′ B ′′ ≡ 2∆ . ′
Now let us assume that W is of general type. We denote by P the minimal model of W , by ρ : W → P the corresponding projection and let B := ρ ∗ ( B ′′ ). Then B is an even divisor linearly equivalent to 2∆, where ∆ := ρ ∗ (∆ ). ′ V is the canonical resolution of the double cover of P branched along B and, by [Hor75, Lemma 6], one has:
$$K _ { S } ^ { 2 } - k = K _ { V } ^ { 2 } =$$
$$( 2. 3 ) \ \chi ( \mathcal { O } _ { S } ) = \chi ( \mathcal { O } _ { V } ) = 2 \chi ( \mathcal { O } _ { P } ) + \frac { 1 } { 2 } ( K _ { P } + \Delta ). \Delta - \frac { 1 } { 2 } \sum _ { i } x _ { i } ( x _ { i } - 1 ) \,,$$
where x i := /floorleft m i 2 /floorright , being m i the multiplicity of the singular point b i of B . We recall the following:
Proposition 2.1 ([Bom73, Proposition 1]) . Let S be a minimal surface of general type. Let C be an irreducible curve on S , then K .C S ≥ 0 and if K .C S = 0 , then C 2 = -2 and C is a rational non-singular curve.
We can now prove:
Proposition 2.2. Let S be a (minimal) surface of general type with K 2 S = 6 , p g ( S ) = 0 and that contains no rational curves except at most a ( -2) -curve L . Let σ be an involution on S such that one of the following holds:
- (i) either Fix σ ( ) contains more than 8 isolated points and a nonrational curve;
- (ii) or Fix σ ( ) is given by 6 isolated points and an elliptic curves C such that C ∩ L = ∅ ;
Then Σ := S/ σ 〈 〉 is not of general type.
- (iii) or Fix σ ( ) is given by 4 isolated points, an elliptic curves C and the ( -2) -curve L : C ∩ L = ∅ .
Proof. We use the notation of above and aiming for a contradiction we assume W of general type.
Since S is a minimal surface of general type with K 2 S = 6 and p g ( S ) = 0, then p g ( P ) = q P ( ) = 0 and from formulas (2.2), (2.3) we get
$$5 - \frac { k } { 2 } = K _ { P } ^ { 2 } + K _ { P }. \Delta + \sum _ { i } ( x _ { i } - 1 ) \,.$$
We claim that K .B P = 2 K . P ∆ > 0, indeed: B = ∑ r l =1 B l + ∑ j ρ A ∗ j , where, by assumption, each B l is an irreducible curve, r > 0 and B t is non-rational for at least one t ∈ { 1 , . . . , r } . By Proposition 2.1, K .B P l ≥ 0 for any l = 1 , . . . , r and K .B > P t 0 and the claim follows.
Case (i) : k ≥ 8 and by (2.4) we get:
$$1 \geq K _ { P } ^ { 2 } + K _ { P }. \Delta + \sum _ { i } ( x _ { i } - 1 ) \,,$$
but this is not possible because K 2 P ≥ 1 and K . P ∆ > 0. Case (ii) : k = 6 and by (2.4) we get:
$$2 = K _ { P } ^ { 2 } + K _ { P }. \Delta + \sum _ { i } ( x _ { i } - 1 ) \,,$$
therefore K 2 P = K . P ∆ = 1 and x i = 1 for any i . Since e V ( ) = 12, it holds e W ( ) = 6 + 1 2 e B ( ′′ ); by assumption B ′′ := ∑ 6 j =1 A j + ˜ with C ˜ := C π h ( -1 ( C )) a smooth elliptic curve: e B ( ′′ ) = 12. By Noether's formula K 2 W = 0, hence ρ is the blow-down of exactly one ( -1)-curve E . In particular E does not intersect any ( -2)-curves. By assumption, S contains no rational curves except at most a ( -2)curve L , such that L ∩ C = ∅ . If S contains such a curve L , then σ maps L onto itself hence two σ -fixed points lie on it; an easy computation shows that π L ( ˜ ) is a ( -2)-curve on W , being ˜ L ⊂ V the strict transform of L . Therefore, the rational curve E must intersect ˜ C in at least 4 points (by Hurwitz' formula) and B contains a singular point with m x
Case (iii) : k = 4 and by (2.4) we get:
i ≥ 4, i.e. i ≥ 2.
$$3 = K _ { P } ^ { 2 } + K _ { P }. \Delta + \sum _ { i } ( x _ { i } - 1 ) \,,$$
therefore K 2 P ≤ 2.
Let ˜ := C π h ( -1 ( C )) and ˜ := L π h ( -1 ( L )), then B ′′ = ∑ 4 j =1 A j + ˜ + ˜ is C L the disjoint union of five rational curves and an elliptic curve: e B ( ′′ ) = 10. Since e V ( ) = 10, e W ( ) = 1 2 e V ( ) + 1 2 e B ( ′′ ) = 10 and by Noether's formula 2 = K 2 W ≤ K 2 P . We get that W = P is minimal, K . P ∆ = 1 and x i = 1 for any i . By (2.3), it follows ∆ 2 = -3 and so -12 = B ′′ 2 = -12. It is direct to show that ˜ is a ( L -4)-curve, hence ˜ C 2 -12 = B ′′ 2 . We have an elliptic curve ˜ C with ˜ C 2 = 0 on a minimal surface of general type, it contradicts Proposition 2.1. /square
## 3. Generalized Burniat type surfaces
In this section we recall the construction of generalized Burniat type surfaces. For further details we refer to [BCF].
For j = 1 2 3, let , , E j = C / 〈 1 , τ j 〉 be an elliptic curve and denote by z j a uniformizing parameter on E j . Let L j be the Legendre L -function for E j : L j is a meromorphic function on E j and L j : E j → P 1 is a double cover branched over four distinct points: ± 1 , ± a j ∈ P 1 \ { 0 , ∞} . It is well known that the following statements hold (see [Ino94, Lemma 3-2] and [BC11, Section 1]):
$$\dot { \bullet } \mathcal { L } _ { j } ( 0 ) = 1, \, \mathcal { L } _ { j } ( \frac { 1 } { 2 } ) ^ { \mu } = - 1, \, \mathcal { L } _ { j } ( \frac { \tau _ { j } } { 2 } ) = a _ { j }, \, \mathcal { L } _ { j } ( \frac { \tau _ { j } + 1 } { 2 } ) = - a _ { j } ;$$
- · let b j := L j ( τ j 4 ), then b 2 j = a j ;
- · L j ( z j ) = L j ( z j +1) = L j ( z j + τ j ) = L j ( -z j ) = -L j ( z j + 1 2 ) ;
- · d L j d z j ( z j ) = 0 if and only if z j ∈ { 0 , 1 2 , τ j 2 , τ j +1 2 } (since these are the ramification points of L j );
- · L j ( z j + τ j 2 ) = a j L j ( z j ) .
For j ∈ { 1 2 3 , , } , we define an action of { ( ζ , η j j , /epsilon1 j ) } ∼ = ( Z / 2 Z ) 3 on E j , as follows:
↦
↦
↦
$$( z _ { j } \mapsto - z _ { j } ) \quad & \hat { = } \quad ( 1, 0, 0 ) \\ ( 3. 1 ) & ( z _ { j } \mapsto - z _ { j } + \frac { \tau _ { j } } { 2 } ) \quad & \hat { = } \quad ( 0, 1, 0 ) \\ ( z _ { j } \mapsto - z _ { j } + \frac { 1 } { 2 } ) \quad & \hat { = } \quad ( 0, 0, 1 ), \\ \text{and we consider the induced action of } \mathcal { G } & = \{ ( \zeta _ { 1 }, \eta _ { 1 }, \epsilon _ { 1 }, \zeta _ { 2 }, \eta _ { 2 }, \cdot \,,$$
$$\pi ^ { \prime } \colon E _ { 1 } \times E _ { 2 } \times E _ { 3 } & \quad \longrightarrow \quad \mathbb { P } ^ { 1 } \times \mathbb { P } ^ { 1 } \times \mathbb { P } ^ { 1 } \\ ( z _ { 1 }, z _ { 2 }, z _ { 3 } ) & \quad \longmapsto \quad \left ( \frac { \mathcal { L } _ { 1 } ( z _ { 1 } ) } { b _ { 1 } }, \frac { \mathcal { L } _ { 2 } ( z _ { 2 } ) } { b _ { 2 } }, \frac { \mathcal { L } _ { 3 } ( z _ { 3 } ) } { b _ { 3 } } \right ).$$
and we consider the induced action of G = { ( ζ , η 1 1 , /epsilon1 1 , ζ 2 , η 2 , /epsilon1 2 , ζ 3 , η 3 , /epsilon1 3 ) } on T := E 1 × E 2 × E 3 . We define also the following map:
$$( z _ { 1 }, z _ { 2 }, z _ { 3 } ) \quad \longmapsto \ \left ( \frac { \mathcal { L } _ { 1 } ( z _ { 1 } ) } { b _ { 1 } }, \frac { \mathcal { L } _ { 2 } ( z _ { 2 } ) } { b _ { 2 } }, \frac { \mathcal { L } _ { 3 } ( z _ { 3 } ) } { b _ { 3 } } \right ).$$
↦
The G -action on T induces, via π ′ , an action of H := ( H 1 ) 3 ∼ = ( Z / 2 Z ) 6 on P 1 := ( P 1 ) 3 , where H ∼ 1 = ( Z / 2 Z ) 2 acts on P 1 in this way:
↦
$$& ( 1, 0 ) \ \stackrel { \, \hat { \equiv } \ \left ( ( s \colon t ) \mapsto ( t \colon s ) \right ), } { ( 0, 1 ) \ \stackrel { \, \hat { \equiv } \ \left ( ( s \colon t ) \mapsto ( s \colon - t ) \right ) } { \, } }, \\ & \text{hain } l o \cdot t \ll \homogeneous \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \sin \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \ cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \ sin \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \os \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \sec \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \colon$$
↦
being ( s : t ) homogeneous coordinates of P 1 . Let Y ⊂ P 1 × P 1 × P 1 be an irreducible Del Pezzo surface of degree 6 invariant under a subgroup H ∼ = ( Z / 2 Z ) 2 /triangleleft H .
The inverse image ˆ := X π ′-1 ( Y ) of Y under π ′ is an irreducible hypersurface in the product of three smooth elliptic curves T := E 1 × E 2 × E 3 , which is of multi degree (2 , 2 2). ,
## Definition 3.1. ˆ X is called a Burniat hypersurface in T .
Remark 3.2 . According to [BCF], every Burniat hypersurface is given by one of the following equations:
$$\hat { X }$$
$$\hat { X } _ { \nu } = \{ ( z _ { 1 }, z _ { 2 }, z _ { 3 } ) \in T \ | \ \nu _ { 1 } ( \mathcal { L } _ { 1 } ( z _ { 1 } ) \mathcal { L } _ { 2 } ( z _ { 2 } ) \mathcal { L } _ { 3 } ( z _ { 3 } ) + b _ { 1 } b _ { 2 } b _ { 3 } ) + \\ \nu _ { 2 } ( \mathcal { L } _ { 1 } ( z _ { 1 } ) b _ { 2 } b _ { 3 } + b _ { 1 } \mathcal { L } _ { 2 } ( z _ { 2 } ) \mathcal { L } _ { 3 } ( z _ { 3 } ) ) = 0 \} \,, \\ \hat { \ }$$
$$( 3. 5 ) \quad \hat { X } _ { \mu } = \left \{ ( z _ { 1 }, z _ { 2 }, z _ { 3 } ) \in T \, | \, \mathcal { L } _ { 1 } ( z _ { 1 } ) \mathcal { L } _ { 2 } ( z _ { 2 } ) \mathcal { L } _ { 3 } ( z _ { 3 } ) = \mu \right \},$$
$$& ( 3. 6 ) \quad \hat { X } _ { b } = \{ ( z _ { 1 }, z _ { 2 }, z _ { 3 } ) \in T \ | \ \mathcal { L } _ { 1 } ( z _ { 1 } ) \mathcal { L } _ { 2 } ( z _ { 2 } ) \mathcal { L } _ { 3 } ( z _ { 3 } ) = b _ { 1 } b _ { 2 } b _ { 3 } \} \,, \\ & \quad \Phi \Phi \dots \dots - T \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots$$
Recall that we are considering only values of ν (resp. µ ) such that ˆ X ν (resp. ˆ X µ ) is irreducible, i.e. ( ν /ν 1 2 ) = ± 1 and µ = 0.
where ν := ( ν 1 : ν 2 ) ∈ P 1 , µ ∈ C and b := b b b 1 2 3 .
/negationslash
/negationslash
Remark 3.3 . By construction, a Burniat hypersurface ˆ X has at most finitely many nodes as singularities. Therefore, denoting by /epsilon1 : X ′ → ˆ X the minimal resolution of its singularities, we have that K X ′ = /epsilon1 ∗ K ˆ X and X ′ is a minimal surface of general type with K 2 X ′ = 48 and χ X ( ′ ) = 8.
Let G 0 ∼ = ( Z / 2 Z ) 6 /triangleleft G ∼ = ( Z / 2 Z ) 3 × ( Z / 2 Z ) 3 × ( Z / 2 Z ) 3 be the group: G 0 := { ( ζ , η 1 1 , /epsilon1 1 , ζ 2 , η 2 , /epsilon1 2 , ζ 3 , η 3 , /epsilon1 3 ) | η 1 = η 2 = η , /epsilon1 3 1 + /epsilon1 2 + /epsilon1 3 = 0 } .
Then we have the following:
Lemma 3.4 ([BCF]) . (1) ˆ X ν is invariant under the group
G ′ 1 := { ( ζ , η 1 1 , 0 , ζ 2 , η 1 , /epsilon1 2 , ζ 3 , η 1 , /epsilon1 3 ) | /epsilon1 2 + /epsilon1 3 = 0 } ∼ = ( Z / 2 Z ) 5 /triangleleft G 0 .
- (2) ˆ X µ is invariant under the group
- G 1 := { ( ζ , 1 0 , /epsilon1 1 , ζ 2 , 0 , /epsilon1 2 , ζ 3 , 0 , /epsilon1 3 ) | /epsilon1 1 + /epsilon1 2 + /epsilon1 3 = 0 } ∼ = ( Z / 2 Z ) 5 /triangleleft G 0 .
- (3) ˆ X b is invariant under G 0 .
Remark 3.5 . Let g ∈ G 0 \ { 0 } be an element fixing points on T . By [BC13, Proposition 4.3], g is then an element in Table 1.
Table 1. The element of G 0 having fixed points on T
| | g 1 | g 2 | g 3 | g 4 | g 5 | g 6 | g 7 | g 8 | g 9 | g 10 | g 11 | g 12 | g 13 | g 14 | g 15 | g 16 | g 17 |
|-------------|-------|-------|-------|-------|-------|-------|-------|-------|-------|--------|--------|--------|--------|--------|--------|--------|--------|
| ζ 1 | 0 | 0 | 1 0 | | 1 | 1 | 0 | 0 | 0 | 1 | 1 | 0 | 0 | 0 | 0 | 1 | 1 |
| η 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 1 | 1 | 1 |
| /epsilon1 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 1 | 0 | 0 | 1 | 1 | 0 | 0 | 1 | 1 |
| ζ 2 | 0 | 1 | 0 | 1 | 0 | 1 | 0 | 0 | 0 | 1 | 0 | 1 | 0 | 0 | 1 | 0 | 1 |
| η 2 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 1 | 1 | 1 |
| /epsilon1 2 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 1 | 0 | 1 | 0 | 1 | 0 | 1 | 0 | 1 |
| ζ 3 | 1 | 0 | 0 | 1 | 1 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 1 | 0 | 1 | 1 | 0 |
| η 3 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 1 | 1 | 1 |
| /epsilon1 3 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 1 | 0 | 0 | 1 | 1 | 0 | 0 | 1 | 1 | 0 |
1) Let ˆ := X ˆ . X b In Table 1, the elements g 1 -g 3 fix pointwise a surface S ⊂ T . Each element g 4 -g 9 fixes pointwise a curve C ⊂ T and its fixed locus has non trivial intersection with ˆ X since ˆ X ⊂ T is an ample divisor. Finally, the elements g 10 -g 17 have isolated fixed points on T ; in particular, the elements g 11 -g 17 have fixed points on ˆ , X while the fixed locus of element g 10 intersects ˆ X only for special choices of the three elliptic curves.
2) The same holds for ˆ X ν := π ′-1 ( Y ν ) (resp. π ′-1 ( Y µ )), considering only the elements g 1 -g 7 , g 10 , g 11 , g 14 , g 15 (resp. g 1 -g 13 ), i.e. the ones belonging to G ′ 1 (resp. G 1 ). In particular, the fixed locus of element g 10 intersects ˆ X only for special choices of the three elliptic curves and the parameter ν (resp. µ ).
Definition 3.6. Let ˆ be a Burniat hypersurface in X E 1 × E 2 × E 3 and let G ∼ = ( Z / 2 Z ) 3 be a subgroup of G 0 acting freely on ˆ . X The minimal resolution S of the quotient surface ˆ := S ˆ X/G is called a generalized Burniat type (GBT) surface . We call ˆ the S quotient model of S .
Remark 3.7 . 1) Since G acts freely and ˆ X has at most nodes as singularities, ˆ is singular if and only if S ˆ X is singular and ˆ has at most S nodes as singularities.
- 2) A generalized Burniat type surface S is a smooth minimal surface of general type with K 2 S = 6 and χ S ( ) = 1.
In [BCF], GBT surfaces have been completely classified. In particular, it has been shown there are exactly four families of GBT surfaces with p g = q = 0:
Theorem 3.8. Let S → ˆ = S ˆ X/G be a regular generalized Burniat type surface S then ( ˆ X,G ) ∈ { ( ˆ X , G ν 1 ) , ( ˆ X ,G µ 2 ) , ( ˆ X , G b j ) , j = 3 4 , } , where the groups G , G , G , G 1 2 3 4 are in Table 2.
| ζ 1 | η 1 | 1 | /epsilon1 | ζ 2 | η 2 | /epsilon1 2 | ζ 3 | η 3 | /epsilon1 3 |
|-------|-------|-----|-------------|-------|-------|---------------|-------|-------|---------------|
| 1 | 0 | 0 | | 1 | 0 | 0 | 1 | 0 | 0 |
| 1 0 | 1 | | 0 | 1 | 1 | 0 | 1 | 1 | 0 |
| 0 | | 0 | 0 | 0 | 0 | 1 | 1 | 0 | 1 |
| 1 0 | 0 | | 0 | 0 | 0 | 1 | 1 | 0 | 1 |
| G 2 | | 0 | 1 | 0 | 0 | 0 | 1 | 0 | 1 |
| G 2 | 0 | 0 | 0 | 1 | 0 | 1 | 0 | 0 | 1 |
| 0 | 1 | 0 | 0 | 0 | 0 | 1 | 1 | 0 | 1 |
| 0 | | 1 | 0 | 0 | 1 | 0 | 1 | 1 | 0 |
| 0 | 0 | 0 | 1 | 1 | 0 | 1 | 1 | 0 | 0 |
| 0 | 1 | 0 | 1 | 0 | 0 | 1 | 1 | 0 | 0 |
| 0 | | 1 | 0 | 0 | 1 | 0 | 1 | 1 | 0 |
| 0 | 0 | 0 | 0 | 1 | 0 | 1 | 1 | 0 | 1 |
To fix the notation, let us call a surface S a generalized Burniat type (GBT) surface of type j if S belongs to the (uniquely determined) family number j in Tables 2.
Remark 3.9 . As shown in [BCF] GBT surfaces of type j (1 ≤ j ≤ 4) have pairwise non isomorphic fundamental groups. In particular, they belong to different connected components of the moduli space of surfaces of general type.
Remark 3.10 . Let ˆ := S ˆ X/G be the quotient model of a regular GBT surface S . According to Theorem 3.8, there are the following possibilities:
- a) ˆ = X ˆ X ν := { ( z , z 1 2 , z 3 ) ∈ T | ν L 1 1 ( z , z 1 2 , z 3 )+ ν L 2 2 ( z , z 1 2 , z 3 ) = 0 } and G = G 1 , where ν := ( ν 1 : ν 2 ) ∈ P 1 , ( ν /ν 1 2 ) = ± 1,
/negationslash
$$L _ { 2 } ( z _ { 1 }, z _ { 2 }, z _ { 3 } ) \ \coloneqq \ \mathcal { L } _ { 1 } ( z _ { 1 } ) b _ { 2 } b _ { 3 } + b _ { 1 } \mathcal { L } _ { 2 } ( z _ { 2 } ) \mathcal { L } _ { 3 } ( z _ { 3 } ) \,. \\ \intertext { N \sim \colon \, + \, \text{how to } \ \theta \ \, - \ \, / 1 \ \circ \, \circ \ 1 \ \circ \ 0 \ 1 \ \circ \ 0 \ \, \cup \ \mathcal { C } \quad \text{and to } \text{and} }$$
L 1 ( z , z 1 2 , z 3 ) := L 1 ( z 1 ) L 2 ( z 2 ) L 3 ( z 3 ) + b b b 1 2 3 and
Note that g 0 := (1 0 0 1 0 0 1 0 0) , , , , , , , , ∈ G 1 , and its fixed locus F 0 = { ( z , z 1 2 , z 3 ) ∈ T | 2 z 1 = 2 z 2 = 2 z 3 = 0 } intersects ˆ X ν if
In other words, if ν ∈ B 0 then G 1 does not act freely on ˆ X ν and therefore does not give rise to a GBT surface.
- ν ∈ B 0 := { ( L 2 ( z , z 1 2 , z 3 ) : -L 1 ( z , z 1 2 , z 3 )) , ( z , z 1 2 , z 3 ) ∈ F 0 } .
/negationslash
- b) ˆ X = ˆ X µ := { ( z , z 1 2 , z 3 ) ∈ T | L 1 ( z 1 ) L 2 ( z 2 ) L 3 ( z 3 ) = µ } , with µ ∈ C , µ = 0 and G = G 2 . Since g 0 ∈ G 2 , if µ ∈ B ′ := {L 1 ( z 1 ) L 2 ( z 2 ) L 3 ( z 3 ) , ( z , z 1 2 , z 3 ) ∈ F 0 } , then G 2 does not act freely on ˆ X µ ; moreover (see e.g. [Ino94]):
B ′ = {± 1 , ± a , i ± a a , i j ± a a a 1 2 3 } with i = j ∈ { 1 2 3 , , } .
/negationslash
We remark that this case gives rise to the family of primary Burniat surfaces (see [BC11, BC13]).
- c) ˆ X = ˆ X b := { ( z , z 1 2 , z 3 ) ∈ T | L 1 ( z 1 ) L 2 ( z 2 ) L 3 ( z 3 ) = b } , with b := b b b 1 2 3 and G = G j , j = 3 4. ,
We already remarked that ˆ X has at most finitely many nodes as singularities. The next statement shows that either ˆ X is smooth or has exactly eight nodes.
Proposition 3.11. Let S → ˆ = S ˆ X/G be a regular generalized Burniat type surface, i.e., ( ˆ X,G ) ∈ { ( ˆ X , G ν 1 ) , ( ˆ X ,G µ 2 ) , ( ˆ X , G b j ) , j = 3 4 , } . Then:
- 1) ˆ X ν is singular if and only if ν ∈ B := { ± ( b 1 : 1) , (1 : ± b 1 ) } and Sing( ˆ X ν ) = { ( z , 1 ± 1 4 , ± 1 4 ) | 2 z 1 = 0 , νb 1 + L 1 ( z 1 ) = 0 } ∪ { ( z , 1 τ 2 2 ± 1 4 , τ 3 2 ± 1 4 ) | 2 z 1 = 0 , b 1 + ν L 1 ( z 1 ) = 0 } ;
- 2) ˆ X µ is smooth;
- 3-4) ˆ X b is singular if and only if b := b b b 1 2 3 ∈ B ′ and Sing( ˆ X b ) = { ( z , z 1 2 , z 3 ) ∈ ˆ X b | 2 z 1 = 2 z 2 = 2 z 3 = 0 } .
In particular, either ˆ X is smooth or its singular locus consists of exactly 8 nodes.
Proof. We start with cases 2) and 3-4).
Let f ( z , z 1 2 , z 3 ) := L 1 ( z 1 ) L 2 ( z 2 ) L 3 ( z 3 ). It is easy to see that since df dz i = d L i dz i L i +1 L i +2 (the indices i ∈ { 1 2 3 , , } have to be considered
- Sing( ˆ X µ ) = { z ∈ ˆ X µ | ∇ f ( z ) = 0 } = { ( z , z 1 2 , z 3 ) ∈ ˆ X µ | 2 z 1 = 2 z 2 = 2 z 3 = 0 }
mod 3). We observe that Sing( ˆ X µ ) = ˆ X µ ∩ F 0 , being F 0 the fixed locus of g 0 := (1 , 0 0 1 0 0 1 0 0). , , , , , , ,
/negationslash
Since g 0 ∈ G 2 , we get that the surfaces of case 2) are smooth. In case 3-4), µ := b = b b b 1 2 3 and ˆ X b is singular if and only if b ∈ B ′ = { f ( z ) | z ∈ F 0 } ; we have to determine the number of nodes. We note that if ( z , z 1 2 , z 3 ) ∈ F 0 then f ( z , z 1 2 , z 3 ) = f ( z 1 + 1 2 , z 2 + 1 2 , z 3 ) = f ( z , z 1 2 + 1 2 , z 3 + 1 2 ) = f ( z 1 + 1 2 , z 2 , z 3 + 1 2 ), and all these points are in F 0 . Since a j = b 2 j = ± 1 and ˆ X b is invariant under ( ∗ , 1 , ∗ ∗ , , 1 , ∗ ∗ , , 1 , ∗ ), it is easy to show that b = ± 1 iff b = ± a a a 1 2 3 and that b = ± a i iff b = ± a i +1 a i +2 (indices have to be considered mod 3); in particular, | B ′ | = 8. Since | F 0 | = 64, it follows that for any choice of b ∈ B ′ exactly 8 points of F 0 belong to ˆ , i.e. X b Sing( ˆ X b ) consist of exactly 8 nodes.
In case 1), let f ν ( z , z 1 2 , z 3 ) := ν L 1 1 ( z , z 1 2 , z 3 ) + ν L 2 2 ( z , z 1 2 , z 3 ). We note that if ν 1 = 0 or ν 2 = 0, arguing as above we get that ˆ X ν is smooth ( g 0 ∈ G 1 ), so we may assume ν 2 = 1 and ν := ν 1 ∈ C \ { 0 } . We recall that i has poles in z i = τ i 1 and zeroes in z i = 1 .
/negationslash
L 2 ± 4 ± 4 τ 1
We start considering charts such that z i = i 2 ± 4 for i = 1 2 3. , , It is easy to see that
$$& \dots \nolimits _ { J } \, \ll \, \dots$$
If ν L 1 + b 1 = 0 for a point in ˆ , X then
$$f _ { \nu } = \mathcal { L } _ { 2 } \mathcal { L$$
ˆ
/negationslash
/negationslash hence ν = ± 1, i.e. X is not irreducible, a contradiction (see Remark 3.2); analogously we can assume ν L L 2 3 + b b 2 3 = 0. Since L i and L ′ i have no common zeroes, L ′ 2 = 0 if and only if L ′ 3 = 0; in this case the solutions of ∇ f ν = 0 are points in F 0 , the fixed locus of g 0 ∈ G 1 . Therefore ( z , z 1 2 , z 3 ) ∈ Sing( ˆ ) if X and only if it satisfies the following equations: L 2 ( z 2 ) = L 3 ( z 3 ) = 0, L ′ 1 ( z 1 ) = 0 and f ν = νb 1 + L 1 ( z 1 ) = 0. It is immediate to see that the last two equations have common solutions if and only if ν ∈ B = {± b , 1 ± b -1 1 } ; if ν ∈ B , we find 4 nodes: z 1 = L -1 1 ( -νb 1 ), z 2 ∈ {± } 1 4 and z 3 ∈ {± 1 4 } .
We now consider charts such that z 2 = ± 1 4 and z 3 = ± 1 4 then the affine equation f ν = 0 can be written as follows:
/negationslash
$$f _ { \nu } = & \ \overline { \mathcal { L } } _ {$$
being L i := L -1 i , i = 2 3. Arguing as above, one gets that ( , z , z 1 2 , z 3 ) ∈ Sing( ˆ ) X if and only if it satisfies the following equations: L 2 ( z 2 ) = L 3 ( z 3 ) = 0, L ′ 1 ( z 1 ) = 0 and f ν = ν L 1 ( z 1 ) + b 1 = 0. The last two equations have common solutions if and only if ν ∈ B = {± b , 1 ± b -1 1 } ; if ν ∈ B , we find other 4 nodes, namely: z 1 = L -1 1 ( -b 1 ν ), z 2 ∈ { τ 2 2 ± } 1 4 and z 3 ∈ { τ 3 2 ± 1 4 } .
Considering the other charts, one finds either no singular points, or four of the eight nodes we found.
/square
Corollary 3.12. Let S → ˆ := S ˆ X/G be a generalized Burniat surface. Then either ˆ S is smooth or its singular locus is given by exactly one node.
Proof. We simply note that G ∼ = ( Z / 2 Z ) 3 acts freely on ˆ ; in particular X it acts transitively on the set of nodes. /square
## 4. the main result
In this section we give a prove (using Corollary 1.5 and Proposition 2.2) that Bloch's conjecture holds for regular generalized Burniat type surfaces.
Remark 4.1 . Let S → ˆ := S ˆ X/G be a GBT and let γ : ˆ X → ˆ := S ˆ X/G be the projection onto the quotient. Let σ be an involution on ˆ , X it defines an involution σ on ˆ : S σ γ x ( ( )) := γ σ x ( ( )) and
$$F i x ( \overline { \sigma } ) = \bigcup _ { g \in$$
being Fix ˆ X ( σ ) := Fix σ ( ) ∩ ˆ . X Moreover, σ lifts to an involution σ ′ := /epsilon1 -1 ◦ σ ◦ /epsilon1 on S .
Lemma 4.2. Let S → ˆ = S ˆ X/G be a regular generalized Burniat type surface with ( ˆ X,G ) = ( ˆ X , G ν 1 ) . Let g ∈ G ′ 1 be an element fixing point on X ν , then its fixed locus Fix g ( ) on ˆ X ν is as in Table 3.
Generalized Burniat type surfaces are constructed considering G ∼ = ( Z / 2 Z ) 3 /triangleleft G 0 acting freely on a Burniat hypersurface ˆ X ⊂ T , hence it is natural to consider involutions in G 0 \ G . We start determining the fixed locus of elements in G 0 .
| | Fix ( g i ) , X ν smooth | Fix ( g i ) , X ν singular |
|-------------------------------------------------------------------------------------------------------------------------------------|----------------------------|------------------------------|
| g 1 := (0 , 0 , 0 , 0 , 0 , 0 , 1 , 0 , 0) g 2 := (0 , 0 , 0 , 1 , 0 , 0 , 0 , 0 , 0) g 3 := (1 , 0 , 0 , 0 , 0 , 0 , 0 , 0 , 0) | 4 genus 5 curves | 4 genus 5 curves Γ |
| g 4 := (0 , 0 , 0 , 1 , 0 , 0 , 1 , 0 , 0) g 5 := (1 , 0 , 0 , 0 , 0 , 0 , 1 , 0 , 0) | 32 pt | 32 pt |
| g 7 := (0 , 0 , 0 , 0 , 0 , 1 , 0 , 0 , 1) | 16 pt, 8 ell. curves | 8 nodes, 8 ell. curves |
| g 11 := (1 , 0 , 0 , 0 , 0 , 1 , 0 , 0 , 1) g 14 := (0 , 1 , 0 , 0 , 1 , 0 , 0 , 1 , 0) g 15 := (0 , 1 , 0 , 1 , 1 , 1 , 1 , 1 , 1) | 32 pt | 32 pt, 8 nodes |
| | | 32 pt |
Table 3.
In Table 3, Γ denotes the disjoint union Γ := C 1 /unionsq C 2 /unionsq D , being C i ( i = 1 2 , ) a genus 5 curve and D the union of 8 elliptic curves each one passing through exactly two nodes and such that a point belongs to two of them if and only if it is a node.
Proof. By Proposition 3.11, ˆ X ν is singular if and only if ν ∈ B := { ± ( b 1 : 1) , (1 : ± b 1 ) } ; in this case the eight nodes on ˆ X ν are fixed by g 11 := (1 , 0 0 0 0 1 0 0 1). , , , , , , , g 1 : since Fix g ( 1 ) ∩ Fix g ( 11 ) = ∅ , the fixed locus of g 1 is independent from ν . It fixes the points ( z , z 1 2 , z 3 ) with 2 z 3 = 0: z 3 ∈ { 0 , 1 2 , τ 3 2 , τ 3 +1 2 } , hence it cuts on ˆ X ν four disjoint curves of genus 5: each one is given by an equation of multidegree (2,2) in E 1 × E 2 .
g 2 : this case is analogous to the previous one.
g 3 : it fixes the points with 2 z 1 = 0 that is z 1 ∈ V := { 0 , 1 2 , τ 1 2 , τ 1 +1 2 } .
If ˆ X ν is singular ( ν ∈ B ), we rewrite the equation of ˆ X ν as follows:
If ˆ X ν is smooth, this case is analogous to g 1 : we get four disjoint smooth curves of genus 5 on ˆ X ν .
$$\begin{array} {$$
For a fixed value ν ∈ B , there exists a unique z 1 ∈ V such that νb 1 + L 1 ( z 1 ) = 0, and the equation of ˆ X ν is satisfied if and only if L 2 ( z 2 ) = 0 or L 3 ( z 3 ) = 0. We get four elliptic curves on ˆ X ν fixed by g 3 : ( z , 1 ± 1 4 , z 3 ), ( z , z 1 2 , ± 1 4 ). Analogously, considering the element z ′ 1 := z 1 + τ 1 2 ∈ V we get other four elliptic curves ( z , ′ 1 τ 1 2 ± 1 4 , z 3 ) and ( z , z ′ 1 2 , τ 1 2 ± 1 4 ) on ˆ X ν fixed by g 3 . We observe that that a point belongs to two of these eight curves if and only if it is a node.
Considering z 1 ∈ V \ { z , z 1 ′ 1 } , we get two disjoint curves of genus 5: both given by an equation of multidegree (2,2) in E 2 × E 3 .
/negationslash since Fix g ( 4 ) ∩ Fix g ( 11 ) = ∅ , the fixed locus of g 4 is independent from ν . It fixes the points ( z , z 1 2 , z 3 ) with 2 z 2 = 2 z 3 = 0 and 2 z 1 = 0, since g 0 := (1 0 0 1 0 0 1 0 0) , , , , , , , , ∈ G 1 has no fixed points: for any pair ( z , z 2 3 ), the equation defining ˆ X ν has two distinct solutions, whence g 4 fixes 32 points on ˆ X ν .
these cases are analogous to g 4 .
The involution g 7 fixes the points ( z , z 1 2 , z 3 ) with 2 z 2 = 1 2 , 2 z 3 = 1 2 . Let ( L i ( z i ) 0 : L i ( z i ) 1 ) be the homogeneous coordinates of the point L i ( z i ). The equation of ˆ X ν is then
ν [ L 1 ( z 1 ) 0 L 2 ( z 2 ) 0 L 3 ( z 3 ) 0 + b b b 1 2 3 L 1 ( z 1 ) 1 L 2 ( z 2 ) 1 L 3 ( z 3 ) 1 ]+
[ b b 2 3 L 1 ( z 1 ) 0 L 2 ( z 2 ) 1 L 3 ( z 3 ) 1 + b 1 L 2 ( z 2 ) 0 L 3 ( z 3 ) 0 L 1 ( z 1 ) 1 ] = 0 .
It follows easily from the properties of the Legendre L -function that ( L i ( 1 4 ) 0 : L i ( 1 4 ) ) 1 = (0 : 1) , ( L i ( 1 4 + τ i 2 ) 0 : L i ( 1 4 + τ i 2 ) ) 1 = (1 : 0) . If z 2 = 1 and z 3 = 1 + τ 3 or z 2 = 1 + τ 2 and z 3 = 1 , then the
± 4 ± 4 2 ± 4 2 ± 4
g
4
:
g 5 -g 6 :
g
7
:
g
11
:
An analogous argument holds if z 2 = ± 1 4 + τ 2 2 and z 3 = ± 1 4 + τ 3 2 .
equation is satisfied for any z 1 ∈ E 1 , i.e g 7 fixes 8 disjoint elliptic curves contained in the smooth locus of ˆ X ν .
If z 2 = ± 1 4 and z 3 = ± 1 4 then the equation becomes νb 1 L 1 ( z 1 ) 1 + L 1 ( z 1 ) 0 = 0 that has two solutions if ν / B ∈ (i.e. ˆ X ν is smooth) and one solution if ν ∈ B ; in other words, if ˆ X ν is smooth g 4 fixes 8 isolated points, else g 7 fixes 4 nodes. Analogously, if z 2 = ± 1 4 + τ 2 2 and z 3 = ± 1 4 + τ 3 2 and if ˆ X ν is smooth g 7 fixes other 8 isolated points, else it fixes the other 4 nodes.
we observe that Fix g ( 11 ) = Fix g ( 7 ) ∩ { 2 z 1 = 0 } and arguing as above we distinguish three cases: if z 2 = ± 1 4 and z 3 = ± 1 4 + τ 3 2 or z 2 = ± 1 4 + τ 2 2 and z 3 = ± 1 4 , then the equation of ˆ X ν is satisfied for any z 1 ∈ E 1 , but 2 z 1 = 0 hence g 11 fixes 32 smooth points on ˆ X ν . If z 2 = ± 1 4 and z 3 = ± 1 4 then the equation of ˆ X ν is νb 1 L 1 ( z 1 ) 1 + L 1 ( z 1 ) 0 = 0, since 2 z 1 = 0 we get no solution if ˆ X ν is smooth ( ν / B ∈ ) and one solution (a node) if ˆ X ν is singular ( ν ∈ B ).
Therefore, if ˆ X ν is smooth g 11 fixes 32 isolated points, else g 11 fixes 32 smooth isolated points and 8 nodes.
g 14 : since Fix g ( 14 ) ∩ Fix g ( 11 ) = ∅ , the fixed locus of g 14 is independent from ν . It fixes 64 points on E 1 × E 2 × E 3 , namely
Observe that L k ( ± τ k 4 ) = b k and L k ( ± τ k 4 + ) = 1 2 -b k . It is a straightforward computation to show that exactly 32 of them lie on ˆ X ν . g : since Fix g ( ) ∩ Fix g ( ) = ∅ , the fixed locus of g is independent
$$z \in \left \{ \frac { 1 } { 4 } \left ( \begin{array} { c } \pm \tau _ { 1 } \\ \pm \tau _ { 2 } \\ \pm \tau _ { 3 } \end{array} \right ) + \frac { 1 } { 2 } ( \mathbb { Z } / 2 \mathbb { Z } ) ^ { 3 } \right \} \,. \\ \text{hat $\mathcal{L}_{k}(\pm \frac{t_{k} } { 4 } ) = b_{k}$ and $\mathcal{L}_{k}(\pm \frac{t_{k} } { 4 } + \frac { 1 } { 2 } ) = - b_{k}$. It} \\ \text{and $\mathcal{L}_{k}(\pm \frac{t_{k} } { 4 } ) = b_{k}$ and $\mathcal{L}_{k}(\pm \frac{t_{k} } { 4 } + \frac { 1 } { 2 } ) = - b_{k}$. It}$$
15 15 11 15 from ν . It fixes 64 points on E 1 × E 2 × E 3 , namely
Observe now that
$$\nu \, \infty \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varp \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varphi \, \varprod \, \intertext { r e w o n w o t a t } / \, 1 \quad - \, \cdots \, ^ { 2 } / \, 1 \quad - \, \cdots \, ^ { 1 \cdots \, ^ { 2 } }$$
$$\mathcal { L } _ { k } \left ( \frac { 1 } { 4 } + \frac { \tau _ { k } } { 4 } \right ) ^ { 2 } = \mathcal { L } _ { k } \left ( \frac { 1 } { 4 } + \frac { \tau _ { k } } { 4 } + \frac { 1 } { 2 } \right ) ^ { 2 } = - a _ { k } \,,$$
whence {L k ( 1 4 + τ k 4 ) , L i k ( 1 4 + τ k 4 + 1 2 ) } = { √ -1 b k , -√ -1 b k } . It is a straightforward computation to show that exactly 32 of them lie on ˆ X ν .
/square
Lemma 4.3. Let S → ˆ = S ˆ X/G be a regular generalized Burniat type surface with ( ˆ X,G ) = ( ˆ X , G µ 2 ) . Let g ∈ G 1 be an element fixing point on X µ , then its fixed locus Fix g ( ) on ˆ X µ is as in Table 4.
Proof. Noting that g 0 = (1 0 0 1 0 0 1 0 0) , , , , , , , , ∈ G 2 , the same arguments of the proof of Lemma 4.2 hold, and the statement follows.
| | | Fix ( g i ) , X µ |
|------------|-------------------------------------------------------------------------------|----------------------|
| g 1 := | (0 , 0 , 0 , 0 , 0 , 0 , 1 , 0 , 0) | 4 genus 5 curves |
| g 2 := | (0 , 0 , 0 , 1 , 0 , 0 , 0 , 0 , 0) | 4 genus 5 curves |
| g 3 | := (1 , 0 , 0 , 0 , 0 , 0 , 0 , 0 , 0) | 4 genus 5 curves |
| g 4 | := (0 , 0 , 0 , 1 , 0 , 0 , 1 , 0 , 0) | 32 pt |
| g 5 := g 6 | (1 , 0 , 0 , 0 , 0 , 0 , 1 , 0 , 0) := (1 , 0 , 0 , 1 , 0 , 0 , 0 , 0 , 0) | 32 pt |
| g 7 | := (0 , 0 , 0 , 0 , 0 , 1 , 0 , 0 , 1) | 16 pt, 8 ell. curves |
| g 8 g | := (0 , 0 , 1 , 0 , 0 , 0 , 0 , 0 , 1) | 16 pt, 8 ell. curves |
| 9 g 11 | := (0 , 0 , 1 , 0 , 0 , 1 , 0 , 0 , 0) := (1 , 0 , 0 , 0 , 0 , 1 , 0 , 0 , 1) | 32 pt |
| g 12 | := (0 , 0 , 1 , 1 , 0 , 0 , 0 , 0 , 1) | 32 pt |
| g 13 | := (0 , 0 , 1 , 0 , 0 , 1 , 1 , 0 , 0) | 32 pt |
/square
Finally, we consider the case ˆ = X ˆ ; we study the fixed locus only X b of the elements in G 0 having fixed locus of dimension one on T , since it is enough for our purposes.
Lemma 4.4. Let S → ˆ = S ˆ X/G be a regular generalized Burniat type surface with ( ˆ X,G ) = ( ˆ X , G b j ) , j ∈ { 3 4 , } . Let g ∈ G 0 be an element having fixed locus of dimension one on T then its fixed locus Fix g ( ) on ˆ X b is as in Table 5.
| | Fix ( g i ) , X b smooth | Fix ( g i ) , X b singular |
|----------------------------------------------------------------------------------------------------------------------------------|-------------------------------|------------------------------|
| g 4 := (0 , 0 , 0 , 1 , 0 , 0 , 1 , 0 , 0) g 5 := (1 , 0 , 0 , 0 , 0 , 0 , 1 , 0 , 0) g 6 := (1 , 0 , 0 , 1 , 0 , 0 , 0 , 0 , 0) | 32 pt | 16 pt, 8 nodes |
| g 7 := (0 , 0 , 0 , 0 , 0 , 1 , 0 , 0 , 1) g 8 := (0 , 0 , 1 , 0 , 0 , 0 , 0 , 0 , 1) g 9 := (0 , 0 , 1 , 0 , 0 , 1 , 0 , 0 , 0) | 16 pt, 8 ell. curves Table 5. | 8 nodes, 8 ell. curves |
/negationslash
Proof. By Proposition 3.11, ˆ X b is singular if and only if b ∈ B ′ = {± 1 , ± a , i ± a a , i j ± a a a 1 2 3 } , with i = j ∈ { 1 2 3 , , } ; in this case the eight nodes on ˆ X b are fixed by g 0 = (1 0 0 1 0 0 1 0 0). , , , , , , , , g 4 : it fixes the points ( z , z 1 2 , z 3 ) with 2 z 2 = 2 z 3 = 0. If b / B ∈ ′ , for every pair ( z , z 2 3 ), 2 z 1 = 0, whence g 4 fixes 32 points on ˆ . X b
If b ∈ B ′ , for 8 choices of ( z , z 2 3 ) there are two values of z 1 verifying the equation of ˆ , while for the other 8 possibilities there is a unique X b
/negationslash
g
5
-
g
6
:
g
7
:
g 8 -g 9 :
value of z 1 verifying the equation of ˆ , whence X b Fix g ( 4 ) is given by 16 smooth points and the 8 nodes of ˆ . X b these cases are analogous to g 4 .
since Fix g ( 7 ) ∩ Fix g ( 0 ) = ∅ , the fixed locus of g 7 is independent from b .
The involution g 7 fixes the points ( z , z 1 2 , z 3 ) with 2 z 2 = 1 2 , 2 z 3 = 1 2 . Let ( L i ( z i ) 0 : L i ( z i ) 1 ) be the homogeneous coordinates of the point L i ( z i ). The equation of ˆ X b is then
It follows easily from the properties of the Legendre L -function that ( L i ( 1 4 ) 0 : L i ( 1 4 ) ) 1 = (0 : 1) , ( L i ( 1 4 + τ i 2 ) 0 : L i ( 1 4 + τ i 2 ) ) 1 = (1 : 0) . If z 2 = ± 1 4 and z 3 = ± 1 4 + τ 3 2 or z 2 = ± 1 4 + τ 2 2 and z 3 = ± 1 4 , then the equation is satisfied for any z 1 ∈ E 1 , i.e g 7 fixes 8 disjoint elliptic curves.
$$\mathcal { L } _ { 1 } ( z _ { 1 } ) _ { 0 } \mathcal { L } _ { 2 } ( z _ { 2 } ) _ { 0 } \mathcal { L } _ { 3 } ( z _ { 3 } ) _ { 0 } = b _ { 1 } b _ { 2 } b _ { 3 } \mathcal { L } _ { 1 } ( z _ { 1 } ) _ { 1 } \mathcal { L } _ { 2 } ( z _ { 2 } ) _ { 1 } \mathcal { L } _ { 3 } ( z _ { 3 } ) _ { 1 } \,. \\ \tau + \epsilon + \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \alpha \,.$$
If z 2 = ± 1 4 and z 3 = ± 1 4 then the equation becomes L 1 ( z 1 ) 1 = 0 that has two solutions. If z 2 = ± 1 4 + τ 2 2 and z 3 = ± 1 4 + τ 3 2 then the equation becomes L 1 ( z 1 ) 0 = 0 that has two solutions, whence g 7 fixes 16 isolated points and 8 disjoint elliptic curves on ˆ . X b these cases are analogous to g 7 .
/square
We are now ready to prove our main result:
Theorem 4.5. Let /epsilon1 : S → ˆ = S ˆ X/G be a regular generalized Burniat type surface: ( ˆ X,G ) ∈ { ( ˆ X , G ν 1 ) , ( ˆ X , G µ 2 ) , ( ˆ X , G b j ) , j = 3 4 , } . Then it verifies the Bloch conjecture.
We recall the following result, which allows to prove that each single surface in the moduli space corresponding to GBT surfaces of type j (1 ≤ j ≤ 4) satisfies Boch's conjecture.
## Theorem 4.6 ([BCF]) .
- i) Let S be a smooth projective surface homotopically equivalent to a GBT surface S i of type i . Then S is a GBT surface of type i , i.e. contained in the same irreducible family as S i .
- ii) The connected components N i of the Gieseker moduli space M can 1 6 , corresponding to GBT surfaces of type i is irreducible, generically smooth, normal and unirational of dimension 4 ( i = 1 2 , ) and of dimension 3 else.
Together with Theorem 4.5 we thus obtain:
Theorem 4.7. Let S be any surface such that its moduli point [ S ] ∈ N i , 1 ≤ i ≤ 4 , then S satisfies Bloch's conjecture.
Proof of Theorem 4.5. We prove the statement case by case. In each case we consider a group H ∼ = ( Z / 2 Z ) 2 < Aut( ˆ ) X which allow us
define a group H ′ ∼ = ( Z / 2 Z ) 2 < Aut( S ) satisfying the assumptions of Corollary 1.5.
- G 1 ) Let us consider the involutions σ 1 := (0 0 0 0 0 0 1 0 0) and , , , , , , , , σ 2 := (0 , 0 0 1 0 0 0 0 0) in , , , , , , , G ′ 1 .
In the coset σ G 1 1 there are four elements fixing points on ˆ X ν :
```
\begin{array} { l l } g _ { 1 } = ( 0, 0, 0, 0, 0, 0, 1, 0, 0 ) \\ g _ { 7 } = ( 0, 0, 0, 0, 1, 0, 0, 1 ) \end{array} \quad \begin{array} { l l } g _ { 6 } = ( 1, 0, 0, 1, 0, 0, 0, 0 ) \\ g _ { 1 5 } = ( 0, 1, 0, 1, 1, 1, 1, 1 ) \end{array}
```
Since g 0 := (1 , 0 0 1 0 0 1 0 0) , , , , , , , ∈ G 1 has no fixed points on ˆ X ν , the 4 sets Fix ˆ X ν ( g k ) ( k ∈ { 1 6 7 15 , , , } ) are pairwise disjoint.
g 6 and g 15 fix 32 points each on ˆ X ν and G 1 acts freely on these two sets: we get 8 points fixed by σ 1 .
g 1 fixes 4 curves of genus 5 on ˆ X ν , G 1 acts transitively on this set of curves and g 0 := (1 0 0 1 0 0 1 0 0) , , , , , , , , ∈ G 1 maps each curve onto itself, hence a genus 3 curve is fixed by σ 1 on ˆ . S
g 7 fixes 8 disjoint elliptic curves and 16 points if ˆ X ν is smooth, 8 nodes otherwise. Since G 1 acts freely on the set of points and transitively on the set of curves, we get that σ 1 fixes one elliptic curve and either 2 points or 1 node.
It follows that the involution σ 1 on ˆ S lifts to an involution σ ′ 1 on S whose fixed locus contains a genus 3 curve, an elliptic curve and 8 isolated smooth points, by Proposition 2.2, S/σ ′ 1 is not of general type.
In the coset σ G 2 1 there are three elements fixing points on ˆ X ν :
$$\begin{array} { c } g _ { 2 } = ( 0, 0, 0, 1, 0, 0, 0, 0, 0 ) \\ g _ { 1 1 } = ( 1, 0, 0, 0, 0, 1, 0, 0, 1 ) \end{array} \quad g _ { 5 } = ( 1, 0, 0, 0, 0, 1, 0, 0 )$$
The 3 sets Fix ˆ X ν ( g k ) ( k ∈ { 2 5 11 , , } ) are pairwise disjoint, since g 0 has no fixed points on ˆ X ν .
g 5 fixes 32 points on ˆ X ν : we get 4 points fixed by σ 2 .
g 2 fixes 4 curves of genus 5 on ˆ X ν , G 1 acts transitively on this set of curves and g 0 := (1 0 0 1 0 0 1 0 0) , , , , , , , , ∈ G 1 maps each curve onto itself, hence a genus 3 curve is fixed by σ 2 on ˆ . S
- If ˆ X ν is smooth g 11 fixes 32 isolated points, else it fixes 32 smooth isolated points and 8 nodes: we get that σ 2 fixes 4 smooth points and, if ˆ is singular, a node too. S
- It follows that the involution σ 2 on ˆ S lifts to an involution σ ′ 2 on S whose fixed locus contains a genus 3 curve, an elliptic curve and 8 isolated smooth points, by Proposition 2.2, S/σ ′ 2 is not of general type.
Let σ 3 := σ 1 + σ 2 . In the coset σ G 3 1 there are three elements fixing points on ˆ X ν :
```
\begin{array} { c } g _ { 3 } = ( 1, 0, 0, 0, 0, 0, 0, 0, 0 ) \\ g _ { 1 4 } = ( 0, 1, 0, 0, 1, 0, 0, 1, 0 ) \end{array} \quad g _ { 4 } = ( 0, 0, 0, 1, 0, 0, 1, 0, 0 )
```
The 3 sets Fix ˆ X ν ( g k ) ( k ∈ { 3 4 14 , , } ) are pairwise disjoint, since g 0 has no fixed points on ˆ X ν .
If ˆ X ν is smooth, g 3 fixes four disjoint smooth curves of genus 5 on ˆ X ν . Let H := 〈 (1 , 0 0 1 0 0 1 0 0) (0 0 0 0 0 1 1 0 1) , , , , , , , , , , , , , , , , 〉 /triangleleft G 1 , each curve is invariant under H , hence σ 3 fixes two disjoint genus 2 curves.
g 4 and g 14 fix 32 points each on ˆ X ν and G 1 acts freely on these two sets: we get 8 points fixed by σ 3 .
If ˆ X ν is singular, g 3 fixes two disjoint smooth curves of genus 5 and 8 elliptic curves such that a point belongs to two of them if and only if it is a node. Looking at the G 1 action on this configuration of curves, one can easily prove that σ 3 fixes on ˆ two elliptic curves intersecting S in a node and a genus 2 curve.
It follows that the involution σ 3 on ˆ lifts to an involution S σ ′ 3 on S whose fixed locus contains a genus 2 curve and 8 isolated smooth points, by Proposition 2.2, S/σ ′ 3 is not of general type.
Applying Corollary 1.5, with ( Z / 2 Z ) 2 = 〈 σ , σ 1 2 〉 , we conclude that S verifies Bloch's conjecture.
G 2 ) Let us consider the involutions σ 4 := (1 0 0 0 0 0 0 0 0) and , , , , , , , , σ 5 := (0 , 0 0 1 0 0 0 0 0) in , , , , , , , G 1 .
In the coset σ G 4 2 , there are four elements fixing points on ˆ X µ :
```
\begin{array} { l l } g _ { 3 } = ( 1, 0, 0, 0, 0, 0, 0, 0 ) \\ g _ { 9 } = ( 0, 0, 1, 0, 0, 1, 0, 0, 0 ) \end{array} \quad \begin{array} { l l } g _ { 4 } = ( 0, 0, 0, 1, 0, 0, 1, 0, 0 ) \\ g _ { 1 2 } = ( 0, 0, 1, 1, 0, 0, 0, 0, 1 ) \end{array}
```
Since g 0 := (1 0 0 1 0 0 1 0 0) , , , , , , , , ∈ G 2 has no fixed points, the 4 sets Fix ˆ X µ ( l j ) ( j ∈ { 3 4 9 12 , , , } ) are pairwise disjoint.
g 4 and g 12 fix 32 points each on ˆ X ν and G 2 acts freely on these two sets: we get 8 points fixed by σ 4 .
g 3 fixes four genus 5 curves that are invariant under the subgroup 〈 (0 , 0 0 1 0 1 0 0 1) (1 0 0 1 0 0 1 0 0) , , , , , , , , , , , , , , , , 〉 /triangleleft G 2 , hence σ ′ 4 fixes two disjoint genus 2 curves.
g 9 fixes 16 isolated points and 8 disjoint elliptic curves on ˆ X µ , each curve is invariant under (1 0 1 0 0 1 0 0 0) , , , , , , , , ∈ G 2 . We get that σ ′ 4 fixes 2 isolated points and two disjoint elliptic curves.
It follows that Fix σ ( ′ 4 ) is given by 10 isolated fixed points, two genus 2 curves and two elliptic curves. By Proposition 2.2, the quotient S/σ ′ 4 is not of general type.
The same argument shows that S/σ ′ 4 and S/ σ ( 4 + σ 5 ) ′ are not of general type, whence S verifies Bloch's conjecture, thanks to Corollary 1.5.
G 3 ) Let us consider the involutions σ 6 := (1 0 0 1 0 0 0 0 0) and , , , , , , , , σ 7 := (1 , 0 0 0 0 0 1 0 0). , , , , , , , In the coset σ G 6 3 , there are two elements
fixing points on ˆ : X
$$g _ { 6 } = ( 1, 0, 0, 1, 0, 0, 0, 0, 0 ) \quad g _ { 8 } = ( 0, 0, 1, 0, 0, 0, 0, 0, 1 )$$
The 2 sets Fix ˆ X b ( h j ) ( j ∈ { 6 8 , } ) are disjoint.
g 6 fixes 32 points on ˆ X b if it is smooth, 16 smooth points and 8 nodes otherwise. Since G 3 acts freely on this set of points we get that σ 6 fixes either 4 points ( ˆ X b smooth) or 2 smooth points and the node.
g 8 fixes 16 isolated smooth points and 8 elliptic curves.
In the smooth case Fix σ ( ′ 6 ) is given by 6 isolated fixed points and an elliptic curve. By Proposition 2.2, the quotient S/σ ′ 6 is not of general type.
In the singular case, σ 6 fixes the node p , 4 smooth points and an elliptic curve. Let Γ := /epsilon1 -1 ( p ), the involution σ 6 lifts to an involution σ ′ 6 on S such that σ ′ 6 (Γ) = Γ = ∼ P 1 and it fixes an elliptic curve, 4 smooth points and either two isolated points on Γ or Γ: ( -2)-curve. In both cases, by Proposition 2.2, S/σ ′ 6 is not of general type.
The same argument shows that S/σ ′ 7 and S/ σ ( 6 + σ 7 ) ′ are not of general type, whence S verifies Bloch's conjecture, thanks to Corollary 1.5.
- G 4 ) Considering involutions (1 , 0 0 1 0 0 0 0 0) and (1 0 0 0 0 0 1 0 0), , , , , , , , , , , , , , , , this case is analogous to the G 3 -case.
/square
## References
[Bar85] R. Barlow. Rational equivalence of zero cycles for some more surfaces with p g = 0. Invent. Math., 79(2):303-308, 1985.
[Bau14] I. Bauer. Bloch's conjecture for Inoue surfaces with p g = 0, K 2 = 7. Proc. Amer. Math. Soc., 2014.
I. Bauer and F. Catanese. Burniat surfaces I: fundamental groups and moduli of primary Burniat surfaces. In Classification of algebraic varieties, EMS Ser. Congr. Rep., pages 49-76. Eur. Math. Soc., Z¨rich, u
[BC11] 2011.
[BC12] I. Bauer and F. Catanese. Inoue type manifolds and Inoue surfaces: a connected component of the moduli space of surfaces with K 2 = 7, p g = 0. In Geometry and arithmetic, EMS Ser. Congr. Rep., pages 23-56. Eur. Math. Soc., Z¨rich, 2012. u
I. Bauer and F. Catanese. Burniat-type surfaces and a new family of surfaces with p g = 0, K 2 = 3. Rend. Circ. Mat. Palermo (2), 62(1):37-
[BC13] 60, 2013.
[BCF] I. Bauer, F. Catanese, and D. Frapporti. Generalized Burniat-type surfaces. Preprint, 2014.
[BCGP12] I. Bauer, F. Catanese, F. Grunewald, and R. Pignatelli. Quotients of products of curves, new surfaces with p g = 0 and their fundamental groups. American Journal of Mathematics, 134(4):993-1049, 2012.
S. Bloch, A. Kas, and D. Lieberman. Zero cycles on surfaces with p g = 0.
[BKL76] Compositio Math., 33(2):135-145, 1976.
[Blo75] S. Bloch. K 2 of Artinian Q -algebras, with application to algebraic cycles. Comm. Algebra, 3:405-428, 1975.
[Bom73]
E. Bombieri. Canonical models of surfaces of general type. Inst. Hautes ´ tudes Sci. Publ. Math., 42:171-219, 1973. E
[Bur66]
P. Burniat. Sur les surfaces de genre P 12 > 1. Ann. Mat. Pura Appl., 71(4):1-24, 1966.
[CC13]
Tsz On Mario Chan and Stephen Coughlan. Kulikov surfaces form a con- nected component of the moduli space. Nagoya Mathematical Journal,
210:1-27, 06 2013.
[Hor75]
E. Horikawa. On deformations of quintic surfaces. Invent. Math., 31(1):43-85, 1975.
[IM79] g
H. Inose and M. Mizukami. Rational equivalence of 0-cycles on some surfaces of general type with p = 0. Math. Ann., 244(3):205-217, 1979.
[Ino94] M. Inoue. Some new surfaces of general type. Tokyo J. Math., 17(2):295-
319, 1994.
[Keu88]
J.H. Keum. Somr new surfaces of general type with p g = 0. Unpublished manuscript, 1988.
[Kim05]
Shun-Ichi Kimura. Chow groups are finite dimensional, in some sense. Math. Ann., 331(1):173-201, 2005.
[Voi92]
Claire Voisin. Sur les z´ro-cycles de certaines hypersurfaces munies d'un e automorphisme. Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4), 19(4):473- 492, 1992.
Mathematisches Institut der Universit¨t Bayreuth, NW II, Univera sit¨tsstr. a 30, 95447 Bayreuth
| null |
[
"Ingrid Bauer",
"Davide Frapporti"
] |
2014-08-04T18:28:00+00:00
|
2014-08-04T18:28:00+00:00
|
[
"math.AG"
] |
Bloch's conjecture for Generalized Burniat Type surfaces with $p_g=0$
|
The aim of this article is to prove Bloch's conjecture, asserting that the
group of rational equivalence classes of zero cycles of degree 0 is trivial for
surfaces with geometric genus zero, for regular generalized Burniat type
surfaces. The technique is the method of "enough automorphisms" introduced by
Inose-Mizukami in a simplified version due to the first author.
|
1408.0761v1
|
Nuclear Physics A
Nuclear Physics A 00 (2014) 1-4
## Understanding the p /π ratio at LHC due to QCD mass spectrum
## Jacquelyn Noronha-Hostler a , Carsten Greiner b
a Instituto de F´ ısica, Universidade de S˜o Paulo, C.P . 66318, 05315-970 S˜o Paulo, SP, Brazil a a b Institut f¨r Theoretische Physik, Goethe Universit¨t, F rankfurt, Germany u a
## Abstract
Thermal fits have consistently reproduced the experimental particles yields of heavy ion collisions, however, the proton to pion ratio from ALICE Pb + Pb √ sNN = 2 76 TeV is over-predicted by thermal models- known at the . p /π puzzle. Here we test the relevance of the extended mass spectrum, i.e., include Hagedorn states (resonances that follow an exponential mass spectrum and have very short life times) on the p /π puzzle. We find that the extended mass spectrum is able to reproduce particle ratios at both RHIC and the LHC as well as being able to match the lower p /π ratio at the LHC through dynamical chemical equilibration.
Keywords: hadron resonance gas, relativistic heavy-ion collisions, extended mass spectrum, Hagedorn states, dynamical hadronic interations, thermal fits
## 1. Introduction
Final state particle ratios and yields have been matched using thermal fit models [1, 2] in order to determine the chemical freeze-out temperature and baryonic chemical potential to aid in describing the nuclear phase diagram [3]. Heavy ion experiments such as RHIC and SPS have found very precise matching to thermal models. Using previous results, predictions were then made for the LHC [4]. However, recent results at ALICE Pb + Pb √ sNN = 2 76 TeV . for the LHC have proven to be di ffi cult to fit by thermal models and the thermal models consistently overpredict the proton to pion ratio, p /π - known as the p /π puzzle [5].
Various attempts have been made in understanding this puzzle [6] and an overview of these attempts can be found in [7]. However, these attempts have not considered the e ff ects of the extended mass spectrum (exponentially increasing mass spectrum of yet to be measured resonances with very short lifetimes) and or multi-mesonic reactions. / Evidence for the extended mass spectrum is found through the exponential behavior of the known hadrons [8]. Previous work on the extended mass spectrum has found that they play a large role in the context of heavy ion collisions. Most significantly, the extended mass spectrum decreases the shear viscosity to entropy density ratio in the hadron gas phase close to the AdS CFT limit near to the critical temperature [9, 10, 11]. The extended mass spectrum is also able / to extend the region of the thermodynamical quantities that we are able to match with lattice QCD [12] from around T = 130 -140 MeV (for the standard hadron resonance gas) up to T ≈ 155 MeV when the added resonances are present [10, 13]. Elliptical flow is also suppressed in hybrid model calculations due to the extended mass spectrum when switching temperatures of TSW ≈ 155 MeV or higher are used [14]. Other e ff ects of the extended mass spectrum have been found: an improvement of thermal fits [2], the phase change order [15], and cumulants and correlations of charge fluctuations [16].
Here we include the e ff ects of the extended mass spectrum in order to explain the p /π ratio at the LHC. Dynamical chemical reactions are used that are catalyzed by quickly decaying non-strange, mesonic Hagedorn states [17]. We
Figure 1. Comparison of the results of our extended mass spectrum model particle ratio calculations vs. experimental data points of both PHENIX and STAR at Au + Au RHIC √ sNN = 200 GeV. The
find that adding in the e ff ects of the extended mass spectrum using out-of-chemical equilibrium dynamics can, indeed, explain the lower p /π ratio at the LHC.
## 2. Setup
The standard modeling of heavy-ion collisions has the following pattern: initial conditions until a time of τ 0 ≈ 0 5 . -1 fm, relativistic hydrodynamics is then initiated and allowed to expand and cool until the fluid cells reach a switching temperature of TSW ≈ 155 MeV, once the fluid is converted into particles it is described by either a hadronic afterburner or hadronic transport model until chemical and kinetic freezeout is reached. For simplicity's sake, we begin a Bjorken expansion with an accelerating radial flow that begins at τ 0 using
$$V ( \tau ) = \pi \, \tau \left ( r _ { 0 } + v _ { 0 } ( \tau - \tau _ { 0 } ) + \frac { 1 } { 2 } a _ { 0 } ( \tau - \tau _ { 0 } ) ^ { 2 } \right ) ^ { 2 }$$
to describe the relativistic hydrodynamical expansion. Here we use the initial radius size of r 0 = 7 1 . f m , and initial flow v 0 = 0 for both RHIC and LHC, whereas initial time is τ 0 = 0 6 and 1 0 fm and the acceleration is . . a 0 = 0 03 and . 0 02 . f m -1 for LHC and RHIC, respectively. We ensure that causality is preserved and the final velocity is reasonable ( vfinal ≈ 0 5 . -0 7). . We assume TSW = 155 MeV for both LHC and RHIC, which corresponds to the temperature region where we expect the extended mass spectrum to be valid [10, 13, 14].
After the switching temperature, we populate the hadrons using multihadronic decay reactions driven through Hagedorn states. To do so we must first establish the form of the mass spectrum, which is used to calculate the thermodynamics and chemical equilibrium values of the hadrons. In this proceedings we consider only the simplest form of the Hagedorn spectrum
$$\rho = A e ^ { m / T _ { H } }$$
where A = 2 84(1 . / GeV ) and TH = 0 252 GeV. However, in [18] we explore di . ff erent descriptions of the extended mass spectrum, which also describe di ff erent types of decays. The Hagedorn state decays that catalyze the other hadrons to quickly reach chemical equilibrium are
$$n \pi \ \leftrightarrow \ H S \, \leftrightarrow \, n ^ { \prime } \pi + X \bar { X }.$$
where a Hagedorn state can either decay into multiple n pions or n ′ pions plus an XX ¯ where XX ¯ = pp KK ¯, ¯ , or ΛΛ ¯ . The exact rate equations used to describe these decays can be found in [18] with further details and tests of the robustness of our current assumptions.
Figure 2. Comparison of the results of our extended mass spectrum model particle ratio calculations vs. experimental data points of ALICE at Pb + Pb LHC √ sNN = 2 76 TeV. .
## 3. Results
In Fig. 1 our results are shown compared to Au + Au RHIC √ sNN = 200 GeV data and in Fig. 2 our results are shown in comparison to Pb + Pb LHC √ sNN = 2 76 TeV data. . The solid black dots represent the situation where there are no initial protons, kaons, and lambdas in our system (while the pions and Hagedorn states begin in chemical equilibrium) whereas the outlined circles represent the scenario when all hadrons begin in chemical equilibrium. Follow these initial conditions the hadrons are then allowed to dynamically equilibrate over the expansion period. The LHC calculations end at Tend = 133 MeV and the RHIC calculations end at Tend = 135 MeV. We caution against using Tend MeV as a chemical equilibration temperature because it is highly dependent on the choice of parameters when describing the extended mass spectrum [18].
In a previous papers [19, 20, 21] we explored the e ff ect of the extended mass spectrum at RHIC and found that they were able to match experimental data points. However, we have since updated our mass spectrum to fit to more recent lattice data [12] and have also updated our modeling method to reflect the current modeling procedures of heavy ion collisions. Thus, it is important to note that even after these changes the extended mass spectrum is able to match experimental particle ratios at RHIC (when the protons begin underpopulated). This is an essential point because the same procedure that is used to match particle yields at RHIC using thermal fits was not able to adequately explain the lower p /π ratio at LHC [5, 4].
One can see in Fig. 2 that once again we are able to match the experimental data points at the LHC, which includes the lower p /π ratio that has been unexplained by thermal fits [5, 4]. However, this is only possible in the scenario where there is an initial underpopulation of protons and lambdas. When the protons start in chemical equilibrium, then the proton to pion ratio is significantly overpopulated, which implies that both the extended mass spectrum combined with dynamical chemical equilibration is needed to explain the p /π puzzle at the LHC.
## 4. Conclusions
Our results indicate that the inclusion of the extended mass spectrum into dynamical hadron gas interactions can then explain the suppressed p /π ratio at the LHC. Our current model is somewhat limited in that we can only consider non-strange, mesonic Hagedorn states. We hope in the future to combine the transport model that includes the extended mass spectrum [22] with a relativistic event-by-event hydrodynamical code [23] to provide a more systematic check of the e ff ect of the extended mass spectrum on the particles yields and collective flow. We also plan on exploring the e ff ect of the extended mass spectrum on multi-strange particles and additionally how strange and or / baryonic Hagedorn states would populate those states.
However, these results provide a strong indication that we could be seeing an e ff ect from the extended mass spectrum at the LHC. Already the extended mass spectrum plays a role in the transport coe ffi cients, specifically the shear viscosity to entropy density ratio [9, 10], and that additional strange baryons can a ff ect cumulants and correlations of charge fluctuations [16]. Thus, it is natural to question what other e ff ects the extended mass spectrum will play on the signals of the Quark Gluon Plasma once they are integrated into hadronic after burners and transport methods.
- J. Noronha-Hostler acknowledges Fundac ¸˜ ao de Amparo ` a Pesquisa do Estado de S˜ ao Paulo (FAPESP) for financial support. This work was supported by the Bundesministerium fur Bildung und Forschung (BMBF), the HGS-HIRe and the Helmholtz International Center for FAIR within the framework of the LOEWE program launched by the State of Hesse.
## References
- [1] P. Braun-Munzinger, K. Redlich and J. Stachel, arXiv:nucl-th 0304013; P. Braun-Munzinger, J. Stachel, J. P. Wessels and N. Xu, Phys. Lett. B / 344 , 43 (1995);Phys. Lett. B 365 , 1 (1996); J. Cleymans, D. Elliott, A. Keranen and E. Suhonen, Phys. Rev. C 57 , 3319 (1998); J. Cleymans, H. Oeschler and K. Redlich, Phys. Rev. C 59 , 1663 (1999); R. Averbeck, R. Holzmann, V. Metag and R. S. Simon, Phys. Rev. C 67 , 024903 (2003); P. Braun-Munzinger, I. Heppe and J. Stachel, Phys. Lett. B 465 , 15 (1999). J. Cleymans, H. Satz, E. Suhonen and D. W. von Oertzen, Phys. Lett. B 242 , 111 (1990); J. Cleymans and H. Satz, Z. Phys. C 57 , 135 (1993); F. Becattini, M. Gazdzicki and J. Sollfrank, Eur. Phys. J. C 5 , 143 (1998); F. Becattini, J. Cleymans, A. Keranen, E. Suhonen and K. Redlich, Phys. Rev. C 64 , 024901 (2001); G. Torrieri and J. Rafelski, Phys. Lett. B 509 , 239 (2001); G. Torrieri, S. Steinke, W. Broniowski, W. Florkowski, J. Letessier and J. Rafelski, Comput. Phys. Commun. 167 , 229 (2005); S. Wheaton and J. Cleymans, Comput. Phys. Commun. 180 , 84 (2009); A. Kisiel, T. Taluc, W. Broniowski and W. Florkowski, Comput. Phys. Commun. 174 , 669 (2006) . C. Spieles, H. Stoecker and C. Greiner, Eur. Phys. J. C 2 , 351 (1998); B. Schenke and C. Greiner, J. Phys. G 30 , 597 (2004). P. Braun-Munzinger, D. Magestro, K. Redlich and J. Stachel, Phys. Lett. B 518 , 41 (2001); W. Florkowski, W. Broniowski and M. Michalec, Acta Phys. Polon. B 33 , 761 (2002); W. Broniowski and W. Florkowski, Phys. Rev. C 65 , 064905 (2002); M. Kaneta and N. Xu, arXiv:nucl-th 0405068; J. Adams / et al. [STAR Collaboration], Nucl. Phys. A 757 , 102 (2005). A. Andronic, P. Braun-Munzinger and J. Stachel, Nucl. Phys. A 772 , 167 (2006). J. Manninen and F. Becattini, Phys. Rev. C 78 , 054901 (2008).
- [2] J. Noronha-Hostler, H. Ahmad, J. Noronha and C. Greiner, Phys. Rev. C 82 , 024913 (2010) [arXiv:0906.3960 [nucl-th]].
- [3] P. Braun-Munzinger and J. Stachel, Nucl. Phys. A 638 , 3 (1998); J. Cleymans and K. Redlich, Phys. Rev. Lett. 81 , 5284 (1998);Phys. Rev. C 60 , 054908 (1999); J. Cleymans, J. Phys. G 35 , 044017 (2008); J. Cleymans, R. Sahoo, D. K. Srivastava and S. Wheaton, Eur. Phys. J. ST 155 , 13 (2008) J. Cleymans, R. Sahoo, D. P. Mahapatra, D. K. Srivastava and S. Wheaton, Phys. Lett. B 660 , 172 (2008); J. Cleymans, H. Oeschler, K. Redlich and S. Wheaton, J. Phys. G 32 , S165 (2006).
- [4] N. Armesto et al. , J. Phys. G 35 , 054001 (2008).
- [5] B. Abelev et al. [ALICE Collaboration], Phys. Rev. Lett. 109 , 252301 (2012) [arXiv:1208.1974 [hep-ex]].
- [6] J. Steinheimer, J. Aichelin and M. Bleicher, Phys. Rev. Lett. 110 , 042501 (2013) . Phys. Rev. D 85 , 014004 (2012) .
- [7]
- M. Floris, Quark Matter 2014 proceedings.
- [8] W. Broniowski, W. Florkowski and L. Y. .Glozman, Phys. Rev. D 70 , 117503 (2004) [hep-ph 0407290]. /
- [9] J. Noronha-Hostler, J. Noronha and C. Greiner, Phys. Rev. Lett. 103 , 172302 (2009) [arXiv:0811.1571 [nucl-th]].
- [10] J. Noronha-Hostler, J. Noronha and C. Greiner, Phys. Rev. C 86 , 024913 (2012) [arXiv:1206.5138 [nucl-th]].
- [11] S. Pal, Phys. Lett. B 684 , 211 (2010) .
- [12] S. Borsanyi, G. Endrodi, Z. Fodor, A. Jakovac, S. D. Katz, S. Krieg, C. Ratti and K. K. Szabo, JHEP 1011 , 077 (2010) [arXiv:1007.2580 [hep-lat]].
- [13] A. Majumder and B. Muller, Phys. Rev. Lett. 105 , 252002 (2010) [arXiv:1008.1747 [hep-ph]].
- [14] J. Noronha-Hostler, J. Noronha, G. S. Denicol, R. P. G. Andrade, F. Grassi and C. Greiner, Phys. Rev. C 89 , 054904 (2014).
- [15] L. G. Moretto, K. A. Bugaev, J. B. Elliott and L. Phair, Europhys. Lett. 76 , 402 (2006); V. V. Begun, M. I. Gorenstein and W. Greiner, J. Phys. G 36 , 095005 (2009); I. Zakout, C. Greiner and J. Scha ff ner-Bielich, Nucl. Phys. A 781 , 150 (2007); I. Zakout and C. Greiner, Phys. Rev. C 78 , 034916 (2008); L. Ferroni and V. Koch, Phys. Rev. C 79 , 034905 (2009); K. A. Bugaev, V. K. Petrov and G. M. Zinovjev, Phys. Rev. C 79 , 054913 (2009); A. I. Ivanytskyi, K. A. Bugaev, A. S. Sorin and G. M. Zinovjev, Phys. Rev. E 86 , 061107 (2012) [arXiv:1211.3815 [nucl-th]].
- [16] A. Bazavov, H. -T. Ding, P. Hegde, O. Kaczmarek, F. Karsch, E. Laermann, Y. Maezawa and S. Mukherjee et al. , arXiv:1404.6511 [hep-lat].
- [17] R. Hagedorn, Nuovo Cim. Suppl. 3 , 147 (1965); R. Hagedorn, Nuovo Cim. A 56 , 1027 (1968).
- [18] J. Noronha-Hostler and C. Greiner, arXiv:1405.7298 [nucl-th].
- [19] C. Greiner et al. J. Phys. G 31 , S725 (2005).
- [20] J. Noronha-Hostler, C. Greiner and I. A. Shovkovy, Phys. Rev. Lett. 100 , 252301 (2008) [arXiv:0711.0930 [nucl-th]].
- [21] J. Noronha-Hostler, M. Beitel, C. Greiner and I. Shovkovy, Phys. Rev. C 81 , 054909 (2010) [arXiv:0909.2908 [nucl-th]].
- [22] M. Beitel, K. Gallmeister and C. Greiner, arXiv:1402.1458 [hep-ph].
- [23] J. Noronha-Hostler, G. S. Denicol, J. Noronha, R. P. G. Andrade and F. Grassi, Phys. Rev. C 88 , 044916 (2013); J. Noronha-Hostler, J. Noronha and F. Grassi, arXiv:1406.3333 [nucl-th].
|
10.1016/j.nuclphysa.2014.08.101
|
[
"Jacquelyn Noronha-Hostler",
"Carsten Greiner"
] |
2014-08-04T18:33:09+00:00
|
2014-08-04T18:33:09+00:00
|
[
"nucl-th",
"hep-ph"
] |
Understanding the $p/π$ ratio at LHC due to QCD mass spectrum
|
Thermal fits have consistently reproduced the experimental particles yields
of heavy ion collisions, however, the proton to pion ratio from ALICE Pb+Pb
$\sqrt{s_{NN}}=2.76$ TeV is over-predicted by thermal models- known at the
$p/\pi$ puzzle. Here we test the relevance of the extended mass spectrum, i.e.,
include Hagedorn states (resonances that follow an exponential mass spectrum
and have very short life times) on the $p/\pi$ puzzle. We find that the
extended mass spectrum is able to reproduce particle ratios at both RHIC and
the LHC as well as being able to match the lower $p/\pi$ ratio at the LHC
through dynamical chemical equilibration.
|
1408.0762v1
|
## Topological full groups of minimal subshifts with subgroups of intermediate growth
Nicol´s Matte Bon a ∗
August 5, 2014
## Abstract
We show that every Grigorchuk group G ω embeds in (the commutator subgroup of) the topological full group of a minimal subshift. In particular, the topological full group of a Cantor minimal system can have subgroups of intermediate growth, a question raised by Grigorchuk; it can also have finitely generated infinite torsion subgroups, as well as residually finite subgroups that are not elementary amenable, answering questions of Cornulier. By estimating the word-complexity of this subshift, we deduce that every Grigorchuk group G ω can be embedded in a finitely generated simple group that has trivial Poisson boundary for every simple random walk.
## 1 Introduction
Let ( X,ϕ ) be a Cantor minimal system , i.e. a minimal dynamical system where X is a compact space homeomorphic to the Cantor set, and ϕ is a homeomorphism of X . Recall that a dynamical system is said to be minimal if every orbit is dense. The topological full group of ( X,ϕ ) is the group [[ ϕ ]] of all homeomorphisms of X that locally coincide with a power of ϕ ; in other words the homeomorphisms g of X so that there exists a continuous function n : X → Z that verifies g x ( ) = ϕ n x ( ) ( x ) for every x ∈ X . Giordano, Putnam and Skau show in [GPS99] that this group characterizes completely the dynamics of ( X,ϕ ), up to replacing ϕ by ϕ -1 .
In this note we are mainly interested in the possible behaviors for the growth of finitely generated subgroups of [[ ϕ ]]. The growth function of a finitely generated group G equipped with a finite symmetric generating set S is the function b G,S : N → N that counts the number of group elements that can be obtained as a product of at most n generators in S . The growth of G is said to be polynomial if there exist C, D > 0 such that b G,S ( n ) ≤ Cn D for every n ∈ N , exponential if there exists c > 1 such that b G,S ( n ) ≥ c n for every n ∈ N , and intermediate otherwise. These properties do not depend on the choice of S .
∗ Universit´ d'Orsay & DMA, ´ cole Normale Sup´ erieure; email: [email protected] e E
Grigorchuk groups G ω , first defined and studied by Grigorchuk in [Gri84], provide examples of groups of intermediate growth. The groups G ω can be defined as certain subgroups of the automorphism group of a binary rooted tree. They are parameterized by an infinite sequence on three symbols, denoted ω ∈ { 0 1 2 , , } N + . We refer to Section 2 for preliminaries on the groups G ω .
Theorem 1.1. For every ω ∈ { 0 1 2 , , } N + there exists a minimal subshift ( X ,ϕ ω ω ) and an embedding of the Grigorchuk group G ω into the topological full group [[ ϕ ω ]] .
Relying on properties of the groups G ω proven in [Gri84, Theorem 2.1, Corollary 3.2] we get
Corollary 1.2. A finitely generated subgroup of the topological full group of a minimal subshift
- 1. can have intermediate growth;
- 2. can be an infinite torsion group;
- 3. can be residually finite without being elementary amenable.
The question whether a subgroup of the topological full group of a Cantor minimal system can have intermediate growth was originally raised by Grigorchuk (private communication), and it appears in Matui [Mat13], where he shows that the group [[ ϕ ]] and its derived subgroup [[ ϕ ]] ′ have exponential growth whenever they are finitely generated; see also Question (2c) in Cornulier's survey [Cor14]. Points 2 and 3 in Corollary 1.2 answer to Questions (2c)-(2d) and to part of Question (2b) asked by Cornulier in the same survey [Cor14].
The susbshift ( X ,ϕ ω ω ) in Theorem 1.1 can be defined as a space of labeled graph structures on the set of integers, that have the same finite patterns as the orbital Schreier graphs for the action of G ω on the boundary of the binary rooted tree. Similar constructions of group actions based on graph colourings have been used, in this context, by Elek and Monod [EM13] and by van Douwen [vD90].
A minor modification of the construction allows to embed the groups G ω in the commutator subgroup of the topological full group of a minimal subshift, this is shown in Corollary 4.4. The latter is a finitely generated simple group by results of Matui [Mat06, Theorem 4.9, Theorem 5.4]. Relying on a result from [MB14] we deduce the following statement in Theorem 4.1: every Grigorchuk group G ω can be embedded in a finitely generated simple group with the Liouville property (in particular, amenable). Note that the topological full group of any Cantor minimal system is amenable by a result of Juschenko and Monod [JM13].
Related constructions appear in Vorobets' articles [Vor10, Vor12]. In [Vor10] he studies dynamical properties of a minimal one-sided substitutional subshift, which can be shown to be a one-sided version of our ( X ,ϕ ω ω ) for the periodic sequence ω = 012 · · · (the group G ω corresponding to this ω is known as the first Grigorchuk group ). In [Vor12] he studies the dynamics of the action of the first Grigorchuk group on the space of marked Schreier graphs.
The paper is structured as follows. Section 2 recalls the definition of Grigorchuk groups and the construction of their orbital Schreier graphs. Proofs concerning this construction are postponed to the Appendix at the end. Section 3 contains the definition of the subshift ( X ,ϕ ω ω ) and to the proof of Theorem 1.1; this section can be read without reading the Appendix, but it requires notations from Section 2. In Section 4 we modify the construction to get an embedding in the commutator subgroup; then we study the complexity of the subshift ( X ,ϕ ω ω ) and deduce the above-mentioned embedding result for Grigorchuk groups.
## Acknowledgements
I am grateful to Anna Erschler for several useful discussions, to Slava Grigorchuk for many interesting comments and for pointing out the connection with Vorobets' work [Vor10, Vor12], and to Gidi Amir for explaining me the useful relation between the Gray code and automata groups, first appeared in [BG ˇ 03, Section 10.3], that is used S in the Appendix.
## 2 Grigorchuk groups and their Schreier graphs
## 2.1 Grigorchuk groups
Grigorchuk groups act on the infinite binary rooted tree by automorphisms. The binary rooted tree is identified with the set of binary strings { 0 1 , } ∗ , by declaring v, w ∈ { 0 1 , } ∗ connected by an edge if there exists x ∈ { 0 1 , } such that w = vx . This tree is rooted at the empty string ∅ .
The groups G ω are parameterized by the set Ω = { 0 1 2 , , } N + , which is endowed with a shift map σ : Ω → Ω (note that the bold digits 0 1 2 , , shall not be confused with the digits 0 , 1 used to index the tree). Given ω ∈ Ω, the Grigorchuk group G ω is the group generated by the four automorphisms a, b ω , c ω , d ω of the tree, that we now define. Let ε be the non-trivial permutation of { 0 1 , } . The generator a does not depend on the choice of the sequence ω , and it acts on binary strings by
$$a ( x v ) = \varepsilon ( x ) v,$$
glyph[negationslash]
where the above formula holds for every x ∈ { 0 1 , } and v ∈ { 0 1 , } ∗ . For i, j ∈ { 0 1 2 , , } let ε ij be the permutation of { 0 1 , } which is equal to ε if i = j and to the identity if i = . The generators j b ω , c ω , d ω are defined by the recursive rules
$$b _ { \omega } ( 0 x v ) & = 0 \varepsilon _ { 2 \omega ( 1 ) } ( x ) v ; \quad b _ { \omega } ( 1 v ) = 1 b _ { \sigma \omega } ( v ) ; \\ c _ { \omega } ( 0 x v ) & = 0 \varepsilon _ { 1 \omega ( 1 ) } ( x ) v ; \quad c _ { \omega } ( 1 v ) = 1 c _ { \sigma \omega } ( v ) ; \\ d _ { \omega } ( 0 x v ) & = 0 \varepsilon _ { 0 \omega ( 1 ) } ( x ) v ; \quad d _ { \omega } ( 1 v ) = 1 d _ { \sigma \omega } ( v ) ;$$
∗
that hold for every x ∈ { 0 1 , } , v ∈ { 0 1 , } and where σω ∈ Ω is the shifted sequence. Let us give another description of the generator b ω . If v ∈ { 0 1 , } ∗ does not contain the digit 0, or if the digit 0 appears only at the last position of v , then b ω ( v ) = v . Otherwise suppose that the first appearance of 0 in v is at position j . If ω j ( ) = 2 then glyph[negationslash]
b ω permutes the letter at position j + 1 in v , while if ω j ( ) = 2 then b ω ( v ) = v . An analogous description holds for c ω , d ω after replacing 2 by 1 0 , , respectively.
Note that for every ω ∈ Ω we have the relations a 2 = b 2 ω = c 2 ω = d 2 ω = b ω c ω d ω = . e
Grigorchuk proved in [Gri84, Theorem 2.1, Corollary 3.2] that
- 1. the group G ω is residually finite and amenable for every sequence ω ;
- 2. if ω eventually constant the group G ω is virtually abelian, otherwise it has intermediate growth and it is not elementary amenable (recall that the class of elementary amenable groups is the smallest class of groups containing finite and abelian groups and which is closed by taking extensions, inductive limits, subgroups and quotients);
- 3. the group G ω is 2-torsion if ω has infinitely many appearances of each of the symbols 0 1 2 , , , otherwise G ω admits elements of infinite order.
## 2.2 Schreier graphs of Grigorchuk groups
The action of Grigorchuk groups on the binary tree extends to the boundary at infinity of the tree, identified with the set of right-infinite strings { 0 1 , } ∞ .
Hereinafter let ρ be the constant 1-string ρ = 111 · · · ∈ { 0 1 , } ∞ , and let G ρ ω be its orbit. We denote by Γ ω the Schreier graph for the action of G ω on the orbit of ρ , with respect to the generating set S = { a, b ω , c ω , d ω } . This is the labelled graph whose vertex set is G ρ ω and such that γ, η ∈ G ρ ω are connected by an edge if there exists s ∈ S such that sγ = η . Such an edge is labelled by s . Loops and multiple edges are allowed. Since all generators are involutions, the orientation of edges is unimportant and we shall work with non-oriented graphs.
The Schreier graphs Γ ω can be described elementary, see Bartholdi and Grigorchuk [BG00, Section 5] where the graph Γ ω is constructed for the sequence ω = 012012 · · · (the corresponding group G ω is known as the first Grigorchuk group ). The construction of the graph Γ ω for other choices of ω is also well-known and can easily be adapted from the case of the first Grigorchuk group. We now recall this construction and some related facts; since we were not able to locate a reference for a generic ω , we refer to the Appendix for proofs.
Figure 1: The beginning of the graph Γ ω for ω = 012012 · · · .
Proposition 2.1. The orbit G ρ ω consists exactly of all γ ∈ { 0 1 , } ∞ that are cofinal with ρ .
Proposition 2.2. As an unlabelled graph, Γ ω does not depend on the sequence ω and is isomorphic to the half-line with additional loops and double edges shown in Figure 1 (apart from labels). The endpoint of the half-line is the vertex ρ . Moreover for every sequence ω
- · the three loops at ρ are labelled by ' b ω ', ' c ω ', ' d ω ';
- · every simple edge is labelled ' a '.
The remaining edges (loops and double edges) are labelled by b ω , c ω , d ω . The labelling of these depends on ω and is described in Lemma 2.4.
Before describing the labelling of Γ ω we fix some notation.
Definition 2.3. 1. We denote Θ Λ , 0 , Λ 1 , Λ 2 the four finite labelled graphs shown in Figure 2, and Ξ the graph with one vertex and three loops, labelled by b ω , c ω , d ω .
- 2. We denote ∗ the operation of gluing two graphs by identifying two vertices. For well-definiteness, we assume that all finite graphs that we consider have a distinguished 'leftmost vertex' and a 'rightmost vertex' (that will be clear from the context); the operation ∗ corresponds to identifying the rightmost vertex of the first graph with the leftmost vertex of the second.
Figure 2: The 'building blocks' of Γ ω .
By Proposition 2.2, with this notation Γ ω has the form
$$\Gamma _ { \omega } = \Xi * \Theta * \Delta _ { 1, \omega } * \Theta * \Delta _ { 2, \omega } * \Theta * \Delta _ { 3, \omega } * \Theta \cdots,$$
where ∆ i,ω ∈ { Λ 0 , Λ 1 , Λ 2 } .
It will be convenient to denote Γ ˜ ω the modification of Γ ω obtained by erasing the three loops at ρ , i.e.
$$\tilde { \Gamma } _ { \omega } = \Theta * \Delta _ { 1, \omega } * \Theta * \Delta _ { 2, \omega } * \Theta * \Delta _ { 3, \omega } \cdots.$$
We now state a lemma, that allows to construct ˜ Γ ω recursively (part (i) is sufficient for this purpose). We refer to the Appendix for a proof.
Lemma 2.4. (i) For every n ≥ 0 , the the graph spanned by the first 2 n +1 vertices of ˜ Γ ω is isomorphic to the graph ˜ Γ n,ω , defined recursively by
$$\begin{cases} \tilde { \Gamma } _ { 1, \omega } = \Theta * \Lambda _ { \omega ( 1 ) } * \Theta ; \\ \tilde { \Gamma } _ { n + 1, \omega } = \tilde { \Gamma } _ { n, \omega } * \Lambda _ { \omega ( n + 1 ) } * \tilde { \Gamma } _ { n, \omega }, \end{cases}$$
where ω ∈ { 0 1 2 , , } N + is the sequence that defines G ω .
(ii) The graph ˜ Γ ω has the following self-similarity property: for every n ≥ 1 we have
$$\tilde { \Gamma } _ { \omega } = \tilde { \Gamma } _ { n, \omega } * \Delta _ { 1, \sigma ^ { n } \omega } * \tilde { \Gamma } _ { n, \omega } * \Delta _ { 2, \sigma ^ { n } \omega } * \tilde { \Gamma } _ { n, \omega } * \Delta _ { 3, \sigma ^ { n } \omega } * \tilde { \Gamma } _ { n, \omega } * \Delta _ { 4, \sigma ^ { n } \omega } \cdots,$$
where σ n ω is the sequence ω shifted by n , ˜ Γ σ n ω = Θ ∗ ∆ 1 ,σ n ω ∗ Θ ∗ ∆ 2 ,σ n ω ··· is the unique decomposition of ˜ Γ σ n ω with ∆ i,σ n ω ∈ { Λ 0 , Λ 1 , Λ 2 } , and ˜ Γ n,ω is the sequence of finite graphs constructed at (i).
## 3 Construction of the subshift
We now turn to the proof of Theorem 1.1.
We first address separately the degenerate case of an eventually constant ω .
Proposition 3.1. Suppose that the sequence ω is eventually constant. Then for every minimal Cantor system ( X,ϕ ) there is an embedding of G ω into [[ ϕ ]] .
glyph[negationslash]
Proof. If ω is eventually constant, the group G ω is virtually abelian, see [Gri84, Theorem 2.1 (3)]. In fact it can be embedded in a semi-direct product of the form Z n glyph[multicloseright] S n for sufficiently large n , where the symmetric group S n acts on Z n by permutation of the coordinates (strictly speaking, it is shown in [Gri84] that it can be embedded in D n ∞ glyph[multicloseright] S n , where D ∞ is the infinite dihedral group; note for instance that D n ∞ glyph[multicloseright] S n embeds in Z 2 n glyph[multicloseright] S 2 n by identifying D ∞ with the subgroup of Z 2 glyph[multicloseright] S 2 generated by (1 , -1) ∈ Z 2 and by the nontrivial glyph[epsilon1] ∈ S 2 ). It is thus sufficient to show that for every minimal Cantor system ( X,ϕ ) and every n ∈ N , the group Z n glyph[multicloseright] S n embeds in [[ ϕ ]]. Let U ⊂ X be a clopen set such that U, ϕ U ( ) , . . . , ϕ n -1 ( U ) are disjoint. For every 0 ≤ i < j ≤ n -1 let σ ij ∈ [[ ϕ ]] be the involution with support ϕ i ( U ) ∪ ϕ j ( U ), which coincides with ϕ j -i on ϕ i ( U ), with ϕ i -j on ϕ j ( U ), and is the identity on the complement of ϕ i ( U ) ∪ ϕ j ( U ). The group generated by the involutions σ ij is isomorphic to the symmetric group S n . Let r 0 ∈ [[ ϕ ]] be any element of infinite order with support contained in U (for instance take the first return map of U extended to the identity on the complement of U ). For i = 0 , . . . , n -1 set r i = ϕ r ϕ i 0 -i . The support of r i is contained in ϕ i ( U ), thus the choice of U implies that r i and r j commute if i = j . It follows that the group generated by r , . . . , r 0 n -1 is free abelian of rank n . The group 〈 σ ij , r l : 0 ≤ i < j ≤ n -1 , 0 ≤ l ≤ n - 〉 1 is then isomorphic to Z n glyph[multicloseright] S n .
Hereinafter we fix ω ∈ Ω and assume that it is not eventually constant.
By a pattern in a labeled graph we mean the isomorphism class of a finite, connected labelled subgraph.
Definition 3.2 (Construction of the subshift ( X ,ϕ ω ω )) . 1. Let X be the the space of edge-labeled, connected graphs with vertex set Z such that any two consecutive integers n, n +1 ∈ Z are connected by one of the four labelled graphs in Figure 2, and there is no edge between n, m ∈ Z if | n -m > | 1. Endow X with the natural product topology and with the shift map ϕ induced by translations of Z , which makes ( X,ϕ ) conjugate to the full shift over the alphabet A = Θ Λ { , 0 , Λ 1 , Λ 2 } .
- 2. Let X ω ⊂ X be the set of all x ∈ X such that every pattern of x appears in ˜ Γ ω (recall that this is the graph obtained from Γ ω after erasing the three initial loops). This defines a closed, shift-invariant subset of X and we denote ( X ,ϕ ω ω ) the subshift obtained in this way.
Remark 3.3 . Equivalently X ω could be defined as the orbit-closure of any x ∈ X isomorphic to the Schreier graph of the orbit of a point γ ∈ { 0 1 , } ∗ lying outside the orbit of ρ (these graphs are isomorphic to bi-infinite lines; their labellings admit similar and only slightly more complicated recursive construction).
A labeled graph is said to be uniformly recurrent if for every finite pattern in the graph exists a constant R > 0 such that every ball of radius R in the graph contains a copy of the pattern.
Lemma 3.4. The graph ˜ Γ ω is uniformly recurrent. Moreover its labelling is not eventually periodic along the half-line.
Proof. We take notations from Definition 2.3 and Lemma 2.4.
By part (i) of Lemma 2.4, for every pattern of Γ ˜ ω there exists n such that the pattern appears in ˜ Γ n,ω . Using this, part (ii) of Lemma 2.4 implies that ˜ Γ ω is uniformly recurrent.
glyph[negationslash]
Suppose that the labelling of Γ ˜ ω is eventually periodic, i.e. that the sequence of graphs (∆ j,ω ) j ≥ 1 is eventually periodic with period T . Since ˜ Γ 1 ,ω = Θ ∗ Λ ω (1) ∗ Θ, part (ii) of Lemma 2.4 for n = 1 implies that all odd terms in the sequence (∆ j,ω ) j ≥ 1 are equal to Λ ω (1) , while the subsequence of even terms is equal to (∆ j,σω ) j ≥ 1 . Since we assume that ω is not eventually constant, part (i) of Lemma 2.4 easily implies that the sequence (∆ j,ω ) j ≥ 1 is also not eventually constant, hence there are infinitely many i s for which ∆ i,σω = Λ ω (1) . We conclude that T is even, and that the sequence(∆ j,σω ) j ≥ 1 is also eventually periodic with period T/ 2. Since ω is not eventually constant, neither is σω , thus we may repeat the same reasoning for Γ ˜ σω to conclude that T/ 2 is even and (∆ j,σ 2 ω j ) ≥ 1 is eventually periodic with period T/ 4. By induction T is a multiple of 2 n for every n , thus T = 0.
## Corollary 3.5. The subshift ( X ,ϕ ω ω ) is infinite and minimal.
Previous results by Vorobets in [Vor10] imply minimality of a one-sided subshift given by a certain substitution, which is conjugate to the one-sided version of ( X ,ϕ ω ω ) for the periodic ω = 012 · · · .
Proof. Let us first check that X ω is not empty. To see this, let y ∈ X be any graph that agrees with Γ ˜ ω on the positive half-line. Then any cluster point of ( ϕ n ( y )) n ≥ 0 belongs to X ω .
Let x ∈ X ω be arbitrary. It is a consequence of Lemma 3.4 that every pattern of ˜ Γ ω appears in x infinitely many times, and that x is uniformly recurrent. Namely every pattern of ˜ Γ ω appears in every sufficiently long segment of ˜ Γ , thus in every sufficiently ω long segment of x by the construction of X ω . This easily implies that the orbit of x is dense in X ω , and that ( X ,ϕ ω ω ) is minimal by a well-known characterization of minimality (see [Got46] or [Que87, Proposition 4.7]). Finally the fact that Γ ˜ ω is not eventually periodic (Lemma 3.4) implies that x is not periodic, since Γ ˜ ω and x have the same finite patterns and periodicity can be easily characterized in terms of these (for instance by Morse and Hedlund's theorem, see [CN10, Theorem 4.3.1]). Thus X ω is infinite.
Remark 3.6 . It follows from the proof that every pattern in ˜ Γ ω appears in every x ∈ X ω .
Proposition 3.7. The Grigorchuk group G ω embeds in the topological full group [[ ϕ ω ]] .
Proof. Let x ∈ X ω . By construction, for every f ∈ { a, b ω , c ω , d ω } the vertex 0 ∈ Z is the endpoint of exactly one edge in x which is labelled by f (this edge is possibly a loop at 0). We define four elements ¯ a, b ¯ ω , c ¯ ω , d ¯ ω of the topological full group [[ ϕ ω ]] as follows: every ¯ f ∈ { ¯ ¯ a, b ω , c ¯ ω , d ¯ ω } translates x ∈ X ω in the direction where the label f is, that is ¯ fx = ϕ ω ( x ) (respectively ¯ fx = ϕ -1 ω ( x ), ¯ fx = x ) if the label f is on an edge connecting 0 and 1 (respectively on an edge connecting 0 and -1, on the loop at 0). It is clear that ¯ a, b ¯ ω , c ¯ ω , d ¯ ω ∈ [[ ϕ ω ]].
↦
We claim that the map ι : a → ¯ a, b ω → ¯ b ω , c ω → ¯ c ω , d ω → ¯ d ω extends to an injective group homomorphism ι : G ω → [[ ϕ ω ]].
↦
↦
↦
glyph[negationslash]
glyph[negationslash]
glyph[negationslash]
To see that ι extends to a well-defined homomorphism, it is sufficient to check that relations are respected, i.e. that for every n ∈ N and h , . . . , h 1 n ∈ { a, b ω , c ω , d ω } such that h 1 · · · h n = e in G ω we have ¯ h 1 · · · ¯ h n = e in [[ ϕ ω ]]. Suppose that ¯ h 1 · · · ¯ h n = e in [[ ϕ ω ]]. Then there exists a point x ∈ X ω such that ¯ h 1 · · · ¯ h x n = x . Write x | n for the finite subgraph of x spanned by the interval [ -n, n ] ⊂ Z . Observe that the fact whether x is fixed or not by ¯ h 1 · · · ¯ h n only depends on x through x | n . By the construction of X ω , the graph x | n isomorphic to a labeled subgraph of Γ ˜ ω , hence of Γ ω . Let γ ∈ Γ ω be the midpoint of this subgraph. It follows from the definition of ¯ a, b ¯ ω , c ¯ ω , d ¯ ω that h 1 · · · h γ n = γ . We conclude that h 1 · · · h n = e in G ω .
glyph[negationslash]
glyph[negationslash]
glyph[negationslash]
glyph[negationslash]
The verification that ι is injective is similar. Let h , 1 · · · , h n ∈ { a, b ω , c ω , d ω } be such that h 1 · · · h n = e in G ω . Then there is a vertex of the rooted tree, say v ∈ { 0 1 , } ∗ , which is moved by h 1 · · · h n . Recall that the orbit G ρ ω consists exactly of all sequences γ ∈ { 0 1 , } ∗ that are cofinal with ρ (Proposition 2.1). In particular, v is a prefix of infinitely many such sequences, each of which is moved by h 1 · · · h n . Thus we can find γ ∈ Γ ω lying at a distance greater than n from ρ , and such that h 1 · · · h γ n = γ . Since we have chosen γ at a distance greater than n from ρ , γ is the midpoint of a connected graph of length 2 n in Γ ω . This graph is also a subgraph of Γ ˜ ω . By Remark 3.6 and by shift-invariance, there exists x ∈ X ω such that x | n is isomorphic to this graph (recall that x | n denotes the finite subgraph of x spanned by [ -n, n ]). Again by construction of ¯ a, b ¯ ω , c ¯ ω , d ¯ , we have ¯ ω h 1 · · · ¯ h x n = ϕ ± k ω ( x ), where k > 0 is the distance by which γ is displaced on Γ ω by the action of h 1 · · · h n . But Corollary 3.5 implies that there is no periodic point for ϕ ω , thus ¯ h 1 · · · ¯ h n ( x ) = ϕ ± k ω ( x ) = x , and so ¯ h 1 · · · ¯ h n = e in [[ ϕ ω ]]. We conclude that ι is injective.
glyph[negationslash]
Remark 3.8 . The usual action of G ω on the Cantor set identified with the boundary of the tree { 0 1 , } ∞ does not belong to the topological full group of a minimal homeomorphism of { 0 1 , } ∞ . To see this, note that b ω , c ω , d ω fix ρ but act non-trivially on every neighbourhood of ρ . This can not happen in the topological full group of a minimal homeomorphism.
## 4 Embedding G ω in a simple Liouville group
It is a classical result, due to Hall [Hal74], Gorjuskin [Gor74] and Schupp [Sch76] that every countable group can be embedded in a finitely generated simple group. These constructions always yield 'large' ambient groups (e.g. non-amenable) even if the starting group is 'small'. In this section we deduce the following consequence for the groups G ω from the construction presented in Section 3.
Theorem 4.1. Every Grigorchuk group G ω can be embedded in a finitely generated, simple group that has trivial Poisson-Furstenberg boundary for every symmetric, finitely supported probability measure on it (in other words, it has Liouville property and in particular it is amenable).
The class of groups with the Liouville property contains the class of subexponentially growing groups (Avez [Ave74]) and it is contained in the class of amenable groups (this is due to Furstenberg, see [KV83, Theorem 4.2]).
Recall that the complexity of a subshift ( X,ϕ ) over a finite alphabet A is the function ρ X : N → N that associates to n the number of finite words in the alphabet A that appear as subwords of sequences in X . The proof of Theorem 4.1 relies on the following result from [MB14].
Theorem 4.2 (Theorem 1.2 in [MB14]) . Let ( X,ϕ ) be a subshift without isolated periodic points whose complexity satisfies ρ X ( n ) = o ( n 2 log 2 n ) . Then every symmetric, finitely supported probability measure on [[ ϕ ]] has trivial Poisson-Furstenberg boundary.
The next lemma allows to modify the construction in Section 3 to get an embedding in the commutator subgroup.
Lemma 4.3. Let ( X,ϕ ) be a Cantor minimal system and let H < [[ ϕ ]] be the subgroup generated by elements of order 2. Then
- (i) there exists a minimal Cantor system ( Y, ψ ) and an embedding of H into [[ ψ ]] ′ ;
- (ii) if moreover ( X,ϕ ) is a subshift with complexity function ρ X , the system ( Y, ψ ) in (i) can be chosen to be a subshift whose complexity ρ Y satisfies
$$\rho _ { Y } ( n ) \leq 2 \rho _ { X } ( \lceil n / 2 \rceil ).$$
Proof. (i). Set Y = X ×{ 1 2 , } and define ψ : Y → Y by
$$\begin{cases} \ \psi ( x, 1 ) = ( x, 2 ) \\ \ \psi ( x, 2 ) = ( \varphi ( x ), 1 ). \end{cases}$$
The system ( Y, ψ ) is clearly minimal as soon as ( X,ϕ ) is. Observe that X ×{ } 1 and X ×{ } 2 are ψ 2 -invariant and that the restrictions of ψ 2 to X ×{ } 1 and to X ×{ } 2 are conjugate to ( X,ϕ ). This naturally defines two commuting subgroups of [[ ψ ]] isomorphic to [[ ϕ ]] that have support X ×{ } 1 and X ×{ } 2 respectively, and are denoted [[ ϕ ]] 1 and [[ ϕ ]] 2 . Consider the diagonal injection δ : [[ ϕ ]] → [[ ϕ ]] 1 × [[ ϕ ]] 2 ⊂ [[ ψ ]]. If g ∈ [[ ϕ ]] has order 2, its image δ g ( ) equals the commutator [ g 1 : τ ] = g τg τ 1 1 , where g 1 ∈ [[ ϕ ]] 1 is the image of g under the natural isomorphism [[ ϕ ]] → [[ ϕ ]] 1 , and τ ∈ [[ ψ ]] is the element that coincides with ψ on X ×{ } 1 and with ψ -1 on X ×{ } 2 . It follows that δ H ( ) is contained in [[ ψ ]] ′ , proving part (i).
(ii). If ( X,ϕ ) is a subshift over the finite alphabet A , the above construction is readily seen to be equivalent to the following one. Consider the alphabet B = Aglyph[unionsq] { z } where z is a letter not in A , and define Y ⊂ B Z to be the subshift consisting of all sequences of the form y = · · · zx -2 zx -1 .zx zx 0 1 · · · or y = · · · zx -2 zx -1 z.x zx 0 1 · · · where x = · · · x -2 x -1 .x 0 x 1 · · · is a sequence in X . There is a natural partition of Y in two sets homeomorphic to X which are given by the parity of the position of
appearances of z . These two sets play the roles of X ×{ } 1 and X ×{ } 2 in the proof of part (i) and the same argument yields an embedding of H in [[ ψ ]] ′ . The claim on the complexity is also clear from this description.
Since the groups G ω are generated by elements of order 2, we get the following improvement of Theorem 1.1.
Corollary 4.4. Every Grigorchuk group G ω embeds in the commutator subgroup of the topological full group of a minimal subshift.
This already gives an embedding of G ω into a finitely generated, simple amenable group by the result of Juschenko and Monod [JM13]. To prove that the Liouville property holds, we estimate the complexity of the subshift ( X ,ϕ ω ω ) constructed in Section 3, then apply part (ii) of Lemma 4.3 and Theorem 4.2. The following lemma is enough to conclude the proof of Theorem 4.1.
Lemma 4.5. For every ω ∈ Ω the complexity of the subshift ( X ,ϕ ω ω ) constructed in Section 3 satisfies ρ X ω ( n ) ≤ 6 n for every n ∈ N .
Remark 4.6 . Here we see ( X ,ϕ ω ω ) as a subshift in the usual sense, over the (formal) alphabet A = { Θ Λ , 0 , Λ 1 , Λ 2 } .
Proof. Let x ∈ X ω and consider a finite subword w of x of length n ∈ N . Let m ∈ N be the smallest integer such that n ≤ 2 m . By part (ii) of Lemma 2.4 the word w appears as a subword of a word of the form Γ ˜ m -1 ,ω ∗ ∆ ∗ ˜ Γ m -1 ,ω , where ∆ ∈ { Λ 0 , Λ 1 , Λ 2 } . Such a word is uniquely determined by ∆ (3 possibilities) and by the position of its first letter (2 m possibilities, since we may assume without loss of generality that the first letter of w is in Γ ˜ m -1 ,ω ∗ ∆). Thus ρ X ω ( n ) ≤ 3 · 2 m ≤ 6 n , where we have used that 2 m -1 < n by the choice of m .
## Appendix: Schreier graphs and Gray code
In this Appendix we give proofs of the statements in Section 2.2, namely Propositions 2.1, Proposition 2.2 and Lemma 2.4.
We first state a remark that follows from the definitions of a, b ω , c ω , d ω .
Remark A.1 . Given any binary string γ ∈ { 0 1 , } ∞ , a generator s ∈ { a, b ω , c ω , d ω } that acts non-trivially on γ can act in two possible ways:
(M1) flip the first digit of γ (if and only if s = a );
(M2) flip the digit after the first appearance of 0 in γ (only if s ∈ { b ω , c ω , d ω } ).
glyph[negationslash]
Moreover for every γ = ρ it is possible to choose s ∈ { b ω , c ω , d ω } that acts on γ by a move of type (M2).
Proof of Proposition 2.1. Since the action of every generator of G ω changes at most one letter of each binary string, every string in the orbit of ρ is cofinal with ρ . Conversely, it is easy to see that starting from ρ and performing only moves of type (M1) and (M2) it is possible to reach every string which is cofinal with ρ .
glyph[negationslash]
glyph[negationslash]
Proof of Proposition 2.2. Since ρ is fixed by b ω , c ω , d ω , it has three loops and it is the endpoint of a simple edge (given by the action of a ). If γ = ρ there exists exactly one generator s ∈ { b ω , c ω , d ω } , such that sγ = γ . The generator s is determined by the value ω m ( ), where m is the position of the first 0 appearing in γ . The two other generators s , s ′ ′′ ∈ { b ω , c ω , d ω } act non-trivially on γ by a move of type (M2), in particular s γ ′ = s ′′ γ . The generator a acts non-trivially on γ by a move of type (M1), and thus aγ = s γ ′ . Hence the vertex corresponding to γ in Γ ω is the endpoint of a loop, a double edge and a simple edge. Finally, observe that the only connected unlabelled graph respecting these local rules is the graph shown in Figure 2.
Before giving the proof of Lemma 2.4 we recall some terminology. The Gray code is a classical non-standard way to code natural numbers with binary strings, having the property that two consecutive strings differ only by one bit. It is a well-known fact that the Gray code can be used to describe the orbital Schreier graphs of several groups acting on rooted trees, including all groups G ω (cf. Bartholdi, Grigorchuk and ˇ uni´ [BG ˇ 03, Section 10.3] where this is mentioned explicitly for the first Grigorchuk S c S group). We recall the construction of the Gray code in the definition below; the reader should be warned that the roles of 0 and 1 are exchanged here with respect to the more usual definition.
Definition A.2. The Gray code order is the numeration of the set { 0 1 , } l of binary sequences of length l , denoted ( r ( ) l j ) 0 ≤ ≤ j 2 l -1 , defined recursively as follows. Set r (1) 0 = 1 and r (1) 1 = 0. For every l ≥ 0 and 0 ≤ j ≤ 2 l -1 set
$$r _ { j } ^ { ( l + 1 ) } = r _ { j } ^ { ( l ) } 1 ; \quad r _ { 2 ^ { l } + j } ^ { ( l + 1 ) } = r _ { 2 ^ { l } - 1 - j } ^ { ( l ) } 0.$$
In plain words, the first 2 l terms of the sequence ( r ( +1) l j ) 0 ≤ ≤ j 2 l +1 -1 are obtained by attaching the digit 1 to the sequence ( r ( ) l j ) 0 ≤ ≤ j 2 l -1 , and the last 2 l terms are obtained by reversing the order of the sequence ( r ( ) l j ) 0 ≤ ≤ j 2 l -1 and attaching the digit 0 at the end.
For example, the Gray code order of binary strings of length 4 is 1111; 0111; 0011; 1011; 1001; 0001; 0101; 1101; 1100; 0100; 0000; 1000; 1010; 0010; 0110; 1110 .
Definition A.3. The Gray code enumeration of the orbit G ρ ω is the sequence ( ρ j ) j ∈ N , taking values in { 0 1 , } ∞ , where ρ j is obtained by attaching infinitely many digits 1 to r ( ) l j , for any l > 0 such that r ( ) l j is defined (i.e. such that j ≤ 2 l -1). Note that, for any such l , the string r ( +1) l j is obtained from r ( ) l j by attaching the digit 1 on the right, in particular ρ j does not depend on the choice of l .
Lemma A.4. The sequence ( ρ j ) coincides with the sequence of vertices of Γ ω , ordered by increasing distance from ρ .
Proof. It follows from the definitions that ρ 0 = ρ , and that ρ j +1 is obtained from ρ j by performing a move of type (M1) if j is even and a move of type (M2) if j is odd (this can be immediately proven by induction on l from the definition of the Gray code order). Thus the sequence ( ρ j ) coincides with the sequence of vertices of Γ ω by Remark A.1.
Proof of Lemma 2.4. For j ≥ 1 let m j be the position of the first 0 digit in ρ j . There is exactly one generator s ∈ { b ω , c ω , d ω } such that sρ j = ρ j and this generator is determined by ω m ( j ) (more precisely s = b ω , c ω , d ω if ω m ( j ) = 2 1 0 , , , respectively). By comparing this observation with Definition 2.3 and with Figure 2, we see that ∆ i,ω = Λ ω m ( 2 i -1 ) . Set a i = m 2 i -1 . The sequence ( a i ) only depends on the definition of the Gray code order, and it is characterized by the recursion
$$\begin{cases} \ a _ { 2 ^ { n } } = n + 1 & \text{for every $n\geq 0$}, \\ \ a _ { 2 ^ { n } + j } = a _ { 2 ^ { n } - j } & \text{for every $1\leq j\leq 2^{n}-1$}. \end{cases}$$
To prove part (i) of Lemma 2.4, proceed by induction on n . First note that a 1 = 1, so that ∆ 1 ,ω = Λ ω (1) and ˜ Γ 1 ,ω = Θ ∗ Λ ω (1) ∗ Θ. Suppose that the first 2 n +1 vertices of ˜ Γ ω span the graph Γ ˜ n,ω . Then, using the recursion for a j , we have
$$\tilde { \Gamma } _ { n + 1, \omega } = \Theta * \Lambda _ { \omega ( a _ { 1 } ) } * \Theta * \cdots * \Lambda _ { \omega ( a _ { 2 ^ { n } } ) } * \cdots * \Theta * \Lambda _ { \omega ( a _ { 2 ^ { n + 1 } - 1 } ) } * \Theta = \tilde { \Gamma } _ { n, \omega } * \Lambda _ { \omega ( n + 1 ) } * \tilde { \Gamma } _ { n, \omega } ^ { R },$$
where Γ ˜ R n,ω denotes ˜ Γ n,ω reversed from right to left. However we may assume by induction that Γ ˜ n,ω is symmetric with respect to this operation, thus Γ ˜ n +1 ,ω = ˜ Γ n,ω ∗ Λ ω n ( +1) ∗ ˜ Γ n,ω concluding the proof of part (i).
To prove part (ii), fix n and use part (i) to prove by induction on m ≥ 1 that ˜ Γ n + m,ω has the form
$$\tilde { \Gamma } _ { n + m, \omega } = \tilde { \Gamma } _ { n, \omega } * \Delta _ { 1, \sigma ^ { n } \omega } * \tilde { \Gamma } _ { n, \omega } * \Delta _ { 2, \sigma ^ { n } \omega } \cdots \tilde { \Gamma } _ { n, \omega } * \Delta _ { 2 ^ { m + 1 } - 1, \sigma ^ { n } \omega } * \tilde { \Gamma } _ { n, \omega },$$
then let m →∞ to conclude the proof.
## References
[Ave74] Andr´ e Avez. Th´ eor`me de Choquet-Deny pour les groupes ` croissance non e a exponentielle. C. R. Acad. Sci. Paris S´r. A e , 279:25-28, 1974.
[BG00] L. Bartholdi and R. I. Grigorchuk. On the spectrum of Hecke type operators related to some fractal groups. Tr. Mat. Inst. Steklova , 231(Din. Sist., Avtom. i Beskon. Gruppy):5-45, 2000.
[BG ˇ 03] S Laurent Bartholdi, Rostislav I. Grigorchuk, and Zoran ˇ uni ´ . Branch groups. S k In Handbook of algebra, Vol. 3 , pages 989-1112. North-Holland, Amsterdam, 2003.
[CN10] Julien Cassaigne and Fran¸ cois Nicolas. Factor complexity. In Combinatorics, automata and number theory , volume 135 of Encyclopedia Math. Appl. , pages 163-247. Cambridge Univ. Press, Cambridge, 2010.
[Cor14] Yves Cornulier. Groupes pleins-topologiques [d'apr`s e Matui, Juschenko, Monod,...]. Ast´ erisque , (361):Exp. No. 1064, 2014. S´minaire Bourbaki. Vol. e 2012/2013.
[EM13] G´ abor Elek and Nicolas Monod. On the topological full group of a minimal Cantor Z 2 -system. Proc. Amer. Math. Soc. , 141(10):3549-3552, 2013.
| [Gor74] | A. P. Gorjuˇkin. s Imbedding of countable groups in 2-generator simple groups. Mat. Zametki , 16:231-235, 1974. |
|-----------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [Got46] | W. H. Gottschalk. Almost period points with respect to transformation semi- groups. Ann. of Math. (2) , 47:762-766, 1946. |
| [GPS99] | Thierry Giordano, Ian F. Putnam, and Christian F. Skau. Full groups of Cantor minimal systems. Israel J. Math. , 111:285-320, 1999. |
| [Gri84] | R. I. Grigorchuk. Degrees of growth of finitely generated groups and the theory of invariant means. Izv. Akad. Nauk SSSR Ser. Mat. , 48(5):939-985, 1984. |
| [Hal74] | P. Hall. On the embedding of a group in a join of given groups. J. Austral. Math. Soc. , 17:434-495, 1974. Collection of articles dedicated to the memory of Hanna Neumann, VIII. |
| [JM13] | Kate Juschenko and Nicolas Monod. Cantor systems, piecewise translations and simple amenable groups. Ann. of Math. (2) , 178(2):775-787, 2013. |
| [KV83] | V. A. Ka˘manovich ı and A. M. Vershik. Random walks on discrete groups: boundary and entropy. Ann. Probab. , 11(3):457-490, 1983. |
| [Mat06] | Hiroki Matui. Some remarks on topological full groups of Cantor minimal systems. Internat. J. Math. , 17(2):231-251, 2006. |
| [Mat13] | Hiroki Matui. Some remarks on topological full groups of Cantor minimal systems II. Ergodic Theory Dynam. Systems , 33(5):1542-1549, 2013. |
| [MB14] | Nicol´s a Matte Bon. Subshifts with slow complexity and simple groups with the Liouville property. Geom. Funct. Anal, to appear , 2014. arXiv:1111.0512. |
| [Que87] | Martine Queff´lec. e Substitution dynamical systems-spectral analysis , volume 1294 of Lecture Notes in Mathematics . Springer-Verlag, Berlin, 1987. |
| [Sch76] | Paul E. Schupp. Embeddings into simple groups. J. London Math. Soc. (2) , 13(1):90-94, 1976. |
| [vD90] | Eric K. van Douwen. Measures invariant under actions of F 2 . Topology Appl. , 34(1):53-68, 1990. |
| [Vor10] | Ya. Vorobets. On a substitution subshift related to the Grigorchuk group. Tr. Mat. Inst. Steklova , 271(Differentsialnye Uravneniya i Topologiya. II):319- 334, 2010. |
| [Vor12] | Yaroslav Vorobets. Notes on the Schreier graphs of the Grigorchuk group. In Dynamical systems and group actions , volume 567 of Contemp. Math. , pages 221-248. Amer. Math. Soc., Providence, RI, 2012. |
| null |
[
"Nicolás Matte Bon"
] |
2014-08-04T18:42:11+00:00
|
2014-08-04T18:42:11+00:00
|
[
"math.DS",
"math.GR"
] |
Topological full groups of minimal subshifts with subgroups of intermediate growth
|
We show that every Grigorchuk group $G_\omega$ embeds in (the commutator
subgroup of) the topological full group of a minimal subshift. In particular,
the topological full group of a Cantor minimal system can have subgroups of
intermediate growth, a question raised by Grigorchuk; it can also have finitely
generated infinite torsion subgroups, as well as residually finite subgroups
that are not elementary amenable, answering questions of Cornulier. By
estimating the word-complexity of this subshift, we deduce that every
Grigorchuk group $G_\omega$ can be embedded in a finitely generated simple
group that has trivial Poisson boundary for every simple random walk.
|
1408.0763v3
|
## A rotating string model versus baryon spectra
## Jacob Sonnenschein and Dorin Weissman
The Raymond and Beverly Sackler School of Physics and Astronomy, Tel Aviv University, Ramat Aviv 69978, Israel
E-mail:
[email protected] , [email protected]
Abstract: We continue our program of describing hadrons as rotating strings with massive endpoints. In this paper we propose models of baryons and confront them with the baryon Regge trajectories. We show that these are best fitted by a model of a single string with a quark at one endpoint and a diquark at the other. This model is preferred over the Y-shaped string model with a quark at each endpoint. We show how the model follows from a stringy model of the holographic baryon which includes a baryonic vertex connected with N c strings to flavor probe branes. From fitting to baryonic data we find that there is no clear evidence for a non-zero baryonic vertex mass, but if there is such a mass it should be located at one of the string endpoints. The available baryon trajectories in the angular momentum plane ( J, M 2 ), involving light, strange, and charmed baryons, are rather well fitted when adding masses to the string endpoints, with a single universal slope α ′ = 0.95 GeV -2 . Most of the results for the quark masses are then found to be consistent with the results extracted from the meson spectra in [1], where the value of the slope emerging from the meson fits was found to be 0.90 GeV -2 . In the plane of radial excitations, ( n, M 2 ), we also find a good agreement between the meson and baryon slopes. The flavor structure of the diquark is examined, where our interest lies in particular on baryons composed of more than one quark heavier than the u and d quarks. For these baryons we present a method of checking the holographic interpretation of our results.
Contents
| 1 | Introduction | Introduction | 2 |
|-----|---------------------------------------------------------------|--------------------------------------------------------------------------|-----|
| 2 | Holographic stringy baryons | Holographic stringy baryons | 4 |
| | 2.1 The stringy models in flat space-time and their stability | 2.1 The stringy models in flat space-time and their stability | 9 |
| 3 | Basic theoretical model | Basic theoretical model | 11 |
| | 3.1 Classical rotating string with massive endpoints | 3.1 Classical rotating string with massive endpoints | 11 |
| | | 3.1.1 Generalizations: Different masses, Y-shape model, and central mass | 13 |
| 4 | Fitting models | Fitting models | 14 |
| | 4.1 | Rotating string model | 14 |
| | | 4.1.1 Rotating Y-shape with central mass | 15 |
| | 4.2 | Fitting procedure | 15 |
| 5 | Fit results | Fit results | 15 |
| | 5.1 | Y-shape and central mass | 16 |
| | 5.2 | Symmetric vs. imbalanced string | 17 |
| | 5.3 | Light quark baryons | 18 |
| | | 5.3.1 Trajectories in the ( J, M 2 ) plane | 19 |
| | | 5.3.2 Trajectories in the ( n, M 2 ) plane | 19 |
| | 5.4 | Strange baryons | 19 |
| | 5.5 | Charmed baryons | 21 |
| 6 | Summary of results | Summary of results | 22 |
| | 6.1 Comparison with meson fits | 6.1 Comparison with meson fits | 24 |
| | 6.2 Structure of the baryon in the quark diquark model | 6.2 Structure of the baryon in the quark diquark model | 25 |
| 7 | Conclusions and future directions | Conclusions and future directions | 28 |
| A | Individual | trajectory fits | 30 |
| | A.1 The states used in the fits | A.1 The states used in the fits | 30 |
| | A.2 Trajectories in the ( J, M 2 ) plane | A.2 Trajectories in the ( J, M 2 ) plane | 32 |
| | A.2.1 | Light quark baryons | 32 |
| | A.2.2 | Strange baryons | 33 |
| | A.2.3 | Charmed baryons | 35 |
| | A.2.4 | Charmed-strange baryons: The Ξ c | 36 |
| | A.3 | Trajectories in the ( n, M 2 ) plane | 37 |
| | A.3.1 Light quark baryons | A.3.1 Light quark baryons | 37 |
## 1 Introduction
The old idea that hadrons admit a stringy behavior has been reincarnated in recent years in the context of the holographic duality between gauge and string theories. Although the main focus of holography has been the description of properties of operators of the boundary gauge theory in terms of fields that reside either in the bulk or on flavor branes, strings in the bulk and strings ending on flavor branes have also played a major role in dualizing the gauge dynamics. For instance strings and D 1 branes with fixed endpoints on the boundary where shown to be the duals of Wilson and 't Hooft lines respectively. In [1] it was argued that the basic property of the mesonic spectra, the Regge trajectories, cannot generically be accounted for by fields (in this case the fluctuations of the flavor branes) whereas holographic rotating strings naturally admit this behavior. Using a map developed in [2] we transformed the holographic strings of mesons into rotating strings with massive endpoints in flat spacetime. This model, which is characterized by the string tension, the endpoint masses, and an intercept, was compared to PDG data and was found to be a universal setup describing mesons built from light quarks as well as those constructed from heavy ones.
It is very well known that mesons in nature admit Regge trajectories, but it is a lesser known fact that baryons also furnish Regge trajectories[3][4][5]. Thus, the spectra of baryons that follow from holography should admit such behavior. Like for the mesons, there are two candidates for holographic duals of the baryons. Namely, they are either a field configuration or a string configuration. It was shown [6] that in terms of the former, baryons correspond to flavor instantons which are a static configuration of the 5 dimensional flavor gauge theory that resides on the N f flavor branes, compactified, for instance in the Sakai-Sugimoto model [7], on an S 4 . It was worked out, mainly in the context of a generalization of that model, that these instantons can adequately describe the static properties of baryons, and be responsible for appropriate (large N c ) nuclear interactions and nuclear matter [8][9][10]. However, the spectrum of these instantons does not admit the Regge trajectories behavior neither for M 2 as a function of the angular momentum J , nor as a function of the radial excitation number n . Therefore, we are led again to consider strings as the holographic duals of baryons.
Since a single string that ends on the boundary or on a flavor brane corresponds to an external or a dynamical quark respectively, a stringy baryonic configuration has to connect N c strings. It was shown in [11] for the AdS 5 × S 5 background and later for confining backgrounds [12][13] that a D p brane that wraps a non-trivial p cycle with an RR flux of N c flowing out of it, must be connected to N c strings. The other end of each of these strings can be either on the boundary or on the flavor brane, thus constituting an external or dynamical baryon respectively. Whereas in the original model [11] the baryonic vertex is a D 5 wrapping the fluxed S 5 , in the confining models of [7] it is a D 4 brane wrapping an S 4 , and in the
model of [14] it is a D 3 wrapping the three cycle of the deformed conifold. The location of the baryonic vertex in the radial direction and on the flavor brane world-volume coordinates is determined by minimizing the energy of the configuration [15]. In section 2 we briefly review two possible setups of stringy baryons: the spherically symmetric one and the totally asymmetric configuration.
As we did in [1] for the mesons while following [2], we map the baryonic holographic configurations to configurations of strings with a baryonic vertex and massive endpoints in flat space-time. In this paper, like in [1], we take these stringy configurations in flat space-time as our theoretical models to compare with experimental data. We would like to emphasize that we do not provide a systematic controlled transformation from the holographic domain characterized by large N c , large λ , and N f N c << 1 to the realistic regime of N c = 3, λ of order one, and N f = 6. The comparison between the predictions of our model and the experimental data enables us to decide what is the preferable configuration (in terms of the location of the baryonic vertex) and furthermore to extract the parameters that characterize the models. These parameters are the string tension T (or equivalently its inverse, the slope of the trajectories α ′ ), the endpoint masses m sep , the mass of the baryonic vertex m bv and the intercept a . Whereas the former parameters show up in the classical stringy model, the intercept emerges only upon quantization. It is well known that for a string with no massive endpoints, namely the case of the linear Regge trajectory, the passage from the classical to quantum trajectories is via the replacement J = α E ′ 2 → J + n = α E ′ 2 + , where a n is the quantum radial excitation number. The relevant questions to our analysis are what is the theoretical value of the intercept for the massless case and moreover how does it depend on the string endpoint masses. In [16] it was found, somewhat surprisingly, that for the case of a rotating bosonic string with angular momentum in a single plane, the intercept of the massless case is independent of the dimensionality of the space-time D and takes the value of D -2 24 + 26 -D 24 = 1. The first contribution to the intercept is the usual 'Casimir term' and the second one is the Polchinski-Strominger term. For the rotating string with massive endpoints a similar determination of the intercept has not yet been written down even though certain aspects of the quantization of such a system have been addressed [17][18][19].
The main result of the paper is that the best model providing a universal setup of the baryon is the model of a single string with a diquark at one endpoint and a quark at the other. This model is preferred over the model of a Y-shaped string with a quark at each endpoint. There is no clear evidence for a non-zero baryonic vertex mass, but if there is such a mass it should be located at one of the string endpoints. We see that we can fit the available baryon trajectories in the angular momentum plane ( J, M 2 ) rather well when adding masses to the endpoints, and we can do it, if we wish, with a single universal slope α ′ = 0 95 GeV . -2 . Most of the results for the quark masses are then found to be consistent with the results extracted from the meson spectra in [1], where the value of the slope emerging from the meson fits - 0.90 GeV -2 - is close to the value obtained here for the baryons. In the plane of string excitations, ( n, M 2 ), we fitted the trajectories of light baryons and found that there too there is a good agreement between the meson and baryon slopes.
Figure 1 . Schematic picture of holographic baryons. On the left is an external baryon with strings that end on the boundary, while on the right is a dynamical baryon with strings ending on a flavor brane.
The paper is organized as follows. In section 2 we briefly review the concept of baryonic stringy configurations in holography. Section 3 is devoted to the basic (flat space-time) theoretical model for the baryonic string with massive endpoints. In section 4 we describe the various fitting models, and then move on to the results of the fits in section 5. Section 6 begins with a summary of the results, compares them with the results of the meson fits, and offers a discussion of their implications regarding the structure and composition of the baryons. In our discussion of the structure we also look into (but do not analyze in detail) the decay modes of the baryons. In section 7 we conclude, summarize and mention several future directions of the research program.
## 2 Holographic stringy baryons
Astring that stretches from the boundary of a holographic background corresponds in the dual field theory to an external quark. Similarly a string that stretches from a flavor probe brane is a dual of a dynamical quark. The dual of the meson gauge singlet thus corresponds to a string that starts and ends either on the boundary (external) or on a flavor brane (dynamical). It is thus clear that a stringy holographic baryon has to include N c strings that are connected together and end on a flavor brane (dynamical baryon) or on the boundary (external baryon). The question is what provides the 'baryonic vertex' that connects together N c strings. In [20] it was shown that in the AdS 5 × S 5 background, which is equipped with an RR flux of value N c , a D 5 brane that wraps the S 5 has to have N c strings attach to it. This property can be generalized to other holographic backgrounds so that a D p brane wrapping a nontrivial p cycle with a flux of an RR field of value N c provides a baryonic vertex. These two possible stringy configurations are schematically depicted in figure (1). Whereas the dual of the original proposal [20] was a conformal field theory, Baryons can be constructed also
in holographic backgrounds that correspond to confining field theories. A prototype model of this nature is the model of N c D 4 branes background compactified on a circle with an N f D 8-5 D 8 U-shaped flavor branes [7]. In this model the baryonic vertex is made out of a D 4 brane that wraps an S 4 . Another model for baryons in a confining background is the deformed conifold model with D 7D 7 U-shaped flavor branes [21]. In this model the baryon is a D 3 brane that wraps the three-cycle of the deformed conifold.
The argument why a D p brane wrapping a fluxed p cycle is a baryonic vertex is in fact very simple. The world-volume action of the wrapped D p brane has the form of S = S DBI + S CS . The CS term takes the following form
$$S _ { C S } = \int _ { S _ { p } \times R _ { 1 } } \sum _ { i } c _ { p _ { i } } \wedge e ^ { F } = \int _ { S _ { p } \times R _ { 1 } } c _ { p - 1 } \wedge F = - \int _ { S _ { p } \times R _ { 1 } } F _ { p } \wedge A = - N _ { c } \int _ { R _ { 1 } } A \quad \quad ( 2. 1 )$$
where (i) the sum over i is over the RR p -forms that reside in the background, (ii) from the sum one particular RR form was chosen, the c p -1 -form that couples to the Abelian fieldstrength F, (iii) for simplicity's sake we took the p cycle to be S p , (iv) A is the Abelian connection that resides on the wrapped brane, (v) F p is the RR p -form field strength, and (vi) in the last step we have made use of the fact that ∫ S p F p = N c . This implies that there is a charge N c for the Abelian gauge field. Since in a compact space one cannot have nonbalanced charges and since the endpoint of a string carries a charge one, there must be N c strings attached to it. It is interesting to note that a baryonic vertex rather than being a 'fractional' D 0 brane of the form of a D p brane wrapping a p cycle, can also be a D 0 brane in an N c fluxed background. This is the case in the non-critical string backgrounds like [22] where there is no non-trivial cycle to wrap branes over, but an 'ordinary' D 0 brane in this background will also have a CS term of the form N c ∫ R 1 A .
Next we would like to determine the location of the baryonic vertex in the radial dimension. In particular the question is whether it is located on the flavor branes or below them. This is schematically depicted in figure (2) for the model of [7].
In [15] it was shown by minimizing the 'mechanical energy' of the N c strings and the wrapped brane that it is preferable for the baryonic vertex to be located on the flavor branes in the model of [7]. For the baryonic vertex of the model of [14] it was shown that if the tip of the U-shaped flavor brane is close to the lowest point of the deformed conifold the baryonic vertex does dissolve in the flavor branes. It is interesting to note that for a background that corresponds to the deconfining phase of the dual gauge theory the baryonic vertex falls into the 'black hole' and thus the baryon dissolves.
The locations of the baryonic vertex and the ends of the N c strings on the flavor branes is a dynamical issue. In figure (3) we draw two possible setups. In one the baryonic vertex is located at the center and the N c ends of the strings are located around it in a spherically symmetric way. In the 'old' stringy model of baryons for three colors this is the analog of the Y-shape configuration that we will further discuss in the sections describing the flat space-time models. Another possibility is that of a baryonic vertex connected by N c -1 very
Figure 2 . The location of the baryonic vertex along the radial direction of the Sakai-Sugimoto model. u 0 is the tip of the flavor branes and u kk is the 'confining scale' of the model.
Figure 3 . The location of the baryonic vertex on the flavor brane and the corresponding configuration of the baryon for some large N c . When N c = 3, the left is the analog of the quark-diquark flat spacetime model, and the right the analog of the Y-shape model.
short strings and with one long string to the flavor branes. Since the product of N c -1 fundamental representations includes the anti-fundamental one, this configuration can be viewed as a string connecting a quark with an anti-quark. For the case of N c = 3 this is the analog of what will be discussed below as the quark-diquark stringy configuration. This
Figure 4 . The holographic setup downlifted to N c = 3 of a quark and a diquark is mapped to a similar configuration in flat space-time. The vertical segments of the holographic string is mapped into masses of the endpoints.
latter string configuration (for any N c ) is similar to the stringy meson, but there is a crucial difference, which is the fact that the stringy baryon includes a baryonic vertex.
It was shown in [2] that the classical rotating holographic stringy configuration of the meson can be mapped into that of a classical rotating bosonic string in flat space-time with massive endpoints. A similar map applies also to holographic stringy baryons that can be transformed into stringy baryons in flat space time with massive endpoints. We will proceed now to discuss this map for the central and asymmetric layouts of figure (3). The asymmetric holographic configuration of a quark and N c -1 quarks on the two ends of the holographic string depicted in figure (4) is mapped into a similar stringy configuration in flat space-time where the vertical segments of the string are transferred into massive endpoints of the string. On the left hand side of the string in flat space-time there is an endpoint with mass m sep given by
$$m _ { s e p } = T \int _ { u \Lambda } ^ { u _ { f } } d u \sqrt { g _ { 0 0 } g _ { u u } }$$
where u λ is the location of the 'wall', u f is the location of the flavor brane, and g 00 and g uu are the 00 and uu components of the metric of the background.
On the right hand side the mass of the endpoint is m sep + m bv . This is the sum of the energy associated with the vertical segment of the string, just like that of the left hand side, and the mass of the baryonic vertex. Note that even though on the right endpoint of the string there are in fact two endpoints or 'two quarks' the mass is that of a single quark since there is a single vertical string segment, belonging to the string connecting the baryonic vertex with the lone quark at the other endpoint. This string setup is obviously very similar to that of the meson. The only difference is the baryonic vertex that resides at the diquark endpoint. Since we do not know how to evaluate the mass of the baryonic vertex, it will be left as a free parameter to determine by the comparison with data. Our basic task in this
Figure 5 . The map between the holographic central configuration (for N c = 3) and the Y-shaped string in flat space-time.
case will be to distinguish between two options: (i) the mass of the baryonic vertex is much lighter than the endpoint mass, m bv << m sep , in which case the masses at the two endpoints will be roughly the same, and (ii) an asymmetric setup with two different masses if the mass of the baryonic vertex cannot be neglected.
The configuration with a central baryonic vertex can be mapped into the analog of a Y-shaped object with N c massive endpoints and with a central baryonic vertex of mass m c = m bv + N m c sep . The factor of N c is due to the fact that there are N c strings that stretch from it vertically from the flavor brane to the 'wall', as can be seen in figure (5) for the case of N c = 3. In this case, regardless of the ratio between the mass of the baryonic vertex and that of the string endpoint, there is a massive center which is at least as heavy as three sting endpoints.
So far we have considered stringy holographic baryons that attach to one flavor brane. In holographic backgrounds one can introduce flavor branes at different radial locations thus corresponding to different string endpoint masses, or different quark masses. For instance, a setup that corresponds to u and d quarks of the same m sep mass, a strange s quark, a charm c quark, and a bottom b quark is schematically drawn in figure (6). A B meson composed of a bottom quark and a light ¯ u/d ¯ anti-quark was added to the figure.
Correspondingly there are many options of holographic stringy baryons that connect to different flavor branes. This obviously relates to baryons that are composed of quarks of different flavor. In fact there are typically more than one option for a given baryon. In addition to the distinction between the central and quark-diquark configurations there are more than one option just to the latter configuration. We will demonstrate this situation in section 6.2, focusing on the case of the doubly strange Ξ baryon ( ssd or ssu ). The difference between the the two holographic setups is translated to the two options of the diquark being
Figure 6 . Holographic setup with flavor branes associated with the u, d quark, s , c and b quarks.
either composed of two s quarks, whereas the other setup features a ds or us diquark. Rather than trying to determine the preferred configuration from holography we will use a comparison with experimental data to investigate this issue.
## 2.1 The stringy models in flat space-time and their stability
There are some words to be said about the models emerging when mapping the holographic rotating strings to flat space-time. In section 3, we will write the equations of motion of the string with massive endpoints in flat space-time and present the rotating solution. In this section, we briefly discuss another matter: the stability of the rotating solution. Specifically, we claim that the Y-shape model of the baryon is (classically) unstable. In our analysis of the spectrum we disregard this potential instability of the model and use the expressions for the energy and angular momentum of the unperturbed rotating solution of the Y-shape model as one of our fitting models, but it is important to remember that there is this theoretical argument against it as a universal setup for baryons before we test it out as a phenomenological model.
Other than the Y-shape and quark-diquark configurations which we analyze, there are two more possible stringy models for the baryon when considering a purely flat space-time point of view. These are drawn, together with the quark-diquark and Y-shape models, in figure (7). The additional models are the two-string model where one of the quarks is located at the center of the baryon, and the other two are attached to it by a string. The second is the ∆-shape model, in which each quark is connected to the other two. It can be looked at as a closed string with three points along it that carry finite momentum (which may be either massive or massless). While the two-string model may have its analog in holography, with
Figure 7 . Four different stringy models for the baryon in flat space-time. (a) is the quark-diquark model, (b) the two-string model, (c) the three-string Y-shape, and (d) the closed ∆-shape. (a) and (c) are the models discussed in the preceding discussion of holographic strings and their mappings to flat space-time.
the baryonic vertex lying with the quark at the center of mass, the ∆-shaped string cannot be built if we impose the constraint that the three quarks should all be connected to a baryonic vertex.
Two independent analyses of the three-string model [23][24] concluded that the rotating solution of the Y-shape configuration is unstable, even before taking quantum effects into account. In another work, it was found that the instability does not show itself in first order perturbation theory [25], but the claim of the unstable nature of the Y-shape model has been verified using numerical methods in [26], where the instability was observed in simulations and its dependence on endpoint masses was examined. To summarize the results, in the three-string Y-shape model, a perturbation to one of the three arms would cause it to shorten until eventually the Y-shape collapses to a form like that of the straight two-string model. From this model in turn a different kind of instability is expected [23][24]. The quark in the center of the baryon will move away from the center of mass given a small perturbation and as it approaches one of the quarks at the endpoints quantum effects will induce a collapse to the single string quark-diquark model, it being energetically favorable for two of the quarks to form an diquark bound state in the anti-fundamental color representation. It would seem that all other models have an instability that would cause them to eventually collapse to the quark-diquark configuration as two quarks get close enough to each other. 1
From a phenomenological point of view the models differ mainly in their prediction of the slope of the Regge trajectory. Assuming the strings in baryons have the same tension as those in mesons, we can see which of the models offers the best match. We will see that it is the configuration we know to be stable, that of a single string with a quark and a diquark at
1 More discussion and detailed analyses of the different stringy models of the baryon and their stability are found in the work of G.S. Sharov, most recently in [27].
its endpoints. Therefore, our fitting analysis will be focused on this model. 2
## 3 Basic theoretical model
As explained in the introduction our theoretical model is not the holographic stringy model, but a model in flat space-time that incorporates two ingredients motivated by the holographic picture. Namely, they are massive particles at the ends of the strings and a baryonic vertex that connects N c = 3 strings.
## 3.1 Classical rotating string with massive endpoints
We describe the string with massive endpoints (in flat space-time) by adding to the NambuGoto action,
$$S _ { N G } & = - T \int d \tau d \sigma \sqrt { - h } \\ h _ { \alpha \beta } & \equiv \eta _ { \mu \nu } \partial _ { \alpha } X ^ { \mu } \partial _ { \beta } X ^ { \nu }$$
a boundary term - the action of a massive chargeless point particle
$$S _ { p p } = - m \int d \tau \sqrt { - \dot { X } ^ { 2 } } \\ \dot { X } ^ { \mu } \equiv \partial _ { \tau } X ^ { \mu }$$
$$( 3. 2 )$$
at both ends. There can be different masses at the ends, but here we assume, for simplicity's sake, that they are equal. We also define σ = ± l to be the boundaries, with l an arbitrary constant with dimensions of length.
The variation of the action gives the bulk equations of motion
$$\partial _ { \alpha } \left ( \sqrt { - h } h ^ { \alpha \beta } \partial _ { \beta } X ^ { \mu } \right ) = 0$$
and at the two boundaries the condition
$$T \sqrt { - h } \partial ^ { \sigma } X ^ { \mu } \pm m \partial _ { \tau } \left ( \frac { \dot { X } ^ { \mu } } { \sqrt { - \dot { X } ^ { 2 } } } \right ) = 0$$
It can be shown that the rotating configuration
$$X ^ { 0 } = \tau, X ^ { 1 } = R ( \sigma ) \cos ( \omega \tau ), X ^ { 2 } = R ( \sigma ) \sin ( \omega \tau )$$
solves the bulk equations (3.3) for any choice of R σ ( ). We will use the simplest choice, R σ ( ) = σ , from here on. 3 Eq. (3.4) reduces then to the condition that at the boundary,
$$\frac { T } { \gamma } = \gamma m \omega ^ { 2 } l$$
2 The quark-diquark model was also used to analyze the baryon spectrum in [28]. Another discussion of diquarks as building blocks for hadrons is in [29].
3 Another common choice is X 0 = τ, x 1 = sin( σ ) cos( ωτ , X ) 2 = sin( σ ) sin( ωτ ).
with γ -1 ≡ √ 1 -ω l 2 2 4 .
We then derive the Noether charges associated with the Poincar´ invariance of the ace tion, which include contributions both from the string and from the point particles at the boundaries. Calculating them for the rotating solution, we arrive at the expressions for the energy and angular momentum associated with this configuration:
$$E = - p _ { 0 } = 2 \gamma m + T \int _ { - l } ^ { l } \frac { d \sigma } { \sqrt { 1 - \omega ^ { 2 } \sigma ^ { 2 } } }$$
$$J = J ^ { 1 2 } = 2 \gamma m \omega l ^ { 2 } + T \omega \int _ { - l } ^ { l } \frac { \sigma ^ { 2 } d \sigma } { \sqrt { 1 - \omega ^ { 2 } \sigma ^ { 2 } } } \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \ \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \cdots \, \quad \quad \quad$$
Solving the integrals, and defining q ≡ ωl -physically, the endpoint velocity - we write the expressions in the form
$$E = \frac { 2 m } { \sqrt { 1 - q ^ { 2 } } } + 2 T l \frac { \arcsin ( q ) } { q }$$
$$J = 2 m l \frac { q } { \sqrt { 1 - q ^ { 2 } } } + T l ^ { 2 } \left ( \frac { \arcsin ( q ) - q \sqrt { 1 - q ^ { 2 } } } { q ^ { 2 } } \right )$$
The terms proportional to m are the contributions from the endpoint masses and the term proportional to T is the string's contribution. These expressions are supplemented by condition (3.6), which we rewrite as
$$T l & = \frac { m q ^ { 2 } } { 1 - q ^ { 2 } }$$
This last equation can be used to eliminate one of the parameters l, m, T, and q from J and E . Eliminating the string length from the equations we arrive at the final form
$$E = 2 m \left ( \frac { q \arcsin ( q ) + \sqrt { 1 - q ^ { 2 } } } { 1 - q ^ { 2 } } \right )$$
$$J = \frac { m ^ { 2 } } { T } \frac { q ^ { 2 } } { ( 1 - q ^ { 2 } ) ^ { 2 } } \left ( \arcsin ( q ) + q \sqrt { 1 - q ^ { 2 } } \right )$$
These two equations are what define the Regge trajectories of the string with massive endpoints. They determine the functional dependence of J on E , where they are related through the parameter 0 ≤ q < 1 ( q = 1 when m = 0). Since the expressions are hard to make sense of in their current form, we turn to two opposing limits - the low mass and the high mass approximations.
In the low mass limit where the endpoints move at a speed close to the speed of light, so q → 1, we have an expansion in ( m/E ):
$$\xi, \cdots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \\ J = \alpha ^ { \prime } \, E ^ { 2 } \left ( 1 - \frac { 8 \sqrt { \pi } } { 3 } \left ( \frac { m } { E } \right ) ^ { 3 / 2 } + \frac { 2 \sqrt { \pi ^ { 3 } } } { 5 } \left ( \frac { m } { E } \right ) ^ { 5 / 2 } + \cdots \right )$$
4 Notice that in addition to the usual term γm for the mass, the tension that balances the 'centrifugal force' is T γ .
from which we can easily see that the linear Regge behavior is restored in the limit m → 0, and that the first correction is proportional to √ E . The Regge slope α ′ is related to the string tension by α ′ = (2 πT ) -1 .
The high mass limit, q → 0, holds when ( E -2 m / m ) 2 glyph[lessmuch] 1. Then the expansion is
$$J = \frac { 4 \pi } { 3 \sqrt { 3 } } \alpha ^ { \prime } \, m ^ { 1 / 2 } ( E - 2 m ) ^ { 3 / 2 } + \frac { 7 \pi } { 5 4 \sqrt { 3 } } \alpha ^ { \prime } \, m ^ { - 1 / 2 } ( E - 2 m ) ^ { 5 / 2 } + \cdots$$
## 3.1.1 Generalizations: Different masses, Y-shape model, and central mass
The generalization of the symmetric model to the case where there are two different masses is simple. The angular momentum and energy were calculated by summing two equal contributions - from two halves of the string ( σ ∈ ( -l, 0) and σ ∈ (0 , l )) and two identical point particles. To generalize this we simply replace the factor of two with a sum over two different, but similar, contributions.
The two masses would still rotate with the same angular velocity, ω , and the string tension remains the same for both string segments. The difference between the two contributions stems from different endpoint velocities, q i , and the different radii of rotation, related to the velocities by ωl i = q i .
The energy and angular momentum in this case are, then,
$$E = \sum _ { i = 1, 2 } m _ { i } \left ( \frac { q _ { i } \arcsin ( q _ { i } ) + \sqrt { 1 - q _ { i } ^ { 2 } } } { 1 - q _ { i } ^ { 2 } } \right )$$
$$J = \sum _ { i = 1, 2 } \pi \alpha ^ { \prime } \, m _ { i } ^ { 2 } \frac { q _ { i } ^ { 2 } } { ( 1 - q _ { i } ^ { 2 } ) ^ { 2 } } \left ( \arcsin ( q _ { i } ) + q _ { i } \sqrt { 1 - q _ { i } ^ { 2 } } \right )$$
The velocities q 1 and q 2 can be related using the boundary condition (3.6), from which we have
$$\frac { T } { \omega } = m _ { 1 } \frac { q _ { 1 } } { 1 - q _ { 1 } ^ { 2 } } = m _ { 2 } \frac { q _ { 2 } } { 1 - q _ { 2 } ^ { 2 } }$$
With q 1 and q 2 thus related, the massive Regge trajectory is obtained from the parametric curve
$$E = E \left ( q _ { 1 }, q _ { 2 } ( q _ { 1 } ) \right ) \quad J = J \left ( q _ { 1 }, q _ { 2 } ( q _ { 1 } ) \right )$$
where 0 ≤ q 1 < 1
The equations of motion for the three segments of the Y-shaped string are unchanged from the simple straight string. The only difference is an added boundary condition at the point where the three strings connect. While this added condition is important when analyzing the stability of the model, as done e.g. in [23], it holds trivially in the unperturbed rotating solution. Therefore the only adjustment we need make to eqs. (3.16) and (3.17) to get to the trajectories of the Y-shape model is in the summation index i -we now sum over three contributions from three string segments and three endpoint masses, so i = 1 2 3. , ,
In the case where we have three identical end point masses we have
$$E = 3 m \left ( \frac { q \arcsin ( q ) + \sqrt { 1 - q ^ { 2 } } } { 1 - q ^ { 2 } } \right )$$
$$J = \frac { 3 } { 2 } \frac { m ^ { 2 } } { T } \frac { q ^ { 2 } } { ( 1 - q ^ { 2 } ) ^ { 2 } } \left ( \arcsin ( q ) + q \sqrt { 1 - q ^ { 2 } } \right )$$
This model is completely equivalent (in terms of the Regge trajectory) with a single string model with the same total mass at the endpoints and with an effective higher string tension. That is to say, a Y-shaped string with three identical masses m Y and the string tension T Y has the same exact Regge behavior as a string with two masses m 1 = m 2 = 3 m / Y 2 and the tension T , provided T Y = 3 2 T . In particular, in the massless case the Y-shape model yields the result
$$J = \frac { 2 } { 3 } \alpha ^ { \prime } \, E ^ { 2 }$$
Another generalization would be to add a central mass to the string around which the string rotates. This mass would be stationary so its only contribution would be a constant shift in the energy, which amounts to the modification E → E -m bv in previous equations. 5
## 4 Fitting models
## 4.1 Rotating string model
We define the linear fit by
$$J + n = \alpha ^ { \prime } \, E ^ { 2 } + a$$
where the fitting parameters are the slope α ′ and the intercept a .
For the massive fit, we use the general expressions for the mass and angular momentum of the rotating string, eqs. (3.16) and (3.17), for the case of two different masses, and we add to them, by hand, an intercept and an extrapolated n dependence, assuming the same replacement of J → J + n -a .
$$E = \sum _ { i = 1, 2 } m _ { i } \left ( \frac { q _ { i } \arcsin ( q _ { i } ) + \sqrt { 1 - q _ { i } ^ { 2 } } } { 1 - q _ { i } ^ { 2 } } \right )$$
$$J + n = a + \sum _ { i = 1, 2 } \pi \alpha ^ { \prime } \, m _ { i } ^ { 2 } \frac { q _ { i } ^ { 2 } } { ( 1 - q _ { i } ^ { 2 } ) ^ { 2 } } \left ( \arcsin ( q _ { i } ) + q _ { i } \sqrt { 1 - q _ { i } ^ { 2 } } \right )$$
With the relation between q 1 and q 2 as in eq. (3.18):
$$\frac { T } { \omega } = m _ { 1 } \frac { q _ { 1 } } { 1 - q _ { 1 } ^ { 2 } } = m _ { 2 } \frac { q _ { 2 } } { 1 - q _ { 2 } ^ { 2 } }$$
5 The name m bv stems from a possible identification of such a central mass with the presence of a baryonic vertex encountered in the holographic models.
With the two additions of n and a , the two equations reduce to that of the linear fit in (4.1) in the limit where both masses are zero.
Now the fitting parameters are a and α ′ as before, as well as the the two endpoint masses m 1 and m 2 .
## 4.1.1 Rotating Y-shape with central mass
We also examine fits using the Y-shape model described in section 3.1.1, as well as the model where the string rotates about a stationary mass, m bv .
The Y-shape fit uses the same expressions as the massive fit with the summation changed from i = 1 2 to , i = 1 2 3. The central mass , , m bv may be inserted to the fits by the replacement E → E -m bv in eq. (4.2), the same way we have inserted the intercept a to J . Explicitly, the expressions then are
$$E = m _ { b v } + \sum _ { i = 1 } ^ { 3 } m _ { i } \left ( \frac { q _ { i } \arcsin ( q _ { i } ) + \sqrt { 1 - q _ { i } ^ { 2 } } } { 1 - q _ { i } ^ { 2 } } \right )$$
$$J + n = a + \sum _ { i = 1 } ^ { 3 } \pi \alpha ^ { \prime } \, m _ { i } ^ { 2 } \frac { q _ { i } ^ { 2 } } { ( 1 - q _ { i } ^ { 2 } ) ^ { 2 } } \left ( \arcsin ( q _ { i } ) + q _ { i } \sqrt { 1 - q _ { i } ^ { 2 } } \right )$$
The relation between the three velocities q i is again through
$$\frac { T } { \omega } = m _ { i } \frac { q _ { i } } { 1 - q _ { i } ^ { 2 } }$$
for i = 1 2 3. In the limit where all endpoint masses are zero, these equations reduce to , ,
$$J + n - a = \frac { 2 } { 3 } \alpha ^ { \prime } \left ( E - m _ { b v } \right ) ^ { 2 }$$
## 4.2 Fitting procedure
We measure the quality of a fit by the dimensionless quantity χ 2 , which we define, as we did in [1], by
$$\chi ^ { 2 } = \frac { 1 } { N - 1 } \sum _ { i } \left ( \frac { M _ { i } ^ { 2 } - E _ { i } ^ { 2 } } { M _ { i } ^ { 2 } } \right ) ^ { 2 }$$
M i and E i are, respectively, the measured and calculated value of the mass of the i -th particle, and N the number of points in the trajectory.
## 5 Fit results
This section offers a discussion of the fit results for the baryons. We start by briefly discussing the results using the Y-shaped string model for the baryon, and the effects of including a massive baryonic vertex at the center of mass, via the replacement M → M -m bv . The results prove to be against these options, so the rest of the section discusses the results when
using the quark-diquark model for the baryons. This means we describe the baryons simply as a single string with two masses at its endpoints.
We separate the results into three sections, one for the light quark baryons, the next for strange baryons, and the third for charmed baryons. In the light baryon section we also examine the radial trajectories of the N and ∆ baryons. The rest of the sections have trajectories only in the angular momentum plane ( J, M 2 ).
The detailed individual trajectory fits and the specification of all the states used are found in appendix A. The experimental data used in this paper is taken from the Particle Data Group's Review of Particle Physics[30].
## 5.1 Y-shape and central mass
As explained in section 3.1.1, the Y-shaped string model is equivalent in terms of the Regge trajectories to a single string model with a higher effective string tension. In the picture we have before adding endpoint masses, we may assume (as a phenomenological model) linear trajectories for both the meson and baryon trajectories,
$$J = \alpha ^ { \prime } \, M ^ { 2 } + a$$
with different slopes, α ′ b and α ′ m for the baryons and mesons, respectively. Now, the assumption that the baryons are Y-shaped strings while the mesons are straight single strings, and that there is a single universal string tension, would lead us to expect the relation
$$\alpha ^ { \prime } _ { \, b } & = \frac { 2 } { 3 } \alpha ^ { \prime } _ { \, m }$$
between the baryon and meson Regge slopes. When we fit the data we see no such relation. What we see in our results is that in fact, the meson and baryon slopes are very similar -α ′ b ≈ α ′ m -supporting the same single string model for the baryons that was used for the mesons. This also excludes the triangle-shaped closed string baryon, which we have not analyzed in detail but predicts an effective slope α ′ b of between 3 8 α ′ m and 1 2 α ′ m , depending on the type of solution [31].
Our addition of endpoint masses does not change this picture, as we would still need to see a similar relation between the baryon and meson slopes, with the baryon slope being consistently lower. The slopes obtained from the baryon trajectory fits, which we present in the next sections, still remain too high for the Y-shape model of the baryon to be consistent with experimental data. In many cases the baryon slope is actually higher than the meson slope.
As for the baryonic vertex mass, the assumption that there is a central mass that contributes to the total mass of a state but not to the angular momentum was also found to be unsupported by the data, neither in the Y-shape model nor in the straight string model. 6 We
6 The two-string model discussed in section 2.1 with one quark at the center of mass has the same slope as the quark-diquark model, but we exclude it here (after excluding it on grounds of its instability in section 2.1) because we see no evidence, in any trajectory, of a massive central point.
plot one of these fits in a following section (bottom-right plot in figure (8)). This does not rule out the presence of a mass due to a holographic baryonic vertex, but means it is most likely located at the string endpoints and not at its center.
With these results in mind, we continue to present our fits using only the single string model with two masses at its endpoints.
## 5.2 Symmetric vs. imbalanced string
Now we turn to the different mass configurations in the single string model. As mentioned in the last subsection, there is no evidence to support a mass located at the center of the string. To understand the structure of the baryon we would like to be able to tell how the mass is distributed between the two endpoints, but this is information we cannot gather from the Regge trajectory fits. In the low mass approximation for the single string, the leading order correction is proportional to α ′ ( m 3 2 / 1 + m 3 2 / 2 ) √ E . This is the generalization of the low mass expansion of eq. (3.14). Therefore, for small masses - ( m 1 + m /E 2 ) glyph[lessmuch] 1 we cannot distinguish from the Regge trajectory fits alone between different configurations with equal m 3 2 / 1 + m 3 2 / 2 . There are of course higher order corrections, but our fits are not sensitive enough to them, and in practice, we see that fits with m 3 2 / 1 + m 3 2 / 2 = Const. are nearly equivalent even for masses of a few hundred MeV (when using a typical value of the slope, in the neighborhood of 0 9 GeV . -2 ). An example of this is in the bottom-left plot of figure (9).
When expanding J in E for two heavy masses, the combination in the leading term would be m 1 + m 2 . In the mid range that cannot be described accurately by either expansion there is a transition between the two different type of curves. This can be best seen in figure (10), which shows the fits of the charmed Λ c baryon. For higher masses, the variation of the intercept along these curves is big enough to be easily measured, but as the intercept takes reasonable values for both symmetric and asymmetric massive fits we cannot use this measurement to decide between the two configurations. We do not have accurate predictions for the intercept, and neither will we find emerging in our fit results remarkably consistent results for it. To give some numbers, the symmetric fit of the Λ c with 2 m = 2010 has a = 0 09, . while the asymmetric configuration which gives a comparable fit has m 1 = 1720, m 2 = 90 and a = -0 13. . The slope is roughly equal between the two fits, 1 13 GeV . -2 for the former and 1 22 GeV . -2 for the latter. Since we have no reason to prefer a = 0 09 over . a = -0 13 we . cannot use this information.
In both cases the symmetric fit where m 1 = m 2 = m gives an indication of the total mass we can add to the endpoints for a good fit of a given trajectory. The best fitting masses are on a curve in the ( m ,m 1 2 ) plane. The choice m 1 = m 2 maximizes the total mass m 1 + m 2 , whether the masses are light or heavy.
It should also be noted that for the trajectories we analyze we either have fits with low masses, where the m 3 2 / approximation is valid, or trajectories with only 3 data points where we would by default expect the optimum to be located on a curve in the ( m ,m 1 2 ) plane, seeing how there are four fitting parameters in total.
Figure 8 . The light baryon trajectory fits. Top-left: The N trajectory and fits, showing the even/odd effect. Top-right: χ 2 vs. the two masses for the N trajectory. Bottom-left: χ 2 vs. ( α , ′ 2 m ) for the N . Bottom-right: χ 2 vs. 2 m and a central vertex mass m bv for the ∆ trajectory. The χ 2 plots are for fits to the ( J, M 2 ) trajectories, and use only the even L states. The darkest areas in the χ 2 plots have χ 2 m /χ 2 l < 1, lightest areas are χ 2 m /χ 2 l < . 1 1.
While the presentation in the following sections of the results is for the symmetric fit, this does not mean that we have found it is actually preferred by the data. The symmetric fit tells us whether there is a preference for non-zero endpoint masses or not, and allows us to obtain the values of the slope for a given total endpoints mass.
If we assume χ 2 is constant along curves of the form m 3 2 / 1 + m 3 2 / 2 = Const. , then there is not much difference between the symmetric case and the case m 1 = 2 m 2 when considering the total sum m 1 + m 2 . So for a given symmetric fit with 2 m , the configuration with ( m ,m 1 2 ) = ( 2 3 m, 4 3 m ) will be almost equivalent to the one with m 1 = m 2 = m (and this is true at both high and low masses). If one of the masses is zero, say m 2 = 0, then as a rule of thumb one can take m 1 = 2 2 3 / m ≈ 0 8 . × 2 m to move from the symmetric fit result to the totally imbalanced mass configuration.
## 5.3 Light quark baryons
In the light baryon sector, we look at the N and ∆ resonances.
## 5.3.1 Trajectories in the ( J, M 2 ) plane
One of the most interesting features of the baryon Regge trajectories is the splitting of the trajectories of even and odd L states. The states with even and odd orbital angular momentum do not lie on one single trajectory, but on two parallel linear trajectories, the odd L states being higher in mass and lying above the trajectory formed by the even states. The plot in figure (8) shows this effect for the ( J, M 2 ) trajectory of the N .
In our analysis, we fit the even and odd trajectories together, with the same endpoint masses and slope, and allow the intercept to carry the difference between the even and odd states.
In this way, we get that the N trajectory is best fitted with a slope of around 0 95 GeV . -2 , and that the linear fit is optimal. Only small masses, up to a total mass of 2 m = 170 MeV, are allowed. 7 Trying a fit using only the four highest J states (two even and two odd), we achieve a weaker χ 2 dependence on the mass, and we can add a total mass of up to 640 MeV, with the slope being near 1 GeV -2 for the highest masses.
The ∆ is also best fitted by the linear, massless, trajectory, but it allows for higher masses. The maximum for it is 450 MeV. The slope, for a given mass, is lower than that of the N . It is between 0.9 GeV -2 for the linear fit, and 0.97 GeV -2 for the maximal massive fits.
As for the even-odd effect, we quantify it by looking at the difference between the intercept obtained for the even L trajectory and the one obtained for the odd L trajectory. The magnitude of the even-odd splitting is higher for the ∆ than it is for the N baryons. For the ∆, the difference in the intercept is of almost one unit a e -a o ≈ 1, while for the N it is less than half that: a e -a o ≈ 0 45. . The difference in M 2 is obtained by dividing by α ′ , so it is 0 5 GeV . -2 for the N and 1 1 GeV . 2 for the ∆.
## 5.3.2 Trajectories in the ( n, M 2 ) plane
The radial trajectories we analyze are also best fitted with small, or even zero, endpoint masses.
For the ∆ we have three states with J P = 3 2 / + . The slope is between 0.92 and 0.94 GeV -2 and the maximal allowed total mass is less than 200 MeV.
For the N we use a total of fifteen states: four with J P = 1 2 / + (the neutron/proton and higher resonances), three with 3 / 2 -, and four pairs with other J P assignments. They are all fitted with the same slope and mass. The results show a lower slope here, from 0.82 GeV -2 for the linear fit to 0 85 GeV . -2 for the highest mass fit, this time with 2 m = 425 MeV.
## 5.4 Strange baryons
In the strange section there are several trajectories we analyze.
7 Masses in what we call the 'allowed' range give a value of χ 2 that is within 10% of its optimal value for that specific trajectory.
Figure 9 . The strange baryon trajectory fits. Top-left: χ 2 vs. ( α , ′ 2 m ) for the Λ trajectory. Top-right: same for the Σ. Bottom: χ 2 vs. ( m ,m 1 2 ) for the Ξ trajectory, for α ′ = 0 950 (left) and . adjustable α ′ (right). The red curve in the bottom-left plot is m 3 / 2 1 + m 3 / 2 2 = 2 × (425) 3 / 2 .
The first is that of the Λ. There are five states in this trajectory, enough for us to see that the even-odd effect is not present - or too weak to be noticeable. The linear fit, with α ′ = 0 95 . GeV -2 , is the optimal fit, and only small masses of the order of 60 MeV are allowed at each endpoint. Even if one of the masses is zero, the mass at the other end could not exceed 100 MeV. This is a puzzling result because the results of the meson fits, which will be compared in detail to the baryon fits in a later section, point toward a mass of 200 -400 MeV for the s quark.
On the other hand, the other strange baryon trajectories we examine point to very high masses, with correspondingly high values of the slope. These are the results of the two trajectories of the Σ baryon we examine. The first has three states with J P = 1 2 / + , 3 2 / -, and 5 2 / + . The second has the parity reversed (for a given value of J ): J P = 3 2 / + , 5 / 2 -, and 7 / 2 + .
Since there are only three states per trajectory we cannot determine from the data alone whether or not there is an even-odd splitting effect present here (and this is the case with all following trajectories). Assuming that there is no even-odd splitting, the best fits have 2 m ≈ 1200 MeV and a slope of about 1 4 . -1 5 GeV . -2 . Assuming splitting, we find in the
Figure 10 . χ 2 vs. the two endpoint masses for the Λ c trajectory, for the two cases: α ′ optimized (left) and α ′ fixed at 0.95 GeV -2 (right). In the adjustable slope case the optimum has α ′ = 1 2 . GeV -2 . The red curves in the left plot are m 1 + m 2 = 2 × 1000 and m 3 / 2 1 + m 3 / 2 2 = 2 × (1000) 3 / 2 . The real curve on which χ 2 is optimal, for which we do not know the exact parametrization, is between those two.
case of the Σ that the linear fit connecting the two even states has α ′ ≈ 0 9 GeV . -2 .
A third option, is fixing the slope at a more reasonable low value - we chose α ′ = 0 95 . GeV -2 -and redoing the massive fits (with the assumption that there is no splitting). The best fits then for the Σ are at around 2 m = 500 MeV, with the mass being somewhat higher in the trajectory beginning with of the 1 / 2 + state. This is certainly the choice that is most consistent with previous results, as we can distribute the total mass so there is a mass of m s ≈ 400 MeV at one end and up to 100 MeV at the other. The cost in χ 2 is fairly high: for the first trajectory χ 2 is approximately 10 -4 for the higher slope and ten times larger for α ′ = 0 95, while for the second . χ 2 is almost zero for the high slope fit 8 and about 5 × 10 -4 for the fixed slope fit. In any case, the fits with α ′ = 0 95 GeV . -2 and with the added masses have a better χ 2 than the linear massless fits.
There is one more possible trajectory we examine, of the doubly strange Ξ baryon. The best fit overall is again with α ′ ≈ 1 5 GeV . -2 , and at a somewhat higher mass of 2 m = 1320. Fixing the slope at 0 95 GeV . -2 results in 2 m = 850 being optimal. This is again the best choice in terms of consistency - the total mass is exactly in the range we would expect to see where there are two s quarks present. In χ 2 , the fit with the high slope has χ 2 ≈ 10 -4 while the latter has χ 2 ≈ 4 × 10 -4 .
## 5.5 Charmed baryons
In the charmed baryon section we have only one trajectory we can use, comprised of three states, that of the Λ c baryon. The best fits are again at a relatively high slope, 1.1 GeV -2 ,
8 This is often the case with three point trajectories, where we may find a choice of the parameters for which the trajectory passes through all three data points. This makes the error in the measurement hard to quantify.
Table 1 . Summary table for the baryon fits. The ranges listed have χ 2 within 10% of its optimal value. N is the number of points in the trajectory. [a] is the ( J, M 2 ) trajectory of the N baryons when taking only the four highest J states, [b] is a fit to radial trajectories of the N . The fifteen states used are four states with J P = 1 2 / + , three with 3 / 2 -, and four pairs with other values of J P . [c] is the radial trajectory of the ∆ (3 / 2 + ). [d] is a trajectory beginning with the state Σ(1385) 3 / 2 + , as opposed to the 1 / 2 + Σ ground state. The rest of the trajectories are all leading trajectories in the ( J, M 2 ) plane, and do not exclude any states.
| Traj. | N | m | α ′ | a |
|---------|-----|---------------|-------------------|-------------------------------------------------------------------|
| N | 7 | 2 m = 0 - 170 | 0 . 944 - 0 . 959 | = ( - 0 . 32) - ( - 0 . 23) a o = ( - 0 . 75) - ( - 0 . 65) |
| N [ a ] | 4 | 2 m = 0 - 640 | 0 . 949 - 1 . 018 | e = ( - 0 . 34) - 0 . 50 a o = ( - 0 . 98) - ( - 0 . 13) |
| N [ b ] | 15 | 2 m = 0 - 425 | 0 . 815 - 0 . 878 | / 2+ = ( - 0 . 22) - 0 . 07 a 3 / 2 - = ( - 0 . 36) - ( - 0 . 06) |
| ∆ | 7 | 2 m = 0 - 450 | 0 . 898 - 0 . 969 | a e = 0 . 14 - 0 . 54 a o = ( - 0 . 84) - ( - 0 . 42) |
| ∆ [ c ] | 3 | 2 m = 0 - 175 | 0 . 920 - 0 . 936 | a = 0 . 11 - 0 . 21 |
| Λ | 5 | 2 m = 0 - 125 | 0 . 946 - 0 . 955 | a = ( - 0 . 68) - ( - 0 . 61) |
| Σ | 3 | 2 m = 1190 | 1 . 502 | a = ( - 0 . 15) |
| Σ [ d ] | 3 | 2 m = 1255 | 1 . 459 | a = 1 . 37 |
| Ξ | 3 | 2 m = 1320 | 1 . 455 | a = 0 . 50 |
| Λ c | 3 | 2 m = 2010 | 1 . 130 | a = 0 . 09 |
with the mass 2 m = 2010 MeV. This fit's χ 2 tends to zero. The fit with the slope fixed at 0.95 GeV -2 takes the mass down to 2 m = 1760 MeV with χ 2 = 3 × 10 -5 . The high slope fit is equivalent to a fit with m 1 = 1720 MeV and m 2 = 90, while a the fit with m 1 = 1400 and m 2 = 90 is roughly equal to the latter fixed slope fit. We plot χ 2 as a function of the two endpoint masses for both the fixed and adjustable slope case in figure (10).
We can also do a fit using two Ξ c states. These states are charmed and strange and are composed of dsc (Ξ ) or 0 c usc (Ξ + c ). Since we only have two states, we do only a fit with the fixed slope, α ′ = 0 95 GeV . -2 . The best massive fit then has 2 m = 2060 MeV.
## 6 Summary of results
We begin by presenting the two summary tables: in table (1) we list the results of the general fits, while in table (2) are the results of the fits with the fixed slope, α ′ = 0 95 GeV . -2 .
The internal structure of the baryon is more complex than that of the meson, and as a result a unified stringy model of the baryon is harder to construct. Out of the various stringy models we have examined, the model of a quark and diquark is best supported by experiment. This is based mostly on the observation that the Regge slope for baryons is roughly equal to the meson slope - so if we assume a universal string tension, the baryons must be described by a single string model as the mesons are. As explained in section 5.2, we present our results in terms of the total mass of the endpoints, as we cannot determine the distribution of the mass between them from the Regge trajectory fits alone.
Table 2 . J, M 2 ) fits done with the slope fixed at α ′ = 0 950 GeV . -2 . Fits with m 1 = m 2 generally maximize m 1 + m 2 . In this table we may also include a fit for two Ξ c states. The ranges listed have χ 2 within 10% of its optimal value. N is the number of points in the trajectory.
| Traj. | N | m | a |
|---------|-----|-----------------|-----------------------------------------------------------------|
| N | 7 | 2 m = 0 - 180 | a e = ( - 0 . 33) - ( - 0 . 22) a o = ( - 0 . 77) - ( - 0 . 65) |
| ∆ | 7 | 2 m = 300 - 530 | a e = 0 . 31 - 0 . 66 a o = ( - 0 . 71) - ( - 0 . 26) |
| Λ | 5 | 2 m = 0 - 10 | a = ( - 0 . 68) - ( - 0 . 61) |
| Σ | 3 | 2 m = 530 - 690 | a = ( - 0 . 29) - ( - 0 . 04) |
| Σ* | 3 | 2 m = 435 - 570 | a = 0 . 15 - 0 . 38 |
| Ξ | 3 | 2 m = 750 - 930 | a = ( - 0 . 22) - 0 . 10 |
| Λ c | 3 | 2 m = 1760 | a = ( - 0 . 36) |
| Ξ c | 2 | 2 m = 2060 | a = ( - 1 . 13) |
Our massive fits for the baryons are not always consistent. For the light quark trajectories, of the N and the ∆, we have seen there is no evidence for a string endpoint mass of the light quarks. These states are best fitted by linear trajectories with a slope similar to that of the light mesons - around 0.95 GeV -2 for the N and 0 90 GeV . -2 for the ∆. The ( J, M 2 ) trajectories also exhibit a splitting between the trajectories of states with even and odd orbital angular momentum. In our fits we incorporate this difference into the intercept alone.
Since we do not see the splitting effect in the trajectory of the strange Λ baryon, which has five data points, we assume this effect is only significant for the light baryons. The next heavier trajectories after the Λ are comprised of only three data points and therefore cannot be used to determine whether this is a correct assumption or not. In the following we summarize our results using this assumption.
For the strange baryon trajectories there is a significant discrepancy between the Λ, which is best (and very well) fitted by a linear, massless, trajectory, and the Σ baryons which are optimally fitted with a high total mass, of around 1200 MeV and an unusually high slope, 1.5 GeV -2. The fits when fixing the slope at the value obtained from the lighter baryon fits give a total mass in the more reasonable 500 -600 MeV range. We have also looked into the doubly strange Ξ baryon, where we have a similar picture: the same high slope (1.5) and masses (1300) or α ′ = 0 95 GeV . -2 and a total mass of around 800 MeV. We noted that the latter choice, of a fixed slope, is not only more consistent with the slopes obtained from the lighter baryons, but also with the mass obtained for the s quark in the meson fits, m s ≈ 400 MeV.
The last results are those of the charmed baryons, the heaviest baryons for which we have a trajectory. The charmed Λ c is again best fitted by a high slope, 1.2GeV -2 and a total mass of a little over 1800 MeV. Once more we can bring down the mass by fixing α ′ at 0.95 GeV -2 and get 2 m = 1760 MeV as the best fit. For the charmed-strange Ξ c trajectory, including only two data points and therefore fitted only with the fixed slope of 0.95 GeV -2 , we find a fit with 2 m = 2060 MeV.
## 6.1 Comparison with meson fits
This section will briefly summarize the results of the fits to the Regge trajectories of mesons done in [1], and offer a comparison between them and the results presented in this paper for the baryon trajectories.
The mesons are expected to have a simpler structure than the baryons. Therefore, one could say they are a better source from which we can begin to extract the parameters of our models. We have only one stringy model of the meson - in holography it is the rotating open string connected at both ends to flavor branes, and in the mapping to flat space-time it is a single string with two endpoint masses.
The first result, that is easiest to compare, is that of the Regge slope. For the ( J, M 2 ) trajectories of the mesons we found that, with added masses, all trajectories involving u , d , s , and c quarks are well fitted with a Regge slope of around 0 9 GeV . -2 9 . For the baryons, we have seen that the light ( N and ∆) baryons, as well as the strange Λ, are best fitted with a slope of 0.90-0.95 GeV -2 . This similarity between the meson and baryon results is what drove us to use the single string quark-diquark model for the baryon, which predicts an equal Regge slope between baryons and mesons.
Moving on to the heavier baryons, we found higher slopes (of up to 1.5 GeV -2 ) but also found that we can get reasonably good fits using 0.95 GeV -2 as a 'universal' slope, common to all available trajectories (light, strange, and charmed). We have seen this happen for the mesons as well, but the differences there between the optimal fits and the 'universal slope' fits were smaller - there it was the difference between 1.0-1.1 and 0.9 GeV -2 .
In the ( n, M 2 ) plane the light meson trajectories had a similar slope to their respective orbital trajectories, but always slightly lower. A generic value obtained there would be 0.8 GeV -2 . For the baryons, we only analyzed the trajectories of the N and the ∆. For the N we have seen the same behavior we have for the mesons, with approximately the same slope, but for the ∆ the slope in ( n, M 2 ) is the higher one.
Knowing the different mesons' compositions, we can extract from them the quark masses directly. The two endpoint masses correspond directly to the quark and anti-quark making up the meson, and so we naturally identify the masses obtained in our fits with quark masses.
The mass of the u and d quarks we could not determine from the meson fits. We found no evidence that clearly states that the light quarks have a non-zero mass, but we did not exclude a mass of up to about 100 MeV. For the s quark meson fits we found that a non-zero mass for the s was always preferable. The results were generally in the range m s = 200 -400 MeV. The c quark was found to have a mass close to the value of its mass as a constituent quark, around 1500 MeV.
9 The only exception, in the ( J, M 2 ) plane, was the trajectory of the Υ ( bb ¯ ), which had a slope of 0.64 GeV -2 . For the baryons, we do not have trajectories of bottom baryons, and therefore, no point of comparison. In [1] we noted that since our model is based on a long string approximation it might not hold for the heavy bb ¯ mesons, and hence the discrepancy in the slope. We do not have trajectories of baryons containing b quarks, and the problem of short strings is less apparent in the current analysis.
Figure 11 . The two holographic setups for the Ξ -baryon, with different compositions of the diquarks. The vertical segments of the strings contribute the measured endpoint masses.
The light baryon trajectories seem to prefer massless endpoints without completely excluding masses of a few dozen MeV, or up to a hundred MeV. This is similar to the meson result. The strange Λ offers the biggest discrepancy in terms of the mass. The Λ trajectory is best fitted by a simple linear fit with no endpoint masses where we would have expected the presence of the s quark to contribute a mass of at least 200 MeV. The other strange baryons, on the other hand, seem to be consistent with the meson results, especially when fixing the slope at its universal value. The results for the charmed Λ c baryon are also consistent with a mass of 1500 MeV for the c quark, as is the result for the charmed-strange baryon, which is compatible with a mass of 1500 MeV on one end (for the charmed), and a mass of 400 MeV on the other.
## 6.2 Structure of the baryon in the quark diquark model
For the light baryons our analysis of the spectrum cannot offer much new insight regarding the different baryons' structure, 10 in particular because we have no way to distinguish between the two light quarks, given their small - possibly zero - masses, but also because we cannot in general make any comments regarding the mass distribution within the different baryons (both light and heavy). In spite of this, there is one interesting implication when interpreting our results in light of the underlying holographic models and the way they map the diquarks into flat space-time.
10 [28] offers a discussion of the composition of the light baryons in a model of a quark and diquark joined by a flux tube. In the analysis of the spectrum done there, the light baryons are assigned different configurations of the diquark based on the energetics of the ud diquarks.
In our analysis of the meson spectrum, we argued that the mass parameter relevant to the analysis is the mass of the quark as a string endpoint, which generically was found to be between the usual QCD and constituent masses attributed to the respective quark. For the diquark the identification between string length and mass can have another implication, as illustrated in figure (11). If the relevant mass parameter is the length of the vertical segment of the string connected to the flavor brane, and if the two quarks forming the diquark and the baryonic vertex to which they are both connected all lie close to each other on the flavor brane, then we would expect the mass of the diquark - in the holographic picture, as a string endpoint - to be approximately equal to the mass of a single quark:
$$m _ { q q } & \approx m _ { q }$$
This is because we have only one contribution to the mass from the string connecting the lone quark outside the diquark and the baryonic vertex. This is a prediction that can serve as a test of the holographic interpretation of the string endpoint masses. Since we do not have an accurate figure for the mass of the light u and d quarks, and since we lack data for charmed and heavier baryons, our best avenue for verifying this experimentally is by examining the doubly strange Ξ baryon.
The two options for the Ξ quark are one where the diquark is composed of an s and a light quark, and another where the diquark is composed of two s quarks. For the first option, our holographic interpretation would lead us to expect there to be two masses at the endpoints approximately equal to the s quark mass, leading to a total mass of 2 m ≈ 2 m s at the endpoints. In the second option, the two s quarks in the diquark would contribute only once to the total mass we measure in the Regge trajectory fits, so the result for the total mass 2 m should be around, possibly a little higher than, the mass of a single s quark.
The result from the fixed slope fit of the Ξ trajectory, 2 m = 750 -930 MeV, is consistent, from the holographic point of view, with a ds or us diquark, as opposed to ss . This is because we expect the mass of the s to be somewhere near 400 MeV. Of course, from a purely flat space-time perspective, an ss diquark with a mass of roughly 2 m s is not excluded.
If we look at the decay modes of the states used in the Ξ fits we might learn something about their structure [32][33]. We look at a baryon's strong decays into a baryon and a meson, and our assumption is that in these decays the diquark and baryonic vertex go into the outgoing baryon while the third lone quark ends up in the meson. An illustration of this type of decay is in figure (12).
The lightest doubly strange state does not have the phase space for strong decays. If we look at the two next states in the trajectory, the Ξ(1820) and the Ξ(2030), we see that they decay mainly into Λ K or Σ K . The fact that they decay into a strange meson and strange baryon is against the ss diquark configuration. The leading modes of decay should leave the diquark intact, so if the diquark were ss the leading mode of decay would be Ξ π (as it is for some of the other observed doubly strange baryons).
Figure 12 . A doubly strange baryon with an ss diquark decaying into a doubly strange baryon and a non-strange meson. First the string tears, and then the endpoints reconnect to the flavor brane, forming a quark-anti-quark pair.
For the baryons with a single strange quark, the Λ and the Σ, we have seen an odd discrepancy between the obtained mass values. The mass in the Λ baryon was less than 100 MeV, while in the Σ we have seen masses of above 400 MeV. We cannot explain this discrepancy in terms of different configurations of the diquarks, as we expect the s to contribute to the mass whether it is in the diquark or not. The decay modes do not give a straightforward answer regarding the compositions of the Λ and Σ, as the states decay both to NK and Σ π /Λ π . We do not see a systematic preference for decays where the s remains in the baryon (implying it is near the baryonic vertex in the diquark) or vice versa in either of the trajectories.
For the charmed-strange Ξ c we see a mass compatible compatible with m 1 + m 2 = m s + m c . This implies to us that the possibility of a cs diquark is excluded, since we see both quarks' masses (from the holographic point of view we expect the diquark mass to be m cs ≈ m c ). Unfortunately we cannot test this based on the decay modes. If we look at the decays of the Ξ c baryons, we find that the Ξ /Ξ 0 c -c does not have the phase space to decay strongly, and the next state we take in the trajectory, Ξ (2815), is also too light to provide information c that would be useful to us. The Ξ (2815) cannot decay to a charmed meson and a strange c baryon (which is the decay mode we will naively expect if the Ξ c is a us ds / diquark joined to a c quark), simply because the lightest of these, D ± / D 0 and Λ respectively, are still heavy enough so that the sum of their masses exceeds the mass of the Ξ (2815). c
For the charmed Λ c baryon, with one c and two light u/d quarks, we have no prediction based on the masses, because in any case we expect to see a total mass of approximately
m c = 1500 MeV. The first state in the trajectory heavy enough to decay into a charmed meson and a baryon, the Λ (2800) c + , was observed to decay both to pD 0 and Σ c π , but there is no quantitative data to indicate which of these modes (if any) is dominant.
## 7 Conclusions and future directions
If we seek a universal stringy model of the baryon the best option seems to be the model of a quark and a diquark at the endpoints of a rotating string. We have seen how this model can be obtained from a mapping of rotating holographic strings to flat space-time, and it is the only stringy configuration in flat space-time which we know to be classically stable.
From a phenomenological point of view, unlike the three string Y-shape or the closed string ∆-shape models, the quark-diquark model is supported by the simple fact that the baryons lie on Regge trajectories with a slope roughly equal to that of the meson trajectories. In our fitting analysis we have presented fits of trajectories of baryons composed of u , d , s , and c quarks, an in particular fits with the fixed slope α ′ = 0 95 GeV . -2 . The fixed slope fits, while not optimal, are for the most part consistent with the meson fit results of [1].
In section 6.2 we have explained how one could test the holographic interpretation of the endpoint masses as the lengths of the vertical segments of the string along the radial dimension by examining the trajectories of doubly heavy baryons. We have attempted this test for the doubly strange Ξ baryon, but found no conclusive answer. In that section we also briefly discussed our approach to the holographic decay of hadrons. A detailed analysis of hadronic decays, not only qualitative as offered in this paper but also quantitative, is one avenue for continuing the use of our model for research. The static properties of baryons have also been examined in holography [6][15], and it is left to see how one can describe some of them using a stringy model.
Another prediction from holography is the presence of a baryonic vertex. Our fits exclude the presence of a baryonic vertex mass at the center of mass of the rotating baryon, but in the quark-diquark model which we prefer it is expected to be found at one of the string endpoints, with the diquark. If there is a baryonic vertex at an endpoint, there is no evidence to suggest it contributes greatly to the endpoint mass.
We could also attempt enhancements of the model. Adding spin degrees of freedom to the endpoints would give us a much better chance of constructing a universal model that would describe the entire baryon spectrum. In [28] the distinction between spin zero and spin one diquarks played an important part in analyzing the spectrum, while we have only discussed the flavor structure of the diquark and ascribed its mass to the holographic string alone (with a possible small addition from the baryonic vertex). Spin interactions could also help explain the even-odd splitting observed in the light baryons - our simple classical model will not explain a phenomenon that distinguishes between symmetric and anti-symmetric states without some additional interaction.
We should also strive to gain a better understanding of the intercept. While all previous sections discuss results for the slope and endpoint masses alone, the intercept, the results
for which are listed in the summary tables of section (6), is an interesting parameter from a theoretical point of view, and understanding its behavior is an important goal in constructing a truly universal model of the baryon. Added interactions should contribute, in leading order, a correction to the intercept, which should also be affected by the endpoint masses, and understanding the intercept's behavior may also help us distinguish between different configurations of the baryon without requiring additional information from experiment.
Table 3 . The baryon states used in the ( J, M 2 ) trajectory fits.
| Traj. | I | J P | State | Status | Traj. | I | J P | State | Status |
|---------|-----|--------------------------------------------------|--------------------------------------------------|------------------------------|-------------------|-----|-------------------------------------------------|-------------------------------------------------|------------------------------|
| N | 1 2 | 1 / 2 + 3 / 2 - 5 / 2 + 7 / 2 - 9 / 2 + 11 / 2 - | n/p N (1520) N (1680) N (2190) N (2220) N (2600) | **** **** **** **** **** *** | Λ S = - 1 Σ | 0 1 | 1 / 2 + 3 / 2 - 5 / 2 + 7 / 2 - 9 / 2 + 1 / 2 + | Λ(1116) Λ(1520) Λ(1820) Λ(2100) Λ(2350) Σ(1193) | **** **** **** **** *** **** |
| ∆ | 3 2 | 13 / 2 + 3 / 2 + 5 / 2 - 7 / 2 + | N (2700) ∆(1232) ∆(1930) ∆(1950) | ** **** *** **** | S = - 1 Σ S = - 1 | 1 | 3 / 2 - 5 / 2 + 3 / 2 + 5 / 2 - | Σ(1670) Σ(1915) Σ(1385) Σ(1775) | **** **** **** **** |
| | | 9 / 2 - 11 / 2 + 13 / 2 - 15 / 2 + | ∆(2400) ∆(2420) ∆(2750) ∆(2950) | ** **** ** ** | Ξ S = - 2 | 1 2 | 7 / 2 + 1 / 2 + 3 / 2 - 5 / 2 + | Σ(2030) Ξ 0 / Ξ - Ξ(1820) Ξ(2030) | **** **** *** *** |
| | | | | | Λ c C = 1 | 0 | 1 / 2 + 3 / 2 - 5 / 2 + | Λ c (2286) + Λ c (2625) + Λ c (2880) + | **** *** *** |
| | | | | | Ξ c C = 1 S = - 1 | 1 2 | 1 / 2 + 3 / 2 - | Ξ + c / Ξ 0 c ) Ξ c (2815) + | *** *** |
## A Individual trajectory fits
This section presents the data and individual trajectory fits. The experimental data is taken from the Particle Data Group's (PDG) Review of Particle Physics [30].
## A.1 The states used in the fits
The states we have used in our analysis, and their assignment into trajectories, are summarized in tables (3) and (4). The first table is for the ( J, M 2 ) plane and the latter for the ( n, M 2 ) trajectories. For the baryons, we also indicate the overall 'status' of their PDG listing, ranking the resonances from one to four stars based on how well established are they and their properties in experiment. As a general rule we consider states with three or four stars safe to use as we see fit. We include lesser resonances only when they complement a trajectory formed by well established states.
To build the trajectories, we assume the relation between a states orbital angular momentum and its parity is P = ( -1) L , so even and odd states in a given trajectory in the ( J, M 2 ) have alternating parity, as they do for mesons. We do fits only to leading trajectories
Table 4 . The baryon states used in the ( n, M 2 ) trajectory fits and their assignments.
| Traj. | I ( J P ) | n | State | Status | Traj. | I ( J P ) | n | State | Status |
|---------|----------------|-----|----------|-----------|---------|----------------|-----|-------------------|----------|
| N | 1 2 (1 / 2 + ) | 0 1 | n/p N | **** **** | N | 1 2 (1 / 2 - ) | 0 1 | N (1535) N (1895) | **** ** |
| | | | (1440) | | | | | | |
| | | 2 | N (1880) | ** | N | 1 2 (3 / 2 + ) | 0 | N (1720) | **** |
| | | 3 | N (2100) | * | | | 1 | N (2040) | * |
| N | 1 2 (3 / 2 - ) | 0 | N (1520) | **** | N | 1 2 (5 / 2 - ) | 0 | N (1675) | **** |
| | | 1 | N (1875) | *** | | | 1 | N (2060) | ** |
| | | 2 | N (2150) | ** | ∆ | 3 2 (1 / 2 + ) | 0 | ∆(1232) | **** |
| N | 1 2 (5 / 2 + ) | 0 | N (1680) | **** | | | 1 | ∆(1600) | *** |
| | | 1 | N (2000) | ** | | | 2 | ∆(1920) | *** |
(with n = 0), so we always select the lightest mass state for a given angular momentum and with the appropriate quantum numbers for a given trajectory.
The majority of states used in the ( J, M 2 ) fits are states with a PDG status of four or three stars. The few exceptions are the high spin states of the N and ∆ baryons, where we use some two star states. These states generally fit in well with their respective trajectories, which might be considered as evidence for the existence of those high J states at their given masses. The only exception is the ∆(2950) which lies above its predicted place in the trajectory. Its mass should be about 100 MeV lower than the mass the PDG lists for it.
The strange baryons used are all well established states, but there one comment to be made about the Σ states chosen. In the first trajectory, beginning with the J P = 1 2 / + Σ ground state, the next state chosen (with 3 / 2 -) is the Σ(1670). The PDG lists another state with the same J P at a lower mass, Σ(1580), but it is a one star state and its existence is very uncertain. If we use the Σ(1580) as the 3 / 2 -state in the trajectory, we see that it is well fitted by a linear trajectory with α ′ = 0 90 GeV . -2 . The choice of Σ(1670) is more compatible with the massive fit, and has similar results to the ones obtained from the second Σ trajectory, where we use the lightest known states with J P = 3 2 / + , 5 / 2 -, and 7 / 2 + , all well established.
For the doubly strange trajectories we use, alongside the ground state Ξ 0 / Ξ ± and the Ξ(1820), the state Ξ(2030). This is a three star state in the PDG, but its parity and angular momentum are not exactly known. The PDG places the bound J ≥ 5 2 on its / angular momentum, and our analysis of the Regge trajectory seems consistent when identifying this state as the J P = 5 2 / -state in the leading Ξ trajectory.
Going to the charmed baryon sector, we still have well established states we can use, but some of these states' quantum numbers are yet to be measured directly. The second state in the Λ c trajectory is the Λ (2625), for which c J and P have not been measured, but the PDG offers the assignment J P = 3 2 / -. In the charmed-strange sector, the two states Ξ c states we use are again without a direct measurement of their spin-parity, and are only assigned J P = 1 2 / + and 3 2 / -. Our analysis of the Regge trajectories gives us no reason to doubt
these assignments.
We also include some trajectories of the light baryons in the ( n, M 2 ) plane. For the ∆ baryons we have two states with three stars following the well known and well established ground state. For the N baryons, we present an assignment of fifteen states into six trajectories, each with a different spin-parity. Here is the only place where we use some dubious one star states. We include them mostly to observe how this assignment results in parallel trajectories in the ( n, M 2 ) plane. The conclusions we eventually draw from the fits pertain to the value of the slope obtained, and the slope does not change significantly if we include only the best established states, the J P = 1 2 / + or 3 / 2 -N baryons with three or four PDG stars. N resonances with three or four stars that we do not use in these fits are the N (1650)1 / 2 -, N (1700) 3 / 2 -, N (1710) 1 / 2 + , and N (1900) 3 / 2 + . For each of these we find at least one lower resonance with the same J P , and all are too low in mass to be the next states in their respective trajectories.
## A.2 Trajectories in the ( J, M 2 ) plane
## A.2.1 Light quark baryons
The states in this section are all comprised of u and d quarks only.
The N trajectory: The states, with their J P values, are N (939) 1 + 2 , N (1520) 3 2 -, N (1680) 5 + 2 , N (2190) 7 2 -, N (2220) 9 + 2 , N (2600) 11 2 -, and N (2700) 13 + 2 . The trajectory and its fits are depicted in (13).
The linear fit is
$$\alpha ^ { \prime } \, = 0. 9 4 4, a _ { e } = - 0. 3 2, a _ { o } = - 0. 7 5$$
with χ 2 l = 12 15 . × 10 -4 . It is optimal, and the highest mass fit with a good χ 2 is
$$2 m = 1 7 0, \alpha ^ { \prime } \, = 0. 9 5 9, a _ { e } = - 0. 2 3, a _ { o } = - 0. 6 5$$
with χ 2 m /χ 2 l = 1 10. .
When taking only the four highest J states in the trajectory (i.e. the states starting from J = 7 2), the mass dependence is weaker. The linear fit, now /
$$\alpha ^ { \prime } \, = 0. 9 4 9, a _ { e } = - 0. 3 4, a _ { o } = - 0. 9 8$$
is still optimal, with χ 2 l = 1 10 . × 10 -4 , but we can go to higher masses, such as
$$2 m = 6 4 0, \alpha ^ { \prime } \, = 1. 0 1 8, a _ { e } = 0. 5 0, a _ { o } = - 0. 1 3$$
which has χ 2 m /χ 2 l = 1 09. .
The ∆ trajectory: Here we use the states ∆(1232) 3 + 2 , ∆(1930) 5 2 -, ∆(1950) 7 + 2 , ∆(2400) 9 2 -, ∆(2420) 11 + 2 , ∆(2750) 13 2 -, and ∆(2950) 15 + 2 . The linear fit is again optimal
$$\alpha ^ { \prime } \, = 0. 8 9 8, a _ { e } = 0. 1 4, a _ { o } = - 0. 8 4$$
Figure 13 . Top: The N with linear (red, dashed line) and maximal mass fits (blue line). The maximal mass is defined as the highest mass with χ 2 within 10% of its optimal value. Top-left: all the states and fit with 2 m = 170. Top-right: fit to highest J states alone leads to a maximal mass of 2 m = 640. The states marked in red are those excluded from the fits. Bottom: Trajectory of the ∆ with the maximal mass 2 m = 450.
and it has χ 2 l = 12 51 . × 10 -4 . The highest mass fit is
$$2 m = 4 5 0, \alpha ^ { \prime } \, = 0. 9 6 9, a _ { e } = 0. 5 4, a _ { o } = - 0. 4 2$$
with χ 2 m /χ 2 l = 1 10. This trajectory is depicted in the bottom plot of figure (13). .
## A.2.2 Strange baryons
I = 0 . The Λ trajectory: In the left side plot of figure (14), we have the Λ baryons. These contain two light quarks ( u/d ), and an s quark. The states we use are Λ(1116)1 / 2 + , Λ(1520)3 / 2 -, Λ(1820)5 / 2 + , Λ(2100)7 / 2 -, and Λ(2350)9 / 2 + .
The even/odd effect is not present in this trajectory. The best linear fit is
$$\alpha ^ { \prime } \, = 0. 9 4 6, a = - 0. 6 8$$
Figure 14 . The strange baryon trajectories. Left: The Λ, for which the linear fit is optimal and the maximal mass is 2 m = 125. Right: The two Σ trajectories, with their massive fits - 2 m = 1190 for the trajectory beginning with J P = 1 2 / + and 2 m = 1255 for the trajectories whose lowest state has J P = 3 2 / + .
with χ 2 l = 0 71 . × 10 -4 . Once again, it is optimal, and masses can only go up to
$$2 m = 1 2 5, \alpha ^ { \prime } = 0. 9 5 5, a = - 0. 6 1$$
where χ 2 m /χ 2 l = 1 10. .
I = 1 . The Σ trajectory: Here we use the states Σ(1193)1 / 2 + , Σ(1670)3 2 / -, and Σ(1915)5 / 2 + . The best linear fit is:
$$\alpha ^ { \prime } \, = 0. 8 4 3, a = - 0. 7 1$$
with χ 2 l = 26 42 . × 10 -4 . The best massive fit is at the high value of
$$2 m = 1 1 9 0, \alpha ^ { \prime } \, = 1. 5 0 2, a = 0. 5 0$$
with χ 2 m /χ 2 l = 0 06. . We also do a fit with the slope fixed at α ′ = 0 950, to match the result . of the N , ∆, and Λ, and get as a result the fit
$$2 m = 6 2 0, \alpha ^ { \prime } \, = 0. 9 5 0, a = - 0. 1 5$$
as the optimum with χ 2 m /χ 2 l = 0 77. . If, instead, we assume there is an even-odd effect, then we are left only with two data points. The line connecting them is
$$\alpha ^ { \prime } \, = 0. 8 9 0, a _ { e } = - 0. 7 6$$
and then the intercept determined from the single odd state and the given slope is a o = -0 98. .
I = 1 . The reverse parity Σ trajectory: We have three more Σ states with the parity reversed relative to the states used in the previous trajectory: Σ(1385)3 2 / + , Σ(1775)5 / 2 -, and Σ(2030)7 / 2 -. For them, the optimal linear fit is
$$\alpha ^ { \prime } \, = 0. 8 8 2, a = - 0. 2 0$$
It has χ 2 l = 5 96 . × 10 -4 , and the optimal massive fit,
$$2 m = 1 2 5 5, \alpha ^ { \prime } \, = 1. 4 5 9, a = 1. 3 7$$
has χ 2 m = 3 × 10 -6 ( χ 2 m /χ 2 l = 0 005). The optimal fit with the slope . α ′ = 0 950 is .
$$2 m = 5 0 5, \alpha ^ { \prime } \, = 0. 9 5 0, a = 0. 2 7$$
with χ 2 m /χ 2 l = 0 81. The two Σ trajectories are in figure (14). .
Doubly strange baryons. The Ξ trajectory: This potential trajectory is comprised of the Ξ (1315) 0 / Ξ (1322)1 2 -/ + , Ξ(1820)3 / 2 -, and takes Ξ(2030) to be the 5 / 2 + state. Its fits are plotted in figure (15). The best linear fit is
$$\alpha ^ { \prime } \, = 0. 7 8 8, a = - 0. 9 0$$
with χ 2 l = 51 12 . × 10 -4 . The optimal symmetric fit is
$$2 m = 1 3 2 0, \alpha ^ { \prime } \, = 1. 4 5 5, a = 0. 5 0$$
with χ 2 m /χ 2 l = 0 20, and the best fit with . α ′ = 0 950 is .
$$2 m = 8 5 0, \alpha ^ { \prime } = 0. 9 5 0, a = - 0. 0 4$$
with χ 2 m /χ 2 l = 0 77. .
## A.2.3 Charmed baryons
The Λ c trajectory: Here we used the following three states: Λ (2286) c + 1 2 / + , Λ (2625) c + 3 2 / -, and Λ (2880) c + 5 2 / + . These can be seen in the top-right plot of figure (15). Assuming that, like the strange Λ, no even-odd effect is present, the best linear fit we have is
$$\alpha ^ { \prime } \, = 0. 6 4 2, a = - 2. 8 8$$
with χ 2 l = 1 39 . × 10 -4 , and the optimal fit is at a high mass
$$2 m = 2 0 1 0, \alpha ^ { \prime } \, = 1. 1 3 0, a = 0. 0 9$$
with χ 2 m = 2 × 10 -9 . The best fixed slope fit is
$$2 m = 1 7 6 0, \alpha ^ { \prime } \, = 0. 9 5 0, a = - 0. 3 6$$
Figure 15 . Left: The doubly strange Ξ baryon and its fit, 2 m = 1320. Right: The charmed Λ c and its fit with m 1 = 90, m 2 = 1720. Bottom: The charmed-strange Ξ c with its fit, 2 m = 2060.
and it has χ 2 m 0 24 . × 10 -4 ( χ 2 m /χ 2 l = 0 17). . Some fits with an asymmetric distribution of the masses are
$$m _ { 1 } = 1 7 2 0, m _ { 2 } = 9 0, \alpha ^ { \prime } \, = 1. 2 2 1, a = - 0. 1 3$$
with χ 2 m = 6 × 10 -10 , or in fixed slope case, we find values near
$$m _ { 1 } = 1 4 0 0, m _ { 2 } = 9 0, \alpha ^ { \prime } \, = 0. 9 5 0, a = - 0. 6 8$$
which has χ 2 m = 0 30 . × 10 -4 ( χ 2 m /χ 2 l = 0 22). .
## A.2.4 Charmed-strange baryons: The Ξ c
We include here a fit done to a trajectory containing only two points. To reduce the number of fitting parameters, we do the fits with the assumption m 1 = m 2 = m and with the fixed slope α ′ = 0 95 GeV . -2 . This leaves us with only two fitting parameters: the total mass 2 m and the quantum intercept a . The states used are the Ξ + c / Ξ 0 c 1 2 / + and Ξ (2815) 3 c / 2 -. The best fit with the slope fixed is
$$2 m = 2 0 6 0, \alpha ^ { \prime } \, = 0. 9 5 0, a = - 0. 1 3$$
The linear fit connecting the two points is
$$\alpha ^ { \prime } \, = 0. 5 4 7, a = - 2. 8 3$$
χ 2 ≈ 0 for both these fits. They are plotted in figure (15).
## A.3 Trajectories in the ( n, M 2 ) plane
## A.3.1 Light quark baryons
The N radial trajectory: We use a total of 15 states belonging to six different trajectories with different J P assignments. They are N (939), N (1440), N (1880), and N (2100) with J P = 1 2 / + , N (1520), N (1875), and N (2150) which have J P = 3 2 / -, and N (1680) and N (2000) with J P = 5 2 / + . Also, we have the states with reverse parity, N (1535) and N (1895) with 1 / 2 -, N (1720) and N (2040) with 3 / 2 + , and finally N (1675) and N (2060) with 5 / 2 -. The best linear fit has
$$\alpha ^ { \prime } \, = 0. 8 1 5$$
$$a _ { 1 / 2 + } = - 0. 2 2, a _ { 3 / 2 - } = - 0. 3 6, a _ { 5 / 2 + } = 0. 1 5$$
$$a _ { 1 / 2 - } = - 1. 4 2, a _ { 3 / 2 + } = - 0. 9 2, a _ { 5 / 2 - } = 0. 1 6$$
and it has χ 2 l = 4 90 . × 10 -4 . It is optimal and the highest good mass fit is
$$2 m = 4 2 5, \alpha ^ { \prime } \, = 0. 8 7 8$$
with χ 2 m /χ 2 l = 1 10 and the intercepts .
$$a _ { 1 / 2 + } = 0. 0 7, a _ { 3 / 2 - } = - 0. 0 6, a _ { 5 / 2 + } = 0. 4 6$$
$$a _ { 1 / 2 - } = - 1. 1 1, a _ { 3 / 2 + } = - 0. 6 1, a _ { 5 / 2 - } = 0. 4 7$$
The radial trajectories are depicted in figure (16).
The ∆ radial trajectory: Also in figure (16) we have the radial trajectory of the ∆. Here we have three states: ∆(1232), ∆(1600), and ∆(1920), all with 3 / 2 + . The linear fit
$$\alpha ^ { \prime } \, = 0. 9 2 0, a = 0. 1 1$$
is optimal with χ 2 l = 1 78 . × 10 -4 . The highest mass fits which are still close to the linear fit are around
$$2 m = 1 7 5, \alpha ^ { \prime } \, = 0. 9 3 6, a = 0. 2 1$$
$$\text{with } \chi _ { m } ^ { 2 } / \chi _ { l } ^ { 2 } = 1. 1 0.$$
Figure 16 . Top: The radial trajectories of the N . Top-left are the J P = 1 2 / + , 3 / 2 -, and 5 2 / + states. Top right are 1 / 2 -, 3 2 / + , and 5 / 2 -. All are fitted with the same slope and mass. Bottom: Radial trajectory of the ∆ baryon.
Table 5 . Predictions for the next states in the ( J, M 2 ) plane based on the optimal massive fits, with their J PC and mass (in MeV) values. The ranges listed correspond to the ranges in table (1). For the Σ baryons, the Ξ and the Λ c the first value is using the higher mass and slope of table (1), and the second is the value from the α ′ = 0 95 fixed slope fits in table (2). .
| Trajectory | Next states | Next states | Next states | Next states |
|--------------|---------------|---------------|---------------|---------------|
| N | 15 / 2 - : | 2950 - 2960 | 17 / 2 + : | 3050 - 3060 |
| ∆ | 17 / 2 - : | 3115 - 3250 | 19 / 2 + : | 3170 - 3230 |
| Λ | 11 / 2 - : | 2555 | 13 / 2 + : | 2750 - 2755 |
| Σ | 7 / 2 - : | 2210 / 2075 | 9 / 2 + : | 2445 / 2270 |
| Σ | 9 / 2 - : | 2245 / 2295 | 11 / 2 + : | 2430 / 2520 |
| Ξ | 7 / 2 - : | 2270 / 2325 | 9 / 2 + : | 2460 / 2560 |
| Λ c | 7 / 2 - : | 3095 / 3060 | 9 / 2 + : | 3280 / 3275 |
| Ξ c | 5 / 2 + : | -- / 3080 | 7 / 2 - : | -- / 3315 |
Table 6 . Predictions for the next states in the ( n, M 2 ) plane based on the optimal massive fits. Mass are in MeV. The ranges listed correspond to the ranges in table (1).
| Trajectory | Next states | Next states | Next states | Next states |
|--------------|---------------|---------------|---------------|---------------|
| N ( 1 2 + ) | n = 4: | 2395 - 2405 | n = 5: | 2625 - 2650 |
| N ( 3 2 - ) | n = 3: | 2425 - 2440 | n = 4: | 2655 - 2680 |
| N ( 5 2 + ) | n = 2: | 2295 - 2310 | n = 3: | 2540 - 2560 |
| N ( 1 2 - ) | n = 2: | 2180 - 2195 | n = 3: | 2435 - 2455 |
| N ( 3 2 + ) | n = 2: | 2315 - 2330 | n = 3: | 2555 - 2580 |
| N ( 5 2 - ) | n = 2: | 2290 - 2310 | n = 3: | 2535 - 2560 |
| ∆( 3 2 + ) | n = 3: | 2180 - 2185 | n = 4: | 2415 - 2420 |
## B Predictions for higher states
In tables (5) and (6) are the predictions for higher J and n baryons, based on the results of our fits.
## References
- [1] J. Sonnenschein and D. Weissman, Rotating strings confronting PDG mesons , JHEP 1408 (2014) 013, [ arXiv:1402.5603 ].
- [2] M. Kruczenski, L. A. Pando Zayas, J. Sonnenschein, and D. Vaman, Regge trajectories for mesons in the holographic dual of large-N(c) QCD , JHEP 0506 (2005) 046, [ hep-th/0410035 ].
- [3] A. Inopin and G. Sharov, Hadronic Regge trajectories: Problems and approaches , Phys.Rev. D63 (2001) 054023, [ hep-ph/9905499 ].
- [4] A. Tang and J. W. Norbury, Properties of Regge trajectories , Phys.Rev. D62 (2000) 016006, [ hep-ph/0004078 ].
- [5] E. Klempt and J.-M. Richard, Baryon spectroscopy , Rev.Mod.Phys. 82 (2010) 1095-1153, [ arXiv:0901.2055 ].
- [6] H. Hata, T. Sakai, S. Sugimoto, and S. Yamato, Baryons from instantons in holographic QCD , Prog.Theor.Phys. 117 (2007) 1157, [ hep-th/0701280 ].
- [7] T. Sakai and S. Sugimoto, Low energy hadron physics in holographic QCD , Prog.Theor.Phys. 113 (2005) 843-882, [ hep-th/0412141 ].
- [8] V. Kaplunovsky and J. Sonnenschein, Searching for an Attractive Force in Holographic Nuclear Physics , JHEP 1105 (2011) 058, [ arXiv:1003.2621 ].
- [9] V. Kaplunovsky, D. Melnikov, and J. Sonnenschein, Baryonic Popcorn , JHEP 1211 (2012) 047, [ arXiv:1201.1331 ].
- [10] V. Kaplunovsky and J. Sonnenschein, Dimension Changing Phase Transitions in Instanton Crystals , JHEP 1404 (2014) 022, [ arXiv:1304.7540 ].
- [11] E. Witten, Anti-de Sitter space, thermal phase transition, and confinement in gauge theories , Adv.Theor.Math.Phys. 2 (1998) 505-532, [ hep-th/9803131 ].
- [12] A. Brandhuber, N. Itzhaki, J. Sonnenschein, and S. Yankielowicz, Baryons from supergravity , JHEP 9807 (1998) 020, [ hep-th/9806158 ].
- [13] J. Callan, Curtis G., A. Guijosa, K. G. Savvidy, and O. Tafjord, Baryons and flux tubes in confining gauge theories from brane actions , Nucl.Phys. B555 (1999) 183-200, [ hep-th/9902197 ].
- [14] A. Dymarsky, S. Kuperstein, and J. Sonnenschein, Chiral Symmetry Breaking with non-SUSY D7-branes in ISD backgrounds , JHEP 0908 (2009) 005, [ arXiv:0904.0988 ].
- [15] S. Seki and J. Sonnenschein, Comments on Baryons in Holographic QCD , JHEP 0901 (2009) 053, [ arXiv:0810.1633 ].
- [16] S. Hellerman and I. Swanson, String Theory of the Regge Intercept , arXiv:1312.0999 .
- [17] A. Chodos and C. B. Thorn, Making the Massless String Massive , Nucl.Phys. B72 (1974) 509.
- [18] M. Baker and R. Steinke, Semiclassical quantization of effective string theory and Regge trajectories , Phys.Rev. D65 (2002) 094042, [ hep-th/0201169 ].
- [19] J. Zahn, The excitation spectrum of rotating strings with masses at the ends , JHEP 1312 (2013) 047, [ arXiv:1310.0253 ].
- [20] E. Witten, Baryons and branes in anti-de Sitter space , JHEP 9807 (1998) 006, [ hep-th/9805112 ].
- [21] A. Dymarsky, D. Melnikov, and J. Sonnenschein, Attractive Holographic Baryons , JHEP 1106 (2011) 145, [ arXiv:1012.1616 ].
- [22] S. Kuperstein and J. Sonnenschein, Non-critical, near extremal AdS(6) background as a holographic laboratory of four dimensional YM theory , JHEP 0411 (2004) 026, [ hep-th/0411009 ].
- [23] G. 't Hooft, Minimal strings for baryons , hep-th/0408148 .
- [24] G. Sharov, Quasirotational motions and stability problem in dynamics of string hadron models , Phys.Rev. D62 (2000) 094015, [ hep-ph/0004003 ].
- [25] E. Federovsky, Stringy baryons and scattering amplitudes , Master's thesis, Tel Aviv University, August, 2010.
- [26] G. Harpaz, Simulating stringy baryons , Master's thesis, Tel Aviv University, June, 2008.
- [27] G. Sharov, String Models, Stability and Regge Trajectories for Hadron States , arXiv:1305.3985 .
- [28] A. Selem and F. Wilczek, Hadron systematics and emergent diquarks , hep-ph/0602128 .
- [29] T. Friedmann, QCD vs. the Centrifugal Barrier: a New QCD Effect , EPJ Web Conf. 70 (2014) 00027.
- [30] Particle Data Group Collaboration, J. Beringer et al., Review of Particle Physics (RPP) , Phys.Rev. D86 (2012) 010001.
- [31] G. Sharov, String baryon model 'triangle': Hypocycloidal solutions , Phys.Rev. D58 (1998) 114009, [ hep-th/9808099 ].
- [32] K. Peeters, J. Sonnenschein, and M. Zamaklar, Holographic decays of large-spin mesons , JHEP 0602 (2006) 009, [ hep-th/0511044 ].
- [33] J. Sonnenschein and D. Weissman, On decays of stringy hadrons [work in progress] , .
|
10.1007/JHEP02(2015)147
|
[
"Jacob Sonnenschein",
"Dorin Weissman"
] |
2014-08-04T18:43:08+00:00
|
2015-01-01T18:10:01+00:00
|
[
"hep-ph",
"hep-th"
] |
A rotating string model versus baryon spectra
|
We continue our program of describing hadrons as rotating strings with
massive endpoints. In this paper we propose models of baryons and confront them
with the baryon Regge trajectories. We show that these are best fitted by a
model of a single string with a quark at one endpoint and a diquark at the
other. This model is preferred over the Y-shaped string model with a quark at
each endpoint. We show how the model follows from a stringy model of the
holographic baryon which includes a baryonic vertex connected with N_c strings
to flavor probe branes. From fitting to baryonic data we find that there is no
clear evidence for a non-zero baryonic vertex mass, but if there is such a mass
it should be located at one of the string endpoints. The available baryon
trajectories in the angular momentum plane (J,M^2), involving light, strange,
and charmed baryons, are rather well fitted when adding masses to the string
endpoints, with a single universal slope of 0.95 GeV^-2. Most of the results
for the quark masses are then found to be consistent with the results extracted
from the meson spectra in a preceding paper, where the value of the slope
emerging from the meson fits was found to be 0.90 GeV^-2. In the plane of
quantum radial excitations, (n,M^2), we also find a good agreement between the
meson and baryon slopes. The flavor structure of the diquark is examined, where
our interest lies in particular on baryons composed of more than one quark
heavier than the u and d quarks. For these baryons we present a method of
checking the holographic interpretation of our results.
|
1408.0765v2
|
## Modulation Classification via Gibbs Sampling Based on a Latent Dirichlet Bayesian Network
Yu Liu, Osvaldo Simeone, Member, IEEE, Alexander M. Haimovich, Fellow, IEEE, Wei Su, Fellow, IEEE
## Abstract
A novel Bayesian modulation classification scheme is proposed for a single-antenna system over frequency-selective fading channels. The method is based on Gibbs sampling as applied to a latent Dirichlet Bayesian network (BN). The use of the proposed latent Dirichlet BN provides a systematic solution to the convergence problem encountered by the conventional Gibbs sampling approach for modulation classification. The method generalizes, and is shown to improve upon, the state of the art.
## Index Terms
modulation classification, Gibbs sampling, Bayesian network, latent Dirichlet
## I. INTRODUCTION
Recognition of the modulation format of unknown received signals is an important functionality of intelligent, or cognitive, radios for both military and civilian applications [1]. In most practical scenarios, the modulation classification task must cope with the fact that transmission takes place over an unknown frequency-selective channel. A generalized likelihood ratio test is proposed in [2] to tackle this problem. The method, however, fails to identify nested signal constellations such as QPSK and 8-PSK [3], [4]. An alternative approach is to use a preliminary
Y. Liu, O. Simeone and A. M. Haimovich are with the Center for Wireless Communications and Signal Processing Research (CWCSPR), ECE Department, New Jersey Institute of Technology (NJIT), Newark, NJ 07102, USA (email: {yl227, osvaldo.simeone, haimovic}@njit.edu).
W. Su is with the U.S. Army Communication-Electronics Research Development and Engineering Center, I2WD, Aberdeen Proving Ground, MD 21005, USA (email: [email protected]).
blind equalization stage in order to compensate for the frequency-selective channel [1], [5]. The joint estimation of channel and modulation is analytically intractable, and hence Markov Chain Monte Carlo (MCMC) methods [6] provide viable solutions. A classification method based on MCMC was proposed in [5] for a single-antenna system by leveraging Gibbs sampling and by introducing a 'superconstellation' in order to solve the convergence problem of conventional Gibbs sampling (see Sec. III).
In this paper, a novel Bayesian modulation classification scheme is proposed for a singleantenna system over frequency-selective fading channels. The method is based on Gibbs sampling as applied to a latent Dirichlet Bayesian network (BN). The use of the proposed latent Dirichlet BN provides a systematic solution to the convergence problem encountered by the conventional Gibbs sampling approach. The method in [5] based on 'superconstellation' can be interpreted as an approximation of the proposed approach. Furthermore, with an appropriate selection of the prior distribution of the latent variable, our technique is shown to improve over [5].
Notation : The superscripts T and H are used to denote matrix or vector transpose and Hermitian, respectively. We use the notation ∝ to denote proportionality. The cardinality of a set S is denoted as |S| . Notation 1 ( ) · represents the indicator function. The following notations for some important distributions will be used: we denote by CN ( µ , C ) the circularly symmetric complex Gaussian distribution with mean vector µ and covariance matrix C ; we use U S ( ) to denote a uniform distribution on a set S , i.e., all elements of S are equiprobable; notation Cat S ( p ) is used for a categorical distribution on a set S with a length|S| vector p of probabilities; the inverse gamma distribution with shape parameter a and scale parameter b is denoted as IG ( a, b ) .
## II. SYSTEM MODEL
We consider a single-antenna system over a frequency-selective fading channel. The transmitted symbols s n are taken from a finite constellation A , such as M -PSK or M -QAM. We assume that the constellation A belongs to a known finite set A of possible constellations. The received baseband signal r ( ) t at the output of the matched filter is given by
$$r \left ( t \right ) = \sum _ { n = - \infty } ^ { \infty } s _ { n } h \left ( t - n T \right ) + w ( t ),$$
where h t ( ) represents the overall pulse shape, which includes the effect of transmit and receive filters and of the channel [7], and w t ( ) is complex white Gaussian noise with zero mean and
variance σ 2 . Without loss of generality, the average power of the transmitted symbols is assumed to be unity, i.e., E s ( | n | 2 ) = 1 . Moreover, the pulse shape h t ( ) is assumed to be of finite duration LT , for a given integer L . Sampling at multiples of the symbol period T , the k th received baseband sample is
$$r \left ( k T \right ) = \sum _ { n = - \infty } ^ { \infty } s _ { n } h \left ( k T - n T \right ) + w ( k T ).$$
Processing is performed in blocks of size N . Within this block, the received samples (2) can be collected in an N × 1 vector r = [ r (0) , r ( T , ) · · · , r (( N -1) T )] T , which, using (2), can be written as
$$r = S h + w.$$
In (3), we have defined the L × 1 channel vector h as h = [ h (0) , h ( T , ) · · · , h (( L -1) T )] T ; the vector w = [ w (0) , w ( T , ) · · · , w (( N -1) T )] T ∼ CN ( 0 , σ 2 I ) is the complex white Gaussian noise; and the N × L transmitted symbol convolution matrix is defined as
$$\mathbf S = \left [ \mathbf s _ { 0 }, \mathbf s _ { 1 }, \cdots, \mathbf s _ { N - 1 } \right ] ^ { T },$$
where s n = [ s n , s n -1 , · · · , s n -L +1 ] T .
Given the received signal vector r in (3), the goal of the receiver is to classify the modulation format A while being uninformed about the transmitted symbols s , defined as s = { s n } N -1 n = -L +1 , the channel vector h and the noise power σ 2 .
## III. PRELIMINARIES
In this work, as in [5], we perform the modulation classification task outlined above by using a Bayesian approach via MCMC methods. In this section, we review some key preliminary concepts.
## A. Bayesian Approach
The joint posterior probability density function (pdf) of the unknown quantities ( A, s h , ,σ 2 ) can be expressed as
$$p \left ( A, \mathbf s, \mathbf h, \sigma ^ { 2 } \Big | \mathbf r \right ) \in p \left ( \mathbf r \Big | A, \mathbf s, \mathbf h, \sigma ^ { 2 } \right ) p \left ( A, \mathbf s, \mathbf h, \sigma ^ { 2 } \right ),$$
where the likelihood function p ( r | A, s h , , σ 2 ) is such that
$$\mathbf r \Big | \left ( A, \mathbf s, \mathbf h, \sigma ^ { 2 } \right ) \sim \mathcal { C N } ( \mathbf S h, \sigma ^ { 2 } \mathbf l ),$$
and the term p A, ( s h , , σ 2 ) represents the available prior information on the unknown variables. We assume that this prior distribution factorizes as
$$p ( A, \mathbf s, \mathbf h, \sigma ^ { 2 } ) = p \left ( A \right ) \left \{ \prod _ { n = - L + 1 } ^ { N - 1 } p \left ( s _ { n } | A \right ) \right \}$$
where we have A ∼ U ( A ) , s n | A ∼ U ( A ) , h ∼ CN ( 0 , α I ) and σ 2 ∼ IG ( α , β 0 0 ) , where ( α, α , β 0 0 ) are fixed parameters. One typically selects α and β 0 to be sufficiently large and α 0 to be sufficiently small in order to obtain uninformative priors [5].
The Bayesian approach to modulation classification aims at estimating the posterior probability of the modulation A when conditioned on the received signal r , namely
$$p \left ( A | { \mathbf r } \right ) = \sum _ { \mathbf s } \int p \left ( A, \mathbf s, { \mathbf h }, \sigma ^ { 2 } | { \mathbf r } \right ) d { \mathbf h } d \$$
The computation of (8) involves a multidimensional integration, which is generally infeasible. In the following, we illustrate how this task can be accomplished by MCMC techniques.
## B. Bayesian Network
In order to facilitate the introduction of MCMC methods for modulation classification in the next section, we first recall some basic facts about BNs [8]. A BN is a directed graph, whose nodes are the random variables in the domain of interest and whose edges encode the direct probabilistic influence of one variable on another. Specifically, for a set of random variables { X k } K k =1 , a BN encodes a factorization of the joint distribution of the variables at hand of the form
$$p \left ( X _ { 1 }, \cdots, X _ { K } \right ) = \prod _ { k = 1 } ^ { K } p \left ( X _ { k } | \text{Pa} _ { X _ { k } } \right ), \quad \quad \quad$$
where Pa X k represents a subset of the variables ( X , 1 · · · , X k -1 ) . By the chain rule, the factorization (9) states that, when conditioning on all the variables ( X , 1 · · · , X k -1 ) , each variable X k is only influenced by the 'parent' variables Pa X k (i.e., we have the Markov chain ( X , 1 · · · , X k -1 ) -Pa X k -X k ). This statistical dependence between variable X k and the set of parent variables Pa X k is encoded in the BN by introducing a directed edge between all variables Pa X k and X k . As an example, the BN encoding factorization (5), (7) is shown in Fig. 1.
## C. Markov Chain Monte Carlo
MCMC methods provide a general approach for generating M samples x (1) , · · · , x ( M ) from an arbitrary target distribution p ( X ) with the aim of estimating ensemble averages, and hence multidimensional integrals such as in (8). MCMC methods simulate a Markov chain whose equilibrium distribution is p ( X ) in order to produce such samples [8]. For instance, the marginal of a joint distribution p ( X ) with respect to any variable X i in X can be estimated by MCMC as
$$p \left ( X _ { i } = x _ { i } \right ) \approx \frac { 1 } { M } \sum _ { m = M _ { 0 } + 1 } ^ { M _ { 0 } + M } 1 \left ( \mathbf x _ { i } ^ { ( m ) }$$
where x ( m ) i is the i th element of the m th sample of the simulated Markov chain and x i is a value in the domain of X i . Note that, in (10), the first M 0 samples generated by the Markov chain are not used in order to limit the impact of the initialization.
Gibbs sampling [8] is a classical MCMC algorithm, whereby, at each step m , a new sample of a given random variable X i is generated according to the conditional distribution p X ( i | X -i = x ( m -1) -i ) , where X -i denotes all variables in X except X i , which are fixed to the current value x ( m -1) -i . Gibbs sampling is known to provide asymptotically correct estimates (10) (with probability one) under appropriate conditions. A sufficient condition for convergence is that the conditional distributions p X ( i | X -i ) are strictly positive in their domains for all i [8, Ch. 12]. As we will see, this condition is not satisfied by the distribution (5) for the problem under study. This suggests that more sophisticated strategies than conventional Gibbs sampling are needed, as discussed in the next section.
## IV. GIBBS SAMPLING FOR MODULATION CLASSIFICATION
In this section, we design a Gibbs sampler that performs modulation classification in the presence of frequency-selective fading. As outlined in the previous section, the goal is to estimate the posterior probability p ( A | r ) in (8).
## A. Conventional Gibbs Sampling
We first elaborate on the conventional Gibbs sampler for the calculation of the posterior p ( A | r ) that is based directly on the joint distribution (5)-(7). The corresponding BN G 1 is shown in Fig. 1.
Figure 1. BN G 1 for modulation classification based on the factorization (5).
As per the discussion in the previous section, Gibbs sampling requires the knowledge of the conditional distributions of each variable given all others. Calculating these conditional probabilities requires to multiply all the factors in the factorization (5)-(7) that contain the variable of interest and then normalize the resulting distribution [8, Ch. 12] . This leads to the following (see also [5]):
$$p \left ( A \Big | \text{s}, \text{h}, \sigma ^ { 2 }, \text{r} \right ) \in p \left ( A \right ) \left \{ \prod _ { n = - L + 1 } ^ { N - 1 } p \left ( s _ { n } | A \right ) \right \},$$
$$p \left ( s _ { n } \Big | A, s _ { - n }, \mathbf h, \sigma ^ { 2 }, \mathbf r \right ) \in p \left ( s _ { n } | A \right ) p \left ( \mathbf r \Big | A, s, \mathbf h, \sigma ^ { 2 } \right ),$$
$$h \Big | \, ( A, s, h, \sigma ^ { 2 }, r ) \sim \mathcal { C N } ( h _ { * }, \Sigma _ { * } ),$$
$$\text{and } \sigma ^ { 2 } \Big | \left ( A, \{ s _ { n } \} _ { n = - L + 1 } ^ { N - 1 }, \text{h,r} \right ) \sim \mathcal { I } \mathcal { G } \left ( \alpha, \beta \right ),$$
where h ∗ = Σ Σ ∗ ( -1 0 h 0 + S H r σ 2 ) , Σ -1 ∗ = Σ -1 0 + S H S σ 2 , α = α 0 + N and β = β 0 + ‖ r -Sh ‖ 2 . Note that (13) follows from standard MMSE estimation results (see, e.g., [8]) and that (14) is a consequence of the fact that the inverse Gamma distribution is the conjugate prior for the Gaussian likelihood [9].
Gibbs sampling starts with an arbitrary feasible initialization for all variables. In particular, one needs to initialize the constellation A to some value A = a and, correspondingly, the transmitted symbols s to source values belonging to the constellation a . Using (11) and (12), it can be easily seen that conventional Gibbs sampling will never select values of A different from the initial value a . This is due to the fact that the conditional distribution p ( s n | A ) gives zero probability to all values of s n not belonging to A . As a result, Gibbs sampling fails to converge to the
posterior distribution (see also Sec. III). Next, we demonstrate how this problem can be solved by the proposed approach based on the latent Dirichlet BN.
## B. Gibbs Sampling Based on Latent Dirichlet BN
In order to avoid the problem described above, we propose to base the Gibbs sampler on the BN G 2 shown in Fig. 2. In this BN, each transmitted symbol s n is distributed according to a random mixture of uniform distributions on the different constellations. Specifically, we introduce a random vector P A to represent the mixture weights, so that P A ( a ) is the probability that s n takes values in the constellation a ∈ A . The prior distribution of P A is Dirichlet, so that we have P A ∼ Dirichlet ( γ ) for a given set of nonnegative parameters γ = [ γ , 1 · · · , γ |A| ] T [8] . 1 When conditioned on P A , the transmitted symbol variables s n are independent and distributed according to a mixture of uniform distributions, i.e., p ( s n | P A ) = ∑ a s : n ∈ a P A ( a ) / a | | .
The BN G 2 , while departing from the original model (5)-(7), has the advantage that a Gibbs sampler based on it is not limited by the zeros present in the distribution (5)-(7). In particular, thanks to the introduction of the latent variable P A , Gibbs sampling is able to explore different constellations irrespective of its initialization. The idea of introducing the latent Dirichlet variable P A is inspired by [10], where a similar quantity was used to account for the distribution of topics within a document. According to the BN in Fig. 2, the joint pdf p ( P A , s h , , σ 2 , r ) can be factorized as
$$& p \left ( \mathbf P _ { A }, s, \mathbf h, \sigma ^ { 2 }, \mathbf r \right ) \\ = & p \left ( \mathbf P _ { A } \right ) \left \{ \prod _ { n = - L + 1 } ^ { N - 1 } p \left ( s _ { n } | \mathbf P _ { A } \right ) \right \} p \left ( \mathbf h \right ) \cdot \\ & \quad \cdot p \left ( \sigma ^ { 2 } \right ) p \left ( \mathbf r \Big | s, \mathbf h, \sigma ^ { 2 } \right ), \\. \times \dots \dots \dots \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots \mathbf r \dots$$
where, as discussed above, we have P A ∼ Dirichlet( γ ) and p ( s n | P A ) = ∑ a s : n ∈ a P A ( a ) / a | | , while the remaining conditional distributions are as in (6) and (7).
To apply Gibbs sampling based on the factorization (15), the conditional distributions for P A , s n , h and σ 2 conditioned on all other variables are required. It can be shown that the
1 Intuitively, the parameter γ a can be interpreted as the number of symbols in constellation a ∈ A observed during some preliminary measurements.
Figure 2. BN G 2 for the proposed modulation classification scheme based on the Dirichlet latent variable P A .
conditional distribution for h and σ 2 are (13) and (14) respectively. The other required conditional distributions of P A and s n are as follows:
$$\mathbf P _ { A } \Big | \left ( s, \mathbf h, \sigma ^ { 2 }, \mathbf r \right ) \sim \text{Dirichlet} \left ( \gamma + c \right ),$$
where c = [ c , 1 · · · , c |A| ] T and c a is the number of symbols that belong to constellation a ∈ A ;
glyph[negationslash]
$$& \text{and } p \left ( s _ { n } \Big | \mathbf P _ { A }, \{ s _ { k } \} _ { k \neq n } ^ { N - 1 }, \mathbf h, \sigma ^ { 2 }, \mathbf r \right ) \\ & \quad \infty p \left ( s _ { n } | \mathbf P _ { A } \right ) p \left ( \mathbf r \Big | s, \mathbf h, \sigma ^ { 2 } \right ).$$
Note that (16) follows the fact that different modes of the mixture distribution is distributed as a categorical distribution and Dirichlet distribution is the conjugate prior for categorical distribution [8].
The task of modulation classification is achieved by computing the posterior distribution p ( P A | r ) following the Gibbs procedure discussed above. From the posterior p ( P A | r ) , we can then obtain an estimate ˆ A for the constellation as
$$\hat { A } = \arg \max _ { a \in \mathcal { A } } E \left [ \mathbf P _ { A } \left ( a \right ) \, | \, \mathbf r \right ],$$
where the expectation is taken over the distribution p ( P A | r ) .
Remark 1 : The method proposed in [5], based on the introduction of a 'superconstellation', can be seen as an approximation of the approach presented above. Specifically, the scheme of [5] is obtained by setting γ = 0 and by choosing P A to be equal to c / ∑ |A| a =1 c a , at each Gibbs iteration, where we recall that c a is the number of symbols that belong to constellation a ∈ A . Furthermore, the computational complexity of the proposed scheme is comparable to the superconstellation Gibbs sampler [5, Table I].
Remark 2 : At high signal-to-noise ratios (SNR), the relationship between r and s defined by (6) is almost deterministic. Following the discussion in the previous section, this may create convergence problems. This issue can be tackled via the idea of annealing [8, Ch. 12]. Accordingly, the distribution (6) is modified as r ∣ ∣ ∣ ( A, s h , , σ 2 ) ∼ CN ( Sh , ρσ 2 I ) , where ρ is a 'temperature' parameter. The procedure starts with a high value of ρ to prevent the mentioned convergence problems, and then cools to a lower temperature to produce the desired target distribution. Effective cooling schedules include logarithmic and linear decreases of the temperature, whose parameters can be determined based on preliminary runs [11].
## V. NUMERICAL RESULTS AND CONCLUDING REMARKS
In this section, we evaluate the performance of the proposed scheme for the recognition of three modulation formats, namely QPSK, 8-PSK and 16-QAM. We assume Rayleigh fading channels, which are normalized so that E [ ‖ h ‖ 2 ] = 1 . The average SNR is defined as SNR = 1 /σ 2 . The number of samples used by Gibbs sampling are M = 300 and M 0 = 100 in (10). No annealing is used. The performance criterion of interest is probability of correct classification (PCC).
In Fig. 3, we plot the PCC for L = 3 independent taps with relative powers given by [0dB , -0 9dB . , -4 9dB] . . The performance of the proposed method is compared to the superconstellation Gibbs sampler of [5]. The prior distribution Dirichlet ( γ ) for P A is selected so that all elements of the vector γ are identical and equal to a parameter γ . In order to investigate the impact of prior distributions for P A on the classification performance, five values, 0 1 . , 0 5 . , 1 , 10 and 15 are considered for γ . It is observed that the performance is enhanced with a larger γ , especially for higher SNR values. This is because increasing γ enhances the relative importance of the prior distribution and helps improve the convergence properties of the algorithm (see Remark 2). We observe that in practice the value of the hyperparameter γ , as is the case for all of MCMC methods [12], can be determined based on offline preliminary runs. Finally, for sufficiently high SNRs, the superconstellation Gibbs sampler achieves a PCC of about 83% , while the proposed scheme achieves a PCC of 95% with γ = 15 .
Fig. 4 shows the PCC for channels with two paths with non-integer delays [0 , 1 3 . T ] and with relative powers [0dB , -0 9dB] . [13, Ch. 3]. A raised cosine pulse shape with roll-off factor 0.3 is assumed and we set L = 6 . For N = 100 , while there is performance degradation as compared to 3-tap channels due to more severe frequency selectivity, the proposed scheme still can achieve
Figure 3. Probability of correct classification using the proposed method and the superconstellation Gibbs sampling of [5] versus SNR with L = 3 independent taps.
above 90% PCC at sufficiently large SNR, while the method of [5] achieves a PCC of 80% . For N = 400 , both classification schemes obtain performance gains, but the proposed scheme attains a PCC of 95 6% . , while the superconstellation method achieves a PCC of 90% .
Figure 4. Probability of correct classification using the proposed method and the superconstellation Gibbs sampling of [5] versus SNR with two paths and non-integer delays ( L = 6 ).
In summary, as demonstrated by the discussed numerical results, the proposed Gibbs sampling method based on latent Dirichlet Bayesian network provides significant performance gains over
the state of the art.
## REFERENCES
- [1] O. A. Dobre, A. Abdi, Y. Bar-Ness and W. Su, 'A survey of automatic modulation classification techniques: classical approaches and new developments,' IET Communications , vol.1, no. 2, pp. 137-156, Apr. 2007.
- [2] N. Lay and A. Polydoros, 'Modulation classification of signals in unknown ISI environments,' in Proc. IEEE MILCOM , pp. 170-174, San Diego, CA, Nov. 1995.
- [3] P. Panagiotou, A. Anastasopoulos, and A. Polydoros, 'Likelihood ratio tests for modulation classification," in Proc. IEEE MILCOM , pp. 670-674, Los Angeles, CA, Oct. 2000,
- [4] F. Hameed, O. A. Dobre, and D. C. Popescu, 'On the likelihood-based approach to modulation classification," IEEE Trans. Wireless Commun ., vol. 8, no. 12, pp. 5884-5892, Dec. 2009.
- [5] T. A. Drumright and Z. Ding. 'QAM constellation classication based on statistical sampling for linear distortive channels,' IEEE Trans. Signal Process ., vol. 54, no. 5, pp. 1575-1586, May 2006.
- [6] A. Doucet and X. Wang, 'Monte Carlo methods for signal processing,' IEEE Signal Process . Mag ., vol. 22, no. 6, pp. 152-170, Nov. 2005.
- [7] J. R. Barry, E. A. Lee and D. G. Messerschmitt, Digital Communication , Springer, 2003.
- [8] D. Koller and N. Friedman, Probabilistic Graphical Models: Principles and Techniques , MIT Press, 2009.
- [9] K. P. Murphy, 'Conjugate Bayesian analysis of the Gaussian distribution,' Univ. of British Columbia, Canada, Tech. Rep ., 2007 [Online]. Available: http://www.cs.ubc.ca/~murphyk/Papers/bayesGauss.pdf
- [10] D. Blei, A. Ng, and M. Jordan. 'Latent Dirichlet allocation,' Journal of Machine Learning Research , vol. 3, pp. 993-1022, Jan. 2003.
- [11] Y. Nourani and B. Andresen, 'A comparison of simulated annealing cooling strategies,' J. Phys . A , vol. 31, no. 41, pp. 8373-8385, July 1998.
- [12] R. M. Neal, 'Sampling from multimodal distributions using tempered transitions,' Statistics and computing , vol. 6, no. 4, pp. 353-366, June 1996.
- [13] A. Goldsmith, Wireless Communications , Cambridge University Press, 2005.
|
10.1109/LSP.2014.2327193
|
[
"Yu Liu",
"Osvaldo Simeone",
"Alexander M. Haimovich",
"Wei Su"
] |
2014-08-04T18:47:24+00:00
|
2014-08-20T02:04:25+00:00
|
[
"cs.IT",
"cs.CV",
"math.IT"
] |
Modulation Classification via Gibbs Sampling Based on a Latent Dirichlet Bayesian Network
|
A novel Bayesian modulation classification scheme is proposed for a
single-antenna system over frequency-selective fading channels. The method is
based on Gibbs sampling as applied to a latent Dirichlet Bayesian network (BN).
The use of the proposed latent Dirichlet BN provides a systematic solution to
the convergence problem encountered by the conventional Gibbs sampling approach
for modulation classification. The method generalizes, and is shown to improve
upon, the state of the art.
|
1408.0768v1
|
## NuSTAR Observations of the Magnetar 1E 2259+586
Julia K. Vogel 1 2 , , Romain Hasco¨ et ,Victoria M. Kaspi , Hongjun An , Robert Archibald , 3 4 4 4 Andrei M. Beloborodov , Steven E. Boggs , Finn E. Christensen , William W. Craig 3 5 6 2 5 , , Eric V. Gotthelf , Brian W. Grefenstette , Charles J. Hailey , Fiona A. Harrison , Jamie A. 3 7 3 7 Kennea , Kristin K. Madsen , Michael J. Pivovaroff , Daniel Stern , William W. Zhang 8 7 2 9 10
1 Corresponding author, [email protected]
2 Physics Division, Physical and Life Sciences Directorate, Lawrence Livermore National Laboratory, Livermore, CA 94550, USA
3
Physics Department and Columbia Astrophysics Laboratory, Columbia University, New
York, NY 10027, USA
4
Department of Physics, McGill University, Montreal, Quebec H3A 2T8, Canada
5
Space Sciences Laboratory, University of California, Berkeley, CA 94720, USA
6
DTU Space, National Space Institute, Technical University of Denmark, Elektrovej 327,
DK-2800 Lyngby, Denmark
7 Cahill Center for Astronomy and Astrophysics, California Institute of Technology,
Pasadena, CA 91125, USA
8 Department of Astronomy and Astrophysics, Pennsylvania State University, University Park, PA 16802, USA
9
Jet
Propulsion
Laboratory,
California
Institute of
Technology,
Pasadena,
CA 91109,
USA
10 Goddard Space Flight Center, Greenbelt, MD 20771, USA
## ABSTRACT
We report on new broad band spectral and temporal observations of the magnetar 1E 2259+586, which is located in the supernova remnant CTB 109. Our data were obtained simultaneously with the Nuclear Spectroscopic Telescope Array (NuSTAR) and Swift , and cover the energy range from 0 5 . -79 keV. We present pulse profiles in various energy bands and compare them to previous RXTE results. The NuSTAR data show pulsations above 20 keV for the first time and we report evidence that one of the pulses in the double-peaked pulse profile shifts position with energy. The pulsed fraction of the magnetar is shown to increase strongly with energy. Our spectral analysis reveals that the soft X-ray spectrum is well characterized by an absorbed double-blackbody or blackbody plus power-law model in agreement with previous reports. Our new hard X-ray data, however, suggests that an additional component, such as a power-law, is needed to describe the NuSTAR and Swift spectrum. We also fit the data with the recently developed coronal outflow model by Beloborodov (2013a) for hard X-ray emission from magnetars. The outflow from a ring on the magnetar surface is statistically preferred over outflow from a polar cap.
Subject headings: pulsars: individual (1E 2259+586) - stars: magnetars - stars: neutron - X-rays: bursts
## 1. Introduction
Magnetars are a class of young, isolated neutron stars (NSs). Their defining features are their sporadic outbursts in X-rays and soft gamma-rays along with luminosities often much larger than their expected spin-down powered X-ray emission. To date, 21 confirmed magnetars and 5 candidates have been reported 1 (Olausen & Kaspi 2014). Spin periods for magnetars lie in the narrow range of P = 2 1 . -11 8 s with period derivatives of the order of . ˙ P = 10 -11 s s -1 . The characteristic ages of these objects ( τ = P/ P 2 ˙ ) imply that magnetars are rather young neutron stars with ages of the order of several thousand years. However, especially for young objects, the spindown-inferred age is only a crude estimate. The young ages are, however, generally supported for most magnetars by their specific locations as well as their association with supernova remnants (SNRs). Both magnetar outbursts and a substantial fraction of their persistent X-ray emission are believed to be powered by their intense magnetic fields (Thompson & Duncan 1993; 1996), with strengths of 10 14 -10 15 G inferred assuming magnetic dipole braking in vacuum ( B ∝ √ PP ˙ ).
The magnetar model was originally formulated to explain the behavior of a sub-class of magnetars, the so-called Soft Gamma Repeaters (SGRs), the most active of these objects. However, observations of similar behavior such as bursting activity (Gavriil et al. 2002; Kaspi et al. 2003) for Anomalous X-ray Pulsars (AXPs), another previously identified magnetar sub-class, indicate that SGRs and AXPs are likely to be at different ends of a continuous magnetar activity spectrum.
One of the most studied magnetars to date is 1E 2259+586. It was the first AXP to be discovered (Fahlman & Gregory 1981; 1983) and also played a key role in the unification of these neutron stars with very high magnetic fields, when it exhibited a
1 http://www.physics.mcgill.ca/ ~ pulsar/magnetar/main.html
Fig. 1.ROSAT PSPC image of SNR CTB 109 (0 1 . -2 4 keV). The white frame indicates . the NuSTAR field of view (FoV), but no corresponding data have been included here. The bright point source is the magnetar 1E 2259+586.
series of over 80 X-ray bursts in 2002 (Kaspi et al. 2003). This AXP has shown a very stable spin-down rate (Kaspi et al. 1999) as well as pulsed flux emission apart from glitches 2 in 2002 and 2007 (Icdem et al. 2012; Dib & Kaspi 2014) and a rare anti-glitch in 2012 (Archibald et al. 2013). 1E 2259+586 is located near the center of SNR CTB 109 (G109 1 . -1 0), . which is known for its half-shell morphology in both radio and X-rays (see Fig. 1). It is at a distance of 4 0 . ± 0 8 kpc (Tian et al. . 2010). The spin period of 1E 2259+586 is P = 6 98 s with a spin-down rate of . ˙ P = 0 05 . × 10 -11 s s -1 . This implies a surface dipole magnetic field of 0 59 . × 10 14 G, which is towards the lower end
2 A glitch is a sudden spin-up of a neutron star. In one case a sudden spin-down has been observed (Archibald et al. 2013), which is referred to as an anti-glitch.
of the typical range of magnetar magnetic fields, and the characteristic age is estimated to be 230 kyr. While no radio counterpart at the position of the X-ray pulsar has been identified down to a level of 7 µ Jy (Archibald et al. 2013), persistent emission has been detected from the mid- to far-infrared, all the way up to X-rays (see, e.g., Kuiper et al. 2006 for a brief review of previous observations). The soft X-ray spectrum is typical of AXPs and, in the 0 5 . -7 0 keV band, was described by Patel et al. . (2001) as a blackbody spectrum with temperature kT = 0 412(6) keV and a soft power-law with Γ = 3 6(1) for . . N H = 0 93(3) . × 10 22 cm -2 . Zhu et al. (2008) updated these values in the 0 6 . -12 0 keV . band with XMM-Newton data taken post-outburst in 2002, yielding kT = 0 400(7) keV . and Γ = 3 75(3) for . N H = 1 012(7) . × 10 22 cm -2 . It is interesting that significant spectral changes were observed by Woods et al. (2004) at energies < 10 keV when comparing preand post-burst data for the 2002 outburst. Timing and spectral properties for hard X-ray emission (8 -24 keV) from 1E 2259+586 were first derived by Kuiper et al. (2006) using RXTE data. They also compared their results for the pulsed spectrum to upper limits from IBIS ISGRI and COMPTEL (Den Hartog et al. 2006). For the RXTE data, an absorbed double power-law model yields a good fit to the pulsed spectrum with soft and hard power-law indices of Γ 1 = 4 26(1) and Γ . 2 = -1 02 . +0 24 . -0 13 . for N H = 0 93(3) . × 10 22 cm -2 . Therefore, this AXP follows others in exhibiting an onset of dramatic hardening above 10 keV. For 1E 2259+586, Kuiper et al. (2006) found that the power-law components become equally strong at E = 15 8 . ± 2 3 keV. For energies beyond 20 keV, conclusive . confirmation of the dramatic hardening for 1E 2259+586 has not been available until now.
In this paper, we report on the spectral and temporal properties of the magnetar 1E 2259+586 in the 0 5 . -79 keV band. Our analysis is based on data acquired with the Nuclear Spectroscopic Telescope Array ( NuSTAR ) and the Swift X-Ray Telescope (XRT). In Section 2, we describe the observations, and following this we present the results of our data analysis in Section 3. We apply the coronal outflow model of Beloborodov (2013a) to
the hard X-ray emission data, and show that the model is consistent with the phase-resolved spectra. A discussion of our results can be found in Section 4, and in Section 5 we conclude and summarize our findings.
## 2. Instruments and Observations
NuSTAR is the first hard X-ray focusing telescope in space with sensitivity in the energy range of 3 -79 keV (Harrison et al. 2013). The instrument consists of two co-aligned focusing optics (Hailey et al. 2010) with CdZnTe detectors in the focal plane (Harrison et al. 2010). Each focal-plane module consists of four detector chips and the two telescope modules are dubbed FPMA and FPMB. NuSTAR 's energy resolution is 400 eV at 10 keV (Full Width at Half Maximum, FWHM) and the observatory provides an angular resolution of 58 ′′ HPD 3 (18 ′′ FWHM). Its temporal resolution of 2 µ s is more than adequate for studying the 6 98 s AXP 1E 2259+586. . For more information on NuSTAR , see Harrison et al. (2013).
Swift (Gehrels et al. 2004) is a multi-wavelength mission dedicated to the study of gamma-ray bursts. The Swift XRT (Burrows et al. 2005) is one of three instruments flown on the satellite. It is a Wolter-I type optic with a Charge-Coupled Device (CCD) focal plane detector. The instrument is sensitive in the energy range from 0 2 to 10 keV with an . energy resolution of 140 eV at 5 9 keV and a point spread function of 22 . ′′ (HPD) at 8 1 keV. .
The NuSTAR observations of 1E 2259+586 began on 24 April 2013 at UT 21:51:07 with the last of four observations concluding on 18 May 2013 at UT 06:01:07. The total net exposure for the combination of all four observations is 157 4 ks. . A simultaneous Swift observation (XRT, photon-counting mode) was carried out to extend the spectral coverage
3 Half-Power Diameter
down to ∼ 0 5 keV. While . NuSTAR is very sensitive in the hard X-ray band, the low energy data from the Swift XRT (0 2 . -10 keV) improve our ability to constrain the thermal components which are softer and emit mostly below the NuSTAR pass-band. The Swift observation started at UT 00:23:55 on 25 April 2013 and was split into three parts with a total exposure time of 29 9 ks. .
The NuSTAR data were processed using the standard NuSTAR Data Analysis Software NuSTARDAS version 1.2.0 along with CALDB version 20131007 and HEASOFT version 6.13 as available on HEASARC . For the 4 Swift data, cleaned event files were produced using xrtpipeline along with the HEASARC remote CALDB 5 employing the standard filtering procedure of Capalbi et al. (2005). Further processing steps applied to these files are outlined below. Details about the different observations are shown in Table 1, which summarizes all data sets used in this analysis. Figure 2 shows the unsmoothed, exposure-corrected NuSTAR image of 1E 2259+586 in two energy bands, chosen to yield a comparable number of counts in the signal extraction region of radius 60 . ′′
## 3. Analysis and Results
## 3.1. Timing Analysis
For the timing analysis, we extracted photons in a circular region of 60 ′′ radius around the nominal source position for the NuSTAR data, while for Swift we used a 22 ′′ radius. The energy bands used for spectral analysis were 5 -79 keV and 0 5 . -10 keV for NuSTAR and Swift , respectively. Since 1E 2259+586 is located in CTB 109, the SNR background must be considered. Sasaki et al. (2004) pointed out that no emission above 4 keV was
4 See http://heasarc.gsfc.nasa.gov
5 See http://heasarc.nasa.gov/docs/heasarc/caldb/caldb\_remote\_access.html
Fig. 2.- Images of unsmoothed, exposure-corrected NuSTAR data in two energy ranges (Left: 3 -4 keV, right: 4 -79 keV) with comparable numbers of counts in the signal extraction region of 60 ′′ radius.
observed in the Chandra data apart from the pulsar and the NuSTAR data agrees with this. Since in one of the two NuSTAR modules (FPMA) stray light is observable at the edge of the field of view, we excluded this region when defining the background extraction area. For NuSTAR , background was extracted from a circular region of radius 100 ′′ nearby the source region. We also compared our results using these background regions to those obtained using the mission's background model nuskybgd (Wik et al. 2014), and confirmed that all results were compatible within the corresponding uncertainties.
Due to the good spatial resolution of the Swift XRT, background could be extracted from an annular region around the neutron star with an inner radius of 80 ′′ and an outer radius of 200 . ′′ Before continuing we checked if any pile-up events were present in the data.
We followed the standard Swift procedure 6 and determined the count rate in a circular region of 47 2 . ′′ radius (20 pixels) centered on the source to be above 0 5 ct/s, indicating that . the central 1-2 pixel radius region of the bright core might suffer from pile-up. We removed this area for the subsequent analysis by substituting the circular source extraction region by an annulus with inner and outer radii of 5 ′′ and 22 , respectively. ′′ We also re-generated the ARFs to correct the flux of the spectrum for the loss of counts due to the use of an annular region as well as the correction factor for the light curves 7 .
In the next step, we applied a barycentric correction to the selected source events using the multi-mission tool barycorr with the corresponding orbital and clock correction files at the Chandra position of α = 23 01 h m 08 s . 295 and δ = +58 52 44 ◦ ′ ′′ . 45 (J2000.0) reported by Patel et al. (2001). Following the H-test method (De Jager et al. 1989), we searched for pulsations and determined the best period for the significant pulsations to be P ( NuSTAR ) = 6 97914(2) s and . P ( Swift ) = 6 97915(2) s. . The measured periods were in agreement with those obtained from the ephemeris of the Swift monitoring program of 1E 2259+586 (see Table 2).
The resulting pulse profiles are shown in Figure 3 for six energy ranges: 3-4 keV, 4-8 3 keV, 8 3-11 9 keV, 11 9-16 3 keV, 16 3-24 0 keV, 24 0-79 0 keV. The bands up to . . . . . . . . . 24 0 keV were chosen to coincide with those reported by Kuiper et al. . (2006) for direct comparison. All pulse profiles, with the exception of the highest energy band (24 -79 keV), are background-subtracted. The statistical significance of the observed pulsations in all
6 See http://www.swift.ac.uk/analysis/xrt/pileup.php
7 See http://www.swift.ac.uk/analysis/xrt/arfs.php
Table 1: NuSTAR and Swift observations of 1E 2259+586 used in this study.
| Observatory | Mode | ObsID | Date [MJD] a | Date | Exp [ks] |
|---------------|--------|-------------|----------------|----------------|------------|
| NuSTAR | · · · | 30001026002 | 56406 | April 24, 2013 | 37.3 |
| NuSTAR | · · · | 30001026003 | 56407 | April 25, 2013 | 15.4 |
| NuSTAR | · · · | 30001026005 | 56408 | April 26, 2013 | 16.3 |
| NuSTAR | · · · | 30001026007 | 56428 | May 16, 2013 | 88.4 |
| Swift | PC | 00080292002 | 56407 | April 25, 2013 | 13.1 |
| Swift | PC | 00080292003 | 56408 | April 26, 2013 | 14.1 |
| Swift | PC | 00080292004 | 56410 | April 28, 2013 | 2.7 |
a At the start of data acquisition.
Table 2: Swift Ephemeris for 1E 2259+586 (Archibald et al. 2013).
| Parameter | Value |
|------------------------------------------|-------------------------|
| Spin frequency f 0 [s - 1 ] | 0 . 14328414235(81) |
| Spin frequency derivative ˙ ν 0 [s - 2 ] | - 10 . 14(12) × 10 - 15 |
| Epoch | MJD 56250 . 000 |
| Start observing epoch | MJD 56201 . 284 |
| Stop observing epoch | MJD 56439 . 246 |
energy bands is larger than 99% (p-value 8 smaller than 0 01). . For the lower five energy bands, the pulse profiles agree generally with those reported by Kuiper et al. (2006); however, in the energy ranges between 11 9 and 24 0 keV slight differences are apparent. . . In the highest energy band (24 -79 keV), significant pulsations are observable for the first time. We note, however, that due to limited statistics we did not subtract the background in order to avoid negative bin counts. The nominal background count rate in this case is 6 56 . × 10 -4 cts / s.
As previously pointed out by Kuiper et al. (2006), there is a gradual change in pulse morphology observable with increasing energy. The double-peak profile at low energies (below ∼ 8 keV) evolves into a less pronounced double-peak or possibly a single peak profile for intermediate energies (about 8 -16 keV) and back to a double-peak structure at higher energies. A possible way to describe this development is that while for energies below ∼ 8 keV, the pulse near phase 0 9 ('Peak A') tends to dominate, it becomes less significant . for increasing energy with respect to the second peak ('Peak B') near phase 1 45. . In the intermediate energy range (8 -16 keV), both peaks are approximately equally pronounced, while peak B starts to slightly dominate at energies above ∼ 16 keV. A difference between NuSTAR and RXTE profiles is apparent in the energy bands of 11 9 . -16 3 keV and . 16 3 . -24 0 keV. While Kuiper et al. . (2006) observe a clearly dominating peak B in this energy range, our analysis shows only a small lead for this pulse over peak A.
In order to quantify the energy dependence of the pulse morphology, we decompose the pulse profiles in terms of their Fourier components. This decomposition is shown in Figure 4. We find six harmonics to be sufficient, and we included these in the plots. We
8 The p-value is the probability of obtaining a test statistic which is at least as extreme as the one that was actually observed under the assumption that the null hypothesis (in our case, no pulsation) is true.
Fig. 3.- Pulse profiles for 1E 2259+586 from NuSTAR data in various energy bands. All but the pulse profile in the band from 24 -79 keV, which has limited statistics, are backgroundsubtracted. Energy bands up to 24 keV were chosen to enable direct comparison with Kuiper et al. (2006). The dashed lines near phase 0 9 and 1 45 are meant to serve as a guide to the . . eye to compare alignment of peak A and peak B. Note that the y-axis labels differ and the zero is suppressed in some of the plots. Also shown is the Fourier representation of the pulse profiles (red) used to estimate the pulsed emission level, denoted by the horizontal line.
observe that the ratio of the second harmonic to the first harmonic changes with energy. For energies below about 8 keV, the second harmonic dominates over the first, while in the intermediate range of 8 -16 keV the first harmonic takes the lead, before the second harmonic dominates again for 16 -24 keV. If the main contribution originates from the first
Fig. 4.- Evolution of the Fourier power harmonic distribution of the pulse profile for 1E 2259+586 as a function of energy. The energy bands shown correspond to those displayed in Figure 3. Power levels have been normalized to the total power of the first six harmonics. Note that the error bars for the 24 -79 keV band are artificially small, since the corresponding pulse profile has not been background-subtracted. Errors will significantly increase in this energy band (due to low statistics) if background is taken into consideration.
harmonic number, this indicates that there is either one single peak in the pulse profile or a double peak with maxima of similar magnitude, while a leading second harmonic number points towards a double peak with pulses of different height and/or shape. This Fourier decomposition therefore reflects the change in dominance of one peak over the other in the
double-peak morphology or, likewise, a possible change from a 2-peak structure to one peak and back. For the NuSTAR band ( E > 24 keV), the first and third harmonics dominate (2, 4, and 6 contribute at the 10% level). When interpreting this power spectrum (lower right plot of Figure 4) one should keep in mind that background was not subtracted and the peak-to-valley ratio is not very large. Here the first and (small) second harmonic take care of the wide peak B (from phase 1 -1 8, including the small feature in the main minumum), . while the third harmonic describes the narrower, subdominant peak A.
As previously reported by Kuiper et al. (2006) as well as Gavriil & Kaspi (2002), a small pulse-like feature can be observed in the main minimum around phase 0 60 . -0 65. . This feature is the main reason for the presence of higher orders of harmonics ( n ≥ 3) in the Fourier decomposition (see Fig. 4) of the pulse profile.
We also note that there is a hint of a small feature in the intermediate minimum between peak A and B around phase 1 2, best seen in the lowest two energy bands. . With the current statistics, however, we do not consider this feature to be significant.
Another observation is a possible shift of the peak B position in phase towards lower values with increasing energy. Dashed lines near phase 0 9 and 1 45 have been added to . . Figure 3 to guide the eye for alignment comparison of peak A and peak B. In order to test this pulse-shift hypothesis we determined the mean value of the peak position for each peak in various energy bands by considering the unbinned group phase between phase 0 75 . -1 1 . for peak A and 1 2 . -1 55 for peak B. In addition, we crosschecked our results by fitting . a Gaussian to the peaks. The results for the mean and its error are listed in Table 3 and indicate that peak A remains at the same phase position of 0 937(17) throughout all energy . ranges (0 933(3) for . E /lessorsimilar 8 keV, 0 940(16) for . E /greaterorsimilar 8 keV). For peak B, there is a slight tendency to change position towards lower phase values for higher energies. Below ∼ 8 keV, peak B can be found at 1 384(3), above these energies it is located at 1 349(17). . . We note
that the shift is not very significant at the given statistics, and more data are needed to confirm or rule out the observed tendency.
In addition, we used the ephemeris-folded profiles to determine the pulsed fraction (PF) as a function of energy for our observations. A pulse profile that changes with energy or time makes it challenging to determine the pulsed fraction of the source accurately and different methods are commonly used in the literature (see e.g. Archibald et al. 2007, Archibald et al. 2014 for a discussion). All of them have their advantages and disadvantages and therefore there are some caveats to keep in mind. The area pulsed fraction is probably the most natural and physically meaningful definition. It is defined as the difference between the pulsed flux and the constant flux integrated over a full phase cycle and can be calculated according to
$$P F _ { \text{area} } = \frac { \frac { 1 } { N } \sum _ { j = 1 } ^ { N } p _ { j } - p _ { \min } } { \frac { 1 } { N } \sum _ { j = 1 } ^ { N } p _ { j } }, \\ r \text{ of events in the $i$-th nhase bin. $N$ is the total number of bins. and}$$
where p j is the number of events in the j -th phase bin, N is the total number of bins, and p min is the minimum flux. Note that the determination of the true minimum and its error
Table 3: Mean peak position for peak A and B in various energy bands.
| Energy band [keV] | Position peak A | Error peak A | Position peak B | Error peak B |
|---------------------|-------------------|----------------|-------------------|----------------|
| 3 . 0 - 4 . 0 | 0 . 93 | < 0 . 01 | 1 . 39 | < 0 . 01 |
| 4 . 0 - 8 . 3 | 0 . 94 | < 0 . 01 | 1 . 38 | < 0 . 01 |
| 8 . 3 - 11 . 9 | 0 . 94 | 0 . 01 | 1 . 35 | 0 . 01 |
| 11 . 9 - 16 . 3 | 0 . 94 | 0 . 01 | 1 . 35 | 0 . 01 |
| 16 . 3 - 24 . 0 | 0 . 93 | 0 . 01 | 1 . 36 | 0 . 01 |
is the main challenge and the biggest caveat when using PF area , because both noise and binning show a tendency to bias the values of the area pulsed fraction upwards. In Figure 3 we included the results for p min as the minimum off-pulse value determined from the Fourier representation (horizontal magenta line). The results for the area pulsed fraction using the p min determined in this way are shown in Figure 5. Note that the NuSTAR energy bands here (black squares) correspond to those presented in Kuiper et al. (2006) for the RXTE PCA (five bands in the range from 3 0 . -24 0 keV) and HEXTE data (14 8 . . -27 0 keV). . The Swift data (blue diamonds) covers the two lowest RXTE PCA bands in Kuiper et al. (2006) plus an additional band from 0 3 . -3 0 keV. Background was subtracted in all energy . bands.
The small inset plot in the same Figure displays the standard pulsed fraction (peak-to-peak pulsed fraction), conventionally defined as ratio of peak flux minus minimum flux to the total flux
$$P F _ { \text{P2P} } = \frac { F _ { \max } - F _ { \min } } { F _ { \max } + F _ { \min } }.$$
This quantity is extensively used in the literature since it is straightforward to calculate, although the difficulty in determining the maximum and minimum flux values accurately is biased in a way similar to the area pulsed fraction. Furthermore, the PF P2P cannot provide information on the total energy since the peak width does not feed into the flux variation. The PF area and PF P2P derived from our data agree with each other within uncertainties and show the same trend: the pulsed fraction increases dramatically with increasing energy. For energies around 10 keV, PF area is 60% ± 12%, while for energies around 20 keV the fraction reaches 96% ± 14%.
Following the method presented in Gonzalez et al. (2010) (see also the forthcoming publication by Archibald et al. 2014 for details), we additionally used a root-mean-square
Fig. 5.- Area pulsed fraction as function of energy for 1E 2259+586 from NuSTAR (black squares) and Swift data (blue diamonds). The energy bands correspond to those presented in Kuiper et al. (2006) for the RXTE PCA and HEXTE data. The small inset plot shows the conventionally defined PF P2P . In both cases the pulsed fraction strongly increases with energy.
(RMS) estimator as a measure for the deviation of the pulsed flux from its mean defined as
$$P F _ { \text{RMS} } = \frac { \sqrt { 2 \sum _ { k = 1 } ^ { 6 } \left ( \left ( a _ { k } ^ { 2 } + b _ { k } ^ { 2 } \right ) - \left ( \sigma _ { a _ { k } } ^ { 2 } + \sigma _ { b _ { k } } ^ { 2 } \right ) \right ) } } { a _ { 0 } }$$
where the Fourier coefficients a k and b k are
$$a _ { k } = \frac { 1 } { N } \sum _ { j = 1 } ^ { N } p _ { j } \cos \left ( \frac { 2 \pi k j } { N } \right ),$$
$$b _ { k } = \frac { 1 } { N } \sum _ { j = 1 } ^ { N } p _ { j } \sin \left ( \frac { 2 \pi k j } { N } \right )$$
and σ a k , σ b k are the uncertainties of a k and b k , respectively,
$$\sigma _ { a _ { k } } ^ { 2 } = \frac { 1 } { N ^ { 2 } } \sum _ { j = 1 } ^ { N } \sigma _ { p _ { j } } ^ { 2 } \cos ^ { 2 } \left ( \frac { 2 \pi k j } { N } \right ),$$
$$\sigma _ { b _ { k } } ^ { 2 } = \frac { 1 } { N ^ { 2 } } \sum _ { j = 1 } ^ { N } \sigma _ { p _ { j } } ^ { 2 } \sin ^ { 2 } \left ( \frac { 2 \pi k j } { N } \right ).$$
p j is again the number of events in the j -th phase bin, σ p j is the uncertainty of p j , and N is the total number of bins. The number of Fourier harmonics included is n = 6. Note that PF RMS is the RMS deviation of the flux from its average and can be defined in the Fourier domain. It does not cover the full range of values from 0 -1 without additional (pulse shape dependent) renormalization, i.e. it is not a pulsed fraction in the conventional sense, even though it is often referred to as the RMS pulsed fraction in the literature. This definition yields a very robust measure for the pulsed power that is less sensitive to noise and therefore signficantly less biased than the area pulsed fraction and more meaningful than the peak-to-peak pulsed fraction. Figure 6 shows the RMS variation for NuSTAR (black squares) and Swift data (blue diamonds). Note that the energy bands again correspond to those presented in Kuiper et al. (2006) for the RXTE PCA (five bands in the range from 3 0 . -24 0 keV) and HEXTE data (14 8 . . -27 0 keV) and represent background-subtracted . data in all energy bands. Since the definition of PF RMS differs from those of PF area and PF P2P as mentioned above, PF RMS does not have the same values as the other definitions in each energy band, but the same trend is observable in all cases: the pulsed fraction increases dramatically with increasing energy. Without renormalization we determined that the PF RMS is 32% ± 9% for energies around 10 keV, while for energies around 20 keV it reaches 71% ± 15%.
Fig. 6.- RMS deviation of the flux from the mean as function of energy for 1E 2259+586 from NuSTAR (black squares) and Swift data (blue diamonds). The energy bands correspond to those presented in Kuiper et al. (2006) for the RXTE PCA and HEXTE data. Similarly to the area and peak-to-peak pulsed fraction also the PF RMS strongly increases with energy, even though the absolute values differ due to the definition of the RMS estimator.
## 3.2. Spectral Analysis
3.2.1. Phase-Averaged Spectral Analysis
In this Section, we use the Swift and NuSTAR observations to derive a phase-averaged spectrum. The same extraction regions as defined in Section 3.1 were used and pile-up for the Swift data was corrected for as described above.
We started with fitting the Swift data only in order to compare the results to previous soft X-ray measurements (Zhu et al. 2008) as well as to obtain a best-fit N H to be used for the combined of Swift and NuSTAR data sets. All extracted spectra
were grouped to have at least 50 counts per bin using the grppha tool. We fit the resulting spectra with XSPEC v12.8.1 using an absorbed blackbody (BB) plus power-law (PL) model ( tbabs*(bbody+powerlaw) ) as well as an absorbed double blackbody ( tbabs*(bbody+bbody) ). The Swift spectrum is a little softer (Γ = 4 1(1)) than previously . reported (Γ = 3 75(3), Zhu et al. . 2008), but consistent within uncertainties with these earlier results. Our best-fit result yields N H = 1 10(6) . × 10 22 cm -2 in the case of the blackbody plus power-law and N H = 0 60(4) . × 10 22 cm -2 for the double-blackbody model. The first value agrees well with estimates of Durant & van Kekwijk (2006) obtained from fitting individual absorption edges of O, Fe, Ne, Mg and Si in the XMM-Newton RGS spectra. The latter value is in agreement with Zhu et al. (2008) and consistent with the best-fit N H -value for CTB 109, N H = (0 5 . -0 7) . × 10 22 cm -2 , as measured by Sasaki et al. (2004).
In a next step we fit the spectra from 0 5 . -79 keV by using the NuSTAR and Swift data sets together. All model parameters were tied except for the cross-normalization factors, which were set to 1 0 for . NuSTAR 's FPMA and left to vary for FPMB and Swift . Spectra were rebinned to have at least 50 counts per bin. We started out with fitting an absorbed double blackbody ( tbabs*(bbody+bbody) ) but the fit was poor, yielding a χ 2 per degree of freedom (dof) of 1448 / 827. The fit improved slightly by using an absorbed blackbody plus power-law ( tbabs*(bbody+powerlaw) ), with χ 2 over dof of 1152/827. Adding an additional component improved the fit significantly and therefore a more complex model (BB+double PL or BB+broken PL, χ 2 /dof=689/825 or 747/825, respectively) than suggested by previous data was favored. The F-test supports this conclusion, yielding a probability significantly smaller than 0 05. . All spectral fit results can be found in Table 4, while Figure 7 shows our best-fit model for the spectra of NuSTAR and Swift .
Table 4: Best-fit parameters for 1E 2259+586 for phase-averaged and phase-resolved analysis.
| Phase | Data a Energy [keV] | Fit | Model b [ × | N H c 10 22 cm - 2 ] | kT [keV] | Γ s /kT d [-/keV] | E break e [keV] | Γ h f | Flux g | Ratio h χ 2 / | dof |
|-----------|-----------------------|--------------|---------------|------------------------|------------|---------------------|-------------------|----------|-----------|-----------------|------------|
| 0-1 | S | 0 . 5-10 . 0 | BB+BB | 0 . 60(4) | 0 . 35(2) | 0 . 71(6) | · · · | · · · | 1 . 9(2) | 1 . 00(1) | 331 / 537 |
| 0-1 | S | 0 . 5-10 . 0 | BB+PL | 1 . 10(6) | 0 . 42(2) | 4 . 1(1) | · · · | · · · | 2 . 1(1) | 1 . 71(9) | 402 / 537 |
| 0-1 | S/N | 0 . 5-79 . 0 | BB+BB | 0 . 60(4) | 0 . 423(3) | 1 . 78(4) | · · · | · · · | 2 . 25(5) | 0 . 2(1) | 1448 / 827 |
| 0-1 | S/N | 0 . 5-79 . 0 | BB+PL | 1 . 10(6) | 0 . 38(1) | 3 . 91(3) | · · · | · · · | 2 . 2(1) | 2 . 5(3) | 1152 / 827 |
| 0-1 | S/N | 0 . 5-79 . 0 | BB+2PL | 1 . 10(6) | 0 . 410(7) | 4 . 34(3) | · · · | 0.4(1) | 2 . 36(7) | 0 . 31(2) | 689 / 825 |
| 0-1 | S/N | 0 . 5-79 . 0 | BB+bPL | 1 . 10(6) | 0 . 409(9) | 4 . 08(3) | 11 . 5(3) | 1 . 2(1) | 3 . 0(1) | 2 . 2(2) | 747 / 825 |
| Peak A | S/N | 0.5-79.0 | BB+2PL | 1 . 10(6) | 0.416(6) | 4.57(7) | · · · | 1.0(3) | 0.68(3) | 0.29(4) | 716/803 |
| Peak B | S/N | 0.5-79.0 | BB+2PL | 1 . 10(6) | 0.399(5) | 4.41(5) | · · · | 0.3(1) | 0.73(3) | 0.53(4) | 645/812 |
| Off-pulse | S/N | 0.5-79.0 | BB+2PL | 1 . 10(6) | 0.404(6) | 4.87(9) | · · · | 1.0(4) | 0.48(3) | 0.37(6) | 879/1031 |
| Peak A | S/N | 0.5-79.0 | BB+bPL | 1 . 10(6) | 0.397(5) | 4.37(4) | 10.8(5) | 1.38(1) | 1.51(4) | 0.78(2) | 776/803 |
| Peak B | S/N | 0.5-79.0 | BB+bPL | 1 . 10(6) | 0.386(5) | 4.26(5) | 10.9(4) | 0.9(2) | 1.46(2) | 0.85(1) | 696/812 |
| Off-pulse | S/N | 0.5-79.0 | BB+bPL | 1 . 10(6) | 0.393(5) | 4.66(5) | 9.6(4) | 1.4(4) | 1.43(3) | 0.47(1) | 911/1031 |
a NuSTAR 5 -79 keV (N), Swift 0 5 . -10 keV (S)
b Blackbody (BB), power-law (PL), broken power-law (bPL)
c N H was frozen to the value obtained from the phase-averaged spectral analysis for the phase-resolved analysis.
d
Soft photon index Γ s if BB+PL model is used. Blackbody temperature of second (hot) BB if BB+BB is fit.
e Break energy for the broken power-law (where applicable).
f Photon index of the hard power-law component.
g Unabsorbed flux in units of 10 -11 erg cm -2 s -1 from 2 -10 keV for Swift and 2 -79 keV if NuSTAR data are included for all model components (BB+BB, BB+PL, BB+bPL) or two PL components (BB+2PL).
h Ratio of flux in the 2 -10 keV range for Swift and 2 -79 keV range if NuSTAR is included. We show F BB(hot) /F BB(cold) (BB+BB), F PL /F BB (BB+PL), F bPL /F BB (BB+bPL) and F PL(hard) /F PL(soft) ( BB + 2 PL ).
Note. - Numbers in parentheses indicate the 1 σ uncertainty in least significant digit. Cross-normalization factors were used when data from NuSTAR and Swift were combined. These factors were set to 1 0 for module . A of NuSTAR and to 1 for Swift if no NuSTAR data were included. Cross-normalization factors were frozen to values obtained for phase-averaged spectral analysis, when performing phase-resolved spectral analysis simultaneously for NuSTAR and Swift . Fluxes are unabsorbed fluxes measured using XSPEC's cflux model.
:1;/<=(=+>?(307@4
A@<BC+*!
D(E
*
Fig. 7.- Phase-averaged spectra of Swift data (green) and all four NuSTAR observations with the best-fit model (see Table 4). The additive components are included in the plot. The model shown is an absorbed blackbody plus a double power-law.
## 3.2.2. Phase-Resolved and Pulsed Spectral Analysis
In this Section we present results from a phase-resolved spectral analysis based on the NuSTAR and Swift data to study the double-peak structure of the pulse profiles shown in Figure 3. The phase intervals containing peak A and peak B were chosen to be 0 80 . -1 16 . and 1 16 . -1 52, respectively. . The off-peak spectrum covers the range of phase 0 52 . -0 80. . Before fitting the spectra we used grppha to regroup the data such that there are at least 50 counts per spectral bin. We also use the same cross-normalization factors as obtained for the phase-averaged spectroscopy ( NuSTAR : 1.0 for FPMA, 1.06 for FPMB; Swift : 0.94) and froze them for this fit along with the N H value obtained above for the BB+PL model ( N H = 1 10(6) . × 10 22 cm -2 ). To compare the spectra of the three phase intervals, we fit
them with an absorbed blackbody plus a broken power-law model or a blackbody plus a double power-law. All fit results are included in Table 4. Some of the spectral parameters differ slightly for the three chosen phase ranges: the photon indices are slightly harder for peak B than they are for peak A and the softest power-law is obtained for the off-peak spectra. The blackbody temperatures for all three regions are consistent and the break point for the energy of the broken power-law fit is located around 10 keV.
We also investigated the pulsed spectrum by subtracting the unpulsed emission from the total emission, but the statistics are too low to extract conclusive results. We cannot rule out a simple power-law or confirm a hardening at high energies (above ∼ 15 keV) as suggested by Kuiper et al. (2006).
## 3.2.3. Spectral Fits with the Coronal Outflow Model
Similar to what was done by An et al. (2013) for NuSTAR observations of 1E 1841 -045, we apply the coronal outflow model of Beloborodov (2013a) to the observations of 1E 2259+586. In this model the hard X-ray component is produced by a decelerating e ± outflow in a twisted magnetic loop (the 'j-bundle'). The relativistic e ± flow is produced near the star with a Lorentz factor γ /greaterorsimilar 10 3 via continual discharge. As the outflow expands and fills the j-bundle, it experiences a radiative drag (deceleration) due to resonant scattering of thermal photons (with energies in the keV range) around the neutron star. Beloborodov (2013a,b) showed that the Lorentz factor of the flow decreases proportionally to the local magnetic field
$$\gamma \sim 1 0 0 \frac { B } { B _ { Q } },$$
where B Q = m c / 2 e 3 /planckover2pi1 e /similarequal 4 44 . × 10 13 G. Close to the star, where B /greaterorsimilar 10 13 G, the energy of scattered photons,
$$E _ { \text{sc} } \sim \gamma ( B / B _ { Q } ) m _ { e } c ^ { 2 } \sim 5 0 ( B / B _ { Q } ) ^ { 2 } \text{ MeV} \sim 5 \gamma ^ { 2 } \text{ keV} \,,$$
is large enough to immediately convert to e ± pairs. At larger radii, where B /lessorsimilar 10 13 G, the scattered photons escape, and the flow radiates almost all its kinetic energy before it reaches the magnetic equator, where the e ± pairs annihilate. The predicted hard X-ray spectrum has an average photon index of -1 5 (this follows from Equations 8 and 9) and . cuts off in the MeV band. The hard X-rays are beamed along the magnetic field lines, and the observed spectrum varies greatly depending on the line of sight (see Figure 7 in Beloborodov 2013b).
To fit the NuSTAR data, we make the simple assumption of an axisymetric j-bundle. The poloidal magnetic field is assumed to be close to the dipole configuration, and the magnetic dipole moment is fixed to the value inferred from spindown, µ sd /similarequal 5 9 . × 10 31 G. The model is described by 5 parameters: (1) the angular position θ j (magnetic colatitude) of the j-bundle footprint, (2) the angular width ∆ θ j of the j-bundle footprint, (3) the power L of the e ± outflow along the j-bundle, (4) the angle α mag between the rotation axis and the magnetic axis, and (5) the angle β obs between the rotation axis and the observer's line of sight. In addition, the reference point of the rotational phase, φ 0 , must be introduced as a free parameter when fitting the phase-resolved spectra. Note that ∆ θ j = θ j would describe a polar cap. In this paper, we allow the j-bundle to have a ring-shaped footprint, which corresponds to ∆ θ j < θ j .
To test observations against the model, we follow the two-step method proposed by Hasco¨ et et al. (2014). We first explore the whole parameter space by fitting the phase-averaged spectrum of the total (pulsed+unpulsed) emission together with three phase-resolved spectra of the pulsed emission. We only consider data above 16 keV where
the observed hard component starts to dominate over the soft component as apparent from
Figure 7. The results are shown in Figure 8. We find that the model can fit the data. However, the fit does not give strong constraints on the parameters of the model; a large region in the α mag -β obs plane is allowed. The only significant constraint derived from this analysis is that the polar-cap shape of the j-bundle footprint does not give a good fit; we find that a ring-shaped footprint is statistically preferred with 0 4 . /lessorsimilar θ j /lessorsimilar 0 75 and . ∆ θ /θ j j /lessorsimilar 0 2 (at the 1 . σ level).
Fig. 8.- Map of p-values for the fit of the hard X-ray component with the coronal outflow model. The p-values are shown in the plane of ( α mag , β obs ) and maximized over the other parameters. The p-value scale is shown on the right. While the hatched green region has p-values smaller than 0 001, the white region has p-values greater than 0 1. . . Low p-values show parameters which are statistically disfavored (or excluded). Interchanging the values of α mag and β obs does not change the model spectrum, as long as the j-bundle is assumed to be axisymmetric. Therefore, the map of p-values is symmetric about the line of β obs = α mag .
At the second step, we freeze the best-fit parameters for the outflow model and fit the phase-averaged spectrum in the 0 5 . -79 keV range using both NuSTAR and Swift data.
Table 5: Parameters of the best-fit model for the soft X-ray component ( NuSTAR and Swift phase-averaged spectra).
| Model | N H | kT 1 | kT 2 | E tail | L 1 a | L 2 a | Γ | νL ν a | χ 2 /dof | p-value |
|--------------------|---------------|----------|---------|-----------|---------|----------|---------|----------|------------|--------------|
| | (10 22 cm - 2 | (keV) | (keV) | (keV) | | | | @ 1 keV | | |
| BB+BB | 0.44(2) | 0.430(4) | 1.22(2) | · · · | 1.7(2) | 0.073(3) | · · · | · · · | 632/497 3 | . 8 × 10 - 5 |
| BB+PL | 1.17(4) | 0.43(1) | · · · | · · · | 0.6(1) | · · · | 4.28(5) | 3.3(3) | 526/497 | 0.18 |
| BB tail | 0.52(2) | 0.401(6) | · · · | 2 . 59(5) | 1.13(9) | · · · | · · · | · · · | 539/498 | 0.098 |
| BB 1 + BB tail , 2 | 0.62(4) | 0.33(3) | 0.65(9) | 4 . 3(6) | 1.1(1) | 0.3(2) | · · · | · · · | 522/496 | 0.20 |
a In units of 10 35 ( D/D 0 ) 2 erg s -1 , where D 0 = 4 0 . kpc is the distance inferred for 1E 2259+586 (Tian et al. 2010).
We test different models for the soft component, taking into account the extension of the outflow contribution to low energies. Among these models BB tail is a phenomenological modification of the blackbody (BB), where the Wien tail is replaced by a power-law smoothly connected to the Planck distribution (see Hasco¨t et al. e 2014). The energy E tail , at which the power-law starts, is a free parameter. All results are summarized in Table 5.
## 4. Discussion
In this paper we report on observations of 1E 2259+586 in X-rays obtained with the NuSTAR and Swift satellites. We observe a double-peak structure above 24 keV for the first time at a statistical significance larger than 3 σ . Furthermore we find a hint of a phase shift towards lower phase values for one of the peaks in the pulse profile. We also find the pulsed fraction of 1E 2259+586 increases strongly with energy. To describe the phase-averaged total spectrum of the magnetar we use an absorbed blackbody plus a broken power-law.
Furthermore, we test the coronal outflow model of Beloborodov (2013a) using the observed phase-averaged and phase-resolved spectra.
## 4.1. Pulse Profile
The NuSTAR pulse profiles of 1E 2259+586 generally agree with those presented previously in Kuiper et al. (2006) based on RXTE data. A change in peak predominance occurs with increasing energy at the boundary between the high-energy end of the soft component (where thermal emission from the star is the dominant source) and the beginning of the hard component (where the e ± outflow becomes the dominant source).
The two pulses of the double-peak profile seen in 1E 2259+586 are separated by half a period and have similar peak fluxes. This might be an indication that the neutron star is seen from the same magnetic latitude 9 (in absolute value) every half period (assuming that the emission pattern around the neutron star is relatively axisymmetric). This is possible if (and only if) α mag ≈ π/ 2 (corresponding to a nearly orthogonal rotator) or β obs ≈ π/ 2. Note that no specific line of sight is a priori favored and β obs ≈ π/ 2 is the most probable observer angle. We also note that the center of pulse B shifts from 1 38 to 1 35(2) for . . energies above 8 keV, while the initially dominant peak A remains centered at the same phase independent of energy. The phase shift of peak B is, however, not significant with current statistics, and more data are needed to confirm or rule out the observed trend. The e ± outflow and its emission are probably not perfectly axisymetric, which can easily result in an energy-dependent phase shift of the peak maxima. Modeling such a minor shift with
/negationslash
9 While the rotation axis of the neutron star is fixed, the orientation of the magnetic axis relative to the observer changes with phase if α mag = 0. As a result the magnetic latitude of the observer line of sight also changes with phase.
a small number of parameters is difficult and would require a different, more complicated fitting procedure.
Both the area and the peak-to-peak pulsed fraction as well as PF RMS show an increase with energy. The area pulsed fraction is about 60% for energies around 10 keV and rises to close to 100% at 20 keV, PF RMS is around 30% and 70% at these energies, respectively. The Swift and NuSTAR values for area and peak-to-peak pulsed fraction are consistent with those reported by Patel et al. (2001) as peak-to-peak pulsed fraction for E = 0 5 . -7 0 keV of . 35 8% . ± 1 4%. The use of different analysis techniques precludes a direct comparison of the . pulsed fractions we determined and those established in some previous publication (spectral pulsed fraction of /greaterorsimilar 43% for energies below 10 keV as reported by Kuiper et al. 2006). We can compare, however, the general behavior for the pulsed fraction versus energy observed for 1E 2259+586 with that for other magnetars. Kuiper et al. (2006) reported an increase of pulsed fraction with energy for 1E 1841 -045, 4U 0142+61 and 1RXS J1708 -4009. An et al. (2013) observed the same tendency for 1E 1841 -045, although the trend was not quite as pronounced (24% ± 4% RMS pulsed fraction at 20 keV and 41% ± 18% at 80 keV). For 1E 2259+586 we clearly see a dramatic increase in pulsed fraction following the general trend observed for other magnetars. Such an increase in the pulsed fraction is naturally expected in the coronal outflow model: photons of higher energy are more strongly beamed along the magnetic field lines as they are produced where the flow moves with a higher Lorentz factor (see Equation 9).
## 4.2. Spectrum
We found that the spectral parameters of Swift and NuSTAR for the phase-averaged spectrum (see Section 3.2.1) agree well with those reported by Patel et al. (2001) and Zhu et al. (2008). These results support the fact that especially the soft-band spectrum
Fig. 9.- Spectral turnover (Γ s -Γ ) versus magnetic field h B (left) and spin-down ˙ ν (right) from Kaspi & Boydstun (2010) for all magnetars presented there as well as our new result. Blue circles represent total flux, green triangles represent pulsed flux. Our total flux result is indicated by the red star. 1 σ uncertainties are shown. For details on full list of magnetar names, included data and references, see Kaspi & Boydstun (2010).
(below ∼ 10 keV) has been stable over a period of ∼ 13 yrs from Chandra observations in 2000 (Patel et al. 2001) to 2013 despite an anti-glitch (Archibald et al. 2013) and glitches (Kaspi et al. 2003; Icdem et al. 2012), that temporarily altered the spectral parameters.
With the new data we were able to test if an additional power-law component is required and find that it is. Therefore 1E 2259+586 exhibits the general tendency for magnetars to get harder at X-ray energies above 10 keV. We parametrized the NuSTAR plus Swift spectrum with a blackbody plus double power-law model as well as a blackbody plus a broken power-law. This phenomenological parameterization is a convenient way to characterize the presence of a hard X-ray component; the physical model of a coronal outflow is discussed in Section 4.3 below.
We note that the results of our analysis support the anticorrelation of spectral turnover (Γ s -Γ ) with magnetic field h B and spin-down ˙ ν suggested by Kaspi & Boydstun (2010). The authors demonstrated that there is a trend suggesting that those magnetars with the highest (lowest) B or ˙ ν have the smallest (largest) spectral turnover. In Figure 9, we summarize the results of Kaspi & Boydstun (2010) for pulsed (green triangles) and total fluxes (blue circles) and add our new findings for 1E 2259+586 (red star) which further strengthens their case. Errors shown are 1 σ uncertainties. It is interesting to point out that the observed spectral turnover for energetic rotation-powered pulsars (RPPs) is practically zero despite their much lower magnetic fields. Therefore Kaspi & Boydstun (2010) concluded that the production mechanism for X-rays above 10 keV in RPPs differs significantly from the one in magnetars. We also find a value for the hardness ratio, i.e. F /F h s (flux ratio of hard to soft spectral component in the 2 -79 keV band), that agrees well with the observed correlation of hardness ratio and characteristic age that has been inferred from the spin-down rate, as reported by Enoto et al. (2010). They found a decrease in hardness ratio with increasing characteristic age or, likewise, an increase of F /F h s with magnetic field B .
Furthermore, we investigate the phase-resolved spectra for NuSTAR and Swift and find that the PL indices were slightly harder for peak B than for peak A, while the breaking point for the energy is located around 10 keV for both pulses. We also examine pulsed spectra extracted from the NuSTAR observation, but due to low statistics we are not able to rule out a single power-law. More high-energy data are needed in order to compare the results to those of Kuiper et al. (2006).
We comment further that our analysis did not find any spectral lines or absorption features in the phase-averaged and phase-resolved spectra that could be interpreted as cyclotron lines (Tiengo et al. 2013). We also searched for short bursts in the data, but in
contrast to 1E 1048.1 -5937, for which recently six bright X-ray bursts have been reported in NuSTAR data (An et al. 2014), we did not observe a similar behavior for 1E 2259+586.
## 4.3. Spectral Modeling with the Coronal Outflow Model
We find that the phase-resolved spectra of 1E 2259+586 are consistent with the model of Beloborodov (2013a), although the available data do not allow us to derive strong constraints on the parameters of the model. In contrast to 1E 1841 -045 (An et al. 2013), 4U 0142+61, and 1RXS J1708 -4009 (Hasco¨ et et al. 2014), a broad range of α mag and β obs is so far allowed for 1E 2259+586. The reason for this degeneracy is twofold: (i) the statistics of NuSTAR data of 1E 2259+586 are not as good as for the other objects; (ii) the magnetic dipole moment of 1E 2259+586 is rather low, one order of magnitude smaller than for 1E 1841 -045. As a result, the boundary of the radiative zone (where B ∼ 10 13 G) is closer to the star. This leads to a larger allowed range of the j-bundle latitudes and to more flexible predictions of the model. Better statistics (with longer exposure times), extension of observations to the MeV range (where the emission of the e ± outflow peaks), or polarization measurements would help to break the degeneracy of α mag and β obs .
The footprint location of the j-bundle is better constrained, with 0 4 . /lessorsimilar θ j /lessorsimilar 0 75 and . ∆ θ /θ j j /lessorsimilar 0 2 (at the 1 . σ level). This means that a ring-like footprint located at rather high colatitudes 10 is statistically preferred to a polar cap footprint covering low colatitudes. This contrasts with the results obtained for 1E 1841 -045 (An et al. 2013), 4U 0142+61, and 1RXS J1708 -4009 (Hasco¨ et et al. 2014), where a polar cap footprint provided a good fit. If confirmed with future, higher-statistics data, it would point to diversity in the crustal motions responsible for the twist of the magnetosphere.
10 Complementary angle of the latitude, i.e. difference between 90 ◦ and latitude.
We also analyzed the soft component, taking into account the extension of the outflow emission to low energies. The results are not very sensitive to the choice of parameters for the coronal outflow, because all good fits have similar extensions to low energies, which cut off below ∼ 5 keV. We find that the models BB+PL and BB+BB tail provide equally good fits to the soft component; the fit with a modified blackbody (BB tail ) is only slightly worse but still acceptable, while the two-blackbody model (BB+BB) is statistically unacceptable. Even though BB+PL provides a good fit, it is physically problematic because the power-law component is brighter than the blackbody (at all frequencies) and its energy content diverges at the low energy end of the spectrum (in case of no cut off). It likely overpredicts the (sub) keV flux and the hydrogen column density N H . Our physically preferred model is BB+BB tail , where the cold blackbody corresponds to emission from most of the neutron star surface and the hot modified blackbody to a hot spot on the star (Hasco¨t et al. e 2014). Such a spot may be expected at the footprint of the j-bundle, as some particles produced in the e ± discharge bombard the footprint and heat it (Beloborodov 2009). The area of the cold and hot thermal components are A ≈ 1 0 7 . A NS and A ≈ 2 0 02 . A NS , respectively, where A NS represents the surface area of a neutron star with a typical radius of R NS = 10 km. These results are similar to those obtained for 1E 1841 -045, 4U 0142+61, and 1RXS J1708 -4009 (Hasco¨ et et al. 2014). Note however that 'BB tail only' also provides an acceptable fit for the soft component in 1E 2259+586, and the parameters of the hot spot in this source are not strongly constrained from these data.
## 5. Conclusion
We present an analysis of simultaneous NuSTAR and Swift observations of the AXP 1E 2259+586, and report on our spectral and temporal results. We find that the double-peak pulse profile in different energy bands generally agrees with previous RXTE
results, indicating a gradual change in peak dominance with energy. We used Fourier analysis techniques to quantify the similarities and differences and also show for the first time pulsations in the energy band above 24 keV for non-background-subtracted NuSTAR data. There is a hint at a change of position in phase for one of the two peaks in the pulse profile with increasing energy, but more data are needed to confirm this observation. This might suggest the pulsar is not a nearly orthogonal rotator. We also present the area and peak-to-peak pulsed fractions along with the RMS variation, showing that they all increase with energy and reach 96% ± 14% ( PF area ) and 71% ± 15% ( PF RMS ) at around 20 keV, similar to what is seen in other magnetars (Kuiper et al. 2006). The phase-averaged spectral analysis for the magnetar finds that there is good agreement between previous Chandra and RXTE data and the new observations. We also report that a model with an addition PL component (BB+2PL or BB+broken PL) is statistically preferred for the pulsar spectrum including the hard X-ray NuSTAR data. Our result for parametrization of the spectral turnover supports the anticorrelation hypothesis of Kaspi & Boydstun (2010) and we obtain a value of Γ s -Γ h = 4 0(1) for . B = 0 59 . × 10 14 G. Our findings also agree with the observed correlation of hardness ratio and characteristic age as reported by Enoto et al. (2010). The observed phase-resolved spectra are consistent with the coronal outflow model of Beloborodov (2013a). The fits of the spectra by the model prefer a ring-like j-bundle, in contrast to 1E 1841 -045, 4U 0142+61, and 1RXS J1708 -4009 (Hasco¨ et et al. 2014). Unfortunately the available data do not provide significant constraints on the magnetic inclination ( α mag ) or the observer line of sight ( β obs ).
Part of this work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344 with support from the LDRD program through grant 13-ERD-033. This work was supported under NASA Contract No. NNG08FD60C, and made use of data from the NuSTAR mission,
a project led by the California Institute of Technology, managed by the Jet Propulsion Laboratory, and funded by the National Aeronautics and Space Administration. We thank the NuSTAR Operations, Software and Calibration teams for support with the execution and analysis of these observations. This research has made use of the NuSTAR Data Analysis Software (NuSTARDAS) jointly developed by the ASI Science Data Center (ASDC, Italy) and the California Institute of Technology (USA). VMK receives support from an NSERC Discovery Grant and Accelerator Supplement, from the Centre de Recherche en Astrophysique du Qu´ ebec, an R. Howard Webster Foundation Fellowship from the Canadian Institute for Advanced Study, the Canada Research Chairs Program and the Lorne Trottier Chair in Astrophysics and Cosmology. AMB is supported by the NASA ATP grant NNX 13AI34G. This work made use of data supplied by the UK Swift Science Data Centre at the University of Leicester. We also thank Dr. A. M. Archibald for helpful discussions.
## REFERENCES
An, H., Hasco¨ et, R., Kaspi, V. M. et al. 2013, ApJ, 779, 163.
An, H., et al. 2014, ApJ, submitted.
Archibald, A. M., Dib, R., Kaspi, V. M., 2007, CASCA 2007, http: \\ www.physics.mcgill.ca/ ∼ aarchiba/pulsed-flux-CASCA-07.pdf.
Archibald, A. M. et al. 2014, publication in preparation.
Archibald, R. F., Kaspi, V. M., Ng, C.-Y et al. 2013, Nature, 497, 591.
Beloborodov, A. M. 2009, ApJ, 703, 1044.
Beloborodov , A. M. 2013a, ApJ, 762, 13.
Beloborodov , A. M. 2013b, ApJ, 777, 114.
Burrows, D. N., Hill, J. E., Nousek, J. A. et al. 2005, Space Science Reviews, 120, 165.
Capalbi, M., Perri, M., Saija, B., Tamburelli, F., Angelini, L. 2005, Technical Report 1.2, The Swift XRT Data Reduction Guide, http://swift.gsfc.nasa.gov/analysis/xrt\_swguide\_v1\_2.pdf .
De Jager, O. C., Swanepoel, J. W. H., & Raubenheimer, B. C. 1989, A&A, 221, 180.
Den Hartog, P. R., Hermsen, W., Kuiper, L. et al. 2006, A&A, 451, 587.
Dib, R., Kaspi, V. M., 2014, ApJ, 784, 37 [arXiv:1401.3085].
Durant, M. & van Kerkwijk, M. H. 2006, ApJ, 650, 1082.
Enoto, T., Nakazawa, K., Makishima, K. et al. 2010, ApJ, 722, L162.
Fahlman, G. G. & Gregory, P. C. 1981, Nature, 293, 202.
Fahlman, G. G. & Gregory, P. C. 1983, in IAU Symposium, Vol. 101, Supernova Remnants and their X-ray Emission, ed. J. Danziger & P. Gorenstein, 445.
Gavriil, F. P., Kaspi, V. M., & Woods, P. M. 2002, Nature, 419, 142.
Gehrels, N., Chincarini, G., Giommi, P. et al. 2004, ApJ, 611, 1005.
Gonzalez, M. E, Dib, R., Kaspi, V. M. et al. 2010, ApJ, 716, 1345.
Hailey, C. J., An, H., Blaedel, K. L. et al. 2010, Proc. SPIE, 7732, 77320T.
Harrison, F. A., Cook, W. R., Miyasaka, H. et al. 2010, in Semiconductor Radiation Detection Systems (Boca Raton, FL: CRC Press), 67.
Harrison, F. A., Craig, W. W., Christensen, F. E. et al. 2013, ApJ, 770, 103.
Hasco¨ et, R., Beloborodov, A. M., den Hartog, P. R. 2014, ApJL, submitted, [arXiv:1401.3406].
Icdem, B., Baykal, A., & Inam, S. C. 2012, MNRAS, 419, 3109.
Kaspi, V. M., Chakrabarty, D., Steinberger, J. 1999, ApJ, 525, L33.
- Kaspi, V. M., Gavriil, F. P., Woods, P. M. et al. 2003, ApJ, 588, L93.
Kaspi, V. M., Boydstun, K. 2010, ApJ, 710, L115.
Kuiper, L., Hermsen, W., den Hartog, P. R., & Collmar, W. 2006, ApJ, 645, 556.
Olausen, S. A. and Kaspi, V. M. 2014, ApJS, submitted [arXiv:1309.4167].
Patel, S. K., Kouveliotou, C., Woods, P. M. et al. 2001, ApJ, 563, L45.
Sasaki, M.,Plucinsky, P. P., Gaetz, T. J. et al. 2004, ApJ, 617, 322.
Thompson, C. & Duncan, R. C. 1993, ApJ, 408, 194.
```
-
<?Y="
: | + \
|
+
-
<!Y=" |
|
```
This manuscript was prepared with the AAS LT E X macros v5.2. A
|
10.1088/0004-637X/789/1/75
|
[
"Julia K. Vogel",
"Romain Hascoet",
"Victoria M. Kaspi",
"Hongjun An",
"Robert Archibald",
"Andrei M. Beloborodov",
"Steven E. Boggs",
"Finn E. Christensen",
"William W. Craig",
"Eric V. Gotthelf",
"Brian W. Grefenstette",
"Charles J. Hailey",
"Fiona A. Harrison",
"Jamie A. Kennea",
"Kristin K. Madsen",
"Michael J. Pivovaroff",
"Daniel Stern",
"William W. Zhang"
] |
2014-08-04T19:03:11+00:00
|
2014-08-04T19:03:11+00:00
|
[
"astro-ph.HE"
] |
NuSTAR Observations of the Magnetar 1E 2259+586
|
We report on new broad band spectral and temporal observations of the
magnetar 1E 2259+586, which is located in the supernova remnant CTB 109. Our
data were obtained simultaneously with the Nuclear Spectroscopic Telescope
Array (NuSTAR) and Swift, and cover the energy range from 0.5-79 keV. We
present pulse profiles in various energy bands and compare them to previous
RXTE results. The NuSTAR data show pulsations above 20 keV for the first time
and we report evidence that one of the pulses in the double-peaked pulse
profile shifts position with energy. The pulsed fraction of the magnetar is
shown to increase strongly with energy. Our spectral analysis reveals that the
soft X-ray spectrum is well characterized by an absorbed double-blackbody or
blackbody plus power-law model in agreement with previous reports. Our new hard
X-ray data, however, suggests that an additional component, such as a
power-law, is needed to describe the NuSTAR and Swift spectrum. We also fit the
data with the recently developed coronal outflow model by Beloborodov for hard
X-ray emission from magnetars. The outflow from a ring on the magnetar surface
is statistically preferred over outflow from a polar cap.
|
1408.0769v2
|
## Spiral spin textures of bosonic Mott insulator with SU(3) spin-orbit coupling
Tobias Graß , Ravindra W. Chhajlany 1 1 2 , , Christine A. Muschik 1 3 , , and Maciej Lewenstein 1 4 , 1 ICFO-Institut de Ci`ncies Fot`niques, e o Av. C.F. Gauss 3, 08860 Barcelona, Spain 2 Faculty of Physics, Adam Mickiewicz University, Umultowska 85, 61-614 Pozna´, Poland n 3 Institute for Quantum Optics and Quantum Information,
Austrian Academy of Sciences, 6020 Innsbruck, Austria and
4 ICREA-Instituci´ Catalana de Recerca i Estudis Avan¸ cats, Lluis Campanys 23, 08010 Barcelona, Spain o
We study the Mott phase of three-component bosons, with one particle per site, in an optical lattice by mapping it onto an SU(3) spin model. In the simplest case of full SU(3) symmetry, one obtains a ferromagnetic Heisenberg model. Introducing an SU(3) analog of spin-orbit coupling, additional spin-spin interactions are generated. We first consider the scenario of spin-dependent hopping phases, leading to Dzyaloshinskii-Moriya-type interactions. They result in the formation of spiral spin textures, which in one dimension can be understood by a local unitary transformation. Applying classical Monte Carlo simulations, we extend our study to two-dimensional systems, and systems with 'true' spin-orbit coupling, involving spin-changing hoppings.
## I. INTRODUCTION
Quantum many-body physics is to a large extent the physics of interacting spins. Literally, the spin is an intrinsic property of particles, formally described by a representation of the SU(2) group. For example, the fundamental representation of SU(2), expressed in terms of the Pauli matrices, describes the 1/2-spin, as present, for instance, in electrons. It requires only a small amount of abstraction to extend the meaning of spin to an arbitrary internal degree of freedom carried by the particles. Such an extended spin picture is very common in quantum engineering, where it may denote the different states in which cold atoms or trapped ions are prepared. While these states may in principal belong to one hyperfine manifold of the atoms, such that the true atomic spin would distinguish between the states, this is not necessarily the case. Then the different states form a so-called pseudospin manifold, which, in contrast to real spin, is independent from the quantum-statistical properties of the particles 1 . In an even wider sense, one might also map external degrees of freedom onto a spin system. For instance, empty or occupied sites on a lattice could be seen as a spin-1/2 pointing up or down. This is what occurs in the hard boson limit of the Bose Hubbard model (cf. Ref. 2). The most common way to derive spin models from Hubbard models is to consider the limit of strong interactions, and derive an effective Hamiltonian, describing super-exchange interactions 3,4 .
terms of a single SU(2) spin matrix. As SU(2) has only three generators, there can only be one spin-raising operator, say S + , which raises state |-〉 to | 0 , 〉 and | 0 〉 to | + , but a direct link from 〉 |-〉 to | + is missing. To con-〉 nect these two states, one has to apply twice the raising operator S + . It may then be more convenient to turn to a new spin, or better 'isospin', picture, in which all operators acting on the degree of freedom span an SU(3) group. Since this group provides three different raising operators, any unitary operation on the three-level Hilbert space can then be expressed by a single SU(3) spin matrix, see Fig. 1b. This demonstrates that SU(3) spins may play an important role in systems with threelevel constituents ranging from high-energy systems of elementary particles with their intrinsic three-fold degrees of freedom like flavor or color 5 , to condensed matter systems with a three-dimensional pseudospin manifold, or quantum optical systems of three level atoms . 6 In these contexts SU(3) spin models and their ground states have recently been investigated 7-9 . Quantum-chaotic behavior in a SU(3) spin system built of trapped ions has been studied in Ref. 10.
What is common to these different spin pictures, is its formal aspect: the operators describing the degree of freedom under scrutiny belong to some representation of the SU(2) group. In particular, for the case of pseudospin, one typically has a two-level system. In this case, indeed any local unitary operation on the spin degree of freedom can be expressed through a single Pauli matrix. However, if the (pseudo)spin degree of freedom involves more than two states, the spin picture might become cumbersome. As illustrated in Fig. 1a for a three-level system, the transition from state | + to state 〉 |-〉 cannot be described in
In this paper, we focus on multi-component quantum gases, which have recently attracted a lot of attention in the wide field of quantum simulation and quantum control 2,11-13 . In particular, we study those models which arise from a three-fold pseudospin degree of freedom combined with variants of spin-orbit coupling. The latter is the important mechanism, which connects external and internal degrees of freedom of the particles. We stress that in Nature, spin-orbit coupling exists due to the electric charges of the particles and the magnetic moments associated with the spins. In atomic gases, spin-orbit coupling has to be engineered artificially, as has been pioneered in Ref. 14 for a Bose-Einstein condensate with pseudospin-1/2. Such spin-orbit coupling leads to a rich phase diagram 15,16 , with order-by-disorder induced by quantum fluctuations 17 . The great versatility of cold atoms allows one to study the effects of spin-orbit coupling in different physical scenarios, from Josephson physics 18 to fractional quantum Hall effect
FIG. 1: SU(2) vs. SU(3) : Transitions in a three-level system might span a SU(2) algebra with a single raising (lowering) operator S ± , as depicted in (a), but most generally span a SU(3) algebra with three pairs of raising lowering operators, T ± , U ± , and V ± .
physics 19 . There are also several proposals to engineer spin-orbit coupling for atoms in an optical lattice 20,21 , which would affect the Mott transition 22,23 , give rise to exotic superfluidity 22 , and generate a surprisingly rich variety of spin textures in the Mott phase 24-28 , mostly via superexchange terms of the Dzyaloshinskii-Moriya type 29,30 .
Formally, spin-orbit coupling is a Berry connection, which can be minimally coupled to the momentum. In the lattice, one makes the Peierls substitution
$$- t \Psi _ { i + x } ^ { \dagger } \Psi _ { i } \rightarrow - t \Psi _ { i + x } ^ { \dagger } e ^ { - i A _ { x } } \Psi _ { i },$$
that is, a particle hopping along x -direction (with tunneling amplitude t ) from site i to site ( i + ) experiences x a spin rotation described by the gauge potential A x acting on the spinor Ψ . i In the case of artificial spin-orbit coupling for pseudospin larger than 1/2, which can be engineered within a n -pod scheme 31 , this gauge potential may belong to SU( N ) rather than SU(2). An SU(3) spin-orbit coupling has been investigated in Ref. 32 in the tight-binding limit, and has been shown to produce a topologically non-trivial bandstructure.
It is the goal of this paper to extend the study of SU(3) spin-orbit coupling to the strongly interacting regime: therefore, in Sec. II, we derive effective Hamiltonians for three-component bosons in the Mott phase. After discussing the differences between spinor and pseudospin gases in the interaction term, we introduce (pseudo)spin-dependent hopping phases, and arrive at an effective Hamiltonian with Dzyaloshinskii-Moriyatype interactions. In Sec. III, we solve this Hamiltonian analytically by means of a local unitary transformation. We find spiral spin textures, with a periodicity controlled by the hopping phase. Applying exact diagonalization and classical Monte Carlo simulations, we extend our study to cases not covered by our analytical solution: chains with periodic boundary conditions, for which a devil staircase behavior is found, and the case of two- dimensional lattices. In Sec. IV, we consider systems with a more complicated spin-orbit coupling, including spin-changing terms. Finally, in Sec. V, we comment on the experimental realization of such systems, and possible ways of detecting the spiral order.
## II. HEISENBERG MODEL FOR A MOTT PHASE OF THREE-COMPONENT BOSONS
Three-level bosons have attracted a lot of attention already since the early days of Bose-Einstein condensation 33-35 , as it occurs quite frequently that bosonic atoms carry the spin one. Let us therefore start from this most familiar case of spinor gases, and view it from a generalized SU(3) point of view. This prepares for a convenient description of pseudospin gases. Finally, we will turn our attention to SU(3) models with Dzyaloshinskii-Moriya-type interactions, relevant for systems with (pseudo)spin-dependent hopping.
## A. Effective Mott Hamiltonian for spinor bosons
In spinor gases, the interaction is characterized by two paramters U 0 and U 2 . Discretized on a lattice, the interaction Hamiltonian reads:
$$\mathbb { n } _ { \text{spinor} } ^ { \text{int} } = \frac { U _ { 0 } } { 2 } \sum _ { i } n _ { i } ( n _ { i } - 1 ) + \frac { U _ { 2 } } { 2 } \sum _ { i } ( \mathbf S _ { i } ^ { 2 } - 2 n _ { i } ), \quad ( 1 ) \\ \quad \text{--} \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \ddots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots\cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdets \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots$$
where n i is the occupation number on site i , and S i the (total) spin operator acting on the particle on site i .
The kinetic energy of bosons in a standard optical lattice is well described by nearest-neighbor tunneling,
$$H _ { \text{standard} } ^ { \text{kin} } = - t \sum _ { \langle i, j \rangle } \sum _ { \sigma \in \{ +, 0, - \} } ( a _ { i \sigma } ^ { \dagger } a _ { j \sigma } + \text{H.c} ), \quad ( 2 )$$
with t the tunneling amplitude. For t /lessmuch U , U 0 2 , the system is in the Mott phase, and insight in its physics can be gained by taking the kinetic part as a second-order perturbation to the interactions. It has been shown that this leads to a biquadratic SU(2) Heisenberg model 36,37
$$\text{live} \quad \ H _ { S U ( 2 ) } = - J \sum _ { \langle i, j \rangle } \left [ \cos \theta ( \mathbf S _ { i } \cdot \mathbf S _ { j } ) + \sin \theta ( \mathbf S _ { i } \cdot \mathbf S _ { j } ) ^ { 2 } \right ], \ \ ( 3 ) \\ \text{inor} \quad \ \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots$$
with the parameters J and θ depending on t, U 0 , and U 2 . This effective Hamiltonian exhibits spin-nematic and ferromagnetic phases, a prediction which recently has been confirmed by a full quantum Monte-Carlo study of the problem 38 .
As argued in the introduction, the biquadratic SU(2) Hamiltonian (3) can be rewritten in terms of SU(3) spin matrices, which reduces it to a quadratic form. While the most general expression is provided in the appendix, a particularly simple special case is the one of spin-independent interactions, U 2 = 0, corresponding to
θ = π/ 4. The SU(3) symmetry of the system then leads to an effective SU(3) Heisenberg Hamiltonian
$$H _ { S U ( 3 ) } = - \frac { t ^ { 2 } } { 2 U _ { 0 } } \sum _ { \langle i, j \rangle } \sum _ { \nu = 1 } ^ { 8 } \lambda _ { i } ^ { ( \nu ) } \lambda _ { j } ^ { ( \nu ) } = - \frac { J } { 2 \sqrt { 2 } } \sum _ { \langle i, j \rangle } \lambda _ { i } ^ { T } \cdot \lambda _ { j },$$
where we have inserted J = -√ 2 t 2 /U 0 . The λ ( ν ) are the eight generators of SU(3), which, for convenience, have been arranged together to a eight-component vector λ = ( λ (1) , . . . , λ (8) ) T on the right side of Eq. (4). The form of Eq. (4) holds for any filling factor if the λ ( ν ) are chosen a symmetric SU(3) representation, corresponding to the desired particle number. An interesting case is the filling with three bosons per site: As follows from Young tableaux calculus, the Hilbert space of three qutrits consists of a fully symmetric decuplet, an fully antisymmetric singlet, and two octets of mixed symmetry, 3 ⊗ 3 ⊗ 3 = 10 ⊕ 8 ⊕ 8 ⊕ 1 . According to the bosonic nature of the atoms, the low-energy Hilbert space is given by the 10 representation of SU(3), which is implemented in Nature as the baryon decuplet.
In the following, however, we restrict ourselves to the simpler case of one boson per site. The low-energy Hilbert space then is three-dimensional, and one works in the fundamental representation of SU(3). Accordingly, we associate the eight Gell-Mann matrices with the λ ( ν ) , cf. Ref. 5. Note that six of these matrices are related to the raising/lowering operators of Fig. 1b: ( λ (1) ± iλ (2) ) = 2 T ± , ( λ (4) ± iλ (5) ) = 2 V ± , and ( λ (6) ± iλ (7) ) = 2 U ± . The other two matrices are diagonal: λ (3) = diag(1 , -1 0) , , and λ (8) = 1 √ 3 diag(1 1 , , -2).
## B. Effective Mott Hamiltonian for bosons with SU(3) pseudospin
The biquadratic form of Eq. (3) holds due to the invariance of H = H kin standard + H int spinor under spin rotations. If, instead of considering Bose gases with an intrinsic spin, one switches to pseudospin-1 systems, that is, a trinary mixture of bosons, this symmetry is likely to be broken. In particular, one might not expect interactions as described by Eq. (1). Neglecting spin-changing collisions, the most general two-body contact interaction would be given in terms of six interaction parameters U σσ ′ :
$$H _ { \text{pseudospin} } ^ { \text{int} } = U _ { \sigma \sigma ^ { \prime } } \sum _ { i } | \sigma \sigma ^ { \prime } \rangle _ { i } \, \langle \sigma \sigma ^ { \prime } | _ { i } \,, \quad \ ( 5 ) \quad \ \text{for t} \\ \quad \ \ \ \ A _ { y }. \, \\ \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
for pairs of particles in state | σ 〉 and | σ ′ 〉 interacting on site i . The effective Mott Hamiltonian to H = H kin standard +
H int pseudospin is of the form
$$H _ { \text{eff} } = & - \sum _ { \langle i, j \rangle } \left [ \sum _ { \nu = 1 } ^ { 8 } J _ { \nu } \lambda _ { i } ^ { ( \nu ) } \lambda _ { j } ^ { ( \nu ) } + J _ { 3 8 } ( \lambda _ { i } ^ { ( 3 ) } \lambda _ { j } ^ { ( 8 ) } + \\ & \lambda _ { i } ^ { ( 8 ) } \lambda _ { j } ^ { ( 3 ) } ) \right ] + \sum _ { i } \left [ h _ { 3 } \lambda _ { i } ^ { ( 3 ) } + h _ { 8 } \lambda _ { i } ^ { ( 8 ) } \right ]. \quad \quad ( 6 ) \\ \text{Now the interaction parameters } J _ { \nu } \text{ are spin-dependent,}$$
Now the interaction parameters J ν are spin-dependent, there is an additional interaction J 38 , and magnetic-fieldlike terms h 3 and h 8 . All parameters are given in appendix A2.
## C. Effective Mott Hamiltonian in the case of spin-dependent tunneling phases and generalized Dzyaloshinskii-Moriya interactions
We now turn our attention to systems where the standard hopping is replaced by a spin-orbit coupled kinetic term. Recently, spin-orbit coupling in two-component Bose gases has been shown to give rise to effective Dzyaloshinskii-Moriya interactions in the low energy description of the Mott insulator phase 24-27 , that is an antisymmetric spin-spin interaction of the form D · ( S i × S j ), characterized by the Dzyaloshinskii-Moriya vector D . The presence of such interaction leads to a variety of different spin textures in the ground state.
In this section, we derive an effective Hamiltonian for three-component bosons, which exhibits DzyaloshinskiiMoriya type interactions generalized to SU(3) spins. For this, it is sufficient to consider pseudospin-dependent tunneling phases. If these phases were spatially dependent, this would mimic a magnetic field acting differently on the three components, corresponding to different electric charges of the pseudospin states. Note that such a situation occurs naturally in the quark model, where the up quark has an electric charge of 2 / 3 , while down quark e and strange quark have a charge of -e/ 3.
We start our analysis by writing down the general form of the kinetic term in the presence of a gauge potential,
$$H _ { S O C } ^ { \text{kin} } = - t \sum _ { i } \sum _ { \mathbf r \in \{ \hat { e } _ { x }, \hat { e } _ { y } \} } \mathbf a _ { i + \mathbf r } ^ { \dagger } e ^ { - i \mathbf A \cdot \mathbf r } \mathbf a _ { i } + \mathbf H. c., \quad ( 7 )$$
where we have introduced a vector notation for the three bosonic components a † i = ( a † i + , a † i 0 , a † i -). Spin-orbit coupling is obtained by choosing the gauge potential to be sensitive to the (psedo)spin. We have assumed a twodimensional square lattice in the notation above. A onedimensional system is obtained by freezing out the hopping between adjacent one-dimensional chains.
Except in Sec. IV, we shall consider the simplest choice for the gauge potential corresponding to diagonal A x and A y . In the generic 2-dimensional case we choose standard gauge-free hopping in one direction, say y , i.e. A y = 1 . The non-trivial gauge choice is made in the x-direction (also chosen as the chain direction for 1-d lattices) with
A x = diag( α, β, γ ). In the effective Mott Hamiltonian, this will lead to modified spin interactions along the x -direction, as the superexchange terms gain a phase factor.
Let us, in the first place, assume the important special case of SU(3)-symmetric interactions of strength U . The hopping in x -direction then yields the effective Hamiltonian:
$$h _ { x } & = - \, \frac { t ^ { 2 } } { 2 U } \sum _ { i } \left [ \sum _ { \nu = 1 } ^ { 8 } J _ { \nu } \lambda _ { i } ^ { ( \nu ) } \lambda _ { i + x } ^ { ( \nu ) } + \quad \text{when} \\ & \quad \sum _ { \nu, \mu } V _ { \nu \mu } \left ( \lambda _ { i } ^ { ( \nu ) } \lambda _ { i + x } ^ { ( \mu ) } - \lambda _ { i } ^ { ( \mu ) } \lambda _ { i + x } ^ { ( \nu ) } \right ) \right ], \quad \text{(8)} \\ \text{with}$$
with
$$& J _ { 1 } = J _ { 2 } = \cos ( \alpha - \beta ), \ J _ { 4 } = J _ { 5 } = \cos ( \alpha - \gamma ), \quad & \text{for} \\ & J _ { 6 } = J _ { 7 } = \cos ( \beta - \gamma ), \ J _ { 3 } = J _ { 8 } = 1, \quad & ( 9 ) \quad & \text{ramet} \\ & V _ { 1 2 } = \sin ( \alpha - \beta ), \ V _ { 4 5 } = \sin ( \alpha - \gamma ), \ V _ { 6 7 } = \sin ( \beta - \gamma ). \quad & \text{Dzyal} \\ & \dots \quad. \quad. \quad. \quad. \quad. \quad 2 D \text{ sv}$$
All other V νµ are zero. Apparently, the second term in Eq. (8) describes an antisymmetric spin-spin interaction. To make the analogy to the Dzyaloshinskii-Moriya interaction as close as possible, we introduce a 'vector product' of eight-dimensional SU(3) vectors:
$$\mathbf u \times \mathbf v \equiv f ^ { i j k } u _ { i } v _ { j } \hat { e } _ { k }, \quad \quad ( 1 0 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
with f ijk the antisymmetric structure constants of SU(3), cf. Ref. 5. If sin( α -β ) = sin( α -γ ) -sin( β -γ ), the effective Hamiltonian h x can be written as
$$h _ { x } = - \frac { t ^ { 2 } } { 2 U } \sum _ { i } \left [ \sum _ { \nu = 1 } ^ { 8 } J _ { \nu } \lambda _ { i } ^ { ( \nu ) } \lambda _ { i + x } ^ { ( \nu ) } + \mathbf D \cdot ( \lambda _ { i } \times \lambda _ { i + x } ) \right ], \\ ( 1 1 ) \quad \text{ acting} \quad \text{ that}$$
with the J 's given in Eq. (9), and the non-zero entries of the Dzyaloshinskii-Moriya vector given by
$$D _ { 3 } = \sin ( \alpha - \beta ), \ D _ { 8 } = \frac { 1 } { \sqrt { 3 } } [ \sin ( \alpha - \gamma ) + \sin ( \beta - \gamma ) ]. \quad \begin{smallmatrix} \text{ that it} \\ \text{ to eac} \\ \text{ Eq. (t)} \end{smallmatrix}$$
In the discussion below, we consider Hamiltonian's of the type given in Eq. (11) determined by the parameter α , with the choice γ = α, β = 0.
As will be discussed in greater detail below, the Dzyaloshinskii-Moriya interaction does not fully lift the huge degeneracy of the ferromagnetic SU(3) Heisenberg model. One way to obtain a unique ground state is to break the SU(3) symmetry of the interactions. For instance, by strengthening interactions between identical particles, U , relative to interactions between particles of different (pseudo)spin, U inter , one might force the system into a fully unpolarized state, with equal occupation numbers in all spin states, N + = N 0 = N -. The effective spin Hamiltonian resulting from these interactions is given again by Eq. (8), but now with the parameters
$$\varnothing \cdots \varkappa \cdots \varkappa _ { J } \dashvarpi \cdots \varkappa _ { J } \dashvarpi \cdots \varkappa _ { J } \dashvarpi \cdots \varkappa _ { J } \dashvarpi \cdots \\ J _ { 1 } = J _ { 2 } = \frac { 1 } { \mu } \cos \alpha, \ J _ { 4 } = J _ { 5 } = \frac { 1 } { \mu }, \\ J _ { 6 } = J _ { 7 } = \frac { 1 } { \mu } \cos \alpha, \ J _ { 3 } = J _ { 8 } = \frac { 2 \mu - 1 } { \mu }, \quad \ \ ( 1 3 ) \\ V _ { 1 2 } = \frac { 1 } { \mu } \sin \alpha, \ V _ { 4 5 } = 0, \ V _ { 6 7 } = - \frac { 1 } { \mu } \sin \alpha, \\ \dashvarpi \cdots \varkappa \cdots = \tau \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
where µ ≡ U inter /U is the ratio between the two interaction strengths.
## III. SPIRAL ORDER PHASES
For simplicity, we consider first spin models stemming from SU(3) symmetric interactions, Eq. (8) with parameters from Eq. (9), for discussing the effect of the Dzyaloshinskii-Moriya term. We will separately study 1D systems with open and closed boundary, as well as 2D systems with closed boundary.
## A. Effect of the Dzyaloshinskii-Moriya term in 1D with open boundary
Freezing the dynamics in the y -direction, it is possible to solve the problem analytically under open boundary conditions. It is convenient to introduce a spin rotation matrix U α i [ ]:
$$U _ { i } [ \alpha ] \equiv \left ( \begin{array} { c c c } e ^ { - i \alpha } & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & e ^ { - i \alpha } \end{array} \right ), \quad \text{(14)} \\ \text{on site $i$. For a chain of $N$ spins, labeled $1,\dots,N$}$$
acting on site i . For a chain of N spins, labeled 1 , . . . , N we introduce
$$\mathbf U \equiv \otimes _ { k = 1 } ^ { N } U _ { k } [ ( N - k ) \alpha ], \text{ \quad \ \ } ( 1 5 )$$
that is, each pair of neighboring spins is rotated relative to each other by an angle α . The Hamiltonian h x from Eq. (8) can then be written as
$$h _ { x } = - \frac { t ^ { 2 } } { 2 U } \mathbf U ^ { \dagger } \left ( \sum _ { i = 0 } ^ { N - 2 } \lambda _ { i } ^ { T } \lambda _ { i + 1 } \right ) \mathbf U. \quad \quad ( 1 6 ) \\ _ { \dots \dots \dots \lambda _ { i } \dots \lambda _ { i } \dots \dots \lambda _ { i } \dots \dots \lambda _ { i } \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \end{array}$$
This means that h x can be mapped onto the ferromagnetic SU(3) Heisenberg model via a unitary transformation, and both Hamiltonians have the same energy spectrum. In particular, the ground state degeneracy of the ferromagnetic Heisenberg Hamiltonian is not lifted by the Dzyaloshinskii-Moriya term.
However this solution allows one to gain insight into the magnetic correlations in the SU(3) system by mapping the ground states of ferromagnetic model to the Dzyaloshinskii-Moriya Hamiltonian. The ground state manifold of the SU(3) symmetric Heisenberg ferromagnet
corresponds to the maximal symmetric representation of SU(3) on N sites and is therefore spanned by the symmetric Dicke states, i.e. , the symmetric superposition of states with N + particles in | + , 〉 N 0 particles in | 0 , and 〉 N -particles in |-〉 , where N + + N 0 + N -= N . Therefore, the number of degenerate ground states is given by d = ∑ N N + =0 ∑ N -N + N 0 =0 = 1 2 ( N + 1)( N + 2). In the presence of the Dzyaloshinskii-Moriya interactions, each Fock state contributing to an eigenstate gains a phase through the unitary transformation, Eq. (15).
To illustrate this phase acquirement, let us first consider a chain of N = 3 spins. This yields d = 10
ground states. Three of them are given in terms of a single Fock state, namely the states | +++ 〉 ≡ | + 〉 i =1 ⊗ | + 〉 i =2 ⊗ | + 〉 i =3 , | 000 , 〉 and |- - -〉 . These states are, apart from a irrelevant overall phase, invariant under U . The remaining seven ground states are transformed by the Dzyaloshinskii-Moriya interactions. Let us, as an important example, pick out the fully unpolarized state with one + particle, one -particle, and one 0 particle. This is the unique ground state of the isotropic ferromagnetic SU(3) model in a subspace of the Hilbert space with N + = N -= N 0 . Note that these numbers are constants of motion.
For N = 3, the fully unpolarized Dicke state reads
$$\Psi _ { + 0 - } \equiv \frac { 1 } { \sqrt { 6 } } ( | + 0 - \rangle + | + - 0 ) + | 0 + - \rangle + | 0 - + \rangle + | - + 0 ) + | - 0 + \rangle ).$$
The ground state of h x in the fully unpolarized sector reads
$$\tilde { \Psi } _ { + 0 - } = \mathbf U ^ { \dagger } \Psi _ { + 0 - } = \frac { 1 } { \sqrt { 6 } } ( e ^ { 2 i \alpha } | + 0 - ) + e ^ { 3 i \alpha } | + - 0 ) + e ^ { i \alpha } | 0 + - ) + e ^ { i \alpha } | 0 - + ) + e ^ { 3 i \alpha } | - + 0 ) + e ^ { 2 i \alpha } | - 0 + \rangle ). \quad ( 1 8 )$$
This is easily generalized to any N being a multiple of 3. We note here that breaking the SU(3) symmetry of the interactions with µ < 1, as in Eq.(13) (and α = 0), yields a unique ground state originating from the unpolarized Dicke state. The overlap of this ground state with the unpolarized Dicke state is large for the system sizes for which we have performed exact diagonalization, e.g for µ = 0 75 . , N = 9, it is approximately 0.95. The effect of the phases α = 0 can again be taken into account by the transformation Eq.(15), which at any α produces the correct ground state with a fidelity > 0 99 . for µ = 0 75 . , N = 9. Therefore the discussion below holds to a good approximation also for the case when the Dzyaloshynskii-Moriya type Hamiltonian ground state is unique.
The spin structure can thus be viewed as spirals in the λ (1) -λ (2) - and the λ (6) -λ (7) -plane, and a ferromagnet in the λ (4) -λ (5) -plane.
/negationslash
We note that spiral order has also been found in SU(2) systems with spin-orbit coupling 24-26 , but the coexistence of ferromagnetic and spiral order is a feature only found in the SU(3) system. Moreover, the phase angle α can be used to freely tune the periodicity of the spirals. In contrast to the systems in Refs. 24-26, we have not assumed off-diagonal spin-orbit coupling, easing the experimental realization of the system studied here.
It can easily be seen that due to strong entanglement in this state, the single site density matrices are all propotional to the identity and therefore the spin (traceless Gellmann matrices) averages to zero, 〈 λ ( ) i 〉 = 0, for arbitrary N . By combinatoric analysis of the Dicke state, the spin-spin correlations can be evaluated and they show an interesting behavior as a function of α :
$$\int _ { \real } \real - r \real - \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \int _ { \real } \\ \langle \lambda _ { i } ^ { ( 1 ) } \lambda _ { i + r } ^ { ( 1 ) } + \lambda _ { i } ^ { ( 2 ) } \lambda _ { i + r } ^ { ( 2 ) } \rangle & = \frac { 4 N } { 9 ( N - 1 ) }, \\ \langle \lambda _ { i } ^ { ( 6 ) } \lambda _ { i + r } ^ { ( 6 ) } + \lambda _ { i } ^ { ( 7 ) } \lambda _ { i + r } ^ { ( 7 ) } \rangle & = \frac { 4 N } { 9 ( N - 1 ) } \cos ( r \alpha ), \\ \langle \lambda _ { i } ^ { ( 3 ) } \lambda _ { i + r } ^ { ( 3 ) } \rangle & = 0, \\ \langle \lambda _ { i } ^ { ( 8 ) } \lambda _ { i + r } ^ { ( 8 ) } \rangle & = 0.$$
## B. Effect of the Dzyaloshinskii-Moriya term with periodic boundary conditions
The unitary transformation that gauges away the Dzyaloshinskii-Moriya term in the Hamiltonian in the previous section fails, in general, to achieve this completely in the case of a periodically closed boundary. Indeed, the transformation gauges away the DzyaloshinskiiMoriya term in the entire bulk of the lattice leaving only an uncompensated, frustrated boundary term. The reason for this is the phase relation between the first and the last spin demanded by the periodic boundary condition. Following the logic of the previous section, fixing the phase transformations on the first and last spins results in a phase mismatch of Nα on the bond connecting these spins.
As such the magnetic ordering is therefore expected to be qualitatively very similar to that described in the last section. This is exactly true when the unitary trans-
formation has no conflict with the periodic boundary as in the special cases when 0 = mod[ Nα, 2 π ], i.e. , for α being an integer multiple of 2 π/N . Interestingly, our results from exact diagonalization suggest that the ground state is well described by a unitary transformation for any Hamiltonian parameter α . Certainly, the input parameter to the unitary transformation have to be chosen from the discrete set ˜ = α m π/N 2 . The integer m is a kind of a 'winding number' counting the number of times a spin spiral wraps around the axis throughout the system. This transformation parameter, ˜, depends on the α Hamiltonian parameter, α , through a devil staircase-like function:
$$m = j \text{ for } j \in \mathcal { Z } \colon \frac { 2 \pi } { N } ( j - \frac { 1 } { 2 } ) \leq \alpha < \frac { 2 \pi } { N } ( j + \frac { 1 } { 2 } ). \text{ } ( 2 0 ) \text{ } \underset { \text{by sta} } { \mined }$$
We emphasize here, that this unitary transformation whilst providing a description of the ground state, for arbitrary α , does not unitarily connect between the Heisenberg Hamiltonian and the Dzyaloshinskii-Moryia-type Hamiltonians. This reflects in an energy spectrum depending on α .
Due to the discreteness of the input parameters to the unitary transformation, the finite system with periodic boundary conditions exhibits clearly separated phases, characterized by the winding number m . From this, one obtains the periodicity of the spin pattern, expressed in the size of the unit cell P , as
$$P = \frac { L } { m } p, \quad \quad \ \ ( 2 1 ) \quad \ \ i z \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
where L is the length of the system, and p is the smallest integer such taht P is integer. If m and L are coprime, the system is incommensurate, and we have P = L . It is worth to notice that for m = N/ 2, that is in the vicinity of α = π/ 2, the unit cell size is 2, that is, the spiral order becomes antiferromagnetic.
## C. Effect of the Dzyaloshinskii-Moriya term in 2D
We now turn to a two-dimensional system, under periodic boundary conditions, with spin-orbit coupling in one direction, and standard hopping in the other direction. Again, we are not able to map the problem onto the Heisenberg model by means of the unitary transformation of Eq. (15).
We obtain here the classical ground state phase of the 2-D model. In order to achieve this, the 8-dimensional SU(3) spins are treated classically by replacing them with their expectation values and finding the classical spin configuration on the lattice minimizing the energy. The on-site vector /vector λ = √ 3 2 〈 λ 〉 can be seen as a generalized Bloch vector, where the prefactor is needed to attain unit normalization. It is important to note a crucial difference w.r.t the well-known SU(2) case: A pure state of a two-level system is characterized by two angles, a mixing angle, and a relative phase, which perfectly maps onto a sphere. A pure state in a three-level system is, up to a irrelevant overall phase, of the form:
$$\underset { n \text{ for } } { \text{ound} } \quad | \Psi \rangle = e ^ { i \xi _ { 1 } } \sin \theta \cos \phi | + \rangle + e ^ { i \xi _ { 2 } } \sin \theta \sin \phi | 0 \rangle + \cos \theta | - \rangle \,, \\ \text{$t$ pa-}$$
with 0 ≤ θ, φ < π 2 , and 0 ≤ ξ 1 , ξ 2 < 2 π . This means that the state is parametrized by only four real numbers, and the eight-component vector /vector λ has to be restricted to the corresponding four-dimensional subspace.
In the classical limit, the spin-spin Hamiltonian provides an energy functional which is numerically minimized by simulated annealing. The latter is implemented by starting from an arbitrary initial configuration, which is locally updated through a Metropolis algorithm. We have repeatedly carried out the simulated annealing for a lattice with 30 × 30 spins.
For the case of SU(3)-symmetric interaction, Eq. (9), the simulated annealing always led to a ferromagnet fully polarized within the λ (3) -λ (8) plane. This can be understood by noting that on the classical level the Dzyaloshinskii-Moriya interactions do not affect the part of the system which is polarized in the λ (3) -λ (8) plane, in which the energy per spin may remains minimal by forming a ferromagnet.
To force the system into a less trivial configuration than the ferromagnetic one, we have to make the polarization in the λ (3) -λ (8) plane less favorable. This can again be achieved by tuning pseudospin-dependent interactions away from the SU(3)-symmetry point, as described in Eq. (13). For µ < 1, polarization within the λ (3) -λ (8) plane becomes energetically less favorable, and the effect of the Dzyaloshinskii-Moriya interactions is expected to be enhanced.
Accordingly, our simulated annealing lattice with pseudospin interactions characterized by µ = 0 75 yields in-. teresting spin patterns: As illustrated in Fig. 2, there is spiral order along the x -axis, and ferromagnetic order along the y -axis with the standard hopping. The spiral ordering occurs only in the 1-,2-,6-, and 7-components of /vector λ , that is, in those components directly affected by the Dzyaloshinskii-Moriya term. As expected, the 3- and 8components of /vector λ are vanishingly small, and the 4- and 5-components of /vector λ , unaffected by the DzyaloshinskiiMoriya interaction, show ferromagnetic order.
The periodicity of these textures along the x -axis depends on the hopping phase α , in exactly the same way as in the 1D system with periodic boundary conditions, cf. Eqs. (20-21). We illustrate this by plotting the winding number of the spirals and the size of the unit cell in Fig. 3 for a 30 × 30 lattice. Indeed, the spiral order found along x is precisely the classical version of the quantum solution we obtained in 1D. To see this, we evaluate spin-spin correlations in the classical solution. The site-dependent parametrization of /vector λ that is in agreement
FIG. 2: (Color online) Spin textures in the presence of spin-dependent tunneling phases : Simulating annealing results for a 30 × 30 lattice with periodic boundary for different tunneling phases α . We plot projections of the local SU(3) spin vectors λ into different spin planes. The λ (3) -and λ (8) -component is not plotted, as it is zero. Since the system behaves ferromagnetic along y -direction, we only show the behavior along x . In (a), despite non-zero α , the system remains ferromagnetic in all spin components. In (b-d), only the λ (4) -and λ (5) -component behaves ferromagnetic, while the other spin components show spiral order. In (b), we plot the case with only one spiral, while in (c) we plot the case with maximum winding number, that is, N/ 2 spirals, or antiferromagnetic order. In (d), we give an example for an incommensurate winding number, m = 7. For all cases, we plot the corresponding unit cell.
Winding number
Unit cell size
FIG. 3: (Color online) System periodicity in the presence of a spin-dependent tunneling phase α : We consider a system with 30 sites along the direction of non-trivial hopping, and plot the winding number of the spirals and the corresponding size of the unit cell as a function of α . The figure represents our MonteCarlo results for a two-dimensional system, cf. 2, but also illustrates the essence of the staircase formula, Eq. (20).
with the results of simulated annealing, is given by:
$$\vec { \lambda } _ { x, y } = \frac { 1 } { \sqrt { 3 } } \left ( \cos ( x \tilde { \alpha } ), \sin ( x \tilde { \alpha } ), 0, 1, 0, \cos ( x \tilde { \alpha } ), - \sin ( x \tilde { \alpha } ) \right ), \quad \text{$Fourth$} \\ ( 2 3 ) \quad \text{$on the$} \quad \text{$Fur$}$$
more repulsive that atoms in different pseudospin states. Furthermore, even in this case, the spiral structures could be traced back to effects of a gauge transformation acting on the (anisotropic) Heisenberg SU(3) model.
where ˜ α is an integer multiple of 2 π/L x , with L x the size of the lattice along the x -direction. The value of ˜ α depends on α according to the devil staircase function (20) described in the previous section. Using this, the correlations are:
$$\text{elations are: } & \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\text{$\lambda_{i}^{(1)} \lambda_{i+r}^{(1)} } + \lambda_{i}^{(2)} \lambda_{i+r}^{(2)} = \frac { 1 } { 3 } \cos ( r \alpha ), \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\quad \text{$\alpha_{i}^{(4)} \lambda_{i+r}^{(4)} } + \lambda_{i}^{(5)} \lambda_{i+r}^{(5)} = \frac { 1 } { 3 }, \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\text{\em img into account the normalization constant in the } \quad \text{also} \quad \text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$$\text$$\text{$\text{$\text{$\text{$\text$$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text \text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\end}$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$
\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\begin{array} { c } } } \end{array}$$
Taking into account the normalization constant in the definition of /vector λ , the spin-spin correlators in the classical configuration are identical to the quantum ones of Eq. (19) in the limit of large number of spins N .
## IV. OFF-DIAGONAL SPIN-ORBIT COUPLING
In the previous section, we discussed the simplest spinorbit coupling, consisting only of spin-dependent tunneling phases or abelian spin-orbit coupling. We found that this can give rise to interesting pseudospin textures, but it is not sufficient to fully remove the huge ground state degeneracy of the Heisenberg model. The latter was achieved by breaking the SU(3)-symmetry in the interactions, e.g. by making atoms in the same pseudospin state
Furthermore, the precise interaction Hamiltonian for spin-orbit coupled quantum gases is a delicate issue. One has to bear in mind that schemes to generate the spinorbit coupling typically dress the atoms. Such dressing might have significant consequences for the interactions, cf. Refs. 39,40. It seems, however, to be a reasonable assumption to consider spin-independent interactions. Nature provides atomic manifolds with (almost) SU( N )-symmetric interactions, and if the dressed states are composed only from states of such manifolds, interactions should remain spin-independent.
In this section, we therefore analyze a more complicated (non- abelian) spin-orbit coupling, which includes spin-changing hopping processes with the assumption of SU(3) symmetric interactions. As we shall see, such terms again lead to non-trivial spin textures, but they also remove the degeneracies of the Heisenberg model.
A truly SU(3) spin-orbit coupling has been discussed by Barnett, Boyd, and Galitski in Ref. 32. It is described by the vector potential
$$( A _ { x }, A _ { y } ) = \frac { \pi } { 3 } \left ( - \frac { 2 } { \sqrt { 3 } } ( \lambda ^ { ( 2 ) } - \lambda ^ { ( 5 ) } + \lambda ^ { ( 7 ) } ), \lambda ^ { ( 3 ) } + \sqrt { 3 } \lambda ^ { ( 8 ) } \right ).$$
With this choice, motivated by a mapping onto the SU( N ) Hofstadter model, the non-interacting system has been shown to exhibit a topological non-trivial bandstructure.
We now derive the effective Hamiltonian in the strongly interacting limit corresponding to the spin-orbit coupling in Eq.(25). We first note that along the y -direction the spin-orbit coupling remains diagonal, and
FIG. 4: (Color online) Spin textures in the presence of off-diagonal spin-orbit coupling : The plots show the projection of the local SU(3) spin vectors λ into different spin planes, for a Mott system with the spin-orbit coupling of Eq. (25). We have studied a lattice of 30 sites in each direction, but here we only plot the unit cell. In (a), we restrict the dynamics to the x -direction, while (b) shows the results for the two-dimensional system.
Hopping from the left to the right therefore promotes a particle counterclockwise along the triangle of Fig. 1b: Upon hopping to the right, a | + 〉 particle becomes | 0 , 〉 a | 0 〉 becomes |-〉 , and a |-〉 becomes | + . 〉 This makes it straightforward to write down the hopping in terms of raising/lowering operators, and one obtains a relatively compact second-order Hamiltonian:
$$h _ { x } = & - \frac { t ^ { 2 } } { U } \left \{ \lambda _ { i } ^ { ( 4 ) } \lambda _ { i + 1 } ^ { ( 6 ) } - \lambda _ { i } ^ { ( 5 ) } \lambda _ { i + 1 } ^ { ( 7 ) } - \lambda _ { i } ^ { ( 1 ) } \lambda _ { i + 1 } ^ { ( 4 ) } + \lambda _ { i } ^ { ( 2 ) } \lambda _ { i + 1 } ^ { ( 5 ) } - \lambda _ { i } ^ { ( 6 ) } \lambda _ { i + 1 } ^ { ( 1 ) } - \lambda _ { i } ^ { ( 7 ) } \lambda _ { i + 1 } ^ { ( 2 ) } \\ & - \frac { 1 } { 2 } \left [ \lambda _ { i } ^ { ( 3 ) } \lambda _ { i + 1 } ^ { ( 3 ) } + \lambda _ { i } ^ { ( 8 ) } \lambda _ { i + 1 } ^ { ( 8 ) } + \sqrt { 3 } \left ( \lambda _ { i } ^ { ( 3 ) } \lambda _ { i + 1 } ^ { ( 8 ) } - \lambda _ { i } ^ { ( 8 ) } \lambda _ { i + 1 } ^ { ( 3 ) } \right ) \right ] \right \}. \\ & \coprod$$
As in the previous section, the simulated annealing algorithm has been used to grasp the spin textures in the ground state of this Hamiltonian. First, we freeze the hopping in the y -direction, and study h x alone. The spins now avoid the λ (3) -λ (8) plane due to the spin-orbit coupling. In the remaining planes, a three-periodic pattern is exhibited, see Fig. 4(a). Different from the case of diagonal spin-orbit coupling, in none of the planes does the system exhibit ferromagnetic alignment.
the antiferromagnetic correlations due to the superexchange have been observed 52 .
If, additionally to the spin-orbit coupling in x -direction, one also assumes the diagonal spin-orbit term along the y -direction, according to Eq. (25), the two terms together gives rise to periodic spin patterns in both direction. As shown in Fig. 4(b), the unit cell of the spin lattice then becomes two-dimensional, containing 3 × 3 sites. These ground state patterns have been consistently obtained for various system sizes under open boundary conditions, as well as for periodic boundary conditions and linear dimensions divisible by 3. Incommensurate textures were obtaind in the case of periodic boundaries and system sizes not being divisible by 3. We therefore expect the 3x3 unit cell to survive in the thermodynamic limit.
## V. EXPERIMENTAL REALIZATION
In this section, we comment on the realization of the scenarios discussed in this paper, and discuss the different requirements.
- a. Three-component bosons. There is a variety of bosonic atoms providing three or more internal spin degrees of freedom 41 . In particular, cold atom experiments frequently work with 87 Rb in the F = 1 hyperfine manifold, cf. Ref.42 for an experiment in an optical lattice. Such species allow for the realization of SU(3) BoseHubbard like systems.
b. Atomic Mott phase. Next, the experimental setup has to realize the strongly interacting limit, where the particle fluctuations are strongly suppressed as described theoretically e.g. in Refs. 2,36,43-45. In this limit, superexchange interactions may lift the huge degeneracy between different spin states, but the corresponding energy scale, given by t 2 /U , is necessarily very small, since t /lessmuch U . The stability of the spin textures therefore requires an immense cooling effort. Advances in cooling technologies have, during the past years, allowed for experimentally demonstrating superexchange interactions 46-51 . In a recent experiment with fermions, c. Spin-orbit coupling. In recent years, a variety of proposals for engineering spin-orbit coupling in ultra-cold atoms has been made, see for instance Refs. 53-56. The first experimental realization of artificial spin-orbit coupling was achieved in Ref. 14 using Raman coupling within a pseudospin-1/2 manifold of F = 1 87 Rb atoms. In the experiment, a quadratic Zeeman shift detunes the third state from resonance in order to mimic SU(2), whereas an SU(3) spin-orbit coupling would arise naturally without this detuning. A theoretical proposal which explicitly considers spin-1 spin-orbit coupling within the very flexible N -pod scheme is found in Ref. 31.
While such flexibility might be needed to engineer the off-diagonal spin-orbit coupling of Sec. IV, much simpler schemes are sufficient to obtain a diagonal spin-orbit coupling, that is, a spin-dependent hopping phase. The implementation of a diagonal SU(2) spin-orbit coupling has been proposed recently in Ref. 57, stressing its great advantages by avoiding the heating related to spin-flip processes. The proposal is based on a lattice tilt to replace normal hopping by laser-assisted Raman tunneling. This technique has already given rise to artificial magnetic fields for atoms in optical lattices 58,59 . In contrast to an artifificial magnetic field, the artificial spin-orbit coupling does not require spatially dependent hopping phases, but instead, a spin dependence of the hopping phases is needed. The latter can be achieved via a lattice tilt induced by a magnetic field gradient, and the use of hyperfine states with different magnetic moments. In our case, two spin states experience the same hopping phase, while the third state does not gain a hopping phase. Accordingly, two hyperfine states should have the same magnetic moment, sufficiently different from the magnetic moment of the third state.
d. Detection of the spin textures. Once the desired Hamiltonian has been engineered at sufficiently low temperature, the main experimental concern is how to detect the ground state and its properties. The observable related to the spin textures reported in this paper are the spin-spin correlations, see Eqs. (19) and (24). Correlation measurements in a Bose gas have been pioneered using noise interferometry in Ref. 60. Nowadays, it is also possible to address single atoms in a spin-resolved way, using a quantum gas microscope 61 . Information about the periodicity of the spin pattern is encoded in the magnetic structure factor, which is directly measurable with a spin-dependent Bragg scattering 62 . This technique makes use of the fact that the light scattered from
an atom depends on the atom's internal state, rendering the interference pattern in a Bragg scattering experiment spin dependent with constructive interferences along angles corresponding to peaks in the magnetic structure factor. This technique has recently been used to observe antiferromagnetic correlations of a quantum gas in the superexchange limit of the Fermi-Hubbard model 52 .
## VI. SUMMARY AND OUTLOOK
We have derived the effective Hamiltonian of threecomponent bosons corresponding to the Mott phase in the presence of spin-orbit coupling. We have focused on two simple cases: a purely diagonal hopping with spin-dependent phases, as well as a particular purely off-diagonal spin-orbit coupling. The abelian spin-orbit coupling allows us to introduce an SU(3) analog of the Dzyaloshinskii-Moriya interaction (antisymmetric in the pseudo-spin positions). We studied the one-dimensional chain and the square lattice. Both, the diagonal and the off-diagonal coupling, lead to SU(3) spin spiral textures in the ground state. The periodicities of the spin textures are controlled by the spin-orbit coupling parameters. For diagonal spin-orbit coupling, remaining SU(3) degeneracies are lifted by asymmetry between the interand intra-particle interactions, µ < 1. Alternatively, the spiral patterns are obtained by restricting to a Hilbert space with fixed spin polarization.
nealing technique, yielding, for diagonal spin-orbit coupling, phases that can be understood from the corresponding one-dimensional case. On the other hand, a generic unique classical ground state spiral phase was obtained for symmetric interactions with non-abelian spinorbit coupling.
More generally, the interplay between diagonal and off-diagonal spin-orbit coupling leads to a much richer landscape of spin Hamiltonians than discussed here. The work considered here could be extended to the study of the competition between the asymmetry of interactions µ and the various types of spin-orbit coupling.
A completely different physical scenario might arise in the two-dimensional case, if one changes the lattice geometry. Assume, for instance, a triangular lattice with spindependent hopping phases along one direction, and trivial hopping on the other bonds: It is clear that the spiral order favored by the spin-dependent hopping cannot be implemented without frustrating at least one of the ferromagnetic bonds along the other direction. This results in a highly complex quantum problem for which the predictive power of numerical simulations is restricted, and which might best be addressed by a quantum simulation.
## Acknowledgments
On the quantum level, the spin textures can be traced back to a local gauge transformation over Dicke-like states. The classical ground state phase on the twodimensional lattice was studied using the simulated an-
We acknowledge support from ERC advanced Grant OSYRIS, EU IP SIQS, EU STREP EQuaM (FP7/20072013 Grant No. 323714), FOQUS, and Fundaci´ Cellex. o R.W.C. acknowledges a Mobility Plus fellowship from the Polish Ministry of Science and Higher Education and the (Polish) National Science Center Grant No DEC2011/03/B/ST2/01903.
## Appendix A: Effective Hamiltonians
## 1. SU(3) notation of biquadratic Heisenberg Hamiltonian
The SU(2) spin-1 matrices read
$$S ^ { x } = \frac { 1 } { \sqrt { 2 } } \begin{pmatrix} 0 & 1 & 0 \\ 1 & 0 & 1 \\ 0 & 1 & 0 \end{pmatrix}, \ \ S ^ { y } = \frac { 1 } { \sqrt { 2 } } \begin{pmatrix} 0 & - i & 0 \\ i & 0 & - i \\ 0 & i & 0 \end{pmatrix}, \ \ S ^ { z } = \begin{pmatrix} 1 & 0 & 0 \\ 0 & 0 & 0 \\ 0 & 0 & - 1 \end{pmatrix}.$$
The spin-1 vector is thus written in terms of Gell-Mann matrices as
$$\mathbf S = \frac { 1 } { \sqrt { 2 } } \left ( \lambda ^ { ( 1 ) } + \lambda ^ { ( 6 ) }, \lambda ^ { ( 2 ) } + \lambda ^ { ( 7 ) }, \frac { 1 } { \sqrt { 2 } } ( \lambda ^ { ( 3 ) } + \sqrt { 3 } \lambda ^ { ( 8 ) } ) \right ).$$
Using this equation, it is straightforward to translate any spin-1 Hamiltonian into the SU(3) picture. In particular, for the biquadratic Heisenberg Hamiltonian, Eq. (3), we find the following contributions
$$S _ { i } \cdot S _ { j } = & \frac { 1 } { 2 } \left ( \lambda _ { i } ^ { ( 1 ) } \cdot \lambda _ { j } ^ { ( 1 ) } + \lambda _ { i } ^ { ( 2 ) } \cdot \lambda _ { j } ^ { ( 2 ) } + \lambda _ { i } ^ { ( 6 ) } \cdot \lambda _ { j } ^ { ( 6 ) } + \lambda _ { i } ^ { ( 7 ) } \cdot \lambda _ { j } ^ { ( 7 ) } \right ) \\ & + \frac { 1 } { 2 } \left ( \lambda _ { i } ^ { ( 1 ) } \cdot \lambda _ { j } ^ { ( 6 ) } + \lambda _ { i } ^ { ( 6 ) } \cdot \lambda _ { j } ^ { ( 1 ) } + \lambda _ { i } ^ { ( 2 ) } \cdot \lambda _ { j } ^ { ( 7 ) } + \lambda _ { i } ^ { ( 7 ) } \cdot \lambda _ { j } ^ { ( 2 ) } \right ) \\ & + \frac { 1 } { 4 } \left ( \lambda _ { i } ^ { ( 3 ) } \cdot \lambda _ { j } ^ { ( 3 ) } + 3 \lambda _ { i } ^ { ( 8 ) } \cdot \lambda _ { j } ^ { ( 8 ) } + \sqrt { 3 } \lambda _ { i } ^ { ( 3 ) } \cdot \lambda _ { j } ^ { ( 8 ) } + \sqrt { 3 } \lambda _ { i } ^ { ( 8 ) } \cdot \lambda _ { j } ^ { ( 2 ) } \right ), \\ \left ( S _ { i } \cdot S _ { j } \right ) ^ { 2 } = & - \frac { 1 } { 2 } \left ( \lambda _ { i } ^ { ( 1 ) } \cdot \lambda _ { j } ^ { ( 6 ) } + \lambda _ { i } ^ { ( 6 ) } \cdot \lambda _ { j } ^ { ( 1 ) } + \lambda _ { i } ^ { ( 2 ) } \cdot \lambda _ { j } ^ { ( 7 ) } + \lambda _ { i } ^ { ( 7 ) } \cdot \lambda _ { j } ^ { ( 2 ) } \right ) \\ & + \frac { 1 } { 4 } \left ( \lambda _ { i } ^ { ( 3 ) } \cdot \lambda _ { j } ^ { ( 3 ) } - \lambda _ { i } ^ { ( 8 ) } \cdot \lambda _ { j } ^ { ( 8 ) } - \sqrt { 3 } \lambda _ { i } ^ { ( 3 ) } \cdot \lambda _ { j } ^ { ( 8 ) } - \sqrt { 3 } \lambda _ { i } ^ { ( 8 ) } \cdot \lambda _ { j } ^ { ( 2 ) } \right ) \\ & + \frac { 1 } { 2 } \left ( \lambda _ { i } ^ { ( 4 ) } \cdot \lambda _ { j } ^ { ( 4 ) } + \lambda _ { i } ^ { ( 5 ) } \cdot \lambda _ { j } ^ { ( 5 ) } \right ).$$
Apparently, this sum of both contributions yields the SU(3) Heisenberg Hamiltonian.
## 2. Pseudospin interactions - XYZ-type Hamiltonian
For pseudospin interactions as given in Eq. (5), one obtains the effective Mott Hamiltonian of Eq. (6), with the parameters:
$$J _ { 1 } = J _ { 2 } = \frac { t ^ { 2 } } { 2 U _ { + 0 } },$$
$$J _ { 4 } = J _ { 5 } = \frac { \stackrel { \cdot \cdot \cdot } { t ^ { 2 } } } { 2 U _ { + - } },$$
$$J _ { 6 } = J _ { 7 } = \frac { t ^ { 2 } } { 2 I J _ { \mathfrak { a } } },$$
$$J _ { 1 } & = J _ { 2 } = \frac { t ^ { 2 } } { 2 U _ { + 0 } }, \\ J _ { 4 } & = J _ { 5 } = \frac { t ^ { 2 } } { 2 U _ { + - } }, \\ J _ { 6 } & = J _ { 7 } = \frac { t ^ { 2 } } { 2 U _ { 0 } }, \\ & \quad J _ { 3 } = \frac { t ^ { 2 } } { 2 } \left ( \frac { 1 } { U _ { + + } } + \frac { 1 } { U _ { 0 0 } } - \frac { 1 } { U _ { + 0 } } \right ), \\ & \quad J _ { 8 } = \frac { t ^ { 2 } } { 6 } \left ( \frac { 1 } { U _ { + + } } + \frac { 1 } { U _ { 0 0 } } + \frac { 4 } { U _ { - - } } + \frac { 1 } { U _ { + 0 } } - \frac { 2 } { U _ { + - } } - \frac { 2 } { U _ { 0 - } } \right ), \\ & \quad J _ { 3 8 } = \frac { \sqrt { 3 t ^ { 2 } } } { 6 } \left ( \frac { 1 } { U _ { + + } } - \frac { 1 } { U _ { - - } } + \frac { 1 } { U _ { 0 - } } - \frac { 1 } { U _ { + } } \right ), \\ & \quad h _ { 3 } = \frac { 1 } { 3 } \left ( \frac { 1 } { U _ { + + } } - \frac { 1 } { U _ { 0 0 } } \right ), \\ & \quad h _ { 8 } = \frac { \sqrt { 3 t ^ { 2 } } } { 1 8 } \left ( \frac { 2 } { U _ { + + } } + \frac { 2 } { U _ { 0 0 } } - \frac { 4 } { U _ { - - } } + \frac { 2 } { U _ { + 0 } } - \frac { 1 } { U _ { + - } } - \frac { 1 } { U _ { 0 - } } \right ),$$
## 3. Spinor interaction and tunneling phases
Assuming spinor interactions, Eq. (1), in combination with a tunneling phase α for particles in |±〉 , the following second-order Hamiltonian is obtained along the direction of tunneling:
$$\text{Assuming spinor interactions, Eq. (1), in combination with a tunneling phase $\alpha$ for particles in |=fty, the following} \\ \text{second-order Hamiltonian is obtained along the direction of tunneling:} \\ H _ { \text{eff} } = \frac { - t ^ { 2 } } { U _ { 0 } + U _ { 2 } } \sum _ { i } \left \{ \left [ \cos \alpha \left ( \lambda _ { i } ^ { ( 1 ) } \cdot \lambda _ { i + 1 } ^ { ( 1 ) } + \lambda _ { i } ^ { ( 2 ) } \cdot \lambda _ { i + 1 } ^ { ( 2 ) } + \lambda _ { i } ^ { ( 6 ) } + \lambda _ { i } ^ { ( 6 ) } + \lambda _ { i } ^ { ( 7 ) } \cdot \lambda _ { i + 1 } ^ { ( 7 ) } \right ) } & \quad ( A ^ { 1 } ) \\ + \sin \alpha \left ( \lambda _ { i } ^ { ( 1 ) } \cdot \lambda _ { i } ^ { ( 2 ) } - \lambda _ { i } ^ { ( 2 ) } \cdot \lambda _ { i + 1 } ^ { ( 1 ) } + \lambda _ { i } ^ { ( 6 ) } \cdot \lambda _ { i + 1 } ^ { ( 7 ) } - \lambda _ { i } ^ { ( 7 ) } \cdot \lambda _ { i + 1 } ^ { ( 6 ) } + \lambda _ { i } ^ { ( 8 ) } \cdot \lambda _ { i + 1 } ^ { ( 8 ) } \\ + \frac { 2 \sqrt { 3 } } { 3 } \left ( \lambda _ { i } ^ { ( 3 ) } \cdot \lambda _ { i + 1 } ^ { ( 8 ) } + \lambda _ { i } ^ { ( 8 ) } \cdot \lambda _ { i + 1 } ^ { ( 3 ) } \right ) + \frac { 2 } { 3 } \lambda _ { i } ^ { ( 3 ) } - \frac { 2 \sqrt { 3 } } { 9 } \lambda _ { i } ^ { ( 8 ) } \right ] \\ - \frac { 3 t ^ { 2 } } { 3 U _ { 0 } - U _ { 2 } ( 1 - 4 \cos \alpha ) } \left [ \frac { 2 } { 3 } \left ( \lambda _ { i } ^ { ( 1 ) } \cdot \lambda _ { i } ^ { ( 6 ) } + \lambda _ { i } ^ { ( 6 ) } \cdot \lambda _ { i + 1 } ^ { ( 1 ) } + \lambda _ { i } ^ { ( 2 ) } \cdot \lambda _ { i } ^ { ( 7 ) } + \lambda _ { i } ^ { ( 7 ) } \cdot \lambda _ { i + 1 } ^ { ( 2 ) } \right ) \\ + \frac { 1 } { 3 } \left ( \lambda _ { i } ^ { ( 4 ) } \cdot \lambda _ { i + 1 } ^ { ( 4 ) } + \lambda _ { i } ^ { ( 5 ) } \cdot \lambda _ { i + 1 } ^ { ( 1 ) } + 2 \lambda _ { i } ^ { ( 3 ) } \cdot \lambda _ { i + 1 } ^ { ( 4 ) } \right ) \\ - \frac { \sqrt { 3 } } { 3 } \left ( \lambda _ { i } ^ { ( 3 ) } \cdot \lambda _ { i + 1 } ^ { ( 8 ) } + \lambda _ { i } ^ { ( 8 ) } \cdot \lambda _ { i + 1 } ^ { ( 3 ) } \right ) - \frac { 2 } { 3 } \lambda _ { i } ^ { ( 3 ) } + \frac { 2 \sqrt { 3 } } { 9 } \lambda _ { i } ^ { ( 8 ) } \right ] \\ - \frac { 3 t ^ { 2 } } { 3 U _ { 0 } - 2 U _ { 2 } ( 1 + 2 \cos \alpha ) } \left [ \frac { 2 } { 3 } \left ( \lambda _ { i } ^ { ( 4 ) } \cdot \lambda _ { i } ^ { ( 4 ) } + \lambda _ { i } ^ { ( 5 ) } \cdot \lambda _ { i + 1 } ^ { ( 5 ) } \right ) \\ - \frac { 2 } { 3 } \left ( \lambda _ { i } ^ { ( 1 ) } \cdot \lambda _ { i } ^ { ( 6 ) } + \lambda _ { i } ^ { ( 6 ) } \cdot \lambda _ { i + 1 } ^ { ( 1 ) } + \lambda _ { i } ^ { ( 2 ) } \cdot \lambda _ { i + 1 } ^ { ( 7 ) } + \lambda _ { i } ^ { ( 7 ) } \cdot \lambda _ { i + 1 } ^ { ( 2 ) } \right ) \\ - \frac { \sqrt { 3 } } { 3 } \left ( \lambda _ { i } ^ { ( 3 ) } \cdot \lambda _ { i + 1 } ^ { ( 8 ) } + \lambda _ { i } ^ { ( 8 ) } \cdot \lambda _ { i + 1 } ^ { ( 3 ) } \right ) + \frac { 1 } { 3 } \left ( \lambda _ { i } ^ { ( 3 ) } \cdot \lambda _ { i + 1 } ^ { ( 3 ) } - \lambda _ { i } ^ { ( 8 ) } \cdot \lambda _ { i + 1 } ^ { ( 8 ) } \right ) \right ] \right \}. & \quad ( A ^ { 1 } ) \\ \text{For non-zero tunneling phases, a SU(3)-field is acting in the $\lambda ^ { ( 3 ) } - \lambda ^ { ( 8 ) } \, \text{plane, favoring the polarization of the system} \\ \text{within this plane.}$$
For non-zero tunneling phases, a SU(3)-field is acting in the λ (3) -λ (8) plane, favoring the polarization of the system within this plane.
- 1 C. J. Myatt, E. A. Burt, R. W. Ghrist, E. A. Cornell, and C. E. Wieman, Phys. Rev. Lett. 78 , 586 (1997), URL http://link.aps.org/doi/10.1103/PhysRevLett.78.586 .
- 11 X. Zhang, M. Bishof, S. L. Bromley, C. V. Kraus, M. S. Safronova, P. Zoller, A. M. Rey, and J. Ye, Science 345 , 1467 (2014).
- 3 P. W. Anderson, Phys. Rev. 79 , 350 (1950), URL http://link.aps.org/doi/10.1103/PhysRev.79.350 .
- 2 M. Lewenstein, A. Sanpera, and V. Ahufinger (Oxford University Press, New York, 2012).
- 12 F. Scazza, C. Hofrichter, M. H¨ ofer, P. C. D. Groot, I. Bloch, and S. F¨lling, Nat. Phys. o 10 , 779 (2014).
- 4 P. W. Anderson, Phys. Rev. 115 , 2 (1959), URL http://link.aps.org/doi/10.1103/PhysRev.115.2 .
- 13 G. Cappellini, M. Mancini, G. Pagano, P. Lombardi, L. Livi, M. Siciliani de Cumis, P. Cancio, M. Pizzocaro, D. Calonico, F. Levi, et al., Phys. Rev. Lett. 113 , 120402 (2014), URL http://link.aps.org/doi/10.1103/PhysRevLett.113.120402 .
- 6 B. W. Shore, M. A. Johnson, K. C. Kulander, and J. I. Davis, Opt. Express 8 , 28 (2001).
- 5 M. Gell-Mann and Y. Ne'eman (Benjamin, New York, 1964).
- 14 Y. J. Lin, K. Jim´nez-Garc´ ıa, and I. B. Spielman, Nature e 471 , 83 (2011).
- 7 M. Greiter, S. Rachel, and D. Schuricht, Phys. Rev. B 75 , 060401 (2007), URL http://link.aps.org/doi/10.1103/PhysRevB.75.060401 .
- 8 T. A. T´ oth, A. M. L¨ auchli, F. Mila, and K. Penc, Phys. Rev. Lett. 105 , 265301 (2010), URL http://link.aps.org/doi/10.1103/PhysRevLett.105.265301
- 9 G. Go, J.-H. Park, and J. H. Han, Phys. Rev. B 87 , 155112 (2013), URL http://link.aps.org/doi/10.1103/PhysRevB.87.155112 .
- 10 T. Graß, B. Juli´-D´ ıaz, a M. Ku´ s, and M. Lewenstein, Phys. Rev. Lett. 111 , 090404 (2013), URL http://link.aps.org/doi/10.1103/PhysRevLett.111.090404
- . . 15 T. D. Stanescu, B. Anderson, and V. Galitski, Phys. Rev. A 78 , 023616 (2008), URL http://link.aps.org/doi/10.1103/PhysRevA.78.023616 . 16 S. Sinha, R. Nath, and L. Santos, Phys. Rev. Lett. 107 , 270401 (2011), URL http://link.aps.org/doi/10.1103/PhysRevLett.107.270401 . 17 R. Barnett, S. Powell, T. Graß, M. Lewenstein, and S. Das Sarma, Phys. Rev. A 85 , 023615 (2012), URL http://link.aps.org/doi/10.1103/PhysRevA.85.023615 . 18 M. A. Garcia-March, G. Mazzarella, L. Dell'Anna, B. Juli´-D´ ıaz, a L. Salasnich, and A. Polls, Phys. Rev. A 89 , 063607 (2014), URL http://link.aps.org/doi/10.1103/PhysRevA.89.063607 .
- D. M. Stamper-Kurn and M. Ueda, Rev. Mod. Phys. 85 , 1191 (2013), URL
.
.
.
54
55
56
57
.
58
59
60
.
61
62
- 19 T. Graß, B. Juli´-D´ ıaz, a M. Burrello, and M. Lewenstein, J. Phys. B: At. Mol. Opt. Phys. 46 , 134006 (2013), URL http://stacks.iop.org/0953-4075/46/i=13/a=134006 . 20 K. Osterloh, M. Baig, L. Santos, P. Zoller, and M. Lewenstein, Phys. Rev. Lett. 95 , 010403 (2005). 21 P. Hauke, O. Tieleman, A. Celi, C. ¨ lschl¨ger, O a J. Simonet, J. Struck, M. Weinberg, P. Windpassinger, K. Sengstock, M. Lewenstein, et al., Phys. Rev. Lett. 109 , 145301 (2012), URL http://link.aps.org/doi/10.1103/PhysRevLett.109.145301 22 T. Graß, K. Saha, K. Sengupta, and M. Lewenstein, Phys. Rev. A 84 , 053632 (2011), URL http://link.aps.org/doi/10.1103/PhysRevA.84.053632 . 23 A. T. Bolukbasi and M. Iskin, Phys. Rev. A 89 , 043603 (2014), URL http://link.aps.org/doi/10.1103/PhysRevA.89.043603 . 24 W. S. Cole, S. Zhang, A. Paramekanti, and N. Trivedi, Phys. Rev. Lett. 109 , 085302 (2012), URL http://link.aps.org/doi/10.1103/PhysRevLett.109.085302 25 J. Radi´ c, A. Di Ciolo, K. Sun, and V. Galitski, Phys. Rev. Lett. 109 , 085303 (2012), URL http://link.aps.org/doi/10.1103/PhysRevLett.109.085303 26 Z. Cai, X. Zhou, and C. Wu, Phys. Rev. A 85 , 061605 (2012), URL http://link.aps.org/doi/10.1103/PhysRevA.85.061605 . 27 S. Peotta, L. Mazza, E. Vicari, M. Polini, R. Fazio, and D. Rossini, Journal of Statistical Mechanics: Theory and Experiment 2014 , P09005 (2014), URL http://stacks.iop.org/1742-5468/2014/i=9/a=P09005 . 28 T. Graß, A. Celi, and M. Lewenstein, Phys. Rev. A 90 , 043628 (2014), URL http://link.aps.org/doi/10.1103/PhysRevA.90.043628 . 29 I. Dzyaloshinskii, J. Phys. Chem. Solids 4 , 241 (1958). 30 T. Moriya, Phys. Rev. 120 , 91 (1960), URL http://link.aps.org/doi/10.1103/PhysRev.120.91 . 31 G. Juzeli¯nas, u J. Ruseckas, and J. Dalibard, Phys. Rev. A 81 , 053403 (2010), URL http://link.aps.org/doi/10.1103/PhysRevA.81.053403 . 32 R. Barnett, G. R. Boyd, and V. Galitski, Phys. Rev. Lett. 109 , 235308 (2012), URL http://link.aps.org/doi/10.1103/PhysRevLett.109.235308 33 T.-L. Ho, Phys. Rev. Lett. 81 , 742 (1998), URL http://link.aps.org/doi/10.1103/PhysRevLett.81.742 . 34 C. K. Law, H. Pu, and N. P. Bigelow, Phys. Rev. Lett. 81 , 5257 (1998), URL http://link.aps.org/doi/10.1103/PhysRevLett.81.5257 . 35 T. Ohmi and K. Machida, J. Phys. Soc. Jpn. 67 , 1822 (1998). 36 A. Imambekov, M. Lukin, and E. Demler, Phys. Rev. A 68 , 063602 (2003), URL http://link.aps.org/doi/10.1103/PhysRevA.68.063602 . 37 S. K. Yip, Phys. Rev. Lett. 90 , 250402 (2003), URL http://link.aps.org/doi/10.1103/PhysRevLett.90.250402 38 L. de Forges de Parny, F. H´ ebert, V. G. Rousseau, and G. G. Batrouni, Phys. Rev. B 88 , 104509 (2013), URL http://link.aps.org/doi/10.1103/PhysRevB.88.104509 . 39 R. A. Williams, L. J. LeBlanc, K. Jim´nez-Garc´ ıa, M. C. e Beeler, A. R. Perry, W. D. Phillips, and I. B. Spielman, Science 335 , 314 (2012). 40 B. Juli´-D´ ıaz, a T. Graß, O. Dutta, D. E. Chang, and M. Lewenstein, Nat. Commun. 4 , 2046 (2013). 41 43 44 45 46 47 48 49 50 51 52 53
- http://link.aps.org/doi/10.1103/RevModPhys.85.1191 . 42 K. Jim´ enez-Garc´ ıa, L. J. LeBlanc, R. A. Williams, M. C. Beeler, A. R. Perry, and I. B. Spielman, Phys. Rev. Lett. 108 , 225303 (2012), URL http://link.aps.org/doi/10.1103/PhysRevLett.108.225303 . K. Eckert, /suppress L. Zawitkowski, M. Laskinen, A. Sanpera, and M. Lewenstein, New J. Phys. 9 , 133 (2007). O. Romero-Isart, K. Eckert, and A. Sanpera, Phys. Rev. A 75 , 050303(R) ((2007)). G. De Chiara, M. Lewenstein, and A. Sanpera, Phys. Rev. B 84 , 054451 (2011). M. Anderlini, P. J. Lee, B. L. Brown, J. Sebby-Strabley, W. D. Phillips, and J. V. Porto, Nature 448 , 452 (2007). P. J. Lee, M. Anderlini, B. L. Brown, J. Sebby-Strabley, W. D. Phillips, and J. V. Porto, Phys. Rev. Lett. 99 , 020402 (2007). S. Trotzky, P. Cheinet, S. F¨ olling, M. Feld, U. Schnorrberger, A. M. Rey, A. Polkovnikov, E. A. Demler, M. D. Lukin, and I. Bloch, Science 319 , 295 (2008). Y.-A. Chen, S. Nascimb` ene, M. Aidelsburger, M. Atala, S. Trotzky, and I. Bloch, Phys. Rev. Lett. 107 , 210405 (2011). J. Imriˇ ska, M. Iazzi, L. Wang, E. Gull, D. Greif, T. Uehlinger, G. Jotzu, L. Tarruell, T. Esslinger, and M. Troyer, Phys. Rev. Lett. 112 , 115301 (2014). D. Greif, T. Uehlinger, G. Jotzu, L. Tarruell, and T. Esslinger, Science 340 , 1307 (2013). R. A. Hart, P. M. Duarte, T.-L. Yang, X. Liu, T. Paiva, E. Khatami, R. T. Scalettar, N. Trivedi, D. A. Huse, and R. G. Hulet, arXiv:1407.5932 (2014). J. Dalibard, F. Gerbier, G. Juzeli¯nas, u and P. ¨ hberg, O Rev. Mod. Phys. 83 , 1523 (2011), URL http://link.aps.org/doi/10.1103/RevModPhys.83.1523 . F. Gerbier and J. Dalibard, New J. Phys. 12 , 033007 (2010). L. Mazza, A. Bermudez, N. Goldman, M. Rizzi, M. A. Martin-Delgado, and M. Lewenstein, New J. Phys. 14 , 015007 (2012). L. Mazza, M. Rizzi, M. Lewenstein, and J. Cirac, Phys. Rev. A 82 , 043629 (2010). C. J. Kennedy, G. A. Siviloglou, H. Miyake, W. C. Burton, and W. Ketterle, Phys. Rev. Lett. 111 , 225301 (2013), URL http://link.aps.org/doi/10.1103/PhysRevLett.111.225301 . H. Miyake, G. A. Siviloglou, C. J. Kennedy, W. C. Burton, and W. Ketterle, Phys. Rev. Lett. 111 , 185302 (2013), URL http://link.aps.org/doi/10.1103/PhysRevLett.111.185302 . M. Aidelsburger, M. Atala, M. Lohse, J. T. Barreiro, B. Paredes, and I. Bloch, Phys. Rev. Lett. 111 , 185301 (2013), URL http://link.aps.org/doi/10.1103/PhysRevLett.111.185301 . S. F¨lling, F. Gerbier, A. Widera, O. Mandel, T. Gericke, o and I. Bloch, Nature 434 , 481484 (2005), ISSN 0028-0836, 10.1038/nature03500. W. S. Bakr, J. I. Gillen, A. Peng, S. F¨ olling, and M. Greiner, Nature 462 , 7477 (2009), ISSN 0028-0836, 10.1038/nature08482. T. A. Corcovilos, S. K. Baur, J. M. Hitchcock, E. J. Mueller, and R. G. Hulet, Phys. Rev. A 81 , 013415 (2010), URL http://link.aps.org/doi/10.1103/PhysRevA.81.013415 .
|
10.1103/PhysRevB.90.195127
|
[
"Tobias Graß",
"Ravindra W. Chhajlany",
"Christine A. Muschik",
"Maciej Lewenstein"
] |
2014-08-04T19:04:12+00:00
|
2014-11-18T14:14:49+00:00
|
[
"cond-mat.quant-gas",
"quant-ph"
] |
Spiral spin textures of bosonic Mott insulator with SU(3) spin-orbit coupling
|
We study the Mott phase of three-component bosons, with one particle per
site, in an optical lattice by mapping it onto an SU(3) spin model. In the
simplest case of full SU(3) symmetry, one obtains a ferromagnetic Heisenberg
model. Introducing an SU(3) analog of spin-orbit coupling, additional spin-spin
interactions are generated. We first consider the scenario of spin-dependent
hopping phases, leading to Dzyaloshinskii-Moriya-type interactions. They result
in the formation of spiral spin textures, which in one dimension can be
understood by a local unitary transformation. Applying classical Monte Carlo
simulations, we extend our study to two-dimensional systems, and systems with
"true" spin-orbit coupling, involving spin-changing hoppings.
|
1408.0770v2
|
## Electrons and holes in phosphorene
Pengke Li ∗ and Ian Appelbaum
Department of Physics and Center for Nanophysics and Advanced Materials, University of Maryland, College Park, MD 20742
We present a symmetry analysis of electronic bandstructure including spin-orbit interaction close to the insulating gap edge in monolayer black phosphorus ('phosphorene'). Expressions for energy dispersion relation and spin-dependent eigenstates for electrons and holes are found via simplification of a perturbative expansion in wavevector k away from the zone center using elementary group theory. Importantly, we expose the underlying symmetries giving rise to substantial anisotropy in optical absorption, charge and spin transport properties, and reveal the mechanism responsible for valence band distortion and possible lack of a true direct gap.
## I. INTRODUCTION
The experimental isolation of atomically thin 2dimensional layers from van der Waals-bonded 3dimensional solids has exposed many new opportunities for revealing unconventional electron transport physics. This research area, famously begun with exfoliation of the semimetal graphene from bulk graphite, has now vastly expanded to include work on related group-IV structures (silicene 1 , germanene , and stanene ), and on 2 3 binary semiconductors such as transition-metal dichalcogenides (WS 2 4 , etc.), topological insulators such as Bi Se 2 3 5 , and the group-III-V insulator boron nitride . 6 However, until recently, little has been done to explore the possibility that other forms of elemental compounds beyond group-IV can be exfoliated into few- or singlelayer structures like graphite can. Among candidate bulk source materials, orthorhombic black phosphorus (elemental group V) has emerged as a contender.
which can be compared to empirical dispersion relations obtained using Angle-Resolved Photo-Emission Spectroscopy from the clean surface. 29 Among the intriguing bandstructure features found are p-type semiconducting bandgap in the visible or infrared region and large excitonic binding energy, 20,30 prominent anisotropy of effective mass and hence carrier mobility, 25,26 ultraflat valence band dispersion and possible indirect bandgap, straininduced gap modification, 28 high optical efficiency, 31 etc.
The classical literature on this substance, reviewed in Ref. 7, is fairly complete at first glance. Single crystals of this ambient-stable allotrope are typically produced using high-pressure Bridgman growth 8,9 from which atomic structure was first determined via X-ray diffraction more than 50 years ago. 10,11 Early experimental results on charge transport 12 were complemented by investigations into both electrical and optical properties 13,14 and compared with theoretical predictions 15,16 . Unexpected phenomena including superconductivity 17 up to 13 K and evidence of 2D transport 18 motivated a modest resurgence of interest several decades ago.
Following in the footsteps of graphene's rise in the past decade, more recent experimental work on black phosphorus has focused on field-effect transistor action using thin multi-layered exfoliated flakes as channel conductor. 19,20 Single layer black phosphorus, dubbed 'phosphorene', is of particular interest but has not yet been incorporated into electrical devices. The properties of this 2-dimensional semiconductor have been studied in detail only by using the familiar machinery for bandstructure calculation. Several groups have addressed this problem using different approaches, including the empirical tight-binding 2122 , pseudopotential 23 , and ab initio (Density Functional Theory (DFT)) 24-28 methods, all of
Despite this abundance of bandstructure results in the available literature, several elementary questions remain unanswered, all of which become crucially important once single-layer phosphorene devices are experimentally realized. For example, what is the origin of the large valence-band effective mass anisotropy? What mechanisms determine whether this material has a truly direct bandgap? What are the optical transition selection rules? What are the dominant wavefunction components that dictate spin-dependent properties? etc. The answers to these questions are essential in providing insight for predictions of the properties of electrons and holes affecting charge and spin transport in this material, and are therefore necessary in developing possible device applications.
Various brute-force numerical schemes can churn out the relevant quantities needed to answer the questions above, but they often come at the expense of obscuring the physics at their root, i.e. the fundamental symmetries manifest in the structure of this physical materials system. In the present paper, we exploit the discrete lattice/wavefunction symmetries in phosphorene using the formal results of group theory to directly answer these questions. By first identifying the symmetry properties of wavefunctions at the Brillouin zone center, we simplify k ˆ · p perturbation theory using the method of invariants 32 and matrix element theorem to identify terms contributing to the dispersion and spin-dependent eigenstates of all relevant bands at nearby momenta. All the symmetryprotected properties are captured by k · ˆ p parameters that can be easily verified by numerical calculations and empirically determined by further experiments.
The present paper is organized as follows: Section II provides general information on the symmetry of the phosphorene 2D lattice, from which we analyze the symmetries of constituent atomic orbitals and the nearly free
FIG. 1. Orthorhombic lattice of phosphorene. (a) In realspace, two types of bonds between neighboring atoms are indicated by d 1 and d 2 . The origin of the Cartesian coordinates (red arrows) is chosen to be at the center of a d 2 bond. (b) Top view, showing the two lattice vectors (blue arrows). Inset: 2D projection of the reciprocal lattice Brillouin zone, with high symmetry points indicated.
electron model. In Section III, we investigate the spinindependent part of the Hamiltonian using the method of invariants and reveal the fundamental origins of the effective mass anisotropy, especially those interactions resulting in an ultraflat valence band. Optical selection rules are also provided. Using the same approach, in Section IV we focus on the spin-orbit interaction and construct the spin-dependent eigenstates. From these results, we analyze spin relaxation anisotropy before providing a summary and outlook in Section V.
## II. SYMMETRY CONSIDERATIONS
## A. Lattice symmetry and space group operators
The black phosphorus monolayer is a 2D hexagonal lattice that is buckled, or 'puckered', along the armchair direction. This geometry results in two types of bonds [see Fig. 1(a)]. P atoms connected by bonds parallel to the 2D plane (Type I, with bond length d 1 = 2 224 ˚ ) . A form upper and lower sublayers, while bonds connecting P atoms between these two sublayers (Type II) have a length d 2 = 2 244 ˚ and are oriented 71 7 . A . ◦ out of the plane. In the following discussion, we use Cartesian coordinates with the origin at the center of one Type II
TABLE I. Character table of the Γ-point space group, including basis functions of the IRs. x , y , and z are components of polar vectors, while A x , A y and A z are components of axial vectors.
| E | τC 2 x C | 2 y τC 2 z | i τR x R y τR z |
|-------------|------------|--------------|-------------------|
| Γ + 1 1 1 + | 1 | 1 1 1 | 1 1 -1 1 -1 |
| Γ 2 1 -1 | 1 | -1 1 | A y |
| Γ + 3 1 1 | -1 | -1 1 1 | -1 -1 A x |
| Γ + 4 1 -1 | -1 | 1 1 | -1 -1 1 A z |
| Γ - 1 1 1 | 1 | 1 -1 -1 | -1 -1 |
| Γ - 2 1 -1 | 1 | -1 | -1 1 -1 1 y |
| Γ - 3 1 1 | -1 | -1 -1 | -1 1 1 x |
| Γ - 4 1 -1 | -1 | 1 | -1 1 1 -1 z |
bond. The z axis is chosen to be out of plane, with the in-plane x axis along the armchair direction and the y axis transverse to it. In this rectangular x -y basis, the two Bravais lattice constants are a x = 4 376 ˚ and . A a y = 3 314 ˚ [see Fig. 1(b)], and within a unit cell there . A are four P atoms. The inset in Fig. 1(b) shows the first Brillouin zone of this 2D orthorhombic lattice with high symmetry points labeled: Γ-point is the zone center, and X and Y points are at ( π a x , 0) and (0 , π a y ), respectively, half of the reciprocal lattice vectors.
Phosphorene shares the same in-plane translation symmetry with its bulk counterpart black phosphorus, whose space group is base-centered orthorhombic with international number and symbol 64 : Cmca . 33 The lattice structure in Fig. 1 provides all necessary information about its nonsymmorphic space group, whose factor group is isomorphic to the point group D 2 h . There are eight elements in this nontrivial factor group; each of them is a coset about the direct product of a symmetry operator and the lattice vector translation subgroup T . Among these eight symmetry operators, four are pure rotations (either proper or improper), including the identity operator E , the space inversion operator i , the operator corresponding to 180 ◦ rotation around the y axis C 2 y , and the operator corresponding to reflection with respect to the y = 0 plane R y . The remaining four are a translation τ = ( a x 2 , a y 2 ) in addition to pure rotations, including τC 2 x , τC 2 z , τR x and τR z . Since the D 2 h group is abelian (commutative), each group element forms a single class.
The most important information about the symmetry of this space group and its eight irreducible representations (IRs) is included in the character table (see Table I). Since this is the same group as that of the Brillouin zone center Γ-point, these IRs are denoted by Γ +( -) i , with subscript i = 1 2 3 4, and superscript + ( , , , -) indicating even (odd) parity under the inversion operator. In this table we also list the basis functions of some IRs. Here, x y , and z are components of a polar vector (e. g. the momentum operator ˆ p ) while A x , A y and A z are those of an axial, or 'pseudo-', vector (essential for analyzing the effect of spin-orbit interaction with vector field ∝ ∇ V × ˆ p ).
In the following subsection, we will briefly discuss the
FIG. 2. Four configurations of the p y orbitals within a unit cell, together with their associated IRs, all of which are odd under the reflection operator R y . They represent the four zone-center eigenstates composed solely from the pure p y atomic orbitals. From left to right, the eigenenergies of the four orbital states decrease [ E (Γ + 3 ) > E (Γ -2 ) > E (Γ + 4 ) > E (Γ -1 )], according to the bonding or antibonding nature of the covalent bonds between neighboring atoms. Note that the Γ -2 configuration also represents the long-wavelength acoustic phonon mode where all atoms move in-phase along the zigzag direction.
symmetries of the atomic orbitals and the empty lattice 'nearly free' electron band structure at the Γ-point, using the information in Table I. These considerations provide intuition on the symmetry-related coupling of the eigenstates near the zone center and are important for understanding the overall electronic structure.
## B. Atomic orbital symmetries at the center of the Brillouin zone
The band structure of black phosphorus was studied by Takao et. al. 21 with a spin-independent tight-binding model using a basis of one s orbital and three p orbitals. With four P atoms within a unit cell, there are therefore sixteen bands. By calculating the hopping energy and orbital overlaps between neighboring atoms, they determined the band gaps of black phosphorus in the form of monolayer and bulk, and indexed the symmetries of eigenstates at the Γ-point. In general, wavefunctions constructed this way will consist of sp 3 hybridized atomic orbitals. However, at the zone center, the p y orbital (which is odd under R y ) is isolated from the remaining s , p x and p z orbitals that remain mixed (similar to the case of pure p z orbitals in monolayer graphene due to the R z operator).
According to the character table (Table I), all the sixteen eigenstates at the Γ-point are nondegenerate, since all the IRs are one-dimensional. Four of these IRs, associated with the four p y orbital configurations illustrated in Fig. 2, are odd under the reflection operator R y : Γ + 3 , Γ + 4 , Γ -1 and Γ -2 . The bonding energy is dominated by Type I bonds (within each sublayer) that hybridize ppπ and ppσ covalent chemical bonds, the latter of which is much stronger due to higher orbital overlap. Type II bonds (connecting the two sublayers) are pure but weaker ppπ bonds and have only a secondary contribution to the bonding energy. By considering the bonding and antibonding nature of the p y orbitals, we give in Fig. 2 the relative order of the four eigenenergies, which matches the tight-binding calculation 21 .
Takao et. al. gave the p z orbital configurations at the
FIG. 3. Same as Fig. 2, here for (a) p x , (b) p z and (c) s orbitals that are even under the reflection operator R y . IRs in each row are listed from left to right in descending order of total covalent bond energy. Similar to Γ -2 in Fig. 2, here we see configurations corresponding to the remaining two long wavelength acoustic phonon modes: Γ -3 in (a) corresponds to in-plane motion along the armchair direction while Γ -4 in (b) corresponds to out-of-plane (flexural) phonons.
bandgap edge (Γ + 2 for the valence band and Γ -4 for the conduction band). In Fig. 3(a), (b), and (c), we list the configurations of all the p x , p z and s orbitals, respectively, for the four IRs Γ + 1 , Γ + 2 , Γ -3 and Γ -4 that are even under the reflection operator R y , and order them according to their bond energy. Within the sp 3 tight-binding model, each of these four representations corresponds to three bands, among which the contributions of p x , p z and s orbital components with the same symmetry vary.
Similar to our discussion of the p y orbital, we now examine the bonding or antibonding characteristics of these atomic orbital configurations and determine the relative energy ordering of the eigenstates. In the following sections we will focus on the band edge states which belong to Γ + 2 and Γ -4 , both dominated by the p z orbital. In this case [see Fig. 3(b)], ppσ bonds are within the Type II bonds, which are bonding (antibonding) in Γ + 2 (Γ -4 ) with lower (higher) energy assigning it the top of the valence band (bottom of the conduction band).
Using the first-principles calculation package Quantum ESPRESSO 34 , we studied this energy order of the Γ-point eigenstates. Different density functionals and pseudopotentials chosen as input to the ab initio calculation vary the detailed values of energy differences at the Γ point as well as the dispersion curves away from it. However, the band ordering at the Γ point is consistent with those given by Takao et. al. (only a few bands very close in energy switch places). This verifies that the order of the Γ point eigenstates is generally determined by the symmetries of the atomic orbitals alone.
## C. Empty-lattice band structure
The atomic orbitals give a perturbative picture of the electronic structure in the tight-binding regime. A useful
FIG. 4. Free-electron bandstructure, including symmetry labels (yellow) of zone-center eigenstates formed from superpositions of plane-waves centered at the reciprocal lattice points indicated (purple).
approximation in the opposite (delocalized) extreme is the nearly-free electron model that describes the emptylattice band structure, in which the electronic states are pure planewaves with wave numbers given by the reciprocal lattice vectors ( G n ). Using projection operators, we obtain the symmetrized wavefunctions which are linear combinations of planewaves degenerate at the Γpoint. When the atomic potentials are introduced into this nearly free electron model, only those symmetrized wavefunctions belonging to the same IR can be linearly combined to form the real eigenstates (the symmetry of the eigenstates from empty-lattice to the real lattice is maintained) and result in broken degeneracy with the possibility of bandgaps.
An important potential concern with this approach is that it does not account for the transformation properties of the wavefunctions under out-of-plane reflection. Therefore, in principle, wavefunctions constructed in this way provide only incomplete symmetry properties. However, since the dynamical characteristics of interest involve only in-plane symmetries, the simple nearly-free electron model can indeed provide sufficient information.
Because of the orthorhombic symmetry of monolayer black phosphorus, the degeneracy of nearly-free electron states at the Γ-point can only be singlet ( G n = 0), doublet ( G n on the k x or k y axes) or quartet (general G n ). Fig. 4 shows the empty-lattice band structure including the lowest nine Γ-point eigenstates. In the following sections, we will show that the in-plane momentum matrix elements 〈 ˆ p x,y 〉 of the zone center eigenstates play a fundamental role in determining the electronic structure of phosphorene. The nearly-free electron model explicitly shows that 〈 ˆ p x,y 〉 are nonzero only between degenerate states with relevant wavevector components. For instance, for the doubly degenerate first excited states in the form of symmetrized wavefunctions with G n = ( ± 2 πx a x , 0), we have
$$\langle \Gamma _ { 2 } ^ { + } | \hat { p } _ { x } | \Gamma _ { 4 } ^ { - } \rangle = \langle \cos \frac { 2 \pi x } { a _ { x } } | \frac { \hbar { } } { i } \frac { \partial } { \partial x } | \sin \frac { 2 \pi x } { a _ { x } } \rangle = \frac { 2 \pi \hbar { } } { a _ { x } }, \quad ( 1 )$$
while 〈 Γ + 2 | ˆ p y | Γ -4 〉 = 0. Similarly, ˆ p x and ˆ p y between either of these two states and any other remaining Γ-point planewave state vanish, consisent with the matrix element theorem as discussed below.
## III. SPIN-INDEPENDENT PROPERTIES
In this section, we study the spin-independent band structure properties close to the bandgap in phosphorene. We construct the k · ˆ p Hamiltonian near the zone center using the method of invariants, 32 and show that the dispersion relation of the conduction band and valence band can be captured by an effective mass approximation only in the k x direction, while in the k y direction the unique coupling from remote bands can potentially lead to an indirect bandgap.
## A. Hamiltonian and method of invariants
Following conventional k ˆ · p theory, the Hamiltonian is given by
$$H = H _ { 0 } + H _ { \mathbf k \cdot \mathfrak { P } } + H _ { \mathrm S O } + H _ { \mathrm S O, \mathbf k }, \quad \quad ( 2 )$$
where
$$H _ { 0 } = \frac { \hbar { ^ { 2 } } } { 2 m _ { 0 } } ( k _ { x } ^ { 2 } + k _ { y } ^ { 2 } ),$$
$$H _ { \mathbf k \cdot \mathfrak { p } } = \frac { \hbar } { m _ { 0 } } ( k _ { x } \hat { p } _ { x } + k _ { y } \hat { p } _ { y } ),$$
$$H _ { \text{SO} } = \frac { \hslash } { 4 m _ { 0 } ^ { 2 } c ^ { 2 } } \nabla V \times \hat { \mathbf p } \cdot \vec { \sigma }, \text{ and } \quad \quad \quad ( 5 )$$
$$H _ { \text{SO}, \text{k} } = \frac { \hslash ^ { 2 } } { 4 m _ { 0 } ^ { 2 } c ^ { 2 } } & \left [ ( k _ { x } \sigma _ { y } - k _ { y } \sigma _ { x } ) \frac { \partial V } { \partial z } + \\ k _ { y } \sigma _ { z } \frac { \partial V } { \partial x } - k _ { x } \sigma _ { z } \frac { \partial V } { \partial y } \right ]. \quad \text{(6)}$$
Here, H 0 is the in-plane free electron dispersion, and H k ˆ · p is the k · ˆ p term to be treated perturbatively. In this section we will focus on these two spin-independent terms. The spin-related properties are captured by the k -independent H SO and k -dependent H SO , k terms, and are discussed in the subsequent section. Note that for light atoms like phosphorus, the interaction strength hierarchy is generally H k ˆ · p glyph[greatermuch] H SO glyph[greatermuch] H SO , k .
Given the symmetry of the problem, we naturally choose to represent this Hamiltonian in a basis defined by the spin-independent Γ-point eigenstates. Matrix elements of perturbative terms in Eq. (2) can then be determined by the method of invariants 32 : only if the IR associated with the invariant component operator is included
TABLE II. Table of invariants
| IRs | Γ + 1 | Γ + 2 | Γ + 3 | Γ + 4 | Γ - 2 | Γ - 3 | Γ - 4 |
|------------|---------------|---------|---------|---------|-----------------|---------------|-------------------|
| Invariants | k 2 x + k 2 y | σ y | σ x | σ z | k y , - k x σ z | k x , k y σ z | k x σ y - k y σ x |
TABLE III. Direct product between Γ + 2 (Γ -4 ) and all IRs.
| | Γ + 1 Γ + 2 Γ + 3 Γ + 4 Γ - 1 Γ - 2 Γ - 3 Γ - 4 |
|-------|---------------------------------------------------|
| Γ + 2 | Γ + 2 Γ + 1 Γ + 4 Γ + 3 Γ - 2 Γ - 1 Γ - 4 Γ - 3 |
| Γ - 4 | Γ - 4 Γ - 3 Γ - 2 Γ - 1 Γ + 4 Γ + 3 Γ + 2 Γ + 1 |
in the direct sum decomposition of the direct product of the two IRs of the basis functions can the matrix element be nonzero. In Table II, we list the association of the IRs with all the invariant components of terms in Eq. (2) according to their transformation properties under the symmetry operators in Table I.
Our main focus is on the lowest conduction band and the highest valence band, which belong to the IRs Γ -4 and Γ + 2 , respectively. For convenience, we list the direct product of these two IRs with all IRs in Table III. In particular, we will reveal the origin of the anisotropy of the energy dispersion relation in the k x and k y directions, which is naturally endowed by the orthorhombic symmetry of the crystal lattice.
## B. Effective mass of electrons and holes in the k x direction
With the help of Table II, we see that the perturbative term proportional to k p x ˆ x in Eq. (4) belongs to Γ -3 . This term directly couples the lowest conduction band and the highest valence band in off-diagonal matrix elements (Γ -3 = Γ -4 ⊗ Γ + 2 ; see Table III). Similarly, the term proportional to k 2 x in Eq. (3) belongs to Γ + 1 and corresponds to diagonal matrix elements Γ -4 ⊗ Γ -4 and Γ + 2 ⊗ Γ + 2 . Notice that the second lowest conduction band also belongs to Γ + 2 and therefore is expected to affect the Γ -4 bottom conduction band significantly as well. Given the fact that all other remote bands belonging to Γ + 2 or Γ -4 are much further away in energy, we can describe the k x direction dispersion relation close to the bandgap by a minimal 3 × 3 Hamiltonian with a basis of { Γ + 2c , Γ -4c , Γ + 2v } (where the subscripts v and c indicate 'valence' and 'conduction', respectively):
$$H _ { 3 \times 3 } = \left ( \begin{array} { c c c } E _ { 1 } + E _ { g } + \frac { h ^ { 2 } k _ { x } ^ { 2 } } { 2 m _ { 0 } } & P _ { x 2 } k _ { x } & 0 \\ P _ { x 2 } k _ { x } & E _ { g } + \frac { h ^ { 2 } k _ { x } ^ { 2 } } { 2 m _ { 0 } } & P _ { x 1 } k _ { x } \\ 0 & P _ { x 1 } k _ { x } & \frac { h ^ { 2 } k _ { x } ^ { 2 } } { 2 m _ { 0 } } \end{array} \right ), \, ( 7 ) & E _ { h }$$
where E g is the bandgap at the Γ-point, E 1 is the energy difference between Γ + 2c , and Γ -4c , and the off-diagonal ma-
ky glyph[negationslash]
FIG. 5. Schematic bandstructure of phosphorene near the zone center, illustrating the relevant perturbative interactions allowed by symmetry. Red double arrows indicate dominant coupling terms, and pink arrows highlight important nonnegligible additional interactions. The gray region represents the forbidden gap. See Eq. (19) and related text for conditions resulting in indirect gap ( E , k 0 0 = 0). The plotting of bands are not to scale, especially E 0 is exaggerated for better illustration.
trix elements are dependent on
$$P _ { x 1 } = \frac { \hbar } { m _ { 0 } } \langle \Gamma _ { 4 c } ^ { - } | \hat { p } _ { x } | \Gamma _ { 2 \nu } ^ { + } \rangle, \text{ and } \text{ \quad \ \ } ( 8 )$$
$$P _ { x 2 } = \frac { \hbar } { m _ { 0 } } \langle \Gamma _ { 4 c } ^ { - } | \hat { p } _ { x } | \Gamma _ { 2 c } ^ { + } \rangle.$$
Such couplings are shown by double arrows in Fig. 5 on the k x side of the Γ-point in a schematic bandstructure. Note that, for zone center wavefunctions, we can always adjust their overall phases so that P x 1 and P x 2 are real numbers.
Applying the L¨wdin partitioning method o 35 to lowest order, Eq. (7) yields analytic expressions for the dispersion relations of electrons and holes in the k x direction:
$$E _ { e } ( k _ { x } ) = E _ { g } + \frac { \hslash ^ { 2 } } { 2 m _ { 0 } } k _ { x } ^ { 2 } - \frac { P _ { x 2 } ^ { 2 } } { E _ { 1 } } k _ { x } ^ { 2 } + \frac { P _ { x 1 } ^ { 2 } } { E _ { g } } k _ { x } ^ { 2 }, \text{ and } \text{ } ( 1 0 )$$
$$\L ^ { \cdot \cdot \cdot \cdot } _ { h } & E _ { h } ( k _ { x } ) = \frac { \hbar { 2 } } { 2 m _ { 0 } } k _ { x } ^ { 2 } - \frac { P _ { x 1 } ^ { 2 } } { E _ { g } } k _ { x } ^ { 2 }. & \L ^ { \cdot } & & ( 1 1 )$$
From these eigenenergies, we can then calculate the ef-
fective masses via [ 1 glyph[planckover2pi1] 2 d E 2 dk 2 x ] -1 :
$$\frac { 1 } { m _ { e, x } } = \frac { 1 } { m _ { 0 } } - \frac { 2 P _ { x 2 } ^ { 2 } } { E _ { 1 } } + \frac { 2 P _ { x 1 } ^ { 2 } } { E _ { g } }, \text{ and } \quad \text{ (12)}$$
$$\frac { 1 } { m _ { h, x } } = - \frac { 1 } { m _ { 0 } } + \frac { 2 P _ { x 1 } ^ { 2 } } { E _ { g } }. \quad \begin{array} { c c c } \cdot \\ \end{array} \quad \begin{array} { c c c } \cdot \\ \end{array} \quad \begin{array} { c c c } \cdot \\ \end{array} \quad \begin{array} { c c c } \cdot \\ \end{array} \quad \begin{array} { c c c } \cdot \\ \end{array} \quad \begin{array} { c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c c c c } \cdot \\ \end{array} \quad \begin{array} { c c c c c c c } \cdot \\ \end{array} \quad \begin{array} {$$
To evaluate m e and m h , we utilize the results of the free electron model in Sec. II C. Our DFT calculations show that both Γ -4c and Γ + 2v have a dominant first excited state planewave component with G = ( 2 π a x , 0), while for Γ + 2 c , this component is small and the majority is the second excited state with G = (0 , 2 π a y ), regardless of the DFT details. Thus, the magnitudes of both m e,x and m h,x are largely dictated by the term related to P x 1 [illustrated in Fig. (5) by the large red double arrow on the k x side], with amplitude on the order of 2 π glyph[planckover2pi1] 2 m a 0 x [see Eq. (1)], while the P x 2 -related term is a small correction to m e,x that reduces the difference between m -1 e,x and m -1 h,x .
It should be noted that a recently-proposed two-band model yields a similar dispersion relation in the k x direction. 28 In that model, it is claimed that the effect of the Γ + 2c conduction band (as well as the contribution of remote bands) has been lumped into off-diagonal matrix elements by terms quadratic in k after L¨wdin paro titioning. We are compelled to point out that this result is flawed, however, since k 2 x is not an invariant of Γ -4 ⊗ Γ + 2 = Γ -3 , and L¨ owdin partitioning always maintains the underlying symmetry. Furthermore, inclusion of specious odd-power terms in the eigenenergies (due to the coupling of the off-diagonal linear and quadratic terms) breaks time-reversal symmetry such that dispersions in + k x and -k x directions are unphysically different. For the same reason, the off-diagonal k 2 y term in that two-band model (leading to a k 4 y dispersion relation) is also incorrect. In the following subsection, we will show the correct origins of the flat valence band and possible indirect bandgap in the k y direction.
## C. Dispersion relation in the k y direction
Unlike the case in the k x direction, direct coupling of the band edge states Γ + 2v and Γ -4c is absent in the k y direction; energetically-remote bands must therefore be taken into account. The situation for the conduction band is relatively simple: according to Tables II and III, the invariant component k y belongs to Γ -2 and couples Γ -4c to Γ + 3 states. As discussed in Sec. II B, within the sp 3 tightbinding model giving the lowest sixteen bands, Γ + 3 is the highest energy state among the four pure p y orbitals. Numerical calculation shows that it lies beyond Γ -4c 21 , consistent with our DFT calculation. Together with other (even higher) states with the same symmetry, Γ + 3c repels the dispersion of the lowest conduction band downward via the matrix element
$$P _ { y 1 } = \frac { \hbar } { m _ { 0 } } \langle \Gamma _ { 3 c } ^ { + } | \hat { p } _ { y } | \Gamma _ { 4 c } ^ { - } \rangle. \text{ \quad \ \ } ( 1 4 )$$
This interaction is schematically shown with a doublesided pink arrow in Fig. 5. Since the planewave component of G = (0 , 2 π a y ) in Γ + 4c is relatively small, P y 1 is not large enough to reverse the conduction band's positive curvature. However, it does result in a value of the effective mass larger than m 0 .
The valence band state Γ + 2v requires a different analysis. According to Table III, and reflected in Fig. 5, Γ -1v can directly couple to Γ + 2v via k p y ˆ y perturbation. This state, the lowest among the four pure p y orbital states (see Fig. 2), compels us to consider a matrix element
$$P _ { y 2 } = \frac { \hbar } { m _ { 0 } } \langle \Gamma _ { 1 \nu } ^ { - } | \hat { p } _ { y } | \Gamma _ { 2 \nu } ^ { + } \rangle. \quad \quad ( 1 5 )$$
Compared with P x 1 , this quantity is relatively small due to the minority (0 , 2 π a y ) planewave component in Γ + 2v (despite its dominant role in Γ -1v ). However, without a counterbalancing interaction, its presence would repel the Γ + 2 v , band upward, leading to an electron-like positive effective mass and close the bandgap.
Previous DFT calculation by others has already shown that monolayer black phosphorus potentially posseses an indirect bandgap, in which the maximum of the valence band is located along the k y direction away from the zone center. 28 Using various input density functionals and pseudopotentials, we have verified the persistence of this feature which leads to a small positive energy difference ( E 0 ∼ meV, depending on the details of the numerical procedures) between the valence band maximum at wavevector k 0 ( ≈ 10% from BZ edge) and eigenenergy at the zone center. In contrast to an ordinary quadratic dispersion relation, understanding such unusual band structure requires careful examination of the couplings between Γ + 2v , Γ -1v , Γ + 2c , and further upper conduction bands of Γ -1 symmetry.
We have mentioned that for both Γ -1v and Γ + 2c states, planewave component (0 , 2 π a y ) dominates the wavefunctions. As indicated in Fig. 5 by a large red double arrow, the direct coupling of these two states by k p y ˆ y in
$$P _ { y 3 } = \frac { \hbar } { m _ { 0 } } \langle \Gamma _ { 1 \nu } ^ { - } | \hat { p } _ { y } | \Gamma _ { 2 c } ^ { + } \rangle \sim \frac { 2 \pi \hbar { ^ { 2 } } } { m _ { 0 } a _ { y } } \quad \quad ( 1 6 )$$
is very large. Thus, the Γ -1v band is strongly repelled downward along the k y direction, giving a dispersion
$$E _ { 1 v } ( k _ { y } ) \approx - E _ { 2 } - \frac { P _ { y 3 } ^ { 2 } } { E _ { 1 } + E _ { g } + E _ { 2 } } k _ { y } ^ { 2 }, \quad ( 1 7 )$$
where E 2 is the energy difference from Γ + 2v to Γ -1v . Both the free electron dispersion and the effect from Γ + 2v [Eq. (15)] are neglected in this expression, due to their minor contributions compared with the influence
of P y 3 . As k y increases, the energy difference between Γ + 2v and Γ -1v therefore quickly grows, further diminishing their coupling.
An additional interaction is still needed to induce a hole-like negative effective mass for Γ + 2v to preserve the open bandgap. This role is played by conduction band states with Γ -1 symmetry, as shown in Fig. 5 by the matrix element
$$P _ { y 4 } = \frac { \hbar } { m _ { 0 } } \langle \Gamma _ { 1 c } ^ { - } | \hat { p } _ { y } | \Gamma _ { 2 \nu } ^ { + } \rangle. \quad \quad ( 1 8 ) \quad \stackrel { \dots } { y - \Gamma } \\ ( \Gamma _ { 4 } ^ { - } )$$
Under the basis functions of { Γ -1c , Γ + 2c , Γ + 2v , Γ -1v } , one could construct a 4 × 4 Hamiltonian similar to Eq. (7), including all important k p y ˆ y interactions affecting the valence band. However, just from the Hamiltonian matrix element analysis in Eqs.(15)-(18), the application of L¨ owdin partitioning already gives the unusual valence band dispersion in the k y direction approximated by
$$E _ { h } ( k _ { y } ) = \frac { \hslash ^ { 2 } } { 2 m _ { 0 } } k _ { y } ^ { 2 } - \frac { P _ { y 4 } ^ { 2 } } { E _ { 3 } } k _ { y } ^ { 2 } - \frac { P _ { y 2 } ^ { 2 } k _ { y } ^ { 2 } } { E _ { 1 v } ( k _ { y } ) }, \quad ( 1 9 ) \quad \begin{pmatrix} \text{in} \\ \text{spin} \end{pmatrix}$$
where E 3 is the energy difference from Γ -1c to Γ + 2v . Note that, in the denominator of the last term, we use the complete dispersion of the Γ -1v band [Eq. (17)] which is negative and depends on k y , rather than E 2 , the fixed energy difference between Γ -1v and Γ + 2v .
glyph[negationslash]
E 2 2 m 0 2
Eq. (19) guarantees a hole-like dispersion if the second term (repulsion from Γ -1c ) overcomes the first (free electron) term. The ultraflat valence band along k y is thus due to the counteracting effects of Γ -1v and Γ -1c close to the zone center; phosphorene is indirect-gap ( k 0 = 0) when the matrix element magnitudes satisfy P 2 y 2 + glyph[planckover2pi1] 2 >
P y 4 E 3 > glyph[planckover2pi1] 2 2 m 0 . Although this condition is predicted by DFT, other indications of inaccuracies or variations in that numerical method (for example, gross underestimation of the bandgap or the details of relaxed structure) 36 suggest that we cannot exclude the possibility that repulsion from Γ -1c is so strong that P 2 y 4 E 3 > P 2 y 2 E 2 + glyph[planckover2pi1] 2 2 m 0 , resulting in a direct gap. The true nature of the valence band must therefore be revealed by experiment.
## D. Optical selection rules
The interaction between matter and radiation e m A ˆ p · has the same symmetry as H k ˆ · p (the electric field of light transforms like an ordinary polar vector). Therefore, we can use the perturbative coupling superimposed as arrows on Fig. 5 to reveal the selection rules of optical transitions between conduction and valence bands in phosphorene.
The dominant transition paths are represented by the two red double arrows. When the electric field polarization is parallel to the x -direction, photons with glyph[planckover2pi1] ω ≥ E g can cause excitation across the bandgap from Γ + 2v to
Γ -4c . On the other hand, photons with orthogonal polarization parallel to the y -direction can excite lower valence band Γ -1v electrons to the upper conduction band Γ + 2c , which requires a frequency deep into the ultraviolet regime ( E g + E 1 + E 2 ∼ 5 eV photon energy). In addition to the bands shown in Fig. (5) that are relevant to the bandedge dispersion, we note that there are also a Γ + 4v band between Γ + 2v and Γ -1v , as well as a Γ -3c band in between Γ + 2c and Γ + 3c . Optical transition with y -polarization is also allowed between these two states (Γ + 4v ⊗ Γ 3c -= Γ -2 ), which are seperated by ∼ 5 eV but extend along the Γ -X axis and reach a minimum energy separation at the location of a valence band satellite valley. This could explain the ∼ 3 7 eV onset photon en-. ergy in the calculated absorption spectrum of y -polarized light. 30
The sensitivity of the absorption spectrum to electric field orientation, confirmed by DFT calculation, 26 may make phosphorene an ideal photoconductive polarimeter in near-infrared frequencies.
In anticipation of the next section where we consider spin properties, it is appropriate to point out in the present discussion of radiative selection rules that optical orientation in phosphorene is hopelessly inefficient: Unlike in bulk semiconductors with cubic symmetry, where degeneracy of the p -like valence band is preserved in the spin-independent Hamiltonian allowing strong mixing of spin and orbital angular momentum, here with inversionsymmetric but orthorhombic phosphorene we have only nondegenerate and energetically well-separated bands. Spin-orbit interaction can thus only have a small perturbative mixing of spin states. Illumination with linear polarization almost entirely preserves spin during absorptive transitions, so that excitation of the spin-unpolarized valence band will result only in population of a nearly spin-unpolarized conduction band.
## IV. SPIN-DEPENDENT PROPERTIES
The spin-orbit interaction (SOI) in semiconductors composed of light atoms is expected to be weak, resulting in relatively pure spin states and long spin lifetime in the intrinsic regime. By this reasoning, SOI in phosphorus allotropes should resemble that of its neighbor on the periodic table, silicon, an indirect-gap group-IV bulk semiconductor that has a long spin lifetime for electrons in the conduction band 37-39 and a small split-off energy (44 meV) in the valence band. The latter feature of the bandstructure is a result of the SOI perturbing an otherwise three-fold degenerate p-like valence band extremum. Phosphorene, on the other hand, has only non-degenerate and energy-isolated bands (other than the few accidental crossings). When SOI is included in the calculation, we therefore expect only that some of the bands will be unnoticeably shifted, as confirmed by DFT calculation. 26
Although SOI has a negligible effect on the band structure (Hamiltonian eigenvalues) in phosphorene, it is still
worthwhile to examine the effect it has on spin-dependent eigenstates, especially because the in-plane anisotropy is expected to extend to spin-related phenomena such as relaxation mechanisms. In this section, we will now derive the spin-dependent Hamiltonian including H SO and H SO , k , again using the method of invariants, and calculate the spin-dependent eigenstates of electrons and holes that capture the underlying symmetries.
Before starting our discussion, we compare some other spin-related properties between phosphorene and silicon, the latter of which is believed to be a promising material candidate for spintronic devices. 40 Besides the low atomic number already mentioned, there are two important factors leading to the rather long spin lifetime in Si. One is the centrosymmetric property of the diamond lattice that (along with Kramers' time-reversal symmetry) preserves spin degeneracy, and precludes the DyakonovPerel spin relaxation mechanism where momentum scattering causes a fluctuating effective magnetic field driving spin flips. 41 The other is the absence of nuclear spin in the most abundant isotope 28 Si, so that there is little hyperfine interaction affecting the electron spin states.
Whereas the high abundance of 31 P isotope (with halfinteger nuclear spin) makes hyperfine interaction in phosphorene worth considering, it is outside the scope of this paper. However, phosphorene is indeed centrosymmetric (see Table I), and so like Si, Dyakonov-Perel spin relaxation is absent. The dominant spin relaxation mechanism in both materials is then the Elliott and Yafet processes.
Generally speaking, perturbative spin-orbit coupling between different bands causes a mixing of pure Pauli spinors so that the two degenerate spin states within the same i th band can be written as glyph[negationslash]
$$| i \uparrow \rangle = C _ { i } \left ( | i \uparrow \rangle + \sum _ { j \neq i } a _ { j } | j \uparrow \rangle + b _ { j } | j \downarrow \rangle \right ), \text{ and } \quad ( 2 0 ) \quad \text{where} \quad \delta$$
glyph[negationslash]
$$| i \Downarrow \rangle = & C _ { i } \left ( | i \downarrow \rangle + \sum _ { j \neq i } a _ { j } ^ { * } | j \downarrow \rangle - b _ { j } ^ { * } | j \uparrow \rangle \right ), \quad \quad ( 2 1 ) \quad \text{ while } \text{ band}.$$
where C i is a normalization factor. The dominant orbital component in both | i ⇑〉 and | i ⇓〉 is | i 〉 , whose amplitude is very close to unity. However, the coefficients | b j | glyph[lessdblequal] | a j | glyph[lessmuch] 1 depending on their perturbation origins. In our following discussion, we ignore the prefactor C i ≈ 1, and for clarity, write the spin-dependent eigenstates in the form of 2-row matrices such as
$$| i \uparrow \rangle = \begin{pmatrix} a _ { 1 } & a _ { 2 } & \dots & 1 & \dots \\ b _ { 1 } & b _ { 2 } & \dots & 0 & \dots \end{pmatrix} \,, \text{ and } \quad \ \ ( 2 2 ) \quad \ \ \ \ \ t h e$$
$$| i \Downarrow \rangle = \begin{pmatrix} - b _ { 1 } ^ { * } & - b _ { 2 } ^ { * } & \dots & 0 & \dots \\ a _ { 1 } ^ { * } & a _ { 2 } ^ { * } & \dots & 1 & \dots \end{pmatrix} \,, \quad \quad \ ( 2 3 ) \quad \text{$usu$} \quad \text{$who$} \quad \text{$dne}$$
where quantities in the top (bottom) row indicate coefficients of |↑〉 ( |↓〉 ) states, and each column corresponds to a given band.
Spin flips are induced by an interaction 〈⇑| Ξ |⇓〉 , where Ξ is a scattering potential. The spin-independent part of this potential, normally leading to momentum scattering, will then couple the same spin components of |⇑〉 and |⇓〉 , resulting in the Elliot mechanism 42 , whereas the spin-dependent part of the scatterer will couple opposite spin components and give rise to the Yafet term 43 . These two quantities coherently interfere to yield the total spin relaxation rate τ -1 s , proportional to the square of the matrix element according to Fermi's golden rule. A major goal of the following subsections is to examine the symmetry of these processes and understand the spin-relaxation anisotropy in terms of the spin-dependent eigenstate components.
## A. Hamiltonian and spin-dependent eigenstates
The incorporation of spin-orbit interactions [Eqs. (5) and (6)] within the framework of the method of invariants is as straightforward as our treatment of H k ˆ · p . We start with a discussion of the valence band Hamiltonian matrix of the k -independent SOI ( H SO ). From Table II, one finds that the three invariant components of H SO belong to the IRs of Γ + 2 , Γ + 3 and Γ + 4 , which couples the highest valence band Γ + 2v to remote bands belong to Γ + 1 , Γ + 4 and Γ + 3 , respectively. Therefore, in a basis { Γ + 3 , Γ + 4 , Γ + 1 , Γ + 2v } , the matrix form of H SO can be written
$$\iismsup _ { \substack { \text{ses.} \\ \text{using} \\ \text{aux} } } \quad H _ { \text{SO} } ^ { \text{hole} } = \begin{pmatrix} & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & & \\ & \\ & & \\ & \\ & & \\ & \\ & \\ & \\ & \\ & \\ & \\ & \\ & \\ & \\ & \\ & \\ & \\ \\ & \\ & \\ \\ & \\ \\ & \\ \\ & \\ \\
\end{pmatrix}, \ ( 2 4 )$$
where
$$\delta _ { l 2 } = \sum _ { \Gamma _ { l } ^ { + } } \frac { i \hbar } { 4 m _ { 0 } ^ { 2 } c ^ { 2 } } \langle \Gamma _ { l } ^ { + } | \frac { \partial V } { \partial x _ { m } } \hat { p } _ { n } - \frac { \partial V } { \partial x _ { n } } \hat { p } _ { m } | \Gamma _ { 2 \nu } ^ { + } \rangle, \quad ( 2 5 )$$
which are real numbers taking into account all remote bands belonging to the IR Γ + l . Here, { lmn } is a cyclic permutation of { 1,2,3 } and { x 1 , x 2 , x 3 } correspond to { x , y , z } . By expanding the Pauli matrices, H hole SO in Eq. (24) is an 8 × 8 matrix, where each basis function is the direct product of a spin-independent wave function and a spinor ( |↑〉 or |↓〉 , eigenstates of σ z ). Note that the lack of matrix elements other than those involving Γ + 2v in Eq. (24) does not imply that they are zero; rather, we focus here only on the matrix elements that are important in determining the valence band eigenvectors to lowest order.
H SO , k in Eq. (6) gives a small correction to H SO that is usually negligible, since | glyph[planckover2pi1] k | is comparable with | p | only when k reaches the zone edge. In phosphorene, however, due to the unusually flat valence band, k y of hole states can be relatively large ( glyph[greaterorsimilar] 10% of π/a y ), and the resulting H SO , k cannot be ignored. The components of H SO , k transform like polar vectors as shown in Table II and for simplicity we only keep the more important k y -related terms. Applying the same procedure as above, we obtain
the matrix form of H SO , k for hole states as
$$H _ { \text{SO}, \mathbf k } ^ { \text{hole} } = \begin{pmatrix} \frac { i \alpha _ { 4 2 } k _ { y } \sigma _ { z } \ - i \alpha _ { 3 2 } k _ { y } \sigma _ { x } } { - i \alpha _ { 4 2 } k _ { y } \sigma _ { z } } \\ i \alpha _ { 3 2 } k _ { y } \sigma _ { x } \end{pmatrix}, \ \ ( 2 6 ) \quad \text{before} \\ \quad \text{have} \ \ 1 \\ \quad \text{basis} \ \ 1$$
which is a 6 × 6 matrix in the basis { Γ + 2v , Γ -4 , Γ -3 }⊗{↑ , ↓} . Again, we are only interested in the matrix elements that couple to Γ + 2v in lowest order. The α parameters in Eq. (26) are
$$\alpha _ { 3 2 } = \sum _ { \Gamma _ { 3 } ^ { - } } \frac { i \hslash ^ { 2 } } { 4 m _ { 0 } ^ { 2 } c ^ { 2 } } \langle \Gamma _ { 3 } ^ { - } | \frac { \partial V } { \partial z } | \Gamma _ { 2 \nu } ^ { + } \rangle, \text{ and } \quad ( 2 7 ) \quad \text{ enter} \\.. \circ \quad \widetilde {.. } \ \widetilde {.. } \$$
$$\alpha _ { 4 2 } = \sum _ { \Gamma _ { 4 } ^ { - } } ^ { \kappa } \frac { i \hbar { ^ { 2 } } } { 4 m _ { 0 } ^ { 2 } c ^ { 2 } } \langle \Gamma _ { 4 } ^ { - } | \frac { \partial V } { \partial x } | \Gamma _ { 2 \nu } ^ { + } \rangle. \quad \ \ ( 2 8 ) \begin{array} { c } \stackrel { \bullet \circ } { \text{to s} } \\ \text{orth} \\ \text{the} \\ \end{array} \end{array}$$
The essential spin-orbit interaction of hole states is now captured by H hole SO ⊕ H hole SO , k , reducible to a 12 × 12 matrix because of the redundant Γ + 2v . In combination with the spin-independent H 0 and H k ˆ · p operators discussed in the previous section, one can diagonalize the total Hamiltonian and obtain the spin-dependent eigenstates for holes.
Several simplifying approximations can be made. Because it appears in the same matrix element and is much smaller than the P x 1 k x terms in Eq. (7), α 42 k σ y z in Eq. (26) can reasonably be ignored. In addition, the P k y y terms (coupling between Γ + 2v and Γ -1 bands) can be neglected for two reasons: ( i. ) the amplitudes of P y 2 and P y 4 are small; and ( ii. ) the effect from Γ -1 bands below and above Γ + 2v counteract each other near the Γ-point (reflected by the flat band there).
To avoid lengthy summations and energy denominators, we define the following quantities related to the δ , α and P parameters:
$$\Delta _ { l 2 } = \sum _ { \Gamma _ { l } ^ { + } } \frac { i \hbar { \, } } { 4 m _ { 0 } ^ { 2 } c ^ { 2 } } \frac { \langle \Gamma _ { l } ^ { + } | \frac { \partial V } { \partial x _ { m } } \hat { p } _ { n } - \frac { \partial V } { \partial x _ { n } } \hat { p } _ { m } | \Gamma _ { 2 \nu } ^ { + } \rangle } { E _ { \Gamma _ { l } ^ { + } } - E _ { \Gamma _ { 2 \nu } ^ { + } } }, \quad ( 2 9 ) \quad \begin{pmatrix} \Gamma \log \cdot \\ \text{the s} \\ \text{and} \end{pmatrix} \\ \psi \quad. \quad \partial \omega \quad \dots$$
$$A _ { 3 2 } k _ { y } = \sum _ { \substack { \Gamma _ { 3 } ^ { - } \\ \sim \cdot } } \frac { i \hslash ^ { 2 } } { 4 m _ { 0 } ^ { 2 } c ^ { 2 } } \frac { \langle \Gamma _ { 3 } ^ { - } | \frac { \partial V } { \partial z } | \Gamma _ { 2 \nu } ^ { + } \rangle k _ { y } } { E _ { \Gamma _ { 3 } ^ { - } } - E _ { \Gamma _ { 2 \nu } ^ { + } } }, \text{ and } \quad ( 3 0 ) \quad \Delta _ { 1 2 } ) \quad \frac { \partial B ( 1 ) } { \partial z } \, \mathbb { H } _ { \substack { \partial V \\ \text{bhor} } }$$
$$\Pi _ { x 1 } k _ { x } & = \frac { P _ { x 1 } k _ { x } } { E _ { g } }. & ( 3 1 ) & \quad \text{phor} \\ & \quad \text{oppc}$$
Because of their origins in the H k ˆ · p glyph[greatermuch] H SO glyph[greatermuch] H SO , k perturbation terms, the hierarchy between these unitless parameters is A 32 k y glyph[lessmuch] ∆ l 2 glyph[lessmuch] Π x 1 k x glyph[lessmuch] 1.
Each spin-dependent eigenvector includes 12 coefficients corresponding to the 12 basis functions which span the subspace of six IRs { Γ + 3 , Γ + 4 , Γ + 1 , Γ + 2v , Γ -4c , Γ -3 } and the two spinors. We write the spin-dependent eigenvectors in the form of Eqs. (22) and (23) giving 2 × 6 matrices:
$$| h \uparrow _ { \perp } \rangle = \left ( \begin{array} { c c c } i \Delta _ { 3 2 } & 0 & 0 & 1 & \Pi _ { x 1 } k _ { x } & 0 \\ 0 & i \Delta _ { 4 2 } & \Delta _ { 1 2 } & 0 & 0 & - i A _ { 3 2 } k _ { y } \end{array} \right ) \,, \text{ and } \quad \text{ term } i \\. \quad \text{ not ffi}$$
$$| \mathbf h \Downarrow _ { \perp } \rangle = \begin{pmatrix} 0 & i \Delta _ { 4 2 } & - \Delta _ { 1 2 } & 0 & 0 & - i A _ { 3 2 } k _ { y } \\ - i \Delta _ { 3 2 } & 0 & 0 & 1 & \Pi _ { x 1 } k _ { x } & 0 \end{pmatrix} \,. \ \ ( 3 3 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
Here 'h' stands for 'hole' and the subscript ' ⊥ ' indicates that the spin orientation z is out of plane. Notice that, before the trivial normalization, the dominant coefficients have the value of 1, corresponding to the | Γ + 2v ↑〉 ( Γ | + 2v ↓〉 ) basis function in | h ⇑ ⊥ 〉 ( h | ⇓ ⊥ 〉 ).
Eqs. (32) and (33) explicitly indicate the spin-purity of the eigenstates. One can evaluate the total square amplitude of the minority-spin components (the so-called 'spin mixing' coefficient), which is approximately ∆ 2 42 + ∆ 2 12 (the A 2 42 k 2 y term is a small correction). In Eq. (29), the energy denominators are several eV or more, while the remainder are within the same order of the δ l 2 parameters in Eq. (25). It is known that the dominant contribution to spin-orbit coupling is from the part of wavefunctions orthogonal to the core states, which are populated in the vicinity of the nucleus where the atomic potential changes drastically. 44,45 As in Si, these core electrons are 2 p states, so we can estimate using the same parameters: δ l 2 ∼ meV. 38 This value then gives ∆ l 2 ∼ 10 -3 , so the spin mixing coefficient for holes in phosphorene should be similar to that of conduction band electrons in Si, which is approximately 10 -5 -10 -6 .
Eqs. (32) and (33) are fundamental in helping us understand the spin-flip process, in which a transition occurs between | h ⇑ ⊥ 〉 and | h ⇓ ⊥ 〉 during momentum scattering. As an example, we analyze the symmetry of the Elliott-Yafet (EY) spin-flipping process due to scattering by smallk long-wavelength acoustic phonons (in-phase quasi-uniform vibration of atoms). This is the dominant spin relaxation mechanism that limits the intrinsic spinlifetime in such a centrosymmetric system at finite temperature.
The Elliott term 42 proportional to ∇ V has the same symmetry of a polar vector or the p orbitals of Γ -2 in Fig. 2 , Γ -3 in Fig. 3(a) and Γ -4 in Fig. 3(b). It couples the same spin components between | h ⇑ ⊥ 〉 and | h ⇓ ⊥ 〉 . Eqs. (32) and (33) show that this happens between Γ + 2v and Γ -3 (with the coefficients 1 and -A 32 k y ), as well as between Γ -4c and Γ + 1 (with the coefficients Π x 1 k x and ∆ 12 ), and the responsible phonon mode is the Γ -4 -related ∂V ∂z Elliott term (out-of-plane motion of atoms, or flexural phonons).
On the other hand, the Yafet operator that couples opposite spin components 43 between | h ⇑ ⊥ 〉 and | h ⇓ ⊥ 〉 is proportional to ∇ H SO , which is the gradient of an axial vector with symmetry
$$& ( \Gamma _ { 4 } ^ { - } \oplus \Gamma _ { 3 } ^ { - } \oplus \Gamma _ { 2 } ^ { - } ) \otimes ( \Gamma _ { 4 } ^ { + } \oplus \Gamma _ { 3 } ^ { + } \oplus \Gamma _ { 2 } ^ { + } ) \\ = & 3 \Gamma _ { 1 } ^ { - } \oplus 2 \Gamma _ { 2 } ^ { - } \oplus 2 \Gamma _ { 3 } ^ { - } \oplus 2 \Gamma _ { 4 } ^ { - }.$$
Examining Eqs. (32) and (33), one sees that the dominant coupling is between Γ + 2v and Γ -4c by one of the two Γ -3 IRs in Eq. (34) that is related to ∂ ∂z ( ∂V ∂z p x -∂V ∂x p z ) σ y . This term also corresponds to flexural phonons [the other term belonging to Γ -3 IR is ∂ ∂y ( ∂V ∂x p y -∂V ∂y p x ) σ z , which does not flip the spin].
The same formalism can be applied to study the spindependent conduction band electrons. Here, we list the
expressions for Hamiltonian matrices and spin-dependent eigenstates of the lowest eigenvalue (conduction band minimum). In a basis { Γ -3 , Γ -2 , Γ -1 , Γ -4c } , H SO for Γ -4c electrons is
$$H _ { \text{SO} } ^ { \text{electron} } = \begin{pmatrix} & & \text{rectio} \\ & & \text{i} \delta _ { 3 4 } \sigma _ { y } \\ & & \text{i} \delta _ { 2 4 } \sigma _ { x } & - i \delta _ { 1 4 } \sigma _ { z } \end{pmatrix}, & & \text{int} \\ & & \text{depend} \\ & & \text{orient} \\ & & \text{most} \\ & & \text{fixed} \\ & & & \text{contacts}. \end{pmatrix}$$
where
$$\delta _ { l 4 } = \sum _ { \Gamma _ { l } ^ { - } } \frac { \hslash ^ { 2 } } { 4 m _ { 0 } ^ { 2 } c ^ { 2 } } ( \Gamma _ { l } ^ { - } | \frac { \partial V } { \partial x _ { m } } \hat { p } _ { n } - \frac { \partial V } { \partial x _ { n } } \hat { p } _ { m } | \Gamma _ { 4 c } ^ { - } ). \quad ( 3 6 ) \quad \text{level} \\ \text{plane} \\ \text{invari}$$
As in Eq. (26) for holes, the H SO , k matrix for electrons in the k y direction, using the basis { Γ -4c , Γ + 2 , Γ + 1 } , reads
$$H _ { \text{SOL} } ^ { \text{electron} } = \begin{pmatrix} \frac { i \alpha _ { 2 4 } k _ { y } \sigma _ { z } \ - i \alpha _ { 1 4 } k _ { y } \sigma _ { x } } { - i \alpha _ { 2 4 } k _ { y } \sigma _ { z } } \\ i \alpha _ { 1 4 } k _ { y } \sigma _ { x } \end{pmatrix}, \quad \begin{pmatrix} \text{r} \, \text{ir} \\ \text{hole} \, \, \,$$
where the α parameters are
$$\alpha _ { 2 4 } = \sum _ { \Gamma _ { 2 } ^ { + } } \frac { i \hslash ^ { 2 } } { 4 m _ { 0 } ^ { 2 } c ^ { 2 } } \langle \Gamma _ { 2 } ^ { + } | \frac { \partial V } { \partial z } | \Gamma _ { 4 c } ^ { - } \rangle, \text{ and } \quad ( 3 8 ) \quad \text{char} \\ \dots \theta \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots$$
$$\alpha _ { 1 4 } = \sum _ { \Gamma _ { 1 } ^ { + } } \frac { i \hbar { ^ { 2 } } } { 4 m _ { 0 } ^ { 2 } c ^ { 2 } } \langle \Gamma _ { 1 } ^ { + } | \frac { \partial V } { \partial x } | \Gamma _ { 4 c } ^ { - } \rangle. \quad \ ( 3 9 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
As in Eqs. (29) and (30), we define
$$\Delta _ { l 4 } = \sum _ { \Gamma _ { l } ^ { - } } \frac { i \hbar } { 4 m _ { 0 } ^ { 2 } c ^ { 2 } } \frac { \langle \Gamma _ { l } ^ { + } | \frac { \partial V } { \partial x _ { m } } \hat { p } _ { n } - \frac { \partial V } { \partial x _ { n } } \hat { p } _ { m } | \Gamma _ { 4 c } ^ { - } \rangle } { E _ { \Gamma _ { l } ^ { - } } - E _ { \Gamma _ {$$
$$A _ { 1 4 } k _ { y } = \sum _ { \Gamma _ { 1 } ^ { + } } \frac { i \hslash ^ { 2 } } { 4 m _ { 0 } ^ { 2 } c ^ { 2 } } \frac { \langle \Gamma _ { 1 } ^ { + } | \frac { \partial V } { \partial z } | \Gamma _ { 4 c } ^ { - } \rangle k _ { y } } { E _ { \Gamma _ { 1 } ^ { + } } - E _ { \Gamma _ { 4 c } ^ { - } } }. \quad \quad ( 4 1 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
The resulting spin-dependent electron eigenstates in the basis { Γ -3 , Γ -2 , Γ -1 , Γ -4c , Γ + 2v , Γ + 1 } are
$$| e \uparrow _ { \perp } \rangle = & \left ( \begin{array} { c c c c } 0 & 0 & i \Delta _ { 1 4 } & 1 & \Pi _ { x 1 } k _ { x } & 0 \\ \Delta _ { 3 4 } & i \Delta _ { 2 4 } & 0 & 0 & 0 & - i A _ { 1 4 } k _ { y } \end{array} \right ) \,, \text{ and } \quad \text{ Here,} \\ \quad \quad ( 4 2 ) & \text{cases} \quad \text{cases} \quad \text{$\quad$} \quad \text{$\quad$} \quad \text{$\quad$}.$$
$$| e \Downarrow _ { \perp } \rangle = \begin{pmatrix} - \Delta _ { 3 4 } & i \Delta _ { 2 4 } & 0 & 0 & 0 & - i A _ { 1 4 } k _ { y } \\ 0 & 0 & - i \Delta _ { 1 4 } & 1 & \Pi _ { x 1 } k _ { x } & 0 \end{pmatrix} \,. \ ( 4 3 ) \quad \text{$\Pi_{x1} d$} \\ \quad \text{$\label} \text{li} \\ \Delta _ { \dots } \text{i\nu}$$
## B. Anisotropy of spin-dependent properties
We naturally expect that the orthorhombic inequivalence of the three orthogonal armchair, zigzag, and out-of-plane axes will induce anisotropy of the spindependent properties of electrons and holes. Already from our analysis of the band dispersion, we can see an obvious anisotropy in the spin-diffusion length λ s =
√ τ D s , where the carrier diffusion coefficient D is related to the anisotropic effective mass. However, there is an additional contribution: the spin lifetime τ s also has an anisotropy, related in this case not to the wavevector direction, but rather to the spin orientation.
In the previous subsection, we have derived the spindependent Hamiltonian and eigenstates with the spin orientation in the out-of-plane z -direction. However, in most spin-injection experiments, the spin orientation is fixed by the magnetization of ferromagnetic thin-film contacts with an in-plane easy axis. It is therefore more relevant to study the spin-dependent eigenstates and related properties with the spin quantization axis z inplane. Of course, the spatial symmetry of the system is invariant regardless of the coordinate system we choose, and so are the properties of IRs and the relations between them (Table III); only the coordinate labels change.
First, we calculate the spin-dependent properties of hole states under a new coordinate system where z is chosen to be along the armchair, x along the zigzag and y along the out-of-plane direction. We apply the cyclic permutation xyz → zxy to the coordinates in Fig. 1(a), and the corresponding invariant components in Table II (note that one could also keep the coordinate system unchanged but alternatively derive spin-dependnet eigenstates from linear combination of | h ⇑ ⊥ 〉 and | h ⇓ ⊥ 〉 , and | e ⇑ ⊥ 〉 and | e ⇓ ⊥ 〉 , according to the chosen spin orientation). The spin-dependent Hamiltonian can be derived straightforwardly, so we do not repeat the previous procedures but rather give the result of the spin-dependent eigenstates.
Like Eqs. (32) and (33), the spin-dependent eigenvectors for valence-band holes under the new coordinate system also includes 12 coefficients corresponding to the 12 basis functions which expand the subspace of six IRs { Γ + 3 , Γ + 4 , Γ + 1 , Γ + 2v , Γ -4c , Γ -3 } and the two spinors. In the form of Eqs. (22) and (23), they read
$$\int _ { \substack { | h \uparrow _ { a } \rangle = \begin{pmatrix} 0 & i \Delta _ { 4 2 } & 0 & 1 & \Pi _ { z 1 k _ { z } } & - i A _ { 3 2 k _ { x } } \\ \Delta _ { 3 2 } & 0 & i \Delta _ { 1 2 } & 0 & A _ { 4 2 k _ { x } } & 0 \end{pmatrix} }, \text{ and } \\ \int the \begin{pmatrix} 0 & i \Delta _ { 4 2 } & 0 & 1 & \Pi _ { z 1 k _ { z } } & - i A _ { 3 2 k _ { x } } \\ \Delta _ { 3 2 } & 0 & A _ { 4 2 k _ { x } } & 0 \end{pmatrix}, \text{ and } \\ \int \, \text{ and }$$
$$\xi _ { 1 } \left | h \Downarrow _ { \text{ac} } \right \rangle = \left ( \begin{matrix} - \Delta _ { 3 2 } & 0 & i \Delta _ { 1 2 } & 0 & - A _ { 4 2 } k _ { x } & 0 \\ 0 & - i \Delta _ { 4 2 } & 0 & 1 & \Pi _ { z 1 } k _ { z } & i A _ { 3 2 } k _ { x } \end{matrix} \right ) \,. \ ( 4 5 )$$
Here, the subscript 'ac' stands for 'armchair' and indicates the spin orientation. Π z 1 has the same value as Π x 1 defined by Eq. (31) since we have only changed the labeling from x to z . Also, similar to the definition of A 32 in Eq. (30), we have
$$A _ { 4 2 } k _ { y } = \sum _ { \Gamma _ { 4 } ^ { - } } \frac { i \hslash ^ { 2 } } { 4 m _ { 0 } ^ { 2 } c ^ { 2 } } \frac { \langle \Gamma _ { 4 } ^ { - } | \frac { \partial V } { \partial z } | \Gamma _ { 2 \nu } ^ { + } \rangle k _ { y } } { E _ { \Gamma _ { 4 } ^ { - } } - E _ { \Gamma _ { 2 \nu } ^ { + } } }. \quad ( 4 6 )$$
Eqs. (44) and (45) directly show that spin-mixing to lowest order is ∆ 2 32 +∆ 2 12 , different from the previous case (∆ 2 42 +∆ ) when the spin orientation 2 12 z is out-of-plane. It will be shown that if z is along the zigzag direction, the spin mixing is ∆ 2 32 +∆ 2 42 [see Eqs. (47) and 48)]. The
anisotropy of the spin-mixing is generally reflected in the definition of Eq. (29) by the energy denominators, while the numerators scale similarly, due to their major origins related to the 2 p core states, as previously discussed. 44,45 Numerical calculation shows that, for the Γ + 2v hole band, the closest Γ + 3 is the fourth conduction band approximately 3.5 eV above, while the closest neighboring Γ + 4 and Γ + 1 bands are the second and third valence bands (not shown in Fig. 5) that are approximately 1.8 eV below. We therefore estimate that the spin mixing with z in-plane is approximately half of that with z out-ofplane, and the Elliott-process-limited spin lifetime is consequently ∼ 4 times longer.
A more intriguing feature of the anisotropy is that of the phonon polarization in spin-flip scattering. Again, we take the interaction with long wavelength acoustic phonons as an example. In contrast to the case when z is out-of-plane, here the coupling of the dominant Γ + 2v band by the Elliott operator to Γ -3 vanishes, since the coefficient for Γ -3 ↑ ( ↓ ) in | h ⇓ ac ( ⇑ ac ) 〉 is zero. However, the coupling between Γ + 2v and Γ -4 survives, with the coefficient -A 42 k y . More importantly, the interaction belongs to Γ -3 = Γ + 2v ⊗ Γ -4 , which still corresponds to ∂V ∂z , but is now related to in-plane, instead of flexural, phonons.
It is well-known that, due to their quadratic dispersion relation, the thermal population of flexural phonons in 2D materials diverges when the phonon wavevector q approaches zero. Such a singularity does not exist in the case of in-plane phonons with linear dispersion. The intrinsic spin lifetime in 2D materials is thus essentially limited by the interaction with flexural phonons while the influence from in-plane phonons is much less crucial. 4 However, here in phosphorene, for spins oriented in the armchair direction the Elliott spin-flip coupling of the dominant Γ + 2v component via flexural phonons is excluded by symmetry.
We have seen that scattering by acoustic phonons ultimately determines the upper bound of the spin lifetime in phosphorene. However, this intrinsic mechanism will be superseded by various extrinsic spin relaxation mechanisms in anything other than the most pure, well-isolated samples. Carrier spins in 2D materials are especially sensitive to extrinsic effects such as interactions with substrate and contacts, scattering with impurities or defects, and the influence of deformation such as nanoripples and strain. As long as these spin-flip processes are within the perturbative regime, the spin-dependent eigenvectors can be effectively used to determine whether a certain process is symmetry-allowed, as well as its orientation dependence. The strength of the interaction can be evaluated by the coefficients in these eigenvectors, calculated with the assistance of numerical schemes able to yield the coupling amplitude between basis functions.
Before closing this section, we derive expressions for the spin-dependent hole eigenvectors for the remaining orientation with z along the zigzag direction by applying the permutation xyz → yzx to the table of invariants. Using a basis { Γ + 3 , Γ + 4 , Γ + 1 , Γ + 2v , Γ -4c , Γ -3 } , the eigenvec- tors read
$$& \text{while} \\ & \underset { 4, 4 5 } { \text{$line} } \\ & \text{$4,45$} \\ & \text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text
$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{ $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text}$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text#\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{#\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test}$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test
$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test.$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text`$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text'$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$$
$$\frac { \infty } { \Gamma _ { 4 } ^ { + } } \quad | h \Downarrow _ { z z } \rangle = \begin{pmatrix} i \Delta _ { 3 2 } - \Delta _ { 4 2 } & 0 & 0 & i A _ { 4 2 } k _ { z } & A _ { 3 2 } k _ { z } \\ 0 & 0 & - i \Delta _ { 1 2 } & 1 & \Pi _ { y 1 } k _ { y } & 0 \end{pmatrix}.$$
The subscript 'zz' stands for 'zigzag'. Also, we give the spin-dependent eigenvectors of the conduction-band electron states for the spin orientation aligned with both inplane axes, in a basis { Γ -3 , Γ -2 , Γ -1 , Γ -4c , Γ + 2v , Γ + 1 } . For z along the armchair direction, the eigenvectors are
$$\underset { \substack { \text{sign} \\ \text{astic} \\ \Gamma _ { 2 \nu } ^ { + } } } { \text{sign} } \quad | e \uparrow _ { a c } \right \rangle = \left ( \begin{array} { c c c } 0 & i \Delta _ { 2 4 } & 0 & 1 & \Pi _ { z 1 } k _ { z } & - i A _ { 1 4 } k _ { x } \\ i \Delta _ { 3 4 } & 0 & \Delta _ { 1 4 } & 0 & - A _ { 2 4 } k _ { x } & 0 \end{array} \right ) \,, \text{ and }$$
$$\stackrel { \ e v } { c o - } } _ { \text{$c o -} } \left | e \Downarrow _ { a c } \right \rangle = \begin{pmatrix} i \Delta _ { 3 4 } & 0 & - \Delta _ { 1 4 } & 0 & A _ { 2 4 } k _ { x } & 0 \\ 0 & - i \Delta _ { 2 4 } & 0 & 1 & \Pi _ { x 1 } k _ { x } & i A _ { 1 4 } k _ { y } \end{pmatrix} \,. \ ( 5 0 )$$
For z along the zigzag direction, they are
$$\begin{smallmatrix} \text{ut is} \\ 3. \\ \text{per-} \\ \text{nons} \end{smallmatrix} \quad | e \uparrow _ { z z } \rangle = \left ( \begin{smallmatrix} i \Delta _ { 3 4 } & 0 & 0 & 1 & \Pi _ { y 1 } k _ { y } & 0 \\ 0 & \Delta _ { 2 4 } & i \Delta _ { 1 4 } & 0 & i A _ { 2 4 } k _ { z } & A _ { 1 4 } k _ { z } \end{smallmatrix} \right ) \,, \text{ and} \\ \quad.$$
$$\underset { \substack { \text{ctor} \\ \text{exit} \\ \text{--i} \dots } } { \text{ctor} } \quad | e \Downarrow _ { z z } \rangle = \begin{pmatrix} 0 & - \Delta _ { 2 4 } & i \Delta _ { 1 4 } & 0 & i A _ { 2 4 } k _ { z } & - A _ { 1 4 } k _ { z } \\ - i \Delta _ { 3 4 } & 0 & 0 & 1 & \Pi _ { y 1 } k _ { y } & 0 \end{pmatrix}. \ ( 5 2 )$$
## V. SUMMARY
We have detailed the myriad ways in which phosphorene's structural in-plane asymmetry is manifest in the anisotropy of charge and spin properties of its electrons and holes, all of which are otherwise obscured by the many numerical methods previously applied to the problem. We have elucidated the origin of the band dispersion anisotropy near the gap edge, which is due to the specific directional preference of the coupling between conduction and valence bands and governs various transport and optical properties. By analyzing the symmetry of spinorbit coupling, we have derived compact spin-dependent eigenstates of electrons and holes for all high-symmetry quantization axes. As an example of the utility of these eigenstates, we investigated the valence band anisotropy of the Elliott-Yafet spin-relaxation process. By incorporating the relevant invariant components into our model, such as those of external applied fields, mechanical strain, and quantum confinement, this theory is highly extensible to the analysis of many other relevant circumstances of interest.
Importantly, our theory provides guidance for experimental efforts to empirically confirm and quantify these charge and spin phenomena. Valence band properties like the effective mass anisotropy and possible indirect bandgap are most directly compared to results from ARPES. 25,27 This ultra-high vacuum surface-sensitive
method is especially convenient since electrical contact to phosphorene transport layers has not yet been realized, in part due to environmental sensitivity and subsequent degradation. 20 Even without metallic electrodes, basic transport properties can be determined with e.g. microwave Hall mobility measurement. 46 However, investigation into many other properties (such as thermal and excitonic transport, weak localization or antilocalization, etc) may require fabrication of true electronic devices.
Optical non-contact techniques are also very useful in verifying the linear polarimetry in the visible spectrum, enabled by the dipole selection rules. However, with the expected inefficiency of optical orientation in this material, electrical techniques are required to measure anisotropy of the spin relaxation. Once single phosphorene layers can be stabilized for device processing and fabrication, four-terminal nonlocal geometry devices in the presence of oblique magnetic fields (to precess spins out-of-plane) will be especially applicable. 47
In closing, we are obliged to point out the presence of
- ∗ [email protected]
- 1 B. Aufray, A. Kara, S. B. Vizzini, H. Oughaddou, C. L´ eAndri, B. Ealet, and G. Le Lay, Appl. Phys. Lett. 96 , 183102 (2010); B. Lalmi, H. Oughaddou, H. Enriquez, A. Kara, S. B. Vizzini, B. N. Ealet, and B. Aufray, Appl. Phys. Lett. 97 , 223109.
- 2 M. E. Dvila, L. Xian, S. Cahangirov, A. Rubio, and G. L. Lay, New Journal of Physics 16 , 095002 (2014).
- 3 A. Barfuss, L. Dudy, M. R. Scholz, H. Roth, P. H¨pfner, o C. Blumenstein, et al. , Phys. Rev. Lett. 111 , 157205 (2013).
- 4 Y. Song and H. Dery, Phys. Rev. Lett. 111 , 026601 (2013).
- 5 M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82 , 3045 (2010).
- 6 N. Alem, R. Erni, C. Kisielowski, M. D. Rossell, W. Gannett, and A. Zettl, Phys. Rev. B 80 , 155425 (2009).
- 7 A. Morita, Appl. Phys. A 39 , 227 (1986).
- 8 P. W. Bridgman, J. Am. Chem. Soc. 36 , 1344 (1914).
- 9 Y. Maruyama, S. Suzuki, K. Kobayashi, and S. Tanuma, Physica B+C 105 , 99 (1981).
- 10 R. Hultgren, N. S. Gingrich, and B. E. Warren, J. Chem. Phys. 3 , 351 (1935).
- 11 A. Brown and S. Rundqvist, Acta Cryst. 19 , 684 (1965).
- 12 R. W. Keyes, Phys. Rev. 92 , 580 (1953).
- 13 D. Warschauer, J. Appl. Phys. 34 , 1853 (1963).
- 14 S. Narita, Y. Akahama, Y. Tsukiyama, K. Muro, S. Mori, S. Endo, M. Taniguchi, M. Seki, S. Suga, A. Mikuni, and H. Kanzaki, Physica B+C 117-118, Part 1 , 422 (1983).
- 15 H. Asahina and A. Morita, J. Phys. C 17 , 1839 (1984).
- 16 T. Low, A. S. Rodin, A. Carvalho, Y. Jiang, H. Wang, F. Xia, and A. H. Castro Neto, Phys. Rev. B 90 , 075434 (2014).
- 17 H. Kawamura, I. Shirotani, and K. Tachikawa, Solid State Comm. 49 , 879 (1984).
- 18 M. Baba, Y. Nakamura, Y. Takeda, K. Shibata, A. Morita, Y. Koike, and T. Fukase, J. Phys.: Cond. Mat. 4 , 1535 (1992).
secondary features of the conduction and valence bands that cannot be captured simply by the zone-center symmetries. For example, satellite valleys may play a role in transport properties of phosphorene when a high electric field accelerates and heats mobile charge carriers. Interestingly, secondary conduction band valleys are predicted by DFT calculation along the k y direction, whereas the valence band valleys are along the k x direction, several hundred meV from the band edge.
## ACKNOWLEDGMENTS
We thank Dr. Yang Song for carefully reading the manuscript and providing valuable suggestions. We gratefully acknowledge support from the Office of Naval Research under contract N000141410317, the National Science Foundation under contract ECCS-1231855, and the Defense Threat Reduction Agency under contract HDTRA1-13-1-0013.
- 19 S. P. Koenig, R. A. Doganov, H. Schmidt, A. H. Castro Neto, and B. Ozyilmaz, Appl. Phys. Lett. ¨ 104 , 103106 (2014).
- 20 H. Liu, A. T. Neal, Z. Zhu, Z. Luo, X. Xu, D. Tomnek, and P. D. Ye, ACS Nano 8 , 4033 (2014).
- 21 Y. Takao and A. Morita, Physica B+C 105 , 93 (1981); Y. Takao, H. Asahina, and A. Morita, J. Phys. Soc. Japan 50 , 3362 (1981).
- 22 A. N. Rudenko and M. I. Katsnelson, Phys. Rev. B 89 , 201408 (2014).
- 23 H. Asahina, K. Shindo, and A. Morita, J. Phys. Soc. Japan 51 , 1193 (1982).
- 24 Y. Du, C. Ouyang, S. Shi, and M. Lei, J. Appl. Phys. 107 , 093718 (2010).
- 25 F. Xia, H. Wang, and Y. Jia, Nature Comm. 5 , 4458 (2014).
- 26 J. Q. andXianghua Kong andZhi-Xin Hu andFeng Yang and W. Ji, Nature Comm. 5 , 4475 (2014).
- 27 L. Li, Y. Yu, G. J. Ye, Q. Ge, X. Ou, H. Wu, D. Feng, X. H. Chen, and Y. Zhang, Nature Nano. 9 , 372 (2014).
- 28 A. Rodin, A. Carvalho, and A. Castro Neto, Phys. Rev. Lett. 112 , 176801 (2014).
- 29 T. Takahashi, K. Shirotani, S. Suzuki, and T. Sagawa, Solid State Communications 45 , 945 (1983).
- 30 V. Tran, R. Soklaski, Y. Liang, and L. Yang, Phys. Rev. B 89 , 235319 (2014).
- 31 M. Buscema, D. J. Groenendijk, S. I. Blanter, G. A. Steele, H. S. J. van der Zant, and A. Castellanos-Gomez, Nano Lett. 14 , 3347 (2014).
- 32 G. L. Bir and G. E. Pikus, Symmetry and Strain-Induced Efiects in Semiconductors (New York, Wiley, 1974).
- 33 C. Bradley and A. Cracknell, The mathematical theory of symmetry in solids: representation theory for point groups and space groups (Clarendon, 1972).
- 34 P. Giannozzi et al. , J. Phys.: Cond. Mat. 21 , 395502 (2009).
- 35 P. L¨wdin, J. Chem. Phys. o 19 , 1396 (1951).
- 36 V. Tran and L. Yang, Phys. Rev. B 89 , 245407 (2014).
- 37 B. Huang, D. Monsma, and I. Appelbaum, Phys. Rev. Lett. 99 , 177209 (2007).
- 38 P. Li and H. Dery, Phys. Rev. Lett. 107 , 107203 (2011).
- 39 J. L. Cheng, M. W. Wu, and J. Fabian, Phys. Rev. Lett. 104 , 016601 (2010).
- 40 R. Jansen, Nature Mat. 11 , 400408 (2012).
- 41 M. Dyakonov and V. Perel, Sov. Phys. Solid State 13 , 3023 (1972).
- 42 R. Elliott, Phys. Rev. 96 , 266 (1954).
- 43 Y. Yafet, in Solid State Physics , Vol. 14, edited by F. Seitz and D. Turnbull (Academic Press, New York, 1963) pp. 1 - 98.
- 44 L. Liu, Phys. Rev. 126 , 1317 (1962).
- 45 G. Weisz, Phys. Rev. 149 , 504 (1966).
- 46 A. M. Portis and D. Teaney, J. Appl. Phys. 29 , 1692 (1958).
- 47 W. Han, K. Pi, K. M. McCreary, Y. Li, J. J. I. Wong, A. G. Swartz, and R. K. Kawakami, Phys. Rev. Lett. 105 , 167202 (2010).
|
10.1103/PhysRevB.90.115439
|
[
"Pengke Li",
"Ian Appelbaum"
] |
2014-08-04T19:04:40+00:00
|
2014-10-02T19:33:00+00:00
|
[
"cond-mat.mtrl-sci",
"cond-mat.mes-hall"
] |
Electrons and holes in phosphorene
|
We present a symmetry analysis of electronic bandstructure including
spin-orbit interaction close to the insulating gap edge in monolayer black
phosphorus ('phosphorene'). Expressions for energy dispersion relation and
spin-dependent eigenstates for electrons and holes are found via simplification
of a perturbative expansion in wavevector $k$ away from the zone center using
elementary group theory. Importantly, we expose the underlying symmetries
giving rise to substantial anisotropy in optical absorption, charge and spin
transport properties, and reveal the mechanism responsible for valence band
distortion and possible lack of a true direct gap.
|
1408.0771v2
|
## Statistical Thermodynamics of Clustered Populations
Themis Matsoukas ∗
Department of Chemical Engineering, Pennsylvania State University, University Park, PA 16802 (Dated: August 19, 2014)
We present a thermodynamic theory for a generic population of M individuals distributed into N groups (clusters). We construct the ensemble of all distributions with fixed M and N , introduce a selection functional that embodies the physics that governs the population, and obtain the distribution that emerges in the scaling limit as the most probable among all distributions consistent with the given physics. We develop the thermodynamics of the ensemble and establish a rigorous mapping to regular thermodynamics. We treat the emergence of a so-called 'giant component' as a formal phase transition and show that the criteria for its emergence are entirely analogous to the equilibrium conditions in molecular systems. We demonstrate the theory by an analytic model and confirm the predictions by Monte Carlo simulation.
PACS numbers: 02.50.Ey, 87.23.Cc 05.70.Ln 68.43.De Keywords: Statistical thermodynamics, ensemble theory
## I. INTRODUCTION
Gibbs's ensemble method [1] represents a remarkable success of mathematical physics: it provides the complete macroscopic description of a complex stochastic system and prescribes its conditions of equilibrium and stability in terms of a small number of variables. The phenomenology of thermodynamics has proven useful in other areas. Systems of evolving discrete populations present a close analogy to thermodynamic microstates and have lead to mathematical treatments that borrow from the language of equilibrium thermodynamics [2-6]. Polymer gelation [7-9], shattering in fragmentation [10], the spread of epidemics [11-13] and the emergence of connectivity in artificial and neural networks [14-17] are all systems that involve the emergence of a giant coherent structure, a process that is commonly discussed in the phenomenology of phase transitions. These diverse physical problems point to a common, yet unclear, link to thermodynamics. Jaynes drew a powerful connection between statistical mechanics and information theory [18] that suggests a broader applicability of thermodynamics. Jaynes considered the problem of a random variable whose probability distribution cannot be accessed except through limited data. He suggested that the unknown distribution can be identified as the one that is most probable among all distributions that are consistent with the available data. This amounts to a constrained maximization of the entropy of the distribution. Given appropriate constraints, practically any distribution may be interpreted as a maximum entropy distribution. The success of the maximum entropy formalism suggests that thermodynamic concepts could be extended outside the realm of interacting particles to any physical problem that involves a distributed population. The unknown distribution would then be determined by constrained maximization of en- tropy. There is a problem however: there is no obvious connection between the constraints that are required to produce a maximum entropy distribution and the physics from which that distribution emerges. For example, depending on the rate law, particle growth by surface deposition may produce a distribution that is Gaussian, if the rate is independent of size, or exponential, if the rate is proportional to particle mass [19]. Both distributions can be derived by maximum entropy arguments, constrained either by fixing the mean (exponential), or the mean and the variance (Gaussian); there is no physical link, however, between these particular constraints and the rate rate laws from which they arise.
∗ [email protected]
Here we develop a thermodynamic theory that circumvents these difficulties and connects a generic population to the laws that govern its evolution. Our approach is motivated by Jaynes in that we view the distribution in the scaling limit as the one that is most probable among all that satisfy the physics of the problem. We depart from Jaynes in that the distributions of the ensemble are not equally probable but biased by a selection functional , a quantity that embodies the physics under which the population evolves. The paper is organized as follows. In section II we define the microcanonical cluster ensemble and introduce the selection bias and its mathematical properties. In section III we pass to the thermodynamic limit, obtain the most probable distribution and derive the fundamental relationships among the primary variables of the ensemble. In section IV we discuss the analogies between the cluster ensemble and the familiar thermodynamic ensemble. In section V we develop the thermodynamic criterion for the emergence of a giant cluster ('gel') and demonstrate the theory with an example, which we solve analytically and test by stochastic simulation. In section VI we discuss the findings in the broader context of stochastic processes and finally summarize the conclusions in section VII.
## II. MICROCANONICAL CLUSTER ENSEMBLE
We begin with a population of M indistinguishable individuals divided into N distinguishable clusters ( N < M ). We define a configuration of masses to be an ordered list of N clusters with total mass M and notate it in vector form as m = ( m ,m , 1 2 · · · , m N ). A configuration is characterized by its distribution n = ( n , n 1 2 , · · · ), such that n i is the number of clusters with i members. We refer to i as the size or mass of the cluster. The microcanonical ensemble consists of all possible configurations of M members are divided into N clusters. All distributions of the ensemble satisfy the constraints
$$\sum _ { i } n _ { i } = N, \ \sum _ { i } i n _ { i } = M. \quad \quad ( 1 ) \quad \text{$^{\ g e n }$}$$
The maximum possible cluster size in a distribution of the ( M,N ) ensemble is M -N +1, but if we adopt the convention n i = 0 for all i > M -N + 1, the limits in the summations may be assumed to run from i =1 to ∞ and will not be explicitly shown. Each distribution is associated with a multiplicity factor that represents the number of configurations that have the same distribution. This is equal to the number of permutations in the order of cluster masses in the configuration and is given by the multinomial coefficient,
$$\mathbf n! = \frac { N } { n _ { 1 }! \, n _ { 2 } \cdots }. \quad \quad \ \ ( 2 )$$
The log of the multiplicity factor is the entropy of distribution:
$$S = \log n!. \quad \quad \ \ ( 3 ) \quad \ i \, \colon$$
In the Stirling approximation this reverts to the familiar functional. We now introduce the selection bias W ( n ), a functional of n that biases selection such that the probability of distribution n is proportional not only to its multiplicity n ! but also to its selection bias W ( n ):
$$P ( \mathbf n ) = \mathbf n! \frac { W ( \mathbf n ) } { \Omega _ { M, N } }. \quad \quad ( 4 )$$
Here Ω MN is the partition function and satisfies the normalization condition,
$$\Omega _ { M, N } = \sum _ { n } n! \, W ( n ), \quad \quad ( 5 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
with the summation taken over all distributions of the ensemble. The selection bias is a fundamental property of the cluster ensemble and embodies the physics of the problem. Here, the bias will remain general and unspecified; the only condition we impose is that it must produce distributions with proper extensive behavior in the thermodynamic limit. This requires log W to be homogeneous in n with degree 1 (see the Appendix for details).
It then follows from Euler's theorem that log W is a linear combination of its derivatives with respect to n i :
$$\underbrace { e \text{ in-} } _ { N } \, < \, \underset { \substack { 1 \text{ or-} \\ 1 \text{ or-} } } \, \log W ( \mathbf n ) = \sum _ { i } n _ { i } \left ( \frac { \partial \log W ( \mathbf n ) } { \partial n _ { i } } \right ) _ { n _ { j } } \equiv \sum _ { i } n _ { i } \log w _ { i }. \ ( 6 )$$
The derivatives log w i are an important element of the theory; they represent the contribution of cluster size i to log W and will be referred to as cluster bias . If the log of the selection bias is a linear functional of n [20], then the w i are intrinsic functions of cluster size i , and thus independent of the distribution itself. A special case of linear bias is W ( n ) = 1 (log w i = 0), which gives equal weight to all distributions (unbiased ensemble). In the general case the w i 's will depend not only on i but also on the distribution n to which they refer, as Eq. (6) implies.
The microcanonical ensemble defined here is closely related to that discussed in Pitman [21] and in Berestycki and Pitman [3] in the context of integer partitions. The main difference is in the natural multiplicity of distributions in the ensemble, which depends on whether cluster configurations are taken to be ordered (as in the cluster ensemble), or not (as in Refs. [3, 21]). Such differences can be reconciled between the two ensembles. References [3, 21] also present solutions for special cases of linear bias functionals that arise in coagulation and fragmentation. Here our interest is not in specific solutions but rather in the fundamental relationships between variables of the cluster ensemble.
## Exchange Reactions
Suppose that two clusters in distribution n with sizes i and j , respectively, exchange members to produce two new clusters with sizes k and l such that i + j = k + . l This binary exchange process converts one distribution of the ensemble into another and can be represented by the reversible reaction,
$$\mathbf n \stackrel { K } { \Longrightarrow } \mathbf n ^ { \prime }$$
If the equilibrium constant is set to
$$K _ { n = n ^ { \prime } } = \frac { W ( { \mathbf n } ^ { \prime } ) } { W ( { \mathbf n } ) },$$
the process will then produce an ensemble of distributions whose probability is given by Eq. (4). The exchange reaction, allows us to associate the cluster ensemble, a static collection of distributions, with a dynamic ensemble that obeys detailed balance with respect to exchange reactions. The practical implication is that we may sample the ensemble by Metropolis Monte Carlo simulation of binary exchange reactions with acceptance probability proportional to W ( n ′ ) /W ( n ). 1
1 This extends to any type of exchange reaction, not necessarily binary with respect to clusters, as long as both M and N are
## III. THERMODYNAMIC LIMIT
We now pass to the thermodynamic limit on the premise that when M and N are large the ensemble reduces to a single distribution, ˜. n From Eq. (5) we then have
$$\log \Omega \to \tilde { S } + \log \tilde { W }, \quad \quad \ ( 9 ) \quad \dot { \text{of} }$$
where ˜ = S S ( ˜ n ) is the entropy of the most probable distribution and log ˜ W = log W ( ˜ n ) is its log-bias. The most probable distribution is obtained by standard Lagrange maximization, and the result is
$$\frac { \tilde { n } _ { i } } { N } = \tilde { w } _ { i } \frac { e ^ { - \beta i } } { q }, \quad \quad \ \ \ \ ( 1 0 )$$
where log ˜ w i is the cluster bias in the most probable distribution. The Lagrange multipliers β and log q correspond to the two constraints in Eq. (1). They are given in terms of the ratio M/N in implicit form by
$$\frac { M } { N } = \sum _ { \substack { i \, \tilde { w } _ { i } e ^ { - \beta i } } } \Big / \sum \tilde { w } _ { i } e ^ { - \beta i }, \quad \ \ ( 1 1 ) \quad \ e n s { \alpha }$$
$$q & = \sum \tilde { w } _ { i } e ^ { - \beta i }, & ( 1 2 ) \quad \text{app}$$
which we obtain by inserting the most probable distribution into the constraints. We now take the log of the most probable distribution, multiply by ˜ , and perform n i the summation (see Appendix):
$$\log \Omega _ { M, N } = \beta M + \left ( \log q \right ) N, \quad \ \ ( 1 3 ) \quad \text{Hor}$$
This result of remarkable simplicity represents a fundamental relationship between the primary variables of the ensemble. Since S and log W are extensive in M and N , it follows from Eq. (9) that so is log Ω. In other words, log Ω is homogeneous in M and N , with degree 1. By Euler's theorem we have
$$\beta = \left ( \frac { \partial \log \Omega } { \partial M } \right ) _ { N } ; \ \ \log q = \left ( \frac { \partial \log \Omega } { \partial N } \right ) _ { M }, \quad ( 1 4 ) \quad \stackrel { \dots \dots } { \text{dyna} }$$
and
$$d \log \Omega _ { M N } = \beta \, d M + ( \log q ) \, d N. \quad \ ( 1 5 ) \quad \stackrel { \_ \_ } { \text{tial} }$$
Equation (14) identifies β and q as the partial derivatives of the microcanonical log-partition function with respect to its extensive variables, while Eq. (15) governs the evolution of a population under a quasistatic change of state (d M, d N ).
conserved. Binary reactions are convenient because they are the simplest to implement by simulation.
## Canonical Ensemble
To complete the theory we derive the statistics of the canonical ensemble. We start with a large microcanonical ensemble of M ′ individuals distributed into N ′ clusters. The ensemble is characterized by fixed β , q . We sample randomly N clusters ( N glyph[lessmuch] N ′ ) from a configuration of this microcanonical pool and seek the probability of distribution n that is sampled in this manner. In the thermodynamic limit the probability to pick a cluster of size i is given by the relative frequency of cluster size i ensemble, P i = w i exp( -βi /q ) . For small N relative to N ′ the probability to sample distribution n = ( n , n 1 2 , · · · ) is n ! P n 1 1 P n 2 2 · · · . Upon expanding the product the result becomes
$$P ( \mathbf n ) = \mathbf n! W ( \mathbf n ) \frac { e ^ { - \beta M } } { q ^ { N } }, \text{ \quad \ \ } ( 1 6 )$$
where M is the total mass (number of members) in the sampled distribution. Strictly, this derivation applies to linear bias because it assumes the w i in Eq. (16) to be the same for all distributions. Nonetheless, a sharply peaked ensemble in the vicinity of the most probable distribution is adequately described by a linearized bias and the result applies to general bias in the scaling limit.
The canonical partition function is the sum of the canonical weights n ! W ( n !) exp( -βM ). Since probabilities in Eq. (16) are properly normalized, by summation over all n on both sides we obtain
$$Q = q ^ { N }.$$
Here we recognize log q as the intensive form of the logarithm of the canonical partition function, (log Q /N ) .
## IV. CONNECTION TO THERMODYNAMICS
An analogy to molecular thermodynamics now emerges. Compare Eq. (15) with the familiar thermodynamic relationship in the nV E ensemble,
$$\frac { d S } { k } = \frac { d E } { k T } + \frac { p \, \mathrm d V } { k T } - \frac { \mu \, \mathrm d n } { k T }.$$
In both cases, on the left-hand side we have the differential of a quantity whose maximization defines the equilibrium state, expressed on the right-hand side in terms of a set of extensive variables that fix the macroscopic state of the system. A one-to-one correspondence can be drawn between the two ensembles, as shown in Table I (to complete the analogy we must also set µ = 0). Notice that the maximized quantity of the cluster ensemble is not the entropy of distribution, as in statistical mechanics, but the microcanonical partition function. In statistical mechanics the selection bias is set by the postulate of equal a priori probabilities. This corresponds to uniform bias W = 1 in our theory. Equation (9) then gives log Ω = S
TABLE I. Correspondence between properties of the cluster ensemble and thermodynamics. Other mappings are possible.
| cluster ensemble | Thermodynamics |
|--------------------|------------------------------|
| log Ω | → (thermodynamic Entropy) /k |
| M | → Energy |
| β | → 1 /kT |
| N | → Volume |
| log q | → p/kT |
and thus we recover the familiar nV E ensemble. In general, log Ω and entropy in the cluster ensemble are distinct properties, and between the two, it is the partition function that is of fundamental importance.
The thermodynamic mapping shown here is not unique as one may choose to associate β with -µ/kT , for example; this mapping is implied in Ziff et al. [22], who refer to the quantity e -β as 'fugacity' (symbol ξ in the original reference). In this sense, the correspondence between variables of the cluster ensemble and thermodynamics should not be taken literally but ought be viewed as a mathematical mapping that allows one to obtain relationships between variables of the cluster ensemble by translation of the corresponding result in thermodynamics. For example, by direct analogy to the thermodynamic result, C V > 0, we may write, ( ∂M/∂ ) N β < 0, a result that we can independently obtain by stability analysis on log Ω. Still, the mapping in Table I has a certain intuitive appeal. For example, a process may be viewed as 'compression' if it causes the average size to increase (we imagine N to decrease at constant M ), or 'expansion,' if the average size decreases.
## V. A PHASE TRANSITION: THE GIANT CLUSTER
If populations are thermodynamic entities, do they undergo the equivalent of phase equilibrium? In the context of the cluster ensemble, a 'phase' is a distinct distribution, and 'phase equilibrium' refers to the coexistence of two separate distributions within the same population, such that they both remain stable under exchange of members. We address this problem by considering the emergence of a giant cluster, namely, the transition from a population of finite clusters to one that contains one cluster that is of the order of M . The emergence of the giant component has long invited an analogy to condensation, but now we are in position to make a rigorous connection.
When M individuals are placed into N clusters, the maximum possible cluster size is i max = M -N + 1. The size range i max / 2 < i ≤ i max is special: it can accommodate at most one cluster ; the mass balance is not satisfied otherwise. A cluster in this region contains a finite fraction of the total mass (its size is of the the order of M ) but a vanishingly small fraction of the total number of clusters. We refer to clusters in this range as the 'gel' phase and to all others as 'sol' phase. Such system distributes members between the two phases so that the partition function of the combined system is maximized under the constraints M sol + M gel = M and N sol + N gel = N with N gel = 1. This variational principle leads to the condition
$$\left ( \frac { \partial \log \Omega _ { \text{sol} } } { \partial M _ { \text{sol} } } \right ) _ { N _ { \text{sol} } } = \left ( \frac { \partial \log \Omega _ { \text{gel} } } { \partial M _ { \text{gel} } } \right ) _ { N _ { \text{gel} } } \equiv \beta. \quad ( 1 8 )$$
This result is obtained by maximizing the partition function of the two-phase system (see Appendix) and establishes the condition of thermal equilibrium between the phases. Since the number of clusters in each phase is fixed, q is not required to equilibrate. The emergence of a giant cluster is akin to osmotic equilibrium.
## A Gelling Selection Bias
We demonstrate the application of the theory using a selection bias that can be solved analytically. We take the cluster bias to be w i = i -3 , which corresponds to the selection functional,
$$W ( { \mathbf n } ) = \prod _ { i } i ^ { - 3 n _ { i } }.$$
Let us examine whether it is possible to construct a twophase solution for this bias. The distribution of the sol phase is
$$\left ( \frac { n _ { i } } { N } \right ) ^ { \text{sol} } = i ^ { - 3 } \frac { e ^ { - \beta i } } { q }, \text{ \quad \ \ } \text{(19)}$$
with β and q given by
$$\frac { M } { N } = \frac { \text{Li} _ { 2 } \left ( e ^ { - \beta } \right ) } { \text{Li} _ { 3 } \left ( e ^ { - \beta } \right ) } \quad \quad \quad ( 2 0 )$$
$$q = \text{Li} _ { 3 } \left ( e ^ { - \beta } \right ). \text{ \quad \ \ } \text{(2)}$$
where Li n ( x ) is the polylogarithm function. These are obtained From Eqs. (11) and (12) by letting the upper limit to go to infinity (scaling limit). To obtain β and q we solve Eqs. (20) and (21) numerically for the given M/N . The sol distribution is therefore completely defined if M/N is fixed.
Suppose the system forms a two-phase state that consists of a gel cluster with mass M gel in a equilibrium with sol ( M sol = M -M gel , N sol = N -1 → N ). For the linear bias functional, the partition function of the gel-phase is Ω gel = w M ( gel ), where w M ( gel ) is the cluster bias of the giant cluster. This follows from Eq. (9) and the fact that the gel phase consists of a single cluster ( S gel = 0). Its temperature is
$$\beta ^ { \text{gel} } = \frac { \text{d} \log w _ { i } } { \text{d} i } \Big | _ { M _ { \text{gel} } } = - 3 / M _ { \text{gel} }, \quad \ \ ( 2 2 )$$
TABLE II. Phase behavior of linear cluster bias w i = i -3 . The system forms a giant cluster at M/N = 1 368. .
| | M/N < 1 . 368 single sol | M/N ≥ 1 . 368 sol+gel |
|----------------------------------------------------|------------------------------------------|-------------------------------------------------------------------------------|
| sol distribution gel distribution β q gel fraction | n i N = e - βi q i - 3 - Eq. 20 Eq. 21 0 | n i N = 0 . 832 i - 3 n gel i = δ i,M gel 0 1 . 202 M gel M = 1 - 1 . 368 M N |
FIG. 1. (Color online) Phase diagram of a gelling system with cluster bias w i = i -3 : (a) gel fraction, (b) mean sol cluster, (c) temperature β , and (d) pressure q , as a function of the progress variable θ = 1 -N/M . The simulations are conducted with M = 200. The theoretical lines are based on the equations in Table II, which assume an infinite system.
and in the scaling limit, β gel → 0. The corresponding equilibrium sol phase is obtained from Eq. (19) with β = 0, q = 1 202. This produces the power-law distribution .
$$\left ( \frac { \tilde { n } _ { i } } { N } \right ) ^ { * } = 0. 8 3 2 i ^ { - 3 }, \text{ \quad \ \ } ( 2 3 ) \text{ \ \ } \text{co}$$
whose mean cluster size is i ∗ = 1 368. The phase behav-. ior is now fully determined. For M/N < i ∗ = 1 368 we . have a single sol phase whose distribution is,
$$\frac { n _ { i } } { N } = \frac { i ^ { - 3 } e ^ { - i \beta } } { \text{Li} _ { 3 } \left ( e ^ { - \beta } \right ) }. \quad \quad \ \ ( 2 4 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\quad \frac { \text{gi} } ( \begin{matrix} 2 \\ N \\ \end{matrix} ) \end{matrix}$$
If M/N is larger than the equilibrium sol cluster, the system splits in two phases: a sol phase whose distribution is given in Eq. (23), and one giant cluster whose fraction is given by the mass balance condition (see Appendix)
$$\phi _ { \text{gel} } = 1 - i ^ { * } \left ( \frac { N } { M } \right ), \text{ \quad \ \ } ( 2 5 )$$
where i ∗ = 1 368 is . the equilibrium mean sol cluster. Notice that the normalized equilibrium sol distribution ( n /N i ) sol remains constant in the entire sol/gel region. The conversion of the sol phase into the gel along the path N → 1 at fixed M occurs isothermally until the sol disappears completely.
We test these predictions by Monte Carlo simulation using the method of exchange reactions on a finite population with M = 200 members. We begin with a configuration (ordered list) of N cluster masses with total mass M ; the initial distribution is approximately uniform but any other distribution may be used. At each step we pick two clusters at random, merge them into a single cluster, then break it up randomly in two. The new configuration is accepted with probability proportional to the equilibrium constant in Eq. (8) and the process is repeated until the system reaches equilibrium. The mean distribution is calculated as an average of 1 to 4 × 10 6 realizations. The simulation is repeated with N ranging from M -1 to 2.
First we analyze the sol and gel phases obtained in the simulation. We calculate the gel fraction, φ gel , as the fraction of the total mass that resides in the region i > M ( -N + 1) / 2. The mean cluster size in the sol ( ¯ ) i is calculated as the average mass in the region 1 ≤ 1 < ( M -N + 1) / 2, and the parameters β and q are obtained from the finite version of Eqs. (20) and (21). These results ( φ , ¯ , i β and q ) are shown in Fig. 1 as a function of the progress variable θ = 1 -N/M , where θ = 0 refers to a fully dispersed population (perfect sol), and θ = 1 a fully gelled population (all particles have joined the giant gel). The simulations are in general agreement with the theory though discrepancies are noted due to finite-size effects. The gel point is predicted at θ ∗ = 1 -1 /i ∗ = 0 2690 ( . N ∗ = 146), while the simulation gives θ ∗ ≈ 0 3 ( . N ∗ ≈ 140). The gel fraction is seen to be a linear function of θ with small deviations seen near the gel point. The mean cluster in the postgel region converges to the predicted value i ∗ = 1 368; . its small value makes it quite more sensitive to finite-size effects. Recall that in these simulations N is varied at constant M . In thermodynamics terms, in the direction of increasing θ the system undergoes compression. Accordingly its temperature 1 /β increases, and so does its pressure log q .
Sample distributions are shown in Fig. 2. At N = 170 > N ∗ the population consists of a single sol distribution that decays monotonically within the sol region and is in excellent agreement with Eq. (19). At N = 75 < N ∗ the population is compressed above its gel point, a gel phase is present, and the sol distribution is now given by Eq. (23). As N is reduced further, the sol phase 'shrinks in place,' i.e., the number of sol
FIG. 2. (Color online) Selected distributions for cluster bias w i = i -3 in a finite population with M = 200 members. The giant cluster appears near N ∗ = 146. The shaded area is the simulated distribution and the solid line is theory. The dashed line marks the phase boundary between the sol phase and the gel phase.
clusters decreases while the normalized distribution remains fixed, and the giant cluster moves to larger size as more particles are converted from the sol. The theoretical distribution of the gel phase is a Kronecker delta at M gel = M (1 -1 368 . N/M ). The apparent distribution of gel clusters represents not the relative number of gel clusters in a typical distribution (there is at most one cluster in this range) but fluctuations in the position of the giant cluster from one configuration to the next.
We examine these fluctuations in more detail in Fig. 3, which shows the mass fraction of the giant cluster,
FIG. 3. (Color online) Fluctuations of the gel cluster with cluster bias w i = i -3 over the course of exchange reactions in a population with M = 200 members. (a) N = 140 (right about the gel point); (b) N = 100 (above the gel point); (c) N = 5 (very close to complete gelation). The dashed line marks the boundary between the sol and the gel phase. When the gel cluster falls below this line, it enters the sol phase and the gel fraction drops to zero.
M gel /M , within individual configurations of the ensemble and how it evolves as a result of exchange reactions. The dashed line marks the sol-gel boundary. A cluster above this line is a gel cluster; if the maximum cluster falls below the dashed line (sol region), no gel cluster exists, i.e. the giant cluster evaporates into the sol and its mass is zero. Crossings above and below the dashed line indicate condensation/evaporation, respectively, of the gel cluster. These fluctuations therefore reflect transfer between the two phases. At N = 140 ( θ = 0 3) the . system is close to the gel point. The nucleation of the gel phase is indicated by short-lived excursions into the gel region. The gel fraction, calculated over all states (whether they contain a gel cluster or not) is φ gel = 0 069. . Even though the giant cluster appears abruptly in the size range i > i max = 30, the ensemble-average gel frac-
FIG. 4. (Color online) The cluster bias w i = i 3 results in a stable single-phase population for all N (no gel phase). The shaded area is the simulated distribution and the solid line is theory. The dashed line marks the phase boundary between sol and the region of the giant cluster.
tion at the gel point is continuous. At N = 100 ( θ = 0 5) . the system is above the gel point. Most states contain a gel cluster that makes frequent crossings into the sol region. The mean gel fraction is φ gel = 0 22. . At N = 4 the system is very close to complete gelation with most of its mass residing on the giant cluster ( φ gel = 0 98). . Even at this highly gelled state, fluctuations reach the sol-gel boundary and occasionally the giant cluster evaporates but these excursions are both less frequent and short-lived.
The phase behavior of the gelling bias studied here bears remarkable similarity to Stockmayer's theory of gelation [23]. Stockmayer constructs an ensemble of polymer chains, similar to our cluster ensemble, and selects chains in proportion to the number of ways that i monomers can be combined to form an i -mer. He obtains the most probable distribution by maximizing the partition function and finds the gel fraction to be a linear function of the fraction of unreacted bonds, a quantity analogous to our θ . While Stockmayer raises numerous analogies between his gelation model and the thermodynamics of phase transitions, he obtains phase diagram by analysis of the convergence radius of the cluster distribution, not by thermodynamics, and identifies the gel point as the limiting condition that guarantees convergence of the moments of order 0 and 1. We can now make the thermodynamic connection rigorous. The cluster bias in Stockmayer's model is 2 [23]
$$w _ { i } = \frac { f! ( f i - i )! } { i! \, ( f i - 2 i + 2 )! },$$
where f is the functionality of the monomer. For f = 3 (the general case f > 3 can be treated similarly) this is functionally equivalent to the cluster bias (see the Appendix)
$$w _ { i } = i ^ { - 5 / 2 }.$$
This is of the same power-law type as the gelling bias discussed above and leads to the same isothermal condition at the sol-gel point, β gel = β sol = 0. This condition also defines the radius of convergence of the summations in Eqs. (11) and (12): the first and second moments of ˜ n i converge in the region β ≥ 0, while the second moment converges in β > 0 but diverges at the gel point. This is precisely the method by which Stockmayer obtained the gel point. Here we arrived at the same result by strict thermodynamic analysis.
Not every selection bias leads to a gelling population. Consider the cluster bias w i = i 3 , the inverse of the case discussed above, whose sol distribution is
$$n _ { i } / N = i ^ { 3 } e ^ { - \beta i } / q.$$
The temperature of the giant cluster, if one forms, is β gel = 3 /M gel . Again, in the scaling limit we obtain β gel → 0 but now the corresponding sol distribution diverges and cannot produce a sol that satisfies the two constraints of the ensemble. It is not possible to construct an equilibrium two-phase mixture this case. Even though this selection bias favors distributions with heavy tails, these are fully contained in the sol region. This prediction is confirmed by Monte Carlo simulation (Fig. 4), which demonstrates that the system forms a single sol distribution for all M/N .
## VI. DISCUSSION
The main element of the theory is the selection bias W , which alone determines the distribution in the scaling limit. The connection to statistical mechanics may now be seen more clearly. In statistical mechanics we
2 Notice that the w i in Stockmayer's notation refers to i w ! i in ours.
begin with the postulate of equal a priori probabilities, which corresponds to the unbiased ensemble with W = 1, and obtain the exponential distribution of microstates as the most probable distribution in the in the scaling limit. By allowing the selection bias to be an arbitrary functional of the distribution n , it is possible to obtain any distribution in the scaling limit. Conversely, any distribution of M members clustered in N groups can be associated with an equilibrium ensemble that obeys thermodynamics, provided that the corresponding selection bias is identified. The broader implication is that stochastic variables may be treated in the formalism of thermodynamics by associating its probability density function with the corresponding equilibrium ensemble. In this sense, thermodynamics should be viewed as a probabilistic calculus that is applicable not only to systems of physical particles, but to stochastic processes in general. Complex problems such as gelation, percolation and network connectivity may then be treated as formal phase transitions that obey thermodynamics. The crux of the problem then is the determination of the selection bias. This must be done on a case-by-case basis, based on the physical laws that govern the system at hand. In the Stockmayer model, for example, the selection bias arises from a combinatorial model for polymer chains. In populations undergoing dynamic transformations, it is determined by the corresponding rate laws. Examples will be discussed elsewhere.
We close with a final note on dynamics. Although time is not an explicit element of the theory, the connection to dynamics is provided by Eq. (15). This equation describes the evolution of the ensemble along a quasi-static trajectory, which we define as a path along which the selection bias remains invariant, while M and N change due to dynamic processes in the population. Essentially, Eq. (15) gives the evolution of the system in terms of the mean cluster size M/N rather than time. At fixed ( M,N ) the system is at 'equilibrium,' which is understood to mean that the ensemble relaxes to the most probable distribution that corresponds to the present values of M and N . Even if the system is on an irreversible path, i.e., unidirectional in time, the state at fixed ( M,N ) may be described using the calculus of equilibrium thermodynamics. The validity of this approach is confirmed in Fig. 1: in moving from θ = 0 to θ = 1 we follow a polymerization reaction that leads to the irreversible formation of a gel.
## VII. CONCLUSIONS
In summary, we have presented a thermodynamic theory for a generic population of M individuals clustered into N groups. The distribution of this population is viewed as the most probable distribution that emerges among all possible partitions of M into N elements, under a functional that biases the selection of individual partitions. By associating the population with a corre- sponding ensemble we are able to express the distribution in terms of thermodynamic quantities (partition function, 'temperature,' 'pressure') that obey relationships analogous to those in molecular systems. An important result of the theory is a thermodynamic criterion (maximization of the partition function) to determine whether a giant cluster (gel phase) forms. Although it is common in the literature to refer to the emergence of a giant component as a phase transition, the theory presented here associates this process, for the first time, with the maximization of a thermodynamic functional. This unlocks the toolbox of thermodynamics and makes it available to the study of population balances and stochastic phenomena in general.
## APPENDIX
## Appendix A: Derivations and Additional Notes
## 1. Homogeneity Condition Eq. (6)
Here we show that in order for the ensemble to have proper extensive behavior in the thermodynamic limit, the log of the selection bias must be a homogeneous function of n i with degree 1. Proper extensive behavior means that if M and N are both increased by a factor λ , the number of clusters ˜ n i in the most probable distribution must increase by the same factor. Equivalently, the ratio ˜ n /N i must be an intensive property, i.e., a function of M/N . The most probable distribution
We start with the relationship between β , q and M/N in Eqs. (11), (12), which we rewrite as
$$\frac { M } { N } = \sum i e ^ { - \beta i + \log \tilde { w } _ { i } } \Big / \sum e ^ { - \beta i + \log \tilde { w } _ { i } }, \quad ( A 1 )$$
$$q = \sum e ^ { - \beta i + \log \tilde { w } _ { i } }.$$
The ratio ˜ n /N i will be intensive only if β , q , and log ˜ w i are intensive; in turn, β and q will be intensive only if the log ˜ w i are intensive. This implies that if a distribution n is multiplied by λ , which will result in all cluster numbers being multiplied by λ while M/N remains constant, the cluster bias log w i must remain unchanged:
$$\log w _ { i } = \left ( \frac { \partial \log W ( \lambda \mathbf n ) } { \partial ( \lambda n _ { i } ) } \right ) = \left ( \frac { \partial \log W ( \mathbf n ) } { \partial n _ { i } } \right ).$$
This requires
$$\log W ( \lambda \mathbf n ) = \lambda \log W ( \mathbf n )$$
and states that log W ( n ) is homogeneous function of n of degree 1.
## 2. Derivation of Fundamental Eq. (13)
Take the log of Eq. (10)
$$\log \frac { \tilde { n } _ { i } } { N } = \log \tilde { w } _ { i } - \beta i - \log q,$$
multiply both sides by ˜ n i and sum over all i :
$$\sum \tilde { n } _ { i } \log \frac { \tilde { n } _ { i } } { N } = \sum \tilde { n } _ { i } \log \tilde { w } _ { i } - \beta M - ( \log q ) N,$$
Using S = log ˜ n !, and the Euler relationship in Eq. (6), we obtain
$$S + \log \tilde { W } = \beta M + ( \log q ) N,$$
which, combined with Eq. (9), leads to the final result
$$\log \Omega = \beta M + ( \log q ) N. \quad \quad ( A 3 ) \quad \text{If } \xi$$
If we write this result as
$$\log \Omega = M \left ( \beta + \frac { N } { M } \log q \right ) = N \left ( \frac { M } { N } \beta + \log q \right ), \ \ ( A 4 ) \quad \begin{array} { c } \Gamma \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \sin \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \ sin \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \tan \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \ cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \osigma \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \colon \sin \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \sin \sin \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \osigma$$
it will become clear that log Ω is homogeneous of degree 1 with respect to both M and N ( β , log q and M/N are all intensive), and satisfies Euler's theorem:
$$\log \Omega = M \left ( \frac { \partial \log \Omega } { \partial M } \right ) + N \left ( \frac { \partial \log \Omega } { \partial N } \right ). \quad ( A 5 ) \quad \text{Ther}$$
Comparison with Eq. (A3) leads to
$$\beta = \left ( \frac { \partial \log \Omega } { \partial M } \right ) _ { N }, \ \ \log q = \left ( \frac { \partial \log \Omega } { \partial N } \right ) _ { M },$$
which appears as Eq. (14) in the text.
## 3. Derivation of Isothermal Condition Eq. (18)
At equilibrium, a sol-phase system maximizes the microcanonical weight of the composite system. For a given partitioning of mass between the two phases, the sol phase maximizes its own microcanonical weight when it relaxes to the most probable distribution that corresponds to M sol = M -M gel , N sol = N -1. Therefore, we seek to maximize the partition function of the combined system,
$$\log \Omega _ { M, N } = \log \Omega _ { M _ { s o 1 }, N - 1 } + \log \Omega _ { g e l }.$$
Suppose that the giant cluster exchanges mass δm with the sol. Such exchange must leave the overall partition function unchanged:
$$0 = \delta \log \Omega _ { M _ { \text{sol} }, N - 1 } + \delta \log \Omega _ { \text{gel} }.$$
We divide both sides by δm , and noting that δM gel = -δM sol , the result is
$$\frac { \delta \log \Omega _ { \text{sol} } } { \delta M _ { \text{sol} } } = \frac { \delta \log \Omega _ { \text{gel} } } { \delta M _ { \text{gel} } }.$$
Since the exchange is at constant number of clusters, we identify these derivatives as the temperatures in each phase and thus arrive at the isothermal condition:
$$\beta _ { \text{sol} } = \beta _ { \text{gel} }.$$
Since the number of clusters does not fluctuate during exchange reactions, pressures do not equalize and each phase may be at its own pressure.
## 4. The Sol-Gel Boundary
Here we determine the cluster size that separates the sol from the gel. Suppose that i sol is the maximum possible cluster size in the sol region and i gel is the minimum possible cluster size in the gel region. Since i gel and i sol are neighboring masses, they meet the condition
$$i _ { \text{gel} } - i _ { \text{sol} } = 1. \text{ \quad \ \ } ( \text{A6} )$$
If a mass i gel is placed in the gel, we are left in the sol with M sol = M -i gel , N sol = N -1. Then the maximum possible cluster size in the sol is
$$\Lambda \right ) \quad i _ { \text{sol} } = M _ { \text{sol} } - N _ { \text{sol} } + 1 = M - N - i _ { \text{gel} } + 2. \quad ( \text{A7} )$$
Combining with Eq. (A6) we find
$$i _ { s o l } = \frac { M - N + 1 } { 2 }, \ \ i _ { g e l } = \frac { M - N + 3 } { 2 }. \quad ( A 8 )$$
Therefore, we identify the sol and the gel regions by the conditions:
$$& \text{sol} \, \colon \quad 1 \leq i \leq i _ { \max } / 2 ; \\ & \text{gel} \, \colon \quad i _ { \max } / 2 < i \leq i _ { \max },$$
where i max = M -N +1 is the maximum possible cluster in the ( M,N ) ensemble.
## 5. Stockmayer's model
Here we derive the asymptotic form of Eq. (26) for trifunctional monomers. Using the Stirling approximation, x ! ≈ x x exp( -x ) √ 2 πx , and f = 3 the Stockmayer weight becomes
$$w _ { i } = \frac { e ^ { 2 } } { \sqrt { \pi } } ( 1 2 i ) ^ { i } ( i + 2 ) ^ { - i - 5 / 2 }. \quad \quad ( A 9 )$$
Noting that
$$( i + 2 ) ^ { - i - 5 / 2 } \sim i ^ { - i - 5 / 2 } e ^ { - 2 - 5 / i } \sim i ^ { - i - 5 / 2 } e ^ { - 2 }, \ \ ( A 1 0 )$$
the previous result simplifies to
$$w _ { i } = \frac { 1 2 ^ { i } } { \sqrt { \pi } } \, i ^ { - 5 / 2 }. \quad \quad \quad ( A 1 1 )$$
The corresponding selection bias W ( n ) on arbitrary distribution n = ( n , n 1 2 , · · · ) is
$$W ( { \mathbf n } ) = \prod _ { i } w _ { i } ^ { n _ { i } } = \pi ^ { - N / 2 } 1 2 ^ { M } \prod _ { i } i ^ { - 5 n _ { i } / 2 }, \quad ( { \mathbf A } 1 2 )$$
where we have used ∑ n i = N and ∑ in i = M . The constant factors raised to N or M are selection-neutral (they make identical contribution to all distributions of the ensemble) and may be dropped from the selection bias so that the Stockmayer bias reduces to
$$w _ { i } = i ^ { - 5 / 2 }. \text{ \quad \ \ } ( A 1 3 )$$
The expressions for the cluster bias in Eqs. (A11) and (A13) are functionally equivalent, i.e., they produce identical ensembles.
- [1] J. W. Gibbs, Elementary Principles in Statistical Mechanics (Ox Bow Press, Woodbridge, CT, 1981), (reprint of the 1902 edition).
- [2] R. D. Vigil, Journal of Colloid and Interface Science 336 , 642 (2009), URL http://www.sciencedirect.com/ science/article/pii/S002197970900455X .
- [3] N. Berestycki and J. Pitman, Journal of Statistical Physics 127 , 381 (2007), URL http://dx.doi.org/10. 1007/s10955-006-9261-1 .
- [4] E. M. Hendriks, Z. Phys. B Condensed Matter 57 , 307 (1985).
- [5] E. M. Hendriks, J. L. Spouge, M. Eibl, and M. Schreckenberg, Zeitschrift f¨r Physik B Condensed Matter u 58 , 219 (1985), URL http://dx.doi.org/10.1007/BF01309254 .
- [6] J. L. Spouge, Macromolecules 16 , 121 (1983), URL http: //pubs.acs.org/doi/abs/10.1021/ma00235a024 .
- [7] P. J. Flory, Journal of the American Chemical Society 63 , 3091 (1941), http://pubs.acs.org/doi/pdf/10.1021/ja01856a062, URL http://pubs.acs.org/doi/abs/10.1021/ja01856a062 .
- [8] D. J. Aldous, Bernoulli 5 , 3 (1999).
- [9] A. A. Lushnikov, Phys. Rev. Lett. 93 , 198302 1 (2004).
- [10] E. D. McGrady and R. M. Ziff, Phys. Rev. Lett. 58 , 892 (1987), URL http://journals.aps.org/prl/abstract/ 10.1103/PhysRevLett.58.892 .
- [11] R. Pastor-Satorras and A. Vespignani, Phys. Rev. Lett. 86 , 3200 (2001), URL http://link.aps.org/doi/10. 1103/PhysRevLett.86.3200 .
- [12] M. E. J. Newman, Phys. Rev. E 66 , 016128 (2002), URL http://link.aps.org/doi/10.1103/PhysRevE. 66.016128 .
- [13] P. L. Krapivsky, S. Redner, and E. Ben-Naim, A Kinetic View of Statistical Physics (Cambridge University Press, Cambridge, 2010).
- [14] P. Erd˝s and A. R´nyi, Math. Inst. Hungar. Acad. Sci o e 5 , 17 (1960).
- [15] M. E. J. Newman, Phys. Rev. Lett. 103 , 058701 (2009), URL http://link.aps.org/doi/10.1103/ PhysRevLett.103.058701 .
- [16] J. Gao, S. V. Buldyrev, H. E. Stanley, and S. Havlin, Nat Phys 8 , 40 (2012), URL http://dx.doi.org/10.1038/ nphys2180 .
- [17] I. Breskin, J. Soriano, E. Moses, and T. Tlusty, Phys. Rev. Lett. 97 , 188102 (2006), URL http://link.aps. org/doi/10.1103/PhysRevLett.97.188102 .
- [18] E. T. Jaynes, Phys. Rev. 106 , 620 (1957).
- [19] T. Matsoukas and Y. Lin, Phys. Rev. E 74 , 031122 (pages 9) (2006), URL http://link.aps.org/abstract/ PRE/v74/e031122 .
- [20] I. M. Gelfand and S. V. Fromin, Calculus of Variations (Dover, Mineola, NY, 2000), (reprint of the 1963 edition).
- [21] J. Pitman, Combinatorial Stochastic Processes , vol. 1875 (Springer, 2006).
- [22] R. M. Ziff, M. H. Ernst, and E. M. Hendriks, Journal of Physics A: Mathematical and General 16 , 2293 (1983), URL http://stacks.iop.org/0305-4470/16/2293 .
- [23] W. H. Stockmayer, The Journal of Chemical Physics 11 , 45 (1943), URL http://link.aip.org/link/?JCP/11/ 45/1 .
|
10.1103/PhysRevE.90.022113
|
[
"Themis Matsoukas"
] |
2014-08-01T18:58:15+00:00
|
2014-08-18T14:39:00+00:00
|
[
"q-bio.PE"
] |
Statistical Thermodynamics of Clustered Populations
|
We present a thermodynamic theory for a generic population of $M$ individuals
distributed into $N$ groups (clusters). We construct the ensemble of all
distributions with fixed $M$ and $N$, introduce a selection functional that
embodies the physics that governs the population, and obtain the distribution
that emerges in the scaling limit as the most probable among all distributions
consistent with the given physics. We develop the thermodynamics of the
ensemble and establish a rigorous mapping to thermodynamics. We treat the
emergence of a so-called "giant component" as a formal phase transition and
show that the criteria for its emergence are entirely analogous to the
equilibrium conditions in molecular systems. We demonstrate the theory by an
analytic model and confirm the predictions by Monte Carlo simulation.
|
1408.0772v4
|
## Symmetry energy of nucleonic matter with tensor correlations
Or Hen, 1, ∗ Bao-An Li, 2 Wen-Jun Guo, 2, 3 L.B. Weinstein, 4 and Eli Piasetzky 1
1 School of Physics and Astronomy, Tel Aviv University, Tel Aviv 69978, Israel 2 Department of Physics and Astronomy, Texas A & M University-Commerce, Commerce, Texas 75429-3011, USA 3 College of Science, University of Shanghai for Science and Technology, Shanghai, 200093, China 4 Old Dominion University, Norfolk, Virginia 23529,USA
(Dated: February 9, 2015)
The nuclear symmetry energy ( E sym ( ρ )) is a vital ingredient of our understanding of many processes, from heavy-ion collisions to neutron stars structure. While the total nuclear symmetry energy at nuclear saturation density ( ρ 0 ) is relatively well determined, its value at supranuclear densities is not. The latter can be better constrained by separately examining its kinetic and potential terms and their density dependencies. The kinetic term of the symmetry energy, E kin sym ( ρ 0 ), equals the difference in the per-nucleon kinetic energy between pure neutron matter (PNM) and symmetric nuclear matter (SNM), often calculated using a simple Fermi gas model. However, experiments show that tensor force induced short-range correlations (SRC) between proton-neutron pairs shift nucleons to high-momentum in SNM, but have almost no effect in PNM. We present an approximate analytical expression for E kin sym ( ρ ) of correlated nucleonic matter. In our model, E kin sym ( ρ 0 ) = -10 MeV, which differs significantly from +12 5 MeV for the widely-used free Fermi gas model. . This result is consistent with our analysis of recent data on the free proton-to-neutron ratios measured in intermediate energy nucleus-nucleus collisions as well as with microscopic many-body calculations, and previous phenomenological extractions. We then use our calculated E kin sym ( ρ ) in combination with the known total symmetry energy and its density dependence at saturation density to constrain the value and density dependence of the potential part and to extrapolate the total symmetry energy to supranuclear densities.
PACS numbers: 25.70.-z, 25.60.-t, 25.80.Ls, 24.10.Lx
The nuclear symmetry energy, E sym ( ρ ), where ρ is the nuclear density, is related to the difference in the energy per nucleon of pure neutron matter (PNM) and symmetric nuclear matter (SNM). It determines many nuclear and astrophysical properties, such as the cooling of proto-neutron stars [1], the mass-radius relations of neutron stars [2], properties of nuclei involved in r -process nucleosynthesis [3], and heavy-ion collisions [4-6].
Much effort is being invested in improving our knowledge of E sym ( ρ ). In particular, several major radioactive beam facilities being built around the world have all listed constraining the symmetry energy as one of their major science drivers, see, e.g., Ref. [7]. Moreover, observations of neutron stars from current missions such as the Chandra X-ray and XMM-Newton observatories, and upcoming missions such as the Neutron star Interior Composition ExploreR (NICER) [8] will provide high precision data to infer more accurately neutron star radii which are very sensitive to the symmetry energy [9-12]
Significant progress has been made in recent years in constraining E sym ( ρ ) especially around ρ ≈ ρ 0 , the nuclear saturation density, using data from both terrestrial laboratory experiments and astrophysical observations [13-18]. Recent surveys of model analyses of world data found that the mean values of the symmetry energy and its density dependence at ρ 0 are consistent with 29 ≤ E sym ( ρ 0 ) ≤ 33 MeV and 40 ≤ L = 3 ρ ∂E sym ( ρ ) ∂ρ | ρ 0 ≤ 60 MeV [19, 20]. However, the decomposition of the symmetry energy into its kinetic and potential parts and its behavior at both sub-saturation ( ρ < ρ 0 ) and suprasaturation ( ρ > ρ 0 ) densities are still poorly known.
A common method to improving our knowledge of the total symmetry energy, E sym ( ρ ), is to separate it into its potential ( E kin sym ( ρ )) and kinetic ( E pot sym ( ρ )) parts,
$$E _ { s y m } ( \rho ) = E _ { s y m } ^ { k i n } ( \rho ) + E _ { s y m } ^ { p o t } ( \rho ) \quad \quad ( 1 )$$
and probing them separately [6, 12, 18]. The kinetic part of the symmetry energy, E kin sym ( ρ ), can be readily calculated from the nuclear momentum distribution. The much less well understood potential part can then be calculated as E pot sym ( ρ ) = E sym ( ρ ) -E kin sym ( ρ ).
This separation is valuable for several reasons. As E kin sym ( ρ ) and E pot sym ( ρ ) have different density dependencies (typically parametrized as E kin sym ( ρ 0 )( ρ ρ 0 ) α and E pot sym ( ρ 0 )( ρ ρ 0 ) γ ) the total symmetry energy can be more reliably extrapolated to higher densities by extrapolating its kinetic and potential parts separately. Secondly, knowledge of E pot sym ( ρ ) is important for constraining key parameters in calculations of the symmetry energy, such as three-body forces [21] and high-order chiral effective interactions [22]. These improved models then allow extrapolation of E pot sym ( ρ ) to supra-saturation densities with improved accuracy [18, 23-25]. Thirdly, knowing E kin sym ( ρ ) and E pot sym ( ρ ) separately is required to describe heavy-ion reactions and describe the isovector dynamical observables. For example, the density dependence of E sym ( ρ ) as extracted from heavy-ion collisions depends on models of E kin sym ( ρ ) [26].
The kinetic part is often approximated in a nonrelativistic free Fermi gas model [6, 12] as the per-nucleon difference between the kinetic energy of pure neutron matter at a density ρ and the kinetic energy of symmetric nuclear matter where the protons and neutrons each have density ρ/ 2:
$$E _ { s y m } ^ { k i n } ( \rho ) | _ { \text{FG} } = ( 2 ^ { \frac { 3 } { 2 } } - 1 ) \frac { 3 } { 5 } E _ { F } ( \rho ) \approx 1 2. 5 \text{MeV} ( \rho / \rho _ { 0 } ) ^ { 2 / 3 } \ ( 2 ) \quad \text{$20.\ F$} \quad \text{that} \ c \quad \text{in SN}$$
where E F ( ρ ) is the Fermi energy at density ρ .
However, short-range correlations (SRC) due to the tensor force acting predominantly between neutronproton pairs significantly increase the average momentum and hence the kinetic energy in SNM but have almost no effect in PNM. They thus reduce significantly the kinetic symmetry energy, possibly even to negative values. This has been shown recently in both phenomenological models [27] and microscopic many-body theories [28-31]. At fixed symmetry energy, E sym ( ρ ), the SRC induced decrease of E kin sym ( ρ ) increases E pot sym ( ρ ) beyond its Fermi Gas model limit of E pot sym ( ρ 0 ) = E sym ( ρ 0 ) -E kin sym ( ρ 0 ) | FG ≈ 19 1 MeV. This is important for trans-. port model simulations of heavy-ion collisions [4-6, 32, 33].
In this paper we provide a phenomenological analytical expression for the kinetic symmetry energy of correlated nucleonic matter based on calculations of nuclear momentum distributions and on data at saturation density ( ρ 0 ) from inclusive ( e, e ′ ) and exclusive ( e, e pN ′ ) scattering experiments at the Thomas Jefferson National Accelerator Facility (JLab) [34-38]. We give credence to our model by comparing to a transport model analysis of nucleon emission data in intermediate energy heavy-ion collisions [39, 40] and to many-body theoretical calculations of nuclei and nuclear matter [28-31, 41, 42]. Last we use the known values of the total symmetry energy, E sym ( ρ ), and its density dependence, L , at saturation density to extract the total symmetry energy at supranuclear densities and to constrain the value and density dependence of the potential part of the symmetry energy.
It has long been known that the tensor force induced SRC leads to a high-momentum tail in the single-nucleon momentum distribution around 300-600 MeV/c [43, 44]. This high-momentum tail scales, i.e., its shape is almost identical for all nuclei from deuteron to infinite nuclear matter, see, e.g. Refs. [42, 45, 46]. This is shown by the constancy of the ratio of the per nucleon inclusive ( e, e ′ ) cross sections for nucleus A to the deuteron, a 2 ( A ), for Bjorken scaling parameter x B between about 1 5 and 1 9 . . [34-36, 47]. The ratio of the momentum distribution in nucleus A to the deuteron for 300 ≤ k ≤ 600 MeV/c is just the cross section ratio a 2 ( A ). Extrapolation of the measured a 2 ( A ) to infinite SNM using three different techniques [48-50], yield an average value of a 2 ( ∞ ) = 7 ± 1. The uncertainty in the extrapolation represents about 50% of the difference between a 2 ( A ) ≈ 5 for heavy nuclei and a 2 ( ∞ ) = 7 for SNM.
Exclusive two-nucleon knockout experiments [37, 38, 51-53] show that, for 300 ≤ k ≤ 600 MeV/c, proton knockout is accompanied by a recoil second nucleon and that second nucleon is predominantly a neutron, i.e., that np -SRC pairs dominate over pp pairs by a factor of about 20. For recent reviews, see Refs. [54, 55]. This implies that correlations are about 20 times smaller in PNM than in SNM. Since the integral of the deuteron momentum distribution from 300 to 600 MeV/c is about 4% [56] and a 2 ( ∞ ) = 7 ± 1, the probability to find a high-momentum nucleon in SNM is about 28% and in PNM is about 1 -2%.
The deuteron momentum distribution, n d ( k ), decreases as 1 /k 4 for 300 ≤ k ≤ 600 MeV/c [57]. Since the nuclear momentum distribution, n A ( k ), in that range is predominantly due to np -SRC pairs and since it is proportional to the deuteron distribution, we can write that n A ( k/k F )( k/k F ) 4 = R a d 2 ( A ), where R d = 0 64 . ± 0 10 . is extracted from the deuteron momentum distribution, and k F is the Fermi momentum [57]. At higher momenta, the momentum distribution n k ( ) drops much more rapidly.
This is supported by 'exact' variational Monte Carlo (VMC) momentum distributions calculated [41] for 4 He and 10 B which decrease as k -4 for np pairs with small pair center-of-mass momentum for nucleon momenta 1 2 . < k/k F < 3 to within about 10%.
FIG. 1: (Color online) The per-nucleon kinetic energy calculated using the Correlated Fermi Gas (CFG) model (diagonal red band) for atomic nuclei from 12 C to 208 Pb. The calculated kinetic energy is shown as a function of λ , the highmomentum tail cutoff parameter. The vertical blue band shows the constraints on λ from the deuteron momentum distribution. The diagonal red band reflects the model uncertainties. Also shown are the results from the uncorrelated Fermi Gas model (dashed purple line) and a horizontal black band spanning the results from many-body nuclear calculations for various nuclei from 12 C to 208 Pb [42] and from exact variational Monte Carlo (VMC) calculations for 12 C [41].
We therefore model n k ( ) for SNM with a depleted Fermi gas region and a correlated high-momentum tail:
$$n _ { S N M } ^ { S R C } ( k ) = \begin{cases} \begin{array} { c c } A _ { 0 } & k < k _ { F } \\ C _ { \infty } / k ^ { 4 } & k _ { F } < k < \lambda k _ { F } ^ { 0 } \\ 0 & k > \lambda k _ { F } ^ { 0 } \end{array} \quad & ( 3 ) \end{cases} \quad & \Sigma _ { \infty }$$
where C ∞ = R a d 2 ( ∞ ) k F ≡ c 0 k F is the phenomenological height factor [57], c 0 = 4 16 . ± 0 95, . k 0 F is the Fermi momentum at ρ 0 and λ ≈ 2 75 . ± 0 25 is the high-momentum . cutoff obtained from the momentum distribution of the deuteron [57]. A 0 is a constant given by
$$A _ { 0 } = \frac { 3 \pi ^ { 2 } } { ( k _ { F } ^ { 0 } ) ^ { 3 } } \frac { \rho _ { 0 } } { \rho } \left [ 1 - \left [ 1 - \frac { 1 } { \lambda } ( \frac { \rho } { \rho _ { 0 } } ) ^ { 1 / 3 } \right ] \frac { c _ { 0 } } { \pi ^ { 2 } } \right ], \quad ( 4 )$$
determined by the normalization
$$\frac { 4 \pi } { ( 2 \pi ) ^ { 3 } } \int _ { 0 } ^ { \lambda k _ { F } ^ { 0 } } 2 n _ { S N M } ^ { S R C } ( k ) k ^ { 2 } d k \equiv 1. \quad \ \ ( 5 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
Based on the JLab data [38], fewer than 2% of neutrons belong to nn -SRC pairs. We thus use the free Fermi gas model for PNM and include the 2% upper limit for correlated neutrons in our estimate of the uncertainty band. In what follows we refer to this as the Correlated Fermi Gas (CFG) model.
The per nucleon kinetic energy of nuclei and of symmetric nuclear matter can then be calculated from the
FIG. 2: (Color online) The per-nucleon kinetic energy for symmetric nuclear matter calculated using the Correlated Fermi Gas (CFG) model (red band). The calculated kinetic energy is shown as a function of λ , the high-momentum tail cutoff parameter. The blue band shows the constraints on λ from the deuteron momentum distribution. The red band reflects the model uncertainties. Also shown are the results from the uncorrelated Fermi Gas model (dashed purple line), the Brueckner-Hartree-Fock (BHF) model using the AV-18 interaction [28], and the Self-Consistent Greens Function (SCGF) approach using the CDBonn, N3LO, and AV18 nucleon-nucleon interactions [30, 31].
FIG. 3: (Color online) The per-nucleon kinetic symmetry energy at saturation density, E kin sym ( ρ 0 ), calculated using the Correlated Fermi-Gas model (diagonal red band) as a function of λ , the high-momentum tail cutoff parameter. The dashed purple line shows the results of the uncorrelated Fermi Gas model. The green band shows the results from transport model analyses of Sn+Sn collisions described in the text. Also shown for comparison are the results from microscopic calculations: Brueckner-Hartree-Fock (BHF) [28], Fermi-HyperNetted-Chain (FHNC) [29] and the Self-Consistent Greens Function (SCGF) using the CDBonn, N3LO, Nij1, and AV18 nucleon-nucleon interactions [30, 31].
momentum distribution using
$$E _ { k i n } = \frac { 4 \pi } { ( 2 \pi ) ^ { 3 } } \int _ { 0 } ^ { \infty } \frac { \hbar { ^ { 2 } } k ^ { 2 } } { 2 m } n ( k ) k ^ { 2 } d k. \quad \quad ( 6 )$$
Fig. 1 shows the resulting kinetic energy for finite nuclei, calculated within the CFG model using a 2 ( A ) = 5 ± 0 3. . The CFG kinetic energy is much larger than that of the uncorrelated Fermi Gas. It agrees with the kinetic energies from many-body nuclear calculations for 12 C, 16 O, 40 Ca, 56 Fe, and 208 Pb [42] and from VMC calculations for 12 C [41].
Fig. 2 shows the average nucleon kinetic energy for SNM, E kin SNM ( ρ 0 ) calculated at saturation density, and shown as a function of λ . The CFG calculation is done using a 2 ( ∞ ) = 7 ± 1 and R d = 0 64 . ± 0 10, and is com-. pared with the free Fermi gas model and the predictions of several microscopic models [28-31]. The error band on the CFG results combines estimated uncertainties in R d and a 2 ( ∞ ). The self-consistent Green's function (SCGF) calculations of the kinetic energy of symmetric nuclear matter, E kin SNM ( ρ 0 ) [30, 31], agree with our CFG calculation (Fig. 2).
Almost all phenomenological and microscopic manybody theories lead to Equations of State (EOS) of asymmetric nucleonic matter that vary quadratically with the isospin-asymmetry δ = ( ρ n -ρ p ) / ρ ( n + ρ p ) according to the so-called empirical parabolic law E ρ, δ ( ) = E ρ, δ ( = 0) + E sym ( ρ δ ) 2 + O δ ( 4 ) . The coefficient of the δ 4 term
at ρ 0 has been found to be less than 1 MeV [33]. The symmetry energy can thus be calculated equally accurately from either the energy difference between PNM and SNM, i.e., E sym ( ρ ) = E ρ, ( 1) -E ρ, ( 0) or the curvature E sym ( ρ ) = 1 2 ∂ 2 E ρ,δ ( ) ∂δ 2 at any δ .
However, it has never been tested whether the empirical parabolic law is valid separately for the kinetic and potential parts of the EOS. While the free Fermi gas kinetic energy satisfies the parabolic law, models that include SRC may not [58]. To be consistent and compare with the free Fermi gas model and microscopic many-body theories, we will define the kinetic symmetry energy of correlated nucleonic matter as E kin sym ( ρ ) = E kin PNM ( ρ ) -E kin SNM ( ρ . ) We add a SRC correction term to the Fermi Gas symmetry energy to get the full kinetic symmetry energy:
$$E _ { s y m } ^ { k i n } ( \rho ) = E _ { s y m } ^ { k i n } ( \rho ) | _ { \text{FG} } - \Delta E _ { s y m } ^ { k i n } ( \rho ) \quad \ \ ( 7 ) \quad \text{ use.}$$
where the SRC correction term is:
$$\Delta E ^ { k i n } _ { s y m } \equiv \frac { E ^ { 0 } _ { F } } { \pi ^ { 2 } } c _ { 0 } \left [ \lambda ( \frac { \rho } { \rho _ { 0 } } ) ^ { 1 / 3 } - \frac { 8 } { 5 } ( \frac { \rho } { \rho _ { 0 } } ) ^ { 2 / 3 } + \frac { 3 } { 5 } \frac { 1 } { \lambda } ( \frac { \rho } { \rho _ { 0 } } ) \right ]. \quad \text{the $\chi^{i}$} \\ \mathfrak { k } - \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \ dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots\dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \delta$}$$
As one expects, the SRC correction increases with both the height ( c 0 = C ∞ /k F = R a d 2 ( ∞ )) and width ( λ ) of the high-momentum tail in SNM.
Fig. 3 shows the kinetic symmetry energy, E kin sym ( ρ 0 ) calculated at saturation density assuming a free Fermi gas model for PNM and shown as a function of λ . The error band on the CFG results combines estimated uncertainties in R , a d 2 ( ∞ ) and the amount of SRC in PNM ( < 2%). Within the uncertainty range of the parameter λ = 2 75 . ± 0 25, . E kin sym ( ρ 0 ) is found to be between -2 5 and . -17 5 MeV, much less than the free Fermi gas . result of ≈ +12 5 MeV. The microscopic many-body the-. ories yield results that are significantly smaller than the free Fermi gas prediction but significantly larger than our CFG model. Despite the agreement between our CFG model and the SCGF calculations of the kinetic energy of symmetric nuclear matter, E kin SNM kin ( ρ 0 ) [30, 31], the SCGF symmetry energy, E kin sym ( ρ 0 ) = E PNM ( ρ 0 ) -E kin SNM ( ρ 0 ), is significantly larger than our model's. This is because the SCGF calculations include about 10% correlations in PNM.
To further validate our CFG model, we perform a transport model analysis of nucleon emission data in intermediate energy heavy-ion collisions. The dynamics of heavy-ion collisions around the Fermi energy are sensitive to the density dependence of the nuclear symmetry energy around ρ 0 [32, 33]. Specifically, the ratio of free neutrons to protons emitted in heavy-ion collisions was found to be sensitive to the symmetry energy [4]. This ratio has been measured recently in 124 Sn+ 124 Sn and 112 Sn+ 112 Sn reactions at E beam /A = 50 and 120 MeV at MSU [40] with improved precision as compared to earlier measurements [39]. The data are given for the double ratio of neutrons to protons in 124 Sn+ 124 Sn to 112 Sn+ 112 Sn reactions to reduce systematic errors associated with neutron detection.
Using the Isospin-dependent Boltzmann-UehlingUhlenbeck (IBUU) transport model [33], analysis of this double ratio was done by introducing two parameters, η and γ , to describe the potential symmetry energy:
$$\mathbf n _ { \mathbb { L } S } \quad E _ { s y m } ^ { p o t } ( \rho ) = [ E _ { s y m } ( \rho _ { 0 } ) - \eta E _ { s y m } ^ { k i n } ( \rho _ { 0 } ) | _ { \text{FG} } ] ( \rho / \rho _ { 0 } ) ^ { \gamma }. \quad ( 9 )$$
Without considering the momentum dependence of nuclear potentials, the corresponding symmetry potential is then
$$\begin{array} { c c c } - & \\ \text{to} & V _ { \text{sym} } ^ { n / p } ( \rho, \delta ) & = & [ E _ { \text{sym} } ( \rho _ { 0 } ) - \eta E _ { \text{sym} } ^ { k i n } ( \rho _ { 0 } ) | _ { \text{FG} } ] ( \rho / \rho _ { 0 } ) ^ { \gamma } \\ \text{ic} & & \times & [ \pm 2 \delta + ( \gamma - 1 ) \delta ^ { 2 } ], & ( 1 0 ) \end{array}$$
where the 2 δ term dominates. The ± sign is due to the fact that neutrons and protons feel repulsive and attractive symmetry potentials respectively.
We varied η and γ on a large 2D fine lattice to minimize the χ 2 between the model calculations and the MSU data at both beam energies. We then performed a covariance analysis to find the uncertainties of η and γ corresponding to a ± 1 σ error band using the method reviewed recently in Refs. [59, 60]. We used an impact parameter of 3 fm, consistent with that estimated for the data [61]. Free nucleons are identified as those with local densities less than ρ / 0 8 at the time of their final freeze-out from the reaction. Calculations using a phase-space coalescence model lead to similar results within the error band [26].
Fig. 4 shows the double free neutron/proton ratios in the two 124 and 112 Sn+Sn reactions at E beam /A = 50 MeV/nucleon [40]. The calculations (red band) shown used the optimized parameters η 0 = -0 30(1 . ± 18 53%) .
FIG. 4: (Color online) The calculated double ratio of free neutron/protons in the two reactions in comparison with the MSU data for transversely emitted nucleons in the angular range of 70 0 ≤ θ cms ≤ 110 0 [40]. The bands represent 1 σ uncertainty of the calculations.
TABLE I: Density dependence parameter, γ , of the potential part of the symmetry energy extracted within the Correlated Fermi Gas (CFG) and Free Fermi Gas (FG) models, assuming a total symmetry energy of E sym ( ρ 0 ) ≈ 31 MeV. Also shown are the value of γ and its 1 σ and 2 σ confidence intervals, extracted from analysis of heavy ion collision data [6] and neutron stars observations [12], assuming a Free FG model. The assumed value of the kinetic symmetry energy at saturation density used in each extraction is also listed.
| | E kin sym ( ρ 0 ) [MeV] | γ ± 1 σ (2 σ ) |
|---------------------|---------------------------|-------------------------------------------------|
| CFG | - 10 ± 3 | 0 . 25 ± 0 . 05 |
| FG | - 10 ± 3 0 12 . 5 | 0 . 58 ± 0 . 05 0 . 55 ± 0 . 06 0 . 48 ± 0 . 10 |
| | 17 . 0 | 0 . 41 ± 0 . 13 |
| Steiner et al. [12] | 17 . 0 | 0 . 3 +0 . 1(0 . 5) - 0 . 1 |
that corresponds to E kin sym ( ρ 0 ) = -(3 8 . ± 0 7) MeV and . γ 0 = 0 80(1 . ± 5 98%) with . a χ 2 0 = 8. This value of E kin sym ( ρ 0 ) was determined without considering the momentum dependence of the symmetry potential known to decrease somewhat the free neutron/proton ratio [62]. It thus represents an upper bound on the kinetic symmetry energy used to reproduce the MSU data within the IBUU model. For comparison, results with a χ 2 = 21 using E kin sym ( ρ 0 ) | FG = 12 5 MeV and . γ = 0 8 are also . shown. Calculations with E kin sym ( ρ 0 ) | FG and other values of γ between 0.4 and 1 leads to even higher χ 2 values.
The value of E kin sym ( ρ 0 ) determined from the IBUU transport analysis of the neutron to proton ratios in Sn+Sn collisions is consistent with that calculated using our CFG model (see Fig. 3).
We now turn to extracting the total symmetry energy at supra-nuclear densities and the density dependence of its potential part using the CFG model. We use the general form of the total symmetry energy given by Eq. 1, with the CFG corrections to the kinetic energy term given by Eq. 7 and 8. As detailed above, by comparing the CFG model results to the known values of the total symmetry energy ( E sym ( ρ ) = 31 0 . ± 1(1 σ ) MeV [19]) we can extract the value of the potential part of the symmetry energy at saturation density: E pot sym ( ρ 0 ) = E sym ( ρ 0 ) -E kin sym ( ρ 0 ). Simillarly, using the known density dependence of the total symmetry energy at saturation density ( L = 50 ± 5(1 σ ) MeV [19]) we can extract the density dependence of the potential part of the symmetry energy:
$$\gamma = \frac { \frac { 1 } { 3 } L - \frac { d E _ { s y m } ^ { k i n } ( \rho ) } { d \rho } | _ { \rho _ { 0 } } } { E _ { s y m } ( \rho _ { 0 } ) - E _ { s y m } ^ { k i n } ( \rho _ { 0 } ) }. \quad \quad ( 1 1 ) \quad \begin{array} { c } \text{os} \\ \text{den} \\ \text{svn} \end{array}$$
Our results are summarized in Table I where we list the value of γ extracted using the CFG model. This is compared with free fermi gas model results (i.e. α = 2 3) / assuming different values for the kinetic symmetry en-
FIG. 5: (color online) The density dependence of the kinetic, potential and total symmetry energy extracted using the CFG and FG models. See text for details.
ergy (i.e. E kin sym ( ρ 0 ) = -10 0 12 5 17 MeV), and with , , . , recent analyses of heavy ion collisions [6] and neutron star data [12] which also assume a free fermi gas model (i.e. α = 2 3). / As can be seen, even within the FG model, the value of γ varies significantly depending on the value of the kinetic symmetry energy. Furthermore, CFG and FG results for the same kinetic symmetry energy also differ due to the density dependence of the SRC correction term (eq. 8). The value of γ obtained from the neutron star analysis of Ref. [12] is very similar to that of the CFG model.
Fig. 5 shows the density dependence of the kinetic, potential and total symmetry energy obtained using both the CFG and FG models. While the two models differ significantly in the values and density dependences of their kinetic and potential parts, their total symmetry energies are almost identical.
To summarize, we provide an analytical expression for a kinetic symmetry energy of correlated nucleonic matter at ρ = ρ 0 , using the dominance of short-range correlated neutron-proton pairs at high momentum observed in electron scattering data. Our model yields E kin sym ( ρ 0 ) = -10 ± 7 5 . MeV, significantly lower than E kin sym ( ρ 0 ) = +12 5 MeV of the widely-used free Fermi . gas model. This result is consistent with our analysis of recent data on the free proton-to-neutron ratios measured in intermediate energy nucleus-nucleus collisions as well as with microscopic many-body calculations, and previous phenomenological extractions. We also extract the density dependence of E pot sym ( ρ ) and E sym ( ρ ) from our model of E kin sym ( ρ ) together with the value of the total symmetry energy and its density dependence at saturation density. While the total symmetry energy exacted using different models is consistent, its separation into kinetic and potential parts is not.
We thank F.J. Fattoyev, X. H. Li, W.G. Newton, Z.Z
Shi, A. Gal, M. M. Sargsian, G. A. Miller, D. W. Higinbotham, M. Strikman, and L. Frankfurt for helpful discussions. O. Hen and E. Piasetzky are supported by the Israel Science Foundation. B.A. Li is supported in part by the US National Science Foundation under Grant No. PHY-1068022, US National Aeronautics and Space Administration under Grant No. NNX11AC41G issued through the Science Mission Directorate, the CUSTIPEN (China-U.S. Theory Institute for Physics with Exotic Nuclei) under DOE grant number DE-FG02-13ER42025 and the National Natural Science Foundation of China under Grant No. 11320101004. W.J. Guo is supported by the National Natural Science Foundation of China (10905041) and the China Scholarship Council Foundation (201208310156). L.B. Weinstein is supported by the US Department of Energy under Grants de-sc00006801 and DE-FG02-96ER40960.
- ∗ Contact Author [email protected]
- [1] J.M. Lattimer, C.J. Pethick, M. Prakash, P. Haensel, Phys. Rev. Lett. 66, 2701 (1991).
- [2] M. Prakash, T.L. Ainsworth, J.M. Lattimer, Phys. Rev. Lett. 61, 2518 (1988).
- [3] N. Nikolov, N. Schunck, W. Nazarewicz, M. Bender, and J. Pei, Phys. Rev. C 83, 034305 (2011).
- [4] B.A. Li, C.M. Ko, Z.Z. Ren, Phys. Rev. Lett. 78, 1644 (1997).
- [5] L.W. Chen, C.M. Ko, B.A. Li, Phys. Rev. Lett. 94, 032701 (2005).
- [6] M.B.Tsang et al., Phys. Rev. Lett. 102, 122701,(2009).
- [7] A. B. Balantekin et al., Modern Physics Letters A 29 , 30010 (2014).
- [8] C. Kouveliotou et al., Enduring Quests-Daring Visions (NASA Astrophysics in the Next Three Decades). arXiv: 1401.3741, January 2014.
- [9] A.W. Steiner, M. Prakash, J.M. Lattimer, P.J. Ellis, Phys. Rep. 411 , 325 (2005).
- [10] B. A. Li and A. W. Steiner, Phys. Lett. B 642 , 436 (2006).
- [11] W.G. Newton, M. Gearheart and B.A. Li, APJ Supplementary Series 204, 9 (2013).
- [12] A. W. Steiner, J. M. Lattimer, and E. F. Brown, Astrophys. J. 722, 33 (2010).
- [13] W.G. Lynch et al. , Prog. Nucl. Part. Phys. 62 , 427 (2009).
- [14] W. Trautmann and H. H. Wolter, Int. J. Mod. Phys. E 21 , 1230003 (2012).
- [15] M. B. Tsang, et al., Phys. Rev. C 86 , 015803 (2012).
- [16] B.A. Li, A. Ramos, G. Verde, and I. Vida˜a, eds., 'Topn ical issue on nuclear symmetry energy', Eur. Phys. J. A 50 , No. 2, (2014).
- [17] C.J. Horowitz et al., J. of Phys. G 41 , 093001 (2014).
- [18] J. M. Lattimer, Annu. Rev. Nucl. Part. Sci. 62 , 485 (2012).
- [19] J.M. Lattimer and Y. Lim, Astrophys. J. 771, 51 (2013).
- [20] B.A. Li, X. Han, Phys. Lett. B 727 , 276 (2013).
- [21] S. Gandolfi, J. Carlson, S. Reddy, A. W. Steiner, and R. B. Wiringa, Eur. Phys. J A 50 (2014) 10.
- [22] K. Hebeler and A. Schwenk, Eur. Phys. J A 50 (2014) 11.
- [23] A. W. Steiner, S. Gandolfi Phys. Rev. Lett. 108 , 081102 (2012).
- [24] K. Hebeler, J. M. Lattimer, C. J. Pethick, A. Schwenk Astrophys.J. 773 (2013) 11.
- [25] A. Gezerlis, I. Tews, E. Epelbaum, S. Gandolfi, K. Hebeler, A. Nogga, A. Schwenk Phys. Rev. Lett. 111 , 032501 (2013).
- [26] B.A. Li, W.J. Guo and Z.Z. Shi, arXiv:1408.6415 (2014).
- [27] C. Xu, A. Li, B.A. Li, J. of Phys: Conference Series 420, 012190 (2013).
- [28] I. Vidana, A. Polls, C. Providencia, Phys Rev C 84, 062801(R) (2011).
- [29] A. Lovato, O. Benhar, S. Fantoni, A. Yu. Illarionov, and K. E. Schmidt, Phys. Rev. C 83, 054003 (2011).
- [30] A. Carbone, A. Polls, A. Rios, Eur. Phys. Lett. 97, 22001 (2012).
- [31] A. Rios, A. Polls, W. H. Dickhoff, Phys. Rev. C 89 , 044303 (2014).
- [32] V. Baran, M. Colonna, V. Greco, and M. Di Toro, Phys. Rep. 410 , 335 (2005).
- [33] B. A. Li, L. W. Chen, and C. M. Ko, Phys. Rep. 464 , 113 (2008).
- [34] K. Egiyan, et al. (The CLAS Collaboration), Phys. Rev. C 68, 014313 (2003).
- [35] K. Egiyan, et al. (The CLAS Collaboration), Phys. Rev. Lett. 96, 082501 (2006).
- [36] N. Fomin, et al. (The Hall C Collaboration), Phys. Rev. Lett. 108, 092502 (2012).
- [37] O. Hen et al. (The CLAS Collaboration), Science 346, 614 (2014).
- [38] R. Subedi, et al. (The Hall A Collaboration), Science 320 , 1476 (2008).
- [39] M.A. Famiano et al., Phys. Rev. Lett. 97, 052701 (2006).
- [40] D.S. Coupland et al., arXiv:1406.4546
- [41] R.B. Wiringa, R. Schiavilla, Steven C. Pieper, and J. Carlson, Phys. Rev. C 89 , 024305 (2014).
- [42] C. Ciofi degli Atti and S. Simula, Phys. Rev. C 53, 1689 (1996).
- [43] H.A. Bethe, Ann. Rev. Nucl. Part. Sci. 21 , 93 (1971).
- [44] A.N. Antonov, P.E. Hodgson and I.Z. Petkov, Nucleon Momentum and Density Distributions in Nuclei (Clarendon Press, Oxford, 1988).
- [45] S. Fantoni and V.R. Pandharipande, Nucl. Phys. A 427 , 473 (1984).
- [46] S.C. Pieper, R.B. Wiringa and V.R. Pandharipande, Phys. Rev. C 46 ,1741 (1992).
- [47] D.B. Day et al. , Phys. Rev. C 40 , 1011 (1989).
- [48] C. Ciofi degli Atti, E. Pace, and G. Salme, Phys Rev C 43 , 1155 (1991).
- [49] Michael McGauley and Misak M. Sargsian, arXiv 1102.3973.
- [50] E. Piasetzky, O. Hen, and L. B. Weinstein, AIP Conf. Proc. 1560, 355 (2013).
- [51] E. Piasetzky et al., Phys. Rev. Lett. 97 , 162504 (2006).
- [52] A. Tang et al., Phys. Rev. Lett. 90 , 042301 (2003).
- [53] H. Baghdasaryan, et al. (The CLAS Collaboration), Phys. Rev. Lett. 105 , 222501 (2010).
- [54] L. Frankfurt, M. Sargsian and M. Strikman, Int. Jour. Mod. Phys. A 23 , 2991 (2008).
- [55] J. Arrington, D.W. Higinbotham, G. Rosner G and M. Sargsian, Prog. Part. Nucl. Phys. 67, 898 (2012).
- [56] I. Passchier et al., Phys. Rev. Lett. 88 , 102302 (2002).
- [57] O. Hen, L. B. Weinstein, E. Piasetzky, G. A. Miller, M. M. Sargsian, Y. Sagi, arXiv:1407.8175.
- [58] X.H. Li et al., arXiv:1403.5577 (2014).
- [59] J. Dobaczewski, W. Nazarewicz, P.-G. Reinhard, J. Phys. G: Nucl. Part. Phys. 41, 074001 (2014).
- [60] J. Piekarewicz, Wei-Chia Chen and F.J. Fattoyev,
- arXiv:1407.0911
- [61] Z.Chajecki and Betty Tsang, private communications (2014).
- [62] B.A. Li, C. B. Das, S. Das Gupta, C. Gale, Nucl. Phys. A 735, 563 (2004).
|
10.1103/PhysRevC.91.025803
|
[
"Or Hen",
"Bao-An Li",
"Wen-Jun Guo",
"L. B. Weinstein",
"Eli Piasetzky"
] |
2014-08-04T19:05:38+00:00
|
2015-02-06T17:24:45+00:00
|
[
"nucl-ex",
"astro-ph.SR",
"nucl-th"
] |
Symmetry Energy of Nucleonic Matter With Tensor Correlations
|
The nuclear symmetry energy (E_sym(\roh)) is a vital ingredient of our
understanding of many processes, from heavy-ion collisions to neutron stars
structure. While the total nuclear symmetry energy at nuclear saturation
density (\rho_0) is relatively well determined, its value at supranuclear
densities is not. The latter can be better constrained by separately examining
its kinetic and potential terms and their density dependencies. The kinetic
term of the symmetry energy, E_sym^kin(\rho_0), equals the difference in the
per-nucleon kinetic energy between pure neutron matter (PNM) and symmetric
nuclear matter (SNM), often calculated using a simple Fermi gas model. However,
experiments show that tensor force induced short-range correlations (SRC)
between proton-neutron pairs shift nucleons to high-momentum in SNM, but have
almost no effect in PNM. We present an approximate analytical expression for
E_sym^kin(\rho) of correlated nucleonic matter. In our model, E_sym^kin(\rho_0)
= -10 MeV, which differs significantly from +12.5 MeV for the widely-used free
Fermi gas model. This result is consistent with our analysis of recent data on
the free proton-to-neutron ratios measured in intermediate energy
nucleus-nucleus collisions as well as with microscopic many-body calculations,
and previous phenomenological extractions. We then use our calculated
E_sym^kin(\rho) in combination with the known total symmetry energy and its
density dependence at saturation density to constrain the value and density
dependence of the potential part and to extrapolate the total symmetry energy
to supranuclear densities.
|
1408.0773v3
|
## Pre-Breakdown Processes in Dielectric Fluid in Inhomogeneous Pulsed Electric Fields
Mikhail N. Shneider 1,* and Mikhail Pekker 2
- 1 Department of Mechanical and Aerospace Engineering, Princeton University, Princeton, NJ, 08544
2 MMSolution, 6808 Walker Street, Philadelphia, PA, 19135
We consider the development of pre-breakdown cavitation nanopores appearing in the dielectric fluid under the influence of the electrostrictive stresses in the inhomogeneous pulsed electric field. It is shown that three characteristic regions can be distinguished near the needle electrode. In the first region, where the electric field gradient is greatest, the cavitation nanopores, occurring during the voltage nanosecond pulse, may grow to the size at which an electron accelerated by the field inside the pores can acquire enough energy for excitation and ionization of the liquid on the opposite pore wall, i.e., the breakdown conditions are satisfied. In the second region, the negative pressure caused by the electrostriction is large enough for the cavitation initiation (which can be registered by optical methods), but, during the voltage pulse, the pores do not reach the size at which the potential difference across their borders becomes sufficient for ionization or excitation of water molecules. And, in the third, the development of cavitation is impossible, due to an insufficient level of the negative pressure: in this area, the spontaneously occurring micropores do not grow and collapse under the influence of surface tension forces. This paper discusses the expansion dynamics of the cavitation pores and their most probable shape.
## I. Introduction.
A mechanism for the rapid breakdown in the fluid, associated with the occurrence of cavitation ruptures under the influence of the electrostrictive forces near the needle electrode, was proposed in [1]. Later, a hydrodynamic model of compressible fluid motion under the influence of the ponderomotive electrostrictive forces in a non-uniform time dependent electric field was suggested in [2]. As shown in [2], if the voltage on the needle electrode is growing fast enough (a few nanoseconds), the negative stress is created in liquid, and it may be sufficient for the cavitation formation. A nanosecond breakdown, the beginning of which can be explained by the formation of cavitation in a stretched liquid due to electrostriction, was investigated experimentally in [3-7]. In [8], based on a theoretical model [2], it had been shown experimentally that the initial stage of development of a nanosecond breakdown in liquids is associated with the appearance of discontinuities in the liquid (cavitation) under the influence of electrostriction forces. The comparison of the experimentally measured area dimensions and its temporal development was found to be in a good agreement with the theoretical calculations. A theory of the cavitation initiation in inhomogeneous pulsed electric field was developed in [9], as well as the method that allows to determine the critical parameters, at which cavitation begins, on the basis of the comparison between the experiment and the simulation results within the framework of hydrodynamics of compressible fluids, was proposed.
In Part II, it is shown in the model example of a spherical electrode that if the applied voltage on the electrode is growing fast enough (~ few nanoseconds), the negative stress induced in liquid by the ponderomotive electrostrictive forces does not have time to be compensated by the rising of the positive hydrostatic pressure, related to the fluid influx into the negative pressure region. Т herefore, the conditions arise for the formation of the fluid discontinuities (nanovoids), i.e., cavitation.
* [email protected]
Part III presents the computed probabilities of the critical bubble appearance in the vicinity of the electrode at different time moments
Part IV presents the equations for the nanovoid expansion in the presence of a strong inhomogeneous electric field and the approaches to this problem, which will be considered, are formulated. The presented numerical results of the nanopores expansion in the case of the spherical electrode can be easily generalized to an arbitrarily shaped electrode. It is shown that there are three characteristic regions in the vicinity of the electrode. In the first region, where the electric field gradient is greatest, the occurring cavitation nanopores have enough time during the nanosecond voltage pulse to grow to a size at which an electron can gain enough energy for the excitation and ionization of the liquid molecules on the pore wall. In the second region, the electrostrictive negative pressure reaches values at which the cavitation development becomes possible (that can be registered by the optical methods), but the nanovoids, appearing during the voltage pulse, do not have enough time to grow to the size at which the potential difference across their borders becomes sufficient for the ionization or excitation of water molecules. And, in the third, the development of cavitation is impossible, since the spontaneously occurring nanovoids do not grow, because the value of the electrostrictive negative pressure is relatively small and cannot compete with the forces of surface tension.
Part V discusses the form of an expanding cavitation micropore. It is shown that a perturbation of the electric field caused by the micropores near the equator (in the plane perpendicular to the direction of the undisturbed electric field) produces a flow directed to the pore, and near the poles - out of the pore. The velocities of the arising flow are comparable to the velocities of the spherical pores expansion and an order of magnitude greater than the flow velocities caused by the unperturbed nonuniform electric field due to the applied voltage on the electrode. As a consequence, the micropore is stretched along the electric field, as its extension is decelerated in the equatorial plane due to the counterflow of the liquid, and the electrostrictive negative pressure is compensated by the hydrostatic pressure.
Note, that the evolution of gas filled bubbles in a dielectric liquid within a strong electric field is theoretically and experimentally studied [10, 11]. Their elongation along the field was due to the fact that the component of the compressive pressure on the poles of the electric field in a spherical bubble in 1/ ε ( ε is the dielectric permittivity of the liquid) times smaller than at the equator. In these papers, the effect of the ponderomotive force on the bubble has not been considered because in a stationary (or slowly varying in time) electric field, the negative pressure in the fluid is compensated by the hydrostatic pressure. In this paper, we consider the case of a rapid variation of the nonuniform electric field when the ponderomotive forces cannot be neglected and they (not the gas in the bubble) lead to a rapid expansion of the micropore.
In the final part of this work, the conclusions are formulated.
## II. Motion of the fluid under the action of the ponderomotive forces
The dynamics of a dielectric liquid (water) in a pulsed inhomogeneous electric field in the approximation of the compressible fluid dynamics can be consider within the standard system of the continuity of mass and momentum equations [12]:
$$& \frac { \partial \rho } { \partial t } + \nabla ( \rho \vec { u } ) = 0 \\ & \rho \frac { \vec { u } } { \partial } + \rho ( \vec { u } \cdot \nabla ) \vec { u } = - \frac { \varepsilon _ { 0 } } { 2 } E ^ { 2 } \nabla \varepsilon + \frac { \varepsilon _ { 0 } } { 2 } \nabla \left ( E ^ { 2 } \frac { \partial \varepsilon } { \partial \rho } \rho \right ) - \nabla p + \nu \partial \vec { u } \\ & \text{where the first and the second terms in the right part of the equation (1) are the volumetric}$$
where the first and the second terms in the right part of the equation (1) are the volumetric densities of ponderomotive forces [13,14,15], 0 ε is the vacuum dielectric permittivity, ρ is the liquid density, u r - velocity, is the electric field, 6 10 -= ν m /s is the kinematic viscosity of 2 water. The relation of the hydrodynamic pressure p with the compressible fluid density ρ (in our case - water) is given by the Tait equation of the state [16,17]: E r
$$\text{$\text{$eval $size$} ] $ is given $voy $ \int $voy $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \int $ \\ p = & \left ( p _ { 0 } + B \right ) \left ( \frac { \rho } { \rho _ { 0 } } \right ) ^ { \gamma } - B \right ), \\ \quad \quad \quad$$
$$\begin{array} { c } \dots \mapsto _ { 0 } \, / \\ \rho _ { 0 } = 1 0 0 0 \ k g / m ^ { 3 }, \ \ p _ { 0 } = 1 0 ^ { 5 } \ P a, \ \ B = 3. 0 7 \cdot 1 0 ^ { 8 } \ P a, \ \ \gamma = 7. 5 \\ \text{For the polar dielectrics fluid (water)} \\ \frac { \partial \varepsilon } { \partial \rho } \, \rho = \alpha \varepsilon \end{array}$$
$$\frac { \partial \varepsilon } { \partial \rho } \rho = \alpha \varepsilon$$
where 5 is the empirical factor for most of the studied polar dielectric liquids, . 1 3 . 1 ~ -α including water, [18, 19] and for the nonpolar dielectrics ( )⋅ ( ) ε ε + ε α = 3 / 2 1 [20].
In the Figures 1 and 2, the results of the calculations performed for a spherical electrode of radius m are shown for the voltage amplitude 100 0 μ = r kV , 54 0 = U 81 and the electric field = ε varying linearly with time:
$$\stackrel { \lambda } { E ( r, t ) } = \frac { U _ { 0 } r _ { 0 } ^ { \lambda } } { r ^ { 2 } } \frac { t } { t _ { 0 } } \,,$$
Here, 0 t is the duration of the pulse front. The results shown in Fig. 1 correspond to ns t 3 0 = , and in Fig. 2, to ns t 100 0 = .
Fig. 1. (a) The pressure dependence on the distance from the electrode at t = 0=3ns. 1 - is the pressure t associated with the electrostriction forces ( 2 0 5 . 0 E PE ε αε -= ), 2 - the hydrostatic pressure associated with the change in density of the fluid, 3 - the total pressure equal to the sum of the electrostrictive and the hydrostatic pressures. The dotted line indicates the distance, at which the pressure is greater than the critical, assumed, for example, ≈ cr P -30 MPa. (b) The dependence of the velocity of fluid from the electrode distance.
0
50
100
150
r-r
0
,
μ
m
Fig. 2. Same as in Fig. 1, at = t t 0 =100ns. There is no region of developed cavitation.
With the rapid rise of the applied voltage on the electrode (on the order of a few nanoseconds), the liquid (water) does not have enough time to come in motion and results in a change of the hydrostatic pressure, compensating the ponderomotive force (Fig. 1). Consequently, the ponderomotive forces cause a significant electrostrictive tensile stress in the dielectric liquid, which can lead to a disruption of the continuity of liquid (creating nanopores), similar to those observed in [21, 22, 23]. For the formation of discontinuities, it is necessary that the absolute value of the negative pressure in the fluid reach the values on the order of 10-30MPa [22]. In this case, the density fluctuations, which always exist due to thermal motion, lead to the formation of growing cavitation voids.
With a relatively slow increase in the voltage on the electrode, the fluid moves to the electrode, and the total pressure does not reach the critical value, at which the conditions required to initiate the fluid discontinuities (Fig. 2) are created. It should be noted that with the rapid and slow increase in voltage, the fluid velocity is less than the speed of sound by more than two orders of magnitude (Fig. 1 (b), 2 (b)) and less than the expansion velocity of nanopores by more than an order of magnitude (see Part III).
If the voltage on the electrodes grows relatively slowly (the voltage pulse front durations are hundreds nanoseconds or longer), the influence of ponderomotive forces on the fluid dynamics can be neglected. In this case, the conditions of works [10,11], in which the dynamics of a gas bubble in a strong electric field was studied, are valid.
## III. The probability of the critical bubble appearance
The probability of the appearance of the critical bubble in the volume V over the time t due to the development of thermal fluctuations, in accordance with the theory [24, 25], is
$$W _ { p o r e } = 1 - \exp \left ( - \iint _ { 0 \, V } \Gamma d t _ { 1 } d V \right ).$$
Here, Γ [m -3 s -1 ] characterizes the rate of the cavitation voids appearance in unit volume per second.
(b)
u, m/s
0
-2
-4
-6
-8
-10
-12
-14
-16
200
250
300
$$\Gamma = \frac { 3 k _ { b } T } { 1 6 \pi ( \sigma \cdot k _ { \sigma } ) ^ { 3 } } \frac { | P | ^ { 3 } } { 4 \pi \hbar } \exp \left ( - \frac { 1 6 \pi ( k _ { \sigma } \sigma ) ^ { 3 } } { 3 k _ { b } T \cdot P ^ { 2 } } \right )$$
Here, T is the temperature of fluid in Kelvin, B k is the Boltzmann constant, h -Plank constant. The parameter ( ) cr R k / 2 1 / 1 δ σ + = characterizes the dependence of the coefficient of surface tension on the critical radius of the nanopores cr R , which is determined by the balance of the tensile and the surface tension forces [9], δ is the so-called Tolman coefficient [26], which is determined from the experiment.
If the critical pressure, at which cavitation begins is 30 MPa [22], then, as follows from -≈ cr P [9], 26, . 0 ≈ σ k 24 nm, and the Tolman coefficient . 1 ≈ cr R 8 nm. That allows calculating . 1 = δ the probability of cavitation, the region where cavitation develops, and the concentration of the cavitation ruptures generated during the voltage pulse. Figure 3 shows the normalized dependencies of the rate of generation of cavitation voids Γ on the distance from the electrode at different time moments during the voltage pulse. The calculations were made at ns t 3 0 = , μ m , and 100 0 = r kV . For all generated cavitation nanopores, regardless of the local 54 0 = U negative pressure value cr P t r P > ) , ( , 26 was assumed. Note, that we do not take into . 0 = σ k account that the rate of cavitation voids generation near the electrode reaches saturation or dramatically decreases, during the voltage pulse, due to the breakdown development.
Fig. 3 . The computed dependencies of the rate of cavitation voids generation ) , normalized by , ( t r Γ ( ) 25 10 5.9 MPa 30 at ⋅ ≈ = Γ = Γ P norm m -3 s -1 . The curve 1 corresponds to t = 2 ns, 2 - 2.5ns, 3 - 3ns.
According to [1], the main reason of the breakdown developing in fluid in a nonuniform field at nanosecond times may be related to a rupture of the continuity of fluid induced by the electrostrictive forces in the inhomogeneous electric field. Thus, in the vicinity of a needle electrode, a region saturated by micropores is created. In the pores, the primary electrons are accelerated by the electric field to the energies exceeding the potential of ionization of a water molecule. The sources of primary electrons in the microcavities can include a background radiation, as well as the field emission from the surface of the cavities.
Figure 4 shows a schematic picture of the development of cavitation micropores arising in the vicinity of the electrode.
Fig. 4. region 1, where the electric field gradient is greatest, the occurring cavitation nanopores have enough time during the nanosecond voltage pulse to grow to a size at which an electron can gain enough energy for the excitation and the ionization of molecules of the liquid on the pore wall. In the region 2, the electrostrictive negative pressure reaches values at which the cavitation development becomes possible (which can be registered by the optical methods), but nanovoids appearing during the voltage pulse do not have enough time to grow to the size at which the potential difference across their borders becomes sufficient for the ionization or the excitation of water molecules. And, in the region 3 - the development of cavitation is impossible, electrostrictive negative pressure is relatively small and cannot compete with the forces of surface tension. Schematic picture of cavitation pore formation and growth in the vicinity of the electrode. In the since the spontaneously occurring nanovoids do not grow, because the value of the
## IV. Development of micropores in а dielectric liquid in inhomogeneous pulsed electric fields
Let us assume a spherical microcavity of the radius fluctuations. Considering the discontinuity of the dielectric constant at the liquid-vacuum boundary, the surface forces exerted by the electric field on the surface of the pore per unit area are [15]: R formed in water as a result of thermal
$$& \text{are} \lfloor 1 5 \rfloor \colon \\ & F _ { s _ { n } } = \frac { \varepsilon _ { 0 } } { 2 } \left ( \alpha \left ( \frac { E _ { p, n } ^ { 2 } } { \varepsilon } + \varepsilon E _ { p, t } ^ { 2 } \right ) - \left ( \varepsilon - 1 \right ) \left ( E _ { p, t } ^ { 2 } + \frac { E _ { p, n } ^ { 2 } } { \varepsilon } \right ) \right ) - \frac { 2 k _ { \sigma } \sigma } { R } - p \ [ N / m ^ { 2 } ].$$
Here, In (7) we took into account the surface tension of the liquid. Since the critical initial sizes of the cavitation pores are about a few nanometers, and the unperturbed electric field induced by the high voltage potential of the electrode is changing on a length scale of order of the electrode size (~ 10-100 microns, typical for experimental conditions in [3-8]). Therefore, the external unperturbed electric field good accuracy n p E , , t p E , are the normal and tangential components of the electric field inside the pore. 0 E in the vicinity of the pore can be considered as homogeneous with
>>
$$E _ { 0 } / ( d E _ { 0 } / d r ) > > R \,.$$
In this case, the electric field is uniform inside the ellipsoidal pore [13]. Wherein the expressions for the normal and tangential components of the electric field on the surface of the spherical pore are [13], [15]:
$$E _ { p, n } = \frac { 3 \varepsilon } { 1 + 2 \varepsilon } E _ { 0 } \cos ( \theta ), \quad E _ { p, t } = \frac { 3 \varepsilon } { 1 + 2 \varepsilon } E _ { 0 } \sin ( \theta ).$$
Here, θ is the azimuthal angle in a spherical coordinate system.
Since the objective of this work is to give the quantitative estimates and a qualitative picture of the development the cavitation micropores during the pulse strong nonuniform electric field, we will not consider changing the shape of the pores. Also, we assume that the force averaged over the surface acts on the surface of micropores. A form of the expanding micropore in a pulsed electric field will be discussed in Section V.
We obtain the mean pressure acting on the micropore surface, averaging (7) over the sphere surface, 4 2 R π :
$$P _ { a v } = & ( 3 / 4 ) \left ( \alpha - 1 \right ) \varepsilon _ { 0 } \varepsilon E ^ { 2 } - 2 k _ { \sigma } \sigma / R - p \,,$$
Wherein, if 0 , then the formed micropore collapses, but if < av P 0 , the micropore starts to > av P expand.
Let us consider the problem of expanding the spherical pores under the influence of the electrostrictive ponderomotive forces and the surface tension. As was shown above, during the relatively short voltage pulse ( 5 ns), the fluid does not have enough time to be set in 1 ~ 0 -t motion (see Fig.1 (b), 2 (b)), and, consequently, the hydrostatic pressure component in (10), p , is much smaller than the electrostrictive. Therefore, we can neglect with the hydrostatic pressure. Next, we follow the analogy with the problem considered in the classic work of Rayleigh [27]. However, in contrast to [27], where the bubble was kept from collapsing by the gas pressure inside, in our case, the bubble is held by the electrostriction forces induced in the liquid.
The fluid velocity at the distance r ′ from the center of an expanding nanopores follows from the continuity equation (1), at which ρ is assumed to be constant, is:
$$u = \frac { R ^ { 2 } } { r ^ { \prime 2 } } U \, \\ \quad \. \quad \ \mathfrak { n } \. \. \. \. \. \. \. \. \. \. \. \. \. \. \.$$
Here, R is the radius of nanopores, R r ≥ ′ ; U is the rate of expansion. In this case, the kinetic energy of the fluid is:
$$W _ { K } = & \, \frac { 1 } { 2 } \int _ { r = R } ^ { \infty } 4 \pi \rho u ^ { 2 } r ^ { \prime 2 } d r ^ { \prime } = 2 \pi \rho U ^ { 2 } R ^ { 4 } \int _ { r ^ { \prime } = R } ^ { \infty } \frac { 1 } { r ^ { \prime 2 } } d r ^ { \prime } = 2 \pi \rho U ^ { 2 } R ^ { 3 }$$
Since the rate of expansion is U = dR dt , the equation (12) can be rewritten as:
$$W _ { \kappa } = 2 \pi \rho R ^ { 2 } \left ( \frac { d R } { d t } \right ) ^ { 2 } \,.$$
The relevant work of the pressure forces (10) for the pores expansion is:
$$W _ { p } = 4 \pi \overset { R } { \underset { R _ { 0 } } { \int } } P _ { a v } r ^ { 2 } d r \\ \epsilon \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot.$$
Since the change in the kinetic energy of the fluid is equal to the work of the pressure forces, we obtain from (13) and (14):
$$\frac { d } { d t } \left ( R ^ { 3 } \left ( \frac { d R } { d t } \right ) ^ { 2 } \right ) = \frac { 4 } { \rho } R \frac { d R } { d t } \left ( \frac { 3 } { 8 } R ( \alpha - 1 ) \varepsilon _ { 0 } \varepsilon E ^ { 2 } - k _ { \sigma } \sigma \right )$$
In (15), we explicitly substituted av P with (10) and neglected the hydrostatic pressure component.
As in Part I, we assume that the electric field, from the spherical electrode to which the voltage is applied linearly, is increasing over time. Also, to simplify the formulas and the calculations presented below, we assume the parameter 1 const ) ( = = R k σ . This assumption is justified since we are interested in a significant expansion of the nanovoids compared to the size of the critical pores, i.e. δ ~ cr R R >> .
When substituting (4) into (15), we obtain:
$$& \text{When substituting (4) into (15), we obtain: } \\ & \frac { d } { d t } \left ( R ^ { 3 } \left ( \frac { d R } { d t } \right ) ^ { 2 } \right ) = \frac { 4 \sigma } { \rho } R \frac { d R } { d t } \left ( \frac { 3 \widetilde { P } _ { \max } R } { 4 \sigma } \left ( \frac { r _ { 0 } } { r } \right ) ^ { 4 } \left ( \frac { t } { t _ { 0 } } \right ) ^ { 2 } - 1 \right ) \\ & \widetilde { P } _ { \max } = 0. 5 \cdot ( \alpha - 1 ) \varepsilon \varepsilon _ { 0 } \left ( \frac { U _ { 0 } } { r _ { 0 } } \right ) ^ { 2 }$$
It is convenient to introduce a variable ( ) 2 / 5 0 / R R x = , where 0 R is the initial size of the pores and 0 / t t = τ . In this case, the equation (16) can be rewritten as
$$& \dots \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \, \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \dots \, \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cd. \, \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cd. \, \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cd� \, \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdt \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \ddots \, \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \dots \}$$
It follows from (17) that at the distance from the center of the electrode, r , the micropore expands at ( ) 5 / 2 4 0 0 max 0 4 ~ 3 x r r R P r ⎟ ⎠ ⎞ ⎜ ⎝ ⎛ = ≥τ σ τ . It is evident that since 1, the region in which the pores ≤ τ
$$\text{can expand to is limited to} \colon r < r _ { 0 } \left ( \frac { 3 \widetilde { P } _ { \max } R _ { 0 } } { 4 \sigma } \right ) ^ { 1 / 4 }. \text{Accordingly, at the time } \tau = \tau _ { 0 } \, \ x = 1 \text{ and}$$
$$\frac { d x } { d \tau } = 0 \,.$$
All calculations presented below were conducted with the same parameters as in Section II: = 0 r 100 µm, 81, = ε 5 , . 1 = α 54 keV, 0 = U 3ns ), 0 = t 072 N/m, . 0 = σ 2nm and 0 = R correspond to the time moment 3ns. 0 = = t t
Figure 5 shows the dependence of the maximum electrostrictive pressure ( ) 4 0 max / ~ ) , ( r r P r E F P E = = and the pore radius R on the distance from the electrode 0 r r -= η . It is seen that, under these conditions, an area, in which, for example, the nanopores of the size 2 nm may be created and expanded, is slightly greater than 1 μ m, whereas, the region, in which the tension generated by the ponderomotive force, is insufficient to cause cavitation at 14 0 > -= r r η μ m.
Figure 6 shows the dependence of the kinetic energy that an electron gets in the growing pore R Ein Δ 2 = φ depending on the distance from the pore to the electrode, where 0 0 5 . 1 2 1 3 E E Ein ≈ + = ε ε is the field inside the pore. It is seen that the electron kinetic energy can significantly exceed ~ 10 eV, i.e., the characteristic ionization potential of water molecules (I H2O ≈ 12.6 eV), in the region 43 . 0 0 < -= r r η μ m. Thus, the excitation (accompanied by UV and visible light) and the ionization starts in the vicinity of the electrode surface, in accordance with the observations of a nanosecond breakdown in the liquid (see, eg, [3-7]). It is apparent that the relative size of the area, in which φ Δ >10 eV, increases at a higher value U 0 or a smaller radius of the electrode 0 r .
Fig.5 . Dependencies of E P (a) and R (b) on 0 r r -= η at the time moment 3 ns. The area to 0 = = t t the left of the dotted vertical line shows where the pores of the initial size 2nm are expanding, and to the right, it shows where they are collapsing. The area where the electrostriction tension (negative pressure) exceeds the cavitation threshold, 30MPa [32] is to the left of the solid vertical line. > E P
Figure 7 shows the dependence of the rate of micropores growth on time. It is seen that the rate of expansion is much less than the velocity of sound and much more than the characteristic velocity of the fluid under the action of ponderomotive forces. Therefore, the assumptions that during the voltage pulse a micropore center is not shifted and the discussed above modification of the Rayleigh model are applicable.
Fig.6. Parameter R Ein 2 = Δφ dependence on 0 r r -= η . The solid vertical line corresponds to = Δφ 10 eV.
Fig.7. Dependencies of the rates of expansion of the pores on time at different 0 r r -= η . Curve 1 corresponds to 0 ; 2 -= η 25; 3 -. 0 = η 5; 4 -. 0 = η 75 and 5 -. 0 = η μ m. 1 = η
## V. On the shape of the expanding micropores
We did not consider the fluid motion caused by the ponderomotive forces in the model of the nanopores expansion discussed above, since the rate of expansion (Fig.7) exceeds essentially the fluid velocity near the electrode (Fig. 1b, 2b). This statement is valid if the distortion of the electric field caused by the presence of the micropore is not taken into account. When considering the potential in the vicinity of the pores determined by an external potential and the dipole potential created by the polarization charges at the boundaries of the pores [13, 15], the expression for the square of the electric field near the spherical pore ) in the assumption , ( θ r Eout ′ of a constant external electric field 0 E (induced by the applied voltage) is:
$$E _ { o u t } ^ { 2 } ( r ^ { \prime }, \theta ) & = E _ { 0 } ^ { 2 } \left ( 1 + \left ( 3 \cos ^ { 2 } \left ( \theta \right ) + 1 \left ( \frac { \left ( \varepsilon - 1 \right ) } { 2 \varepsilon + 1 } \right ) ^ { 2 } \frac { R ^ { 6 } } { r ^ { \prime 6 } } - \left ( 5 \cos ^ { 2 } \left ( \theta \right ) - 1 \right ) \frac { R ^ { 3 } } { r ^ { \prime 3 } } \frac { \left ( \varepsilon - 1 \right ) } { 2 \varepsilon + 1 } \right ) \\ \tau \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots.$$
Here, the angle 0corresponds to the north pole of the pores (to the direction of the external = θ unperturbed electric field), and 2 relates to the equator (normal to the direction of the / π θ = external unperturbed electric field). The dependencies 2 0 2 / E E on R r / ′ at different angles θ are shown in Fig. 8:
Fig. 8 . Dependencies 2 0 2 / E E on R r / ′ at different angles θ . Line 1 coresponds to θ =0, 2 θ = π/8, 3 θ = π/4, 4 θ = 3π/8, 5 θ = π/2.
Taking into account (18) and (3) and neglecting the hydrostatic pressure and viscous friction, the volumetric electrostrictive force near the pore in the equation of fluid motion (1) has the form:
The dependencies on the volumetric force ) , referred to , ( θ r F ′
$$& \text{volumetric electrostructive force near the pore in the equation of fluid motion (1) has the form:} \\ & F ( r ^ { \prime }, \theta ) \approx \frac { 3 } { 2 } \alpha \varepsilon \varepsilon _ { 0 } E _ { 0 } ^ { 2 } \frac { ( \varepsilon - 1 ) } { 2 \varepsilon + 1 } \frac { R ^ { 3 } } { r ^ { \prime 4 } } \left ( \left ( 5 \cos ^ { 2 } ( \theta ) - 1 \right ) - 2 \left ( 3 \cos ^ { 2 } ( \theta ) + 1 \right ) \frac { R ^ { 3 } } { r ^ { \prime 3 } } \frac { ( \varepsilon - 1 ) } { 2 \varepsilon + 1 } \right ). \\ & \text{v.. } \frac { 1 } { 2 } \dots \frac { 1 } { 2 } \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots.}$$
$$& \text{me experience on use yourtime force } r ( r, \sigma ) \,, \text{reference to} \\ & F _ { 0 } = \frac { 3 } { 2 } \alpha \varepsilon \varepsilon _ { 0 } \frac { \left ( \varepsilon - 1 \right ) } { 2 \varepsilon + 1 } \frac { E _ { 0 } ^ { 2 } } { R } \left ( 1 + 2 \frac { \left ( \varepsilon - 1 \right ) } { 2 \varepsilon + 1 } \right ), \, \text{are shown in Fig.9}.$$
FIFO
Fig. 9. dependencies. Line 1 coresponds to dimensional contour plot Normalized volumetric force ) in the vicinity of the pores at different angles , ( θ r F ′ θ =0, 2 θ = π/8, 3 θ = π/4, 4 θ = 3π/8, 5 θ = π/2; b) Twoθ . a) - radial
The fluid moves toward the pore in the vicinity of the equator, due to the volumetric electrostrictive force directed toward the center of the pore (negative direction). However, at the pole, it moves away from the pore, because the electrostrictive force is positive. Therefore, the micropore will be stretched along the electric field, since its expansion in the equatorial plane will be inhibited by the fluid counterflow at the equator (Fig. 10).
Fig. 10. produced by the voltage applied to the electrode. The red arrows indicate the 'electrostatic' pressure inside the pores, the blue arrows demonstrate the directions of the fluid motion in the outer region of the pore induced by the electrostrictive volumetric force. Qualitative picture of the micropores form formation. 0 E r is the unperturbed external field
It should be noted that the above considerations are mostly qualitative and valid only for the fluid outside the pore assumed as an elastic medium, i.e., for a very small deformation of the micropore and fluid in its vicinity. In order to determine the exact shape of the expanding pore we need to consider in the vicinity of the pore following factors, such as: boundary conditions for the ponderomotive forces, tension forces, the hydrostatic pressure, and, finally, the appearance of new microscopic discontinuities (nanopores) in the fluid in the region of large tensile stresses.
## Conclusions
- 1. The equations describing the nanopore expansion in the presence of a strong inhomogeneous pulsed electric field are obtained, and the corresponding approaches are formulated.
- 2. It is shown that the three characteristic regions appear in a liquid dielectric in the vicinity of
the electrode. In the first region, where the electric field gradient is greatest, the occurring cavitation nanopores have enough time during the nanosecond voltage pulse to grow to a size at which an electron can gain enough energy for the excitation and ionization of the liquid molecules on the pore wall. In the second region, the electrostrictive negative pressure reaches values at which cavitation development becomes possible (and can be recorded by the optical methods). However, the nanovoids appearing during the voltage pulse do not have enough time to grow to the size at which the potential difference across their borders becomes sufficient for the ionization or excitation of water molecules. And, in the third region, the development of cavitation is impossible, since the spontaneously occurring nanovoids do not grow, because the value of the electrostrictive negative pressure is relatively small and cannot compete with the forces of surface tension.
- 3. The physical processes affecting the shape of the expanding nanopores are considered. It is shown that a perturbation of the electric field caused by а micropore near its equator (in the plane perpendicular to the direction of the unperturbed external electric field) produces the fluid flow directed towards the pore, and in the vicinity of poles, away from the pore. As a consequence, a micropore will extend along the external electric field.
## References
- [1] M.N. Shneider, M. Pekker, A. Fridman, 'On the initial stage of the sub-nanosecond pulsed breakdown in liquid dielectrics ' IEEE Transactions on Dielectrics and Electrical Insulation 19 , 1579 (2012).
- [2] M. N. Shneider and M. Pekker, 'Dielectric fluid in inhomogeneous pulsed electric field,' Phys. Rev. E 87 , 043004 (2013).
- [3] D. Dobrynin, Y. Seepersad, M. Pekker, M. N. Shneider, G. Friedman, and A. Fridman, 'Non equilibrium nanosecond-pulsed plasma generation in the liquid phase (water, PDMS) without bubbles: Fast imaging, spectroscopy and leader-type model,' J. Phys. D: Appl. Phys. 46 (10), 105201 (2013).
- [4] A. Starikovskiy, 'Pulsed nanosecond discharge development in liquids with various dielectric permittivity constants,' Plasma Sources Sci. Technol. 22 (1), 012001 (2013).
- [5] Y. Seepersad, M. Pekker, M. N. Shneider, D. Dobrynin, and Al. Fridman, 'On the electrostrictive mechanism of nanosecond-pulsed breakdown in liquid phase,' J. Phys. D: Appl. Phys. 46 (16), 162001 (2013).
- [6] I. L. Marinov, O. Guaitella, A. Rousseau, and S. M. Starikovskaia, 'Successive nanosecond discharges in water,' IEEE Trans. Plasma Sci. 39 , 2672 (2011).
- [7] I. L. Marinov, O. Guaitella, A. Rousseau, and S. M. Starikovskaia, 'Cavitation in the vicinity of the high-voltage electrode as a key step of nanosecond breakdown in liquids,' Plasma Sources Sci. Technol. 22 , 042001 (2013).
- [8] M. Pekker, Y. Seepersad, M.N Shneider, A. Fridman, D. Dobrynin , 'Initiation stage of nanosecond breakdown in liquid', J. Phys. D: Appl. Phys. 47 , 025502, (2014).
- [9] M. N. Shneider and M. Pekker, 'Cavitation in dielectric fluid in inhomogeneous pulsed electric field' J. Appl. Phys., 114, 214906 (2013).
- [10] S. M. Korobeinikov, 'Bubble deformation in an electric field,' J. Eng. Phys. Thermophys. 36 (5), 882 (1979).
- [11] S. M. Korobeinikov, A. V. Melekhov, Yu. N. Sinikh, and Yu. G. Soloveichik, 'Effect of strong electric field on the behaviour of bubbles in water,' High Temp. 39 , 368 (2001).
- [12] L. D. Landau and E. M. Lifshitz, Fluid Mechanics, 2nd ed. (Oxford, Pergamon Press, 1987).
- [13] L.D. Landau, E.M. Lifshitz, Electrodynamics of Continuous Media, 2nd ed. (Oxford, Pergamon Press, 1984)
- [14] J. D. Jackson, Classical Electrodynamics, 3rd ed. (Wiley, New York, 1999)
- [15] I. E . Tamm Fundamentals of the Theory of Electricity (Moscow: Mir Publ., 1979) ,
- [16] Yuan-Hui Li, 'Equation of State of Water and Sea Water', J. Geophys. Res., 72 , 2665 (1967)
- [17] R.I. Nigmatulin, R.Bolotnova, ' The equation of state of liquid water under static and shock compression', Proceedings VI Zababakhin Scientific Readings, Snezhinsk, Russian Federal Nuclear Center VNIITF, 2003, www.vniitf.ru/rig/konfer/6zst/6zst.htm.
- [18] J.S. Jakobs, A.W. Lawson, 'An analysis of the pressure dependence of the dielectric constant of polar liquids', J. Chem. Phys. 20 , 1161 (1952)
- [19] V.Y. Ushakov, V.F. Klimkin and S.M. Korobeynikov, Breakdown in liquids at impulse voltage (NTL, Russia, Tomsk, 2005).
- [20] W.K.H. Panofsky and M. Phillips, Classical Electricity and Magnetism (Addison-Wesley Pub. Co.,1962).
[21] V.P. Skripov, Metastable Liquid (Moscow: Nauka, 1972)
[22] E. Herbert, S. Balibar, F. Caupin 'Cavitation pressure in water' Phys. Rev. E 74, 041603 (2006)
[23] V.E. Vinogradov, 'Depression of the cavitation centers in water under pulsed tension conditions', Tech. Phys. Lett. 35 , 54-56 (2009)
[24] Y.B. Zel'dovich, 'On the theory of new phase formation. Cavitation.' Zh. Eksp. Teor. Fiz. 12, 525, 1942.
[25] J.C. Fisher, 'The Fracture of Liquids', J. Appl. Phys. 19 , 1062 (1948)
[26] R.S Tolman 'The superficial density of matter at liquid-vapor boundary', J Chem. Phys. 1949, V17, N2, p. 333
- [27] O. M. F. R. S. Lord Rayleigh, 'On the pressure developed in a liquid during the collapse of a spherical cavity,' Philos. Mag. 34, 94 (1917).
|
10.1063/1.4922244
|
[
"Mikhail N. Shneider",
"Mikhail Pekker"
] |
2014-08-04T19:12:11+00:00
|
2014-12-01T19:19:28+00:00
|
[
"physics.flu-dyn",
"cond-mat.soft"
] |
Pre-breakdown cavitation development in the dielectric fluid in the inhomogeneous, pulsed electric fields
|
We consider the development of pre-breakdown cavitation nanopores appearing
in the dielectric fluid under the influence of the electrostrictive stresses in
the inhomogeneous pulsed electric field. It is shown that three characteristic
regions can be distinguished near the needle electrode. In the first region,
where the electric field gradient is greatest, the cavitation nanopores,
occurring during the voltage nanosecond pulse, may grow to the size at which an
electron accelerated by the field inside the pores can acquire enough energy
for excitation and ionization of the liquid on the opposite pore wall, i.e.,
the breakdown conditions are satisfied. In the second region, the negative
pressure caused by the electrostriction is large enough for the cavitation
initiation (which can be registered by optical methods), but, during the
voltage pulse, the pores do not reach the size at which the potential
difference across their borders becomes sufficient for ionization or excitation
of water molecules. And, in the third, the development of cavitation is
impossible, due to an insufficient level of the negative pressure: in this
area, the spontaneously occurring micropores do not grow and collapse under the
influence of surface tension forces. This paper discusses the expansion
dynamics of the cavitation pores and their most probable shape.
|
1408.0774v3
|
## The lowest limits on the doubly charged Higgs boson masses in the minimal left-right symmetric model
G. Bambhaniya, 1 J. Chakrabortty, 2 J. Gluza, 3 T. Jeli´ski, n 3 and M. Kordiaczy´ska n 3
1 Theoretical Physics Division, Physical Research Laboratory, Navarangpura, Ahmedabad - 380009, India 2 Department of Physics, Indian Institute of Technology, Kanpur-208016, India
3 Institute of Physics, University of Silesia, Uniwersytecka 4, PL-40-007 Katowice, Poland
The doubly charged Higgs bosons H ±± 1 2 , would undoubtedly be clear messengers of the new physics. Wediscuss their mass spectrum and show how experimental data and relations between scalar masses put limits on it. In particular, both the masses of the particles H 0 1 , A 0 1 that play a crucial role in FCNC effects and the masses of the additional gauge bosons W 2 , Z 2 are notably important. For instance, if M H ,A 0 1 0 1 glyph[similarequal] 15 TeV and M W 2 glyph[similarequal] 3 76 TeV then the lowest mass of . H ±± 1 is 465 GeV. In contrast, due to the freedom in the parameter space of the full scalar potential, there is no lowest limit on the mass of H ±± 2 . It is shown to which signals at hadron colliders such relatively light doubly charged scalars might correspond. LHC working at √ s = 14 TeV will enter into the region where existence of such particles with minimal masses can be thoroughly explored for a much wider parameter space of the minimal and manifest version of the left-right symmetric model (MLRSM). Taking into account our considerations and present ATLAS and CMS exclusion limits on M H ±± , there exist already first partial bounds on some of the MLRSM scalar potential parameters.
PACS numbers: 12.60.-i, 12.60.Fr, 14.80.Fd, 14.80.Ec
As argued among others in [1], the Higgs boson discovery at the LHC [2, 3] was a 'Big Deal'. It established the Standard Model (SM) of gauge interactions and proved that the intuition was correct, as far as the mechanism of mass generation is concerned.
Now further investigations are needed as we would like to know the details of the scalar sector. The underlying gauge theory need not to be necessarily the SM which is based on the SU (2) × U (1) symmetry. In fact, there are many arguments that the SM is an effective theory working at relatively low energies [4]. Then, going beyond the SM, one or two open and crucial questions are: what is the representation for the scalar multiplets and what is the shape of the scalar potential of the fundamental theory in particle physics? Are the fundamental forces unified, and if yes, then in which way? So far we do not have any particular answer to these queries. For instance, there are suggestions that the measured values of m H and m t make the electroweak vacuum metastable, and a new physics should be below ∼ 10 10 -10 12 GeV [5-8]. However, there are also arguments for the Standard Model being a valid theory up to the Planck scale [9].
SM scalar sector consists of a doublet of complex scalar field. There are four real scalar fields, three of them are responsible for giving masses to the three gauge bosons W ,Z ± 0 through the Higgs mechanism. The remaining scalar is associated with the neutral Higgs boson which has been discovered at the LHC. However, if we go beyond the SM, we may need to deal with more complex scalar systems, for instance SU (2) triplet which contains charged (singly and(or) doubly) scalar particles. These particles are naturally embedded in the left-right sym- metric models where a new characteristic energy scale exists and the symmetry between left and right SU (2) gauge sectors is broken spontaneously [10, 11]. Here we focus on the so-called minimal and manifest version of the model (MLRSM), see e.g. [10-13].
There is presently large activity in searching for the scalar particles of any sort (neutral, charged) at the LHC. In particular, knowledge of the existence of doubly charged scalar bosons would be crucial for further directions in exploration of particle physics phenomena, for instance, their presence would strongly disfavour the minimal version of the supersymmetric Standard Model. There are two basic ways in which doubly charged particles can be searched for. The first is indirect: looking for rare lepton flavour and number violating processes and precision measurements (deviations from the SM expectations). The second opportunity is provided by the accelerators where new particles can be produced directly at high energies, as in the LHC. There are already many analysis undertaken by the CMS and ATLAS collaborations regarding these kind of searches, and the present limits for doubly charged scalars are:
$$\cdot \quad M _ { H ^ { \pm \pm } } \geq 4 4 5 \, \text{GeV} \, ( 4 0 9 \, \text{GeV} ) \text{ for } \text{ CMS (ATLAS), } ( 1 )$$
in the 100% leptonic branching fraction scenarios [14, 15]. We assume the same framework here. Other scenarios are also possible within MLRSM [16] or in another models, e.g. Higgs Triplet Model [17, 18].
In this paper, we show that theoretical considerations are also important, and we find that in the MLRSM some of the scalar masses are constrained from below. In particular, the lowest limit for the doubly charged scalar H ±± 1 mass is not much beyond the present limit given in
Eq. 1 for many different model parameters.
Recently, it has been shown in [16] that charged scalars with masses at the level of a few hundred GeV can be realized in the MLRSM by allowing a percent-level tuning of the scalar potential parameters, compatible with both the large parity breaking scale v R and severe bounds on neutral scalar masses ( M H ,A 0 1 0 1 ) derived from flavour changing neutral currents (FCNC). This is non-trivial as both v R and M H ,A 0 1 0 1 are at the level of a few TeV and all scalars apart form H 0 0 are naturally heavy, their leading mass terms are proportional to v R . In this paper, we look into further details and show that in fact lowest bounds can be obtained for some of the scalar masses. The MLRSM scalar potential and its minimization followed by the diagonalization have been investigated in [19] and explicit relations between physical and unphysical scalar fields are given in [20]. For our purposes, we repeat here only a subset of formulas which we need for further discussion, they are valid as long as κ 1 glyph[lessmuch] v R , which is true as κ 1 and v R are connected directly with masses of light and heavy charged gauge bosons, and M W 1 glyph[lessmuch] M W 2 [16, 21]
$$M _ { H _ { 0 } ^ { 0 } } ^ { 2 } \, \simeq \, 2 \kappa _ { 1 } ^ { 2 } \lambda _ { 1 }, \, & & ( 2 ) \, \quad \, \text{$\tauea$} \,.$$
$$M _ { H _ { 1 } ^ { 0 } } ^ { 2 } \, \simeq \, \frac { 1 } { 2 } \alpha _ { 3 } v _ { R } ^ { 2 }, \, & \quad \ ( 3 ) \quad \, ^ { ( \text{see} } _ { M _ { H _ { 2 } ^ { 0 } } ^ { \prime } }$$
$$M _ { A _ { 1 } ^ { 0 } } ^ { 2 } \, \simeq \, \frac { \bar { 1 } } { 2 } \alpha _ { 3 } v _ { R } ^ { 2 } - 2 \kappa _ { 1 } ^ { 2 } \, ( 2 \lambda _ { 2 } - \lambda _ { 3 } ) \,, \quad \, ( 4 ) \, \begin{array} { c } \overset { \cdots } { \text{izon} } \\ \text{CM} \end{array} \,.$$
$$M _ { H _ { 3 } ^ { 0 } } ^ { 2 } \, \simeq \, \frac { \bar { 1 } } { 2 } v _ { R } ^ { 2 } \left ( \rho _ { 3 } - 2 \rho _ { 1 } \right ), \, \quad \, \quad \, \quad \, \left ( 5 \right ) \, \quad \, \text{vio} \\ M _ { H ^ { c } } ^ { c }$$
$$M _ { H _ { 1 } ^ { \pm \pm } } ^ { 2 } \, \simeq \, \frac { \bar { 1 } } { 2 } \left [ v _ { R } ^ { 2 } \left ( \rho _ { 3 } - 2 \rho _ { 1 } \right ) + \alpha _ { 3 } \kappa _ { 1 } ^ { 2 } \right ], \quad \ \left ( 6 \right ) \, \substack { M _ { H _ { 1 } ^ { 0 } } \\ \tilde { \delta \rho } \gtrsim }$$
$$M _ { H _ { 2 } ^ { \pm \pm } } ^ { 2 } \, \simeq \, \stackrel { \mu } { 2 } \rho _ { 2 } v _ { R } ^ { 2 } + \frac { 1 } { 2 } \alpha _ { 3 } \kappa _ { 1 } ^ { 2 }. \, \quad \, \quad ( 7 ) \, \stackrel { o \rho } { \sim } \, \\ \ln$$
As we can see, SM-like Higgs boson H 0 0 has a mass proportional to the vacuum expectation value κ 1 ( ∼ electroweak breaking scale, M W 1 ∼ κ 1 [12]). Here, among the neutral scalars only A 0 1 and H 0 1 contribute to the FCNC interactions. To our knowledge, their effects have been discussed for the first time in the context of leftright models in [22], see also [23-28] and the Appendix in [16]. In general their masses need to be at least of the order of 10 TeV, though some alternatives have been also considered in [29]. Also, the parity breaking scale v R is already strongly constrained by a 'golden' decay chain process W R → l 1 N l → l 1 2 l jj [30, 31], M W 2 ≥ 2 8 TeV. . This limit (at 95 % C.L.) is for a genuine left-right symmetric model which we consider here (MLRSM). There are also other model independent limits on the masses of gauge bosons ∼ 4 TeV and we stay conservative while choosing this value that implies v R is already bigger than 8 TeV. For more details and other possible scenarios see [16] and recent [32, 33].
Now, taking into account the above facts, for example with M H ,A 0 1 0 1 ≥ 10 TeV and v R ≥ 8 TeV, the scalar po- tential parameter α 3 can be determined, see Eq. 3. This parameter also enters into the relations of the masses of doubly charged scalars, Eqs. 6 and 7. Thus the only free parameter that constrains M H ±± 1 is δρ ≡ ρ 3 -2 ρ 1 .
FIG. 1: Dependence of the H ±± 1 mass on δρ ≡ ρ 3 -2 ρ 1 for v R = 8TeV and various masses of flavour changing neutral Higgs scalars H 0 1 and A 0 1 . Points marked on the vertical dotted line by square, circle and triangle show minimal M H ±± 1 corresponding to M H ,A 0 1 0 1 = 10, 15 and 20 TeV, respectively (see Tab. I). In the region left to the vertical dotted line, M H 0 3 (dashed line) is smaller than 55 4 GeV (dash-dotted hor-. izontal line). The horizontal grey strip represents the latest CMS exclusion limit on doubly charged Higgs boson mass. The mass spectrum represented by the filled square is obviously ruled out by the LHC. Hence, for v R = 8TeV and M H ,A 0 1 0 1 = 10TeV the minimal allowed value of M H ±± 1 is 445 GeV (empty square) and one gets the following lower bound δρ glyph[greaterorsimilar] 3 · 10 -3 .
In Fig. 1 mass of H ±± 1 is given as a function of δρ for v R = 8TeV and different choices of M H ,A 0 1 0 1 . Without the loss of generality it has been assumed that M H 0 1 = M A 0 1 , which implies that 2 λ 2 = λ 3 , Eq. 4. Here, δρ starts from zero, as we aim at the minimal values of M H ±± 1 . For δρ = 0, the only contribution to the H ±± 1 mass is connected with a second term in Eq. 6, however, also mass of H 0 3 depends directly on δρ , see Eq. 5. Thus it cannot be negative. Moreover, interpreting LEP II data related to e + e -→ γ + E / T , it is possible to find lower bound on M H 0 3 , which is about 55 4 GeV [34]. . This leads to the minimal value for the mass of doubly charged scalar H ±± 1 . It depends both on the minimal allowed values of M H 0 3 and masses of the scalars H ,A 0 1 0 1 that control FCNC. Let us note that for v R ∼ 10 TeV and small values of δρ (e.g. δρ glyph[lessorsimilar] 5 · 10 -4 ), the dependence on δρ is rather weak and the mass of H ±± 1 is dominated by the term √ α κ / 3 1 √ 2 ≈ 171 √ α 3 GeV.
As one can see from Fig. 1, for M H ,A 0 1 0 1 = 10TeV bounds from the LHC yield the following limit: δρ glyph[greaterorsimilar] 3 · 10 -3 . On the other hand, for bigger M H ,A 0 1 0 1 there are no restrictions on δρ from the LHC yet but the model it-
self predicts the minimal possible value of M H ±± masses.
1 It is worthwhile to note that taking into account experimental limits one can obtain the following bound on M H ±± 1 :
$$\sqrt { \min ( M _ { H _ { 3 } ^ { 0 } } ^ { 2 } ) + M _ { H _ { 1 } ^ { 0 } } ^ { 2 } \kappa _ { 1 } ^ { 2 } / v _ { R } ^ { 2 } } \approx \frac { 2. 4 1 \, \text{TeV} ^ { 2 } } { v _ { R } }. \quad ( 8 ) \quad \max _ { \text{thos} } ^ { \infty }$$
The dashed line in Fig. 1 is for M H 0 3 . At δρ ≈ 10 -4 , it crosses the horizontal line which corresponds to the lowest bound on mass of H 0 3 . Hence, the region left to the vertical, dotted line is excluded due to the above mentioned LEP constraints and minimal mass of H ±± 1 can be determined from the points at which solid lines intersect vertical, dotted line. Some precise values of H ±± 1 masses corresponding to such points are presented in the Tab. I.
TABLE I: Minimal masses of a doubly charged Higgs boson H ±± 1 as a function of the parity breaking scale v R of the right sector of the model and the mass of neutral Higgs bosons { A , H 0 1 0 1 } which contribute to the FCNC effects. Corresponding masses of H ±± 2 are fixed by taking in addition ρ 2 = δρ/ 4, what yields M H ±± 2 = M H ±± 1 . No LHC direct limits from Eq. 1 applied. Square, circle and triangle correspond to points marked on Fig. 1.
| v R = 8 TeV ( M W 2 = 3 . 76 TeV) | v R = 8 TeV ( M W 2 = 3 . 76 TeV) | v R = 8 TeV ( M W 2 = 3 . 76 TeV) |
|-------------------------------------|-------------------------------------|-------------------------------------|
| M A 0 1 ,H 0 1 [TeV] | α 3 | min( M ±± H 1 ) [GeV] |
| 10 | 3.13 | 312 ( glyph[squaresolid] ) |
| 15 | 7.03 | 465 ( ) |
| 20 | 12.5 | 617 ( glyph[trianglesolid] ) |
| v R = 10 TeV ( M W 2 = 4 . 7 TeV) | v R = 10 TeV ( M W 2 = 4 . 7 TeV) | v R = 10 TeV ( M W 2 = 4 . 7 TeV) |
| M A 0 1 ,H 0 1 [TeV] | α 3 | min( M ±± H 1 ) [GeV] |
| 10 | 2.0 | 252 |
| 15 | 4.5 | 373 |
| 20 | 8.0 | 495 |
| v R = 12 TeV ( M W 2 = 5 . 64 TeV) | v R = 12 TeV ( M W 2 = 5 . 64 TeV) | v R = 12 TeV ( M W 2 = 5 . 64 TeV) |
| M A 0 1 ,H 0 1 [TeV] | α 3 | min( M ±± H 1 ) [GeV] |
| 10 | 1.39 | 212 |
| 15 | 3.13 | 312 |
| 20 | 5.56 | 413 |
Let us remark that there is, in principle, no restriction from the scalar potential parameters on M H ±± 2 , as the parameter ρ 2 does not play any role to determine the masses of other remaining scalars, H ,A ,H 0 2 0 2 ± 1 , H ± 2 . Thus this parameter is not constrained within perturbative limit. Here, for example we have set ρ 2 = δρ/ 4 in Tab. I. Moreover, note that min( M H ±± 1 ) decreases with v R , see Eq. 8.
It is clear that some of M H ±± 1 in Tab. I are already excluded by the LHC limits, Eq. 1. It means that the first constraints on scalar potential parameters can be derived from those experimental bounds, as already seen in Fig. 1 (point marked by empty square). To this end, it is worthwhile to examine the influence of α 3 on masses of H ±± 1 . Let us focus on v R = 8TeV case. When δρ is smaller than about 10 -4 then mass of doubly charged scalar is dominated by the term √ α κ / 3 1 √ 2. On the other hand, if δρ is bigger than 10 -2 then its mass is mainly driven by v R √ δρ/ 2. On Fig. 2, it is shown how those masses depend on δρ and α 3 in the region of the parameter space where both contributions to masses are important. On that plot, we have shaded the regions of the parameter space which are excluded either by FCNC (i.e. M H ,A 0 1 0 1 glyph[greaterorsimilar] 10 TeV) or LHC. Note that there is also excluded region related to the lower bound on H 0 3 mass (left of the vertical dashed line). It is straightforward to do similar analysis for dependence of H ±± 2 mass on α 3 and ρ 2 .
FIG. 2: Dependence of the H ±± 1 mass (in TeV) on δρ and α 3 for v R = 8TeV. Solid lines divide the parameter space into colored regions in which mass of H ±± 1 is characterized according to the legend. Point marked by a glyph[star] corresponds to benchmark set of parameters (14)-(16). Shaded regions are excluded due to FCNC (under horizontal, dashed line), LEP (left to vertical dashed line) and CMS constraints (under curved dotted line). Squares, circle and traingle correspond to points marked in Fig. 1.
As an example, we present benchmark set of scalar masses which satisfy both FCNC and LHC constraints with light and degenerate masses of doubly charged scalars, assuming v R = 8 TeV (all masses are given in GeV):
$$M _ { H ^ { 0 } _ { 0 } } \ = \ 1 2 5,$$
$$M _ { H _ { 1 } ^ { 0 } } \ = \ 1 6 4 9 2, \ \ M _ { H _ { 2 } ^ { 0 } } = 1 1 3 1 4, \ \ M _ { H _ { 3 } ^ { 0 } } = 2 5 3, ( 1 0 )$$
$$M _ { A _ { 1 } ^ { 0 } } \ = \ 1 6 4 9 6, \ \ M _ { A _ { 2 } ^ { 0 } } = 2 5 3,$$
$$M _ { H _ { 1 } ^ { \pm } } \ = \ 4 3 9, \ \ M _ { H _ { 2 } ^ { \pm } } = 1 6 4 9 6,$$
$$\underset { \infty } { \mathfrak { e } } \quad M _ { H _ { 1 } ^ { \pm \pm } } \ = \ 5 6 7, \ \ M _ { H _ { 2 } ^ { \pm \pm } } = 5 6 7.$$
These masses are outcome of the following parameters of
the scalar potential:
$$\rho _ { 1 } \ = \ 1. 0, \ \rho _ { 2 } = 5 \cdot 1 0 ^ { - 4 }, \ \rho _ { 3 } - 2 \rho _ { 1 } = 2 \cdot 1 0 ^ { - 3 }, ( 1 4 ) \quad \text{charg}$$
$$\lambda _ { 1 } \ = \ 0. 1 3, \ \lambda _ { 2 } = 0, \ \lambda _ { 3 } = 1,$$
$$\alpha _ { 3 } \, = \, 8. 5.$$
This scenario realizes mass spectrum with maximal number (three) of light charged scalars [16]. Of course, yet another setups are also possible. For example, increasing parity breaking scale to 20 TeV, setting α 3 ∼ 0 5, . ρ 1 ∼ ρ 2 glyph[greaterorsimilar] α / 3 4 and tuning δρ ∼ 10 -3 give H 0 3 , A 0 2 , H ± 1 and H ±± 1 masses ∼ v R √ δρ/ 2 ∼ 450 GeV, while the remaining scalars (beside H 0 0 ) have masses larger than v R √ α / 3 2 ∼ 10 TeV. The other option is to keep v R ∼ 10 TeV but also choose ρ 2 ∼ 10 -3 and set δρ ∼ α 3 ∼ 2. As a result, only H ±± 2 is light, with mass ∼ 500 GeV, while other scalars are heavier than 10 TeV.
Let us note that all these spectra are also in agreement with low-energy constraints, here muon ∆ r parameter plays a crucial role [27, 35, 36]. It is interesting to note that scalar and fermionic loop contributions come up with opposite signs to ∆ r and heavy neutrinos can compensate possible large scalar effects [27, 35, 36].
We have focused here on doubly charged Higgs bosons, aiming at their small masses, but we can see that they are entangled by the scalar potential parameters, and in this way also some singly charged and neutral scalars are relatively light. Their effects at hadron colliders will be studied in details elsewhere.
Fig. 3 gives estimation of cross sections for the pair production of doubly charged Higgs bosons at LHC for the center of mass energies at present (8 TeV), forthcoming run (13 -14 TeV) and in the further perspective (100 TeV). For the simplicity, we assume that masses of H ±± 1 and H ±± 2 are degenerate i.e. δρ = 4 ρ 2 . Limits on the doubly charged Higgs boson H ±± 1 masses from recent LHC data have been taken into account, and exclusion limits are explicitly given in Fig. 3.
Here, the cross section for the process pp → ( H ±± 1 H ∓∓ 1 ⊕ H ±± 2 H ∓∓ 2 ) → 4 glyph[lscript] for different centre of mass energies √ s = 8, 14 and 100 TeV at the LHC with only basic cut (lepton p T > 10 GeV) is shown in Fig. 3 for different set of doubly charged scalar masses. We have also implemented the detailed event selection criteria [38], as given in [16, 37] to compute the SM background for this final state coming from tt ¯ ( Z/γ ∗ ) , ( Z/γ ∗ )( Z/γ ∗ ). Using the same set of cuts the signal event cross section is estimated for doubly charged scalar masses [450-900] GeV. We have pointed out four benchmark points for the following masses of doubly charged scalars: 450, 600, 700, and 900 GeV. The corresponding significances [39] of the signal for integrated luminosity 300 fb -1 are 13 8, . 5 7, . 3 3, 1 0 respectively. . . Thus at the LHC with √ s = 14TeV centre of mass energy and 300 fb -1 luminosity up to 600 GeV mass of the doubly charged scalars can be probed with significance ≥ 5. Finally, it is worthwhile to note that due to kinematics for √ s = 100TeV discussed crosssection does not change significantly when mass of doubly charged scalars increases.
FIG. 3: Cross-section σ for the pair production of doubly charged scalars H ±± i decaying to four leptons for √ s = 8, 14 and 100 TeV at the LHC (solid lines) without the detailed selection cuts (see the text for details). The dotted-blue horizontal line is the estimation for the SM background for 4 glyph[lscript] final state with √ s = 14 TeV incorporating the detailed event selection criteria. Also the cross section for the process pp → H ++ 1 H --1 /H ++ 2 H --2 → 4 glyph[lscript] with √ s = 14 TeV after implementing the same full selection cuts is depicted by the dashed-blue line for doubly charged scalar mass range [450900] GeV. The four patches on the 'Signal' line denote the significance of the signal. It is assumed that δρ = 4 ρ 2 , what results in M H ±± 1 = M H ±± 2 , see Eqs. (9)-(13).
In summary, though Beyond Standard Model scenarios (to which MLRSM belongs) usually include mass spectrum at large energy scales, relatively light scalar masses (below TeV level) might be allowed. In this report we have analysed correlations between the scalar masses and the scalar potential parameters within MLRSM framework. We have noted that it is still possible to find some of the scalar fields as light as few hundred GeV even when the new scale is around ∼ 10 TeV, and these spectra are in good agreement with sharp conditions forced by experimental data. Though there are many free parameters in the Higgs sector, present experimental limits on masses of doubly charged Higgs bosons put already first partial constraints on some of these parameters. Moreover, our analysis shows that the lowest limit on doubly charged scalar mass M H ±± 1 can be obtained. The other doubly charged scalar H ±± 2 is theoretically not constrained from below in the discussed model and it can be tuned to any desirable value, in agreement with experimental data. The lowest limit on M H ±± 1 is important, as we know that production of H ±± 1 is much bigger than that of H ±± 2 [16]. Cross sections for the pair productions of doubly charged scalars followed by their leptonic decays have been dis-
cussed toghether with estimation of the SM background at 14 TeV centre of mass energy.
Using an example of the MLRSM model we have shown that we are slowly approaching era at the LHC physics where details of the elusive Higgs sector of the tested theory can be analysed.
This work was supported by Department of Science & Technology, Government of India under the Grant Agreement No. IFA12-PH-34 (INSPIRE Faculty Award), the Research Executive Agency (REA) of the European Union under the Grant Agreement No. PITN-GA-2010264564 (LHCPhenoNet) and Polish National Science Centre (NCN) under the Grant Agreement No. DEC2013/11/B/ST2/04023 and under postdoctoral grant No. DEC-2012/04/S/ST2/00003. MK is supported by the Forszt project co-financed by the EU from the European Social Fund.
- [1] J. Ellis, arXiv:1312.5672.
- [2] G. Aad et al. (ATLAS Collaboration), Phys.Lett. B716 (2012) 1-29, [arXiv:1207.7214].
- [3] S. Chatrchyan et al. (CMS Collaboration), Phys.Lett. B716 (2012) 30-61, [arXiv:1207.7235].
- [4] C. Quigg, Ann.Rev.Nucl.Part.Sci. 59 (2009) 505-555, [arXiv:0905.3187].
- [5] D. Buttazzo, G. Degrassi, P. P. Giardino, G. F. Giudice, F. Sala, et al., JHEP 1312 (2013) 089, [ arXiv:1307.3536].
- [6] V. Branchina and E. Messina, Phys. Rev. Lett. 111 (2013) 241801 [arXiv:1307.5193].
- [7] V. Branchina, E. Messina and A. Platania, JHEP 1409 (2014) 182 [arXiv:1407.4112].
- [8] V. Branchina, E. Messina and M. Sher, Phys. Rev. D91 (2015) 1, 013003 [arXiv:1408.5302].
- [9] F. Jegerlehner, Acta Phys. Polon. B45 (2014) 1167, [arXiv:1304.7813].
- [10] R. Mohapatra and J. C. Pati, Phys.Rev. D11 (1975) 2558.
- [11] G. Senjanovic and R. N. Mohapatra, Phys.Rev. D12 (1975) 1502.
- [12] P. Duka, J. Gluza, and M. Zralek, Annals Phys. 280 (2000) 336-408, [hep-ph/9910279].
- [13] R. Mohapatra, Unification and Supersymmetry. The Frontiers of Quark-Lepton Physics (Springer, New York, 1986).
- [14] CMS Collaboration, Report No. CMS-PAS-HIG-12-005.
- [15] G. Aad et al. (ATLAS Collaboration), Eur.Phys.J. C72 (2012) 2244, [arXiv:1210.5070].
- [16] G. Bambhaniya, J. Chakrabortty, J. Gluza, M. Kordiaczynska, and R. Szafron, JHEP 1405 (2014) 033, [
arXiv:1311.4144].
- [17] A. Melfo, M. Nemevsek, F. Nesti, G. Senjanovic and Y. Zhang, Phys.Rev. D85 (2012) 055018, [arXiv:1108.4416].
- [18] Z. Kang, J. Li, T. Li, Y. Liu and G. -Z. Ning, [arXiv:1404.5207].
- [19] J. Gunion, J. Grifols, A. Mendez, B. Kayser, and F. I. Olness, Phys.Rev. D40 (1989) 1546.
- [20] J. Gluza and M. Zralek, Phys.Rev. D51 (1995) 46954706, [ hep-ph/9409225].
- [21] J. Beringer et al. (Particle Data Group), Phys.Rev. D86 010001 (2012) and 2013 partial update for the 2014 edition. K. A. Olive et al. (Particle Data Group Collaboration), Chin. Phys. C 38 (2014) 090001.
- [22] G. Ecker, W. Grimus, and H. Neufeld, Phys.Lett. B127 (1983) 365.
- [23] R. N. Mohapatra, G. Senjanovic, and M. D. Tran, Phys.Rev. D28 (1983) 546.
- [24] M. Pospelov, Phys.Rev. D56 (1997) 259-264, [hepph/9611422].
- [25] Y. Zhang, H. An, X. Ji and R. N. Mohapatra, Nucl.Phys. B802 (2008) 247, [arXiv:0712.4218].
- [26] A. Maiezza, M. Nemevsek, F. Nesti, and G. Senjanovic, Phys.Rev. D82 (2010) 055022, [arXiv:1005.5160].
- [27] J. Chakrabortty, J. Gluza, R. Sevillano, and R. Szafron, JHEP 1207 (2012) 038, [ arXiv:1204.0736].
- [28] S. Bertolini, A. Maiezza and F. Nesti, Phys.Rev. D89 (2014) 095028, [arXiv:1403.7112].
- [29] D. Guadagnoli and R. N. Mohapatra, Phys.Lett. B694 (2011) 386-392, [arXiv:1008.1074].
- [30] CMS Collaboration, Report No. CMS-PAS-EXO-12-017.
- [31] C. Vuosalo (CMS Collaboration), Report No. CMS-CR2013-155; C. Vuosalo (CMS Collaboration), PoS DIS 2013 (2013) 122 [arXiv:1307.2168].
- [32] F. F. Deppisch, T. E. Gonzalo, S. Patra, N. Sahu, and U. Sarkar, Phys. Rev. D90 (2014) 5, 053014 [arXiv:1407.5384].
- [33] M. Heikinheimo, M. Raidal and C. Spethmann, arXiv:1407.6908.
- [34] A. Datta and A. Raychaudhuri, Phys.Rev. D62 (2000) 055002, [hep-ph/9905421].
- [35] M. Czakon, M. Zralek, and J. Gluza, Nucl.Phys. B573 (2000) 57-74, [hep-ph/9906356].
- [36] M. Czakon, J. Gluza, and J. Hejczyk, Nucl.Phys. B642 (2002) 157-172, [hep-ph/0205303].
- [37] G. Bambhaniya, J. Chakrabortty, S. Goswami, and P. Konar, Phys.Rev. D88 (2013), no. 7 075006, [arXiv:1305.2795].
- [38] These event selection criteria are implemented in PYTHIA. They include lepton transverse momentum ( p T ), pseudorapidity cuts, smearing of leptons, leptonlepton (photon, jet) separation, hadronic activity, missing p T and Z-veto.
- [39] Significance of the signal is conveniently measured by the ratio S/ √ S + B , where S and B are the total number of signal and background events respectively.
|
10.1103/PhysRevD.90.095003
|
[
"G. Bambhaniya",
"J. Chakrabortty",
"J. Gluza",
"T. Jelinski",
"M. Kordiaczynska"
] |
2014-08-04T19:13:05+00:00
|
2015-02-04T23:05:35+00:00
|
[
"hep-ph"
] |
The lowest limits on the doubly charged Higgs boson masses in the minimal left-right symmetric model
|
The doubly charged Higgs bosons $H_{1,2}^{\pm\pm}$ would undoubtedly be clear
messengers of the new physics. We discuss their mass spectrum and show how
experimental data and relations between scalar masses put limits on it. In
particular, both the masses of the particles $H_1^0$, $A_1^0$ that play a
crucial role in FCNC effects and the masses of the additional gauge bosons
$W_2$, $Z_2$ are notably important. For instance, if $M_{H_1^0,A_1^0}\simeq 15$
TeV and $M_{W_2}\simeq 3.76$ TeV then the lowest mass of $H_1^{\pm\pm}$ is 465
GeV. In contrast, due to the freedom in the parameter space of the full scalar
potential, there is no lowest limit on the mass of $H_2^{\pm\pm}$. It is shown
to which signals at hadron colliders such relatively light doubly charged
scalars might correspond. LHC working at $\sqrt{s}=14$ TeV will enter into the
region where existence of such particles with minimal masses can be thoroughly
explored for a much wider parameter space of the minimal and manifest version
of the left-right symmetric model (MLRSM). Taking into account our
considerations and present ATLAS and CMS exclusion limits on $M_{H^{\pm\pm}}$,
there exist already first partial bounds on some of the MLRSM scalar potential
parameters.
|
1408.0775v1
|
## Discovery and characterization of magnetism in sigma-phase intermetallic Fe-Re compounds
J. Cieślak 1* , S. M. Dubiel , M. Reissner 1 2 and J. Tobola 1 1 AGH University of Science and Technology, Faculty of Physics and Applied Computer Science, PL-30-059 Krakow, Poland 2 Institute of Solid State Physics, Vienna University of Technology, A-1040 Wien, Austria
Systematic experimental (vibrating sample magnetometry) and theoretical (electronic structure calculations using charge and spin self-consistent Korringa-Kohn-Rostoker Green's function method) studies were performed on a series of intermetallic sigma-phase Fe100-xRex ( x = 43-53) compounds. Clear evidence was found that all investigated samples exhibit magnetism with an ordering temperature ranging between 65 K for x = 43 and 23 K for x = 53. The magnetism was revealed to be itinerant and identified as a spin-glass (SG) possibly having a re-entrant character. The SG was found to be heterogeneous viz. two regimes could be distinguished as far as irreversibility in temperature dependence of magnetization is concerned: (1) of a weak irreversibility and (2) of a strong one. According to the theoretical calculations the main contribution to the magnetism comes from Fe atoms occupying all five sub lattices. Re atoms have rather small moments. However, the calculated average magnetic moments are highly (ferromagnetic ordering model) or moderately (antiparallel ordering model) overestimated relative to the experimental data.
PACS: 71.15.Mb, 75.30.Cr, 75.50.Bb, 75.50.Lk
## 1. Introduction
The sigma ( ) phase (space group D 14 4h - P42/mnm) is possibly the best-known and most frequently investigated example of Frank-Kasper (FK) phases that occur in alloys in which at least one element is a transition metal. Their characteristic features are high coordination numbers (12-16) and lack of stoichiometry. The originating therefrom topological and chemical complexity as well as the diversity of physical properties that can be tailored by changing constituting elements and/or their relative concentration, make them an attractive yet challenging subject for investigation. The latter is further prompted by the deteriorating effect the FK-phases have on materials in which they precipitate, although attempts have been undertaken to profit from their high hardness and use their precipitation for strengthening purposes e. g. [1,2].
Concerning magnetic properties of in binary alloys, so far only in Fe-Cr and Fe-V systems was definitely revealed to be magnetic [3-5]. However, its magnetism was shown to be more complex than initially regarded viz. in both cases it has a re-entrant character [6]. Furthermore, NMR measurements performed on -FeV samples gave evidence that vanadium atoms present in all five sub lattices are magnetic [7].
Here, based on a systematic experimental (vibrating sample magnetometry, ac susceptibility) and theoretical (electronic structure calculations using charge and spin self-consistent Korringa-Kohn-Rostoker (KKR) Green's function method as well as KKR combined with the coherent potential approximation (CPA) technique) studies a clear evidence is reported in favor of the magnetic properties of in a series of Fe-Re compounds.
## 2. Experimental
## 2. 1. Samples
A series of -Fe100-xRex ( x = 43, 45, 47, 49, 52.6 and 53) was prepared by the following way: powders of elemental iron (3N+ purity) and rhenium (4N purity) were mixed in appropriate proportions and masses (2 g), and next pressed to pellets. The pellets were subsequently melted in an arc furnace under protective atmosphere of argon. The produced ingots were remelted three times to enhance the degree of homogeneity. Next, they were vacuum annealed at 1803 K for 5 hours and, finally, quenched into liquid nitrogen. The mass loses of the fabricated samples were less than 0.01% of their initial values, so it is reasonable to take their nominal compositions as real ones. X-ray diffraction patterns recorded on the powdered samples gave evidence that their crystallographic structure was . More details on structural and electronic properties of in the studied samples can be found elsewhere [8].
## 2. 2. Magnetic measurements
Dc magnetic measurements were performed in a 9 T Physical Properties Measurement System (PPMS) with Vibrating Sample Magnetometer option. Temperature dependence of magnetization was measured under both zero-field cooling (ZFC) and field-cooling (FC) conditions in temperature range 3 to 300 K. Used field was 10 mT. To avoid influence of remanence of superconducting coil and that of earth field, in advance to each measurement field was set exactly to zero with help of a compensation coil. Magnetization curves versus magnetic field were recorded at various temperatures in the range from 4 to 300 K in fields up to 9 T, using field sweep rate of 20 mT/s.
## 3. Theoretical calculations
The charge and spin self-consistent Korringa-Kohn-Rostoker (KKR) Green's function method [9-11] was used to calculate the electronic structure of the studied samples. The crystal potential of muffin-tin (MT) form was constructed within the local density approximation (LDA) framework using the Barth-Hedin formula [12] for the exchangecorrelation part. The group symmetry of the unit cell of the -phase was lowered to a simple tetragonal one to allow for various configurations of Fe/Re atoms. The experimental values of lattice constants and atomic positions were used in all computations [8]. For fully converged crystal potentials electronic density of states (DOS), total, site-composed and -decomposed l DOS (with l max = 2 for Fe and Re atoms) were derived. Fully converged results were obtained for ∼ 120 special k -point grids in the irreducible part of the Brillouin zone but they were also checked for convergence using a denser k -mesh. Electronic DOS were computed using the tetrahedron k -space integration technique and ∼ 700 small tetrahedral. The KKR calculations were carried out for 17 unit cells with different configurations of Fe and Re atoms on all five sub lattices, keeping experimental lattice constants in all cases as constraints. The configurations were chosen in such a way that each possible number of Fe atoms being the nearest-neighbors for a given lattice site, NNFe , had been taken into account. More details relevant to the calculations protocol can be found elsewhere [13, 14].
In order to study the magnetism of the sigma-FeRe alloy spin polarized KKR as well as KKRCPA calculations have been performed. A collinear ordering of magnetic moments on Fe and Re atoms, as allowed by the used computer code, were assumed in a ferromagnetic (FM) state both in the ordered approximants and in the disordered alloys. Parallel a relatively complex model of the magnetic structure, based on antiparallel alignments of magnetic moments within selected sub lattices called APM, was applied (spin-polarized KKR-CPA calculations were also performed for such magnetic unit cell).
## 4. Results and discussion
## 4. 1. Magnetization vs. temperature
FC and ZFC magnetization curves recorded on all samples displayed in Fig. 1 give a clear evidence that all samples (a) are magnetic at temperatures less than 80 K, and (b) show irreversibility. Such behavior is a characteristic feature of spin-glasses and it was also recently revealed for the -phase Fe-Cr and Fe-V intermetallic compounds [6]. The magnetic ordering temperature, TC (susceptibility measurements - discussed below - gave evidence that the magnetic state orders ferromagnetically) was determined from the high temperature side of M T ( ) curves in two ways: (1) from the point of inflection and (2) from extrapolation of the FC -curves to M = 0. As shown in Fig. 2 the two sets of T C-values fit well with each other. It is also evident that the TC decreases with the increase of rhenium content, x , from 65 K at x = 43 to 20 K at x = 53.
Fig. 1(Color online) Field cooled and zerofield cooled magnetization curves of the studied samples recorded in external magnetic field of 10 mT. Figures in the legend stay for the content of rhenium in at%.
Fig. 2 (Color online) Magnetic ordering temperature, TC , versus Re-content, x , as derived from the point of inflection (circles), and from an extrapolation to zero of the MFC(T) curves (triangles). Solid lines represent the linear best-fits to the data.
The temperature, at which the FC and ZFC curves go apart from each other marks the transition into a spin glass (SG) state, and it is known as a spin-freezing temperature, Tf . Below this temperature irreversibility effects characteristic of SG sets in. Another temperature, Tsi , that has been often used in description of spin-glasses is the one at which the ZFC-curve has a maximum. This temperature is interpreted as indicative of the onset of strong irreversibility effects as opposed to weak ones that occur in the range between Tf and Tsi . The dependence of both Tf and Tsi on x for the studied samples is presented in Fig. 3.
Fig. 3
(Color online) Dependence of the spinfreezing temperature T f (triangles) and that of the strong irreversibility temperature T si (circles) on rhenium content, x . The lines represent a guide to the eye.
One can see that both temperatures decrease quite monotonically with x , and eventually, they meet at x 53. Consequently, the weak irreversibility region of the SG defined by the two temperatures shrinks with x . Based on the data shown in Figs. 2 and 3 a magnetic phase diagram of the -phase Fe-Re alloy system could be outlined - see Fig. 4.
Fig. 4
(Color online) Magnetic phase diagram derived from M T ( ) measurements. Paramagnetic state is indicated by PM, ferromagnetic-like one by FM, and spinglass one by RSG. The line marking the transition into the strong irreversibility range of SG is indicated by the dotted line.
From the M T ( )-curves an information on a degree of reversibility of the SG state as a function of temperature can be obtained. For that purpose we calculated a relative difference between MFC(T) and MZFC(T) curves and present it in Fig. 5. It is clear that the degree of irreversibility grows with decreasing temperature. The growth is firstly slow, which reflects a weak irreversibility regime, and next it proceeds much faster indicating a strong irreversibility regime.
Fig. 5
(Color online) Temperature dependence of the relative difference in area under the M FC( T ) and M ZFC( T ) curves shown in Fig. 1. The legend shows the rhenium content in at%.
To get an information on the average degree of irreversibility for a given sample, we calculated an area under the MFC(T) curves, AFC , from the equation:
$$A _ { F C } ( x ) = \underset { 4. 2 } { \overset { 9 0 } { \int } } M _ { F C } ( x, T ) d T \text{ \quad \ \ } ( 1 )$$
And under the MZFC curves, AZFC , from the equation:
$$A _ { Z F C } ( x ) = \underset { 4. 2 } { \overset { 9 0 } { \int } } M _ { Z F C } ( x, T ) d T \quad ( 2 )$$
And next we calculated a relative difference between them, A=100(AFC-AZFC)/AFC , for each value of x . This quantity can be regarded as a measure of the degree of irreversibility. Data obtained in this way are displayed in Fig. 6.
Fig. 6
(Color online) A relative degree of irreversibility, A/A , versus Re content, x , for the studied samples.
It can be easily noticed that the for 43 x
49, A/A, hardly depends on x , while for the two Re-most concentrated samples it has lower values. The latter samples have also the weakest magnetic properties, as shown in Figs. 4 and 10, and the smallest differences between the magnetic ordering temperature and the spinfreezing one.
## 4.2. AC susceptibility
A cusp in the ac susceptibility, , and its frequency dependence have been regarded as the most prominent signature of the spin-glasses. Having this in mind, corresponding measurements were performed on the -Fe47Re53 sample for the frequency ranging between 8 and 1000 Hz. The results shown in Fig. 7 give a clear evidence that has a characteristic cusp
that shifts to higher temperature values as frequency grows. The figure of merit i.e. a relative shift of the spin-freezing temperature, defined by the cusp, per a decade of frequency has a value of 0.008 which is typical of metallic spin-glasses with long-range interactions between
magnetic moments (Ruderman-KittelKasuya-Yosida).
Fig. 7 Real part of the ac susceptibility, ', measured on the -Fe47Re53 sample.
## 4. 3. Magnetization vs. magnetic field
To get more insight into the magnetism of the investigated -FeRe compounds magnetization measurements in external magnetic field, o H , were performed. The obtained curves are presented in Fig. 8a and 8b. It is evident that even at 9 T the saturation state has not been achieved. The latter is in line with the finding that it is the spin-glass that constitutes the magnetic ground state of the investigated samples.
Fig. 8a
(Color online) Magnetization curves measured at 4.2K as a function of an external magnetic field, o H . Dotted lines represent extrapolation of high-field linear parts of the M -curves. The legend marks the content of Re.
Fig. 8b
(Color online) Magnetization curves measured on the -Fe55Re45 alloy at different temperatures (indicated in K) as a function of an external magnetic field, o H. The straight lines were obtained for T=250 and 300K.
Two information relevant to the magnetism of the -phase Fe-Re compounds can be derived from the M(T) and M(H) measurements. First, one can conclude on the kind of magnetic ordering (ferromagnetic or antiferromagnetic). The conclusion can be drawn from the lowfield susceptibility. The linear part of the reciprocal low-field dc magnetic susceptibility
versus temperature is shown in Fig. 9 for all xvalues. The extrapolations to zero of the linear fragments intersect the temperature axis on its positive side, proving thereby that the paramagnetic Curie temperatures (Weiss constants), , are positive i.e. a coupling of the magnetic moments has a ferromagnetic-like character. However, as evidenced in Fig. 8b, the M(H) -curves measured at different temperatures show curvature, which is normally regarded as a sign of a ferromagnetic coupling, at temperatures as high as 200 K which is much higher than the corresponding paramagnetic Curie temperature (the latter 10-15K higher than the corresponding TC -values) as well as the temperature determined by the inflection point of M(H) and denoted above as TC . However, there is also experimental evidence that such curvature in M(H) measurements was observed in Cr+2.7%Fe100-xVx alloys in a paramagnetic state i.e. x 5 [21]. Both systems i.e. -FeRe and Cr-Fe-V have in common an itinerant character of magnetism. Whether or not the presently determined TC -line represents the real magnetic ordering temperature cannot be uniquely decided based on the present experiments. Neutron diffraction measurements which are planned to be carried out will hopefully deliver the solution.
ig. 9
(Color online) Linear parts of the reciprocal low-field dc susceptibility and their extrapolations to zero value for the investigated samples versus temperature, T . The legend shows the concentration of rhenium.
Second, by extrapolating the high-field linear part of the M(H )-curves to zero field, an average magnetic moment per atom, , is estimated and presented in Fig. 10. It can be seen that is a monotonous function of x and its values range between 0.4 B at x = 43 and 0.2 B at x = 53. In other words, an increase of rhenium content causes weakening of magnetism of the investigated alloys. Interestingly, the values of < > are the highest in comparison to those determined for the -phase Fe-V and Fe-Cr compounds in a similar range of composition [5].
Fig. 10
(Color online) Average magnetic moment per atom, , as estimated from the in-field magnetization measurements (circles), and that calculated based on the FM model (dotted line) and on the APM model (solid line).
## 4. 3. Theoretical calculations
## 4.3.1. KKR calculations
The calculations outlined in section 3 yielded magnetic moments of Fe and Re atoms for all considered atomic configurations in each of the five sub lattices. The results obtained with the FM model can be seen in Fig. 11. For each of the five sub lattices there is a cloud of data representing magnetic moments of particular atomic configurations taken into account in the calculations. The Fe-site -values are positive for nearly all configurations and sub lattices, and for every sub lattice there is an increase of with the number of the nearest-neighbor Fe atoms, NNFe . The maximum calculated value of ranges between 2 B for the site A and 3 B for the other four sites. Based on these data, the average magnetic moment for each sub lattice, < >k, (k = A, B, C, D, E) was calculated. Its values are indicated in Fig. 11 by horizontal lines. Here, the highest value of 2.3 B represent Fe-atoms residing on site C, and the lowest one of 0.9 B those that occupy sites A.
Concerning the Re-site magnetic moments, they are significantly smaller, as shown in the lower panel of Fig. 12, and their values for particular configurations oscillate around zero, the maximum amplitude being that for the site B viz. between ±0.15 B. Also in this case the average -values were determined and presented as horizontal lines. One can see that in this case the average moments on Re-atoms residing on sites C and E are zero, while those occupying sites B and D have weakly negative values i.e. -0.02 B for site A and -0.04 B for site D. Knowing the < >k - values for both elements and all sub lattices and the atomic occupancies of the latter as found elsewhere [8], the average magnetic moment per atom (Re or Fe), < >, was computed.
Fig. 11
(Color online) Theoretically calculated, using the FM model, magnetic moments on Fe and Re atoms occupying the five different sub lattices A, B, C, D and E versus number of nearestneighbor Fe atoms, NNFe . The cloud of data seen for each sub lattice represents magnetic moments of different atomic configurations taken into account. Horizontal lines indicate the average values of the moments.
Its dependence on x is shown as a dotted line in Fig. 10. Although it correctly reproduces the experimental concentration dependence of , yet the absolute values are overestimated by a factor of over 2. This failure of the FM approach was also faced in the case of similar calculations performed for the other two intermetallic -phase compounds that exhibit magnetism viz. Fe-Cr [14-16] and Fe-V systems [17]. Consequently, based on group theory and a symmetry analysis, it was shown that an antiparallel ordering on C and D sub lattices could be possible. Therefore a model (APM approach) which allows antiparallel alignment on sites C and D was applied [16]. This model resulted in a significant reduction of < >, hence a better agreement with the experimental values was achieved. The results of the APM approach applied in the present case can be seen in Fig. 12. The main difference, as compared to the results obtained with the FM model (Fig. 11), concerns the sites C and D for which the clouds have separated into two branches: one corresponding to positive values of , the other one to negative ones. The average -values were calculated for all sub lattices and both atoms, and they are indicated in Fig. 12 as horizontal solid lines. The average values for the positive and negative clouds (sub lattices C and D) are marked by horizontal dotted lines. The < >k-values found with the APM model are strongly reduced relative to the corresponding ones presented in Fig. 12, what obviously results in a reduced average moment per atom. The concentration dependence of the latter is presented as a solid line in Fig. 10. A reduction of the discrepancy between the experimental data and the dotted line is clear, albeit the agreement is still unsatisfying even taking into account that the experimental points are underestimated due to the unsaturated nature of the M(H) curves. It seems that the main reason for the observed disaccord between the experimental and theoretical results is the colinear magnetic ordering assumed in the calculations. The experimental findings reported in this paper give a clear evidence that in the ground state the ordering is not co-linear. Nevertheless, it seems that the APM approach is a step forward in a theoretical description of the -phase magnetism as it describes it as a mixture of ferromagnetic and antiferromagnetic features, a necessary condition for the occurrence of spin-glasses.
Fig. 12
(Color online) Theoretically calculated magnetic moments, using the APM model, on Fe and Re atoms occupying the five different sub lattices A, B, C, D and E versus number of nearestneighbor Fe atoms, NNFe . The cloud of data seen for each sub lattice represent different atomic configurations taken into account. Solid lines indicate the average values of the
magnetic moments, whereas the average values for the positive and negative clouds are marked by dotted lines.
## 4.3.1. KKR-CPA calculations
The aforementioned KKR results can be compared to the KKR-CPA calculations, which partially accounted for a random distribution of Fe and Re in the system. Chemical disorder was treated on B, C and either on D or E crystallographic sites, whereas the A position was assumed to be fully occupied by Fe atoms to simplify computations (In fact, according to experimental finding the population of Fe atoms on A sites was 90% [8]). Otherwise, in the case of binary alloys with atoms distributed over thirty sites, the self-consistent CPA cycle would require computing 2 30 (more than 10 ) configurations, a task numerically impossible to 9 do. The number of configurations reasonably decreases to 2 20 (~10 ) or even less viz. 6 2 16 =65536 configurations in the considered cases. In this partially disordered model, we have calculated a spin-polarized density of states and local magnetic moments on Fe and Re sites for Fe17Re13 and Fe15Re15 compounds considering two types of magnetic arrangements (FM and APM).
Fig. 13 (Color online) Total electronic DOS in the FM APM (green, dashed) KKR-CPA
-Fe15Re15 alloy obtained with the (black, solid) and model, as calculated by the method.
The first important difference of the KKR-CPA results with respect to the averaged KKR results was a slightly smaller total magnetic moment (per atom) calculated both for FM and APM models. This is mainly due to almost vanishing Fe moment on the site A as well as to a smaller value of the moment on the site E. We suppose that the CPA approach, which takes into account all possible local atomic arrangements when computing DOS and magnetic moments, is more accurate to describe electronic and magnetic properties of such highly disordered Fe-Re sigma-phase. Thus, one can expect that the results obtained from (i) averaging of finite number of ordered models (the KKR approach), and (ii) disordered model (the KKR-CPA approach), can be slightly different. In the case of other lattice sites, the agreement between both approaches is really fine.
Noteworthy, the magnetic moment for Fe15Re15 (µ = 0.29 B) as found with the KKR-CPA calculations is in very close agreement with the corresponding experimental value (~0.30 B). It can be also seen (Table I) that the average value slightly increases with increasing Fe concentration, which also fits to the experimental data.
It is also interesting to note that from both KKR and KKR-CPA computations, and also for both considered models of the magnetic ordering, large values of Fe magnetic moments were obtained, whatever the crystallographic sites (except for the less populated A site). Further, the lowering of the total magnetic moment can be achieved only by a relative orientation of the Fe-site magnetic moments (locally anti-parallel), without substantial decrease of their absolute values. The magnetic behavior can be best seen on total and site-decomposed DOS calculated in FM and APM states (Figs 13 and 14) for the Fe15Re15 compound.
Table I. Total and Fe magnetic moments (in B per atom) on the five crystallographic sites in Fe-Re sigma phase as computed with the KKR-CPA calculations both for FM and APM model.
| | Total | A-site | B-site | C-site | D-site | E-site |
|-------------|---------|----------|----------|-------------|------------|----------|
| FM | FM | FM | FM | FM | FM | FM |
| Fe 17 Re 13 | 0.93 | -0.13 | 2.49 | 2.67 | 1.65 | 1.52 |
| Fe 15 Re 15 | 0.78 | -0.36 | 2.68 | 2.66 | 1.63 | 1.74 |
| APM | APM | APM | APM | APM | APM | APM |
| Fe 17 Re 13 | 0.32 | -0.15 | 2.72 | -2.20/2.40 | -1.31/1.54 | 1.61 |
| Fe 15 Re 15 | 0.29 | -0.10 | 2.52 | -2.33/+2.34 | -1.49/1.55 | 1.72 |
For example, DOS of Fe on B and C sites, which exhibit the largest spin-polarization, look very similar for both magnetic ordering models. Noteworthy, in the case when the APM ordering was assumed (C and D sites) two types of DOS, of almost anti-symmetric shape, were obtained for Fe atoms (also for Re atoms, but their spin-polarization is actually negligible), represented by corresponding indices (Feu or Fed). In Fig. 14 we show the corresponding DOS (on C and D sites) only for the case of Feu. Interestingly, the similar shape of DOS was also detected for the remaining three sites (A, D and E) where the spinpolarization is smaller. The finding clearly shows that the absolute values of Fe magnetic moments, in so strongly disordered system, are slightly affected by their relative directions.
## 4. 4. Itinerant character of magnetism
Based on the magnetization measurements the character of the magnetism in the studied compounds can be determined. For that purpose one can use the Rhodes-Wohlfarth criterion [20] according to which a magnetism exhibits an itinerant or localized character depending on the value of a ratio between the effective magnetic moment in a paramagnetic phase, eff , and that in the ordered phase, . If the ratio is higher than 1 than the magnetism is itinerant, otherwise it is localized. The values of eff / calculated for the studied compounds are shown in Fig. 15 which illustrates the figure of merit for several chosen localized and itinerant magnets in form of the socalled Rhodes-Wohlfarth plot. The plot clearly demonstrates that the magnetism of the -phase Fe-Re compounds has a highly itinerant character. Although the latter weakens with the increase of the content of iron, yet it remains strongly itinerant even for the iron-most concentrated sample.
Fig. 14 (Color online) Site-decomposed KKRCPA DOS for the sigma-Fe15Re15 alloy as obtained with the APM model.
Fig. 15
(Color online) Rhodes-Wohlfarth plot for various itinerant and localized magnetic systems (open circles) including the -FeCr compounds (open triangles) [4], and presently investigated samples (full circles). The latter are also shown as inset. The horizontal dotted line indicates the critical value above which the magnetism is itinerant.
## 5. Conclusions
Magnetism in a series of -Fe100-xRex intermetallic compounds with x ranging between 43 and 53 was revealed by means of experimental measurements and theoretical calculations. The results obtained in this study can be summarized by drawing the following conclusions:
- 1. All investigated samples, covering the whole range of the sigma-phase occurrence, exhibit magnetism: the ordering temperature changes monotonically between 65K for -Fe53Re47 and 23K for -Fe47Re53.
- 2. The magnetism has a spin-glass phase as the ground state, possibly with a re-entrant character and a ferromagnetic-like phase as an intermediate state: the spin-freezing temperature decreases in a monotonic way with the increase of rhenium content from 50 K to 18 K for the limiting concentrations.
- 3. The spin-glass state can be subdivided into two domains characterized by: (a) a weak irreversibility, and (b) a strong irreversibility, separated by a peak in the ZFC curve.
- 4. The degree of irreversibility strongly grows with a decrease of temperature for all samples. Its temperature-integrated value hardly depends on Re-content for 43 x 49, and is significantly lower for x 53.
- 5. The magnetism has a highly itinerant character, the degree of which decreases with the increase of iron concentration.
- 6. Theoretical calculations admitting a co-linear ferromagnetic ordering yielded overestimated, by a factor of 2.5, values of the average magnetic moment, while those admitting a co-linear but antiparallel ordering on two of five sub lattices resulted in a significant improvement.
- 7. Theoretically calculated absolute values of Fe magnetic moments are high in all applied calculation methods and models, they are also slightly affected by their relative directions. Lowering the average magnetic moment can be achieved by applying various orientations of the partial moments located on the various lattice sites.
## Acknowledgements
The Ministry of Science and Higher Education, Warszawa, is thanked for a support.
## References
- [1] [M. Shibuya, Y. Toda, K. Sawada, H. Kushima, K. Kimura, Mater Sci Eng. A, 592 (2014) 1
- [2] B. Kuhn, M. Talik, L. Niewolak, J. Zurek, H. Hattendorf, P.J. Ennis, W.J. Quadakkers, T. Beck, L. Singheiser, Mater. Sci. Eng. A, 594 (2014) 372
- [3] D. A. Read, E. H. Thomas and J. B. Forsythe, J. Phys. Chem. Solids, 29 , 1569 (1968)
- [4] J. Cieślak, M. Reissner, W. Steiner and S. M. Dubiel, Phys. Stat. Sol (a), 205, 1794 (2008)
- [5] J. Cieślak, B. F. O. Costa, S. M. Dubiel, M. Reissner and W. Steiner, J. Magn. Magn. Mater., 321, 2160 (2009)
- [6] R. Barco, P. Pureur , G. L. F. Fraga, S. M. Dubiel, J. Phys.: Condens. Matter, 24 (2012) 046002
- [7] S. M. Dubiel, J. R. Tozoni, J. Cieślak, D. C. Braz, E. L. Gea Vidoto, T. J. Bonagamba, Phys. Rev. B, 81 (2010) 184407
- [8] J. Cieslak, S. M. Dubiel, J. Tobola, J. Zukrowski, Mater. Chem. Phys., 139 (2013) 590
- [9] W.H. Butler, P. Dederichs, A. Gonis, R. Weaver, editors. Applications of multiple scattering theory to materials science, vol. 253; 1992. Chapter III. MRS Symposia Proceedings, MRS Pittsburgh.
- [10] T. Stopa, S. Kaprzyk, J. Tobola, J. Phys. Condens. Matter., 16 (2004) 4921
- [11] A. Bansil, S. Kaprzyk, P.E. Mijnarends, J. Tobola, Phys. Rev. B, 60 (1999) 13396
- [12] S. Kaprzyk, P.E. Mijnarends, J. Phys. C, 19 (1986) 1283
- [13] J. Cieslak, J. Tobola, S.M. Dubiel, Phys. Rev. B, 81 (2010) 174203
- [14] J. Cieslak, J. Tobola, S.M. Dubiel, S. Kaprzyk, W. Steiner, M. Reissner, J. Phys.
- Condens. Matter., 20 (2008) 235234
- [15] J. Pavlů, J. Vřeštal, M. Šob, Intermetallics, 18 (2010) 212
- [16] J. Cieslak, J. Tobola, S. M. Dubiel, W. Sikora, Phys. Rev. B, 82 (2010) 224407
- [17] J. Cieslak, J. Tobola, S. M. Dubiel, Intermetallics, 22 (2012) 7
- [18] K. H. Fischer, W. Kinzel, J. Phys. C, 17 (1984) 4479
- [19] K. H. Fischer, Phys. Stat. Sol. (a), 130 (1995) 13
- [20] E. P. Wohlfarth, J. Magn. Magn. Mater., 7 (1978) 113
- [21] R. S. Fishman, V. Yu. Galkin, W. A. Ortiz, J. Phys.: Condens. Matter, 10 (1998) 6347
|
10.1063/1.4900853
|
[
"J. Cieslak",
"S. M. Dubiel",
"M. Reissner",
"J. Tobola"
] |
2014-08-04T19:17:20+00:00
|
2014-08-04T19:17:20+00:00
|
[
"cond-mat.str-el"
] |
Discovery and characterization of magnetism in sigma-phase intermetallic Fe-Re compounds
|
Systematic experimental (vibrating sample magnetometry) and theoretical
(electronic structure calculations using charge and spin self-consistent
Korringa-Kohn-Rostoker Green function method) studies were performed on a
series of intermetallic sigma-phase Fe(100-x)Re(x) (x = 43-53) compounds. Clear
evidence was found that all investigated samples exhibit magnetism with an
ordering temperature ranging between 65 K for x = 43 and 23 K for x = 53. The
magnetism was revealed to be itinerant and identified as a spin-glass (SG)
possibly having a re-entrant character. The SG was found to be heterogeneous
viz. two regimes could be distinguished as far as irreversibility in
temperature dependence of magnetization is concerned: (1) of a weak
irreversibility and (2) of a strong one. According to the theoretical
calculations the main contribution to the magnetism comes from Fe atoms
occupying all five sub lattices. Re atoms have rather small moments. However,
the calculated average magnetic moments are highly (ferromagnetic ordering
model) or moderately (antiparallel ordering model) overestimated relative to
the experimental data.
|
1408.0776v1
|
## Oil and water: a two-type internal aggregation model
Elisabetta Candellero ∗ 1 , Shirshendu Ganguly † 2 , Christopher Hoffman ‡ 3 , and Lionel Levine § 4
1 University of Warwick 2,3 University of Washington 4 Cornell University
August 4, 2014
## Abstract
We introduce a two-type internal DLA model which is an example of a non-unary abelian network. Starting with n 'oil' and n 'water' particles at the origin, the particles diffuse in Z according to the following rule: whenever some site x ∈ Z has at least 1 oil and at least 1 water particle present, it fires by sending 1 oil particle and 1 water particle each to an independent random neighbor x ± 1. Firing continues until every site has at most one type of particles. We establish the correct order for several statistics of this model and identify the scaling limit under assumption of existence.
## 1 Introduction and Main Results
We investigate a new interacting particle system on Z that can be considered as a model of mutual diffusion . Two particle species, called for convenience as oil and water , diffuse on Z until there is no site that has both an oil and water particle. We start with n oil and n water particles at the origin. At each discrete time step, if at least 1 oil particle and at least 1 water particle are present at x ∈ Z then x fires by sending 1 oil particle and 1 water particle each to an independent random neighbor x ± 1 with equal probability. The system fixates when no more firing is possible, that is, when every site has particles of at most one type.
How many firings are required to reach fixation? How far is the typical particle from the origin upon fixation? Our main results address these two questions.
Definition 1. For x ∈ Z let u x ( ) be the total number of times x fires before fixation. The random function u : Z → N is called the odometer of the process.
∗ [email protected]. Supported by the Austrian Science Fund (FWF): E-1503W01230, and by a Marie Curie Career Integration Grant PCIG09-GA2011-293619.
† [email protected]. Partially supported by NSF grant DMS-0847661.
‡ [email protected]. Supported by NSF grant DMS-1308645 and NSA grant H98230-13-10827.
§ http://www.math.cornell.edu/ ~ levine . Partially supported by NSF grant DMS-1243606.
In the above informal description we have assumed that all sites fire in parallel in discrete time, but in fact this system has an abelian property : the distribution of the odometer and of the particles upon fixation do not depend on the order of firings (Lemma 2.2).
Our first result concerns the order of magnitude of the odometer.
Theorem 1.1. There exist positive numbers glyph[epsilon1], c, C such that for large enough n, i.
$$\mathbb { P } \left ( \sup _ { x \in \mathbb { Z } } u ( x ) > C n ^ { 4 / 3 } \right ) < e ^ { - n ^ { \epsilon } }$$
ii.
$$\mathbb { P } \left ( \inf _ { x \colon | x | \leq c n ^ { 1 / 3 } } u ( x ) < c n ^ { 4 / 3 } \right ) < e ^ { - n ^ { \epsilon } }.$$
The next result shows that most particles do not travel very far: all but a vanishing fraction of the particles at the end of the process are supported on an interval of length n α , for any exponent α > 1 3. / Formally, for r > 0 let F r ( ) be the number of particles that fixate outside the interval [ -r, r ].
Theorem 1.2. For sufficiently small glyph[epsilon1] > 0 there exists δ > 0 such that
$$\mathbb { P } \left ( F \left ( n ^ { \frac { 1 } { 3 } + \varepsilon } \right ) > n ^ { 1 - \frac { \epsilon } { 2 } } \right ) < e ^ { - n ^ { \delta } }.$$
Theorem 1.1 shows that the odometer function scales like n 4 3 / . Theorem 1.2 motivates the conjecture that the proper scaling factor in the spatial direction should be n 1 3 / . We conjecture that the scaling limit of the odometer exists, and under the assumption of existence we identify the limiting function.
Let ˜ := u E u .
Conjecture 1.3. (i) For any δ > 0
$$\mathbb { P } \left ( \sup _ { x \in \mathbb { Z } } \left | \frac { u ( x ) - \tilde { u } ( x ) } { n ^ { 4 / 3 } } \right | > \delta \right ) \rightarrow 0, \text{ \ as } n \rightarrow \infty.$$
(ii) There is a function w : R → R such that
$$\frac { \tilde { u } ( \lfloor n ^ { 1 / 3 } \xi \rfloor ) } { n ^ { 4 / 3 } } \, \rightarrow \, w ( \xi ),$$
uniformly in ξ .
Simulations support Conjecture 1.3, as shown in Figure 1. Conditionally on Conjecture 1.3, the following result identifies the limit w x ( ) exactly.
Theorem 1.4. Assuming Conjecture 1.3, the function w appearing in (2) must equal
$$w ( x ) = \begin{cases} \frac { 1 } { 7 2 \pi } \left ( ( 1 8 \pi ) ^ { 1 / 3 } - | x | \right ) ^ { 4 }, & | x | < ( 1 8 \pi ) ^ { 1 / 3 } \\ 0 & | x | \geq ( 1 8 \pi ) ^ { 1 / 3 }. \end{cases}$$
Figure 1: Graph of the odometer function of the oil and water model in Z with n = 360000 particles of each type started at the origin: for each x ∈ Z a box is drawn centered at ( x, u x ( )) where u x ( ) is the number of oil-water pairs fired from x . The curve f ( x ) = 1 72 π ((18 πn ) 1 3 / -| | x ) 4 appears in gray. Red and blue vertical bars represent the final configuration of oil and water particles respectively; the height of the bar is proportional to the number of particles.
Figure 2: Oil and water in Z 2 with n = 2 22 particles of each type started at the origin. Each site where particles stop is shaded red or blue according to whether oil or water particles stopped there. The intensity of the shade indicates the number of particles. We believe that the limit shape is a disk of radius of order n 1 4 / .
The oil and water model can be defined on any graph and in particular on higherdimensional lattices. Figure 2 shows an oil and water configuration in Z 2 . In Section 7 we conjecture the relevant exponents in Z d for d ≥ 2.
## 1.1 Related models: internal DLA and abelian networks
In internal DLA , each of n particles started at the origin in Z d performs a simple random walk until reaching an unoccupied site. The resulting random set of n occupied sites is close to a Euclidean ball [14]. Internal DLA is one of several models known to have an abelian property according to which the order of moves does not affect the final outcome.
Dhar [7] proposed certain collections of communicating finite automata as a broader class of models with this property. Until recently the only examples studied in this class have been unary networks (or their 'block renormalizations' as proposed in [7]). Informally, a unary network is a system of local rules for moving indistinguishable particles on a graph, whereas a non-unary network has different types of particles. It is not as easy to construct non-unary examples with an abelian property, but they exist. Alcaraz, Pyatov and Rittenberg studied a class of non-unary examples which they termed two-component sandpile models [1], and asked whether there is a nontrivial example with two particle species such that the total number of particles of each type is conserved. Oil and water has this conservation property,
but differs from the two-component sandpile models in that any number of particles of a single type may accumulate at the same vertex and be unable to fire.
Bond and Levine [6] developed Dhar's idea into a theory of abelian networks and proposed two non-unary examples, oil and water and abelian mobile agents . Can such models exhibit behavior that is 'truly different' from their unary cousins? This is the question that motivates the present paper. Theorem 1.2 shows that oil and water has an entirely different behavior than internal DLA: all but a vanishing fraction of the 2 n particles started at the origin stop within distance n 1 3 + ε (versus n for internal DLA).
In contrast to internal DLA where there is now a detailed picture of the fluctuations in all dimensions [2, 3, 4, 10, 11, 12], there is not even a limiting shape theorem yet for oil and water. Simulations in Z d indicate a spherical limit shape (Figure 2) with radius of order n 1 ( / d +2) .
## 1.2 Main ideas behind the proofs.
In this section we present informally the key ideas behind the proofs. We start with a definition.
Definition 2. For any x ∈ Z and t ∈ { 0 1 , , . . . ∞} let η 1 ( x, t ) and η 2 ( x, t ) represent the number of oil and water particles respectively, at position x at time t .
From now on, let
$$P _ { t } = \sum _ { x \in \mathbb { Z } } \min ( \eta _ { 1 } ( x, t ), \eta _ { 2 } ( x, t ) )$$
be the number of co-located oil-water pairs at time t . Furthermore, let x t denote the left most site with a co-located oil-water pair at time t .
The process is run in phases: we start by firing a pair at the origin and, inductively, at each time t > 0 we locate the vertex x t and let it fire. More precisely, unless otherwise stated, in the following we will make the tacit assumption that at every time t the process fires one pair from x t . By definition, P 0 = n and the process stops at the first time τ for which P τ = 0. Now denote by l t = η 1 ( x t -1 , t ) -η 2 ( x t -1 , t ) the excess of oil particles over water particles at the left neighbor of x t and by r t = η 1 ( x t +1 , t ) -η 2 ( x t +1 , t ) the excess at the right neighbor.
Conditional upon knowing the process up to time t , the random variable
$$Z _ { t } \coloneqq P _ { t + 1 } - P _ { t }$$
can have four possible distributions. To distinguish them, we denote the conditional Z t +1 by the random variables ξ , ξ 1 2 , ξ 3 , ξ 4 which are described in the following table:
| i | P ( ξ i = - 1) | P ( ξ i = 0) | P ( ξ i = 1) | used when |
|-----|------------------|----------------|----------------|-------------------------------|
| 1 | 1 / 4 | 1 / 2 | 1 / 4 | l t r t < 0 |
| 2 | 1 / 4 | 3 / 4 | 0 | exactly one of l t , r t is 0 |
| 3 | 1 / 2 | 1 / 2 | 0 | l t = r t = 0 |
| 4 | 0 | 1 | 0 | l t r t > 0 |
Note that ξ 1 has mean zero, whereas ξ 2 and ξ 3 have negative means, and ξ 4 is degenerate at 0. Since all such expected values are less than or equal to zero,
The main idea of the proof for the upper bound in Theorem 1.1 is as follows: Define the auxiliary random variables
$$N _ { i } ( t ) = \# \{ s \leq t \ \colon \, Z _ { s } \stackrel { d } { = } \xi _ { i } \}.$$
It is clear from the definition that
$$\sum _ { i = 1 } ^ { \tau } Z _ { i } = P _ { \tau } - P _ { 0 } = - n$$
where τ is the stopping time of the process where all oil and water have been separated ( P τ = 0). Now, informally ∑ τ i =1 Z i is the sum of N 1 ( τ ) + N 2 ( τ ) variables with mean at most -1 4 and / N 3 ( τ ) variables with mean 0.
Because of the negative and zero drifts respectively, the sum of the negative mean variables is of order O ( -( N 1 ( τ )+ N 2 ( τ )) ) , whereas the sum of the zero mean variables is roughly O ( √ N 3 ( τ )) (by square root fluctuations of the symmetric random walk). Thus roughly
$$\sum _ { i = 1 } ^ { \tau } Z _ { i } \leq - N _ { 1 } ( \tau ) - N _ { 2 } ( \tau ) + \sqrt { N _ { 3 } ( \tau ) }.$$
We then argue by contradiction: conditional upon the event that the odometer is 'very large' somewhere (i.e. larger than Cn 4 3 / for a suitable big constant C ), we show that it is likely that N 1 ( τ ) + N 2 ( τ ) is 'sufficiently large' compared to √ N 3 ( τ ). This in turn implies that
$$\sum _ { i = 1 } ^ { \tau } Z _ { i } < - n$$
which is a contradiction.
To prove the lower bound we first establish a gradient bound on the odometer using the upper bound. In other words, we show that there exists a constant c such that, with high probability, for x ∈ [0 , cn 1 3 / ] we have
$$u ( x ) - u ( x + 1 ) \geq n / 2.$$
This in turn implies that with high probability
$$u ( 0 ) = \Omega ( n ^ { 4 / 3 } ),$$
since
$$0 \leq u ( \lfloor c n ^ { 1 / 3 } \rfloor ) \leq u ( 0 ) - \frac { n } { 2 } \lfloor c n ^ { 1 / 3 } \rfloor.$$
Theorem 1.2 follows from the proof of the lower bound.
Lastly we discuss the proof of Theorem 1.4. Clearly by symmetry of the process about the origin the function w ( ) is symmetric as well. · We show that Conjecture 1.3 implies that the limiting function is smooth on the positive real axis and in particular satisfies
$$w ^ { \prime \prime } ( x ) = \sqrt { \frac { 2 } { \pi } } w ( x )$$
with certain boundary conditions. At this point, Theorem 1.4 follows by identifying a solution of the above boundary value problem, and using uniqueness of the solution.
## 1.3 Outline
This article is structured as follows. In Section 2 we give a rigorous definition of the model. Furthermore, in order to facilitate our proofs, we define two different versions of the process and prove that they both terminate in finite time with probability one. We then construct a coupling between the two versions, which in particular implies that they terminate with the same odometer function and the same distribution of particles.
In Section 3 we show a polynomial bound on the stopping time τ . We start by showing that the number of pairs P t can be stochastically dominated by a certain lazy random walk with long holding times, started and reflected at n . In order to get a rough upper bound on τ , it suffices to bound the hitting time of zero for this walk, which we show is of order n 4 . Consequently, we get direct polynomial bounds on the support and the maximum of the odometer function. We then improve the last bound, proving the upper bound in Theorem 1.1 in Section 4.
Section 5 is devoted to proving the lower bound in Theorem 1.1. As a byproduct of the proof, Theorem 1.2 follows.
In Section 6 we prove the conditional Theorem 1.4. Section 7 consists of open questions and a conjecture about the process on higher dimensional lattices.
## 2 Rigorous definition of the model
As the underlying randomness for our model we will take a countable family of independent random variables
$$\omega = ( X _ { k } ^ { x }, Y _ { k } ^ { x } ) _ { x \in \mathbb { Z }, k \in \mathbb { N } }$$
with P ( X x k = ± 1) = P Y ( x k = ± 1) = 1 2 . For each x ∈ Z the sequences ( X x k ) k ≥ 1 and ( Y x k ) k ≥ 1 are called the stacks at x . Denote by Ω the set of all stacks ω .
The stacks will have the following interpretation (described formally below): On the k th firing from site x , an oil particle steps from x to x + X x k and a water particle steps from x to x + Y x k .
For any value K ∈ N , a firing sequence is a sequence of vertices s = ( x , . . . , x 0 K -1 ) with each x k ∈ Z . Recall from Definition 2 that the random variables η i ( x, t ) represent the number of particles of type i (type 1 are oil and type 2 are water particles) at location x at time t . Given such a sequence and an initial state η 1 ( · , 0) , η 2 ( · , 0) we define the oil and water process ( η 1 ( · , k ) , η 2 ( · , k )) K k =0 inductively by
$$\eta _ { 1 } ( \cdot, k + 1 ) = \eta _ { 1 } ( \cdot, k ) - \delta ( x _ { k } ) + \delta ( x _ { k } + X _ { i _ { k } } ^ { x _ { k } } )$$
$$\eta _ { 2 } ( \cdot, k + 1 ) = \eta _ { 2 } ( \cdot, k ) - \delta ( x _ { k } ) + \delta ( x _ { k } + Y _ { i _ { k } } ^ { x _ { k } } )$$
where i k = # { j ≤ k : x j = x k } . Here δ x ( ) is the function taking value 1 at x and 0 elsewhere.
## 2.1 Least action principle and abelian property
Definition 3. Let s = ( x , . . . , x 0 K -1 ) be a firing sequence. We say that s is legal for ( η 1 ( · , 0) , η 2 ( · , 0)) if
$$\min ( \eta _ { 1 } ( x _ { k }, k ), \eta _ { 2 } ( x _ { k }, k ) ) \geq 1,$$
for all k , 0 ≤ k ≤ K -1. We say that s is complete for ( η 1 ( · , 0) , η 2 ( · , 0)) if the final configuration ( η 1 ( · , K ) , η 2 ( · , K )) satisfies
$$\min ( \eta _ { 1 } ( x, K ), \eta _ { 2 } ( x, K ) ) \leq 0,$$
for all x ∈ Z .
Making Definition 1 precise we define the odometer of a firing sequence s to be the function u s : Z → N given by
$$u _ { s } ( x ) = \# \{ 0 \leq k < K \ \colon \, x _ { k } = x \}.$$
How does u s depend on s ? The least action principle for abelian networks addresses this question.
Lemma 2.1. (Least Action Principle, [6]) Let s and s ′ be firing sequences. If s is legal for ( η 1 ( · , 0) , η 2 ( · , 0)) and s ′ is complete for ( η 1 ( · , 0) , η 2 ( · , 0)) , then u s ( x ) ≤ u s ′ ( x ) for all x ∈ Z .
We remark that this statement holds pointwise for any stacks ω , even if s and s ′ are chosen by an adversary who knows the stacks.
In this paper we will not use the full strength of Lemma 2.1. Only the following corollaries will be used.
Lemma 2.2. (Abelian property [6]) For fixed stacks ω and fixed initial state ( η 1 ( · , 0) , η 2 ( · , 0)) ,
- (i) If there is a complete firing sequence of length K then every legal firing sequence has length ≤ K .
- (ii) If firing sequences s and s ′ are both legal and complete, then u s = u s ′ .
- (iii) Any two legal and complete firing sequences result in the same final state ( η 1 ( · , K ) , η 2 ( · , K )) .
## 2.2 Leftmost convention
We fix the initial state
$$\eta _ { 1 } ( \cdot, 0 ) = \eta _ { 2 } ( \cdot, 0 ) = n \delta ( 0 )$$
of n oil and n water particles at the origin and none elsewhere. By Lemma 2.2, for fixed stacks ω , the final state does not depend on the choice of legal and complete firing sequence. However, many of our lemmas will require choosing a particular sequence. Unless otherwise specified, we always fire the leftmost legal site :
$$x _ { k } = \min \{ x \in \mathbb { Z } \ \colon \min ( \eta _ { 1 } ( x, k ), \eta _ { 2 } ( x, k ) ) \geq 1 \}.$$
If the set on the right side is empty, then ( η 1 ( · , k ) , η 2 ( · , k )) is the final state and x k is undefined. We denote by τ ∈ N ∪ {∞} the first time at which this happens:
$$\tau = \min \{ k \geq 0 \ \colon \min ( \eta _ { 1 } ( x, k ), \eta _ { 2 } ( x, k ) ) = 0 \ \text{for all } x \in \mathbb { Z } \}. \quad \quad ( 1 1 )$$
Note that everything so far holds pointwise in the stacks ω . The first statement involving probability is that τ < ∞ , almost surely in ω , and we postpone its proof to Section 3.
## 2.3 Merged stacks
Wedescribe another process ( η ′ 1 ( · , · ) , η ′ 2 ( · , · )) which has the same law as the process ( η 1 ( · , · ) , η 2 ( · , · )) defined in Definition 2. Throughout the rest of the article we will refer to the process ( η 1 ( · , · ) , η 2 ( · , · )) as Version 1 and the modified process as Version 2.
In Version 2, analogous to (6) the source of randomness is a set of independent variables (modified stacks)
$$\omega ^ { \prime } & = ( X _ { i } ^ { x }, Y _ { i } ^ { x }, \bar { X } _ { i } ^ { x }, \bar { Y } _ { i } ^ { x } ) _ { x \in 3 \mathbb { Z }, i \in \mathbb { N } }.$$
with P ( X x i = ± 1) = P Y ( x i = ± 1) = P ( ¯ X x i = ± 1) = P ( ¯ Y x i = ± 1) 1 2 . Denote by Ω ′ the set of all stacks ω . ′ Note that stacks are located only at every x ∈ 3 Z .
Informally, the i th firing from x ∈ 3 Z uses moves X x i and Y x i as before, but the i th firing from the set { x -1 , x +1 } uses moves ± ¯ X x i and ± ¯ Y x i according to whether the firing was from x -1 or x +1 respectively. We refer to ( ¯ X ,Y x i ¯ x i ) i ≥ 1 as the merged stacks of x -1 and x +1.
Formally, given a firing sequence s = ( x , . . . , x 0 K -1 ), we define ( η ′ 1 ( · , k ) , η ′ 2 ( · , k )) inductively as follows. If x k ∈ 3 Z then
$$\eta _ { 1 } ^ { \prime } ( \cdot, k + 1 ) = \eta _ { 1 } ^ { \prime } ( \cdot, k ) - \delta ( x _ { k } ) + \delta ( x _ { k } + X _ { i _ { k } } ^ { x _ { k } } )$$
$$\eta _ { 2 } ^ { \prime } ( \cdot, k + 1 ) = \eta _ { 2 } ^ { \prime } ( \cdot, k + 1 ) - \delta ( x _ { k } ) + \delta ( x _ { k } + Y _ { i _ { k } } ^ { x _ { k } } )$$
$$\binom { r x _ { k } } { i _ { k } } \\ \binom { r x _ { k } } { i _ { k } }$$
where i k = # { j ≤ k : x j = x k } . If x k ∈ 3 Z ± 1 then
$$\eta _ { 1 } ^ { \prime } ( \cdot, k + 1 ) = \eta _ { 1 } ^ { \prime } ( \cdot, k ) - \delta ( x _ { k } ) + \delta ( x _ { k } \mp \bar { X } _ { i _ { k } ^ { \prime } } ^ { x _ { k } \mp 1 } )$$
$$\eta _ { 2 } ^ { \prime } ( \cdot, k + 1 ) = \eta _ { 2 } ^ { \prime } ( \cdot, k + 1 ) - \delta ( x _ { k } ) + \delta ( x _ { k } \mp \bar { Y } _ { i _ { k } ^ { \prime } } ^ { x _ { k } \mp 1 } )$$
$$\overset { \rho } { \rho } x _ { k } \mp 1 \\ \overset { \rho } { \overset { \rho } { \rho } } _ { k } \\ \overset { \rho } { \overset { \rho } { \rho } } _ { k } \mp 1 \\ \overset { \rho } { \overset { \rho } { \rho } } _ { k }$$
where i ′ k = # { j ≤ k : x j ∈ { x , x k k ∓ }} 2 .
To compare the modified process to the original we use the following Proposition.
Proposition 2.3. Let ( Z i ) i ∈ I be independent uniform ± 1 -valued random variables indexed by a countable set I . Let i 1 , i 2 , . . . ∈ I be a sequence of distinct random indices and ξ , ξ 1 2 , . . . a sequence of ± 1 -valued random variables such that for all k ≥ 1 both i k and ξ k are measureable with respect to F k -1 := σ Z ( i glyph[lscript] ) 1 ≤ ≤ -glyph[lscript] k 1 . Then ( ξ k Z i k ) k ≥ 1 is an i.i.d. sequence.
Proof. We proceed by induction on glyph[lscript] to show that ( ξ k Z i k ) 1 ≤ ≤ k glyph[lscript] is i.i.d. Since ξ glyph[lscript] and i glyph[lscript] are F glyph[lscript] -1 -measureable, and i glyph[lscript] is distinct from i 1 , . . . , i glyph[lscript] -1 we have
$$\mathbb { E } ( \xi _ { \ell } Z _ { i _ { \ell } } | \mathcal { F } _ { \ell - 1 } ) = \xi _ { \ell } \mathbb { E } ( Z _ { i _ { \ell } } | \mathcal { F } _ { \ell - 1 } ) = 0.$$
Since ξ Z glyph[lscript] i glyph[lscript] is ± 1-valued the proof is complete.
Recall Ω the set of all stacks ω defined by the original process, and Ω ′ , the set of all stacks ω ′ defined by the modified process. Let τ ( ω ) denote the stopping time of the sequence ω and similarly τ ( ω ′ ). Furthermore, denote by ( o , w τ τ )( ω ) the final configuration after performing ω , and by ( o , w ′ τ ′ τ )( ω ′ ) the final configuration of the process after performing ω ′ .
Lemma 2.4. There is a measure-preserving map φ : Ω ′ → Ω such that with the leftmost convention,
$$( \eta ^ { \prime } _ { 1 } ( \cdot, \cdot ), \eta ^ { \prime } _ { 2 } ( \cdot, \cdot ) ) \stackrel { d } { = } ( \eta _ { 1 } ( \cdot, \cdot ), \eta _ { 2 } ( \cdot, \cdot ) ).$$
In particular, τ ( φ ω ( ′ )) = τ ( ω ′ ) , ( o , w ′ τ ′ τ )( ω ′ ) = ( o , w τ τ )( φ ω ( ′ )) and the odometer counts in both the processes are same.
Proof. Given ω ′ ∈ Ω , let ′ x , x 1 2 , . . . be the resulting leftmost legal firing sequence, and let u x ( ) be the number of firings performed at x in the modified process. We set φ ω ( ′ ) = ( X ,Y x i x i ) x ∈ Z ,i ∈ N where if i ≤ u x ( ) then X x i (resp. Y x i ) is the direction in which the i th oil (resp. i th water) exited x in the modified process.
If i > u x ( ) then we make an arbitrary choice (e.g. split the unused portion of each merged stack into even and odd indices).
It is important to note, that each x k is measurable with respect to the σ -algebra F ′ k -1 generated by the stack variables in ω ′ used before time k , and that each stack variable in ω ′ is used at most once. The stack variables used at time k have the form ξX , ξY where X,Y are stack variables not yet used and ξ = 1 x k ∈ 3 Z ∪ (3 Z +1) -1 x k ∈ 3 Z -1 is a random sign that is measurable with respect to F ′ k -1 . Conditional on F ′ k -1 , ξX and ξY are independent uniform ± 1 random variables, and φ is measure-preserving by Proposition 2.3.
## 2.4 Constant convention
To avoid cumbersome notation, we will often use the same letter (generally C , C ′ , c , c ′ , glyph[epsilon1] and δ ) for a constant whose value may change from line to line.
## 3 Preliminary bound on the stopping time
In this section we prove a preliminary bound on the stopping time τ , defined in (11). The bound is very crude and will be improved in the next section where we prove the upper bound in Theorem 1.1. However this bound will be used in several places throughout the article.
Lemma 3.1. τ is finite almost surely. Moreover i. For any given glyph[epsilon1] > 0 , there exists c = ( ) c glyph[epsilon1] > 0 such that
$$\mathbb { P } \left ( \tau > n ^ { 4 + 2 \epsilon } \right ) \leq e ^ { - n ^ { c } }.$$
$$i i. \ \mathbb { E } ( \tau ) \leq 1 6 n ^ { 4 }.$$
This result has an immediate consequence, given by the next Corollary.
Corollary 3.2. Given glyph[epsilon1] > 0 there exists a constant c such that
$$\mathbb { P } \left ( u ( x ) = 0, \ \forall \, x \in \mathbb { Z } \, \colon | x | \geq n ^ { 4 + \epsilon } \right ) \geq 1 - e ^ { - n ^ { c } }.$$
Before proving Lemma 3.1 we recall from subsection 1.2 that
$$P _ { t } = \sum _ { x \in \mathbb { Z } } \min ( \eta _ { 1 } ( x, t ), \eta _ { 2 } ( x, t ) )$$
denotes the total number of co-located oil and water pairs at time t. By definition, the process stops at τ when P τ = 0 . In order to prove Lemma 3.1 we follow the four steps described in the following.
- · First we define a lazy reflected random walk R i on { 0 1 , , . . . , n } started and reflected at n , and stopped upon hitting 0. Define the following stopping time:
$$\tau ^ { \prime } \coloneqq \min _ { i \geq 1 } \{ R _ { i } = 0 \}.$$
- · Consequently, we use P j to define a series of stopping times t i , and we show that we can bound the tails of the waiting times defined as
$$\text{Wait} _ { i } \coloneqq t _ { i + 1 } - t _ { i }.$$
- · Next, we show that P t i is stochastically dominated by R i .
- · Finally, we prove Lemma 3.1 by combining information about the distributions of τ ′ and the waiting times Wait . i
We now define a lazy random walk. The standard definition is when the simple random walk on Z does not jump with probability 1 / 2 and otherwise jumps uniformly to one of the neighbors. Throughout the rest of the article many arguments involve random walks with varying degree of laziness. Here the lazy reflected random walk R i is defined as follows. It starts with R 0 = n . Inductively, if R i = n then R i +1 = n or n -1, with probability 3 / 4 and 1 4 respectively. / Otherwise, if R i ∈ (0 , n ) then it performs a lazy random walk, i.e.,
$$R _ { i + 1 } = \begin{cases} R _ { i } + 1 & \text{w.p. } 1 / 4 \\ R _ { i } - 1 & \text{w.p. } 1 / 4 \\ R _ { i } & \text{w.p. } 1 / 2 \end{cases}.$$
The walk terminates at the stopping time τ ′ , defined in Equation (14).
To perform the second and third steps, we analyze P t and show that it is a supermartingale. This was described heuristically in subsection 1.2 and here we make this concept formal, as this fact is the key to many of the subsequent arguments.
If at time t a site x emits a pair, then P t +1 is either P t -1 , P , t or P t +1. Moreover the distribution of P t +1 -P t depends on the state of the neighbors of x at time t in a fairly simple way.
Define G t to be the filtration given by the first t firings. Recall from Definition 2 that η 1 ( x, t ) and η 2 ( x, t ) are the number of oil and water particles, respectively, at x at time t .
Definition 4. Define for any x ∈ Z and non negative integer time t including infinity
$$g _ { t } ( x ) \coloneqq \eta _ { 1 } ( x, t ) - \eta _ { 2 } ( x, t ).$$
We say that at time t a site x ∈ Z has type oil , 0 or water depending on the type of the majority of particles at the site, i.e., whether g t ( x ) is positive, zero or negative respectively.
We look at how P t changes conditional on the filtration at time t . Formally, we look at
$$Z _ { t } \coloneqq P _ { t + 1 } - P _ { t }.$$
At this point we observe that Z t can have four possible distributions conditioned on G t . Namely, define four random variables ξ , ξ 1 2 , ξ 3 , ξ 4 as follows:
$$\xi _ { 1 } \coloneqq \begin{cases} \begin{array} { c c } 0 & \text{w.p.} & 1 / 2 \\ 1 & \text{w.p.} & 1 / 4 \\ - 1 & \text{w.p.} & 1 / 4 \end{array} \end{cases} \quad \xi _ { 2 } \coloneqq \begin{cases} \begin{array} { c c } 0 & \text{w.p.} & 3 / 4 \\ - 1 & \text{w.p.} & 1 / 4 \end{cases} \end{cases}$$
$$\xi _ { 3 } \coloneqq \begin{cases} \ 0 \quad w. p. \ 1 / 2 \\ - 1 \quad w. p. \ 1 / 2 \end{cases} \quad \xi _ { 4 } \coloneqq 0 \ w. p. \ 1.$$
Remark 1. Every time that a vertex x fires, we divide the possible types of the neighbors of x into four groups which determine the law of Z t .
$$Z _ { t } \stackrel { \text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text
$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text#\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{ $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text}$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text<$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text.$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text'$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text1$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text`$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text-$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$
$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\$$$
Definition 5. A firing rule is an inductive way to determine a legal firing sequence. More formally it is a function that is defined for any t and any atom A ⊂ G t such that P t > 0. The firing rule outputs an integer z such that there is at least one pair at z at time t in A .
The stacks combined with a firing rule determine the evolution of the process. Given a firing rule, we inductively define the set of stopping times t i . Set t 1 = 1.
Then, inductively, given t 1 , . . . , t i set t i +1 to be the first time j after t i such that Z j does not have distribution ξ 4 . Let t L be the last stopping defined. Thus glyph[negationslash]
$$L \coloneqq \# \{ i \leq \tau \ \colon \ Z _ { i } \stackrel { \text{law} } { \neq } \xi _ { 4 } \}.$$
Note that with this definition of the stopping times we have
$$P _ { t _ { i + 1 } } = P _ { t _ { i } + 1 }.$$
Thus the distribution of P t i +1 -P t i is either ξ , ξ 1 2 or ξ 3 . Finally, define the waiting times
$$\text{Wait} _ { i } = t _ { i + 1 } - t _ { i }.$$
With these definitions we are ready to complete the second and third steps of our outline.
Lemma 3.3. For any firing rule the sequence R i stochastically dominates P t i . This in particular implies that τ ′ stochastically dominates L.
Proof. The proof is by induction, and the starting configuration is given by R 0 = P 0 = n . If R i = n then the distribution of R i +1 -R i is ξ 3 , by definition of the reflected random walk.
Inductively, if P t i ≥ R i = n then P t i must also be n . This means that all sites have type 0 and the distribution of P t i +1 -P t i is ξ 3 . Thus we can couple R i +1 and P t i +1 such that R i +1 ≥ P t i +1 .
On the other hand, if 0 < R < n i then then the distribution of R i +1 -R i is ξ 1 while the distribution of P t i +1 -P t i is either ξ , 1 ξ 2 or ξ 3 . As all of ξ , 1 ξ 2 and ξ 3 are stochastically dominated by ξ 1 we can couple R i +1 and P t i +1 such that R i +1 ≥ P t i +1 .
Lemma 3.4. For any glyph[epsilon1] > 0 there exists δ such that
$$\mathbb { P } ( L > n ^ { 2 + \epsilon } ) \leq \mathbb { P } ( \tau ^ { \prime } > n ^ { 2 + \epsilon } ) < e ^ { - n ^ { \delta } }.$$
where τ ′ and L are defined in (14) and (18) respectively.
Proof. The first inequality follows from the fact that τ ′ stochastically dominates L. . The second inequality is a standard fact about lazy random walk (stated later in Lemma 5.1).
Lemma 3.5. Let E i be an i.i.d. sequence of random variables whose distribution is the same as the time taken by simple symmetric random walk started from the origin to hit ± 2 n . There exists a firing rule such that Wait i is stochastically dominated by E i .
glyph[negationslash]
Proof. At any stopping time t i , our firing rule is to pick the location of the left most pair. For times j = t i for any i , we define the firing rule inductively. If t i < j < t i +1 then at time j we choose to fire the pair at the location of the oil particle that just moved at step j -1. Thus we have an oil particle performing simple random walk until the distribution of Z j is not ξ 4 . We now show that this firing rule allows us to control the waiting times.
Let z be the location of the leftmost pair at time t i and let I = ( a, b ) be the interval of integers containing z where every location has at least one particle at time t i . As there are at most 2 n particles we have that I = ( a, b ) ⊂ ( z -2 n, z +2 ). n
If there are firings only in the interior of I then there are no particles at a or b . If the oil particle reaches the boundary of I at time k then the distribution of Z k is not ξ 4 , as one of the neighbors of the pair to be fired (either a or b ) has no particles and is of type 0. Thus t i +1 ≤ k . This tells us that the distribution of the waiting time Wait i = t i +1 -t i is bounded by the time taken by simple random walk started at the origin to leave the interval ( -2 n, 2 n ). Since this is true for each i independently, the lemma is proven.
We are now ready to prove Lemma 3.1.
Proof of Lemma 3.1. We first notice that by Lemma 2.2, τ defined in (11) is independent of the firing rule. For the purposes of the proof we will fix our firing rule to be the one mentioned in Lemma 3.5. Now by Lemmas 3.3 and 3.5
$$\sum _ { i = 1 } ^ { \tau ^ { \prime } } E _ { i } \, \text{ stochastically dominates } \tau.$$
Thus
$$\mathbb { P } ( \tau > n ^ { 4 + \epsilon } ) \leq \mathbb { P } ( \sum _ { i = 1 } ^ { \tau ^ { \prime } } E _ { i } \geq n ^ { 4 + \epsilon } )$$
If ∑ τ ′ i =1 E i ≥ n 4+ glyph[epsilon1] then either
$$\tau ^ { \prime } >. 1 n ^ { 2 + \epsilon / 2 } \quad \text{ or } \quad \sum _ { i = 1 } ^ {. 1 n ^ { 2 + \epsilon / 2 } } E _ { i } > n ^ { 4 + \epsilon / 2 }.$$
Therefore
$$\text{re} \\ \mathbb { P } ( \tau > n ^ { 4 + \epsilon } ) \leq \mathbb { P } ( \tau ^ { \prime } >. 1 n ^ { 2 + \epsilon / 2 } ) + \mathbb { P } \left ( \sum _ { i = 1 } ^ {. 1 n ^ { 2 + \epsilon / 2 } } E _ { i } > n ^ { 4 + \epsilon } \right )$$
The first term is bounded by Lemma 3.4. The second is bounded using standard bounds on the distribution of the time for simple random walk started at the origin to leave a fixed interval ( -k, k ). Both of these probabilities are bounded by e -n δ for some δ = ( ) δ glyph[epsilon1] > 0.
Furthermore, by Lemma 3.5 we have that the E i 's are all i.i.d., hence we can apply Wald's identity (see e.g., [8, Section 3.1]) and get that
$$\mathbb { E } ( \tau ) \leq \mathbb { E } ( \tau ^ { \prime } ) \mathbb { E } ( E _ { 1 } ) \leq 1 6 n ^ { 4 },$$
which concludes the proof.
## 4 Proof of Theorem 1.1 (Upper bound)
## 4.1 Notation
In order to proceed, we need to introduce some further notation. Let
$$\text{Height} = \max \{ u ( x ) \, \colon \, x \in \mathbb { Z } \}.$$
In this section we will bound from below the probability of the event
$$\mathcal { G } = \{ \text{Height} \leq C n ^ { 4 / 3 } \}.$$
for some large constant C . To this purpose, we define
$$\text{Returns} ( x ) \coloneqq \# \left \{ t \leq \tau \ \colon \ \eta _ { 1 } ( x, t ) = \eta _ { 2 } ( x, t ), \ x _ { t } \in \{ x - 1, x + 1 \} \right \}$$
Returns( x ) is the number of times such that
- 1. x has type 0, (the same number, possibly 0 of oil and water particles), and
- 2. a pair of particles is emitted from x -1 or x +1.
Then define
$$\text{Returns} = \sum _ { x \in \mathbb { Z } } \text{Returns} ( x ).$$
Recall the definition of N t i ( ) from Equation (5) and notice that
$$\text{Returns} & = N _ { 2 } ( \tau ) + 2 N _ { 3 } ( \tau ). & ( 2 1 )$$
Our goal is to show that Returns is large. The advantage of the decomposition in (20) is that it will allow us to show the relationship between Returns and the odometer function.
Finally, recall that G t is the σ -algebra generated by the movement of the first t pairs that are emitted and notice that, from the definition of P t (13) we have that
$$P _ { t } = \sum _ { x \in \mathbb { Z } } \min \left [ \eta _ { 1 } ( x, t ), \eta _ { 2 } ( x, t ) \right ]$$
is the number of pairs at the time of the t th emission. We recall that P t is a supermartingale with respect to G t .
✷
## 4.2 Outline
In order to show that it is unlikely that Height is much larger than n 4 3 / , we rely on three main ideas.
## (i) The odometer is fairly regular. Typically
$$| u ( x ) - u ( x + 1 ) | \leq 2 n + u ( x ) ^ { 1 / 2 }.$$
This regularity implies that under the assumption that u x ( 0 ) is much bigger than n 4 3 / then it is likely that u x ( ) is much bigger than n 4 3 / for all x such that | x -x 0 | < n 1 3 / .
- (ii) The variable Returns is comparable with the odometer function. Fix x ∈ Z and consider η 1 ( x, k ) -η 2 ( x, k ) as a function of k . This function performs a lazy random walk that takes about
$$\frac { 1 } { 2 } \Big ( u ( x - 1 ) + u ( x + 1 ) \Big ) \approx u ( x )$$
steps. Thus we expect Returns( x ) to be on the order of u x ( ) 1 2 / . Summing up over all x we expect
$$\Psi \quad \text{Returns} = \sum _ { x \in \mathbb { Z } } \text{Returns} ( x ) \approx \sum _ { x \in \mathbb { Z } } u ( x ) ^ { 1 / 2 }.$$
Combined with the previous paragraph this implies that if Height is much larger than n 4 3 / then it is likely that Returns is much larger than n.
- (iii) The process P t is a supermartingale. P t starts at n and stops when it hits 0. The sum of the negative drifts until the process terminates is, by Equation (21),
$$- \frac { 1 } { 4 } ( N _ { 2 } ( \tau ) + 2 N _ { 3 } ( \tau ) ) = - \frac { 1 } { 4 } \, \text{Returns} \,.$$
Hence, its stopping time is typically around 4 Returns. We use the Azuma-Hoeffding inequality to show that the probability of the event 'Returns is much larger than 4 n ' is decaying very rapidly. By the previous paragraphs we also will get that the probability that Height is much larger that n 4 3 / is decaying very rapidly.
The main difficulty in implementing this outline comes in the second step as the odometer function and Returns are correlated in a complicated way.
## 4.3 The bad events
To make our outline formal we now define our set of bad events. The first three of these deal with the odometer function. But first we use the odometer function to define
$$\text{Base} = \{ x \, \colon \, u ( x ) > 0 \} \subset \mathbb { Z }.$$
glyph[negationslash]
$$( 0 ) \ \mathcal { B } _ { 0 } = \left \{ \, \text{Base $\not\subset\,[-n^{5},n^{5}]$} \right \}$$
$$( \mathbf i ) \ \mathcal { B } _ { 1 } = \{ \text{Height} \geq n ^ { 4. 0 5 } \}.$$
- (ii) Gradient of the Odometer. In words, B 2 is the event where the gradient of the odometer function, u x ( ) -u x ( +1), is too large or has the wrong sign.
Let
$$m ( x ) = \begin{cases} \min \left \{ n, \min _ { 0 \leq y \leq x } u ( y ) \right \} & \text{for $x>0$ and} \\ \min \left \{ n, \min _ { x \leq y \leq 0 } u ( y ) \right \} & \text{for $x\leq 0$.} \end{cases}$$
We define:
$$\mathcal { B } _ { 2 } \colon = \left \{ \exists \, x \in [ - n ^ { 5 }, n ^ { 5 } ] \, \colon \, u ( x ) \geq n ^ {. 5 } \text{ and } | u ( x + 1 ) - u ( x ) | > 2 m ( x ) + u ( x ) ^ {. 5 1 } \right \} \\ \bigcup \left \{ \exists \, x \in [ 0, n ^ { 5 } ] \, \colon \, u ( x ) \geq n ^ { 5 } \text{ and } ( u ( x + 1 ) - u ( x ) ) > u ( x ) ^ {. 5 1 } \right \} \\ \bigcup \left \{ \exists x \in [ - n ^ { 5 }, 0 ] \, \colon \, u ( x ) \geq n ^ {. 5 } \text{ and } \, ( u ( x - 1 ) - u ( x ) ) > u ( x ) ^ {. 5 1 } \right \}.$$
Remark 2. While the definition of B 2 is quite technical we now give one consequence of it that is representative of how we will use it. Let B c 0 ∩ B c 2 occur and consider the set of x such that n < u x ( ) < n 1 96 . (or equivalently 1 < log n ( u x ( )) < 1 96). . Then
$$| u ( x ) - u ( x + 1 ) | < 2 m ( x ) + u ( x ) ^ {. 5 1 } \leq 2 n + ( n ^ { 1. 9 6 } ) ^ {. 5 1 } \leq 3 n. \text{ \quad \ \ } ( 2 3 )$$
For any x and y in a connected component of this set then (23) implies
$$\text{if } | \log _ { n } ( u ( x ) ) - \log _ { n } ( u ( y ) ) | \geq. 0 5 \quad \text{then } | x - y | > \frac { 1 } { 6 } n ^ {. 0 5 }.$$
This is because our conditions on x and y imply | u x ( ) -u y ( ) | ≥ n 1 05 . -n > . 5 n 1 05 . . So between x and y the odometer changes by at least . 5 n 1 05 . in increments of at most 3 n .
Our proof makes heavy use of this and similar estimates that follow from B c 0 ∩ B c 2 . Much of the complication of our proof comes from the fact that we need to use different estimates depending on whether m x ( ) = n or m x ( ) < n and whether log n ( u x ( )) is greater than 1.96, between 1 and 1.96 or less than 1. These estimates are used in Lemmas 4.5 and 4.6 which are used to prove Lemma 4.7. The conclusion of Lemma 4.7 is useful in showing that Returns is large because of our next bad event.
## (iii) Regularity of Returns
$$\mathcal { B } _ { 3 } = \left \{ \exists i, j, k \, \colon \, | i |, | j | \leq n ^ { 5 }, \ k \, \in [ n ^ { 5 }, n ^ { 5 } ], \ j - i >. 1 n ^ {. 0 1 } \text{ and } u ( x ) \geq k \, \forall x \in [ i, j ] \\ \text{ and } \sum _ { \text{ and } } ^ { j - 1 } \text{$r$aturm} / \varkappa \, \in 1 / i \, - i \, \mathcal { \lambda } / \mathcal { \lambda } \, \right \}$$
$$\text{and} \sum _ { x = i + 1 } ^ { j - 1 } \text{Returns} ( x ) <. 0 1 ( j - i ) \sqrt { k } \Big \}$$
Finally define the event
$$\mathcal { E } = \{ \, \text{Returns} \geq 2 0 \left ( n + \tau ^ {. 5 1 } \right ) \, \}.$$
To complete our outline we show that
- · G c ⊂ E ∪ ⋃ 3 i =0 B i and
- · P ( E ) and the P ( B i ) are small.
## 4.4 P t is a supermartingale.
Recall from (4) that
$$Z _ { k } \coloneqq P _ { k + 1 } - P _ { k }.$$
Recall from Remark 1 that conditional on the past, each Z k has one of four possible distributions ξ , . . . , ξ 1 4 .
Lemma 4.1. P t is a supermartingale and
$$\sum _ { t = 1 } ^ { \infty } \mathbb { E } \left ( Z _ { t + 1 } | \mathcal { F } _ { t } \right ) = - \frac { 1 } { 4 } \, \text{Returns} \,.$$
Proof. This follows from the discussion in Section 1.2. The summands on the left hand side are all 0, -1 4 or -1 2 . By (21) and (5) Returns is the sum of the number of -1 4 terms in the sum plus twice the number of -1 2 terms.
$$\text{Lemma } 4. 2. \text{ There exist } C, C ^ { \prime } \text{ such that } \mathbb { P } ( \mathcal { E } ) < C e ^ { - n ^ { C ^ { \prime } } }.$$
Proof. As Returns ≤ τ we have that
$$\mathcal { E } = \bigcup _ { r \geq 2 0 n } \mathcal { E } \cap \{ \tau = r \}.$$
Note that by Lemma 4.1 and because P τ = 0
$$\mathcal { E } \cap \left \{ \tau = r \right \} \subset \left \{ \sum _ { i = 1 } ^ { r } Z _ { t } - \mathbb { E } \left ( Z _ { t } | \mathcal { F } _ { t - 1 } \right ) \geq 4 n + 5 r ^ {. 5 1 } \right \}.$$
Also note that Z t -E ( Z t |F t -1 ) are the increments of a martingale and are bounded by 1. Thus we can apply the Azuma-Hoeffding inequality to get the bound P ( E∩{ τ = r } ) ≤ e -r . 02 . The lemma follows from the union bound.
## 4.5 Regularity of the odometer
Now define the new event
$$\text{Regular} = \mathcal { G } ^ { c } \cap \mathcal { B } _ { 0 } ^ { c } \cap \mathcal { B } _ { 1 } ^ { c } \cap \mathcal { B } _ { 2 } ^ { c },$$
where all the events are defined in Subsection 4.3. Remember the event G is that the odometer at the origin is less than a large constant times n 4 3 / and the events B c 0 , B c 1 and B c 2 are regularity conditions on the odometer.
## Lemma 4.3.
$$\text{Regular } \cap \mathcal { B } _ { 3 } ^ { c } \subset \mathcal { E }$$
where E was defined in (24) . Thus
$$\mathbb { P } ( \mathcal { G } ^ { c } ) \leq \mathbb { P } ( \mathcal { B } _ { 0 } ) + \mathbb { P } ( \mathcal { B } _ { 1 } ) + \mathbb { P } ( \mathcal { B } _ { 2 } ) + \mathbb { P } ( \mathcal { B } _ { 3 } ) + \mathbb { P } ( \mathcal { E } ).$$
Assuming (25), (26) follows by taking the union bound.
We split the proof of Lemma 4.3 into several lemmas. In particular, it will suffice to show that
$$\frac { 1 } { 2 } \text{ Returns } > \ 2 0 n$$
$$\frac { \bar { 1 } } { 2 } \, \text{Returns } \ > \ 2 0 \tau ^ {. 5 1 }.$$
Lemma 4.4. If the event Regular occurs, then
$$\tau \geq n ^ { 5 / 3 }$$
and if Regular ∩B c 3 occurs, then
$$\frac { 1 } { 2 } \, \text{Returns} > 2 0 n.$$
Recall that on the event Regular the height is much bigger than n 4 3 / but less than n 4 05 . , the odometer is supported inside [ -n , n 5 5 ], and gradient of the odometer is such that B c 2 occurs.
Proof. As usual, we choose to give a direct proof with explicit constants (which might be far from optimal), for sake of exposition.
If B c 0 occurs there exists
$$x _ { \text{Height} } \in [ - n ^ { 5 }, n ^ { 5 } ], \text{ such that } u ( x _ { \text{Height} } ) = \text{Height} \,.$$
The events G c and B c 2 imply that
$$u ( x ) > 7 0 0 0 0 0 n ^ { 4 / 3 } > ( 8 0 0 n ^ { 2 / 3 } ) ^ { 2 }$$
for all x in the interval
$$X _ { 0 } \coloneqq [ x _ { \text{Height} } - 1 0 0 n ^ { 1 / 3 }, x _ { \text{Height} } + 1 0 0 n ^ { 1 / 3 } ] \cap [ - n ^ { 5 }, n ^ { 5 } ] \cap \text{Base} \,.$$
This follows from the event B c 0 ∩ B c 2 by the same argument as in Remark 2. Then | X 0 | ≥ 100 n 1 3 / because by B c 0
$$x _ { \text{Height} } \in [ - n ^ { 5 }, n ^ { 5 } ]$$
so at least one of
$$[ x _ { \text{Height} }, x _ { \text{Height} } + 1 0 0 n ^ { 1 / 3 } ] \text{ or } [ x _ { \text{Height} }, x _ { \text{Height} } + 1 0 0 n ^ { 1 / 3 } ]$$
is entirely in [ -n , n 5 5 ].
Thus looking at the volume of the odometer in X 0 we have
$$\tau \geq \sum _ { x \in X _ { 0 } } u ( x ) \geq 7 0 0 0 0 n ^ { 4 / 3 } | X _ { 0 } | \geq n ^ { 5 / 3 }$$
and
$$\tilde { \Phi } \text{Returns} \geq \sum _ { x \in X _ { 0 } } \text{Returns} ( x ) \geq ( 1 0 0 n ^ { 1 / 3 } ) (. 0 1 ) ( 8 0 0 n ^ { 2 / 3 } ) \geq 4 0 n.$$
The next to last inequality is by (30), the size of X 0 and the definition of B c 3 . Thus we have obtained inequality (27).
## 4.6 Partitioning Base .
Inequality (28) is more involved to verify, and in order to do it we proceed as follows. We partition Base up into smaller intervals:
$$\text{Base} = \bigcup _ { i = - K ^ { \prime } } ^ { K } \text{Base} _ { i }$$
where, for each i , we set each Base i = [ a , b i i ] for some a i and b i which we describe in the following. We inductively define the Base i in a way so that on the event Regular and for most i we have
$$b _ { i } - a _ { i } >. 1 n ^ {. 0 1 }.$$
This will involve a series of estimates like in Remark 2.
Define h j = n 4 05 . -. 05 j for j = 0 to j = 68 and h 69 = 0. Let
$$\mathbf H = \{ h _ { j } \, \colon \, j = 1 \dots 6 8 \}.$$
Let j ′ be such that h j ′ is the closest value in H to u (0). Let glyph[negationslash]
$$b _ { 0 } = \inf \left \{ x \, \colon \, x \geq 0 \text{ and } u ( x + 1 ) \not \in \left ( h _ { j ^ { \prime } + 1 }, h _ { j ^ { \prime } - 1 } \right ) \right \}.$$
We say that Base 0 starts at height h j ′ and ends at height h j ′′ ∈ H where j ′′ is defined so that h j ′′ is the closest element of H to u b ( 0 +1) .
Now we inductively define Base i +1 . Suppose we have defined Base i = [ a , b i i ] which ends at height h k . We will inductively define Base i +1 = [ a i +1 , b i +1 ] . We let a i +1 = 1 + b i and say Base i +1 starts at height h k ∈ H. Then we define glyph[negationslash]
$$b _ { i + 1 } = \inf \left \{ x \, \colon \, x \geq a _ { i } \text{ and } u ( x + 1 ) \not \in \left ( h _ { k + 1 }, h _ { k - 1 } \right ) \right \}.$$
We say the block Base i +1 ends at height h k ′ ∈ H where h k ′ is the closest element of H to u b ( i +1) and h k ′ is the closest element of H to u b ( i +1) .
For the case of negative indices, the procedure is totally analogous. Finally, for all j ∈ { 0 , . . . , 68 } define
$$B _ { j } \coloneqq \text{the union of all } B e s _ { i } \text{ that start at } h _ { j } \text{ and have } | B e s _ { i } | >. 1 n ^ { 0 1 }. \quad ( 3 4 )$$
glyph[negationslash]
Lemma 4.5. If a block Base i with i ≥ 0 starts at height h j and ends at height h j ′ then j ′ = j . Conditional on Regular and j ′ < j < 68 then
$$b _ { i } - a _ { i } >. 1 n ^ {. 0 1 }.$$
Proof. The first statement is true because glyph[negationslash]
$$u ( 1 + b _ { i } ) \notin ( h _ { k + 1 }, h _ { k - 1 } )$$
so the closest element of H to u (1 + b i ) is not h k . Consider an interval Base i with i ≥ 0 where Base i starts at height h j and ends at height h j ′ with j ′ < j . Over the course of such an interval the odometer increased by at least . 49 h j -1 , going from less than . 5( h j + h j -1 ) to at least h j -1 . Since j < 68 we have
$$u ( x ) > h _ { j + 1 } \geq h _ { 6 8 } = n ^ {. 6 5 } > n ^ {. 5 }$$
for all x ∈ Base . i Thus the event Regular implies
$$u ( x + 1 ) - u ( x ) < u ( x ) ^ {. 5 1 } < ( h _ { j - 1 } ) ^ {. 5 1 }.$$
Thus by the choice of the h i these intervals must have width at least
$$b _ { i } - a _ { i } \geq \frac {. 4 9 h _ { j - 1 } } { ( h _ { j - 1 } ) ^ {. 5 1 } } \geq. 4 9 ( h _ { j - 1 } ) ^ {. 4 9 } \geq. 4 9 n ^ {. 3 } >. 1 n ^ {. 0 1 }.$$
The next to last inequality follows because j < 68 so h j -1 ≥ n . 7 .
Lemma 4.6. If Regular occurs and a block Base i with i > 0 starts at height h j and ends at height h j ′ with j ′ > j then
- 1. b i -a i > . 1 n . 01 or
- 2. 68 ≥ j ≥ 60 and for no k , 0 ≤ k < i the block Base k starts at h j and ends at a height h k ′ with k ′ > j .
Proof. Consider an interval Base i with i ≥ 0 where Base i starts at height h j and ends at height h j ′ with j ′ > j . Over the course of such an interval the odometer decreased by at least . 49 h j , going from at least . 5( h j + h j +1 ) to at most h j +1 .
First we consider the case that j < 60. We have
$$u ( x ) > h _ { j + 1 } \geq h _ { 6 0 } = n ^ { 1. 0 5 }$$
for all x ∈ Base . i Thus the event Regular implies
$$u ( x ) - u ( x + 1 ) < 2 m ( x ) + u ( x ) ^ {. 5 1 } = 2 n + u ( x ) ^ {. 5 1 } < 3 ( h _ { j } ) ^ {. 9 6 }.$$
Thus by the choice of the h i these intervals must have width at least
$$b _ { i } - a _ { i } \geq \frac {. 4 9 h _ { j } } { 3 ( h _ { j } ) ^ {. 9 6 } } \geq. 1 5 ( h _ { j } ) ^ {. 0 4 } >. 1 n ^ {. 0 1 }.$$
Next we consider the case that Base i ends at h j with 60 ≤ j ≤ 68 . Suppose there exists k < i such that Base k starts at h j and ends at a height h k ′ with k ′ > j . This implies that j < 68 and
$$m ( x ) \leq h _ { j + 1 } \leq ( h _ { j } ) n ^ { -. 0 5 }$$
for all x ∈ Base . i Thus the event Regular and j ≤ 67 implies u x ( ) > n . 65 and
$$u ( x ) - u ( x + 1 ) < 2 m ( x ) + u ( x ) ^ {. 5 1 } < 3 h _ { j } n ^ { -. 0 5 }.$$
Thus these intervals must have width at least
$$b _ { i } - a _ { i } \geq \frac {. 4 9 h _ { j } } { 3 ( h _ { j } ) n ^ { -. 0 5 } } \geq. 1 5 n ^ {. 0 5 } >. 1 n ^ {. 0 1 }.$$
## 4.7 Consequences of a regular gradient.
Recall the definition of B i from (34).
Lemma 4.7. Conditional on Regular there exists i ∈ { 1 , . . . , 68 } such that
$$\sum _ { x \in B _ { i } } u ( x ) \geq. 0 1 \tau.$$
Proof. Define
We first show that
$$\sum _ { x \in B ^ { * } } u ( x ) \leq 2 0 n ^ { 1. 2 }.$$
The lemma will follow easily from (29), (35) and the pigeonhole principle.
Let I = { i : | Base i | ≤ . 1 n . 01 } . First we show that | I | ≤ 20 . By Lemmas 4.5 and 4.6 for every i ∈ I there exists j , 60 ≤ j ≤ 68 such that Base i starts at height h j . From this we draw two conclusions. First by the definitions of B ∗ and the Base i we have u x ( ) < n 1 15 . on B ∗ . Also for each such j Lemma 4.6 implies there exist at most two i ∈ I with Base i starting at height h j , at most one with i ≥ 0 and at most one with i ≤ 0. These two facts combine to establish (35) which completes the proof.
Remark 3. If we perform the analysis in the previous lemma to non-empty intervals Base i that start at h j ≥ n 1 4 . we get that | Base i | ≥ . 1 n . 4 . This implies
$$\tau \geq. 1 n ^ {. 4 } h _ { j + 1 } =. 1 n ^ {. 3 } h _ { j - 1 }.$$
glyph[negationslash]
Lemma 4.8. If B j = ∅ and Regular occurs, then
$$\frac { \tau ^ {. 4 9 } } { ( h _ { j - 1 } ) ^ {. 5 } } \geq n ^ {. 1 }.$$
Proof. If h j ≤ n 1 35 . then the result follows from the first part of Lemma 4.4. If h j ≥ n 1 4 . then it follows from the previous remark.
## 4.8 Consequences of Regular Returns.
Recall the sequence h i defined in Subsection 4.6.
Lemma 4.9. For every j ∈ { 1 , . . . , 68 } the set B j satisfies
$$\sum _ { x \in B _ { j } } u ( x ) \leq | B _ { j } | h _ { j - 1 }.$$
Conditional on Regular ∩B c 3 for every j ∈ { 0 , . . . , 67 } the set B j satisfies
$$\sum _ { x \in B _ { j } } \text{Returns} ( x ) \geq. 0 1 | B _ { j } | ( h _ { j + 1 } ) ^ { 1 / 2 }.$$
$$B ^ { * } = \bigcup _ { i \colon | \text{Base} _ { i } | \leq. 1 n ^ {. 0 1 } } \text{Base} _ { i } \,.$$
Proof. From the choice of the intervals Base j and B i we have that
$$h _ { i + 1 } \leq u ( x ) \leq h _ { i - 1 },$$
for all x ∈ B i . Also, by definition, B i consists of a union of intervals of width at least . 1 n . 01 (cf. Equation (34)). Therefore the second statement follows from the definition of B c 3 .
Lemma 4.10. For all sufficiently large n conditional on Regular ∩B c 3 we have
$$\frac { 1 } { 2 } \text{ Returns} > 2 0 \tau ^ {. 5 1 }.$$
Proof. As Regular occurs, by Lemma 4.7 we obtain that there exists a j such that
$$\sum _ { x \in B _ { j } } u ( x ) \geq. 0 1 \tau.$$
First, consider the case that j ∈ { 1 , . . . , 67 } . In this case, from Lemma 4.9 it follows that
$$| B _ { j } | h _ { j - 1 } \geq \sum _ { x \in B _ { j } } u ( x ) >. 0 1 \tau.$$
Conditional on the event B c 3 , the set B j satisfies ∑ x ∈ B j Returns( x ) ≥ . 01 | B j | ( h j +1 ) 1 2 / , which implies
$$\sum _ { x \in B _ { j } } \text{Returns} ( x ) \quad & \geq \quad. 0 1 | B _ { j } | ( h _ { j + 1 } ) ^ { 1 / 2 } \\ & \stackrel { ( 3 6 ) } { > } \quad 0 1 \Big ( \underline {. 0 1 \tau } \Big ) ( h _ { \dots } ) ^ { 1 / 2 }$$
$$& \overset { ( 3 6 ) } { \geq } \quad. 0 1 \left ( \frac {. 0 1 \tau } { h _ { j - 1 } } \right ) ( h _ { j + 1 } ) ^ { 1 / 2 } \\ & = \quad. 0 0 0 1 \left ( \frac { h _ { j + 1 } } { h _ { j - 1 } } \right ) ^ { 1 / 2 } \frac { \tau ^ {. 4 9 } } { ( h _ { j - 1 } ) ^ {. 5 } } \quad \tau ^ {. 5 1 } \\ & \geq \quad. 0 0 0 1 n ^ { -. 0 5 } n ^ {. 1 } \tau ^ {. 5 1 } \\ & \geq \quad 4 0 \tau ^ {. 5 1 }. \\ & \quad. \wedge \. \quad.. \quad.. \. \. \. \. \. \. \.$$
The next to last line follows from Lemma 4.8, together with the definition of h j +1 and h j -1 , whereas the last inequality holds whenever n is sufficiently large.
Now consider j = 68. In this case, by Lemma 4.4 we have τ > n 5 3 / so τ . 49 > n . 8 and
$$\sum _ { x \in B _ { 6 8 } } \text{Returns} ( x ) \geq | B _ { 6 8 } | \geq \frac {. 0 1 \tau } { n ^ {. 7 } } = \frac {. 0 1 \tau ^ {. 5 1 } \tau ^ {. 4 9 } } { n ^ {. 7 } } \geq 4 0 \tau ^ {. 5 1 },$$
which ends the proof.
Proof of Lemma 4.3. The proof is an easy consequence of Lemmas 4.4 and 4.10. ✷
## 4.9 The probability of the bad events
Finally, we can use our results to bound the probability of G . In Lemma 4.3 we showed that the event G c is contained in a certain union of the bad events and then used union bound to bound its probability. We now bound the probability of those bad events.
Lemma 4.11. There exist positive constants C, C ′ and γ such that
$$\mathbb { P } ( \mathcal { B } _ { 2 } ) < C e ^ { - C ^ { \prime } n ^ { \gamma } }.$$
We first introduce a definition.
Definition 6. Let ∆ ( x k ) denote the quantity such that, after k pairs have been emitted from x , there are k +∆ ( ) particles that moved to the right (i.e. to x k x +1) and k -∆ ( ) x k particles that have moved to the left (i.e. to x -1). Notice that this is just a function of the stack of variables X ,Y x i x i at the site x .
Proof of Lemma 4.11. Without loss of generality, we assume that x ≥ 0. Recall from Subsection 4.3 (ii) that
$$m ( x ) = \min \left \{ n, \min _ { 0 \leq y \leq x } u ( y ) \right \}.$$
Suppose that at some time t exactly k pairs have been emitted from x , and exactly k ′ pairs have been emitted from x +1. Then, the number of particles to the right of x is
$$k + \Delta ^ { x } ( k ) - ( k ^ { \prime } - \Delta ^ { x + 1 } ( k ^ { \prime } ) )$$
which is between 0 and 2 m x ( ). This holds for all times t , in particular it holds for the time when the process stops. Now set t to be such that k = u x ( ) and k ′ = u x ( +1) . Let ∆ = ∆ ( ( )) and ∆ x x u x x +1 = ∆ x +1 ( u x ( +1)) . Then
$$0 \leq u ( x ) + \Delta ^ { x } - ( u ( x + 1 ) - \Delta ^ { x + 1 } ) \leq 2 m ( x ).$$
Rearranging, we get
$$- \Delta ^ { x } - \Delta ^ { x + 1 } \leq u ( x ) - u ( x + 1 ) \leq 2 m ( x ) - \Delta ^ { x } - \Delta ^ { x + 1 }.$$
Then the event B 2 implies that
$$\exists x \in [ - n ^ { 5 }, n ^ { 5 } ] \text{ and } k, \ n ^ {. 5 } \leq k \leq n ^ { 4. 1 } \colon \, | \Delta ^ { x } ( k ) | >. 5 k ^ {. 5 1 }.$$
The result follows from standard concentration results of random walks (cf. A.1) and union bounding over all possible values of x and k . Thus the result holds for some appropriate C, C ′ and γ . We omit the details.
Lemma 4.12. There exist positive constants C, C ′ and γ such that
$$\mathbb { P } ( \mathcal { B } _ { 3 } ) < C e ^ { - C ^ { \prime } n ^ { \gamma } }.$$
Before proving this result, we need to show a few preliminary results. Furthermore, in this context we work with Version 2 of the model, introduced in Section 2.3. Lemma 2.4 allows us to switch between events defined in one version to the other. For x ∈ 3 Z let
$$W ^ { x } ( \ell ) = \frac { \bar { X } _ { \ell } ^ { x } - \bar { Y } _ { \ell } ^ { x } } { 2 }$$
where ¯ X ,Y x glyph[lscript] ¯ x glyph[lscript] are defined in Section 2.3 (12). This represents the change in the difference of oil and water particles at x when the glyph[lscript] th firing takes place from the set { x -1 , x +1 } . Clearly W x ( glyph[lscript] ) has the same distribution as one step of a symmetric lazy random walk. Now define
$$\tilde { R } ^ { x } ( k ) = \# \left \{ 0 \leq j \leq k \colon \sum _ { \ell = 1 } ^ { j } W ^ { x } ( \ell ) = 0 \right \}.$$
Define B ′ 3 to be the event that there exist three integers i, j, k such that
- (i) | i , | | j | ≤ n 5 ,
- (ii) j -i ≥ . 1 n . 01 ,
- (iii) k ∈ [ n . 5 , n 5 ] and
$$( \begin{matrix} \nu \\ \nu \\ \nu \\ ( \text{iv} ) \end{matrix} ) \sum _ { x \in ( i, j ), 3 | x } ^ { \nu } \tilde { R } ^ { x } (. 9 k ) <. 0 1 \sqrt { k } ( j - i ).$$
We now state a standard fact about number of returns to origin for the simple random walk on Z .
Lemma 4.13. Let { S i } i ≥ 0 be a lazy simple random walk on Z started at the origin. Let
$$Z \text{eros} ( l ) = \# \{ i \, \colon \, 0 \leq i \leq l \ a n d \ S _ { i } = 0 \}.$$
Then for all l
$$\mathbb { P } \left ( Z \text{eros} (. 9 l ) >. 1 \sqrt { l } \right ) \geq \frac { 1 } { 2 }.$$
Proof. See Chapter III, Section 5 of [9].
Lemma 4.14. There exist positive constants C, C ′ and γ such that
$$\mathbb { P } ( \mathcal { B } _ { 3 } ^ { \prime } ) < C e ^ { - C ^ { \prime } n ^ { \gamma } }.$$
Proof. Fix i, j, k and x ∈ ( i, j ). Then, from Lemma 4.13 it follows
$$\mathbb { P } \left ( \tilde { R } ^ { x } (. 9 k ) >. 1 \sqrt { k } \right ) \geq \frac { 1 } { 2 }.$$
Therefore we get
$$\text{core we ge} \\ \mathbb { P } & \left ( \sum _ { x \in ( i, j ), 3 | x } \tilde { R } ^ { x } (. 9 k ) <. 0 1 \sqrt { k } ( j - i ) \right ) \\ & < \mathbb { P } \left ( \# \left \{ x \in ( i, j ), 3 | x \ \colon \ \tilde { R } ^ { x } (. 9 k ) >. 1 \sqrt { k } \right \} < \frac { 1 } { 1 0 } ( j - i ) \right ) \\ & < c e ^ { - c ^ { \prime } n ^ {. 1 } }, \\ \prime & \quad \dots \quad \dots \quad \dots$$
where c, c ′ are positive constants.
As there are at most 4 n 15 choices of i, j and k there exists positive constants c, c ′ and γ ′ such that
$$\mathbb { P } ( \mathcal { B } _ { 3 } ^ { \prime } ) < 4 n ^ { 1 5 } c e ^ { - c ^ { \prime } n ^ { \gamma ^ { \prime } } }$$
so the lemma is true for some choice of C, C ′ and γ .
Proof of Lemma 4.12. Consider the map φ defined in Lemma 2.4. Consider the event
$$\phi ^ { - 1 } ( \mathcal { B } _ { 3 } ).$$
By the measure preserving property of φ
$$\mathbb { P } ( \mathcal { B } _ { 3 } ) = \mathbb { P } ( \phi ^ { - 1 } ( \mathcal { B } _ { 3 } ) )$$
where the two probabilities are in the two different probability spaces mentioned in the statement of Lemma 2.4. Note that the event φ -1 ( B 3 ) implies B ′ 3 . This is because (iv) in the definition of B ′ 3 says √
$$\sum _ { x \in ( i, j ), 3 | x } \tilde { R } ^ { x } (. 9 k ) <. 0 1 \sqrt { k } ( j - i )$$
whereas in the definition of B 3 we have
$$\sum _ { x \in ( i, j ) } \tilde { R } ^ { x } (. 9 k ) <. 0 1 \sqrt { k } ( j - i ).$$
The proof now follows from the above lemma.
Lemma 4.15. There exist positive constants D,C ′ and γ such that
$$\mathbb { P } ( \mathcal { B } _ { 0 } ), \mathbb { P } ( \mathcal { B } _ { 1 } ), \mathbb { P } ( \mathcal { B } _ { 2 } ), \mathbb { P } ( \mathcal { B } _ { 3 } ), \mathbb { P } ( \mathcal { E } ) < D e ^ { - C ^ { \prime } n ^ { \gamma } }.$$
Proof. P ( B 0 ) and P ( B 1 ) are bounded by Lemma 3.1. P ( B 2 ) is bounded by Lemma 4.11. P ( B 3 ) is bounded by Lemma 4.12. P ( E ) is bounded by Lemma 4.2
## 4.10 Proof of the upper bound
Thus by (26) and lemma 4.15, i.e there exists glyph[epsilon1] > 0 such that for large enough n
$$\mathbb { P } ( \mathcal { G } ) > 1 - e ^ { - n ^ { \epsilon } }$$
where G was defined in (19). The proof of the upper bound is hence complete.
## 5 Proof of Theorem 1.1 (Lower bound)
The goal of this section is to prove the lower bound in Theorem 1.1. Theorem 1.2 will follow from the proof of the lower bound. Before proceeding to the proofs we state a few standard results about R i , the lazy simple symmetric random walk on Z whose increments R i +1 -R i are ± 1 with probability 1 4 each, and 0 with probability 1 2 . Let
$$M ( t ) = \sup _ { i < t } | R _ { i } |.$$
Lemma 5.1. (i) Given glyph[epsilon1] > , 0 there exists c = ( ) c glyph[epsilon1] > 0 such that
$$\mathbb { P } \left ( M ( t ) > t ^ { 1 / 2 + \epsilon } \right ) < e ^ { - t ^ { c } }.$$
✷
(ii) Given glyph[epsilon1] > 0 , there exists δ = ( ) δ glyph[epsilon1] > 0 , such that ,
$$\mathbb { P } ( M ( t ^ { 2 + \epsilon } ) < 2 t ) < e ^ { - t ^ { \delta } }.$$
$$\mathbb { E } ( M ( t ) ) = \Theta ( \sqrt { t } ).$$
$$\mathbb { E } ( M ( t ) ^ { 2 } ) = \Theta ( t ).$$
$$\lim _ { t \rightarrow \infty } t ^ { - 1 / 2 } \mathbb { E } | R _ { t } | = \sqrt { \frac { 1 } { \pi } }.$$
Proof. For proofs of parts (i)-(iv) we refer the reader to [15, Sections 21, 23].
By the central limit theorem, t -1 2 / R t d - → Z where Z ∼ N (0 , 1 2 ), hence E | Z | = √ 1 /π . Part (v) follows since the random variables t -1 2 / R t are uniformly integrable; see, for instance, [5, Theorem 3.5].
## Lemma 5.2. Let
$$\left \{ M ^ { i } ( n ^ { \frac { 4 } { 3 } } ) \right \} _ { i = 1 } ^ { n ^ { 1 / 3 } }$$
be a sequence of i.i.d. random variables with the same law as M n ( 4 3 ) . Then there exist positive constants D,γ such that
$$\mathbb { P } \left ( \sum _ { i = 1 } ^ { n ^ { 1 / 3 } } M ^ { i } \left ( n ^ { \frac { 4 } { 3 } } \right ) < D n \right ) > 1 - e ^ { - n ^ { \gamma } }.$$
Proof of Lemma 5.2 is deferred to Appendix A. For x ∈ Z and a positive integer i we define the variable
$$D ^ { x } ( i ) = \mathbf 1 _ { ( X _ { i } ^ { x } = 1 ) } - \mathbf 1 _ { ( Y _ { i } ^ { x } = 1 ) }$$
where the variables X ,Y x i x i appear in (6) in Section 2. Clearly
$$D ^ { x } ( i ) = \begin{cases} \begin{array} { c c } - 1 & \text{w.p. } 1 / 4 \\ 0 & \text{w.p. } 1 / 2 \\ 1 & \text{w.p. } 1 / 4. \end{array} \end{cases}$$
Let C be the number appearing in Theorem 1.1. For any x ∈ Z let
$$S _ { x } = \sup _ { i \leq C n ^ { 4 / 3 } } \left | \sum _ { j = 0 } ^ { i } D ^ { x } ( j ) \right |$$
$$O _ { x } = \sup _ { i \leq C n ^ { 4 / 3 } } \left | \sum _ { j = 0 } ^ { i } \frac { X _ { j } ^ { x } } { 2 } \right |$$
$$W _ { x } = \sup _ { i \leq C n ^ { 4 / 3 } } \Big | \sum _ { j = 0 } ^ { i } \frac { Y _ { j } ^ { x } } { 2 } \Big |$$
We now discuss briefly how the proof proceeds. We first establish the following gradient bound on the odometer using (37): there exists a constant c such that, with high probability, for x ∈ [0 , cn 1 3 / ] we have
$$u ( x ) - u ( x + 1 ) \geq n / 2.$$
This in turn implies the theorem since
$$0 \leq u ( \lfloor c n ^ { 1 / 3 } \rfloor ) \leq u ( 0 ) - \frac { n } { 2 } \lfloor c n ^ { 1 / 3 } \rfloor.$$
## 5.1 Lower bound.
Recall (16) and that τ is the stopping time defined in Definition 11. Thus the total number of particles at a site x at the end of the process is | g τ ( x ) | . Also define
$$D _ { x, x + 1 } & = \# \{ \text{ particles moving from } x \text{ to } x + 1 \}, \\ D _ { x + 1, x } & = \# \{ \text{ particles moving from } x + 1 \text{ to } x \}.$$
Remark 4. For x > 0 the difference D x,x +1 -D x +1 ,x is the total number of particles that stop to the right of x .
Since one side of the origin has at least n/ 2 particles at the end of the process without loss of generality we assume that
$$\sum _ { i = 1 } ^ { \infty } | g _ { \tau } ( i ) | \geq n / 2.$$
Also by definition (see (40)) for any x ∈ Z ,
$$g _ { \tau } ( x ) = \sum _ { i = 1 } ^ { u ( x - 1 ) } D ^ { x - 1 } ( i ) - \sum _ { i = 1 } ^ { u ( x + 1 ) } D ^ { x + 1 } ( i ).$$
Recall G defined in (19). Also given glyph[epsilon1] > 0 define the following events
$$\mathcal { G } _ { 1 } \ = \ \sum _ { x = - \epsilon n ^ { 1 / 3 } } ^ { \epsilon n ^ { 1 / 3 } } | S _ { x } | \leq n / 1 2 \\ \dots$$
$$\mathcal { G } _ { 2 } \ = \ 2 \sum _ { x = - \epsilon n ^ { 1 / 3 } } ^ { \epsilon n ^ { 1 / 3 } } ( O _ { x } + W _ { x } ) \leq n / 6.$$
where S , O x x , W x are defined in (41),(42) and (43). Note that we suppress the dependence on glyph[epsilon1], n in the notations for brevity.
Lemma 5.3. For small enough glyph[epsilon1] there exists c > 0 such that
$$\mathbb { P } ( \mathcal { G } \cap \mathcal { G } _ { 1 } \cap \mathcal { G } _ { 2 } ) \geq 1 - e ^ { - n ^ { c } }.$$
Proof. Follows from (37) and Lemma 5.2.
We now state the following lemma establishing a lower bound on the gradient of the odometer function.
Lemma 5.4. Assume (45) . Then there exists a constant glyph[epsilon1] such that
$$u ( j ) - u ( k ) \geq ( k - j ) \frac { n } { 3 } - \frac { n } { 6 }$$
for all 0 ≤ j ≤ k ≤ glyph[epsilon1]n 1 3 / with failure probability at most e -n c for some positive constant c.
Proof. Recalling Definition 6
$$D _ { x, x + 1 } = u ( x ) + \Delta ^ { x } ( u ( x ) ).$$
Using Remark 4 we have
$$u ( 0 ) + \Delta ^ { 0 } ( u ( 0 ) ) - u ( 1 ) + \Delta ^ { 1 } ( u ( 1 ) ) \ = \ \sum _ { y = 1 } ^ { \infty } | g _ { \tau } ( y ) |.$$
$$u ( 1 ) + \Delta ^ { 1 } ( u ( 1 ) ) - u ( 2 ) + \Delta ^ { 2 } ( u ( 2 ) ) \ = \ \sum _ { y = 2 } ^ { \sigma ^ { \mu } } | g _ { \tau } ( y ) |.$$
.
.
.
$$u ( m - 1 ) + \Delta ^ { m - 1 } ( u ( m - 1 ) ) - u ( m ) + \Delta ^ { m } ( u ( m ) ) \ = \ \sum _ { y = m } ^ { \infty } | g _ { \tau } ( y ) |.$$
.
.
.
Adding the above from any 0 < glyph[lscript] < m to m we get
$$u ( \ell ) - u ( m ) + \Delta ^ { \ell } ( u ( \ell ) ) + 2 \sum _ { i = \ell + 1 } ^ { m - 1 } \Delta ^ { i } ( u ( i ) ) + \Delta ^ { m } ( u ( m ) ) = \sum _ { i = \ell + 1 } ^ { m } \sum _ { y = i } ^ { \infty } | g _ { \tau } ( y ) |. \quad ( 5 0 )$$
We claim that for any x ∈ Z
$$| \Delta ^ { x } ( u ( x ) ) | \mathbf 1 ( \mathcal { G } ) \leq O _ { x } + W _ { x }$$
where O ,W x x are defined in (42) and (43). This follows from definitions and the observation that
$$\Delta ^ { x } ( u ( x ) ) \ & = \ \sum _ { i = 1 } ^ { u ( x ) } [ 1 ( X _ { i } ^ { x } = 1 ) - 1 / 2 ] + \sum _ { i = 1 } ^ { u ( x ) } [ 1 ( Y _ { i } ^ { x } = 1 ) - 1 / 2 ] \\ & = \ \sum _ { i = 1 } ^ { u ( x ) } \frac { X _ { i } ^ { x } } { 2 } + \sum _ { i = 1 } ^ { u ( x ) } \frac { Y _ { i } ^ { x } } { 2 }. \\ \quad \vdots \quad \nu \cap$$
Thus by (50)
$$[ u ( \ell ) - u ( m ) ] { 1 } ( \mathcal { G } ) \geq \sum _ { i = \ell + 1 } ^ { m } \sum _ { y = i } ^ { \infty } | g _ { \tau } ( y ) | { 1 } ( \mathcal { G } ) - 2 \left [ \sum _ { x = \ell } ^ { m } ( O _ { x } + W _ { x } ) \right ] { 1 } ( \mathcal { G } ).$$
Now on the event G ∩ G 1 ∩ G 2 , ∀ j < glyph[epsilon1]n 1 3 /
$$\sum _ { i = j } ^ { \infty } | g _ { \tau } ( i ) | \geq n / 3.$$
To show this we upper bound ∑ j i =1 | g τ ( ) . i | By (46) on the event G we have for all 0 < i ≤ glyph[epsilon1]n 1 3 / ,
$$| g _ { \tau } ( i ) | \leq S _ { i - 1 } + S _ { i + 1 }$$
where S i is defined in (41). Hence on the event G ∩ G 1 ∩ G 2
$$\sum _ { i = 1 } ^ { j } | g _ { \tau } ( i ) | \leq 2 \sum _ { i = 1 } ^ { \epsilon n ^ { 1 / 3 } } S _ { i } \leq \frac { n } { 6 }.$$
Thus by (45) we have for all 0 < j < glyph[epsilon1]n 1 3 / ,
$$\sum _ { i = j } ^ { \infty } | g _ { \tau } ( i ) | \geq \frac { n } { 2 } - \sum _ { i = 1 } ^ { j } | g _ { \tau } ( i ) | \geq \frac { n } { 2 } - \frac { n } { 6 } = \frac { n } { 3 }.$$
Therefore by (51) and (52), on the event G ∩ G 1 ∩ G 2 we have for all 0 ≤ j ≤ k ≤ glyph[epsilon1]n 1 3 /
$$u ( j ) - u ( k ) \ \geq \ \frac { 1 } { 3 } ( k - j ) n - 2 \left [ \sum _ { x = 1 } ^ { k } O _ { x } + W _ { x } \right ]$$
$$\geq \ \frac { 1 } { 3 } ( k - j ) n - \frac { n } { 6 }.$$
Thus we are done.
The proof of the lower bound is now a corollary. Since u k ( ) ≥ 0, for all large enough n and j ≤ glyph[epsilon1]n 1 / 3 2 the above implies that
$$u ( j ) \geq \frac { 1 } { 8 } \epsilon n ^ { 4 / 3 }.$$
Hence by (49) it follows
$$\mathbb { P } \left ( \inf _ { 0 \leq j \leq \frac { 1 } { 2 } \epsilon n ^ { 1 / 3 } } u ( j ) \leq \frac { 1 } { 8 } \epsilon n ^ { 4 / 3 } \right ) \leq e ^ { - n ^ { c } }.$$
To complete the proof we use the symmetric version of (50) to get the following bound: For j ≥ 0
$$u ( 0 ) - u ( - j ) - \Delta ^ { 0 } ( u ( 0 ) ) - 2 \sum _ { i = - j + 1 } ^ { - 1 } \Delta ^ { i } ( u ( i ) ) - \Delta ^ { - j } ( u ( - j ) ) = \sum _ { i = - j } ^ { - 1 } \sum _ { y = - \infty } ^ { i } | g _ { \tau } ( y ) |$$
The following bound
$$\sum _ { y = - \infty } ^ { i } | g _ { \tau } ( y ) | \leq 2 n$$
is trivial since the total number of particles is 2 n. Using the above and definition of G 2 we get for all 0 < j < glyph[epsilon1]n 1 3 /
$$u ( 0 ) - u ( - j ) \leq 2 j n + \frac { n } { 6 }.$$
Thus u (0) ≥ 1 4 glyph[epsilon1]n 4 3 / implies that for all j ≤ 1 16 glyph[epsilon1]n 1 3 /
$$u ( - j ) \geq \frac { 1 } { 9 } \epsilon n ^ { 4 / 3 }.$$
(56) and (57) now completes the proof.
Lemma 5.5. There exists a constant C > 0 such that
$$\sup _ { x \in \mathbb { Z } } \mathbb { E } ( | g _ { \tau } ( x ) | ) < C n ^ { \frac { 2 } { 3 } }.$$
Moreover for any glyph[epsilon1] > 0 there exists c > 0
$$\mathbb { P } \left [ \sup _ { x \in \mathbb { Z } } | g _ { \tau } ( x ) | \geq n ^ { \frac { 2 } { 3 } + \epsilon } \right ] \leq e ^ { - n ^ { c } }.$$
Proof. For any x ∈ Z , we have
$$| g _ { \tau } ( x ) | \leq [ S _ { x - 1 } + S _ { x + 1 } ] \mathbf 1 ( \mathcal { G } ) + 2 n \mathbf 1 ( \mathcal { G } ^ { c } )$$
where the event G is defined in (19). The first term follows from (53) and the second term is obvious since the total number of particles is 2 n. The proof of (58) now follows from ( iii ) of Lemma 5.1 and Theorem (37). Additionally using ( ) of Lemma 5.1 we get that for any i x ∈ Z there exists c > 0 such that
$$\mathbb { P } ( | g _ { \tau } ( x ) | \geq n ^ { \frac { 2 } { 3 } + \epsilon } ) \leq e ^ { - n ^ { c } }.$$
Corollary 3.2 says that with probability at least 1 -e -n c for all | x | ≥ n 5
$$| g _ { \tau } ( x ) | = 0.$$
(59) now follows from (61) by union bound over all x ∈ -[ n , n 5 5 ] .
Remark 5. For small enough glyph[epsilon1] and k = glyph[epsilon1]n 1 3 / , by (49) and (60)
$$\mathbb { P } \left [ \sum _ { i = - k } ^ { k } | g _ { \tau } ( i ) | \leq n \right ] \geq 1 - e ^ { - n ^ { c } }$$
for some positive constant c. Thus at least n particles are supported outside the interval [ -glyph[epsilon1]n 1 3 / , glyph[epsilon1]n 1 3 / ] with failure probability at least e -n c .
## 5.2 Proof of Theorem 1.2
Let x = n 1 3+ / glyph[epsilon1] . Under the assumption that there are at least n 1 -glyph[epsilon1]/ 2 many particles to the right of x , for all glyph[lscript] ≤ x,
$$\sum _ { i = \ell } ^ { \infty } | g _ { \tau } ( i ) | \geq n ^ { 1 - \epsilon / 2 }.$$
Recalling (51) we have
$$( u ( 0 ) - u ( x ) ) \mathbf 1 ( \mathcal { G } ) \geq \sum _ { i = 1 } ^ { x } \sum _ { y = i } ^ { \infty } | g _ { \tau } ( y ) | \mathbf 1 ( \mathcal { G } ) - 2 \left [ \sum _ { i = 1 } ^ { x } ( O _ { i } + W _ { i } ) \right ] \mathbf 1 ( \mathcal { G } ).$$
Now by part ( ) of Lemma 5.1 and union bound over 1 i ≤ glyph[lscript] ≤ x there exists a c > 0 such that with probability at least 1 -e -n c ,
$$\sum _ { \ell = 1 } ^ { x } [ O _ { \ell } + W _ { \ell } ] = O ( n ^ { 1 + 2 \epsilon } ).$$
Thus on the event that there are at least n 1 -glyph[epsilon1]/ 2 many particles to the right of x we have
$$u ( 0 ) - u ( x ) \geq x n ^ { 1 - \epsilon / 2 } - O ( n ^ { 1 + \epsilon } ),$$
except on a set of measure at most e -n c . However this implies that
$$u ( 0 ) \geq n ^ { 4 / 3 + \epsilon / 2 } - O ( n ^ { 1 + \epsilon } ).$$
Hence by the upper bound in Theorem 1.1 we conclude that the probability of the event that there are at least n 1 -glyph[epsilon1]/ 2 many particles to the right of x = n 1 3+ / glyph[epsilon1] is less than e -n c for some positive c > 0. The argument for x = -n 1 3+ / glyph[epsilon1] is symmetric and we omit the details. Thus we are done.
## 6 Scaling limit for the odometer
The goal of this section is to prove Theorem 1.4. The first step will be to show that Conjecture 1.3 implies some regularity of the limiting function: we will argue that w ( ) · is decreasing and three times differentiable on the positive real axis. Moreover it is the solution of the boundary value problem
$$w ^ { \prime \prime } & = \quad \sqrt { \frac { 2 } { \pi } w } \\ \lim _ { h \to 0 ^ { + } } \frac { w ( h ) - w ( 0 ) } { h } & = \quad - 1 \\ \lim _ { h \to \infty } w ( h ) & = \quad 0. \\ \cdots \cdots \cdots \cdots \cdots \cdots \cdots$$
At this point, Theorem 1.4 follows by identifying an explicit solution to the above problem and arguing that it is the unique solution.
## 6.1 Properties of the expected odometer
We first make some easy observations about the expected odometer function, denoted by
$$\tilde { u } ( x ) \coloneqq \mathbb { E } ( u ( x ) ).$$
Existence of ˜( u x ) follows from ( ii ) of Lemma 3.1 which says that the stopping time of the process τ has finite expectation and clearly for all x ∈ Z
$$u ( x ) \leq \tau.$$
Recall the notation D x,x +1 from (44).
Lemma 6.1. For x ∈ Z
$$\tilde { u } ( x ) = \mathbb { E } ( D _ { x, x + 1 } ) = \mathbb { E } ( D _ { x, x - 1 } ).$$
$$\tilde { u } ( x ) - \tilde { u } ( x + 1 ) \leq n \ \text{ for } x \geq 0$$
$$\tilde { u } ( x ) - \tilde { u } ( x - 1 ) \leq n \ \ f o r \ x \leq 0.$$
Moreover we have,
Clearly every time there is an emission at a site x ∈ Z , on average one particle moves to x -1 and another to x +1. Hence, informally for every x ∈ Z , ˜( u x ) is the expected number of particles emitted from x that go to x -1 or x +1.
Proof. Wlog assume x ≥ 0. By definition we have
$$D _ { x, x + 1 } = \sum _ { k = 1 } ^ { \infty } ( \mathbf 1 ( X _ { k } ^ { x } = 1 ) + \mathbf 1 ( Y _ { k } ^ { x } = 1 ) ) \mathbf 1 ( u ( x ) > k - 1 ).$$
Let τ -k = τ -k ( x ) be the stopping time such that at time τ -k +1 the k th pair is emitted from x. Now clearly 1 ( X x k = 1)+ 1 ( Y x k = 1) is independent of the filtration F τ -k ∧ τ . Also 1 ( u x ( ) > k -1) is measurable with respect to F τ -k ∧ τ . Hence E ( 1 ( X x k = 1) + 1 ( Y x k = 1) |F τ -k ∧ τ ) = 1 . Thus,
$$\i \, \text{ music}, \\ \mathbb { E } ( D _ { x, x + 1 } ) & = \sum _ { k = 1 } ^ { \infty } \mathbb { E } \left [ \mathbb { E } \left ( 1 ( X _ { k } ^ { x } = 1 ) + 1 ( Y _ { k } ^ { x } = 1 ) 1 ( u ( x ) > k - 1 ) | \mathcal { F } _ { \tau _ { k } ^ { - } \wedge \tau } \right ) \right ] \\ & = \sum _ { k = 1 } ^ { \infty } \mathbb { E } ( 1 ( u ( x ) > k - 1 ) ) \\ & = \mathbb { E } ( u ( x ) ). \\ \wedge \cdot \cdot \cdot \cdot$$
Similarly
$$\mathbb { E } ( D _ { x, x - 1 } ) = \mathbb { E } ( u ( x ) ).$$
Since the total number of particles is 2 n , using the symmetry of the process with respect to the origin one can conclude that the expected number of particles at the end of the process to the right of x is at most n . Hence using Remark 4 we have
$$\mathbb { E } ( D _ { x, x + 1 } - D _ { x + 1, x } ) \leq n.$$
Thus we are done.
Lemma 6.2. ˜( u x ) satisfies the following properties:
- (i) ˜( u x ) is an even function;
- (ii) restricted to Z + , ˜( u x ) is strictly decreasing;
glyph[negationslash]
- (iii) for every x = 0
$$\Delta \tilde { u } ( x ) > 0$$
where ∆ is the discrete Laplacian i.e.
$$\Delta \tilde { u } ( x ) = \tilde { u } ( x - 1 ) + \tilde { u } ( x + 1 ) - 2 \tilde { u } ( x ).$$
Proof. The proof of ( ) follows from the symmetry of the process with respect to the origin. i
To prove ( ii ) we first recall the definition of the function g t from (16). For any x ≥ 0 by Lemma 6.1, the difference ˜( u x ) -˜( u x +1) is the expected number of particles that stop to the right of x :
$$\tilde { u } ( x ) - \tilde { u } ( x + 1 ) = \mathbb { E } \left ( \sum _ { y = x + 1 } ^ { \infty } | g _ { \tau } ( y ) | \right ).$$
Since the quantity on the right hand side is nonnegative we have that ˜( u x ) is non increasing. To see that it is strictly decreasing, we make the simple observation that given any x > 0, with positive probability all the 2 n particles stop somewhere to the right of x . In other words,
$$\mathbb { P } \left ( \sum _ { y = x + 1 } ^ { \infty } | g _ { \tau } ( y ) | = 2 n \right ) > 0.$$
Hence E ( ∑ ∞ y = +1 x | g τ ( y ) | ) > 0, implying the statement.
To prove ( iii ) we see that by (64)
$$\Delta \tilde { u } ( x ) = \tilde { u } ( x - 1 ) - 2 \tilde { u } ( x ) + \tilde { u } ( x + 1 ) = \mathbb { E } \left ( \sum _ { y = x } ^ { \infty } | g _ { \tau } ( y ) | - \sum _ { y = x + 1 } ^ { \infty } | g _ { \tau } ( y ) | \right ) = \mathbb { E } ( | g _ { \tau } ( x ) | ).$$
By using similar reasoning as in the proof of ( ii ), we see that there is positive chance that | g τ ( x ) | > 0. Hence ∆˜( u x ) > 0.
## 6.2 The differential equation w ′′ = √ 2 w/π
In this section we work toward the proof of Theorem 1.4. We recall Conjecture 1.3 stated in the introduction. Note that we have not assumed a priori that w is continuous. Proving this is our first order of business.
Lemma 6.3. w is continuous, (in fact, 1 -Lipschitz) on R . Moreover it is non-increasing on the positive real axis.
Proof. ( ) of Lemma 6.2 implies that i w is an even function. Hence it suffices to prove that w is 1-Lipschitz on [0 , ∞ ). By Lemma 6.1, for any x, k ∈ Z ≥ 0 we have
$$0 \leq \tilde { u } ( x ) - \tilde { u } ( x + k ) = \sum _ { j = 0 } ^ { k - 1 } [ \tilde { u } ( x + j ) - \tilde { u } ( x + j + 1 ) ] \leq k n.$$
Now let x = glyph[floorleft] n 1 3 / ξ glyph[floorright] and k = glyph[floorleft] n 1 3 / ( ξ + h ) glyph[floorright] -glyph[floorleft] n 1 3 / ξ glyph[floorright] . Dividing by n 4 3 / and taking n →∞ , we obtain
$$0 \leq w ( \xi ) - w ( \xi + h ) \leq h$$
thus completing the proof of the lemma.
Recall the set of random variables
$$X _ { i } ^ { x }, Y _ { i } ^ { x }$$
defined in (6). To go further, we define the following quantities: For y = x ± 1
$$O _ { x, y } ^ { k } \coloneqq \sum _ { \mu } ^ { k } \mathbf 1 ( X _ { i } ^ { x } = \pm 1 ),$$
$$W _ { x, y } ^ { k } \coloneqq \sum _ { \mu } ^ { k } \mathbf 1 ( Y _ { i } ^ { x } = \pm 1 ).$$
$$\coloneqq \sum _ { i = 0 } ^ { k } \mathbf 1 ( X _ { i } ^ { x } = \pm 1 ), \\ \coloneqq \sum _ { i = 0 } ^ { k } \mathbf 1 ( Y _ { i } ^ { x } = \pm 1 ).$$
In other words, x .
O k x,y := # oil particles sent from { x to y within the first k firings at x , } W k x,y := # water particles sent from { x to y within the first k firings at }
Thus
$$g _ { \tau } ( x ) = O _ { x - 1, x } ^ { u ( x - 1 ) } + O _ { x + 1, x } ^ { u ( x + 1 ) } - W _ { x - 1, x } ^ { u ( x - 1 ) } - W _ { x + 1, x } ^ { u ( x + 1 ) }$$
where g t ( ) was defined in (16). · Consider now the analogous expression using a deterministic portion of each stack (recall ˜( u x ) = E u x ( )):
$$\tilde { g } _ { \tau } ( x ) \coloneqq O _ { x - 1, x } ^ { \lfloor \tilde { u } ( x - 1 ) \rfloor } + O _ { x + 1, x } ^ { \lfloor \tilde { u } ( x + 1 ) \rfloor } - W _ { x - 1, x } ^ { \lfloor \tilde { u } ( x - 1 ) \rfloor } - W _ { x + 1, x } ^ { \lfloor \tilde { u } ( x + 1 ) \rfloor }.$$
Because ˜ is deterministic the four terms on the right side are independent. u Moreover each term O k x,y and W k x,y for | x - | y = 1 is a sum of k independent Bernoulli(1 / 2) random variables. So the right side is a sum of glyph[floorleft] ˜( u x -1) glyph[floorright] + ˜( glyph[floorleft] u x + 1) glyph[floorright] independent random variables with the same law as a single step of a lazy symmetric random walk as defined in (40). Setting x = glyph[floorleft] n 1 3 / ξ glyph[floorright] for a real number ξ > 0, by ( ii ) in Conjecture 1.3 we have
$$\lim _ { n \to \infty } \frac { \lfloor \tilde { u } ( x - 1 ) \rfloor + \lfloor \tilde { u } ( x + 1 ) \rfloor } { n ^ { 4 / 3 } } = \lim _ { n \to \infty } \frac { 2 \tilde { u } ( x ) } { n ^ { 4 / 3 } } \ = \ 2 w ( \xi ).$$
This is because by Lemma 6.1, | ˜( u x ) -˜( u x +1) | and | ˜( u x -1) -˜( u x ) | are both less than n . As n →∞ by the central limit theorem, since each variable in (40) has variance 1 / 2, we obtain
$$n ^ { - 2 / 3 } \tilde { g } _ { \tau } ( x ) \stackrel { d } { \longrightarrow } N ( 0, w ( \xi ) ).$$
By Lemma 5.1 ( v ) we also have
$$n ^ { - 2 / 3 } \mathbb { E } | \tilde { g } _ { \tau } ( x ) | \longrightarrow \sqrt { \frac { 2 } { \pi } w ( \xi ) }.$$
Next we observe that under ( ) in Conjecture 1.3 , the same kind of central limit theorem i holds for g τ itself.
Lemma 6.4. Let ξ ≥ 0 . For x = glyph[floorleft] n 1 3 / ξ glyph[floorright] , we have as n →∞
(i)
$$n ^ { - 2 / 3 } g _ { \tau } ( x ) \stackrel { d } { \longrightarrow } N ( 0, w ( \xi ) )$$
(ii)
$$n ^ { - 2 / 3 } \mathbb { E } | g _ { \tau } ( x ) | \longrightarrow \sqrt { \frac { 2 } { \pi } } w ( \xi ).$$
Remark 6. ( ii ) along with (65) implies
$$\lim _ { n \to \infty } \frac { \Delta \tilde { u } ( x ) } { n ^ { 2 / 3 } } \to \sqrt { \frac { 2 } { \pi } w ( \xi ) }.$$
To prove Lemma 6.4 we need the next two results.
Lemma 6.5. E [ τ 1 ( τ > n 5 ) ] = O (1) where τ is defined in (11) .
Corollary 6.6. There exists a constant C > 1 0 such that
$$\sup _ { x \in \mathbb { Z } } \tilde { u } ( x ) \leq C _ { 1 } n ^ { 4 / 3 }$$
where ˜ = u E ( u . )
The proofs of the above two results are deferred to Appendix A.
## 6.2.1 Proof of Lemma 6.4.
By (68) and (69) it suffices to show
$$\lim _ { n \to 0 } n ^ { - 2 / 3 } \mathbb { E } | g _ { \tau } ( x ) - \tilde { g } _ { \tau } ( x ) | = 0.$$
Referring to the definitions of g and ˜ in (66) and (67) respectively, this will be accomplished g if we show that as n →∞ for y = x ± 1
$$n ^ { - 2 / 3 } \mathbb { E } \left | \left ( O _ { y, x } ^ { \lfloor \tilde { u } ( y ) \rfloor } - W _ { y, x } ^ { \lfloor \tilde { u } ( y ) \rfloor } \right ) - \left ( O _ { y, x } ^ { u ( y ) } - W _ { y, x } ^ { u ( y ) } \right ) \right |$$
tend to 0. For y = x ± 1 the above differences have identical distributions. Hence it suffices to look at any one. The quantity ( O ˜( u y ) y,x -W ˜( u y ) y,x ) -( O u y ( ) y,x -W u y ( ) y,x ) is a sum of
$$N = | \lfloor \tilde { u } ( y ) \rfloor - u ( y ) |$$
independent random variables X ,. . . , X 1 N with the same law as in (40). By Conjecture 1.3 (i), N/n 4 3 / → 0 in distribution. Fix glyph[epsilon1] ≥ 0. Let
$$\bullet \ A _ { 1 } = 1 ( N \leq \epsilon n ^ { 4 / 3 } ) \sup _ { 1 \leq m \leq \epsilon n ^ { 4 / 3 } } | \sum _ { i = 1 } ^ { m } X _ { i } |$$
$$\bullet \ A _ { 2 } = \mathbf 1 ( N \geq \epsilon n ^ { 4 / 3 } ) \mathbf 1 ( \mathcal { G } ) \sup _ { 1 \leq m \leq ( C + C _ { 1 } ) n ^ { 4 / 3 } } | \sum _ { i = 1 } ^ { m } X _ { i } |$$
- · A 3 = 2 n 5 1 ( G c ) 1 ( τ ≤ n 5 )
- · A 4 = 2 τ 1 ( G c ) 1 ( τ > n 5 )
where C and C 1 are the constants appearing in the statement of Theorem 1.1 and Corollary 6.6 respectively, τ is defined in (11) and G is defined in (19). We now claim that
$$\left | \left ( O _ { y, x } ^ { \lfloor \tilde { u } ( y ) \rfloor } - W _ { y, x } ^ { \lfloor \tilde { u } ( y ) \rfloor } \right ) - \left ( O _ { y, x } ^ { u ( y ) } - W _ { y, x } ^ { u ( y ) } \right ) \right | \leq A _ { 1 } + A _ { 2 } + A _ { 3 } + A _ { 4 }.$$
The first two terms correspond to the cases
- · N ≤ glyph[epsilon1]n 4 3 /
- · { N ≥ glyph[epsilon1]n 4 3 / } ∩ G .
For the last two terms we use the naive bound that
$$| \sum _ { i = 1 } ^ { N } X _ { i } | \leq N \leq \sup _ { x \in \mathbb { Z } } \tilde { u } ( x ) + \sup _ { x \in \mathbb { Z } } u ( x ) \leq C _ { 1 } n ^ { 4 / 3 } + \tau$$
where the last inequality uses Corollary 6.6. Using the above bound and looking at the events { τ ≤ n 5 } ∩ G c and { τ > n 5 } ∩ G c gives us (72).
- · By Lemma 5.1 ( iii ) , E ( A 1 ) = O ( √ glyph[epsilon1]n 2 3 / )
- · By Cauchy-Schwarz inequality and Lemma 5.1 ( iv )
$$\mathbb { E } ( A _ { 2 } ) = O ( n ^ { 2 / 3 } ) \sqrt { P ( N \geq \epsilon n ^ { 4 / 3 } ) }$$
- · E ( A 3 + A 4 ) = O (1) by Theorem 1.1 and Lemma 6.5 respectively.
Thus for any glyph[epsilon1] > 0
$$\mathbb { E } | g _ { \tau } ( x ) - \tilde { g } _ { \tau } ( x ) | \leq \sum _ { i = 1 } ^ { 4 } \mathbb { E } ( A _ { i } ) = n ^ { 2 / 3 } \left ( O ( \sqrt { \epsilon } + \sqrt { P ( N \geq \epsilon n ^ { 4 / 3 } ) } ) \right ).$$
Hence (71) follows using the above and Conjecture 1.3 ( ) ( i N/n 4 3 / goes to 0 in distribution) and we are done.
Remark 7. Note that we actually prove (71) uniformly over x i.e.
$$\lim _ { n \to 0 } \sup _ { x \in \mathbb { Z } } n ^ { - 2 / 3 } \mathbb { E } | g _ { \tau } ( x ) - \tilde { g } _ { \tau } ( x ) | = 0.$$
We now prove an uniform version of (70).
Lemma 6.7. Given glyph[epsilon1] > 0 and x < y such that w x , w y ( ) ( ) > 0 , for large enough n ,
$$\sup _ { \lfloor x n ^ { 1 / 3 } \rfloor \leq j \leq \lfloor y n ^ { 1 / 3 } \rfloor } \left | \frac { \Delta \tilde { u } ( j ) } { n ^ { 2 / 3 } } - \sqrt { \frac { 2 } { \pi } } w ( \frac { j } { n ^ { 1 / 3 } } ) \right | \leq \epsilon$$
Proof. Since w is continuous by Lemma 6.3 and hence uniformly continuous on [ x, y ], for glyph[epsilon1] > 0 there exists real numbers
$$x = x _ { 0 } < x _ { 1 } < \dots < x _ { k } = y$$
such that
$$\sup _ { 1 \leq i \leq k } ( x _ { i } - x _ { i - 1 } ) & \ \leq \ \epsilon \\ \sup _ { 1 \leq i \leq k } ( w ( x _ { i - 1 } ) - w ( x _ { i } ) ) & \ \leq \ \epsilon.$$
By Conjecture 1.3 ( ii ) and (70) we have for large enough n
$$\sup _ { 1 \leq i \leq k } | \frac { \tilde { u } ( \lfloor n ^ { 1 / 3 } x _ { i } \rfloor ) } { n ^ { 4 / 3 } } - w ( x _ { i } ) | \ \leq \ \epsilon$$
$$\sup _ { 1 \leq i \leq k } | \frac { \Delta \tilde { u } ( \lfloor n ^ { 1 / 3 } x _ { i } \rfloor ) } { n ^ { 2 / 3 } } - \sqrt { \frac { 2 } { \pi } w ( x _ { i } ) } | \ \leq \ \epsilon.$$
Now for any glyph[floorleft] xn 1 3 / glyph[floorright] ≤ j ≤ glyph[floorleft] yn 1 3 / glyph[floorright] find 0 ≤ i < k such that
$$\lfloor x _ { i } n ^ { 1 / 3 } \rfloor \leq j \leq \lfloor x _ { i + 1 } n ^ { 1 / 3 } \rfloor.$$
Clearly it suffices to show,
$$\left | \Delta \tilde { u } ( j ) - \Delta \tilde { u } ( \lfloor x _ { i } n ^ { 1 / 3 } \rfloor ) \right | = O ( \sqrt { \epsilon } n ^ { 2 / 3 } ).$$
or by (65)
$$\left | \mathbb { E } | g _ { \tau } ( j ) | - \mathbb { E } | g _ { \tau } ( \lfloor x _ { i } n ^ { 1 / 3 } ) | \right | = O ( \sqrt { \epsilon } n ^ { 2 / 3 } )$$
Notice that by (71) and Remark 7 we have
$$| \mathbb { E } | g _ { \tau } ( j ) | - \mathbb { E } | \tilde { g } _ { \tau } ( j ) | | \ = \ o ( n ^ { 2 / 3 } )$$
$$\left | \mathbb { E } | g _ { \tau } ( \lfloor n ^ { 1 / 3 } x _ { i } \rfloor ) | - \mathbb { E } | \tilde { g } _ { \tau } ( \lfloor n ^ { 1 / 3 } x _ { i } \rfloor ) \right | \ = \ o ( n ^ { 2 / 3 } )$$
Hence it suffices to show
$$| \mathbb { E } | \tilde { g } _ { \tau } ( j ) | - \mathbb { E } | \tilde { g } _ { \tau } ( \lfloor n ^ { 1 / 3 } x _ { i } \rfloor ) | | \ = \ O ( \sqrt { \epsilon } n ^ { 2 / 3 } )$$
$$| \overset { | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
| \mathbb { E } | | \widetilde { g } _ { \tau } ( | | n ^ { 1 / 3 } x _ { i + 1 } | | _ { 1 } ) | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |. \quad \lambda _ { i + 1 } | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |$$
By (74)
$$\tilde { u } ( j ) - \tilde { u } ( \lfloor n ^ { 1 / 3 } x _ { i } \rfloor ) & \ \leq \ \epsilon n ^ { 4 / 3 } \\ \tilde { u } ( \lfloor n ^ { 1 / 3 } x _ { i + 1 } \rfloor ) - \tilde { u } ( j ) & \ \leq \ \epsilon n ^ { 4 / 3 }.$$
Now by (67) the quantities on the left hand side of (78) and (79) are absolute values of a lazy symmetric random walk run for time ˜( ) u j -˜( u glyph[floorleft] n 1 3 / x i glyph[floorright] ) and ˜( u glyph[floorleft] n 1 3 / x i +1 glyph[floorright] ) -˜( u j ) respectively. The result now follows from Lemma 5.1 ( v ).
## Corollary 6.8.
$$\int _ { 0 } ^ { \infty } \sqrt { \frac { 2 } { \pi } w ( \zeta ) } d \zeta \leq 1$$
$$\lim _ { x \to \infty } w ( x ) = 0.$$
Proof. By Lemma 6.1
$$\mathbb { E } \left ( \sum _ { y = 1 } ^ { \infty } | g _ { \tau } ( y ) | \right ) = \sum _ { y = 1 } ^ { \infty } \Delta \tilde { u } ( y ) = \tilde { u } ( 0 ) - \tilde { u } ( 1 ) \leq n.$$
which in particular implies
Thus for any positive number A
$$\sum _ { y = 1 } ^ { \lfloor A n ^ { 1 / 3 } \rfloor } \Delta \tilde { u } ( y ) \leq n.$$
By Lemma 6.7 and aproximation of integral by Riemann sum we have
$$1 \geq \lim _ { n \to \infty } \sum _ { y = 1 } ^ { A n ^ { 1 / 3 } } \frac { 1 } { n ^ { 1 / 3 } } \frac { \Delta \tilde { u } ( y ) } { n ^ { 2 / 3 } } = \int _ { 0 } ^ { A } \sqrt { \frac { 2 } { \pi } w ( \zeta ) } d \zeta.$$
Since w is non negative it follows that
$$\int _ { 0 } ^ { \infty } \sqrt { \frac { 2 } { \pi } } w ( \zeta ) d \zeta \leq 1.$$
By Lemma 6.3 w is non increasing, hence this implies that
$$\lim _ { x \to \infty } w ( x ) = 0. \ 1$$
Remark 8. Note that we assumed only convergence of ˜ in Conjecture 1.3 ( u ii ) but were able to use a special feature of the oil and water model (namely, the identity ∆˜( u x ) = E | g τ ( x ) ) | to obtain something stronger, convergence of the discrete Laplacian ∆˜. u
Next we use this to argue that the scaling limit w ξ ( ) is actually a twice differentiable function of ξ > 0.
For any glyph[epsilon1] > 0 by Conjecture 1.3 ( ) and the above corollary we can choose i L large enough so that for large enough n ,
$$\sup _ { | \xi | \geq L } \tilde { u } ( \lfloor n ^ { 1 / 3 } \xi \rfloor ) < \varepsilon ^ { 2 } n ^ { 4 / 3 }$$
for all | ξ | > L.
Lemma 6.9. Given glyph[epsilon1] > 0 let L be as chosen above. Then
$$\sum _ { | x | > n ^ { 1 / 3 } L } \Delta \tilde { u } ( x ) \leq \varepsilon ^ { 1 / 2 } n.$$
Proof. Since the n 1 3 / ε differences ˜( u x ) -˜( u x +1) for x = glyph[floorleft] n 1 3 / L , . . . , glyph[floorright] glyph[floorleft] n 1 3 / ( L + ) ε glyph[floorright] -1 are nonnegative and sum to at most ˜( u glyph[floorleft] n 1 3 / L glyph[floorright] ) ≤ ε n 2 4 3 / , the smallest of them (which is the last one) must be at most εn . Therefore
$$\sum _ { | x | > n ^ { 1 / 3 } ( L + \varepsilon ) } \Delta \tilde { u } ( x ) \leq \varepsilon n.$$
Now the fact that
$$\sum _ { x = \lfloor n ^ { 1 / 3 } L \rfloor } ^ { \lfloor n ^ { 1 / 3 } ( L + \varepsilon ) \rfloor } \Delta \tilde { u } ( x ) \leq O ( \varepsilon ) n.$$
follows from Lemma 5.5 and the fact ∆˜( u x ) = E | g τ ( x ) | . Combining the above two results the proof follows.
Lemma 6.10. w is differentiable on the positive real line, and for any ξ > 0
$$w ^ { \prime } ( \xi ) = - \int _ { \xi } ^ { \infty } \sqrt { \frac { 2 } { \pi } } w ( \zeta ) \, d \zeta.$$
Proof. By summation by parts, for positive integers x, k
$$\frac { 1 } { k n } [ \tilde { u } ( x ) - \tilde { u } ( x + k ) ] = \frac { 1 } { k n } \sum _ { j = 1 } ^ { \infty } \min ( j, k ) \Delta \tilde { u } ( x + j ).$$
For positive real numbers ξ, L, h let x = glyph[floorleft] ξn 1 3 / glyph[floorright] , Z = glyph[floorleft] Ln 1 3 / glyph[floorright] , k = glyph[floorleft] hn 1 3 / glyph[floorright] and consider the first part of the sum in (82)
$$\frac { 1 } { h } \sum _ { j = 1 } ^ { Z } \frac { 1 } { n ^ { \frac { 1 } { 3 } } } \frac { \min ( j, k ) } { n ^ { \frac { 1 } { 3 } } } \frac { \Delta \tilde { u } ( x + j ) } { n ^ { \frac { 2 } { 3 } } }.$$
Now given δ > 0 , by Lemma 6.7 for large enough n
$$\left | \frac { 1 } { h } \sum _ { j = 1 } ^ { Z } \frac { 1 } { n ^ { \frac { 1 } { 3 } } } \frac { \min ( j, k ) } { n ^ { \frac { 1 } { 3 } } } \frac { \Delta \tilde { u } ( x + j ) } { n ^ { \frac { 2 } { 3 } } } - \frac { 1 } { h } \sum _ { j = 1 } ^ { Z } \frac { 1 } { n ^ { \frac { 1 } { 3 } } } \min ( \frac { j } { n ^ { \frac { 1 } { 3 } } }, h ) \sqrt { \frac { 2 } { \pi } } w ( \xi + \frac { j } { n ^ { \frac { 1 } { 3 } } } ) \right | \leq \delta L.$$
Notice that
$$\frac { 1 } { h } \sum _ { j = 1 } ^ { Z } \frac { 1 } { n ^ { \frac { 1 } { 3 } } } \min ( \frac { j } { n ^ { \frac { 1 } { 3 } } }, h ) \sqrt { \frac { 2 } { \pi } w ( \xi + \frac { j } { n ^ { \frac { 1 } { 3 } } } ) }$$
is a Riemann sum approximation of the integral
$$\frac { 1 } { h } \int _ { \xi } ^ { L + \xi } \min ( \zeta - \xi, h ) \sqrt { \frac { 2 } { \pi } w ( \zeta ) } \, d \zeta.$$
Thus as n goes to ∞ we see that
$$\frac { 1 } { h } \sum _ { j = 1 } ^ { Z } \frac { 1 } { n ^ { \frac { 1 } { 3 } } } \frac { \min ( j, k ) } { n ^ { \frac { 1 } { 3 } } } \frac { \Delta \tilde { u } ( x + j ) } { n ^ { \frac { 2 } { 3 } } } \to \frac { 1 } { h } \int _ { \xi } ^ { L + \xi } \min ( \zeta - \xi, h ) \sqrt { \frac { 2 } { \pi } } w ( \zeta ) \, d \zeta.$$
Fixing glyph[epsilon1] > 0 and choosing the same L as in the statement of Lemma 6.9 we get that the sum of the remaining terms in (82) 1 kn ∞ ∑ j = +1 Z is at most ε 1 2 / by Lemma 6.9. Hence as n →∞ we get from (82)
$$\frac { w ( \xi ) - w ( \xi + h ) } { h } = \frac { 1 } { h } \int _ { \xi } ^ { L } \min ( \zeta - \xi, h ) \sqrt { \frac { 2 } { \pi } w ( \zeta ) } \, d \zeta + O ( \epsilon ^ { 1 / 2 } ).$$
Sending h to 0 followed by glyph[epsilon1] to 0 ( L to ∞ ) we are done.
The right side of (81) is manifestly a differentiable function of ξ , so we obtain the following.
Corollary 6.11. Under Conjecture 1.3, the function w restricted to the positive real axis is twice continuously differentiable and obeys the differential equation
$$w ^ { \prime \prime } = \sqrt { \frac { 2 } { \pi } w }.$$
Lemma 6.12. w is compactly supported. Moreover on the positive region of support
$$w ( x ) = \left ( - \frac { 1 } { 4 } \left ( \frac { 3 2 } { 9 \pi } \right ) ^ { 1 / 4 } x + b \right ) ^ { 4 }$$
for some b > 0 .
Before proving the above we quote the well known Picard existence and uniqueness result for ODE's.
Theorem 6.13. [13, Theorem 8.13] Consider an initial value problem (IVP)
$$y ^ { \prime } ( x ) \ = \ f ( y ( x ), x )$$
$$y ( x _ { 0 } ) \ = \ y _ { 0 }$$
with the point ( x , y 0 0 ) belonging to some rectangle ( a, b ) × ( A,B ) i.e a < a 0 < b and A < y 0 < B. Also assume that f is M -Lipchitz for some M ≥ 0 i.e.
$$| f ( z, x ) - f ( w, x ) | \leq M | z - w |$$
for all x ∈ ( a, b ) z, w ∈ ( A,B . ) Then there exists a h = h x , y ( 0 0 , M ) > 0 such that
- · Existence: There exists a solution to the IVP on the interval ( x 0 -h, x 0 + h ) .
- · Uniqueness: Any two solutions of the IVP agree on the interval ( x 0 -h, x 0 + h ) .
Proof of Lemma 6.12 . Multiplying (83) by w ′ on both sides we get
$$w ^ { \prime } w ^ { \prime \prime } = \sqrt { \frac { 2 } { \pi } } w w ^ { \prime }.$$
Integrating both sides from ξ to ∞ and using the fact that lim x →∞ w x ( ) = lim x →∞ w x ′ ( ) = 0 (from Corollary 6.8 and Lemma 6.10) and that w ′ is non positive we see that w x ( ) satisfies the first order ODE
$$f ^ { \prime } = - \left ( \frac { 3 2 } { 9 \pi } \right ) ^ { 1 / 4 } f ^ { 3 / 4 }.$$
Now suppose w is positive on the entire real axis. Given any z ∈ R + then w z ( ) and w z ′ ( ) are both non zero. Thus we can find a, b such that
$$( a z + b ) ^ { 4 } = w ( z ) \\ 4 a ( a z + b ) ^ { 3 } = w ^ { \prime } ( z )$$
By (86)
$$4 a = - \left ( \frac { 3 2 } { 9 \pi } \right ) ^ { 1 / 4 }.$$
Because of the particular choice of a and b the function ( ax + ) b 4 also satisfies (86). Now since w z ( ) and w z ′ ( ) are both non zero the function w 3 4 / ( z ) is Lipchitz in a neighborhood of z . Hence by Theorem 6.13 ODE (86) has an unique solution in some neighborhood of z . Thus the functions w x ( ) and ( ax + ) b 4 are equal in a neighborhood of z . Now looking at the biggest interval I containing z such that w x ( ) = ( ax + ) b 4 on I we conclude that w x ( ) = ( ax + ) b 4 on R + ∩ supp w . ( ) In particular since ( ax + ) b 4 is positive only on a compact set this implies that w x ( ) has compact support.
Now we find the value for b which completely determines w.
## Lemma 6.14.
$$- 4 a b ^ { 3 } = \lim _ { h \rightarrow 0 ^ { + } } \frac { w ( 0 ) - w ( h ) } { h } = 1.$$
In particular
$$b = \left ( \frac { 9 \pi } { 3 2 } \right ) ^ { 1 / 1 2 }.$$
$$\cdot ^ { 3 } = \lim _ { h \to 0 ^ { + } } \frac { w ( 0 ) - w ( h ) } { h } = 1. \\ \\ b = \left ( \frac { 9 \pi } { 3 2 } \right ) ^ { 1 / 1 2 }.$$
Proof. That -4 ab 3 = lim h → 0 + w (0) -w h ( ) h follows from Lemma 6.12. To see that the quantity equals 1 fix h > 0 . Consider the telescopic sum
$$\tilde { u } ( 0 ) - \tilde { u } ( h n ^ { 1 / 3 } ) = \sum _ { i = 0 } ^ { h n ^ { 1 / 3 } } \tilde { u } ( i ) - \tilde { u } ( i + 1 ).$$
Now ˜( ) u i -˜( u i +1) is the expected number of particles on the right of i by Lemma 6.1. By symmetry of the process about the origin and Lemma 5.5
$$\sum _ { x > 0 } | g _ { \tau } ( x ) | = n - O ( n ^ { 2 / 3 } ).$$
Moreover for any i > 0
Summing over i we get
$$\tilde { u } ( 0 ) - \tilde { u } ( \lfloor h n ^ { 1 / 3 } \rfloor ) = h n ^ { 4 / 3 } - h ^ { 2 } O ( n ^ { 4 / 3 } ).$$
Dividing throughout by n 4 3 / and taking limit as n goes to infinity we get
$$w ( 0 ) - w ( h ) = h + O ( h ^ { 2 } ).$$
Thus dividing by h and sending h to 0 we are done.
## 6.3 Proof of Theorem 1.4
From Lemmas 6.12 and 6.14 and using the symmetry of w about the origin we get
$$w ( x ) = \left ( \left ( \frac { 9 \pi } { 3 2 } \right ) ^ { 1 / 1 2 } - \left ( \frac { 3 2 } { 9 \pi } \right ) ^ { 1 / 4 } \frac { | x | } { 4 } \right ) ^ { 4 }$$
on the region of support. Rearranging we get
$$w ( x ) = \begin{cases} \frac { 1 } { 7 2 \pi } \left ( ( 1 8 \pi ) ^ { 1 / 3 } - | x | \right ) ^ { 4 }, & | x | < ( 1 8 \pi ) ^ { 1 / 3 } \\ 0 & | x | \geq ( 1 8 \pi ) ^ { 1 / 3 }. \end{cases}$$
$$\tilde { u } ( i ) - \tilde { u } ( i + 1 ) = n - O ( n ^ { 2 / 3 } ) i.$$
## 7 Open Questions
Conjecture 1.3 is an obvious target. In this concluding section we collect some additional open questions.
## 7.1 Location of the rightmost particle
For the oil and water process with n particles of each type started at the origin Z , let R n be the location of the rightmost particle upon fixation. Is the sequence of random variables R /n n 1 3 / tight? Does it converge in distribution to a constant? If it does, then Theorem 1.4 suggests that the limit should be at least (18 π ) 1 3 / (and perhaps equal to this value).
## 7.2 Order of the variance
We believe that the standard deviation of the odometer u is of order n 7 6 / in the bulk. Note that Conjecture 1.3 asserts something weaker, namely o n ( 4 3 / ).
Here is a heuristic argument for the exponent 7 / 6. The total number of particle exits from x is 2 u x ( ); let N x be the total number of particle entries to x . Equating entries minus exits with the number of particles left behind, we find that
$$\Delta u ( x ) \coloneqq u ( x - 1 ) + u ( x + 1 ) - 2 u ( x ) = Z _ { 1 } ( x ) + | Z _ { 2 } ( x ) | - 2 n \delta _ { 0 } ( x ). \quad \quad ( 8 8 )$$
where Z 1 ( x ) = u x ( -1) + u x ( +1) -N x , and Z 2 ( x ) is the signed count of the number of particles remaining at x in the final state (counting oil as positive, water as negative). Both Z 1 ( x ) and Z 2 ( x ) are expressable as sums of indpendent indicators involving the stack elements at x ± 1. The limits of summation are u x ( ± 1). Assuming Conjecture 1.3 and arguing as in Lemma 6.4, we can replace the limits of summation by their expected values ˜( u x ± 1), incurring only a small error. The resulting sums ˜ Z 1 and ˜ Z 2 are asymptotically normal with mean zero and variance of order n 4 3 / (assuming x is in the bulk, | x < | ((18 π ) 1 3 / -ε n ) 1 3 / ). Moreover, the function ˜ Z 1 + | ˜ Z 2 | is 2 -dependent : its values at x and y are independent if | x - | y > 2. By summation by parts,
$$u ( x ) = \sum _ { j = 1 } ^ { \infty } j \Delta u ( x + j ).$$
Since most of the support of u is on an interval of length O n ( 1 3 / ), truncating this sum at Cn 1 3 / for a large constant C should not change its variance by much. Replacing ∆ u by its approximation ˜ Z 1 + ˜ | Z 2 | and using the 2-dependence, we arrive at
$$\text{Var} \, u ( x ) = \sum _ { j = 1 } ^ { C n ^ { 1 / 3 } } j ^ { 2 } O ( n ^ { 4 / 3 } ) = O ( n ^ { 7 / 3 } ).$$
## 7.3 Conjectured exponents in higher dimensions
For the oil and water model in Z d starting with n oil and n water particles at the origin, we believe that the typical order of the odometer (away from 0 and the boundary) is n 4 ( / d +2) and the radius of the occupied cluster is of order n 1 ( / d +2) . The reason is by analogy with Section 6.2: if w : R d → R solves the PDE
$$\Delta w = - \delta _ { 0 } + \sqrt { \frac { 2 } { \pi } } w$$
Figure 3: Contour lines of the odometer function u of the oil and water model in Z 2 with n = 2 22 particles of each type started at the origin. Each site is shaded according to the fractional part of 1 5 u 1 4 / .
then its rescaling
$$v ( x ) = t ^ { 4 } w ( x / t )$$
satisfies
$$\Delta v = - t ^ { d + 2 } \delta _ { 0 } + \sqrt { \frac { 2 } { \pi } } v.$$
If the odometer for n particles has a scaling limit w that satisfies (89), then v is the scaling limit of the odometer for t d +2 n particles. So increasing the number of particles a factor of t d +2 increases the radius by by a factor of t and the odometer by a factor of t 4 . This motivates the following conjecture.
Conjecture 7.1. Let u be the odometer for the oil and water model started from n particles of each type at the origin in Z d . There exists a deterministic function w : R d → R such that for all ξ ∈ R -{ } 0 we have almost surely,
$$\frac { u ( \lfloor n ^ { 1 / ( d + 2 ) } \xi \rfloor ) } { n ^ { 4 / ( d + 2 ) } } \rightarrow w ( \xi ).$$
Moreover, w is rotationally symmetric, twice differentiable on R d -{ } 0 and satisfies
$$\Delta w = \sqrt { \frac { 2 } { \pi } } w$$
on R d -{ } 0 and lim ξ → 0 w ξ ( ) g ξ ( ) = 1 where g is the Green function for the Laplacian on R d .
The fourth power scaling is reflected in the even spacing between contour lines of the odometer function in Figure 3.
## A Appendix
## A.1 Concentration Estimates
Lemma A.1. Suppose for all x ∈ Z we have
$$X _ { 1 } ^ { x }, Y _ { 1 } ^ { x }$$
$$X _ { 2 } ^ { x }, Y _ { 2 } ^ { x }$$
$$Y _ { 1 } ^ { x }, Y _ { 1 } ^ { x } \\ Y _ { 2 } ^ { x }, Y _ { 2 } ^ { x } \\ \vdots$$
a sequence of independent uniform ± 1 valued random variables. The sequences across x are also independent of each other. Then there exists constants C, C , γ > ′ 0 such that for n large enough, with probability at least 1 -C exp( -C n ′ γ ) for all k > √ n and -n 5 < j < n 5 we have
$$( i ) \, \left | \sum _ { i = 1 } ^ { k } \mathbf 1 _ { ( X _ { i } ^ { j } = 1 ) } - k / 2 \right | < k ^ { 1 / 2 + \epsilon } ;$$
$$( i i ) \, \left | \sum _ { i = 1 } ^ { k } { \mathbf 1 } _ { ( Y _ { i } ^ { j } = 1 ) } - k / 2 \right | < k ^ { 1 / 2 + \epsilon } ;$$
$$\left ( i i i \right ) \, \left | \sum _ { i = 1 } ^ { k } \mathbf 1 _ { ( X _ { i } ^ { j } = 1 ) } \mathbf 1 _ { ( Y _ { i } ^ { j } = 1 ) } - k / 4 \right | < k ^ { 1 / 2 + \epsilon } ;$$
$$( i v ) \, \left | \sum _ { i = 1 } ^ { k } \mathbf 1 _ { ( X _ { i } ^ { j } = - 1 ) } \mathbf 1 _ { ( Y _ { i } ^ { j } = - 1 ) } - k / 4 \right | < k ^ { 1 / 2 + \epsilon }.$$
Proof. Proof follows by standard bounds from Azuma-Hoeffding's inequality for Bernoulli random variables and union bound over k ≥ √ n followed by j ∈ -[ n , n 5 5 ].
## A.2 Proof of Lemma 5.2
Let us define the truncated variable
$$Y = M ( n ^ { \frac { 4 } { 3 } } ) { \mathbf 1 } \left ( M ( n ^ { \frac { 4 } { 3 } } ) < n ^ { \frac { 2 } { 3 } + \epsilon } \right )$$
for some small but a priori fixed glyph[epsilon1] . Let Y i be iid copies of Y. Now by using Azuma's inequality,
$$\mathbb { P } \left ( \sum _ { i = 1 } ^ { n ^ { 1 / 3 } } ( Y _ { i } - \mathbb { E } ( Y _ { i } ) ) > t \right ) \leq e ^ { - \frac { t ^ { 2 } } { n ^ { 1 / 3 } n ^ { 4 / 3 + 2 \epsilon } } }.$$
Taking t 2 = n 5 3+3 / glyph[epsilon1] we get that
$$\mathbb { P } \left ( \sum _ { i = 1 } ^ { n ^ { 1 / 3 } } ( Y _ { i } - \mathbb { E } ( Y _ { i } ) ) > t \right ) < e ^ { - n ^ { \epsilon } }$$
Now by ( iv ) Lemma 5.1 E ( Y ) = O n ( 2 3 / ).
$$\text{Thus}$$
$$n ^ { 1 / 3 } \mathbb { E } ( Y ) + t < D n$$
for some large D as t = n 5 6+2 / glyph[epsilon1] . Hence
$$\mathbb { P } \left ( \sum _ { i = 1 } ^ { n ^ { 1 / 3 } } Y _ { i } > D n \right ) \leq e ^ { - n ^ { \epsilon } }$$
This implies that
$$\mathbb { P } \left ( \sum _ { i = 1 } ^ { n ^ { 1 / 3 } } M ^ { i } ( n ^ { \frac { 4 } { 3 } } ) > D n \right ) \leq C e ^ { - n ^ { \gamma } }$$
since by ( ) Lemma 5.1 and union bound, there exists a positive constant i c > 0 such that glyph[negationslash]
$$\mathbb { P } \left ( \exists \ 1 \leq i \leq n ^ { 1 / 3 } \text{ such that } Y _ { i } \neq M ^ { i } \right ) \leq e ^ { - n ^ { c } }.$$
## A.3 Proof of Lemma 6.5
We use the variables E i defined in the statement of Lemma 3.5. Let
$$Y = \sum _ { i = 1 } ^ { \tau ^ { \prime } } E _ { i }$$
where τ ′ was defined in (14). As mentioned in proof of Lemma 3.1 by Lemmas 3.3 and 3.5 τ is stochastically dominated by Y . Thus
$$\mathbb { E } \left [ \tau \mathbf 1 ( \tau > n ^ { 5 } ) \right ] \leq \mathbb { E } \left [ Y \mathbf 1 ( Y > n ^ { 5 } ) \right ].$$
Hence to prove the lemma it suffices to show the right hand side is O (1). Now
$$\mathbb { E } \left [ Y 1 ( Y > n ^ { 5 } ) \right ] \ \leq \ \mathbb { E } \left [ Y 1 ( \tau ^ { \prime } > n ^ { \frac { 5 } { 2 } } ) \right ] + \mathbb { E } \left [ Y 1 ( \tau ^ { \prime } \leq n ^ { \frac { 5 } { 2 } } ) 1 ( Y > n ^ { 5 } ) \right ] \quad \ ( 9 0 )$$
$$\leq \ \mathbb { E } \left [ Y \mathfrak { 1 } ( \tau ^ { \prime } > n ^ { \frac { 5 } { 2 } } ) \right ] + \mathbb { E } \left [ \sum _ { i = 1 } ^ { n ^ { \frac { 5 } { 2 } } } E _ { i } \mathfrak { 1 } ( \sum _ { i = 1 } ^ { n ^ { \frac { 5 } { 2 } } } E _ { i } \geq n ^ { 5 } ) \right ] \ \quad \ \ ( 9 1 )$$
$$\leq \ \mathbb { E } \left [ Y \mathbf 1 ( \tau ^ { \prime } > n ^ { \frac { 5 } { 2 } } ) \right ] + \mathbb { E } \left [ \sum _ { i = 1 } ^ { n ^ { \frac { 5 } { 2 } } } E _ { i } \right ] \left [ \sum _ { i = 1 } ^ { n ^ { \frac { 5 } { 2 } } } \mathbf 1 ( E _ { i } \geq n ^ { \frac { 5 } { 2 } } ) \right ] \quad \quad ( 9 2 )$$
where the last inequality follows from the easy fact
$$1 \left ( \sum _ { i = 1 } ^ { n ^ { \frac { 5 } { 2 } } } E _ { i } \geq n ^ { 5 } \right ) \leq \sum _ { i = 1 } ^ { n ^ { \frac { 5 } { 2 } } } 1 ( E _ { i } \geq n ^ { \frac { 5 } { 2 } } ).$$
We use the following tail estimate for E 1 and τ ′ : there exists a constant c < 1 such that for k ≥ n 2
$$\max ( \mathbb { P } ( \tau ^ { \prime } \geq k ), \mathbb { P } ( E _ { 1 } \geq k ) ) \leq ( 1 - c ) ^ { \lfloor \frac { k } { n ^ { 2 } } \rfloor }$$
which easy follows from the fact that starting from any point in [ -2 n, 2 n ] there exists a constant chance c for the random walk to exit the interval in the next n 2 steps. Using (93), independence of τ ′ , E s ′ i , the theorem now follows from (92). The details are omitted.
## A.4 Proof of Corollary 6.6
The proof follows from the following observation:
$$u ( x ) \leq C n ^ { 4 / 3 } 1 ( \mathcal { G } ) + n ^ { 5 } 1 ( \mathcal { G } ^ { c } ) 1 ( \tau \leq n ^ { 5 } ) + \tau 1 ( \tau \geq n ^ { 5 } ) 1 ( \mathcal { G } ^ { c } )$$
where C is the constant appearing in the statement of Theorem 1.1, τ is defined in (11) and G is defined in (19). The first term follows from the definition of G . For the second and third term we use the trivial bound that
$$u ( x ) \leq \tau.$$
Taking expectation we get
$$\tilde { u } ( x ) \leq C n ^ { 4 / 3 } + n ^ { 5 } \mathbb { P } ( \mathcal { G } ^ { c } ) + \mathbb { E } ( \tau \mathbf 1 ( \tau \geq n ^ { 5 } ) ).$$
The last two terms are O (1) by Theorem 1.1 and Lemma 6.5 respectively. Hence we are done.
## Acknowledgments
We are grateful to Alexander Holroyd and Yuval Peres for valuable discussions, and to Deepak Dhar for bringing reference [1] to our attention. We thank Wilfried Huss and Ecaterina Sava-Huss for a careful reading of an earlier draft, and for detailed comments which improved the paper.
## References
- [1] F. C. Alcaraz, P. Pyatov, and V. Rittenberg. Two-component abelian sandpile models. Physical Review E , 79(4):042102, 2009. arXiv:0810.4053 .
- [2] Amine Asselah and Alexandre Gaudilli`re. From logarithmic to subdiffusive polynomial e fluctuations for internal dla and related growth models. The Annals of Probability , 41(3A):1115-1159, 2013. arXiv:1009.2838 .
- [3] Amine Asselah and Alexandre Gaudilli` ere. Sublogarithmic fluctuations for internal DLA. The Annals of Probability , 41(3A):1160-1179, 2013. arXiv:1011.4592 .
- [4] Amine Asselah and Alexandre Gaudilli` ere. Lower bounds on fluctuations for internal DLA. Probability Theory and Related Fields , 158(1-2):39-53, 2014. arXiv:1111.4233 .
- [5] Patrick Billingsley. Convergence of probability measures . John Wiley & Sons, second edition, 1999.
- [6] Benjamin Bond and Lionel Levine. Abelian networks: foundations and examples. 2013. arXiv:1309.3445 .
- [7] Deepak Dhar. The abelian sandpile and related models. Physica A , 263:4-25, 1999. arXiv:cond-mat/9808047 .
| [8] | Rick Durrett. Probability: theory and examples . Cambridge University Press, fourth edition, 2010. |
|-------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [9] | William Feller. An introduction to probability theory and its applications. Vol. I . Third edition. John Wiley & Sons, Inc., New York-London-Sydney, 1968. |
| [10] | David Jerison, Lionel Levine, and Scott Sheffield. Logarithmic fluctuations for in- ternal DLA. Journal of the American Mathematical Society , 25(1):271-301, 2012. arXiv:1010.2483 . |
| [11] | David Jerison, Lionel Levine, and Scott Sheffield. Internal DLA in higher dimensions. Electronic Journal of Probability , 18(98):1-14, 2013. arXiv:1012.3453 . |
| [12] | David Jerison, Lionel Levine, and Scott Sheffield. Internal DLA and the Gaussian free field. Duke Mathematical Journal , 163(2):267-308, 2014. arXiv:1101.0596 . |
| [13] | Walter G Kelley and Allan C Peterson. The theory of differential equations: classical and qualitative . Springer, 2010. |
| [14] | Gregory F. Lawler, Maury Bramson, and David Griffeath. Internal diffusion limited aggregation. The Annals of Probability , 20(4):2117-2140, 1992. |
| [15] | Frank Spitzer. Principles of random walk . Springer-Verlag, New York, second edition, 1976. Graduate Texts in Mathematics, Vol. 34. |
| null |
[
"Elisabetta Candellero",
"Shirshendu Ganguly",
"Christopher Hoffman",
"Lionel Levine"
] |
2014-08-04T19:21:20+00:00
|
2014-08-04T19:21:20+00:00
|
[
"math.PR",
"cond-mat.stat-mech",
"82C24, 82C22, 60K35"
] |
Oil and water: a two-type internal aggregation model
|
We introduce a two-type internal DLA model which is an example of a non-unary
abelian network. Starting with n "oil" and n "water" particles at the origin,
the particles diffuse in Z according to the following rule: whenever some site
x has at least 1 oil and at least 1 water particle present, it "fires" by
sending 1 oil particle and 1 water particle each to an independent random
neighbor x+1 or x-1. Firing continues until every site has at most one type of
particles. We establish the correct order for several statistics of this model
and identify the scaling limit under assumption of existence.
|
1408.0777v3
|
## A Multiresolution Stochastic Process Model for Predicting Basketball Possession Outcomes
Daniel Cervone , Alex D'Amour , Luke Bornn , and Kirk Goldsberry 1 2 3 4
1 Center for Data Science, New York University, New York, NY 10003
2 Department of Statistics, Harvard University, Cambridge, MA 02138 3 Department of Statistics and Actuarial Science, Simon Fraser University, Burnaby, BC, Canada 4 Institute for Quantitative Social Science, Harvard University, Cambridge, MA 02138
## Author's Footnote:
Daniel Cervone (email: [email protected] ) is Moore-Sloan Data Science Fellow at New York University. Alex D'Amour ( [email protected] ) is PhD candidate, Statistics Department, Harvard University. Luke Bornn ( [email protected] ) is Assistant Professor, Department of Statistics and Actuarial Science, Simon Fraser University. Kirk Goldsberry ( [email protected] ) is Visiting Scholar, Center for Geographic Analysis, Harvard University. Daniel Cervone is funded by a fellowship from the Alfred P. Sloan and Betty Moore foundations. Luke Bornn is funded by DARPA Award FA8750-14-2-0117, ARO Award W911NF-15-1-0172, an Amazon AWS Research Grant, and the Natural Sciences and Engineering Research Council of Canada.
The authors would like to thank Alex Franks, Andrew Miller, Carl Morris, Natesh Pillai, and Edoardo Airoldi for helpful comments, as well as STATS LLC in partnership with the NBA for providing the optical tracking data. The computations in this paper were run on the Odyssey cluster supported by the FAS Division of Science, Research Computing Group at Harvard University.
## Abstract
Basketball games evolve continuously in space and time as players constantly interact with their teammates, the opposing team, and the ball. However, current analyses of basketball outcomes rely on discretized summaries of the game that reduce such interactions to tallies of points, assists, and similar events. In this paper, we propose a framework for using optical player tracking data to estimate, in real time, the expected number of points obtained by the end of a possession. This quantity, called expected possession value (EPV), derives from a stochastic process model for the evolution of a basketball possession. We model this process at multiple levels of resolution, differentiating between continuous, infinitesimal movements of players, and discrete events such as shot attempts and turnovers. Transition kernels are estimated using hierarchical spatiotemporal models that share information across players while remaining computationally tractable on very large data sets. In addition to estimating EPV, these models reveal novel insights on players' decision-making tendencies as a function of their spatial strategy. A data sample and R code for further exploration of our model/results are available in the repository https://github.com/dcervone/EPVDemo .
Keywords : Optical tracking data, spatiotemporal model, competing risks, Gaussian process.
## 1. INTRODUCTION
Basketball is a fast-paced sport, free-flowing in both space and time, in which players' actions and decisions continuously impact their teams' prospective game outcomes. Team owners, general managers, coaches, and fans all seek to quantify and evaluate players' contributions to their team's success. However, current statistical models for player evaluation such as 'Player Efficiency Rating' (Hollinger 2005) and 'Adjusted Plus/Minus' (Omidiran 2011) rely on highly reductive summary statistics of basketball games such as points scored, rebounds, steals, assists-the so-called 'box score' summary of the game. Such models reflect the fact that up until very recently, data on basketball games were only available in this low level of resolution. Thus previous statistical analyses of basketball performance have overlooked many of the high-resolution motifs-events not measurable by such aggregate statisticsthat characterize basketball strategy. For instance, traditional analyses cannot estimate the value of a clever move that fools the defender, or the regret of skipping an open shot in favor of passing to a heavily defended teammate. The advent of player tracking data in the NBA has provided an opportunity to fill this gap.
## 1.1 Player-Tracking Data
In 2013 the National Basketball Association (NBA), in partnership with data provider STATS LLC, installed optical tracking systems in the arenas of all 30 teams in the league. The systems track the exact two-dimensional locations of every player on the court (as well as the three-dimensional location of the ball) at a resolution of 25Hz, yielding over 1 billion space-time observations over the course of a full season.
Consider, for example, the following possession recorded using this player tracking system. This is a specific Miami Heat possession against the Brooklyn Nets from the second quarter of a game on November 1, 2013, chosen arbitrarily among those during which LeBron James (widely considered the best NBA player as of 2014) handles the ball. In this particular
Figure 1: Miami Heat possession against Brooklyn Nets. Norris Cole wanders into the perimeter (A) before driving toward the basket (B). Instead of taking the shot, he runs underneath the basket (C) and eventually passes to Rashard Lewis(D), who promptly passes to LeBron James (E). After entering the perimeter (F), James slips behind the defense (G) and scores an easy layup (H).
possession, diagrammed in Figure 1, point guard Norris Cole begins with possession of the ball crossing the halfcourt line (panel A). After waiting for his teammates to arrive in the offensive half of the court, Cole wanders gradually into the perimeter (inside the three point line), before attacking the basket through the left post. He draws two defenders, and while he appears to beat them to the basket (B), instead of attempting a layup he runs underneath the basket through to the right post (C). He is still being double teamed and at this point passes to Rashard Lewis (D), who is standing in the right wing three position. As defender Joe Johnson closes, Lewis passes to LeBron James, who is standing about 6 feet beyond the three point line and drawing the attention of Andray Blatche (E). James wanders slowly into the perimeter (F), until just behind the free throw line, at which point he breaks towards the basket. His rapid acceleration (G) splits the defense and gains him a clear lane to the basket. He successfully finishes with a layup (H), providing the Heat two points.
## 1.2 Expected Possession Value
Such detailed data hold both limitless analytical potential for basketball spectators and new methodological challenges to statisticians. Of the dizzying array of questions that could be asked of such data, we choose to focus this paper on one particularly compelling quantity of interest, which we call expected possession value (EPV), defined as the expected number of points the offense will score on a particular possession conditional on that possession's evolution up to time t .
For illustration, we plot the EPV curve corresponding to the example Heat possession in Figure 2, with EPV estimated using the methodology in this paper. We see several moments when the expected point yield of the possession, given its history, changes dramatically. For the first 2 seconds of the possession, EPV remains around 1. When Cole drives toward the basket, EPV rises until peaking at around 1.34 when Cole is right in front of the basket. As Cole dribbles past the basket (and his defenders continue pursuit), however, EPV falls rapidly, bottoming out at 0.77 before 'resetting' to 1.00 with the pass to Rashard Lewis.
The EPV increases slightly to 1.03 when the ball is then passed to James. As EPV is sensitive to small changes in players' exact locations, we see EPV rise slightly as James approaches the three point line and then dip slightly as he crosses it. Shortly afterwards, EPV rises suddenly as James breaks towards the basket, eluding the defense, and continues rising until he is beneath the basket, when an attempted layup boosts the EPV from 1.52 to 1.62.
Figure 2: Estimated EPV over time for the possession shown in Figure 1. Changes in EPV are induced by changes in players' locations and dynamics of motion; macrotransitions such as passes and shot attempts produce immediate, sometimes rapid changes in EPV. The black line slightly smooths EPV evaluations at each time point (gray dots), which are subject to Monte Carlo error.
In this way, EPV corresponds naturally to a coach's or spectator's sense of how the actions that basketball players take in continuous time help or hurt their team's cause to score in the current possession, and quantifies this in units of expected points. EPV acts like a stock ticker, providing an instantaneous summary of the possession's eventual point value given all available information, much like a stock price values an asset based on speculation of future expectations.
## 1.3 Related Work and Contributions
Concepts similar to EPV, where final outcomes are modeled conditional on observed progress, have had statistical treatment in other sports, such as in-game win probability in baseball (Bukiet, Harold & Palacios 1997; Yang & Swartz 2004) and football (Lock & Nettleton 2014), as well as in-possession point totals in football (Burke 2010; Goldner 2012). These previous efforts can be categorized into either marginal regression/classification approaches, where features of the current game state are mapped directly to expected outcomes, or processbased models that use a homogeneous Markov chain representation of the game to derive outcome distributions. Neither of these approaches is ideal for application to basketball. Marginal regression methodologies ignore the natural martingale structure of EPV, which is essential to its 'stock ticker' interpretation. On the other hand, while Markov chain methodologies do maintain this 'stock ticker' structure, applying them to basketball requires discretizing the data, introducing an onerous bias-variance-computation tradeoff that is not present for sports like baseball that are naturally discrete in time.
To estimate EPV effectively, we introduce a novel multiresolution approach in which we model basketball possessions at two separate levels of resolution, one fully continuous and one highly coarsened. By coherently combining these models we are able to obtain EPV estimates that are reliable, sensitive, stochastically consistent, and computationally feasible (albeit intensive). While our methodology is motivated by basketball, we believe that this research can serve as an informative case study for analysts working in other application areas where continuous monitoring data are becoming widespread, including traffic monitoring (Ihler, Hutchins & Smyth 2006), surveillance, and digital marketing (Shao & Li 2011), as well as other sports such as soccer and hockey (Thomas, Ventura, Jensen & Ma 2013).
Section 2 formally defines EPV within the context of a stochastic process for basketball, introducing the multiresolution modeling approach that make EPV calculations tractable as averages over future paths of a stochastic process. Parameters for these models, which represent players' decision-making tendencies in various spatial and situational circumstances, are discussed in Section 3. Section 4 discusses inference for these parameters using hierchical models that share information between players and across space. We highlight results from actual NBA possessions in Section 5, and show how EPV provides useful new quantifications of player ability and skill. Section 6 concludes with directions for further work.
## 2. MULTIRESOLUTION MODELING
To present our process model for a basketball possession, we require some formalism. Let Ω represent the space of all possible basketball possessions in full detail, with ω ∈ Ω describing the full path of a particular possession. For simplicity, we restrict our focus to possessions that do not include fouls , so that each possession we consider results in 0, 2, or 3 points 1 scored for the offense, denoted X ω ( ) ∈ { 0 2 3 , , } . For any possession path ω , we denote by Z ω ( ) the optical tracking time series generated by this possession so that Z t ( ω ) ∈ Z , t > 0, is a 'snapshot' of the tracking data exactly t seconds from the start of the possession ( t = 0). Z is a high dimensional space that includes ( x, y ) coordinates for all 10 players on the court, ( x, y, z ) coordinates for the ball, summary information such as which players are on the court and what the game situation is (game location, score, time remaining, etc.), and event annotations that are observable in real time, such as a turnover occurring, a pass, or a shot being attempted and the result of that attempt.
Taking the intuitive view of Ω as a sample space of possession paths, we model Z ω ( ) as a stochastic process, and likewise, define Z t ( ω ) for each t > 0 as a random variable in Z . Z ω ( ) provides the natural filtration F ( Z ) t = σ ( { Z -1 s : 0 ≤ s ≤ } t ), which represents all information available from the optical tracking data for the first t seconds of a possession. EPV is the expected value of the number of points scored for the possession ( X ) given all available data up to time t ( F ( Z ) t ):
Definition The expected possession value , or EPV, at time t ≥ 0 during a possession is ν t = E [ X |F ( Z ) t ].
1 Our data include foul events, but do not specify the type or circumstances of the foul. There are several types of fouls and game situations for which fouls lead to free throws-for instance, shooting fouls, technical/flagrant fouls, clear path fouls, and fouls during the fouling team's 'bonus' period; thus, modeling fouls presents additional complications relative to the other events we model in our EPV model. While drawing fouls can be a valuable and important part of team strategy, we omit modeling such behavior in this paper.
The expectation E [ X |F ( Z ) t ] is an integral over the distribution of future paths the current possession can take. Letting T ω ( ) denote the time at which a possession following path ω ends , the possession's point total is a deterministic function of the full resolution data at 2 this time, X ω ( ) = h Z ( T ω ( ) ( ω )). Thus, evaluating EPV amounts to integrating over the joint distribution of ( T, Z T ):
$$\nu _ { t } = \mathbb { E } [ X | \mathcal { F } _ { t } ^ { ( Z ) } ] & = \int _ { \Omega } X ( \omega ) \mathbb { P } ( d \omega | \mathcal { F } _ { t } ^ { ( Z ) } ) \\ & = \int _ { t } ^ { \infty } \int _ { Z } h ( z ) \mathbb { P } ( Z _ { s } = z | T = s, \mathcal { F } _ { t } ^ { ( Z ) } ) \mathbb { P } ( T = s | \mathcal { F } _ { t } ^ { ( Z ) } ) d z d s.$$
Note that we use probability notation P ( ) · somewhat heuristically, as P ( T = s |F ( Z ) t ) is a density with respect to Lebesgue measure, while Z s mixes both discrete (annotations) and continuous (locations) components. We will also generally omit the dependence on ω when writing random variables, e.g., Z t instead of Z t ( ω ).
## 2.1 Estimator Criteria
We have defined EPV in (1) as an unobserved, theoretical quantity; one could thus imagine many different EPV estimators based on different models and/or information in the data. However, we believe that in order for EPV to achieve its full potential as a basis for highresolution player and strategy evaluation, an EPV estimator should meet several criteria.
First, we require that the EPV estimator be stochastically consistent. Recognizing that EPV is simply a conditional expectation, it is tempting to estimate EPV using a regression or classification approach that maps features from F ( Z ) t to an outcome space, [0 , 3] or { 0 2 3 , , } . Setting aside the fact that our data associate each possession outcome X with process-valued inputs Z , and thus do not conform naturally to input/output structure of such models, such an approach cannot guarantee the estimator will have the (Kolmogorov) stochastic consistency inherent to theoretical EPV, which is essential to its 'stock ticker' interpretation. Using a stochastically consistent EPV estimator guarantees that changes in the resulting EPV curve derive from players' on-court actions, rather than artifacts or inefficiencies of the data analysis. A stochastic process model for the evolution of a basketball possession guarantees such consistency.
The second criterion that we require is that the estimator be sensitive to the fine-grained details of the data without incurring undue variance or computatonal complexity. Applying a Markov chain-based estimation approach would require discretizing the data by mapping the observed spatial configuration Z t into a simplified summary C t , violating this criterion by trading potentially useful information in the player tracking data for computational tractability.
To develop methodology that meet both criteria, we note the information-computation tradeoff in current process modeling strategies results from choosing a single level of resolution at which to model the possession and compute all expectations. In contrast, our method for estimating EPV combines models for the possession at two distinct levels of resolution, namely, a fully continuous model of player movement and actions, and a Markov chain model for a highly coarsened view of the possession. This multiresolution approach
2 The time of a possession is bounded, even for pathological examples, by the 12-minute length of a quarter; yet we do not leverage this fact and simply assume that possession lenghts are almost surely finite.
leverages the computational simplicity of a discrete Markov chain model while conditioning on exact spatial locations and high-resolution data features.
## 2.2 A Coarsened Process
The Markov chain portion of our method requires a coarsened view of the data. For all time 0 < t ≤ T during a possession, let C ( ) · be a coarsening that maps Z to a finite set C , and call C t = C Z ( t ) the 'state' of the possession. To make the Markovian assumption plausible, we populate the coarsened state space C with summaries of the full resolution data so that transitions between these states represent meaningful events in a basketball possession-see Figure 3 for an illustration.
Figure 3: Schematic of the coarsened possession process C , with states (rectangles) and possible state transitions (arrows) shown. The unshaded states in the first row compose C poss . Here, states corresponding to distinct ballhandlers are grouped together (Player 1 through 5), and the discretized court in each group represents the player's coarsened position and defended state. The gray shaded rectangles are transition states, C trans , while the rectangles in the third row represent the end states, C end . Blue arrows represent possible macrotransition entrances (and red arrows, macrotransition exits) when Player 1 has the ball; these terms are introduced in Section 3.
glyph[negationslash]
First, there are 3 'bookkeeping' states, denoted C end , that categorize the end of the possession, so that C T ∈ C end and for all t < T,C t ∈ C end (shown in the bottom row of Figure 3). These are C end = made 2 pt, made 3 pt, end of possession { } . These three states have associated point values of 2, 3, and 0, respectively (the generic possession end state can be reached by turnovers and defensive rebounds, which yield no points). This makes the possession point value X a function of the final coarsened state C T .
Next, whenever a player possesses the ball at time t , we assume C t = (ballcarrier ID at t ) × (court region at t ) × (defended at t ), having defined seven disjoint regions of the court and classifying a player as defended at time t by whether there is a defender within 5 feet of him. The possible values of C t , if a player possesses the ball at time t , thus live in C poss = { player ID } × { region ID } × { 1 [defended] } . These states are represented by the unshaded portion of the top row of Figure 3, where the differently colored regions of the court
diagrams reveal the court space discretization.
Finally, we define a set of states to indicate that an annotated basketball action from the full resolution data Z is currently in progress. These 'transition' states encapsulate constrained motifs in a possession, for example, when the ball is in the air traveling between players in a pass attempt. Explicitly, denote C trans = { shot attempt from c ∈ C poss , pass to c ′ ∈ C poss from c ∈ C poss , turnover in progress, rebound in progress } (listed in the gray shaded portions of Figure 3). These transition states carry information about the possession path, such as the most recent ballcarrier, and the target of the pass, while the ball is in the air during shot attempts and passes . Note that, by design, a possession must pass through 3 a state in C trans in order to reach a state in C end . For simplicity and due to limitations of the data, this construction of C = C poss ∪C trans ∪C end excludes several notable basketball events (such as fouls, violations, and other stoppages in play) and aggregates others (the data, for example, does not discriminate among steals, intercepted passes, or lost balls out of bounds).
## 2.3 Combining Resolutions
We make several modeling assumptions about the processes Z and C , which allow them to be combined into a coherent EPV estimator.
## (A1) C is marginally semi-Markov.
The semi-Markov assumption (A1) guarantees that the embedded sequence of disjoint possession states C (0) , C (1) , . . . , C ( K ) is a Markov chain, which ensures that it is straightforward to compute E [ X C | t ] using the associated transition probability matrix (Kemeny & Snell 1976).
Next, we specify the relationship between coarsened and full-resolution conditioning. This first requires defining two additional time points which mark changes in the future evolution of the possession:
glyph[negationslash]
$$\tau _ { t } = \begin{cases} \min \{ s \, \colon s > t, C _ { s } \in \mathcal { C } _ { \text{trans} } \} & \text{if } C _ { t } \in \mathcal { C } _ { \text{poss} } \\ t & \text{if } C _ { t } \not \in \mathcal { C } _ { \text{poss} } \end{cases}$$
glyph[negationslash]
$$\delta _ { t } = \min \{ s \, \colon s \geq \tau _ { t }, C _ { s } \not \in \mathcal { C } _ { \text{trans} } \}.$$
Thus, assuming a player possesses the ball at time t , τ t is the first time after t he attempts a shot/pass or turns the ball over (entering a state in C trans ), and δ t is the endpoint of this shot/pass/turnover (leaving a state in C trans ). We assume that passing through these transition states, C trans , decouples the future of the possession after time δ t with its history up to time t :
$$( A 2 ) \text{ For all } s > \delta _ { t } \text{ and } c \in \mathcal { C }, \, \mathbb { P } ( C _ { s } = c | C _ { \delta _ { t } }, \mathcal { F } _ { t } ^ { ( Z ) } ) = \mathbb { P } ( C _ { s } = c | C _ { \delta _ { t } } ).$$
Intuitively, assumption (A2) states that for predicting coarsened states beyond some point in the future δ t , all information in the possession history up to time t is summarized by the distribution of C δ t . The dynamics of basketball make this assumption reasonable; when a player passes the ball or attempts a shot, this represents a structural transition in the basketball possession to which all players react. Their actions prior to this transition are
3 The reason we index transition states by the origin of the pass/shot attempt (and destination of the pass) is to preserve this information under a Markov assumption, where generic 'pass' or 'shot' states would inappropriately allow future states to be independent of the players involved in the shot or pass.
not likely to influence their actions after this transition. Given C δ t -which, for a pass at τ t includes the pass recipient, his court region, and defensive pressure, and for a shot attempt at τ t includes the shot outcome-data prior to the pass/shot attempt are not informative of the possession's future evolution.
Together, these assumptions yield a simplified expression for (1), which combines contributions from full-resolution and coarsened views of the process.
Theorem 2.1 Under assumptions (A1)-(A2), the full-resolution EPV ν t can be rewritten:
$$\nu _ { t } = \sum _ { c \in \mathcal { C } } \mathbb { E } [ X | C _ { \delta _ { t } } = c ] \mathbb { P } ( C _ { \delta _ { t } } = c | \mathcal { F } _ { t } ^ { ( Z ) } ).$$
Remark Although we have specified this result in terms of the specific coarsening defined in Section 2.2, we could substitute any coarsening for which (A1)-(A2) are well-defined and reasonably hold. We briefly discuss potential alternative coarsenings in Section 6.
The proof of Theorem 2.1, follows immediately from (A1)-(A2), and is therefore omitted. Heuristically, (4) expresses ν t as the expectation given by a homogeneous Markov chain on C with a random starting point C δ t , where only the starting point depends on the fullresolution information F ( Z ) t . This result illustrates the multiresolution conditioning scheme that makes our EPV approach computationally feasible: the term E [ X C | δ t = c ] is easy to calculate using properties of Markov chains, and P ( C δ t |F ( Z ) t ) only requires forecasting the full-resolution data for a short period of time relative to (1), as δ t ≤ T .
## 3. TRANSITION MODEL SPECIFICATION
The representation in Theorem (2.1) shows that estimating EPV does not require a full-blown model for entire basketball possessions at high resolution. Instead, the priority is to accurately predict the next major 'decoupling' action in the possession, which we have denoted C δ t . At this point, Equation (4) switches resolutions: C δ t depends on the full-resolution possession history F ( Z ) t , after which our EPV estimate only depends on the coarsened state C δ t . This section presents models that operate on these two distinct levels of resolution, using parameterizations that reflect players' reactions to the situational, spatiotemporal predicaments they face on the basketball court.
First, using the full-resolution data, we need to predict C δ t . We achieve this using three models; heuristically speaking, one predicts player movement in space while the ballcarrier remains constant, one predicts the occurrence of events (passes/shots/turnovers) that change the ballcarrier, and one predicts the outcome state ( C δ t ) of such events.
Writing these models requires some additional notation. Fix glyph[epsilon1] > 0, and for any t ≥ 0 during the possession (here, we use 1/25 second since this is the temporal resolution of our data), and let M t ( ) be the event { τ t ≤ t + glyph[epsilon1] } ; for this, we say that a macrotransition occurs during ( t, t + ]. Recalling the definition of glyph[epsilon1] τ t (2), M t ( ) is realized when the possession moves from C poss to C trans , which represents the start of a pass, shot attempt, or turnover (and M t ( ) is continuously realized throughout the duration of this action). We now define:
(M1) The microtransition model , P ( Z t + glyph[epsilon1] | M t ( ) c , F ( Z ) t ), which describes infinitesimal player movement assuming that a major ball movement does not occur.
(M2) The macrotransition entry model , P ( M t ( ) |F ( Z ) t ), which describes the occurrence of a macrotransition (pass/shot/turnover) within the next glyph[epsilon1] time.
(M3) The macrotransition exit model P ( C δ t | M t , ( ) F ( Z ) t ), which gives the outcome of this macrotransition in C .
(M1)-(M3) together allow us to sample from P ( C δ t |F ( Z ) t ), as we alternate draws from (M1) and (M2) until a macrotransition occurs, and then use (M3) to sample the outcome state of this macrotransition. Thus, while models (M1)-(M3) condition on the full-resolution possession history F ( Z ) t in order to predict C δ t , calculating EPV given C δ t only requires a model for transitions between coarsened states C . Due to the Markov assumption (A1), this is easily summarized by a transition probability matrix:
(M4) The Markov transition probability matrix P , with P qr = P ( C ( n +1) = c r | C ( n ) = c q ).
Thus, (M1)-(M4) are sufficient to compute EPV using our multiresolution framework of Theorem 2.1. In the following subsections, we discuss each of these models in greater detail.
## 3.1 Microtransition Model
The microtransition model describes player movement with the ballcarrier held constant. In the periods between transfers of ball possession (including passes, shots, and turnovers), all players on the court move in order to influence the character of the next ball movement (macrotransition). For instance, the ballcarrier might drive toward the basket to attempt a shot, or move laterally to gain separation from a defender, while his teammates move to position themselves for passes or rebounds, or to set screens and picks. The defense moves correspondingly, attempting to deter easy shot attempts or passes to certain players while simultaneously anticipating a possible turnover. Separate models are assumed for offensive and defensive players, as we shall describe.
Predicting the motion of offensive players over a short time window is driven by the players' dynamics (velocity, acceleration, etc.). Let the location of offensive player glyph[lscript] ( glyph[lscript] ∈ { 1 , . . . , L = 461 ) at time } t be z glyph[lscript] ( ) = ( t x glyph[lscript] ( ) t , y glyph[lscript] ( )). t We then model movement in each of the x and y coordinates using
$$x ^ { \ell } ( t + \epsilon ) = x ^ { \ell } ( t ) + \alpha _ { x } ^ { \ell } [ x ^ { \ell } ( t ) - x ^ { \ell } ( t - \epsilon ) ] + \eta _ { x } ^ { \ell } ( t )$$
(and analogously for y glyph[lscript] ( )). t This expression derives from a Taylor series expansion to the ballcarrier's position for each coordinate, such that α glyph[lscript] x [ x glyph[lscript] ( ) t -x glyph[lscript] ( t -glyph[epsilon1] )] ≈ glyph[epsilon1]x glyph[lscript] ′ ( ), and t η glyph[lscript] x ( ) t provides stochastic innovations representing the contribution of higher-order derivatives (acceleration, jerk, etc.). Because they are driven to score, players' dynamics on offense are nonstationary. When possessing the ball, most players accelerate toward the basket when beyond the three-point line, and decelerate when very close to the basket in order to attempt a shot. Also, players will accelerate away from the edges of the court as they approach these, in order to stay in bounds. To capture such behavior, we assume spatial structure for the innovations, η glyph[lscript] x ( ) t ∼ N ( µ glyph[lscript] x ( z glyph[lscript] ( )) t , ( σ glyph[lscript] x ) 2 ), where µ glyph[lscript] x maps player glyph[lscript] 's location on the court to an additive effect in (5), which has the interpretation of an acceleration effect; see Figure 4 for an example.
The defensive components of P ( Z t + glyph[epsilon1] | M t ( ) c , F ( Z ) t ), corresponding to the positions of the five defenders, are easier to model conditional on the evolution of the offense's positions.
Figure 4: Acceleration fields ( µ x ( z ( )) t , µ y ( z ( ))) t for Tony Paker (a)-(b) and Dwight Howard (c)-(d) with and without ball possession. The arrows point in the direction of the acceleration at each point on the court's surface, and the size and color of the arrows are proportional to the magnitude of the acceleration. Comparing (a) and (c) for instance, we see that when both players possess the ball, Parker more frequently attacks the basket from outside the perimeter. Howard does not accelerate to the basket from beyond the perimeter, and only tends to attack the basket inside the paint.
Following Franks, Miller, Bornn & Goldsberry (2015), we assume each defender's position is centered on a linear combination of the basket's location, the ball's location, and the location of the offensive player he is guarding. Franks et al. (2015) use a hidden Markov model (HMM) based on this assumption to learn which offensive players each defender is guarding, such that conditional on defender glyph[lscript] guarding offender k his location z glyph[lscript] ( ) = ( t x glyph[lscript] ( ) t , y glyph[lscript] ( )) should t be normally distributed with mean m k opt ( ) = 0 62 t . z k ( ) + 0 11 t . z bask +0 27 . z ball ( ). t
Of course, the dynamics (velocity, etc.) of defensive players' are still hugely informative for predicting their locations within a small time window. Thus our microtransition model for defender glyph[lscript] balances these dynamics with the mean path induced by the player he is guarding:
$$x ^ { \ell } ( t + \epsilon ) & | m _ { \text{opt}, x } ^ { k } ( t ) \sim \mathcal { N } & \left ( x ^ { \ell } ( t ) + a _ { x } ^ { \ell } [ x ^ { \ell } ( t ) - x ^ { \ell } ( t - \epsilon ) ] \\ & + b _ { x } ^ { \ell } [ m _ { \text{opt}, x } ^ { k } ( t + \epsilon ) - m _ { \text{opt}, x } ^ { k } ( t ) ] + c _ { x } ^ { \ell } [ x ^ { \ell } ( t ) - m _ { \text{opt}, x } ^ { k } ( t + \epsilon ) ], ( \tau _ { x } ^ { \ell } ) ^ { 2 } \right )$$
and symmetrically in y . Rather than implement the HMM procedure used in Franks et al. (2015), we simply assume each defender is guarding at time t whichever offensive player j yields the smallest residual || z glyph[lscript] ( ) t -m j opt ( ) t || , noting that more than one defender may be guarding the same offender (as in a 'double team'). Thus, conditional on the locations of the offense at time t + , (6) provides a distribution over the locations of the defense at glyph[epsilon1] t + . glyph[epsilon1]
## 3.2 Macrotransition Entry Model
The macrotransition entry model P ( M t ( ) |F ( Z ) t ) predicts ball movements that instantaneously shift the course of the possession-passes, shot attempts, and turnovers. As such, we consider a family of macrotransition entry models P ( M t j ( ) |F ( Z ) t ), where j indexes the type of macrotransition corresponding to M t ( ). There are six such types: four pass options (indexed, without loss of generality, j ≤ 4), a shot attempt ( j = 5), or a turnover ( j = 6). Thus, M t j ( ) is the event that a macrotransition of type j begins in the time window
( t, t + ], and glyph[epsilon1] M t ( ) = ⋃ 6 j =1 M t j ( ). Since macrotransition types are disjoint, we also know P ( M t ( ) |F ( Z ) t ) = ∑ 6 j =1 P ( M t j ( ) |F ( Z ) t ).
Weparameterize the macrotransition entry models as competing risks (Prentice, Kalbfleisch, Peterson Jr, Flournoy, Farewell & Breslow 1978): assuming player glyph[lscript] possesses the ball at time t > 0 during a possession, denote
$$\lambda _ { j } ^ { \ell } ( t ) = \lim _ { \epsilon \rightarrow 0 } \frac { \mathbb { P } ( M _ { j } ( t ) | \mathcal { F } _ { t } ^ { ( Z ) } ) } { \epsilon }$$
as the hazard for macrotransition j at time t . We assume these are log-linear,
$$\log ( \lambda _ { j } ^ { \ell } ( t ) ) = [ \mathbf W _ { j } ^ { \ell } ( t ) ] ^ { \prime } \beta _ { j } ^ { \ell } + \xi _ { j } ^ { \ell } \left ( \mathbf z ^ { \ell } ( t ) \right ) + \left ( \tilde { \xi } _ { j } ^ { \ell } \left ( \mathbf z _ { j } ( t ) \right ) \mathbf 1 [ j \leq 4 ] \right ),$$
where W glyph[lscript] j ( ) is a t p j × 1 vector of time-varying covariates, β glyph[lscript] j a p j × 1 vector of coefficients, z glyph[lscript] ( ) is the ballcarrier's 2D location on the court (denote the court space t S ) at time t , and ξ glyph[lscript] j : S → R is a mapping of the player's court location to an additive effect on the log-hazard, providing spatial variation. The last term in (8) only appears for pass events ( j ≤ 4) to incorporate the location of the receiving player for the corresponding pass: z j ( ) t (which slightly abuses notation) provides his location on the court at time t , and ˜ , ξ glyph[lscript] j analogously to ξ glyph[lscript] j , maps this location to an additive effect on the log-hazard. The four different passing options are identified by the (basketball) position of the potential pass recipient: point guard (PG), shooting guard (SG), small forward (SF), power forward (PF), and center (C).
The macrotransition model (7)-(8) represents the ballcarrier's decision-making process as an interpretable function of the unique basketball predicaments he faces. For example, in considering the hazard of a shot attempt, the time-varying covariates ( W glyph[lscript] j ( )) t we use are the distance between the ballcarrier and his nearest defender (transformed as log(1 + d ) to moderate the influence of extremely large or small observed distances), an indicator for whether the ballcarrier has dribbled since gaining possession, and a constant representing a baseline shooting rate (this is not time-varying) . 4 The spatial effects ξ glyph[lscript] j reveal locations where player glyph[lscript] is more/less likely to attempt a shot in a small time window, holding fixed the time-varying covariates W glyph[lscript] j ( ). t Such spatial effects (illustrated in Figure 5) are well-known to be nonlinear in distance from the basket and asymmetric about the angle to the basket (Miller, Bornn, Adams & Goldsberry 2013).
All model components-the time-varying covariates, their coefficients, and the spatial effects ξ, ξ ˜ differ across macrotransition types j for the same ballcarrier glyph[lscript] , as well as across all L = 461 ballcarriers in the league during the 2013-14 season. This reflects the fact that players' decision-making tendencies and skills are unique; a player such as Dwight Howard will very rarely attempt a three point shot even if he is completely undefended, while someone like Stephen Curry will attempt a three point shot even when closely defended.
4 Full details on all covariates used for all macrotransition types are included in Appendix A.1
(d) ξ 6 (turnover)
Figure 5: Plots of estimated spatial effects ξ for LeBron James as the ballcarrier. For instance, plot (c) reveals the largest effect on James' shot-taking hazard occurs near the basket, with noticeable positive effects also around the three point line (particularly in the 'corner 3' shot areas). Plot (a) shows that he is more likely (per unit time) to pass to the point guard when at the top of the arc-more so when the point guard is positioned in the post area.
## 3.3 Macrotransition Exit Model
Using the six macrotransition types introduced in the previous subsection, we can express the macrotransition exit model (M3) when player glyph[lscript] has possession as
$$\mathbb { P } ( C _ { \delta _ { t } } | M ( t ), \mathcal { F } _ { t } ^ { ( Z ) } ) & = \sum _ { j = 1 } ^ { 6 } \mathbb { P } ( C _ { \delta _ { t } } | M _ { j } ( t ), \mathcal { F } _ { t } ^ { ( Z ) } ) \mathbb { P } ( M _ { j } ( t ) | M ( t ), \mathcal { F } _ { t } ^ { ( Z ) } ) \\ & = \sum _ { j = 1 } ^ { 6 } \mathbb { P } ( C _ { \delta _ { t } } | M _ { j } ( t ), \mathcal { F } _ { t } ^ { ( Z ) } ) \frac { \lambda _ { j } ^ { \ell } ( t ) } { \sum _ { k = 1 } ^ { 6 } \lambda _ { k } ^ { \ell } ( t ) },$$
using the competing risks model for M t j ( ) given by (7)-(8). As terms λ glyph[lscript] j ( ) are supplied by t (8), we focus on the macrotransition exit model conditional on the macrotransition type, P ( C δ t | M t , j ( ) F ( Z ) t ).
For each j = 1 , . . . , 4, M t j ( ) represents a pass-type macrotransition, therefore C δ t is a possession state c ′ ∈ C poss for the player corresponding to pass option j . Thus, a model for P ( C δ t | M t , j ( ) F ( Z ) t ) requires us to predict the state c ′ ∈ C poss the j th pass target will occupy upon receiving the ball. Our approach is to simply assume c ′ is given by the pass target's location at the time the pass begins. While this is naive and could be improved by further modeling, it is a reasonable approximation in practice, because with only seven court regions and two defensive spacings comprising C poss , the pass recipient's position in this space is unlikely to change during the time the pass is traveling en route, δ t -t (a noteable exception is the alley-oop pass, which leads the pass recipient from well outside the basket to a dunk or layup within the restricted area). Our approach thus collapses P ( C δ t | M t , j ( ) F ( Z ) t ) to a single state in C poss , which corresponds to pass target j 's location at time t .
When j = 5, and a shot attempt occurs in ( t, t + ], glyph[epsilon1] C δ t is either a made/missed 2 point shot, or made/missed three point shot. For sufficiently small glyph[epsilon1] , we observe at Z t whether the shot attempt in ( t, t + ] is a two- or three-point shot, therefore our task in providing glyph[epsilon1]
P ( C δ t | M t , j ( ) F ( Z ) t ) is modeling the shot's probability of success. We provide a parametric shot probability model, which shares the same form as the macrotransition entry model (7)(8), though we use a logit link function as we are modeling a probability instead of a hazard. Specifically, for player glyph[lscript] attempting a shot at time t , let p glyph[lscript] ( ) t represent the probability of the shot attempt being successful (resulting in a basket). We assume
$$\log i t ( p ^ { \ell } ( t ) ) = [ \mathbf W _ { s } ^ { \ell } ( t ) ] ^ { \prime } \beta _ { s } ^ { \ell } + \xi _ { s } ^ { \ell } ( \mathbf z ^ { \ell } ( t ) )$$
with components in (10) having the same interpretation as their j -indexed counterparts in the competing risks model (8); that is, W glyph[lscript] s is a vector of time-varying covariates (we use distance to the nearest defender-transformed as log(1 + d )-an indicator for whether the player has dribbled, and a constant to capture baseline shooting efficiency) with β glyph[lscript] s a corresponding vector of coefficients, and ξ glyph[lscript] s a smooth spatial effect, as in (8).
Lastly, when j = 6 and M t j ( ) represents a turnover, C δ t is equal to the turnover state in C end with probability 1.
Note that the macrotransition exit model is mostly trivial when no player has ball possession at time t , since this implies C t ∈ C trans ∪ C end and τ t = t . If C t ∈ C end , then the possession is over and C δ t = C t . Otherwise, if C t ∈ C trans represents a pass attempt or turnover in progress, the following state C δ t is deterministic given C t (recall that the pass recipient and his location are encoded in the definition of pass attempt states in C trans ). When C t represents a shot attempt in progress, the macrotransition exit model reduces to the shot probability model (10). Finally, when C t is a rebound in progress, we ignore full-resolution information and simply use the Markov transition probabilities from P 5 .
## 3.4 Transition Probability Matrix for Coarsened Process
The last model necessary for calculating EPV is (M4), the transition probability matrix for the embedded Markov chain corresponding to the coarsened process C (0) , C (2) , . . . , C ( K ) . This transition probability matrix is used to compute the term E [ X C | δ t = c ] that appears in Theorem 2.1. Recall that we denote the transition probability matrix as P , where P qr = P ( C ( +1) i = c r | C ( ) i = c q ) for any c , c q r ∈ C .
Without any other probabilistic structure assumed for C ( ) i other than Markov, for all q, r , the maximum likelihood estimator of P qr is the observed transition frequency, ˆ P qr = N qr ∑ r ′ N qr ′ , where N qr counts the number of transitions c q → c r . Of course, this estimator has undesirable performance if the number of visits to any particular state c q is small (for instance, Dwight Howard closely defended in the corner 3 region), as the estimated transition probabilities from that state may be degenerate.
Under our multiresolution model for basketball possessions, however, expected transition counts between many coarsened states C ( ) i can be computed as summaries of our macrotransition models (M2)-(M3). To show this, for any arbitrary t > 0 let M t r j ( ) be the event
$$M _ { j } ^ { r } ( t ) = \{ \mathbb { P } ( M _ { j } ( t ) \text{ and } C _ { t + \epsilon } = c _ { r } | \mathcal { F } _ { t } ^ { ( Z ) } ) > 0 \}.$$
Thus M t r j ( ) occurs if it is possible for a macrotransition of type j into state c r to occur in
5 Our rebounding model could be improved by using full-resolution spatiotemporal information, as players' reactions to the missed shot event are informative of who obtains the rebound.
( t, t + ]. When applicable, we can use this to get the expected number of glyph[epsilon1] c q → c r transitions:
$$\tilde { N } _ { q r } = \epsilon \sum _ { t \colon C _ { t } = c _ { q } } \lambda _ { j } ^ { \ell } ( t ) \mathbf 1 [ M _ { j } ^ { r } ( t ) ].$$
When c q is a shot attempt state from c q ′ ∈ C poss , (11) is adjusted using the shot probability model (10): ˜ N qr = glyph[epsilon1] ∑ t C : t = c q ′ λ glyph[lscript] j ( ) t p t ( ) 1 [ M t r j ( )] when c r represents an eventual made shot and ˜ N qr = glyph[epsilon1] ∑ t C : t = c q ′ λ glyph[lscript] j ( )(1 t -p t ( )) 1 [ M t r j ( )] when c r represents an eventual miss.
By replacing raw counts with their expectations conditional on higher-resolution data, leveraging the hazards (11) provides a Rao-Blackwellized (unbiased, lower variance) alternative to counting observed transitions. Furthermore, due to the hierarchical parameterization of hazards λ glyph[lscript] j ( ) (discussed in Section 4), information is shared across space and player comt binations so that estimated hazards are well-behaved even in situations without significant observed data. Thus, when c q → c r represents a macrotransition, we use ˜ N qr in place of N qr when calculating ˆ P qr .
## 4. HIERARCHICAL MODELING AND INFERENCE
A critical aspect of the micro- and macrotransition models defined in the previous section is that they are parameterized to capture the variations between actions, players, and court space that play a central role in basketball strategy. This section outlines the procedure for reliably estimating this rich set of model parameters using likelihood-based methods.
Hierarchical models are essential for our problem because by implicitly averaging over all possible future possession paths, calculating EPV requires transition probabilities for situations for which there is no data. For instance, DeAndre Jordan did not attempt a three-point shot in the 2013-14 season, yet any EPV estimate for a possession with him on the court requires an estimate of his shooting ability from everywhere on the court, even though for some of these regions it is unlikely he would attempt a shot. Hierarchical models combine information both across space and across different players to estimate such probabilities.
## 4.1 Conditional Autoregressive Prior for Player-Specific Coefficients
Sharing information between players is critical for our estimation problem, but standard hierarchical models encode an assumption of exchangeability between units that is too strong for NBA players, even between those who are classified by the league as playing the same position. For instance, LeBron James is listed at the same position (small forward) as Steve Novak, despite the fact that James is one of the NBA's most prolific short-range scorers whereas Novak has not scored a layup since 2012. To model between-player variation more realistically, our hierarchical model shares information across players based on a localized notion of player similarity that we represent as an L × L binary adjacency matrix H : H glyph[lscript]k = 1 if players glyph[lscript] and k are similar to each other and H glyph[lscript]k = 0 otherwise. We determine similarity in a pre-processing step that compares the spatial distribution of where players spend time on the offensive half-court; see Appendix A.2 for exact details on specifying H .
Now let β glyph[lscript] ji be the i th component of β glyph[lscript] j , the vector of coefficients for the time-referenced covariates for player glyph[lscript] 's hazard j (8). Also let β ji be the vector representing this component across all L = 461 players, ( β 1 ji β 2 ji . . . β L ji ) . ′ We assume independent conditional autogressive
(CAR) priors (Besag 1974) for β ji :
$$\beta _ { j i } ^ { \ell } | \beta _ { j i } ^ { ( - \ell ) }, \tau _ { \beta _ { j i } } ^ { 2 } & \sim \mathcal { N } \left ( \frac { 1 } { n _ { \ell } } \sum _ { k \colon H _ { \ell } = 1 } \beta _ { j i } ^ { k }, \frac { \tau _ { \beta _ { j i } } ^ { 2 } } { n _ { \ell } } \right ) \\ \tau _ { \beta _ { j i } } ^ { 2 } & \sim \text{InvGamma} ( 1, 1 )$$
where n glyph[lscript] = ∑ k H glyph[lscript]k . Similarly, let β s i = ( β 1 s i β 2 s i . . . β L s i ) be the vector of the i th component of the shot probability model (10) across players 1 , . . . , L . We assume the same CAR prior (12) independently for each component i . While the inverse gamma prior for τ 2 ∗ terms seems very informative, we want to avoid very large or small values of τ 2 ∗ , corresponding to 0 or full shrinkage (respectively), which we know are inappropriate for our model. Predictive performance for the 0 shrinkage model ( τ 2 ∗ very large) is shown in Table 1, whereas the full shrinkage model ( τ 2 ∗ = 0) doesn't allow parameters to differ by player identity, which precludes many of the inferences EPV was designed for.
## 4.2 Spatial Effects ξ
Player-tracking data is a breakthrough because it allows us to model the fundamental spatial component of basketball. In our models, we incorporate the properties of court space in spatial effects ξ , ξ glyph[lscript] j ˜ glyph[lscript] j , ξ glyph[lscript] s , which are unknown real-valued functions on S , and therefore infinite dimensional. We represent such spatial effects using Gaussian processes (see Rasmussen (2006) for an overview of modeling aspects of Gaussian processes). Gaussian processes are usually specified by a mean function and covariance function; this approach is computationally intractable for large data sets, as the computation cost of inference and interpolating the surface at unobserved locations is O ( n 3 ), where n is the number of different points at which ξ glyph[lscript] j is observed (for many spatial effects ξ glyph[lscript] j , the corresponding n would be in the hundreds of thousands). We instead provide ξ with a low-dimensional representation using functional bases (Higdon 2002; Qui˜onero-Candela & Rasmussen 2005), which offers three important n advantages. First, this representation is more computationally efficient for large data sets such as ours. Second, functional bases allow for a non-stationary covariance structure that reflects unique spatial dependence patterns on the basketball court. Finally, the finite basis representation allows us to apply the same between-player CAR prior to estimate each player's spatial effects.
Our functional basis representation of a Gaussian process ξ glyph[lscript] j relies on d deterministic basis functions φ j 1 , . . . , φ jd : S → R such that for any z ∈ S ,
$$\xi _ { j } ^ { \ell } ( { \mathbf z } ) = \sum _ { i = 1 } ^ { d } w _ { j i } ^ { \ell } \phi _ { j i } ( { \mathbf z } ),$$
where w glyph[lscript] j = ( w glyph[lscript] j 1 . . . w glyph[lscript] jd ) ′ is a random vector of loadings, w glyph[lscript] j ∼ N ( ω glyph[lscript] j , Σ glyph[lscript] j ). Letting Φ ( j z ) = ( φ j 1 ( z ) . . . φ jd ( z )) , ′ we can see that ξ glyph[lscript] j given by (13) is a Gaussian process with mean function Φ ( j z ) ′ ω glyph[lscript] j and covariance function Cov[ ξ glyph[lscript] j ( z 1 ) , ξ glyph[lscript] j ( z 2 )] = Φ ( j z 1 ) ′ Σ glyph[lscript] j Φ ( j z 2 ). However, since bases φ ji are deterministic, each ξ glyph[lscript] j is represented as a d -dimensional parameter. Note that we also use (13) for pass receiver spatial effects and the spatial effect term in the shot probability model, ˜ ξ glyph[lscript] j and ξ glyph[lscript] s , respectively. For these terms we have associated bases ˜ φ ji , φ s i and weights, ˜ w ,w glyph[lscript] ji glyph[lscript] s i . As our notation indicates, bases functions Φ ( j z ) differ for each
Figure 6: The functional bases φ ji for i = 1 , . . . , 10 and j corresponding to the shot-taking macrotransition, j = 5 . There is no statistical interpretation of the ordering of the bases; we have displayed them in rough order of the shot types represented, from close-range to long-range.
macrotransition type but are constant across players; whereas weight vectors w glyph[lscript] j vary across both macrotransition types and players.
Using d = 10, we determine the functional bases in a pre-processing step, discussed in Appendix A.3. These basis functions are interpretable as patterns/motifs that constitute players' decision-making tendencies as a function of space; please see Figure 6 for an example, or Miller et al. (2013) for related work in a basketball context. Furthermore, we use a CAR model (12) to supply the prior mean and covariance matrix ( ω glyph[lscript] j , Σ glyph[lscript] j ) for the weights:
$$w _ { j } ^ { \ell } | w _ { j } ^ { - ( \ell ) }, \tau _ { \mathbf w _ { j } } ^ { 2 } & \sim \mathcal { N } \left ( \frac { 1 } { n _ { \ell } } \sum _ { k \colon H _ { \ell k } = 1 } \mathbf w _ { j } ^ { k }, \frac { \tau _ { \mathbf w _ { j } } ^ { 2 } } { n _ { \ell } } \mathbf I _ { d } \right ) \\ \tau _ { \mathbf w _ { j } } ^ { 2 } & \sim \ln \text{YGamma} ( 1, 1 ).$$
As with (12), we also use (14) for terms ˜ w j and w s . Combining the functional basis representation (13) with the between-player CAR prior (14) for the weights, we get a prior representation for spatial effects ξ , ξ glyph[lscript] j ˜ glyph[lscript] j , ξ ˜ glyph[lscript] that is low-dimensional and shares information both across space and between different players.
## 4.3 Parameter Estimation
As discussed in Section 3, calculating EPV requires the parameters that define the multiresolution transition models (M1)-(M4)-specifically, the hazards λ glyph[lscript] j , shot probabilities p glyph[lscript] , and all parameters of the microtransition model (M1). We estimate these parameters in a Bayesian fashion, combining the likelihood of the observed optical tracking data with the prior structure discussed earlier in this section. Using our multiresolution models, we can
write the likelihood for the full optical tracking data, indexed arbitrarily by t :
$$& \text{write the likelihood for the full optical tracking data, indexed arbitrarily by t: } \\ & \prod _ { t } \mathbb { P } ( Z _ { t + \epsilon } | \mathcal { F } _ { t } ^ { ( Z ) } ) = \left ( \prod _ { t } \overbrace { \mathbb { P } ( Z _ { t + \epsilon } | M ( t ) ^ { c }, \mathcal { F } _ { t } ^ { ( Z ) } ) ^ { 1 [ M ( t ) ^ { c } ] } } ^ { L _ { \text{mic} } } \right ) \overbrace { \prod _ { t } \quad j = 1 } ^ { 6 } \mathbb { P } ( Z _ { t + \epsilon } | M _ { j } ( t ), C _ { \delta t }, \mathcal { F } _ { t } ^ { ( Z ) } ) ^ { 1 [ M _ { j } ( t ) ] } } ^ { L _ { \text{rem} } } \\ & \quad \times \left ( \underbrace { \prod _ { t } \mathbb { P } ( M ( t ) ^ { c } | \mathcal { F } _ { t } ^ { ( Z ) } ) ^ { 1 [ M ( t ) ^ { c } ] } } ^ { 6 } \prod _ { j = 1 } ^ { 6 } \mathbb { P } ( M _ { j } ( t ) | \mathcal { F } _ { t } ^ { ( Z ) } ) ^ { 1 [ M _ { j } ( t ) ] } } _ { L _ { \text{entry} } } \right ) \left ( \underbrace { \prod _ { t } \quad j = 1 } ^ { 6 } \mathbb { P } ( C _ { \delta t } | M _ { j } ( t ), \mathcal { F } _ { t } ^ { ( Z ) } ) ^ { 1 [ M _ { j } ( t ) ] } } _ { L _ { \text{exist} } } \right ), \\ & \quad \text{The factorization used in (15) \, \text{hiylibits data features that inform different narrower programs: }$$
The factorization used in (15) highlights data features that inform different parameter groups: L mic is the likelihood term corresponding to the microtransition model (M1), L entry the macrotransition entry model (M2), and L exit the macrotransition exit model (M3). The remaining term L rem is left unspecified, and ignored during inference. Thus, L mic , L entry , and L exit can be though of as partial likelihoods (Cox 1975 b ), which under mild conditions leads to consistent and asymptotically well-behaved estimators (Wong 1986). When parameters in these partial likelihood terms are given prior distributions, as is the case for those comprising the hazards in the macrotransition entry model, as well as those in the shot probability model, the resulting inference is partially Bayesian (Cox 1975 a ).
The microtransition partial likelihood term L mic factors by player:
$$L _ { \text{mic} } \, \infty \, \prod _ { t } \prod _ { \ell = 1 } ^ { L } \mathbb { P } ( \mathbf z _ { \ell } ( t + \epsilon ) | M ( t ) ^ { c }, \mathcal { F } _ { t } ^ { ( Z ) } ) ^ { 1 [ M ( t ) ^ { c } \, \text{and} \, \ell \, \text{on court at time } t ] }.$$
Depending on whether or not player glyph[lscript] is on offense (handling the ball or not) or defense, P ( z glyph[lscript] ( + ) t glyph[epsilon1] | M t ( ) c , F ( Z ) t ) is supplied by the offensive (5) or defensive(6) microtransition models. Parameters for these models (5)-(6) are estimated using R-INLA, where spatial acceleration effects µ , µ glyph[lscript] x glyph[lscript] y are represented using a Gaussian Markov random field approximation to a Gaussian process with Mat´rn covariance (Lindgren, Rue & Lindstr¨m 2011). Appendix A.3 e o provides the details on this approximation. We do perform any hierarchical modeling for the parameters of the microtransition model-because this model only describes movement (not decision-making), the data for every player is informative enough to provide precise inference. Thus, microtransition models are fit in parallel using each player's data separately; this requires L = 461 processors, each taking at most 18 hours at 2.50Ghz clock speed, using 32GB of RAM.
For the macrotransition entry term, we can write
$$L _ { \text{entry} } \in \prod _ { l = 1 } ^ { L } \prod _ { j = 1 } ^ { 6 } L _ { \text{entry} _ { j } } ^ { \ell } ( \lambda _ { j } ^ { \ell } ( \cdot ) ),$$
recognizing that (for small glyph[epsilon1] ),
$$\begin{matrix} l & & \\ & & \\ & & \end{matrix}$$
$$\text{where}$$
$$L _ { \text{entry} _ { j } } ^ { \ell } ( \lambda _ { j } ^ { \ell } ( \cdot ) ) & \in \left ( \prod _ { \substack { t \colon M _ { j } ( t ) \\ t \in \mathcal { T } ^ { \ell } } } \lambda _ { j } ^ { \ell } ( t ) \right ) \exp \left ( - \sum _ { t \in \mathcal { T } ^ { \ell } } \lambda _ { j } ^ { \ell } ( t ) \right ) \\ \text{where} \quad \log ( \lambda _ { j } ^ { \ell } ( t ) ) & = [ \mathbf W _ { j } ^ { \ell } ( t ) ] ^ { \prime } \beta _ { j } ^ { \ell } + \phi _ { j } ( \mathbf z _ { \ell } ( t ) ) ^ { \prime } \mathbf w _ { j } ^ { \ell } + \left ( \tilde { \phi } _ { j } ( \mathbf z _ { j } ( t ) ) ^ { \prime } \tilde { \mathbf w } _ { j } ^ { \ell } 1 [ j \leq 4 ] \right )$$
and T glyph[lscript] is the set of time t for which player glyph[lscript] possesses the ball. Expression (18) is the likelihood for a Poisson regression; combined with prior distributions (12)-(14), inference for β glyph[lscript] j , w w glyph[lscript] j , ˜ glyph[lscript] j is thus given by a hierarchical Poisson regression. However, the size of our data makes implementing such a regression model computationally difficult as the design matrix would have 30.4 million rows and a minimum of L p ( j + d ) ≥ 5993 columns, depending on macrotransition type. We perform this regression through the use of integrated nested Laplace approximations (INLA) (Rue, Martino & Chopin 2009). Each macrotransition type can be fit separately, and requires approximately 24 hours using a single 2.50GHz processor with 120GB of RAM.
Recalling Section 3.3, the macrotransition exit model (M3) is deterministic for all macrotransitions except shot attempts ( j = 5). Thus, L exit only provides information on the parameters of our shot probability model (10). Analogous to the Poisson model in (18), L exit is the likelihood of a logistic regression, which factors by player. We also use INLA to fit this hierarchical logistic regression model, though fewer computational resources are required as this likelihood only depends on time points where a shot is attempted, which is a much smaller subset of our data.
## 5. RESULTS
After obtaining parameter estimates for the multresolution transition models, we can calculate EPV using Theorem 2.1 and plot ν t throughout the course of any possession in our data. We view such (estimated) EPV curves as the main contribution of our work, and their behavior and potential inferential value has been introduced in Section 1. We illustrate this value by revisiting the possession highlighted in Figure 1 through the lens of EPV. Analysts may also find meaningful aggregations of EPV curves that summarize players' behavior over a possession, game, or season in terms of EPV. We offer two such aggregations in this section.
## 5.1 Predictive Performance of EPV
Before analyzing EPV estimates, it is essential to check that such estimates are properly calibrated (Gneiting, Balabdaoui & Raftery 2007) and accurate enough to be useful to basketball analysts. Our paper introduces EPV, and as such there are no existing results to benchmark the predictive performance of our estimates. We can, however, compare the proposed implementation for estimating EPV with simpler models, based on lower resolution information, to verify whether our multiresolution model captures meaningful features of our data. Assessing the predictive performance of an EPV estimator is difficult because the estimand is a curve whose length varies by possession. Moreover, we never observe any portion of this curve; we only know its endpoint. Therefore, rather than comparing estimated EPV curves between our method and alternative methods, we compare estimated transition prob-
abilities. For any EPV estimator method that is stochastically consistent, if the predicted transitions are properly calibrated, then the derived EPV estimates should be as well.
For the inference procedure in Section 4, we use only 90% of our data set for parameter inference, with the remaining 10% used to evaluate the out-of-sample performance of our model. We also evaluated out-of-sample performance of simpler macrotransition entry/exit models, which use varying amounts of information from the data. Table 1 provides the out-ofsample log-likelihood for the macrotransition models applied to the 10% of the data not used in model fitting for various simplified models. In particular, we start with the simple model employing constant hazards for each player/event type, then successively add situational covariates, spatial information, then full hierarchical priors. Without any shrinkage, our full model performs in some cases worse than a model with no spatial effects included, but with shrinkage, it consistently performs the best (highest log-likelihood) of the configurations compared. This behavior justifies the prior structure introduced in Section 4.
Table 1: Out of sample log-likelihood (in thousands) for macrotransition entry/exit models under various model specifications. 'Player' assumes constant hazards for each player/event type combination. 'Covariates' augments this model with situational covariates, W glyph[lscript] j ( ) t as given in (8) . 'Covariates + Spatial' adds a spatial effect, yielding (8) in its entirety. Lastly, 'Full' implements this model with the full hierchical model discussed in Section 4.
| | Model Terms | Model Terms | Model Terms | Model Terms |
|--------------|---------------|---------------|---------------|---------------|
| Macro. type | Player | Covariates | Spatial | Full |
| Pass1 | -29.4 | -27.7 | -27.2 | -26.4 |
| Pass2 | -24.5 | -23.7 | -23.2 | -22.2 |
| Pass3 | -26.3 | -25.2 | -25.3 | -23.9 |
| Pass4 | -20.4 | -20.4 | -24.5 | -18.9 |
| Shot Attempt | -48.9 | -46.4 | -40.9 | -40.7 |
| Made Basket | -6.6 | -6.6 | -5.6 | -5.2 |
| Turnover | -9.3 | -9.1 | -9.0 | -8.4 |
## 5.2 Possession Inference from Multiresolution Transitions
Understanding the calculation of EPV in terms of multiresolution transitions is a valuable exercise for a basketball analyst, as these model components reveal precisely how the EPV estimate derives from the spatiotemporal circumstances of the time point considered. Figure 7 diagrams four moments during our example possession (introduced originally in Figures 1 and 2) in terms of multiresolution transition probabilities. These diagrams illustrate Theorem 2.1 by showing EPV as a weighted average of the value of the next macrotransition. Potential ball movements representing macrotransitions are shown as arrows, with their respective values and probabilities graphically illustrated by color and line thickness (this information is also annotated explicitly). Microtransition distributions are also shown, indicating distributions of players' movement over the next two seconds. Note that the possession diagrammed here was omitted from the data used for parameter estimation.
Analyzing Figure 7, we see that our model estimates largely agree with basketball intuition. For example, players are quite likely to take a shot when they are near to and/or moving towards the basket, as shown in panels A and D. Additionally, because LeBron James
LEGEND
EPV
Figure 7: Detailed diagram of EPV as a function of multiresolution transition probabilities for four time points (labeled A,B,C,D) of the possession featured in Figures 1-2. Two seconds of microtransitions are shaded (with forecasted positions for short time horizons darker) while macrotransitions are represented by arrows, using color and line thickness to encode the value (V) and probability (P) of such macrotransitions. The value and probability of the 'other' category represents the case that no macrotransition occurs during the next two seconds.
is a better shooter than Norris Cole, the value of his shot attempt is higher, even though in the snapshot in panel D he is much farther from the basket than Cole is in panel A. While the value of the shot attempt averages over future microtransitions, which may move the player closer to the basket, when macrotransition hazards are high this average is dominated by microtransitions on very short time scales.
We also see Ray Allen, in the right corner 3, as consistently one of the most valuable pass options during this possession, particularly when he is being less closely defended as in panels A and D. In these panels, though, we never see an estimated probability of him receiving a pass above 0.05, most likely because he is being fairly closely defended for someone so far from the ball, and because there are always closer passing options for the ballcarrier. Similarly, while Chris Bosh does not move much during this possession, he is most valuable
as a passing option in panel C where he is closest to the basket and without any defenders in his lane. From this, we see that the estimated probabilities and values of the macrotransitions highlighted in Figure 7 match well with basketball intuition.
The analysis presented here could be repeated on any of hundreds of thousands of possessions available in a season of optical tracking data. EPV plots as in Figure 2 and diagrams as in Figure 7 provide powerful insight as to how players' movements and decisions contribute value to their team's offense. With this insight, coaches and analysts can formulate strategies and offensive schemes that make optimal use of their players' ability-or, defensive strategies that best suppress the motifs and situations that generate value for the opposing offense.
## 5.3 EPV-Added
Aggregations of EPV estimates across possessions can yield useful summaries for player evaluation. For example, EPV-Added (EPVA) quantifies a player's overall offensive value through his movements and decisions while handling the ball, relative to the estimated value contributed by a league-average player receiving ball possession in the same situations. The notion of relative value is important because the martingale structure of EPV ( ν t ) prevents any meaningful aggregation of EPV across a specific player's possessions. E [ ν t -ν t + glyph[epsilon1] ] = 0 for all t , meaning that on average EPV does not change during any specific player's ball handling. Thus, while we see the EPV skyrocket after LeBron James receives the ball and eventually attack the basket in Figure 2, the definition of EPV prevents such increases being observed on average.
If player glyph[lscript] has possession of the ball starting at time t s and ending at t e , the quantity ν t e -ν r glyph[lscript] ( ) t s estimates the value contributed player by glyph[lscript] relative to the hypothetical leagueaverage player during his ball possession (represented by ν r glyph[lscript] ( ) t s ). We calculate EPVA for player glyph[lscript] (EPVA( )) by summing such differences over all a player's touches (and dividing by glyph[lscript] the number of games played by player glyph[lscript] to provide standardization):
$$\text{EPVA} ( \ell ) = \frac { 1 } { \# \text{ games for } \ell } \sum _ { \{ t _ { s }, t _ { e } \} \in \mathcal { T } ^ { \ell } } \nu _ { t _ { e } } - \nu _ { t _ { s } } ^ { r ( \ell ) }$$
where T glyph[lscript] contains all intervals of form [ t s , t e ] that span player glyph[lscript] 's ball possession. Specific details on calculating ν r glyph[lscript] ( ) t are included in Appendix A.4.
Averaging over games implicitly rewards players who have high usage, even if their value added per touch might be low. Often, one-dimensional offensive players accrue the most EPVA per touch since they only handle the ball when they are uniquely suited to scoring; for instance, some centers (such as the Clippers' DeAndre Jordan) only receive the ball right next to the basket, where their height offers a considerable advantage for scoring over other players in the league. Thus, averaging by game-not touch-balances players' efficiency per touch with their usage and importance in the offense. Depending on the context of the analysis, EPVA can also be adjusted to account for team pace (by normalizing by 100 possession) or individual usage (by normalizing by player-touches).
Table 2 provides a list of the top and bottom 10 ranked players by EPVA using our 201314 data. Generally, players with high EPVA effectively adapt their decision-making process to the spatiotemporal circumstances they inherit when gaining possession. They receive the ball in situations that are uniquely suited to their abilities, so that on average the rest of the league is less successful in these circumstances. Players with lower EPVA are not necessarily
'bad' players in any conventional sense; their actions simply tend to lead to fewer points than other players given the same options. Of course, EPVA provides a limited view of a player's overall contributions since it does not quantify players' actions on defense, or other ways that a player may impact EPV while not possessing the ball (though EPVA could be extended to include these aspects).
Table 2: Top/bottom 10 players by EPVA per game in 2013-14, minimum 500 touches in season.
| Rank | Player | EPVA | Rank | Player | EPVA |
|--------|---------------|--------|--------|------------------|--------|
| 1 | Kevin Durant | 3.26 | 277 | Zaza Pachulia | -1.55 |
| 2 | LeBron James | 2.96 | 278 | DeMarcus Cousins | -1.59 |
| 3 | Jose Calderon | 2.79 | 279 | Gordon Hayward | -1.61 |
| 4 | Dirk Nowitzki | 2.69 | 280 | Jimmy Butler | -1.61 |
| 5 | Stephen Curry | 2.5 | 281 | Rodney Stuckey | -1.63 |
| 6 | Kyle Korver | 2.01 | 282 | Ersan Ilyasova | -1.89 |
| 7 | Serge Ibaka | 1.7 | 283 | DeMar DeRozan | -2.03 |
| 8 | Channing Frye | 1.65 | 284 | Rajon Rondo | -2.27 |
| 9 | Al Horford | 1.55 | 285 | Ricky Rubio | -2.36 |
| 10 | Goran Dragic | 1.54 | 286 | Rudy Gay | -2.59 |
As such, we stress the idea that EPVA is not a best/worst players in the NBA ranking. Analysts should also be aware that the league-average player being used as a baseline is completely hypothetical, and we heavily extrapolate our model output by considering value calculations assuming this nonexistant player possessing the ball in all the situations encountered by an actual NBA player. The extent to which such an extrapolation is valid is a judgment a basketball expert can make. Alternatively, one can consider EPV-added over specific players (assuming player glyph[lscript] 2 receives the ball in the same situations as player glyph[lscript] 1 ), using the same framework developed for EPVA. Such a quantity may actually be more useful, particularly if the players being compared play similar roles on their teams and face similar situations and the degree of extrapolation is minimized.
## 5.4 Shot Satisfaction
Aggregations of the individual components of our multiresolution transition models can also provide useful insights. For example, another player metric we consider is called shot satisfaction . For each shot attempt a player takes, we wonder how satisfied the player is with his decision to shoot-what was the expected point value of his most reasonable passing option at the time of the shot? If for a particular player, the EPV measured at his shot attempts is higher than the EPV conditioned on his possible passes at the same time points, then by shooting the player is usually making the best decision for his team. On the other hand, players with pass options at least as valuable as shots should regret their shot attempts (we term 'satisfaction' as the opposite of regret) as passes in these situations have higher expected value.
Specifically, we calculate
$$\text{SATIS} ( \ell ) = \frac { 1 } { | \mathcal { T } _ { \text{shot} } ^ { \ell } | } \sum _ { t \in \mathcal { T } _ { \text{shot} } ^ { \ell } } \nu _ { t } - \mathbb { E } \left [ X \ | \bigcup _ { j = 1 } ^ { 4 } M _ { j } ( t ), \mathcal { F } _ { t } ^ { ( Z ) } \right ]$$
Table 3: Top/bottom 10 players by shot satisfaction in 2013-14, minimum 500 touches in season.
| Rank | Player | Shot Satis. | Rank | Player | Shot Satis. |
|--------|------------------|---------------|--------|-------------------|---------------|
| 1 | Mason Plumlee | 0.35 | 277 | Garrett Temple | -0.02 |
| 2 | Pablo Prigioni | 0.31 | 278 | Kevin Garnett | -0.02 |
| 3 | Mike Miller | 0.27 | 279 | Shane Larkin | -0.02 |
| 4 | Andre Drummond | 0.26 | 280 | Tayshaun Prince | -0.03 |
| 5 | Brandan Wright | 0.24 | 281 | Dennis Schroder | -0.04 |
| 6 | DeAndre Jordan | 0.24 | 282 | LaMarcus Aldridge | -0.04 |
| 7 | Kyle Korver | 0.24 | 283 | Ricky Rubio | -0.04 |
| 8 | Jose Calderon | 0.22 | 284 | Roy Hibbert | -0.05 |
| 9 | Jodie Meeks | 0.22 | 285 | Will Bynum | -0.05 |
| 10 | Anthony Tolliver | 0.22 | 286 | Darrell Arthur | -0.05 |
where T glyph[lscript] shot indexes times a shot attempt occurs, { t : M t 5 ( ) } , for player glyph[lscript] . Recalling that macrotransitions j = 1 , . . . , 4 correspond to pass events (and j = 5 a shot attempt), ⋃ 4 j =1 M t j ( ) is equivalent to a pass happening in ( t, t + ]. glyph[epsilon1] Unlike EPVA, shot satisfaction SATIS( ) is expressed as an average per shot (not per game), which favors player such as glyph[lscript] three point specialists, who often take fewer shots than their teammates, but do so in situations where their shot attempt is extremely valuable. Table 3 provides the top/bottom 10 players in shot satisfaction for our 2013-14 data. While players who mainly attempt threepointers (e.g. Miller, Korver) and/or shots near the basket (e.g. Plumlee, Jordan) have the most shot satisfaction, players who primarily take mid-range or long-range two-pointers (e.g. Aldridge, Garnett) or poor shooters (e.g. Rubio, Prince) have the least. However, because shot satisfaction numbers are mostly positive league-wide, players still shoot relatively efficiently-almost every player generally helps his team by shooting rather than passing in the same situations, though some players do so more than others.
We stress that the two derived metrics given in this paper, EPVA and shot satisfaction, are simply examples of the kinds of analyses enabled by EPV. Convential metrics currently used in basketball analysis do measure shot selection and efficiency, as well as passing rates and assists, yet EPVA and shot satisfaction are novel in analyzing these events in their spatiotemporal contexts.
## 6. DISCUSSION
This paper introduces a new quantity, EPV, which represents a paradigm shift in the possibilities for statistical inferences about basketball. Using high resolution, optical tracking data, EPV reveals the value in many of the schemes and motifs that characterize basketball offenses but are omitted in the box score. For instance, as diagrammed in Figures 2 and 7, we see that EPV may rise as a player attacks the basket (more so for a strong scorer like LeBron James than for a bench player like Norris Cole), passes to a well-positioned teammate, or gains separation from the defense. Aside from simply tracking changes in EPV, analysts can understand why EPV changes by expressing its value as a weighted average of transition values (as done in Figure 7). Doing so reveals that the source of a high (or low) EPV estimate may come from alternate paths of the possession that were never realized, but were probable enough to have influenced the EPV estimate-an open teammate in a good
shooting location, for instance. These insights, which can be reproduced for any valid NBA possession in our data set, have the potential to reshape the way we quantify players' actions and decisions.
We make a number of assumptions-mostly to streamline and simplify our modeling and analysis pipeline-that could be relaxed and yield a more precise model. The largest assumption is that the particular coarsened view of a basketball possession that we propose here is marginally semi-Markov. While this serves as a workable first-order approximation, there are cases that clearly violate this assumption, for example, pre-set plays that string together sequences of runs and passes. Future refinements of the model could define a wider set of macrotransitions and coarsened states to encapsulate these motifs, effectively encoding this additional possession structure from the coach's playbook.
A number of smaller details could also be addressed. For instance, it seems desirable to model rebound outcomes conditional on high resolution information, such as the identities and motion dynamics of potential rebounders; we do not do this, however, and use a constant probability for each team of a rebound going to either the offense or defense. We also do not distinguish between different types of turnovers (steals, bad passes, ball out of bounds, etc.), though this is due to a technical feature of our data set. Indeed, regardless of the complexity and refinement of an EPV model, we stress that the full resolution data still omits key information, such as the positioning of players' hands and feet, their heights when jumping, and other variables that impact basketball outcomes. As such, analyses based on EPV are best accompanied by actual game film and the insight of a basketball expert.
The computational requirements of estimating EPV curves (and the parameters that generate them) likely limit EPV discussions to academic circles and professional basketball teams with access to the appropriate resources. Our model nevertheless offers a case study whose influence extends beyond basketball. High resolution spatiotemporal data sets are an emerging inferential topic in a number of scientific or business areas, such as climate, security and surveillance, advertising, and gesture recognition. Many of the core methodological approaches in our work, such as using multiresolution transitions and hierarchical spatial models, provide insight beyond the scope of basketball to other spatiotemporal domains.
## APPENDIX A. ADDITIONAL TECHNICAL DETAILS
In this appendix we provide additional details on steps used in fitting multiresolution models and deriving basketball metrics from EPV estimates.
## A.1 Time-Varying Covariates in Macrotransition Entry Model
As revealed in (8), the hazards λ glyph[lscript] j ( ) t are parameterized by spatial effects ( ξ glyph[lscript] j and ˜ ξ glyph[lscript] j for pass events), as well as coefficients for situation covariates, β glyph[lscript] j . The covariates used may be different for each macrotransition j , but we assume for each macrotransition type the same covariates are used across players glyph[lscript] .
Among the covariates we consider, dribble is an indicator of whether the ballcarrier has started dribbling after receiving possession. ndef is the distance between the ballcarrier and his nearest defender (transformed to log(1+ d )). ball lastsec records the distance traveled by the ball in the previous one second. closeness is a categorical variable giving the rank of the ballcarrier's teammates' distance to the ballcarrier. Lastly, open is a measure of how open a potential pass receiver is using a simple formula relating the positions of the defensive players to the vector connecting the ballcarrier with the potential pass recipient.
For j ≤ 4, the pass event macrotransitions, we use dribble , ndef , closeness , and open . For shot-taking and turnover events, dribble , ndef , and ball lastsec are included. Lastly, the shot probability model (which, from (10) has the same parameterization as the macrotransition model) uses dribble and ndef only. All models also include an intercept term. As discussed in Section 4.1, independent CAR priors are assumed for each coefficient in each macrotransition hazard model.
## A.2 Player Similarity Matrix H for CAR Prior
The hierarchical models used for parameters of the macrotransition entry model, discussed in Section 4.1, are based on the idea that players who share similar roles for their respective teams should behave similarly in the situations they face. Indeed, players' positions (point guard, power forward, etc.) encode their offensive responsibilities: point guards move and distribute the ball, small forwards penetrate and attack the basket, and shooting guards get open for three-point shots. Such responsibilities reflect spatiotemporal decision-making tendencies, and therefore informative for our macrotransition entry model (7)-(8).
Rather than use the labeled positions in our data, we define position as a distribution of a player's location during his time on the court. Specifically, we divide the offensive half of the court into 4-square-foot bins (575 total) and count, for each player, the number of data points for which he appears in each bin. Then we stack these counts together into a L × 575 matrix (there are L = 461 players in our data), denoted G , and take the square root of all entries in G for normalization. We then perform non-negative matrix factorization (NMF) on G in order to obtain a low-dimensional representation of players' court occupancy that still reflects variation across players (Miller et al. 2013). Specifically, this involves solving:
$$\hat { G } = \underset { \substack { G ^ { * } } { \arg \min } \{ D ( G, G ^ { * } ) \}, \text{ subject to } G ^ { * } = \left ( \text{U} \right ) \left ( \text{V} \right ) \text{ and } U _ { i j }, V _ { i j } \geq 0 \text{ for all } i, j, \\ \dots \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \dots \,.$$
where r is the rank of the approximation ˆ G to G (we use r = 5), and D is some distance
function; we use a Kullback-Liebler type
$$D ( { \mathbf G }, { \mathbf G } ^ { * } ) = \sum _ { i, j } G _ { i j } \log \left ( G _ { i j } / G _ { i j } ^ { * } \right ) - G _ { i j } + G _ { i j } ^ { * }.$$
The rows of V are non-negative basis vectors for players' court occupancy distributions (plotted in Figure 8) and the rows of U give the loadings for each player. With this factorization, U i (the i th row of U ) provides player i 's 'position'-a r -dimensional summary of where he spends his time on the court. Moreover, the smaller the difference between two players' positions, || U i -U j || , the more alike are their roles on their respective teams, and the more similar we expect the parameters of their macrotransition models to be a priori.
Figure 8: The rows of V (plotted above for r = 5 ) are bases for the players' court occupancy distribution. There is no interpretation to the ordering.
Formalizing this, we take the L × L matrix H to consist of 0s, then set H ij = 1 if player j is one of the eight closest players in our data to player i using the distance || U i -U j || (the cutoff of choosing the closest eight players is arbitrary). This construction of H does not guarantee symmetry, which is required for the CAR prior we use, thus we set H ji = 1 if H ij = 1. For instance, LeBron James' 'neighbors' are (in no order): Andre Iguodala, Harrison Barnes, Paul George, Kobe Bryant, Evan Turner, Carmelo Anthony, Rodney Stuckey, Will Barton, and Rudy Gay.
## A.3 Basis Functions for Spatial Effects ξ
Recalling (13), for each player glyph[lscript] and macrotransition type j , we have ξ glyph[lscript] j ( z ) = ∑ d i =1 w φ glyph[lscript] ji ji ( z ), where { φ ji , i = 1 , . . . , d } are the basis functions for macrotransition j . During the inference discussed in Section 4, these basis functions are assumed known. They are derived from a pre-processing step. Heuristically, they are constructed by approximately fitting a simplified macrotransition entry model with stationary spatial effect for each player, then performing NMF to find a low-dimensional subspace (in this function space of spatial effects) that accurately captures the spatial dependence of players' macrotransition behavior. We now describe this process in greater detail.
Each basis function φ ji is itself represented as a linear combination of basis functions,
$$\phi _ { j i } ( \mathbf z ) = \sum _ { k = 1 } ^ { d _ { 0 } } v _ { j i k } \psi _ { k } ( \mathbf z ),$$
where { ψ , k k = 1 , . . . , d 0 } are basis functions (as the notation suggests, the same basis is used for all j , i ). The basis functions { ψ , k k = 1 , . . . , d 0 } are induced by a triangular mesh of d 0 vertices (we use d 0 = 383) on the court space S . In practice, the triangulation is defined on
glyph[negationslash]
a larger region that includes S , due to boundary effects. The mesh is formed by partitioning S into triangles, where any two triangles share at most one edge or corner; see Figure 9 for an illustration. With some arbitrary ordering of the vertices of this mesh, ψ k : S → R is the unique function taking value 0 at all vertices ˜ = k k , 1 at vertex k , and linearly interpolating between any two points within the same triangle used in the mesh construction. Thus, with this basis, φ ji (and consequently, ξ glyph[lscript] j ) are piecewise linear within the triangles of the mesh.
Figure 9: Triangulation of S used to build the functional basis { ψ , k k = 1 , . . . , d 0 } . Here, d 0 = 383 .
This functional basis { ψ , k k = 1 , . . . , d 0 } is used by Lindgren et al. (2011), who show that it can approximate a Gaussian random field with Mat´rn covariance. e Specifically, let x ( z ) = ∑ d 0 k =1 v ψ k k ( z ) and assume ( v 1 . . . v k ) ′ = v ∼ N (0 , Σ ν,κ,σ 2 ). The form of Σ ν,κ,σ 2 is such that the covariance function of x approximates a Mat´ ern covariance:
$$\text{Cov} [ x ( z _ { 1 } ), x ( z _ { 2 } ) ] = \psi ( z _ { 1 } ) ^ { \prime } \Sigma _ { \nu, \kappa, \sigma ^ { 2 } } \psi ( z _ { 2 } ) \approx \frac { \sigma ^ { 2 } } { \Gamma ( \nu ) 2 ^ { \nu - 1 } } ( \kappa | | z _ { 1 } - z _ { 2 } | | ) ^ { \nu } K _ { \nu } ( \kappa | | z _ { 1 } - z _ { 2 } | | ), \quad \text{(A.3)}$$
where ψ ( z ) = ( ψ 1 ( z ) . . . ψ d 0 ( z )) . ′ As discussed in Section 4.2, the functional basis representation of a Gaussian process offers computational advantages in that the infinite dimensional field x is given a d 0 -dimensional representation, as x is completely determined by v . Furthermore, as discussed in Lindgren et al. (2011), Σ -1 ν,κ,σ 2 is sparse ((A.3) is actually a Gaussian Markov random field (GMRF) approximation to x ), offering additional computational savings (Rue 2001).
The GMRF approximation given by (A.2)-(A.3) is actually used in fitting the microtransition models for offensive players (5). We give the spatial innovation terms µ , µ glyph[lscript] x glyph[lscript] y representations using the ψ basis. Then, as mentioned in Section 4.3, (5) is fit independently for each player in our data set using the software R-INLA.
We also fit simplified versions of the macrotransition entry model, using the ψ basis, in order to determine { v jik , k = 1 , . . . , d 0 } , the loadings of the basis representation for φ , (A.2). This simplified model replaces the macrotransition hazards (8) with
$$\log ( \lambda _ { j } ^ { \ell } ( t ) ) = c _ { j } ^ { \ell } + \sum _ { k = 1 } ^ { d _ { 0 } } u _ { j k } ^ { \ell } \psi _ { k } ( z ^ { \ell } ( t ) ) + 1 [ j \leq 4 ] \sum _ { k = 1 } ^ { d _ { 0 } } \tilde { u } _ { j k } ^ { \ell } \psi _ { k } \left ( \mathbf z _ { j } ( t ) \right ),$$
thus omitting situational covariates ( β glyph[lscript] j in (8)) and using the ψ basis representation in place
of ξ glyph[lscript] j . Note that for pass events, like (8), we have an additional term based on the pass recipient's location, parameterized by { ˜ u glyph[lscript] jk , k = 1 , . . . , d 0 } . As discussed in Section 4.3, parameters in (A.4) can be estimated by running a Poisson regression. We perform this independently for all players glyph[lscript] and macrotransition types j using the R-INLA software. Like the microtransition model, we fit (A.4) separately for each player across L = 461 processors (each hazard type j is run in serial), each requiring at most 32GB RAM and taking no more than 16 hours.
For each macrotransition type j , point estimates ˆ u glyph[lscript] jk are exponentiated , so that [ 6 U j ] glyph[lscript]k = exp(ˆ u glyph[lscript] jk ). We then perform NMF (A.1) on U j :
$$\mathbf U _ { j } \approx \left ( \mathbf Q _ { j } \right ) \left ( \mathbf V _ { j } \right ).$$
Following the NMF example in Section A.2, the rows of V j are bases for the variation in coefficients { u glyph[lscript] jk , k = 1 , . . . , d 0 } across players glyph[lscript] . As 1 ≤ k ≤ d 0 indexes points on our court triangulation (Figure 9), such bases reflect structured variation across space. We furthermore use these terms as the coefficients for (A.2), the functional basis representation of φ ji , setting v jik = [ V j ] ik . Equivalently, we can summarize our spatial basis model as:
$$\xi _ { j } ^ { \ell } ( \mathbf z ) = [ \mathbf w _ { j } ^ { \ell } ] ^ { \prime } \phi _ { j } ( \mathbf z ) = [ \mathbf w _ { j } ^ { \ell } ] ^ { \prime } \mathbf V _ { j } \psi ( \mathbf z ).$$
The preprocessing steps described in this section-fitting a simplified macrotransition entry model (A.4) and performing NMF on the coefficient estimates (A.5)-provide us with basis functions φ ji ( z ) that we treat as fixed and known during the modeling and inference discussed in Section 4.
Note that an analogous expression for (A.6) is used for ˜ ξ glyph[lscript] j in terms of ˜ w glyph[lscript] j and ˜ V j for pass events; however, for the spatial effect ξ glyph[lscript] s in the shot probability model, we simply use V 5 . Thus, the basis functions for the shot probability model are the same as those for the shot-taking hazard model.
## A.4 Calculating EPVA: Baseline EPV for League-Average Player
To calculate the baseline EPV for a league-average player possessing the ball in player glyph[lscript] 's shoes, denoted ν r glyph[lscript] ( ) t in (19), we start by considering an alternate version of the transition probability matrix between coarsened states P . For each player glyph[lscript] 1 , . . . , glyph[lscript] 5 on offense, there is a disjoint subset of rows of P , denoted P glyph[lscript] i , that correspond to possession states for player glyph[lscript] i . Each row of P glyph[lscript] i is a probability distribution over transitions in C given possession in a particular state. Technically, since states in C poss encode player identities, players on different teams do not share all states which they have a nonzero probability of transitioning to individually. To get around this, we remove the columns from each P glyph[lscript] i corresponding to passes to players not on player glyph[lscript] i 's team, and reorder the remaining columns according to the position (guard, center, etc.) of the associated pass recipient. Thus, the interpretation of transition distributions P glyph[lscript] i across players glyph[lscript] i is as consistent as possible.
6 The reason for exponentiation is because estimates ˆ u glyph[lscript] jk inform the log hazard, so exponentiation converts these estimates to a more natural scale of interest. Strong negative signals among the ˆ u glyph[lscript] jk will move to 0 in the entries of U j and not be very influential in the matrix factorization (A.5), which is desirable for our purposes.
Wecreate a baseline transition profile of a hypothetical league-average player by averaging these transition probabilities across all players: (with slight abuse of notation) let P r = ∑ L glyph[lscript] =1 P glyph[lscript] /L . Using this, we create a new transition probability matrix P r ( glyph[lscript] ) by replacing player glyph[lscript] 's transition probabilities ( P glyph[lscript] ) with the league-average player's ( P r ). The baseline (league-average) EPV at time t is then found by evaluating ν r glyph[lscript] ( ) t = E P r ( glyph[lscript] ) [ X C | t ]. Note that ν r glyph[lscript] ( ) t depends only on the coarsened state C t at time t , rather than the full history of the possession, F ( Z ) t , as in ν t (4). This 'coarsened' baseline ν r glyph[lscript] ( ) t exploits the fact that, when averaging possessions over the entire season, the results are (in expectation) identical to using a full-resolution baseline EPV that assumes the corresponding multiresolution transition probability models for this hypothetical league-average player.
## APPENDIX B. DATA AND CODE
The Git repository https://github.com/dcervone/EPVDemo contains a one game sample of optical tracking data (csv), along with R code for visualizing model results and reproducing EPV calculations. Pre-computed results of computationally-intensive steps are also included as Rdata files, and can be loaded to save time and resources. A reproducible knitr tutorial, EPV demo.Rnw , introduces the data and demonstrates core code functionality.
## REFERENCES
Besag, J. (1974), 'Spatial Interaction and the Statistical Analysis of Lattice Systems,' Journal of the Royal Statistical Society: Series B (Methodological) , 36(2), 192-236.
Bukiet, B., Harold, E. R., & Palacios, J. L. (1997), 'A Markov Chain Approach to Baseball,' Operations Research , 45(1), 14-23.
Burke, B. (2010), 'Win Probability Added (WPA) Explained,' www.advancedfootballanalytics.com , (website).
Cox, D. R. (1975 a ), 'A Note on Partially Bayes Inference and the Linear Model,' Biometrika , 62(3), 651-654.
Cox, D. R. (1975 ), 'Partial Likelihood,' b Biometrika , 62(2), 269-276.
Franks, A., Miller, A., Bornn, L., & Goldsberry, K. (2015), 'Characterizing the Spatial Structure of Defensive Skill in Professional Basketball,' Annals of Applied Statistics , .
Gneiting, T., Balabdaoui, F., & Raftery, A. E. (2007), 'Probabilistic forecasts, calibration and sharpness,' Journal of the Royal Statistical Society: Series B (Statistical Methodology) , 69(2), 243-268.
Goldner, K. (2012), 'A Markov Model of Football: Using Stochastic Processes to Model a Football Drive,' Journal of Quantitative Analysis in Sports [online] , 8(1).
Higdon, D. (2002), 'Space and Space-Time Modeling Using Process Convolutions,' in Quantitative Methods for Current Environmental Issues , New York, NY: Springer, pp. 37-56.
Hollinger, J. (2005), Pro Basketball Forecast, 2005-06 , Washington, D.C: Potomac Books.
| Ihler, A., Hutchins, J., & Smyth, P. (2006), 'Adaptive Event Detection with Time-Varying Poisson Processes,' in Proceedings of the 12th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , New York, NY: ACM, pp. 207-216. |
|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| Kemeny, J. G., & Snell, J. L. (1976), Finite Markov chains: with a new appendix' General- ization of a fundamental matrix' Springer. |
| Lindgren, F., Rue, H., & Lindstr¨m, o J. (2011), 'An Explicit Link Between Gaussian Fields and Gaussian Markov Random Fields: the Stochastic Partial Differential Equation Ap- proach,' Journal of the Royal Statistical Society: Series B (Methodological) , 73(4), 423- 498. |
| Lock, D., & Nettleton, D. (2014), 'Using random forests to estimate win probability before each play of an NFL game,' Journal of Quantitative Analysis in Sports , 10(2), 197-205. |
| Miller, A., Bornn, L., Adams, R., & Goldsberry, K. (2013), 'Factorized Point Process In- tensities: A Spatial Analysis of Professional Basketball,' in Proceedings of the 31st International Conference on Machine Learning , pp. 235-243. |
| Omidiran, D. (2011), 'A New Look at Adjusted Plus/Minus for Basketball Analysis,' MIT Sloan Sports Analytics Conference [online] , 2011. |
| Prentice, R. L., Kalbfleisch, J. D., Peterson Jr, A. V., Flournoy, N., Farewell, V., & Bres- low, N. (1978), 'The Analysis of Failure Times in the Presence of Competing Risks,' Biometrics , pp. 541-554. |
| Qui˜onero-Candela, n J., &Rasmussen, C. E. (2005), 'A Unifying View of Sparse Approximate Gaussian Process Regression,' The Journal of Machine Learning Research , 6, 1939- 1959. |
| Rasmussen, C. E. (2006), Gaussian Processes for Machine Learning , Cambridge, MA: MIT Press. |
| Rue, H. (2001), 'Fast sampling of Gaussian Markov random fields,' Journal of the Royal Statistical Society: Series B (Statistical Methodology) , 63(2), 325-338. |
| Rue, H., Martino, S., & Chopin, N. (2009), 'Approximate Bayesian Inference for Latent Gaussian Models by Using Integrated Nested Laplace Approximations,' Journal of the Royal Statistical Society: Series B (Methodological) , 71(2), 319-392. |
| Shao, X., & Li, L. (2011), 'Data-Driven Multi-Touch Attribution Models,' in Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , New York, NY: ACM, pp. 258-264. |
| Thomas, A., Ventura, S. L., Jensen, S. T., & Ma, S. (2013), 'Competing Process Hazard Function Models for Player Ratings in Ice Hockey,' The Annals of Applied Statistics , 7(3), 1497-1524. |
| Wong, W. H. (1986), 'Theory of Partial Likelihood,' The Annals of Statistics , pp. 88-123. Yang, T. Y., & Swartz, T. (2004), 'A Two-Stage Bayesian Model for Predicting Winners in |
|
10.1080/01621459.2016.1141685
|
[
"Daniel Cervone",
"Alex D'Amour",
"Luke Bornn",
"Kirk Goldsberry"
] |
2014-08-04T19:22:47+00:00
|
2016-02-25T18:38:38+00:00
|
[
"stat.AP",
"stat.CO"
] |
A Multiresolution Stochastic Process Model for Predicting Basketball Possession Outcomes
|
Basketball games evolve continuously in space and time as players constantly
interact with their teammates, the opposing team, and the ball. However,
current analyses of basketball outcomes rely on discretized summaries of the
game that reduce such interactions to tallies of points, assists, and similar
events. In this paper, we propose a framework for using optical player tracking
data to estimate, in real time, the expected number of points obtained by the
end of a possession. This quantity, called \textit{expected possession value}
(EPV), derives from a stochastic process model for the evolution of a
basketball possession; we model this process at multiple levels of resolution,
differentiating between continuous, infinitesimal movements of players, and
discrete events such as shot attempts and turnovers. Transition kernels are
estimated using hierarchical spatiotemporal models that share information
across players while remaining computationally tractable on very large data
sets. In addition to estimating EPV, these models reveal novel insights on
players' decision-making tendencies as a function of their spatial strategy.
|
1408.0778v2
|
## Cosmic Matter Flux May Turn Hawking Radiation Off
Javad T. Firouzjaee 1 2 ,
1 School of Astronomy, Institute for Research in Fundamental Sciences (IPM), P. O. Box 19395-5531, Tehran, Iran ∗
## George F R Ellis 2
2
University of
Mathematics and Applied Mathematics Department, Cape Town, Rondebosch, Cape Town 7701, South Africa †
Abstract: An astrophysical (cosmological) black hole forming in a cosmological context will be subject to a flux of infalling matter and radiation, which will cause the outer apparent horizon (a marginal trapping surface) to be spacelike [7]. As a consequence the radiation emitted close to the apparent horizon no longer arrives at infinity with a diverging redshift. Standard calculations of the emission of Hawking radiation then indicate that no blackbody radiation is emitted to infinity by the black hole in these circumstances, hence there will also then be no black hole evaporation process due to emission of such radiation as long as the matter flux is significant. The essential adiabatic condition (eikonal approximation) for black hole radiation gives a strong limit to the black holes that can emit Hawking radiation. We give the mass range for the black holes that can radiate, according to their cosmological redshift, for the special case of the cosmic blackbody radiation (CBR) influx (which exists everywhere in the universe). At a very late stage of black hole formation when the CBR influx decays away, the black hole horizon becomes first a slowly evolving horizon and then an isolated horizon; at that stage, black hole radiation will start. This study suggests that the primordial black hole evaporation scenario should be revised to take these considerations into account.
PACS numbers:
## Contents
| I. Introduction | 2 |
|-----------------------------------------------------------------------|-----|
| II. The Eikonal approximation and the tunneling picture | 4 |
| III. The Geometric optics approximation, the wave phase, and redshift | 5 |
| IV. Eikonal approximation for cosmological black hole models | 7 |
| A. Oppenheimer-Snyder collapse | 7 |
| B. LTB cosmological black hole | 8 |
| C. A two-fluid model | 11 |
| D. Necessary and sufficient conditions | 13 |
| V. The future of the matter flux | 13 |
| A. The CBR flux value | 14 |
| B. The future of the CBR flux | 15 |
| VI. Exponential approximation for the affine parameter | 17 |
| VII. Discussion and Conclusions | 18 |
| References | 20 |
| A. Junction condition inside the horizon | 21 |
∗ Electronic address: [email protected]
† Electronic address: [email protected]
## C. Adiabatic condition for dynamical metric
## I. INTRODUCTION
This paper considers formation of an astrophysical black hole in a cosmological context [1]. Unlike the simple Schwarzchild or Kerr black holes, which are static or stationary respectively, cosmological black holes have dynamic behaviour and are surrounded by different kinds of matter and radiation in a changing background. In the case where the geometry is non-static, the need for a local definition of black holes and their horizons has led to concepts such as an isolated horizon [2], Ashtekar and Krishnan's dynamical horizon (DH) [3], and Booth and Fairhurst's slowly evolving horizon [4]. Although the different aspects of the dynamical nature of the astrophysical black hole [5] and primordial black hole [6] have been studied, there have not been serious attempts to consider black hole particle creation in the vicinity of dynamical horizons.[51]
It was shown in a previous paper [7] that in a realistic cosmological situation, an infalling energy density such as incoming cosmic blackbody radiation (CBR) would make the outer apparent horizon (the Outer Marginally Trapped 3-Surface, or OMOTS) become spacelike. When the apparent horizon is spacelike, it seems likely that some emitted radiation will fall into the singularity, and some will escape to infinity. This is because the radiation is emitted in the vicinity of the apparent horizon, so some may be emitted from far enough away that it will escape the event horizon. One might call this part of the radiation that escapes to infinity 'Hawking radiation'; that is a question of terminology. We rather refer to all radiation from the vicinity of the apparent horizon as Hawking radiation, whether it reaches infinity or not.
Hawking's quantum field theory approach to black hole radiation [8, 9, 15], which applies to late time stationary black holes, is not a suitable method for calculating the Hawking temperature in the case of a fully dynamical black hole, where one has to solve the field equations in a changing background. Even the semiclassical stress tensor approach [16] has limitations related to finding the effective action and taking into account the backscattering problem. In such cases, one should look for alternative approaches allowing one to calculate the Hawking radiation in a dynamical context [11, 12, 14]. This radiation is plausibly emitted from the vicinity of apparent horizons rather than from near the global event horizon [10].
The tunneling picture of Hawking radiation emission [11, 12] suggests that when there is infalling matter or radiation, there will be no emission of blackbody radiation arriving at future infinity from the vicinity of the OMOTS surface, because it is spacelike [7, 13]. Additionally Hawking radiation emitted from the inner apparent horizon (the Inner Marginally Trapped 3-Surface, or IMOTS) would have been trapped, so on this view, no Hawking radiation would escape to infinity [13]. However this is based on a particle picture that can be queried.
Here we show that if one uses the eikonal approximation [14], that result is confirmed: when an infalling matter or radiation flux occurs, no Hawking radiation is emitted from the spacelike OMOTS surface. However additionally, because of the way the phase relates to redshift, the IMOTS surface will also not emit Hawking Radiation. In other words, both tunneling arguments (WKB approximation) [12] and eikonal arguments [10] suggest that a matter or CBR radiation flux turns off Hawking radiation to infinity, implying that no black hole explosions [15] would occur when black holes form in a cosmological context as long as the matter flux is significant. However at very late times, all the matter that can do so will have fallen into the black hole in the Λ CDM background and the CBR will have decayed away to almost zero. An isolated event horizon will then form and, on the eikonal view, will then lead to the standard picture of black hole radiation.
One should note here that while the mathematically clearest derivation of the Hawking effect in the usual non-dynamical context involves calculating the evolution of the two-point function of the quantum field backward in time from infinity to near the horizon, the relevant properties of the calculation in that case can also be seen clearly by using the different method given in Hawking's original paper.
However neither method is appropriate in the context we consider, where the semi-classical radiation we are considering may not escape to infinity at all. That is why we use the methods we do, which are suitable for a dynamical context. Dynamical black holes may or may not emit also Hawking radiation. This is a valid question, and we should not be using any framework that doesn't allow it to be answered because it assumes the geometry is close to Schwarzschild. The detailed discussion in most papers deriving Hawking radiation, for example [9], are confined to the case of non-dynamical black holes, and we are considering the more realistic dynamical case. It has been suggested that the deviations from Schwarzschild due to infalling matter would have a completely negligible effect; however this is not supported by work on modelling dynamical black holes [5, 6].
We also consider Hawking's original argument [8] using Bogoliubov coefficients for deriving the black hole thermal radiation for stationary black holes (when there is no influx of matter or radiation). This argument relates radiation emission to the way the geometry of outgoing null geodesics leads to exponential rescaling of null geodesic generators of past null infinity and future null infinity [10]. However when there is infalling matter or radiation, the fact that the OMOTS surface is spacelike causes the event horizon to shift outwards relative to when there is no infalling energy density, and the spacetime geometry is then such that the relevant outgoing null geodesics never reach infinity [7]: rather they fall into the future spacelike singularity (see Fig.(6)). The Hawking argument does not then apply.
Event horizon and backreaction problem: Generally, we can divide black hole formation and evolution into three steps.
The first steps involves the gravitational collapse of matter. During this step, apparent horizons and the singularity form, and we can define the classical event horizon. Most of the gravitational physics can be described by classical physics. During this step, we can study the nature of the horizons (IMOTS and OMOTS) and their location, shell crossings and matter flux rate, and so on. As discussed in this paper, the eikonal (adiabatic) approximation will not be valid during this step.
The second step is when the black hole becomes quasi-static, and we can apply the adiabatic condition to it. The black hole starts to radiate and lose mass, and we have to consider semiclassical physics in order to examine the time evolution [40]. During this step determining the location of the event horizon is complicated, because in order to find the event horizon location we need to know the whole story of the gravitational collapse for all time. To do this, we need to solve the backreaction problem, but it seems that no one has a comprehensive theory to do so. Since the black hole at this step emits quanta of much smaller energy than the whole mass, the adiabatic approximation used in the quantum calculations of the emission will be valid.
The final step which determines the fate of the black hole is the quantum gravitational step. It seems that we can only examine the fundamental question of black hole information and the event horizon during this step.
Our work here considers the first step - the black hole evolution - and investigates when it reaches the second step. This argument does not cover the back reaction problem and possible eventual black hole evaporation. To consider this, one will need to check what the effective stress-tensor associated with particle production is, and also that radiation emission from the central collapsing fluid [16] does not prevent formation of apparent horizons and the associated singularities [17]. These issues will be the topic of further papers. However the tunneling approach is used by many, as is the eikonal approximation: this paper takes the argument of [7] forward by showing these two approaches agree that existence of the matter flux turns off Hawking radiation and so suppresses black hole explosions.
This paper: We will show the close relation between the eikonal approximation and particle tunneling picture in section II. In Section III, we discuss the eikonal approximation relation to the redshift of null geodesics. Section IV considers this approximation for some cosmological black hole models. In Section V, we consider the matter and radiation flux and its evolution in time. Section VI looks at how the exponential relation between null geodesic parameters on future and past null infinities does not occur when the OMOTs is spacelike. We then conclude in section VII. In carrying out this study, it is convenient to think of three successive approximations. First we consider the standard case [8] with no matter flux effect and a static exterior spacetime. Second, we consider black holes with a constant positive influx of
matter or radiation at all times [7]. Finally, we consider the realistic case where the matter flux rate is non zero at all times, but is decaying away to zero in the late time expanding universe.
## II. THE EIKONAL APPROXIMATION AND THE TUNNELING PICTURE
In general relativity, the classical field φ without potential solves the massless Klein-Gordon equation
$$g ^ { \mu \nu } \nabla _ { \mu } \nabla _ { \nu } \phi = 0.$$
To solve it we can locally use the eikonal approximation
$$\phi & = a e ^ { i \psi }$$
where the amplitude a x ( i ) varies much more slowly than the phase ψ x ( i ) when the phase ψ is rapidly varying: ψ glyph[greatermuch] 1. The eikonal approximation can be presented according to the adiabatic condition which will be discussed in Appendix C. Then (1) gives the eikonal equation
$$g ^ { \mu \nu } \nabla _ { \mu } \psi \nabla _ { \nu } \psi = g ^ { \mu \nu } \partial _ { \mu } \psi \partial _ { \nu } \psi = 0$$
for ψ . In analogy with a wave in Minkowski space time, k i = ∂ ψ i is the wave vector (here Latin indexes run from 1 to 3) and w = -∂ψ ∂t the frequency of the wave measured in the coordinate frame. For a preferred observer with four-velocity u µ , the frequency measured by the observer is w = -u ∂ ψ µ µ .
We consider the case of spherical waves: a = a r, t ( ), ψ = ψ r, t ( ). Over a small region of space time for a local observer the eikonal ψ can then be expanded to first order as
$$\psi = \psi _ { 0 } + \frac { \partial \psi } { \partial t } t + \frac { \partial \psi } { \partial r } r$$
For a stationary space time the eikonal can be written
$$\psi = \psi _ { 0 } + w t \pm \int ^ { r } k ( r ^ { \prime } ) d r ^ { \prime }$$
where the +(-) sign in front of the integral indicates that the radial wave is ingoing (outgoing) and k r ( ) := k r .
The geometrical optics corresponds to the limiting case of small wavelength λ → 0 which satisfies the eikonal equation (3). Then the eikonal equation (3) is like the Hamilton-Jacobi equation
$$g ^ { \mu \nu } \partial _ { \mu } S \partial _ { \nu } S = 0$$
where the action S is related to the momentum by p i = ∂S ∂x i and the Hamiltonian is H = -∂S ∂t . Comparing these formula with the field case, we see that wave vector plays the same role as momentum of the particle and frequency plays the role of the Hamiltonian or energy of the particle in geometrical optics:
$$k _ { i } \Leftrightarrow p _ { i }, \ w \Leftrightarrow E \simeq H.$$
To consider Hawking radiation for a dynamical black hole in the spherically symmetric case in Painlev´e Gullstrand coordinates, Visser [14] has used this ansatz for the s wave:
$$\phi = A ( r, t ) e x p [ \mp i ( w t - \int ^ { r } k ( r ^ { \prime } ) d r ^ { \prime } ) ]$$
which is basically the same as using (5) in (2). This form of the wave is valid provided the geometry is slowly evolving on the timescale of the wave.
On the other hand, Hartle and Hawking [28] obtained particle production in stationary black hole space-times using a semi-classical analysis which does not require knowledge of the wave modes. This
method has been extended to different Schwarzchild coordinates in [30]. In their method, the ratio between emission and absorption probabilities is given by
$$\Gamma \sim e ^ { - \beta w } = \frac { P _ { [ e m i s s i o n ] } } { P _ { [ a b s o r p t i o n ] } },$$
where the probability P is the square of the amplitude of the field: P = | φ | 2 . By inserting (8) in this equation we get
$$\Gamma & \sim e ^ { - 2 I m \, f ^ { r } \, k ( r ^ { \prime } ) d r ^ { \prime } }.$$
The term ∫ r k r ( ′ ) dr ′ has a pole singularity on the apparent horizon and gives an imaginary part in this coordinate system. If other coordinates had been chosen, one should have considered the temporal contribution to the emission rate [18]. If one wants to use the particle picture for the wave and take k r ( ) = p r ( ) then we will get the emission rate in the particle picture [12]
$$\Gamma \sim e ^ { - 2 I m \int ^ { r } p ( r ) d r } = e ^ { - 2 I m S }$$
where S is the particle action which satisfies the Hamilton-Jacobi equation and for the Painlev´ coordinate e system we have
$$I m \int ^ { r } p ( r ) d r & = I m S.$$
As a result, the Visser eikonal method [14] and the Parikh and Wilczek tunneling method [12] are similar although they have different wave and particle pictures. However there is one key difference: the tunneling method cannot apply for particle production whenever the MOTS surface is spacelike, because the whole concept of tunneling only makes sense for a timelike surface (where 'inside' and 'outside' are well defined concepts); however the eikonal method (which is based on a wave rather than particle picture) can give particle production when the surface is spacelike and there is slow evolution of the geometry. This difference will be important in the sequel.
Role of the eikonal approximation: In contrast to Minkowski space time, in which the definition of a particle with momentum k is based on a decomposition of fields into plane waves e i ( wt -kx ) , in the dynamic case the spatial size of the wave (particle) packet varies with the dynamics of the space time, and a plane wave cannot describe it. In spacetimes with a slowly changing geometry, the so-called adiabatic vacuum allows defining a meaningful notion of particles that can be applied for an evolving space time [39]. The adiabatic vacuum prescription relies on the WKB (eikonal) approximation for solutions of the wave equation. As stressed by Barcelo et.al [10], physically the adiabaticity constraint (eikonal approximation) is equivalent to the statement that a photon emitted near the peak of the Planckian spectrum should not see a large fractional change in the peak energy of the spectrum over one oscillation of the electromagnetic field (that is, the change in spacetime geometry is adiabatic as seen by a photon near the peak of the Hawking spectrum). The eikonal approximation for having particle creation may have a more profound meaning if we examine the quantum stress-tensor.
## III. THE GEOMETRIC OPTICS APPROXIMATION, THE WAVE PHASE, AND REDSHIFT
In the vacuum (Schwarzschild) case, the event horizon is associated with infinite redshift in the following sense: if a freely falling object drops into the black hole, as it approaches the event horizon, an external observer will see it with ever increasing redshift; as it crosses the event horizon, the redshift diverges [17]. Anywhere that Hawking radiation is associated with an event horizon [8], it is associated with an infinite redshift surface.
In this section we show that that association is not a coincidence: it is essential to the radiation process, and remains true even in the case of a dynamical horizon.
Consider the eikonal equation for the wave ϕ = A t, r ( ) e ± iψ . It is known that k µ = ∇ µ ψ , the normal vector of the constant phase plane, describes wave propagation, and is a null vector:
$$k _ { \mu } k ^ { \mu } = 0.$$
$$k _ { \mu } \nabla _ { \nu } k ^ { \mu } = 0.$$
Since ∇ ν k µ = ∇ ∇ ν µ ψ = ∇ ∇ µ ν ψ = ∇ µ k ν
$$\psi = \nabla _ { \mu } k _ { \nu } \text{ we get the geodesic equation} \\ k ^ { \nu } \nabla _ { \nu } k _ { \mu } = 0.$$
In other words, the null geodesic vector derived from the eikonal equation is affinely parametrized, and equation (13) is the same as the eikonal equation (3).
Let's look at light propagation from an emitter ( e ) to an observation point ( o ). The frequency which is measured by a observer with 4-velocity ˜ u µ = dx µ dλ is w = k µ ˜ u µ ( λ is proper time for the time like observer). Thus, the redshift at the observer point is
$$1 + z = \frac { \nu _ { e } } { \nu _ { o } } = \frac { ( g _ { \mu \nu } k ^ { \mu } \tilde { u } ^ { \nu } ) _ { e } } { ( g _ { \mu \nu } k ^ { \mu } \tilde { u } ^ { \nu } ) _ { o } }.$$
Note that k µ must be an affinely parametrized null vector. An infinite redshift surface is a surface such that the redshift of light arriving at that surface becomes infinite: (1 + z ) → ∞ . The geometric optics approximation is satisfied near an infinite redshift surface.
Let us expand the wave phase near the eikonal approximation case ψ glyph[greatermuch] 1:
$$\psi - \psi _ { 0 } \ & = \ \nabla _ { \mu } \psi d x ^ { \mu } = k _ { \mu } \frac { d x ^ { \mu } } { d \lambda } d \lambda \\ & = \ \nu d \lambda = ( 1 + z ) \nu _ { 0 } d \lambda.$$
In the second line we have used the redshift equation (15). This equation shows that the phase of the wave near the infinite redshift surface is very big: ψ glyph[greatermuch] 1.
According to the discussion by Visser [14], the eikonal (geometric optics) approximation is an essential feature for occurrence of black hole radiation. Therefore, having geometric optics valid in the close vicinity of the apparent horizon is a necessary condition for demonstrating existence of Hawking radiation by the eikonal method. We now see that we can use existence of an infinite or very large redshift surface as a criterion for when this is satisfied.
glyph[negationslash]
This is actually associated with the exponential piling up of the null affine parameter relationship on future null infinity that is often seen as the key to existence of Hawking radiation ([8] and see Section 6). The reason is as follows: whenever there is a timelike Killing vector field ξ a , the energy E of a photon is given by E = -ξ a k a . On a bifurcate Killing horizon, where a Killing vector field changes from timelike to spacelike [24], the Killing vector parameter ξ : ξ a = dx /dξ a and the geodesic affine parameter v : k a = dx /dv a are related exponentially: v = exp( κξ ), κ = 0, which leads to the affine parameter relationship between past and future null infinity discussed in [8]. It follows that
$$k ^ { a } & = \exp ( - \kappa v ) \xi ^ { a },$$
is parallely propagated on the null horizon [24], which leads to the divergent redshift relation as a null geodesic parallel to the horizon approaches the horizon. Note that there will be no such divergence in the case when the Killing vector field does not change from timelike to spacelike on the horizon, but rather is null in an open neighbourhood. This is the case when the surface gravity vanishes ( κ = 0). This is the reason that a non-zero surface gravity is a necessary condition for Hawking radiation emission [14]. However of course the whole of this argument depends on the existence of the Killing vector field, and so will not be valid in a dynamical spacetime. Our argument above can be applied in that more general case.
In the next section we consider application of the redshift condition discussed here to the case of a cosmological black hole.
## IV. EIKONAL APPROXIMATION FOR COSMOLOGICAL BLACK HOLE MODELS
The study of black holes in stationary and asymptotically flat spacetimes has led to many remarkable insights. But, as we know, our universe is not stationary and is in fact undergoing cosmological expansion in presence of background radiation; that is the context in which we should consider gravitational collapse to form a black hole [7]. There have been many papers constructing solutions of the Einstein equations representing a collapsing central mass in a cosmological context. Gluing of a Schwarzchild manifold to an expanding FRW manifold is one such attempt, made first by Einstein and Straus [25].
Now, a widely used metric to describe gravitational collapse of a spherically symmetric dust cloud is the so-called Lemaˆ ıtre-Tolman-Bondi (LTB) metric [31]. Many people have looked at LTB models describing an overdense region in a cosmological background [1]. Although there were some attempts to investigate Hawking radiation from dynamical black holes through tunneling [11], no one has considered necessary conditions for this method in presence of a matter flux such as that due to cosmological black body background radiation (CBR), which does indeed occur everywhere in the real universe.
In this section we consider the essential features of black hole radiation emission for three models of cosmological black holes that take this effect into account: Oppenheimer-Snyder collapse, an LTB cosmological black hole, and a two-fluid model.
## A. Oppenheimer-Snyder collapse
The Oppenheimer-Snyder solution consists of a dust filled Friedmann model, joined across comoving spheres to a timelike hypersurface in the Schwarzchild solution. To make a cosmological black hole, we can match the exterior Schwarzchild black hole to an internal spatially homogeneous FLRW universe (this is one kind of Einstein-Straus cosmological black hole). Since we want to consider the tunneling effect near the horizon, it is sufficient to consider an Oppenheimer-Snyder solution. The metric inside (sign -) the collapsing dust in case of a flat 3-geometry is given by
$$d s _ { - } ^ { 2 } = - d \tau ^ { 2 } + a ( \tau ) ^ { 2 } ( d \chi ^ { 2 } + \chi ^ { 2 } d \Omega ^ { 2 } ).$$
The Einstein equation shows that a τ ( ) satisfies
$$\dot { a } ^ { 2 } & = \frac { 8 \pi } { 3 } \rho a ^ { 2 }.$$
The equation of conservation, ∇ µ T µν = 0, gives the density as ρ τ ( ) = ρ 0 a 3 . We assume that at an initial time τ = τ 0 the scale factor is a = 1, then we get
$$a ( \tau ) = ( 1 - \frac { 3 } { 2 } \sqrt { \frac { 8 \pi \rho _ { 0 } } { 3 } } ( \tau - \tau _ { 0 } ) ) ^ { \frac { 2 } { 3 } } \,.$$
The star surface is located at χ 0 = constant . The metric outside (sign+) is given by
$$d s _ { + } ^ { 2 } = - f d t ^ { 2 } + f ^ { - 1 } d R ^ { 2 } + R ^ { 2 } d \Omega ^ { 2 },$$
where f = 1 -2 M/R .
Since this coordinate system has a singularity at R = 2 M , we choose the Lemaˆ ıtre coordinate [27] to give a junction with the FLRW region which does not have a coordinate singularity (see Appendix A ).
Using the junction condition one can show
$$M = \frac { 4 \pi } { 3 } \rho a ^ { 3 } \chi _ { 0 } ^ { 3 }$$
which is constant. When the star radius is bigger than its Schwarzchild radius, there is no trapped surface, i.e R > 2 M . When the star falls into it's Schwarzchild radius, there will be two parts of the apparent horizon [7]. The first on the outside is the OMOTS (Outer marginally outer trapped 3-surface)
which is the same as the Schwarzchild event horizon, and the second is the IMOTS (Inner marginally outer trapped 3-surfaces) which is inside the star and reaches the singularity at R = 0. The trapped surface is located at R = 2 M which is given by
$$\chi | _ { A H } & = \frac { 2 M } { a }.$$
If this black hole is embedded in the cosmological CBR radiation, it will not make much difference to the spacetime curvature, so we can treat this radiation in a linear approximation as a propagating field on the LTB background making little difference to the fluid collapse. It will however change the location of the OMOTS surface, which will become spacelike because of the CBR influx, which continually falls into the black hole as discussed in [7, 13].
FIG. 1: Oppenheimer-Snyder collapse in the presence of CBR and matter influx, in comoving coordinates. The left hand diagram is when there is no infalling flux, and the right one is for the case of a matter or radiation influx that decays at late times.
The radiation from the vicinity of the OMOTS will be discussed in the next subsection, which is the more general case. To know whether this black hole emits radiation by tunneling from IMOTS, we have to check the geometric optic approximation for the ingoing and outgoing null geodesics which are emitted from a point near the IMOTS, see Fig.(1). Consider an ingoing null geodesic which comes from a point (emitter) on or near the IMOTS surface and arrives at an observer outside this surface. Since the space time is FLRW and the IMOTS is time like: | dτ dχ null | < | dτ dχ IMOTS | , the ingoing wave can pass between the two points without an infinite change in its phase. The outgoing null geodesic can exit from the timelike IMOTS to outside without seeing any infinity (since the space time is FLRW).
Hence, the geometric optic approximation is not valid for ingoing or outgoing null geodesics emitted near the IMOTS. As a result, the eikonal approximation condition [14] is not valid for this surface, and there is no black hole radiation from the timelike IMOTS surface. The Penrose diagram for the Oppenheimer-Snyder collapse with incoming CBR radiation is depicted in Fig.(2).
## B. LTB cosmological black hole
The LTB [31] metric is a spherically symmetric non-static solution of the Einstein equations with a dust source. It can be written in synchronous coordinates as
$$d s ^ { 2 } = - d t ^ { 2 } + \frac { R ^ { \prime 2 } } { 1 + f ( r ) } d r ^ { 2 } + R ( t, r ) ^ { 2 } d \Omega ^ { 2 }.$$
and represents a pressure-less perfect fluid satisfying
$$\rho ( r, t ) = \frac { 2 M ^ { \prime } ( r ) } { R ^ { 2 } R ^ { \prime } }, \quad \dot { R } ^ { 2 } = f + \frac { 2 M } { R } + \frac { \Lambda } { 3 } R ^ { 2 }.$$
IMOTS
II:
+
>0
S MOTS
Outgoing wave
CBR
I:
+
> 0
FIG. 2: Penrose diagrams of black hole formation with incoming CBR or matter flux. The left figure is the case of a constant infalling flux, and the right one shows the case of a decaying flux at late times. When the OMOTS surface is spacelike because of infalling radiation, as the outgoing null wave approaches that surface, the wave redshift seen for radiation emitted from close to this surface does not diverge.
Since this metric is dynamical and includes the general case of a FLRW metric as well as inhomogeneous pressure free stars, the cosmological black hole can be modeled with this metric. As discussed in [26], we can neglect the cosmological constant (considered as a dark energy candidate) for discussing local effects at the black hole such as black hole horizon dynamics.
In presence of the CBR and matter flux, we assume that most of the matter inside and around the black hole is dust causing a matter influx, and the CBR has a non-zero flux of radiation which falls in across the horizon and which can be treated as a linear perturbation, approximated as a propagating field on the LTB background and making little difference to the fluid collapse. It will however change the location of the OMOTS surface, which will become spacelike because of the CBR and matter influx [7, 13]. Thus the space time geometry can effectively be described by an LTB model embedded in the cosmological expanding background, in which the black horizon is dynamical owing to the non-zero incoming radiation flux.
To consider the geometric optics approximation on the apparent horizon let's look at null geodesics near this surface. Using the apparent horizon equation
$$d R = 2 d M ( r ) = R ^ { \prime } d r + \dot { r } d t | _ { A H },$$
the slope of the null geodesic vector tangent to the apparent horizon tangent vector gives
$$\frac { \frac { d t } { d r } | _ { n u l l } } { \frac { d t } { d r } | _ { A H } } = \frac { R ^ { \prime } } { R ^ { \prime } - M ^ { \prime } ( r ) }$$
glyph[negationslash]
To get this equation we have used the collapsing region condition ˙ R < 0 and equation (25). This equation shows that if there were no matter flux into the black hole, M ′ = 0 ⇔ ρ = 0, the apparent horizon has to be null. If the matter flux is non-zero, the slope of the null geodesic is greater than that of the apparent horizon, which is spacelike [7]. In the case that there is no shell-crossing singularity, R ′ = ∞ , and no other singularity on the apparent horizon, the null geodesics slope is finite and non-zero.
Generally the LTB apparent horizon can be either an IMOTS or OMOTS, with the nature of the latter depending on the matter falling into the black hole [19]. In the case that the matter flux has
III:
OMOTS
+
< 0
positive energy density, the OMOTS apparent horizon is spacelike and lies inside the event horizon.
No incoming CBR: It has been shown [26] that the redshift of the light emitted from the apparent horizon for a dust BH without incoming CBR is not generically infinite, but in the case that the apparent horizon (OMOTS) is a slowly evolving horizon it is infinite. As a result, Hawking radiation will occur from the OMOTS (slowly evolving horizon) for a dust cosmological black hole with no CBR.
Incoming CBR: Now we examine the redshift of the light from apparent horizons (both OMOTS and IMOTS) for a dust cosmological black hole in the presence of CBR radiation. Since the CBR radiation gives a non-zero term for the flux of the matter, it causes the black hole area to grow [3]:
$$\mathcal { F } = \mathcal { F } _ { \text{matter} } + \mathcal { F } _ { \text{CBR} } \coloneqq \frac { 1 } { G } ( M ( r _ { 2 } ) - M ( r _ { 1 } ) ) | _ { A H },$$
which means M > ′ 0. For this black hole, the light redshift is [26]:
$$1 + z \ & = \ c _ { 0 } \ e x p \left ( - \int _ { o } ^ { e } \frac { \dot { R } ^ { \prime } } { \sqrt { 1 + f } } d r \right ) \\ & = \ c _ { 0 } \ e x p ( \int _ { o } ^ { e } \frac { - 1 } { \sqrt { 1 + f } } \left ( \frac { M ^ { \prime } } { R \dot { R } } - \frac { M R ^ { \prime } } { \dot { R } R ^ { 2 } } + \frac { f ^ { \prime } } { 2 \dot { R } } ) d r \right ).$$
FIG. 3: The LTB causal diagram near the IMOTS and OMOTS surfaces in the presence of CBR and matter influx, in comoving coordinates.
Note that we calculate this redshift for a comoving observer u µ who is sitting at r = constant with proper time t . We can choose another observer with 4-velocity u ′ µ = dt dt ′ u µ , and we will see that the results will not change if there is no singularity in the second observer 4-velocity. As we know the apparent horizon is in the collapse region where ˙ R < 0 and R is finite and non-zero there. With the assumption that there is no shell-crossing singularity and no infinity for density on the apparent horizon, we get R ′ and therefore M ′ is finite. As depicted in Fig.(3), the ingoing and outgoing null geodesics which emerge from the space like apparent horizon (OMOTS) to a point inside it, will traverse a finite ∆ because r dt dr | null is finite and non-zero. Similarly, the ingoing and outgoing null geodesic which emerge from the time like apparent horizon (IMOTS) to a point outside or inside it, will traverse a finite ∆ r because dt dr | null is finite and non-zero as depicted in Fig.(3). Therefore, the exponential part of the equation (29) will be finite. Hence, the cosmological black hole apparent horizon in the presence of CBR radiation and matter flux does not satisfy the geometric optics approximation. Consequently, there is no Hawking radiation from either the IMOTS surface or the OMOTS surface in this black hole. Note that, the causal structure in the Fig.(3) does not depend on the coordinates and the apparent horizon does not change this spacelike character, so all the above argument can be applied for other non-singular coordinates.
This is in contrast to the Schwarzchild limit case, M ′ = 0, where there is no incoming CBR and matter flux and the apparent horizon tangent vector is a null vector (parallel to a null geodesic).
Then the outgoing null geodesic traverses an infinite ∆ r → ∞ from the emitter point (on the horizon) to the observer point (this is the essential content of equation (2.16) in [8]). Therefore, in this case the apparent horizon is an infinite redshift surface and there is indeed black body radiation emission.
All the above discussion can be extended to beyond the s-wave (spherical wave) case. As discussed in [14], the near-horizon asymptotic behaviour, the phase pile-up, the continuation of the outgoing modes across horizon and Hawking temperature, are independent of having only an s-wave mode.
Even without these calculations, one can intuitively infer that since we do not have the adiabatic condition [10] or late time apparent horizon (OMOTS) [29] for the dynamical black hole, there is no Hawking radiation. This is because null radiation emitted near the apparent horizon when the CBR and matter flux is significant would not be not tangent to the apparent horizon, which is spacelike, and so would arrive at infinity with a finite redshift ( Fig.(2)).
## C. A two-fluid model
The previous two models can be criticized because they do not model the effect on the spacetime of the CBR energy density. Here we consider a 2-fluid model representing the dynamic effects of both the matter and the CBR radiation. As the latter is isotropic to very high accuracy in a FLRW model, we can represent it as a perfect fluid with 4-velocity u a CBR , density ρ CBR , and pressure p CBR = 1 3 ρ CBR . Thus the first fluid is ordinary matter fluid (non-zero inside the collapsing fluid, zero outside,) and the second one is the CBR radiation fluid. We assume the matter flux to be a perfect fluid, which can represent a realistic picture of gravitational collapse [33] and has the ability to model a cosmological black hole [26].
It can be shown that the sum of the two perfect fluids is a fluid with anisotropic stress π ab and heat flow q a [32]. However we can set the heat flow to zero by choosing the timelike vector u a as the eigenvector of the Ricci tensor: then q a = 0. Because of the spherical symmetry of the problem, we can represent the anisotropic fluid pressure in this frame by two terms: a radial pressure p r and a tangential pressure p θ . For slowly moving matter relative to the CBR, this effective fluid has
$$\rho = \rho _ { m } + \rho _ { C B R }.$$
The collapsing fluid within a compact spherically symmetric spacetime region will be described by the following metric in the comoving coordinates ( t, r, θ, ϕ ):
$$d s ^ { 2 } = - e ^ { 2 \nu ( t, r ) } d t ^ { 2 } + e ^ { 2 \psi ( t, r ) } d r ^ { 2 } + R ( t, r ) ^ { 2 } d \Omega ^ { 2 }.$$
The energy momentum tensor for the fluid will have the form
$$T _ { t } ^ { t } = - \rho ( t, r ), \ T _ { r } ^ { r } = p _ { r } ( t, r ), \ T _ { \theta } ^ { \theta } = T _ { \varphi } ^ { \varphi } = p _ { \theta } ( t, r ),$$
with the weak energy condition satisfied:
$$\rho \geq 0 \ \rho + p _ { r } \geq 0 \ \rho + p _ { \theta } \geq 0.$$
The Einstein equations give,
$$\rho = \frac { 2 M ^ { \prime } } { R ^ { 2 } R ^ { \prime } } \, \ p _ { r } = - \frac { 2 \dot { M } } { R ^ { 2 } \dot { R } },$$
$$\nu ^ { \prime } = \frac { 2 ( p _ { \theta } - p _ { r } ) } { \rho + p _ { r } } \frac { R ^ { \prime } } { R } - \frac { p _ { r } ^ { \prime } } { \rho + p _ { r } },$$
$$- \, 2 \dot { R } ^ { \prime } + R ^ { \prime } \frac { \dot { G } } { G } + \dot { R } \frac { H ^ { \prime } } { H } = 0,$$
where
$$G = e ^ { - 2 \psi } ( R ^ { \prime } ) ^ { 2 } \, \ H = e ^ { - 2 \nu } ( \dot { R } ) ^ { 2 },$$
and M is Misner-Sharp mass defined by
$$G - H = 1 - \frac { 2 M } { R }.$$
As discussed in [26], the surface R = 2 M is the apparent horizon for the collapsing region. In this case the matter flux which falls into the singularity is,
$$\mathcal { F } = \mathcal { F } _ { \text{matter} } + \mathcal { F } _ { \text{CBR} } \coloneqq \frac { 1 } { G } ( M ( t _ { 2 }, r _ { 2 } ) - M ( t _ { 1 }, r _ { 1 } ) ) | _ { A H },$$
As a result of non-zero CBR and matter flux, the density and the pressure are finite on the apparent horizon and then there are non-zero and finite values for M > ′ 0, ˙ M > 0. No shell-focusing singularity, no shell-crossing singularity and the apparent horizon location in the collapsing region give non-zero and finite values for R R , ′ and ˙ R < 0 on the apparent horizon respectively. The null geodesic slope on the apparent horizon is dt dr | null = ± R ′ | ˙ R | which according to the above discussion is finite for the outgoing and ingoing null geodesics. With some calculation one can get following equation on the horizon
$$\frac { \frac { d t } { d r } | _ { n u l l } } { \frac { d t } { d r } | _ { A H } } = \left ( \frac { 1 - \frac { 2 \dot { M } } { R } } { 1 - \frac { 2 M ^ { \prime } } { R ^ { \prime } } } \right ).$$
In the vacuum (Schwarzchild limit) case M ′ = 0 ⇔ ρ = 0 and ˙ M = 0 ⇔ p = 0, so the apparent horizon becomes a null surface. Using the weak energy condition (33) we get 2 M ′ R ′ ≥ 2 ˙ M ˙ R . The case of equality in the weak energy condition ρ = -p is very special and non applicable to a realistic cosmological black hole, so we get
$$\frac { 2 M ^ { \prime } } { R ^ { \prime } } > \frac { 2 \dot { M } } { \dot { R } }.$$
Using this equation in (40) results in the dynamical case where the null geodesics cross the apparent horizon (which is spacelike [7]), and for the outgoing null geodesic we have
$$\frac { \frac { d t } { d r } | _ { n u l l } } { \frac { d t } { d r } | _ { A H } } > 1.$$
In this way we can calculate the light redshift from the apparent horizon of this metric. To this end, we need to calculate the affine parametrized radial null geodesic vector k µ . Without writing the details of the calculation, the result is that the affinely parametrised null vector is
$$k ^ { \mu } = c _ { 0 } e ^ { - \int ( \dot { \nu } + \dot { \psi } + \nu ^ { \prime } e ^ { \nu - \psi } ) \frac { d r } { e ^ { \nu - \psi } } } ( 1, \frac { \dot { R } } { R ^ { \prime } }, 0, 0 )$$
Considering a comoving observer with u µ = ( e -ν , 0 0 0), we can calculate the redshift of the null geodesics , , from the emitter point to the observer point. Using the geometric optics approximation, the question is whether the quantity e -∫ ( ˙ + ν ˙ + ψ ν e ′ ν -ψ ) dt is infinite on the apparent horizon. Let us check all terms in this equation. The term e ν -ψ = ˙ R R ′ on the horizon, which is finite. We rewrite the equation (38) as
$$M = \frac { R } { 2 } ( 1 - e ^ { - 2 \psi } ( R ^ { \prime } ) ^ { 2 } + e ^ { - 2 \nu } ( \dot { R } ) ^ { 2 } ).$$
The time derivative of this equation shows that if ˙ ν and ˙ ψ are infinite, then ˙ M will be infinite, which as discussed above is an unphysical condition on the horizon. Finally, with non-singularity of the pressure on the horizon, equation (35) says that ν ′ is finite. Overall, there is no infinity for the redshift of the null geodesic coming from the apparent horizon. Following the last section's discussion, there is no eikonal approximation for the null geodesics coming from apparent horizon (both OMOTS and IMOTS), and hence no emission of Hawking radiation.
## D. Necessary and sufficient conditions
As pointed out by Visser [14], there are two other essential conditions for having black hole radiation for a Lorentzian metric beside the eikonal approximation: the first is existence of an apparent horizon, and the second is non-zero surface gravity.
As regards the apparent horizon, its existence is necessary for defining the black hole boundary in dynamical collapse [3]. The Oppenheimer-Snyder (OS) model has an apparent horizon that consists of two parts, the IMOTS and the OMOTS [7]. For the LTB gravitational collapse model, it was shown in [1, 26] that an apparent horizon will form at R = 2 M after some time, and every LTB gravitational collapsing model has an apparent horizon. For the third (two-fluid) model of gravitational collapse, the scenario is different, because in such a model the gradient of the pressure acts as a repulsive force and can in principle prevent formation of the apparent horizon. However this depends on the details of the gravitational collapse such as the equation of state [33, 41], and for realistic equations of state a black hole will result if the initial mass of the collapsing object is large enough [27, 42]. For studying black hole radiation in the general spherically symmetric case, the surface R = 2 M is again the black hole boundary, which is an apparent horizon.
The surface gravity definition for the first (OS) model is trivial. But the surface gravity definition for a general dynamical metric is not trivial. Having a covariant definition for quantities in thermodynamics is very important because our physical quantities should not depend on the coordinates that we use to calculate. Fortunately, there is nice formula for the thermodynamic law in the spherically symmetric case [43] where the temperature is proportional to the surface gravity. The surface gravity definition of [14] was rewritten in a covariant formalism in the spherically symmetric black hole case in [11]. As shown in [44], and we calculate in equation (C4) for the LTB metric, the surface gravity is non-zero for spherically symmetric dynamical black holes.
## V. THE FUTURE OF THE MATTER FLUX
As we have noted, the spacelike nature of apparent horizon (coming from the in-falling matter or radiation flux) causes light to pass it without having infinite redshift. Let us quantify this property. Consider the future-directed outgoing and ingoing null normals l a and n a respectively at a point, and the expansions θ glyph[lscript] and θ n of the congruences of curves generated by these vector fields.
Let V a be tangential to H (the MOTS hypersurface in Fig.(4)), and orthogonal to the foliation by marginally trapped surfaces. It is always possible to find a function C and normalization of glyph[lscript] a such that V a = glyph[lscript] a -Cn a . In addition, the definition of V a implies that L V θ glyph[lscript] = 0, which gives an expression for C :
$$C = \frac { \mathcal { L } _ { \ell } \theta _ { \ell } } { \mathcal { L } _ { n } \theta _ { \ell } } \,.$$
When C < 0 the apparent horizon is an IMOTS and C > 0 the apparent horizon is an OMOTS, and if C = 0 it becomes an event (isolated) horizon. The value for the C function is important because it shows the type of the black hole horizon. It shows if a MOTS surface is an OMOTS or IMOTS surface and whether it is timelike or spacelike (see Section 1.2 and Table 1 in[13]), and specially, as discussed in the Appendix C, it is a criterion for where the adiabatic approximation is satisfied.
As discussed in [19], in the case of perfect fluid collapse C ∝ ( ρ + ) on the horizon. To see its behaviour p let us look at the Friedmann equation for the standard model of a Λ CDM universe with FLRW metric
$$( \frac { \dot { a } } { a } ) ^ { 2 } = \frac { 8 \pi G } { 3 } \rho + \frac { \Lambda } { 3 },$$
where energy conservation is
$$\dot { \rho } + 3 ( \rho + p ) \frac { \dot { a } } { a } = 0.$$
The background matter density dilutes as ρ m = ρ m a 3 . On the other hand, we know in the expanding background, only part of matter around the black hole can fall into it. For instance, for a de Sitter universe
representing the late evolution of a Λ CDM model, there is a cosmological event horizon. Therefore, after some time the black hole devours all the available matter around itself and after that there is no matter flux which can make the apparent horizon a spacelike surface, i.e C → 0. Note that this is a rough approximation because the matter around the black hole has pressure and the matter energy density increases when it gets near the black hole horizon, so in a more realistic model, C is greater than this approximation and one needs a black hole simulation to find it.
FIG. 4: Particle creation and annihilation around the apparent horizon for the case if an influx that decays away at late times. The OMOTS surface starts at an S MOTS 2-sphere (bottom) and is first spacelike and dynamical; at this stage virtual particle pairs can annihilate, and no radiation results. It then becomes first a slowly evolving horizon, and then at very late times an isolated horizon. At this stage virtual particle pairs can be separated by the OMOTS apparent horizon, and radiation results. Note that we did not include radiation backreaction in this Figure, which is adapted from arXiv:gr-qc/0308033.
## A. The CBR flux value
In the last subsection we did not consider the CBR radiation. As discussed above, the C function (45) is a criterion for a MOTS surface being spacelike [13, 34]. To calculate the radiation flux we consider the CBR as a null fluid that moves inwards from infinity towards the center. In this case Booth [19] has shown that the C function is
$$C = \frac { 2 G } { c ^ { 4 } } A _ { A H } ( \rho + p )$$
where A AH is apparent horizon area. At first sight, it seems that C is very small for the current CBR energy density ρ CBR glyph[similarequal] 10 -14 J/m 3 , but the basic point is that we have to calculate this energy density on the horizon, when CBR photons come from a large distance and get to the horizon. Therefore, the C function is
$$C = \frac { 8 G } { 3 C ^ { 4 } } A _ { A H } \, \rho _ { C B R } \, \frac { w _ { A H } } { w _ { b a c k g r o u n d } }$$
where the w is the CBR photon frequency for Kodama observers [8]. As shown in the Appendix B, for Kodama observers (which reduce to static observers in the static case as discussed ref [8]), the CBR photon frequency becomes infinitely blue shifted when it gets to the apparent horizon. The infinity cannot have any meaning in classical physics. This is the trans-Planckian problem [35] which says that at the very onset of the formation of the trapping horizon, we must consider semiclassical physics. We can roughly use the Planckian cutoff frequency to estimate the value of w AH , which is 10 43 H z . Let us present the general formula for a black hole with mass M and at cosmological redshift time 1 + Z c = a 0 a t ( ) (the usual time scale in astronomy and cosmology). We know that the CBR background frequency is
proportional w 0 ∝ a 0 a ∝ (1 + Z c ) and the CBR energy changes with time as ρ ∝ a 0 a 4 = (1 + Z c ) 4 . All in all, this leads us to this equation for the CBR flux:
$$C \in \frac { 8 G } { 3 c ^ { 4 } } 1 0 ^ { 8 } ( \frac { M } { M _ { \odot } } ) ^ { 2 } \ \rho _ { C B R _ { 0 } } \ \frac { w _ { A H } } { w _ { 0 } } ( 1 + Z _ { c } ) ^ { 3 }$$
If we use the present properties for CBR radiation ρ CBR 0 ∝ 10 -14 and w 0 ∝ 10 10 H z we get
$$C \varpropto 1 0 ^ { - 1 6 } ( \frac { M } { M _ { \odot } } ) ^ { 2 } \ ( 1 + Z _ { c } ) ^ { 3 }$$
This shows that the CBR flux is important for black holes with mass M glyph[greaterorsimilar] 10 7 M glyph[circledot] at the present time, because then C is not negligible and so we then do not have the adiabatic approximation, as shown in Appendix C. But as we know, even if these black holes were radiating we would not see their temperature, because it will be T < 10 -15 k .
Backreaction effects Assume that there is no infalling flux except the CBR flux, we consider now possible backreaction effects. Let us compare the outgoing Hawking radiation flux which decreases the black hole mass, when that radiation is being emitted, with the infalling CBR flux, which cause black holes to grow. If the CBR flux is greater than the Hawking radiation flux the OMOTS remains space like, and if the radiating flux is greater than the CBR flux, the OMOTS will eventually change to a timelike apparent horizon. Now the question is what is the magnitude of the C function for the Hawking radiation flux? Similarly to equation (48), the C function becomes negative, giving a time like surface for the Hawking radiation-only case:
$$C = - \frac { G } { C ^ { 4 } } A _ { A H } ( \rho + p ) _ { H R }.$$
The black hole temperature is T = 6 × 10 -8 M glyph[circledot] M k . Since the flux of the matter is proportional to the energy density [3], we can write:
$$\frac { \mathcal { F } _ { \text{HR} } } { \mathcal { F } _ { \text{CBR} } } = \frac { \rho + p | _ { H R } } { \rho + p | _ { C B R } } = \frac { C _ { H R } } { C _ { C B R } }$$
We can compare this number with the C function of CBR radiation which tells
$$\frac { C _ { H R } } { C _ { C B R } } = - \frac { [ 6 \times 1 0 ^ { - 8 } \frac { M _ { \odot } } { M } ] ^ { 4 } } { [ 2. 7 \ ( 1 + Z _ { c } ) ] ^ { 4 } }$$
For a solar mass black hole this expression is C HR C CBR = 10 -32 at the present time, which is very small. For black holes with mass M > 10 -8 M glyph[circledot] the infalling CBR flux is greater than the radiation flux and OMOTS remains spacelike. Only for a black hole with mass M glyph[lessorsimilar] 10 -8 M glyph[circledot] are the two fluxes comparable and the black hole can have radiation which decreases the black hole area. This equation actually rules out astrophysical black hole radiation backreaction effects that could make the OMOTS surface timelike, if there indeed is any Hawking radiation. Hence for such black holes the OMOTS surface is necessarily inside the classical event horizon.
However, a primordial black hole mass can be less than this bound, then the CBR flux is negligible for them; but the matter flux value may still be significant, and its effect must be considered by modelling.
## B. The future of the CBR flux
One may ask how long this scenario will remain true for a cosmological black hole when the universe expansion reduces the CBR flux effectively to zero in the far future.
Let us calculate the time evolution of the CBR flux. For a universe dominated by radiation with p = ρ/ 3 one has
$$\frac { \rho _ { C B R } ( t _ { 2 } ) } { \rho _ { C B R } ( t _ { 1 } ) } = \frac { a ( t _ { 1 } ) ^ { 4 } } { a ( t _ { 2 } ) ^ { 4 } }.$$
In the case of a matter dominated universe this term becomes
$$\frac { \rho _ { C B R } ( t _ { 2 } ) } { \rho _ { C B R } ( t _ { 1 } ) } = \frac { ( t _ { 1 } ) ^ { 8 / 3 } } { ( t _ { 2 } ) ^ { 8 / 3 } }$$
while in the dark energy dominant era it is
$$\frac { \rho _ { C B R } ( t _ { 2 } ) } { \rho _ { C B R } ( t _ { 1 } ) } = e ^ { 4 H ( t _ { 1 } - t _ { 2 } ) }$$
where H = ˙ a a . The present value of the Hubble expansion rate is H 0 = 3 241 . × 10 -18 hs -1 where h glyph[similarequal] 0 72 . ± 0 1. . As discussed in last subsection, if we assume that the C value decreases 2 orders of magnitude and OMOTS becomes nearly a null surface relative to the cosmological frame, this approximately means that the energy density decrease 2 orders of magnitude (the C function is key to the validity of adiabatic condition (see Appendix C), hence we can use it to find approximately when the black hole radiation will start.) One can easily see that the times when the energy density of the CBR decreases to 1 100 of that at the present time in these two models are t 2 glyph[similarequal] 2 44 . × 10 18 s and t 2 glyph[similarequal] 7 × 10 17 s from the present time respectively. Similar calculations can be done for the matter flux. Therefore, we can say that the CBR and matter flux effect effectively cannot be diluted by the cosmological expansion in the short term. In the other words, these calculations show that the cosmological expansion is not significant in turning off the matter and CBR flux on cosmologically short time scales.
However, leaving aside strange behaviour such as dynamical dark energy, cosmological particle creation, and big crunch singularity models, this picture says that at very large times there is effectively no CBR flux. Although it never becomes exactly null or timelike ([7], section VIII.2), the OMOTS eventually becomes an isolated horizon that is very close to the classical event horizon. Hawking Radiation emission could occur from that time on as depicted in Fig.(5), because the approximations we have used above in the eikonal analysis would break down at that stage, with associated backreaction effects and possibly eventual black hole evaporation.
FIG. 5: Penrose diagram for a black hole in a Λ CDM expanding universe, where there is a matter and radiation influx, but it dies away to zero in the far future. Note that we do not represent back reaction effects in this picture.
This is the key difference from the particle approximation and associated tunneling picture, because on that view, there would never be Hawking radiation emission, as the OMOTS surface would always be spacelike, so no Hawking Radiation backreaction effects or black hole evaporation would occur even in the very far future.
## VI. EXPONENTIAL APPROXIMATION FOR THE AFFINE PARAMETER
This paper so far has been based on examining the Eikonal approximation and its implications. However the original calculation by Hawking [8] was based on the way that the affine parameters on past null infinity I -and future null infinity I + are related to each other by an exponential transformation, so an infinite range of the parameter v on I -corresponds to a finite range of the parameter u on I + (see Figure 4 of [8]). Following this, in order to find the minimal condition for having thermal-like black hole radiation, Visser et al [10] have shown that any collapsing compact object (regardless of whether or not any type of horizon ever forms) will emit a slowly evolving Planckian flux of quanta, provided the adiabatic condition and exponential approximation hold along the null congruence which travels from past null infinity to future null infinity. Having such a relation for wave packets and applying the standard Hawking calculation [8] to derive time-dependent Bogoliubov coefficients, we can see this gives us a Planckian spectrum with a time-dependent Hawking temperature. Thus it was shown there that the exponential relation between the affine parameters on past null infinity, v in , and future null infinity, u out , is the necessary and sufficient condition for generating a Hawking flux.
In the case of black hole formation in a vacuum context, this leads to Hawking's picture. However in the case of a spacelike dynamical horizon as considered in this paper, the pile up of the u-parameter relative to the v-parameter does not reach infinity, because the OMOTs is spacelike (see Fig.(6)). The relevant null geodesics (for u →∞ ) cross the apparent horizon (the OMOTS surface) and do not reach future null infinity [7]. The null coordinate u does not tend to infinity there, and hence surfaces of constant w of the solution p w will not pile up at this surface as in the vacuum case considered by Hawking ([8], just after Fig 4).
FIG. 6: Penrose diagram for black hole formation with incoming radiation that does not decay away. A spacelike OMOTS surface means the infinite affine parameter rescaling between ingoing and outgoing null geodesics that occurs in the pure vacuum case ([8], Fig. 4) does not reach infinity: it is trapped by the singularity.
Putting this in the LCDM context where future infinity is spacelike [7], although there is a unique
event horizon which is the inmost event horizon associated with the black hole, as the CBR becomes cooler and cooler, asymptotically the OMOTS will (in the preferred cosmological frame) tend to being null from being spacelike, but will never actually become null (Fig.(5)). It is then a delicate issue of how the limits work as to whether black hole radiation results from the null geodesics which pass near the slowly evolving horizon at very late times.
One might be concerned that neither the WKB approximation nor apparent horizons are necessary for the Hawking effect, see [40] and [45]. The point however is that we don't consider the most general features of spherical collapse and Hawking radiation. We look for the Hawking radiation phenomenon in the case of cosmological black holes which do indeed have apparent horizons [7]. We also have used the tunnelling approach [11, 12, 14, 30], for which the WKB approximation is a necessary assumption. As shown above, this means that we need a surface that has a very large redshift (not necessarily infinite) for the received light that is emitted from near the apparent horizon, and this leads to our results. In addition, to check the outcome, we have applied the adiabatic condition for cosmological black holes in Appendix C, which is another approach that confirms our results.
## VII. DISCUSSION AND CONCLUSIONS
There are several effects related to the vacuum in quantum field theory: the zero point energy, Casimir effect, and dynamical Casimir effect (which is like a moving mirror). There are three more quantum vacuum effects that are due to a curved spacetime: metric quantum fluctuations, vacuum polarization, and Hawking particle creation [27]. Particle creation results from the Hawking particle creation effect and dynamical Casimir effect (or moving mirror). Basically, Hawking particle creation is thermal radiation due to the black hole horizon which is observed at a large distance. On the other hand, the dynamical Casimir effect is the production of particles and energy from an accelerated moving mirror (see [46] for a comprehensive review). The question is how we can see this effect in a curved space time or in black hole collapse. In a flat space time, the dynamical Casimir effect appears due to a special boundary condition for wave solutions. We can include this effect in a curved space time by solving the general wave equation (including backscattering and boundary conditions) and finding the appropriate vacuum for it (which is like an initial condition). Hence, we can see traces of this effect in the expectation value of energy momentum tensor < T µν > as vacuum polarization [22].
In our dynamical spherically symmetric space time where the apparent horizon is inside the event horizon, any virtual pair particle created due to these effects cannot fall outside the event horizon, become real, and reach future infinity; however it can be seen locally in < T µν > inside horizon. But for any general realistic model of star collapse, we must first solve the wave equation with suitable boundary conditions (the dynamical Casimir effect has dynamical boundary conditions) and second find the appropriate vacuum corresponding to the collapsing model (like the Unruh vacuum), and only then can we read the general particle creation occurring from the number density operator < N = a a † k k > and < T µν > . It seems that in the analogue gravity models for Hawking radiation, we can see the dynamical Casimir effect [47]. One can consider collapsing star models with special dynamical boundary condition to probe the dynamical Casimir effect for particle creation, which is beyond the scope of this paper. Here we just discuss the blackbody particle creation due to Hawking radiation.
We have shown that when one uses either the tunneling approximation [12], [11] or the eikonal approximation [14], one finds that turning on the CBR and matter flux turns off the Hawking radiation emission from the OMOTS surface, and there is also no such radiation emitted from the IMOTS surface. The key feature leading to this result is shown in the contrast between the two cases in Fig.(1) and Fig.(5):
- · When we have the vacuum case outside the star, the OMOTS surface coincides with the null event horizon. Radiation emitted close enough to this surface reaches infinity with an unboundedly large redshift. It is this divergent redshift that is the reason Hawking radiation is emitted just outside the null OMOTS surface and escapes to infinity. The ultimate source of the divergent redshift is the infinite rescaling that takes place between the affine parameter and the group parameter on a bifurcate Killing horizon [24] (cf. equation (2.16) in [8]); hence it is a consequence of the static nature of the exterior vacuum solution, which allows this symmetry group.
- · When we take into account the infalling CBR radiation and matter flux, Fig.(5), the OMOTS
surface is spacelike and lies inside the null event horizon [7]. Outgoing rays reaching infinity from just inside its classical event horizon, or emitted inside that horizon but outside the OMOTS surface, no longer experience this unbounded redshift; and the same applies to the IMOTS surface. That is the reason that no Hawking radiation is emitted in this case.
This is related to the fact that when the flux falls into the black hole, it's mass increases, hence this is no longer a quasi-static spacetime and there is no external Killing vector field with an associated bifurcate Killing horizon. The conclusion is reinforced by the fact that there will be many other kinds of radiation and matter that will also fall into an astrophysical black hole, and increase its mass further. This is a self consistent approximation: as long as there is CBR and matter influx, the OMOTS surface will remain spacelike at all times because there is no incoming negative density Hawking radiation that could make it timelike [7].
- · As depicted in Fig.(5), in the particle tunnelling scenario, if a pair of a particle and antiparticle are created near the dynamical horizon, both of them will fall into the singularity and annihilate each other, so no Hawking radiation will be emitted. However a particle created near an isolated horizon or slowly evolving horizon can reach future infinity, and Hawking radiation will occur. On the other hand, since the geometric optics approximation does not generally held near the OMOTS, we cannot apply the particle interpretation for this surface.
- · At a late enough time in the very far future, the cosmological expansion decreases the CBR and matter density and black hole will devour all the available matter around itself. The black hole horizon then becomes first a slowly evolving horizon [4] and then an isolated horizon [2] Fig.(5). Black body radiation could then initiate at that stage, and possibly lead to black hole explosions at a later time.
- · As long as the matter flux into the black hole is not negligible there is no Hawking radiation, but the black hole radiation scenario is currently applied to every dynamical black hole. However, all black holes in the real universe are surrounded by different types of matter and radiation leading to substantial positive density energy influx for much of their life, and particularly in the very early universe.
- · Application of this constraint to primordial black hole evaporation modelling may bring in a correction to their abundance in the cosmos. Specifically, primordial black holes are candidate progenitors of unidentified Gamma-Ray Bursts (GRBs) that are supposed to detect by the Fermi Gamma-ray Space Telescope observatory. Their abundance might be lowered when the above considerations are taken into account.
This is all in accord with the discussions in [7, 13], and leads to the conclusion that in a realistic cosmological context, a black hole forming from the collapse of a star in a universe permeated by CBR and matter will not emit Hawking radiation in the past or at the present, and so emission of such radiation from them, or evaporation of such black holes in an explosion, will not occur in the visible universe. To what degree this affects primordial black holes, or thefar future universe, will be very context dependent and will need detailed modelling.
One should contrast the above with the Parikh and Wilczek tunnelling method (in standard general relativity) [12], which cannot apply for particle production whenever the OMOTS surface is spacelike, because the whole concept of tunnelling only makes sense for a timelike surface, where 'inside' and 'outside' are well defined concepts (see [12]: paragraphs just after eqn. (8)). Therefore, the tunnelling picture is only applicable from the moment that the instantaneous Hawking radiation flux is greater than the matter flux and the black hole apparent horizon becomes a timelike surface.
Generally, in broader contexts, we cannot say that the tunnelling method requires the apparent horizon to be timelike, due to particle creation by tunnelling. One can apply this method even for spacelike universal horizons in Einstein-aether theory [48, 49] which has a non-standard causal structure. Note also that the adiabatic approach does not limit the horizon to be a timelike or spacelike surface. The point is that the adiabatic condition can be satisfied for an apparent horizon that becomes spacelike due to the CBR flux. In addition, in our models, we quantify the WKB condition in the tunnelling method as the light redshift from the horizon, which is not a comprehensive approach. One can quantify this as the width of the horizon in order to calculate the lingering time in acoustic black holes or
Lorentz-invariance violating models [49, 50].
As mentioned above and emphasized in [7], this work should be extended in two significant ways:
- · By calculating the expectation value of the stress tensor in this scenario (cf. [16]),
- · By checking that radiation emission from within the collapsing fluid does not prevent horizon formation by back reaction of emitted Hawking Radiation, as has been suggested in [38]. That calculation would be altered in the cosmological context considered here, where inter alia the Hartle-Hawking vacuum is not the appropriate vacuum state to use, and the spacelike nature of the OMOTS surface will modify the way modes propagating through the collapsing fluid [8, 16] reach infinity.
However in terms of calculation methods that are used by various authors for determining the extent of Hawking radiation emission, the result given here seems conclusive. Note that we do not claim to prove that no radiation at all will be emitted. The break down of the adiabatic approximation implies only absence of a Planckian spectrum, not necessarily of any radiation. Indeed, a rapidly evolving apparent horizon would most probably lead to some form of particle production with typical frequencies excited of the order of the inverse timescale of the evolution of the metric (maybe linked to the accretion rate). This is a (geometrically induced) Dynamical Casimir effect, which per se can be far from thermal in character but not necessarily negligible (because the time dependence of the apparent horizon is supposed to be fast/non-adiabatic in the early universe).
## Acknowledgments:
We thank Malcolm Perry and Reza Mansouri for fruitful discussions, and Ritu Goswami, Tim Clifton, David Jacobs and Matt Visser for helpful comments on an earlier version of this paper, as well as a referee for useful comments. We thank the National Research Foundation (South Africa) and the University of Cape Town Research Fund for support.
- [1] A. Krasinski and C. Hellaby, Phys. Rev. D 69, 023502 (2004); W. Valkenburg, Gen.Rel.Grav. 44 (2012) 2449-2476;C. Gao, X. Chen, Y-G. Shen and V Faraoni, Phys.Rev. D84 (2011) 104047; J. T. Firouzjaee, R Mansouri, Gen. Rel. Grav. 42, 2431 (2010); K. Bolejko, M-N. Celerier and A. Krasinski, Class.Quant.Grav. 28 (2011) 164002; J. T. Firouzjaee, M. Parsi Mood and R. Mansouri, Gen. Rel. Grav. 44, 639 (2012).
- [2] A. Ashtekar, C. Beetle, O. Dreyer, S. Fairhurst, B. Krishnan, J. Lewandowski and J. Wisniewski, Phys. Rev. Lett. 85 , 3564 (2000).
- [3] A. Ashtekar and B. Krishnan, Phys. Rev. Lett. 89, 261101 (2002); A. Ashtekar and B. Krishnan, Phys. Rev. D 68, 104030 (2003).
- [4] I. Booth and S. Fairhurst, Phys. Rev. Lett. 92 , 011102 (2004)
- [5] A. Ashtekar and B. Krishnan, Living Rev. Rel. 7 , 10 (2004); J. L. Jaramillo, R. P. Macedo, P. Moesta and L. Rezzolla, Phys. Rev. D 85 , 084030 (2012).
- [6] T. Harada and B. J. Carr, Phys. Rev. D 71 , 104010 (2005); D. C. Guariento, J. E. Horvath, P. S. Custodio and J. A. de Freitas Pacheco, Gen. Rel. Grav. 40 , 1593 (2008); I. Musco, J. C. Miller and L. Rezzolla, Class. Quant. Grav. 22 , 1405 (2005).
- [7] G F R Ellis, R Goswami, A I. M. Hamid and S D. Maharaj arXiv:1407.3577 [gr-qc].
- [8] S W Hawking (1975) Comm Math Physics 43 : 199-220.
- [9] S. Hollands and R. M. Wald, arXiv:1401.2026 [gr-qc].
- [10] C. Barcelo, S. Liberati, S. Sonego and M. Visser, JHEP 1102 , 003 (2011); C. Barcelo, S. Liberati, S. Sonego and M. Visser, Phys. Rev. D 83 , 041501 (2011)
- [11] R. Di Criscienzo, S. A. Hayward, M. Nadalini, L. Vanzo and S. Zerbini, Class. Quant. Grav. 27 , 015006 (2010); J. T. Firouzjaee and R. Mansouri, Europhys. Lett. 97 , 29002 (2012)
- [12] M. K. Parikh and F. Wilczek, Phys. Rev. Lett. 85 , 5042 (2000).
- [13] G. F R Ellis, arXiv:1310.4771 [gr-qc].
- [14] M. Visser, Int. J. Mod. Phys. D 12 , 649 (2003).
- [15] S W Hawking (1974) Nature 248 , 30-31.
- [16] N D Birrell and P C W Davies (1984) Quantum Fields in Curved Space (Cambridge: Cambridge University Press)
- [17] S W Hawking and G F R Ellis (1973) The Large Scale Structure of Space-Time (Cambridge: Cambridge University Press).
- [18] E. T. Akhmedov, T. Pilling, and D. Singleton, Int.J.Mod.Phys.D17:2453-2458, (2008) ; V. Akhmedova, T. Pilling, A. de Gill, and D. Singleton, Phys. Lett. B 666 (2008) 269 .
- [19] I. Booth, L. Brits, J. A. Gonzalez and C. Van Den Broeck, Class. Quant. Grav. 23 , 413 (2006) [gr-qc/0506119].
- [20] R H Boyer (1969) Proc. R. Soc. Lond. A:311 245-252.
- [21] T Clifton (2008) Class.Quant.Grav . 25 :175022 [arXiv:0804.2635]
- [22] P. C. W. Davies (1976) Proc Roy Soc (London) 351 : 129-139.
- [23] P C W Davies, S A Fulling, and W G Unruh, Phys. Rev. D13, 2720 (1976) .
- [24] R. H. Boyer (1969) Proc. R. Soc. Lond. A 311:245-252
- [25] A. Einstein and E.G. Straus, Rev. Mod. Phys. 17, 120 (1945)
- [26] J. T. Firouzjaee, Int. J. Mod. Phys. D 21 , 1250039 (2012)
- [27] V. P. Frolov and I. D. Novikov, Black hole physics: basic concepts and new developments , Kluwer Academic Publishers, Dordrecht 1998.
- [28] J.B. Hartle and S.W. Hawking, Phys. Rev. D 13, 2188 (1976).
- [29] P. Hajicek, Phys. Rev. D 36, 1065 (1987)
- [30] S. Shankaranarayanan, T. Padmanabhan and K. Srinivasan, Class. Quant. Grav. 19 , 2671 (2002).
- [31] R. C. Tolman, Proc. Natl. Acad. Sci. U.S.A. 20, 410 (1934); G. Lemaˆ ıtre, Ann. Soc. Sci. Bruxelles I A53, 51 (1933); H. Bondi, Mon. Not. R. Astron. Soc. 107, 343 (1947); K Bolejko, A Krasinski, C W Hellaby and M-N Clrier, Structures in the Universe by Exact Methods - Formation, Evolution (Cambridge University Press, 2009)
- [32] G F R Ellis, R Maartens and M A H MacCallum, Relativistic Cosmology (Cambridge University Press).
- [33] P. S. Joshi, Gravitational Collapse and Spacetime Singularities (Cambridge University Press).
- [34] W. Kavanagh and I. Booth, Phys. Rev. D 74 , 044027 (2006).
- [35] T. Jacobson, Phys. Rev. D 48 , 728 (1993); L. C. Barbado, C. Barcelo, L. J. Garay and G. Jannes, JHEP 1111 , 112 (2011).
- [36] H. Kodama, Prog. Theor. Phys. 63, 1217 (1980).
- [37] A. B. Nielsen and J. T. Firouzjaee, Gen. Rel. Grav. 45 , 1815 (2013).
- [38] L Mersini-Houghton, arXiv:1406.1525 [hep-th]
- [39] L. Parker, Phys. Rev., 183, 1057 (1969).
- [40] C. Barcelo, S. Liberati, S. Sonego and M. Visser, Phys. Rev. D 77 , 044032 (2008) .
- [41] R. Moradi, J. T. Firouzjaee and R. Mansouri, arXiv:1301.1480 [gr-qc].
- [42] C W Misner, K S Thorne and J A Wheeler (1973) Gravitation (Freeman).
- [43] S. A. Hayward, Class. Quant. Grav. 15 , 3147 (1998).
- [44] M. Pielahn, G. Kunstatter and A. B. Nielsen, Phys. Rev. D 84 , 104008 (2011).
- [45] M. Smerlak and S. Singh, Phys. Rev. D 88 , no. 10, 104023 (2013).
- [46] Michael Bordag, Galina Leonidovna Klimchitskaya, Umar Mohideen, and Vladimir Mikhaylovich Mostepanenko, Advances in the Casimir Effect (Oxford University Press).
- [47] C. Barcelo, S. Liberati and M. Visser, Living Rev. Rel. 8 , 12 (2005) [Living Rev. Rel. 14 , 3 (2011)]; I. Carusotto, S. Fagnocchi, A. Recati, R. Balbinot and A. Fabbri, New J. Phys. 10 , 103001 (2008).
- [48] P. Berglund, J. Bhattacharyya and D. Mattingly, Phys. Rev. Lett. 110 , no. 7, 071301 (2013).
- [49] B. Cropp, S. Liberati, A. Mohd and M. Visser, Phys. Rev. D 89 , 064061 (2014).
- [50] S. Finazzi and R. Parentani, Phys. Rev. D 83 , 084010 (2011).
- [51] It has been commented that horizons (whether apparent horizons or event horizons) do not somehow 'emitted' particles/radiation because quantum field theory in curved spacetime does not describe the Hawking process in this manner at all. But in our view this is just semantics. There are papers that show that the radiation appears to be emitted from the region of space-time near the apparent horizon, which this seems a good enough justification for saying 'it is emitted from the vicinity of the apparent horizon'.
## Appendix A: Junction condition inside the horizon
To construct the Oppenheimer-Snyder model inside the Schwarzchild horizon, we need a coordinate system that does not have a singularity on the horizon. We choose the Lemaˆ ıtre coordinate system [27] which is similar to the FLRW and LTB comoving coordinates. Then
$$d s ^ { 2 } = - d t ^ { 2 } + R ^ { \prime 2 } d r ^ { 2 } + R ( t, r ) d \Omega ^ { 2 }$$
where
$$R = ( 2 M ) ^ { 1 / 3 } \left ( \frac { 3 } { 2 } ( r - ( t - t _ { 0 } ) ) \right ) ^ { 2 / 3 }.$$
The singularity is at R = 0. Since the induced metric must be the same on both sides of the star surface at χ 0 , we get
$$R ( \tau ) | _ { \chi _ { 0 } } & = a ( \tau ) \chi _ { 0 } & ( A 3 )$$
and
$$( \frac { d t } { d \tau } ) ^ { 2 } | _ { x _ { 0 } } - R ^ { \prime 2 } ( \frac { d r } { d \tau } ) ^ { 2 } | _ { x _ { 0 } } = 1$$
The unit normals ( n n µ µ = 1) to the star surface are n µ -= (0 , a, 0 0) and , n + µ = ( -dr dτ , dt dτ , 0 0). , Using these normal vectors to calculate the extrinsic curvature lead us to these equations:
$$K _ { - \theta } ^ { \theta } = K _ { - \phi } ^ { \phi } = \frac { 1 } { a \chi }$$
and
$$K _ { + \theta } ^ { \theta } = K _ { + \phi } ^ { \phi } = \frac { d r } { d \tau } \frac { \dot { R } } { R } + \frac { d t } { d \tau } \frac { 1 } { R ^ { \prime } R }$$
Matching the two extrinsic curvatures on the star surface we get,
$$( \frac { d r } { d \tau } \frac { \dot { R } } { R } + \frac { d t } { d \tau } \frac { 1 } { R ^ { \prime } R } ) | _ { \chi _ { 0 } } = \frac { 1 } { a \chi _ { 0 } }$$
The equations (A3, A4, A7) give the evolution of the star surface in the two space times.
## Appendix B: Areal coordinate for dynamical metric
The r coordinate which we used as a radial coordinate in the fluid is a comoving coordinate, where the time coordinate is proper time for this observer. In contrast to the stationary metric in the vacuum, which has a preferred Killing observer, there is no preferred observer in the general dynamical metric. In the case that we want to calculate the redshift for a CBR photon which comes from a large distance to the apparent horizon, we need better coordinates. In the spherically symmetric case, there is a well defined family of observers which can be attributed to the areal coordinate R t, r ( ) [36], and which are Kodama observers. The function R t, r ( ) is a good candidate to describe the particle distance from center, and is called an areal coordinate (it is the angular diameter distance in cosmology). By taking the areal radius as a new coordinate and using the relation dR = R dr ′ + ˙ Rdt for the LTB metric one obtains
$$d s ^ { 2 } = ( \frac { \dot { R } ^ { 2 } } { 1 + f } - 1 ) d t ^ { 2 } + \frac { d R ^ { 2 } } { 1 + f } - \frac { 2 \dot { R } } { 1 + f } d R d t + R ( t, r ) ^ { 2 } d \Omega ^ { 2 }.$$
The (t,R) coordinates are usually called areal coordinates. This LTB metric form is similar to the Painlev´ e form of the Schwarzschild metric. In the case of f = 0, the metric is the same as the Painlev´ metric form. e
The redshift (43) measured by Kodama observer with 4-velocity K i = √ 1+ f ( √ 1 -2 m R ) R ′ ( R , ′ -˙ R ) [11], can be written as
$$\text{written as} \\ 1 + z \, = \, c _ { 0 } \, \frac { \left ( \frac { \sqrt { 2 ^ { m } } _ { R } + f - \sqrt { 1 + f } } { \sqrt { 1 - \frac { 2 ^ { m } } { R } } } \right ) _ { e } } e x p \left ( - \int _ { o } ^ { e } \frac { \dot { R } ^ { \prime } } { \sqrt { 1 + f } } d r \right ) = c _ { 0 } \ e x p \left ( - \int _ { o } ^ { e } \pm \frac { \dot { R } ^ { \prime } } { \dot { R } + \sqrt { 1 + f } } d R \right ). \ ( B 2 ) \\ \nu _ { 1 - 1 } + \dots \dots + \lambda \dots \dots + \lambda \dots \dots + \lambda \dots \dots + \lambda \dots \dots + \lambda \dots \dots + \lambda \dots \dots + \lambda \dots \dots + \lambda \dots \dots + \lambda \dots \dots + \lambda \dots \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \ dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots - \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \lambda \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \ dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots - \lambda \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \ dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots + \beta \dots - \lambda \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \ dots + \alpha \ dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots - \lambda \dots - \lambda \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \sigma \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \sigma \dots +\alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha + \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \dots + \alpha \eq \\ \nu _ { 1 } }$$
The ± refer to the outgoing and ingoing null geodesic respectively. Using the Einstein equation (25), we get the following equation:
$$\text{get the following equation} \colon \\ 1 + z = c _ { 0 } \, \frac { \left ( \frac { \sqrt { \frac { 2 m } { R } + f } - \sqrt { 1 + f } } { \sqrt { 1 - \frac { 2 m } { R } } } \right ) _ { e } } e _ { \ } e x p \left ( - \int _ { o } ^ { e } \pm \frac { \dot { R } ^ { \prime } } { - \sqrt { \frac { 2 M } { R } + f } + \sqrt { 1 + f } } d R \right ).$$
The exponential part can be either e -∞ or e + ∞ on the apparent horizon. For the ingoing null geodesic (with - sign), which comes from a large distance e to the apparent horizon o , light becomes infinitely blue shifted: (1+ z ) → 0. Note that the coordinate R (comoving for a Kodama observer), like the Schwarzchild areal coordinate given by the R = constant surface, is a spatial coordinate outside the horizon ( R > 2 M ) but becomes a time coordinate inside the horizon. Therefore, this coordinate cannot be used for inside the apparent horizon R < 2 M .
## Appendix C: Adiabatic condition for dynamical metric
An equivalent statement to the eikonal approximation is an adiabatic condition [10, 14]. This condition says that the need for slow evolution of the geometry is hidden in the approximation used to write the modes as e iwt multiplied by a position-dependent fact or. This makes sense only if the geometry is quasistatic on the timescale set by w . This condition can be used as a criterion for having Hawking radiation [37]. Since this condition is an essential condition for black hole radiation, we want to examine it for a dynamical metric in the Painlev´-Gullstrand coordinates, e
$$d s ^ { 2 } = - ( c ( t, r ) ^ { 2 } - v ( t, r ) ^ { 2 } ) \ d t ^ { 2 } - 2 v d t \ d R + d R ^ { 2 } + R ^ { 2 } d \Omega ^ { 2 }$$
The adiabatic condition says that the peak in the Planck spectrum is meaningful when
$$k T \approx w _ { p e a k } \gg m a x \{ | \dot { c } / c |, | \dot { v } / v | \}$$
Now we examine this condition for the LTB metric (B1) in this coordinate. The left hand side of this equation is proportional to the surface gravity [11] which is finite. Without giving the details of the calculation for the v = ˙ R 1+ f function, we get
$$\kappa _ { A H } \gg | \dot { v } / v | = \frac { M ^ { \prime } } { \sqrt { 1 + f R R ^ { \prime } } }$$
Using the surface gravity for the LTB metric [11]
$$\kappa _ { H } = \frac { 1 } { 2 } g ^ { \mu \nu } \nabla _ { \mu } \nabla _ { \nu } R = \frac { 1 } { 2 R } - \frac { M ^ { \prime } } { 2 R R ^ { \prime } },$$
and the C function for LTB [26], C = 2 √ 1 + f M M ′ R ′ -′ , one gets
$$1 \gg C.$$
Therefore, to examine the adiabatic condition for the LTB metric, it is sufficient to check the C function.
Conclusion: Surprisingly, only in the case of a slowly evolving horizon obeying (C5) can a LTB dynamical black hole radiate. As a result, the adiabatic condition is also satisfied for an isolated horizon where C = 0 ⇔ M ′ = 0.
As another example, assume that we have ingoing radiation falling into the black hole. The suitable metric for this case is Vaidya ingoing form:
$$d s ^ { 2 } = - ( 1 - \frac { 2 M ( v ) } { r } ) d v ^ { 2 } + 2 d v d r + r ^ { 2 } d \Omega ^ { 2 },$$
where the apparent horizon is located at r = 2 M . Using equation (C4), we get κ H = 1 4 M v ( ) for the surface gravity. Hence, the adiabatic condition (C2) says
$$1 \gg \frac { d M ( v ) } { d v }.$$
This means that the incoming matter flux must be small to satisfy the adiabatic condition. On the other hand, the C function for a Vaidya metric that has a slowly evolving horizon is [34]
$$1 \gg C = \frac { d M ( v ) } { d v }.$$
As a result, similarly to the LTB case, the adiabatic condition allows Hawking radiation for a Vaidya black hole that has a slowly evolving horizon or an isolated horizon.
|
10.1007/s10714-014-1848-2
|
[
"Javad T. Firouzjaee",
"George F. R. Ellis"
] |
2014-08-04T19:25:31+00:00
|
2014-12-06T05:49:55+00:00
|
[
"gr-qc",
"astro-ph.CO",
"hep-th"
] |
Cosmic Matter Flux May Turn Hawking Radiation Off
|
An astrophysical (cosmological) black hole forming in a cosmological context
will be subject to a flux of infalling matter and radiation, which will cause
the outer apparent horizon (a marginal trapping surface) to be spacelike
\cite{ellisetal14}. As a consequence the radiation emitted close to the
apparent horizon no longer arrives at infinity with a diverging redshift.
Standard calculations of the emission of Hawking radiation then indicate that
no blackbody radiation is emitted to infinity by the black hole in these
circumstances, hence there will also then be no black hole evaporation process
due to emission of such radiation as long as the matter flux is significant.
The essential adiabatic condition (eikonal approximation) for black hole
radiation gives a strong limit to the black holes that can emit Hawking
radiation. We give the mass range for the black holes that can radiate,
according to their cosmological redshift, for the special case of the cosmic
blackbody radiation (CBR) influx (which exists everywhere in the universe). At
a very late stage of black hole formation when the CBR influx decays away, the
black hole horizon becomes first a slowly evolving horizon and then an isolated
horizon; at that stage, black hole radiation will start. This study suggests
that the primordial black hole evaporation scenario should be revised to take
these considerations into account.
|
1408.0779v2
|
## Relativistic superfluidity and vorticity from the nonlinear Klein-Gordon equation
Chi Xiong 1 , Michael R.R. Good 1 , Yulong Guo 2 , Xiaopei Liu 3 , and Kerson Huang 1 4 ; GLYPH<3> 1 Institute of Advanced Studies, Nanyang Technological University, Singapore 639673 2 School of Computer Science and Engineering, South China University of Technology, Guangzhou, China 510641 3 School of Computer Engineering, Nanyang Technological University, Singapore 639673 and 4 Physics Department, Massachusetts Institute of Technology, Cambridge, MA, USA 02139
We investigate superfluidity, and the mechanism for creation of quantized vortices, in the relativistic regime. The general framework is a nonlinear Klein-Gordon equation in curved spacetime for a complex scalar field, whose phase dynamics gives rise to superfluidity. The mechanisms discussed are local inertial forces (Coriolis and centrifugal), and current-current interaction with an external source. The primary application is to cosmology, but we also discuss the reduction to the non-relativistic nonlinear Schr¨dinger equation, which is widely o used in describing superfluidity and vorticity in liquid helium and cold-trapped atomic gases.
Keywords: relativistic superfluidity, quantized vorticity, nonlinear Klein-Gordon equation PACS number: 03.75.Kk, 03.75.Lm, 04.62. + v, 11.10.Lm
## I. INTRODUCTION
In quantum mechanics, a system of particles is describable by a complex wave function, which has a modulus as well as a phase, and the existence of the quantum phase is an essential distinction between quantum mechanics and classical mechanics. Macroscopic phase coherence (correlation of the quantum phase over macroscopic distances) gives rise to superfluidity and occurs in Bose-Einstein condensates (BEC) in diverse systems [1-3]: liquid 4 He at low temperatures; cold trapped atomic gases; central region in relativistic heavyion collisions; Higgs or Higgs-like fields over cosmological scales; superconducting metals; liquid 3 He at low temperatures; interior of neutron stars. The last three refer to superconductivity, which can be viewed as the superfluidity of condensed fermion pairs. Ginzburg and Landau [4] propose a general phenomenological theory that describes the phase coherence in terms of a complex scalar field, which is viewed as an order parameter that emerges in a phase transition below a critical temperature. This phase transition is associated with spontaneous breaking of global gauge symmetry, i.e., the invariance of the wave function under a constant change of phase. We adopt such an approach here, and, in view of cosmological applications, begin with a complex scalar field in curved spacetime, and then consider the flat-spacetime and non-relativistic limits. Our treatment is confined to the neighborhood of the absolute zero of temperature.
We denote the complex scalar field by
$$\Phi = F e ^ { i \sigma }, \quad \ \ \ \ ( 1 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
which satisfies a nonlinear Klein-Gordon equation (NLKG) of the form GLYPH<3> GLYPH<8> + f ( GLYPH<8> ) = 0 ; where the d'Alembertian operator
GLYPH<3> Email: [email protected]
Email: [email protected]
Email: [email protected]
Email: [email protected]
Email: [email protected]
GLYPH<3> is the generalization of @ GLYPH<22> @GLYPH<22> = r 2 GLYPH<0> 1 c 2 @ 2 @ t 2 to curvilinear coordinate frames, and f is a nonlinear function, which contains a potential that has a minimum at GLYPH<8> , 0. Thus, there exists a nonzero vacuum field, which breaks global gauge symmetry spontaneously. The superfluid velocity is proportional to the spatial gradient of the phase r GLYPH<27> , which has quantized circulation:
$$\oint _ { C } \nabla \sigma \cdot d s = 2 \pi n \ ( n = 0, \pm 1, \pm 2, \dots ) \,, \quad \ ( 2 )$$
The integral extends over a spatial closed loop C , and the quantization is a consequence of the fact that GLYPH<8> , and hence the phase factor e i GLYPH<27> , must be a continuous function in space. The contour C encircles a vortex line, which may meander in space but must end on itself, forming a closed loop, or terminate on boundaries. The modulus F must vanish on the vortex line, where it approaches zero continuously, over a characteristic distance, the healing length. Thus the vortex line is in actuality a vortex filament with a finite e GLYPH<11> ective radius.
The superfluid velocity v s is a hydrodynamic quantity that obeys certain conservation laws. As discussed below, it has the form
$$\frac { \mathbf v _ { s } } { c } = \xi _ { s } \nabla \sigma,$$
where GLYPH<24> s is a correlation length that is generally spacetime dependent, with GLYPH<24> s ! ~ = mc in the non-relativistic limit, where m is a mass scale. The fact that the modulus F must vanish at the vortex center e GLYPH<11> ectively makes the space non-simply connected, i.e., there are closed circuits that cannot be shrunken to zero continuously. This is why we can have r GLYPH<2> v s , 0, even though v s is proportional to a gradient (at least nonrelativistically).
In the large-scale motion of a quantum fluid, as in a classical medium such as the atmosphere or the ocean, vorticity is ubiquitous, being induced through di GLYPH<11> erent means in di GLYPH<11> erent systems. In liquid helium in the laboratory, they can be created through rotation of the container, or through local heat perturbations. The latter can create quantum turbulence in the form of a vortex tangle. In the cosmos, quantized vortices in
the background superfluid can be created by a rotating black hole, a rotating galaxy, or colliding galaxies. The big bang era could witness the creation of quantum turbulence. Our main theme is that all these e GLYPH<11> ects can be traced to a universal mechanism characterized by operator terms in the NLKG that can be associated with the Coriolis and the centrifugal force:
$$R _ { \text{Coriolis} } & = \frac { 2 \Omega } { c ^ { 2 } } \frac { \partial ^ { 2 } } { \partial \phi \partial t }, & \quad \prod ( \mathfrak { r } ) \\ R _ { \text{centrifugal} } & = - \frac { \Omega ^ { 2 } } { c ^ { 2 } } \frac { \partial ^ { 2 } } { \partial \phi ^ { 2 } }, & \quad ( 4 )$$
where GLYPH<10> is an angular velocity, and GLYPH<30> the angle of rotation. The angular velocity may be an externally given constant, or it may be a spacetime function arising from dynamics, interaction with external sources, or spacetime geometry.
The various threads of our investigation may be summarized by the chart in Fig.1. We begin in Sec.2 with a general formulation of the NLKG in general relativity, and define the superfluid velocity field v s in terms of the phase of the field. This is followed in Sec.3 by a reduction to flat Minkowski spacetime. In Sec.4, we then consider the NLKG in a global rotating frame for orientation, and exhibit the Coriolis force and centrifugal force (the top link in Fig.1). This enables us to extract local Coriolis and centrifugal terms from the curvedspace equation, by expansion in the local angular velocity, (the left to right link in Fig.1). The local rotating frames associated with the Coriolis force leads to frame-dragging. We then make a non-relativistic reduction in Sec.5, recovering the nonlinear Schr¨ odinger equation, particularly the form used to describe vortex generation in rotating BEC (the link to the bottom square in Fig.1). In Sec.6, we discuss the introduction of an external source via current-current coupling, and note how it can have the same e GLYPH<11> ect as the Coriolis force (the right link in Fig.1). Finally, in Sec.7, we show some results from numerical computations.
## II. SUPERFLUIDITY IN CURVED SPACETIME
We start with the NLKG in a general background metric g GLYPH<22>GLYPH<23> ( GLYPH<22>; GLYPH<23> = 0 1 2 3). The action of the complex scalar field is ; ; ; given by
$$S = - \, \int d t \, d ^ { 3 } x \sqrt { - g } \left ( g ^ { \mu \nu } \partial _ { \mu } \Phi ^ { * } \partial _ { \nu } \Phi + V \right ), \quad ( 5 )$$
where g = det GLYPH<16> g GLYPH<22>GLYPH<23> GLYPH<17> , and V denotes a self-interaction potential that depends only on GLYPH<8> GLYPH<3> GLYPH<8> . The equation of motion is
$$( \Box - V ^ { \prime } ) \, \Phi = 0,$$
where V 0 = dV d = ( GLYPH<8> GLYPH<3> GLYPH<8> ), and
$$\Box \Phi \equiv \nabla ^ { \mu } \nabla _ { \mu } \Phi = \frac { 1 } { \sqrt { - g } } \partial _ { \mu } ( \sqrt { - g } g ^ { \mu \nu } \partial _ { \nu } \Phi ). \quad \ \ ( 7 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
where r GLYPH<22> is the covariant derivative.
FIG. 1. Chart showing various threads of our investigation. The general theme, shown by the central and the left square, is that the mechanisms for vortex creation are the inertia forces (Coriolis and centrifugal) in rotating frames, generated by external means or by the spacetime metric, or by external sources through a current-current interaction. See Sec. I to follow the various threads.
In the phase representation GLYPH<8> = Fe i GLYPH<27> , the real and imaginary parts of (6) lead to the equations:
$$( \Box - V ^ { \prime } ) \, F - F \nabla ^ { \mu } \sigma \nabla _ { \mu } \sigma & = 0, \\ 2 \nabla ^ { \mu } F \nabla _ { \mu } \sigma + F \nabla ^ { \mu } \nabla _ { \mu } \sigma & = 0.$$
The first equation can be rewritten in the form of a relativistic Euler equation
$$\nabla _ { \mu } \left ( F ^ { - 1 } \Box F - V ^ { \prime } \right ) - 2 F ^ { - 2 } \nabla ^ { \lambda } ( F ^ { 2 } \nabla _ { \lambda } \sigma \, \nabla _ { \mu } \sigma ) = 0. \quad ( 9 )$$
The second equation is a continuity equation
$$\nabla ^ { \mu } j _ { \mu } = 0, \quad \quad \quad ( 1 0 )$$
with
$$j _ { \mu } \equiv F ^ { 2 } \nabla _ { \mu } \sigma = F ^ { 2 } \partial _ { \mu } \sigma. \text{ \quad \ \ } ( 1 1 )$$
There exist a conserved charge
$$Q = \, \int d ^ { 3 } x \sqrt { - g } \, j ^ { 0 }, \, \text{ \quad \ \ } \text{(12)}$$
and a covariantly conserved energy-momentum tensor
$$T _ { \mu \nu } = \nabla _ { \mu } \Phi ^ { * } \nabla _ { \nu } \Phi + \nabla _ { \nu } \Phi ^ { * } \nabla _ { \mu } \Phi - g _ { \mu \nu } \nabla ^ { \tau } \Phi ^ { * } \nabla _ { \tau } \Phi - g _ { \mu \nu } V, \ \ ( 1 3 )$$
We turn to the definition of the superfluid velocity v s . The field GLYPH<8> ( x ; t ) corresponds to an order parameter at absolute zero, and thus represents a pure superfluid. Let GLYPH<28> be the proper time along a timelike worldline of a fluid element whose coordinates are
$$x ^ { \mu } = \left ( c \, t ( \tau ), \, x ^ { i } ( \tau ) \right ), \, \text{ \quad \ \ } \text{(1 4)}$$
where i = 1 2 3 ; ; : The superfluid velocity is the 3-velocity of the fluid element:
$$\mathbf v _ { s } = \frac { d \mathbf x } { d t }. \quad \quad \ \ ( 1 5 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
We define a 4-velocity
$$U ^ { \mu } \equiv \frac { d x ^ { \mu } } { d \tau } = ( \gamma c, \, \gamma \bar { \mathbf v } _ { s } ) \,, \, \quad \quad ( 1 6 ) \, \stackrel { \infty } { \text{do} } \, \\ \text{ily}$$
where GLYPH<13> GLYPH<17> dt = d GLYPH<28> . From ds 2 = GLYPH<0> c d 2 GLYPH<28> 2 = g GLYPH<22>GLYPH<23> dx dx GLYPH<22> GLYPH<23> ; we obtain
$$\gamma = \left ( - g _ { 0 0 } - 2 g _ { 0 i } \frac { v ^ { i } } { c } - g _ { i j } \frac { v ^ { i } v ^ { j } } { c ^ { 2 } } \right ) ^ { - 1 / 2 }. \quad \ \ ( 1 7 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
The superfluid density GLYPH<26> s is defined through
$$j ^ { \mu } = \rho _ { s } U ^ { \mu }. \quad \quad \ \ ( 1 8 ) \quad ( t ^ { \prime }$$
Comparison with (11) leads to
$$\rho _ { s } & = \hbar { ( } c \gamma \xi _ { s } ) ^ { - 1 } F ^ { 2 }, & \frac { \text{Th} } { c } \frac { \text{Th} _ { s } } { c } & = \xi _ { s } \nabla \sigma & ( 1 9 ) & \text{tin}$$
where
$$\xi _ { s } = \left ( \partial ^ { 0 } \sigma \right ) ^ { - 1 }. \quad \quad \ \ ( 2 0 )$$
## III. REDUCTION TO MINKOWSKI SPACETIME
In Minkowski metric ( GLYPH<0> 1 1 1 1) ; ; ; ; the Euler equation and the continuity equation become
$$& \partial _ { \mu } \left ( F ^ { - 1 } \Box F - V ^ { \prime } \right ) - 2 F ^ { - 2 } \partial ^ { \lambda } ( F ^ { 2 } \partial _ { \lambda } \sigma \partial _ { \mu } \sigma ) = 0, \quad \text{inter} \quad \text{trans} \quad \text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\ \text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\text{/\quad\tpl}} } } } } } } } } }$$
The 4-velocity is given by
$$U ^ { \mu } = ( \gamma c, \gamma \mathbf v _ { s } ) \,, \ \gamma = \frac { 1 } { \sqrt { 1 - v _ { s } ^ { 2 } / c ^ { 2 } } }. \quad \ \ ( 2 2 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
One can check U U GLYPH<22> GLYPH<22> = GLYPH<0> c 2 ; which guarantees v s < c . The superfluid density and velocity are given by
$$\rho _ { s } & = - \, \frac { \hbar { \sigma } } { c ^ { 2 } \gamma } \, F ^ { 2 }, \\ \frac { \mathbf v _ { s } } { c } & = - \frac { c } { \dot { \sigma } } \nabla \sigma. \quad \quad ( 2 3 ) \quad \ \ g u _ { \log } \\ & \quad \ \ \ \ \..$$
In a stationary solution of the form GLYPH<8> ( x ; t ) = e GLYPH<0> i ! t GLYPH<31> ( x ) ; we have ˙ GLYPH<27> = GLYPH<0> !: Eq. (23) was first derived in Ref.[8].
## IV. SUPERFLUID ROTATION
## A. Rotating frame
Asuperfluid can flow frictionlessly past a wall at low velocities; dissipations can occur only when the velocity exceeds a critical value necessary to excite the system. Similarly, a superfluid contained in a rotating bucket will remain at rest, until the angular frequency of the bucket exceeds a critical value, whereupon a rotational velocity field occurs through the creation of quantized vortices, with vortex lines parallel to the axis of rotation. In a steady state, these vortices form a lattice of specific form. Theoretical treatment of this problem is best done in a coordinate frame rotating with the bucket, for it easily exposes the inertial forces, i.e., the Coriolis force and the centrifugal force, that are responsible for vortex creation. The actual superfluid we are studying may not be contained in a bucket, and we may not be in a rotating frame, but the mechanism for vortex creation may be attributed to a local version of these forces.
Consider a rotating frame with angular velocity GLYPH<10> 0 about the z axis. In spatial spherical coordinates, the lab frame ( t 0 ; r 0 ; GLYPH<18> 0 ; GLYPH<30> 0 ) and the rotating frame ( t ; r ; GLYPH<18>; GLYPH<30> ) are related by the transformation
$$t = t ^ { \prime }, \text{ } r = r ^ { \prime }, \text{ } \theta = \theta ^ { \prime }, \text{ } \phi = \phi ^ { \prime } - \Omega _ { 0 } t ^ { \prime } \text{.} \text{ \quad (2 4 ) }$$
The d'Alembertian operator in the rotating frame in flat spacetime is given by
$$\Box = \Box ^ { ( 0 ) } + R _ { \text{Coriolis} } + R _ { \text{centrifugal} }, \quad \quad ( 2 5 )$$
where GLYPH<3> (0) = r GLYPH<0> 2 1 c 2 @ 2 @ t 2 ; and
$$c ^ { - } \sigma ^ { - } \, \nu \\ R _ { \text{Coriolis} } & = \frac { 2 \Omega _ { 0 } } { c ^ { 2 } } \frac { \partial ^ { 2 } } { \partial t \partial \phi }, \\ R _ { \text{centrifugal} } & = - \frac { \Omega _ { 0 } ^ { 2 } } { c ^ { 2 } } \frac { \partial ^ { 2 } } { \partial \phi ^ { 2 } }. \\ \dots \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \alpha \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \beta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \Delta \colon$$
$$( 2 6 )$$
Eq.(25) can be applied to the study of any Klein-Gordon equation in the rotation frame with interactions, e.g. a selfinteraction V which gives the NLKG. Note that the coordinate transformation (24) is restricted by the condition GLYPH<10> 0 R GLYPH<28> c where R is the radius of system. This constraint is satisfied in most rotating systems, from laboratory experiments on liquid 4 He and BEC to neutron stars. Here the spacetime background is assumed to be flat. Curved-spacetime cases involving fastrotating blackholes are studied in [9] and a general treatment will be given in subsection C.
## B. Feynman's relation
Consider N vortices in rotating bucket of radius R and angular velocity GLYPH<10> : At the wall of the bucket, the superfluid velocity is v s = GLYPH<10> R . Thus, r GLYPH<27> = v s = ( c GLYPH<24> s ) = GLYPH<10> R = ( c GLYPH<24> s ) : From the relation I ds GLYPH<1> r GLYPH<27> = 2 GLYPH<25> N , we obtain
$$\Omega = \pi c \xi _ { s } n _ { \nu },$$
where n v = N = GLYPH<16> GLYPH<25> R 2 GLYPH<17> is the number of vortices per unit area. This formulas can give an estimate of the local vortex density when the superfluid flows with varying local angular velocity. In non-relativistic limit GLYPH<24> s ! ~ = mc , we have GLYPH<10> = ( GLYPH<25> ~ = m n ) v ; which is called Feynman's relation. The derivation is valid only for large N , since we treat the rotating superfluid as if it were a rigid body.
## C. Frame-dragging from spacetime geometry
The transformation to a rotating coordinate frame is equivalent to using the metric in the following line element:
$$d s _ { 0 } ^ { 2 } & = - c ^ { 2 } d t ^ { 2 } + d r ^ { 2 } + r ^ { 2 } \sin ^ { 2 } \theta d \theta ^ { 2 } \quad \text{
} \text{
+ r ^ { 2 } ( d \phi + \Omega _ { 0 } d t ) ^ { 2 }. \quad \text{
( 2 8 ) \quad \text{ \text{ \text{ cout \text{ }
| ] . + - < </ > ) ($$
The cross term 2 GLYPH<10> 0 r 2 d GLYPH<30> dt corresponds to a rotating frame of angular velocity GLYPH<10> 0 . A spacetime-dependent GLYPH<10> 0 may lead to frame-dragging, i.e., local rotating frames. We give a general treatment of this e GLYPH<11> ect in the following.
Consider a stationary axially-symmetric background metric with two Killing vectors GLYPH<24> a = ( @=@ t ) a and a = ( @=@GLYPH<30> ) a ; which characterize the stationary condition and the rotational symmetry, respectively. The metric can be parametrized with coordinates f t ; u v ; ; GLYPH<30> g as follows [5]:
$$d s ^ { 2 } & = - A d t ^ { 2 } + g _ { u u } d u ^ { 2 } + 2 g _ { u v } d u d v + g _ { v v } d v ^ { 2 } \quad \text{ \quad \ \ \text{where} } \\ & + 2 B d \phi d t + C d \phi ^ { 2 }, \quad \text{ \quad \ \ \text{(2)} } \quad \text{ \quad \ \ \text{drop} }$$
where the functions A B C ; ; are related to the Killing vectors:
$$A & = - g _ { t t } = - \xi ^ { a } \xi _ { a }, \\ B & = g _ { t \phi } = \xi ^ { a } \psi _ { a }, \\ C & = g _ { \phi \phi } = \psi ^ { a } \psi _ { a }.$$
In general we should solve the NLKG in the background metric (29) for rotation problems. Here we consider a small B approximation. By expanding the d'Alembertian operator in powers of B ; we find a cross term identifiable with the Coriolis force:
$$2 g ^ { t \phi } \frac { \partial ^ { 2 } } { \partial t \partial \phi } = \frac { 2 B } { A C } \frac { \partial ^ { 2 } } { \partial t \partial \phi } + O \left ( B ^ { 3 } \right ). \quad \quad ( 3 1 )$$
Noting that GLYPH<10> = GLYPH<0> B C = is the coordinate angular velocity of locally non-rotating observers [5], we can write
$$R _ { \text{Coriolis} } = \frac { 2 \Omega } { g _ { t t } } \frac { \partial ^ { 2 } } { \partial t \partial \phi }.$$
In a similar manner we obtain the centrifugal term
$$R _ { \text{centralifugal} } = \frac { \Omega ^ { 2 } } { g _ { t t } } \frac { \partial ^ { 2 } } { \partial \phi ^ { 2 } }. \quad \quad ( 3 3 )$$
Examples of metrics with frame-dragging are the BTZ metric in 2 + 1 dimensional spacetime, and the Kerr metric in 3 + 1 dimensional spacetime, which describe the spacetime curvature around a black hole. Frame-dragging in these metrics arise from the angular velocity of the black hole. Vortex creation due to this purely geometric e GLYPH<11> ect is investigated in [9]. From now on we will focus on the flat spacetime background and study the NLKG in the rotating frame defined by the transformation (24).
## V. NON-RELATIVISTIC LIMIT
## A. From NLKG to NLSE
A solution to the NLKG generally contains frequencies of both signs. A large frequency ! 0 of one sign (say, positive) could develop, due to an initial field with large positive charge, or a large potential energy, and the system will approach the non-relativistic limit. We define a wave function GLYPH<9> through
$$\Phi \left ( { \mathbf x }, t \right ) = e ^ { - i \omega _ { 0 } t } \Psi \left ( { \mathbf x }, t \right ).$$
Introducing a mass scale m by putting ! 0 = mc 2 = ~ , we find the equation for GLYPH<9> :
$$i \hbar { \frac { \partial \Psi } { \partial t } } = - \frac { \hbar { ^ { 2 } } } { 2 m } \nabla ^ { 2 } \Psi + U \Psi + \frac { \hbar { ^ { 2 } } } { 2 m c ^ { 2 } } \frac { \partial ^ { 2 } \Psi } { \partial t ^ { 2 } }, \quad ( 3 5 )$$
where U = GLYPH<16> ~ 2 = 2 m V GLYPH<17> 0 GLYPH<0> GLYPH<16> mc 2 = 2 GLYPH<17> : In the limit c ! 1 , we drop the term @ 2 GLYPH<9> =@ t 2 and obtain the nonlinear Schr¨dinger o equation (NLSE).
Let the phase of the non-relativistic wave function be denoted by GLYPH<12> ( x ; t ) :
$$\Psi = | \Psi | e ^ { i \beta }.$$
The phase of the relativistic field GLYPH<8> is thus given by
$$\sigma = - \omega _ { 0 } t + \beta,$$
Putting
$$u ^ { k } \equiv \partial ^ { k } \sigma = \partial ^ { k } \beta,$$
we obtain, from (21),
$$\partial _ { t } \left [ \left ( 1 - \dot { \beta } / \omega _ { 0 } \right ) F ^ { 2 } u ^ { k } + \frac { F ^ { 2 } } { 4 \omega _ { 0 } } \partial ^ { k } \partial _ { t } \left ( \ln F ^ { 2 } \right ) \right ] \\ + \, \frac { c ^ { 2 } } { \omega _ { 0 } } \partial _ { j } ( F ^ { 2 } u ^ { j } u ^ { k } ) + \partial ^ { k } P = 0, \\ \partial _ { t } \left [ \left ( 1 - \dot { \beta } / \omega _ { 0 } \right ) F ^ { 2 } \right ] + \frac { c ^ { 2 } } { \omega _ { 0 } } \partial _ { j } \left ( F ^ { 2 } u ^ { j } \right ) = 0, \\ \text{ere the n ressure } P \text{ is defined hv}$$
where the pressure P is defined by
$$P = \frac { 1 } { 2 } \left ( F ^ { 2 } V ^ { \prime } - V \right ) - \frac { c ^ { 2 } } { 4 \omega _ { 0 } } F ^ { 2 } \, \nabla ^ { 2 } \left ( \ln F ^ { 2 } \right ). \quad \ \ ( 4 0 )$$
The terms in braces are due to self interaction, while the last term is the 'quantum pressure' [2]. The non-relativistic superfluid density and velocity can be obtained from (23) by putting ! 0 = mc 2 = ~ , and formally let c !1 :
$$\rho _ { s } & = m | \Psi | ^ { 2 } \left ( 1 + \hbar { \dot { \beta } } / m c ^ { 2 } - v _ { s } ^ { 2 } / c ^ { 2 } \right ) + O \left ( c ^ { - 4 } \right ), \\ \mathbf v _ { s } & = \frac { \hbar } { m } \nabla \beta \left ( 1 + \hbar { \dot { \beta } } / m c ^ { 2 } \right ) + O \left ( c ^ { - 4 } \right ).$$
TABLE I. Comparison of some features of the relativistic and nonrelativistic limit based on NLKG and NLSE, respectively.
| Equations | NLKG | NLSE |
|----------------------------------|------------------------------------------------------------------------------------------------------------|-------------------------------------------------|
| Variables | GLYPH<8> ( x ; t ) = Fe i GLYPH<27> | GLYPH<9> ( x ; t ) = j GLYPH<9> j e i GLYPH<12> |
| Superfluid densities GLYPH<26> s | GLYPH<0> ~ ˙ GLYPH<27> c 2 GLYPH<13> F 2 | m j GLYPH<9> j 2 |
| Superfluid velocities v s | GLYPH<0> c ˙ r GLYPH<27> | ~ m r GLYPH<12> |
| Rotating terms | GLYPH<27> 2 GLYPH<10> c 2 @ 2 GLYPH<8> @ t @GLYPH<30> , GLYPH<0> GLYPH<10> 2 c 2 @ 2 GLYPH<8> @GLYPH<30> 2 | i ~ GLYPH<10> @ GLYPH<9> @GLYPH<30> |
| Feynman relation | GLYPH<10> = GLYPH<25> c GLYPH<24> s n v | GLYPH<10> = ( GLYPH<25> ~ = m ) n v |
From (21) we obtain the non-relativistic hydrodynamic equations:
$$& \frac { \partial \rho _ { s } } { \partial t } + \nabla \cdot ( \rho _ { s } \mathbf v _ { s } ) = 0, \\ & \left ( \frac { \partial } { \partial t } + \mathbf v _ { s } \cdot \nabla \right ) \mathbf v _ { s } = - \frac { \nabla P } { \rho _ { s } } - \mathbf R. \quad \quad ( 4 2 ) \quad \text{$to s} \\ - \quad - \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \end{cases}$$
where R = ( rGLYPH<1> v s + v s GLYPH<1> r ln GLYPH<26> s ) v s vanishes for an incompressible, divergenceless fluid.
## B. NLSE in rotating frame
We start with the NLKG in a rotating frame:
$$& \text{We start with the NLKG in a rotating frame:} \\ & \left ( \nabla ^ { 2 } - \frac { 1 } { c ^ { 2 } } \frac { \partial ^ { 2 } } { \partial t ^ { 2 } } - V ^ { \prime } + R _ { \text{Coriolis} } + R _ { \text{entrifugal} } \right ) \Phi = 0, \\ & V ^ { \prime } = \left ( \frac { m c } { \hslash } \right ) ^ { 2 } - 2 m \lambda | \Phi | ^ { 2 }, \quad & \text{Since} \quad \text{city} \quad \text{dragg} \\ & R _ { \text{Coriolis} } = \frac { 2 \Omega } { c ^ { 2 } } \frac { \partial ^ { 2 } } { \partial t \partial \phi }, \\ & R _ { \text{entrifugal} } = - \frac { \Omega ^ { 2 } } { c ^ { 2 } } \frac { \partial ^ { 2 } } { \partial \phi ^ { 2 } } \quad & \text{where} \\ \text{where we use a specific form of the potential. Going to the}$$
where we use a specific form of the potential. Going to the non-relativistic limit yields
$$i \hbar { \frac { } { } } \overline { \partial t } = \left ( - \frac { \hbar { ^ { 2 } } } { 2 m } \nabla ^ { 2 } + i \hbar { \Omega } \frac { \partial } { \partial \phi } + \lambda \left | \Psi \right | ^ { 2 } \right ) \Psi, \quad ( 4 4 )$$
which is the NLSE usually used to describe a rotating BEC [3]. The trapping potential can be included as well. The centrifugal term does not appear, because it is GLYPH<0> GLYPH<16> ~ 2 GLYPH<10> 2 = c 2 GLYPH<17> @ =@GLYPH<30> 2 2 , and thus of order c GLYPH<0> 2 : For comparison some features of the relativistic and non-relativistic limit are listed in Table 1.
## VI. CURRENT-CURRENT INTERACTION WITH A SOURCE
The complex scalar field GLYPH<8> may be coupled to an external source. In Minkowski spacetime, the only Lorentz-invariant interaction in the Lagrangian density is the current-current interaction
$$\mathcal { L } _ { \text{int} } = - \eta J ^ { \mu } j _ { \mu }, \quad \quad \ \ ( 4 5 ) \quad \text{wi}$$
where j GLYPH<22> is the conserved current (11) of the complex scalar field, GLYPH<17> a coupling constant, and J GLYPH<22> is the conserved current of the external source. A direct coupling is possible only if the scalar field has multi-components representing internal symmetry. We can write, more explicitly,
$$\mathcal { L } _ { \text{int} } = - \eta F ^ { 2 } J ^ { \mu } \partial _ { \mu } \sigma,$$
showing that it is a derivative coupling of the phase of the field. The NLKG now reads
$$( \Box - V ^ { \prime } ) \, \Phi - i \eta J ^ { \mu } \partial _ { \mu } \Phi = 0. \text{ \quad \ \ } ( 4 7 )$$
Ref. [8] uses an external current of the form
$$J ^ { \mu } = ( \rho, \mathbf J ) \,, \ \mathbf J = \rho \Omega \times \mathbf r,$$
to simulate the presence of a galaxy in a cosmic superfluid. Here, GLYPH<26> ( x ) is the density of the galaxy, represented by a Gaussian distribution, and GLYPH<10> is its angular velocity. In this case, the NLKG takes the form
$$( \Box - V ^ { \prime } ) \, \Phi + i \eta \rho \left ( \frac { \partial \Phi } { \partial t } + \Omega \times \mathbf r \cdot \nabla \Phi \right ) = 0. \quad ( 4 9 )$$
The last term reads, in a cylindrical coordinate system with azimuthal angle GLYPH<30> about the rotation axis,
$$i \eta \rho \Omega \times { \mathbf r } \cdot \nabla \Phi = i \eta \rho \Omega \frac { \partial \Phi } { \partial \phi } \,. \quad \quad ( 5 0 )$$
Since GLYPH<26> GLYPH<10> e GLYPH<11> ectively gives a spatially dependent angular velocity, this term gives rise to an e GLYPH<11> ect similar to framedragging. We can compare it with a Coriolis term
$$\Omega \, \frac { \partial ^ { 2 } \Phi } { \partial t \partial \phi } = - i \omega \Omega \frac { \partial \Phi } { \partial \phi },$$
where we have put @ GLYPH<8> =@ t = GLYPH<0> i ! GLYPH<8> for a stationary solution.
## VII. NUMERICAL COMPUTATIONS
We solve the NLKG with various inertial or interaction terms in (2 + 1)- and (3 + 1)-dimensional Minkowski spacetime. Our emphasis here is on vortex states and vortex dynamics. For the e GLYPH<11> ects of gravity and dark matter modeling, see Ref.[8].
The equations are solved by finite di GLYPH<11> erence scheme and spectrum methods are used to decouple the discretization for derivatives @ t and @GLYPH<30> in the Coriolis term. Semi-implicit scheme similar to the Crank-Nicolson method is used for the linear terms. The non-linear term is handled explicitly. We impose second order Neumann boundary condition in solving the NLKG in the rotating frame (43) and periodic boundary condition in studying the vortex rings lattice and vortex line reconnection with (49), respectively.
Fig.2 shows in (2 + 1) dimensions the number of vortices in a rotating bucket as a function of the angular velocity and the field frequency. The computational data agrees quite well with the relativistic Feynman relation (27). As examples Fig.3
FIG. 2. The number N of vortices in a rotating bucket as function of the angular velocity GLYPH<10> and the field frequency ! . The straight line represents the relativistic Feynman relation (27), which can be rewritten N = ( R 2 = c 2 ) ! GLYPH<10> . Here, R is the radius of the bucket, and ! is the frequency of the the stationary solution to the NLKG.
FIG. 3. Two lattice states with 8 and 63 vortices respectively: The modulus of the field is plotted on the left side and its phase is on the right side. The color map from blue to red (see web version) indicates a phase change of 2 GLYPH<25> .
shows the modulus (left) and the phase (right) plots of the complex scalar field in stationary states with 8 and 63 vortices, respectively. The locations of vortices are indicated by the dots in the modulus plots, corresponding to positions in the phase plots around which the color changes from blue to red, indicating a phase change of 2 GLYPH<25> . The vortex distribution is mainly determined by the Coriolis term while some nonuniformity is caused by the centrifugal term.
In (3 + 1) dimensions, we introduce an external current of the form (48) to simulate a rotating 'star' immersed in a cosmic superfluid and then solve (49). A vortex-ring solution is found and the magnitude of the modulus gradient is plotted in Fig.4 with the usual rainbow colormap. It shows that the star drags the superfluid into local rotation, and there appear vortex rings surrounding the star. As pointed out in Ref. [7, 8], the so-called 'non-thermal filaments' observed near the center of the Milky Way may be parts of such vortex rings around rotating black holes.
Vortex lines in 3D can cross and reconnect. The originally smooth vortex lines become lines with cusps at the reconnection point, and these cusps spring away from each other rapidly, creating two jets of energy in the superfluid. Through repeated reconnections, large vortex rings can be degraded until they become a vortex tangle of fractal dimensions (quantum turbulence). This mechanism has been used in [6, 7] to create matter in the cosmos during the big bang era. We can simulate such a reconnection via the NLKG, as shown in Fig.5 by plotting the curl of the phase gradient. V ortex reconnection in the NLSE has been studied in Ref. [10]. Note that vortex dynamics has also been studied via the local induction approximation (see e.g. [11]) and its connection to the NLSE has been found in [12]. For recent development along this direction, see [13, 14]. For actual photographs of vortex reconnection in superfluid helium, see Ref. [15].
## VIII. CONCLUSION
We formulate a framework for investigating superfluidity and the mechanism for creating quantized vorticity in the relativistic regime, based on a nonlinear Klein-Gordon equation for a complex scalar field. This framework is constructed in curved spacetime, then is reduced to flat spacetime and also the non-relativistic limit. Our numerical study focuses on the flat-spacetime cases. It shows that quantized vorticity can be created by local inertial forces (Coriolis and centrifugal) and current-current interaction. Feynman's relation relating the number of vortices to the angular frequency is numerically verified in the relativistic regime. Our numerical solutions to the NLKG exhibit such features as vortex lattices, 3D vortex rings around rotating stars, and vortex reconnections. In this paper we only consider pure superfluid at zero temperature. It would be interesting to generalize this relativistic formulation to the finite temperature cases in which a normal fluid component and the corresponding e GLYPH<11> ects, such as mutual friction will emerge.
FIG. 4. System of 3D vortex rings surrounding a rotating 'star', viewed from di GLYPH<11> erent perspectives.
- [3] A. L. Fetter, Rev. Mod. Phys. 81 , 647 (2009).
- [4] V. L. Ginzburg and L. D. Landau, Zh. Eksp. Teor. Fiz. 20 , 1064 (1950).
- [5] R. M. Wald, General Relativity (University of Chicago Press, 1984), Chapter 7.
- [6] K. Huang, H. -B. Low and R. -S. Tung, Class. Quant. Grav. 29 , 155014 (2012). [arXiv:1106.5282 [gr-qc]].
- [7] K. Huang, H. -B. Low and R. -S. Tung, Int. J. Mod. Phys. A 27 , 1250154 (2012). [arXiv:1106.5283 [gr-qc]].
- [8] K. Huang, C. Xiong and X. Zhao, Int. J. Mod. Phys. A 29 , no. 13, 1450074 (2014) [arXiv:1304.1595 [gr-qc]].
- [9] M. Good, C. Xiong, A. Chua and K. Huang, arXiv:1407.5760 [gr-qc].
- [10] J. Koplik and H. Levine, Phys. Rev. Lett. 71 , 1375 (1993).
- [11] K. W. Schwarz, Phys. Rev. B 31 , 5782 (1985).
- [12] H. Hasimoto, J. Fluid Mech. 51 , 477 (1972).
- [13] B. K. Shivamoggi, Eur. Phys. J. B 86 , 275 (2013).
- [14] R. A. Van Gorder, J. Fluid Mech. 740 , 5 (2014).
- [15] M. S. Paoletti, M. E. Fisher and D. P. Lathrop, Physica D 239 , 1367-1377 (2010).
FIG. 5. Frames 1-4 show the progress of a vortex reconnection, at equal time steps. An initial configuration of a 3D vortex-line array is constructed with the current-current source term, which is then turned o GLYPH<11> to allow free evolution and reconnection of the vortex lines.
|
10.1103/PhysRevD.90.125019
|
[
"Chi Xiong",
"Michael R. R. Good",
"Yulong Guo",
"Xiaopei Liu",
"Kerson Huang"
] |
2014-08-04T19:26:00+00:00
|
2014-12-03T19:37:46+00:00
|
[
"hep-th",
"astro-ph.HE",
"gr-qc",
"nlin.PS"
] |
Relativistic superfluidity and vorticity from the nonlinear Klein-Gordon equation
|
We investigate superfluidity, and the mechanism for creation of quantized
vortices, in the relativistic regime. The general framework is a nonlinear
Klein-Gordon equation in curved spacetime for a complex scalar field, whose
phase dynamics gives rise to superfluidity. The mechanisms discussed are local
inertial forces (Coriolis and centrifugal), and current-current interaction
with an external source. The primary application is to cosmology, but we also
discuss the reduction to the non-relativistic nonlinear Schr\"{o}dinger
equation, which is widely used in describing superfluidity and vorticity in
liquid helium and cold-trapped atomic gases.
|
1408.0781v1
|
## Internal Cancellation over SSP Rings
## Huanyin Chen
Department of Mathematics, Hangzhou Normal University Hangzhou 310036, People's Republic of China
## Abstract
A ring is SSP if the sum of two direct summands is a direct summand. A ring has internal cancellation if every its (von Neumann) regular elements are unit-regular. We show that in an SSP ring having internal cancellation, any regular element is special clean. Our main results also imply that for any SSP ring internal cancellation and idempotent stable range 1 coincide with each other. Internal cancellation over SSP was then characterized by special clean elements.
Keywords: Idempotent stable range one; Regular elements; Special clean elements; SSP rings.
2010 Mathematics Subject Classification: 16D70, 16E50.
## 1 Introduction
Let R be a ring with an identity. An element a ∈ R is (unit) regular if there exists some (unit) x ∈ R such that a = axa . A ring R is (unit) regular if and only if every element in R is (unit) regular. As is well known, a ring R is unit-regular if and only if every element in R is the product of an idempotent and a unit. For general theory of regular rings, we refer the reader to [7]. An element a ∈ R is (special) clean if it is the sum of an idempotent e ( aR ⋂ eR = 0) and a unit u . In [3, Theorem 5], Camillo and Yu claimed that every element in unit-regular rings is clean. Unfortunately, there was a gap in their proof. In [2, Theorem 1], Camillo and Khurana improved [3, Theorem 5] and proved that a ring R is unit-regular if and only if every element in R is special clean.
A ring R has stable range 1 provided that Ra + Rb = R with a, b ∈ R implies that a + zb ∈ R is a unit. A regular ring R is unit-regular if and only if R has stable range 1. An interesting problem is to extend Camillo-Khurana's theorem to certain rings having stable range 1. A ring R is an exchange ring if for any a ∈ R here exists an idempotent e ∈ aR such that 1 -e ∈ (1 -a R ) . Clearly, every regular ring is an exchange ring. In [4, Theorem 2.1], the author proved that an exchange ring R has stable range 1 if and only if every regular element in R is special clean.
In [8], Khurana and Lam introduced internal cancellation (IC, for short) over a ring. A ring R is IC if every regular element in R is unit-regular. Obviously, a regular ring R is unit-regular if and only if R is IC. As is well known, a ring R is IC if and only if for any regular a, b ∈ R Ra , + Rb = R implies that there exists a z ∈ R such that a + zb ∈ R is
a unit. Clearly, stable range 1 implies internal cancellation, but the converse is not true, e.g. Z . For an exchange ring R , stable range 1 and internal cancellation coincide with each other [5, Lemma 1.3.1]. Many types of IC rings are investigated in [8]. A further attractive problem is to extend Camillo-Khurana's theorem to certain rings having IC. In [9], Wang et al. consider a kind of idempotent stable range one for rings having IC. In ring theoretic version, the main result in [9] stated
Theorem 1.1. Let R be a ring, let a ∈ R is regular, and let end R ( r a ( )) or end R ( aR ) has stable range 1. If R is IC, then Ra + Rb = R implies that there exists an idempotent e ∈ R such that a + eb ∈ U R ( ) and R = aR ⊕ eR .
Then Camillo-Khurana's theorem will be a special case of this one. But stable range 1 is necessary in the preceding theorem, as the substitution of modules is essential in its proof. The motivation of this short note is to extend Camillo-Khurana's theorem in a broader context with no restriction of stable range 1.
Following Garcia [6], a ring R is said to being the summand sum property (briefly SSP) if the sum of two direct summands of R R is also a direct summand of R . For instance, every (von Neumann)regular ring and every every abelian ring (idempotents are central, e.g., commutative rings, right duo rings) are SSP. We shall prove that for any SSP ring internal cancellation and idempotent stable range 1 coincide with each other. Then in an SSP ring having internal cancellation, any regular element is special clean. Internal cancellation over SSP was then characterized by special clean elements.
Throughout, all rings are associative with an identity. Let R be a ring. U R ( ) will denote the set of all units in R . The right annihilator r a ( ) = { r ∈ R | ar = 0 . }
## 2 The main results
We begin with several lemmas which will be needed in our proof of the main results.
Lemma 2.1. Every special clean element in a ring is unit-regular.
Proof. Let a ∈ R be special clean. Then there exists an idempotent e ∈ R and a unit u ∈ R such that a = e + u and aR ⋂ eR = 0 . Hence, au -1 = eu -1 + 1. Thus, au -1 e = eu -1 e + e ∈ aR ⋂ eR = 0. This yields au -1 ( a -u ) = 0, and so au -1 a = a . Therefore a ∈ R is unit-regular. ✷
Lemma 2.2. [9, Proposition 2.2] Let M a right k -module, and let R = End k ( M ). Then two submodules A and B of M has a common direct complement if and only if there exists an idempotent e ∈ R such that eM = A and e | B : B → A is an isomorphism.
Lemma 2.3. [8, Ehrlichs Theorem 1.1] A ring R is IC if and only if R = A 1 ⊕ B = A 2 ⊕ C with A 1 ∼ = A 2 implies B ∼ = C .
Theorem 2.4. Let R be SSP. Then the following are equivalent:
- (1) R is IC;
- (2) For any regular a, b ∈ R Ra , + Rb = R implies that there exists an idempotent e ∈ R such that a + eb ∈ U R ( ) and aR ⊕ eR = R .
- (3) Every regular element in R is special clean.
Proof. (1) ⇒ (2) Suppose that Ra + Rb = R and a, b ∈ R are regular. Write a = axa and x = xax for some x ∈ R . Let K = r a , D ( ) = xaR,I = aR and C = (1 -ax R ) . Then R = K ⊕ D = C ⊕ I with a | D : D ∼ = I . Since R is IC, we see that K ∼ = C . Clearly, K r b ( ) = 0. Hence, b : K → bK is an isomorphism, and then C ∼ = K ∼ = bK .
$$R = f R \bigcap g R \oplus ( 1 - f ) R \oplus f$$
⋂ Since Ra + Rb = R , we see that Rb (1 -xa ) = R (1 -xa ). We infer that b (1 -xa ) ∈ R is regular, and then bK = (1 b -xa R ) is a direct summand of R . Thus, we have an idempotent f ∈ R such that bK = fR . Set g = 1 -xa . Then C = gR and fR ∼ = gR . As R is IC, we get (1 -f ) R ∼ = (1 -g R ) . In view of [6, Proposition 2.5], R has the summand intersection property, i.e., the intersection of two direct summands of R is a direct summand. Thus, fR ⋂ gR is a direct summand of R . Write R = fR ⋂ gR ⊕ L . Then fR = fR ⋂ gR ⊕ fR ⋂ L and gR = fR ⋂ gR ⊕ gR ⋂ L . This shows that
∼ ∼
As fR ⋂ gR ⊕ (1 -f ) R = fR ⋂ gR ⊕ (1 -g R ) , we get ϕ : fR ⋂ L = gR ⋂ L . Let E = { x + ϕ x ( ) | x ∈ fR ⋂ L } . Clearly, fR + gR = fR + E . If x + ϕ x ( ) ∈ fR ⋂ E for x ∈ fR ⋂ L , then ϕ x ( ) = ( x + ϕ x ( ) ) -x ∈ ( gR ⋂ L ) ⋂ fR = ( fR ⋂ gR ) ⋂ L = 0. As ϕ is an isomorphism, we get x = 0, and so x + ϕ x ( ) = 0. We infer that fR ⋂ E = ∅ . Thus, fR + gR = fR ⊕ E . Likewise, fR + gR = gR ⊕ E . Since R is SSP, there exists a right R -module F such that
$$R = ( f R + g R ) \oplus F = b K \oplus ( E \$$
↦
Hence, C and bK has a common direct complement. In view of Lemma 2.2, we have an idempotent e ∈ R such that eR = C and e : bK → C is an isomorphism. Clearly, R = aR ⊕ C = aR ⊕ eR . Let u := a + eb : R = K ⊕ D → R,x + y → a y ( ) + eb x ( + ) for y any x ∈ K,y ∈ D .
One easily checks that the diagram of exact sequences
$$\begin{array} {$$
commutes. As eb : K → C and a : D → I are both isomorphisms, by using the Five Lemma, we conclude that a + eb ∈ U R ( ), as asserted.
- (2) ⇒ (3) For any regular a ∈ R , we see that Ra + R ( -1) = R . Thus, we can find an idempotent e ∈ R such tht a -e ∈ U R ( ) and aR eR = 0. That is, a ∈ R is special clean.
- ⋂ (3) ⇒ (1) In view of Lemma 2.1, every regular element in R is unit-regular. Therefore R is IC, as asserted. ✷
Remark 2.5. From a similar proof of Theorem 2.4, we actually derive a stronger version: An SSP ring R is IC if and only if for any regular a, b ∈ R r a , ( ) ⋂ r b ( ) = 0 implies that there exists an idempotent e ∈ R such that a + eb ∈ U R ( ) and aR ⊕ eR = R . As SSP and IC are both left-right symmetric properties, we see that an SSP ring R is IC if and only if for any regular a, b ∈ R aR , + bR = R implies that there exists an idempotent e ∈ R such that a + be ∈ U R ( ) and Ra ⊕ Re = R , by applying Theorem 2.4 to the opposite ring R op of R .
Corollary 2.6 [Camillo-Khurana's theorem] Let R be a ring. Then the following are equivalent:
- (1) R is unit-regular.
- (2) Every element in R is special clean.
Proof. This is obvious by Theorem 2.4 and Lemma 2.1.
✷
Corollary 2.7. Let R be abelian. Then for any regular a ∈ R , there exists a unique idempotent e ∈ R and a unit u ∈ R such that a = e + u and aR ⋂ eR = 0 .
Proof. Clearly, R is SSP and IC. Let a ∈ R be regular. Then a ∈ R is special clean, by Theorem 2.4. Assume that there exist idempotents e, e ′ ∈ R and units u, u ′ ∈ R such that a = e + u = e ′ + u ′ and aR ⋂ eR = 0 = aR ⋂ e R. ′ Then au -1 = eu -1 +1. As R is abelian, we see that au -1 (1 -e ) = 1 -e , and so au -1 (1 -e e ) ′ = e ′ (1 -e ) ∈ aR ⋂ e R ′ = 0. This forces that e ′ = e e ′ . Likewise, e = ee ′ . Therefore e = e ′ , as needed.
Example 2.8. The ring Z of all integers is an SSP ring having IC, but it has not stable range 1.
Theorem 2.9 Let R be a ring. Then the following are equivalent:
- (1) R is SSP and IC.
- (2) The product of two regular elements is unit-regular.
- (3) The product of two regular elements is special clean.
Proof. (1) ⇒ (3) Let a, b ∈ R be regular. Since R is IC, we can find idempotents e, f ∈ R and units u, v ∈ R such that a = ue and b = fv . Then ab = u ef ( ) v . Observing that
$$( 1 - e ) R \oplus e f R = ( 1 - e ) R + f R$$
we get R = efR ⊕ (1 -e R ) ⊕ (1 -g R ) . This implies that ef ∈ R is regular. Hence, u ef ( ) v is regular, and then so is ab in R . By virtue of Theorem 2.4, ab ∈ R is special clean, as required.
- (3) ⇒ (2) This is clear in terms if Lemma 2.1.
- (2) ⇒ (1) Clearly, every regular element in R is unit-regular. That is, R is IC. Let e, f ∈ R be idempotents. Then eR + fR = eR + (1 -e fR ) . By hypothesis, there exists an idempotent g ∈ R such that (1 -e fR ) = gR . Then we check that eR + fR = ( e + ) g R where e + g ∈ R is an idempotent. Therefore eR + fR is a direct summand of R , hence that R is SSP. ✷
Corollary 2.10 Let R be a ring. Then the following are equivalent:
- (1) R is SSP and IC.
- (2) The product of finitely many of regular elements is unit-regular.
- (3) The product of finitely many elements is special clean.
Proof. (1) ⇒ (3) Let a = a 1 · · · a n -1 a n ( n ≥ 2) where each a i ∈ R is regular. Then a a 1 2 ∈ R is regular, by Theorem 2.9. By iteration of this process, we see that a a 1 2 · · · a n -1 ∈ R is regular. In light of Theorem 2.9, we conclude that ( a 1 · · · a n -1 ) a n ∈ R is special clean. Therefore a ∈ R is special clean, as desired.
- (3) ⇒ (2) This is obvious by Lemma 2.1.
- (2) ⇒ (1) is trivial from Theorem 2.9.
✷
We note that the number of regular elements in Theorem 2.9 can not be reduced from two to one as the following shows.
Example 2.11. Let R = ( Z 3 Z 3 0 Z 3 ) . One easily checks that R has stable range one. In view of Theorem 1.1, every regular element in R is special clean. Choose e = ( 1 1 0 0 ) and f = ( 0 1 0 1 ) . Then e, f ∈ R are idempotents, and then regular. But ef = ( 0 -1 0 0 ) ∈ R is not regular, as ef ∈ ( ef ) R ef ( ) = 0. Thus, ef ∈ R is not special clean.
/negationslash
## References
- [1] E. Akalan and L. Vas, Classes of almost clean rings, Algebra Represent. Theory , 16 (2013) 843-857.
- [2] V.P. Camillo and D. Khurana, A characterization of unit regular rings, Comm. Algebra, 29 (2001), 2293-2295.
- [3] V. Camillo and H.P. Yu, Exchange rings, units and idempotents, Comm. Algebra, 22 (1994), 4737-4749.
- [4] H. Chen, Exchange rings with stable range one, Czechoslovak Math. J. , 57(132) (2007), 579-590.
- [5] H. Chen, Rings Related Stable Range Conditions , Series in Algebra 11, World Scientific, Hackensack, NJ, 2011.
- [6] J.L. Garcia, Properties of direct summands of modules, Comm. Algebra 17 (1989), 7392.
- [7] K.R. Goodearl, Von Neumann Regular Rings , Pitman, Krieger, Malabar, Fl., 1991.
- [8] D. Khurana and T.Y. Lam, Rings with internal cancellation, J. Algebra 284 (2005), 203-235.
- [9] Z. Wang, J. Chen and D. Khurana, Rings of idempotent stable range one, Algebra Represent. Theory 15 (2012), 195-200.
| null |
[
"Huanyin Chen"
] |
2014-08-04T19:30:21+00:00
|
2014-08-04T19:30:21+00:00
|
[
"math.RA"
] |
Internal Cancellation over SSP Rings
|
A ring is SSP if the sum of two direct summands is a direct summand. A ring
has internal cancellation if every its (von Neumann) regular elements are
unit-regular. We show that in an SSP ring having internal cancellation, any
regular element is special clean. Our main results also imply that for any SSP
ring internal cancellation and idempotent stable range $1$ coincide with each
other. Internal cancellation over SSP was then characterized by special clean
elements.
|
1408.0782v1
|
## Targetable Named Entity Recognition in Social Media
## Sandeep Ashwini
## Jinho D. Choi
Amazon Fulfillment Technologies Amazon.com, LLC Seattle, WA 98109
[email protected]
## Abstract
We present a novel approach for recognizing what we call targetable named entities; that is, named entities in a targeted set (e.g, movies, books, TV shows). Unlike many other NER systems that need to retrain their statistical models as new entities arrive, our approach does not require such retraining, which makes it more adaptable for types of entities that are frequently updated. For this preliminary study, we focus on one entity type, movie title, using data collected from Twitter. Our system is tested on two evaluation sets, one including only entities corresponding to movies in our training set, and the other excluding any of those entities. Our final model shows F1-scores of 76.19% and 78.70% on these evaluation sets, which gives strong evidence that our approach is completely unbiased to any particular set of entities found during training.
## 1 Introduction
Current state-of-the-art systems perform very well on recognizing named entities such as persons, locations, or organizations (Lin and Wu, 2009; Ratinov and Roth, 2009; Turian et al., 2010); however, there is relatively little work on recognizing other types of entities such as movies, books, songs, or TV shows (Ritter et al., 2011; Guo et al., 2013). As more people share information about various topics, named entity recognition for these non-traditional entities becomes more important; especially using data collected from social media, where people's interests are directly expressed.
Even without the difficulties of processing data from social media (Foster et al., 2011; Dent and Paul, 2011), NER for non-traditional entities can be more challenging than traditional ones. First, it is hard to build a large dataset containing enough instances of these entities, which is required by most
Department of Computer Science Emory University Atlanta, GA 30322
[email protected]
state-of-the-art systems using statistical machine learning. Previous studies have shown that these types of entities comprise only a small portion of the entire data consisting of random streams (Ritter et al., 2011; Gattani et al., 2013), which is not sufficient to build a good statistical model. Thus, a more focused dataset needs to be constructed to build a good statistical model for these entities.
Second, non-traditional named entities tend to be updated more frequently than traditional ones (e.g., new movies are released every week), which makes it difficult to build a statistical model that performs well for a long period of time without being retrained. It is possible to keep annotating more data for newly introduced entities (Finin et al., 2010) and retraining new models accordingly; however, the choice of which data to annotate may not be always clear and retraining a new model is not necessarily easy, especially for people bound to use off-the-shelf systems. Thus, it is desirable to build a system that can be dynamically adjusted for new entities without having to be retrained.
Third, traditional named entities mostly occur in the form of noun phrases, whereas non-traditional ones do not necessarily follow the same trend (e.g., movie titles such as Frozen , Ride Along ), which makes them more difficult to recognize. One bright side is that there exists a closed set containing some of these entity types such that the entire space is already known for certain types although it may expand quickly. Thus, even though non-traditional named entities can come in various forms, a system can anticipate the general forms they take given the closed set. Throughout this paper, we refer to these types of entities as 'Targetable Named Entity'; that is, named entities in a targeted set that is closed but can be updated (e.g., a set of movie titles). A general NER system may still perform well for targetable named entities; however, a more focused system can be designed to give similar performance when the entity set is assumed.
Table 1: Distribution of the top 15 most frequent movies in our data. The count columns show the number of entities annotated for each movie.
| Movie | Count | Movie | Count | Movie | Count |
|--------------------|---------|---------------|---------|-------------------|---------|
| Hobbit | 223 | Non Stop | 29 | Son of God | 15 |
| Frozen | 131 | Lego Movie | 22 | American Hustle | 14 |
| Gravity | 126 | Lone Survivor | 20 | Ride Along | 14 |
| 12 Years a Slave | 115 | Nebraska | 16 | Lord of the Rings | 13 |
| Dallas Buyers Club | 41 | Man of Steel | 15 | Titanic | 7 |
In this paper, we give a careful analysis of NER for one targetable named entity type, movie title, using data collected from Twitter. We begin by describing our newly introduced corpus used for developing and evaluating our approach (Section 2). We then present a three-stage NER system, where each stage performs normalization, candidate identification, and entity classification, respectively (Section 3). Finally, we evaluate our approach on two datasets, one including only entities corresponding to movies in our training set, and the other excluding any of those entities (Section 4).
To the best of our knowledge, this is the first time that a targetable named entity type is thoroughly examined. Our approach is general enough to be applied to other targetable named entity types such as TV show, song title, product name, etc.
## 2 Corpus
## 2.1 Data collection
We collected data from Twitter because it was one of the fastest growing social platforms in terms of both users and documents (Smith, 2013). Instead of crawling random tweets, we collected tweets filtered by specific keywords using the Tweepy API. 1 These keywords correspond to movies that were popular during the collection time period, between Dec. 2013 and Feb. 2014 (e.g., Hobbit , Frozen ). It was difficult to find tweets including older movies; however, we managed to collect a few tweets corresponding to movies prior to that time period (e.g., Lord of the Rings , Titanic ). Table 1 shows the distribution of the most frequent movies in our data.
We targeted tweets including specific keywords instead of general tweets for two reasons. One is that targeted streams are more commonly used by commercial organizations for monitoring response to their products (Li et al., 2012). We expect that interested parties would find this type of data col-
1 http://www.tweepy.org
lection more useful than a larger body of random tweets. The other is that the proportion of tweets including movie entities among random tweets is usually too small to build a good statistical model and give a meaningful evaluation. For example, Ritter et al. (2011) annotated 2,400 random tweets with 10 entity types; however, only 34 movie entities were found in that data, which made it difficult for them to show how well their system would perform for this entity type. By restricting the search to a single entity type, we can collect more focused data, which leads to better modeling and evaluation for that entity type.
All redundant tweets and tweets not written in English are removed from our corpus. As a result, a total of 1,096 tweets were collected, of which, 71.9% include at least one movie entity.
## 2.2 Annotation
The following are the annotation guidelines used to label movie entities in our corpus. Table 2 shows annotation examples for each guideline.
- 1. Title match : Each sequence of tokens matching a main title (e.g., the #Hobbit ), a sub title (e.g., the Desolation of Smaug ), or a full title (e.g., the Hobbit 2: The Desolation of Smaug ) is annotated as a single entity.
- 2. Sequel match : Sequel numbers are considered part of entities (e.g., Despicable me 2 ).
- 3. Typo match : Entities including typos are still annotated (e.g., the wolf on wall street , which should be the wolf of wall street ).
- 4. Abbreviation match : Abbreviations indicating movie titles are annotated (e.g., 12Yrs , which should be 12 Years a Slave ).
- 5. Hashtag match : Individual hashtags used to indicate movie titles are annotated (e.g., #DesolationOfSmaug is annotated, whereas #uncontrollablyexcited is not).
- 5 Not sure between #hobbit , #anchorman2 , or #americanhustle this weekend? Gonna see #DesolationOfSmaug finally!!! In 3D! OMG, #uncontrollablyexcited!!
| # | Example | Note |
|-----|-----------------------------------------------------------------------------------------------------------------------------------------------------------|--------------------------------------------------|
| 1 | Can't wait to see the #Hobbit tomorrow! You better believe ' the Desolation of Smaug ,' was badass! I really like the Hobbit 2: The Desolation of Smaug . | Main Title Sub Title Full Title |
| 2 | How many times can you watch Despicable me 2 ? Going to see #Hobbit 2 this afternoon. | Sequel #: 2 Sequel #: 2 |
| 3 | Dedicated 3 hours of my life to the wolf on wall street last night... Finally got 12 Years as a Slave on Blu-Ray! | on → of as →∅ |
| 4 | I literally just watched her in 12Yrs on Saturday. I'm gonna have a Hobbit & LoT marathon... | 12Yrs → 12 Years a Slave LoT → Lord of the Rings |
Table 2: Examples for each guideline in Section 2.2, where annotation is shown in bold. When a hashtag is used to indicate a movie title, it is considered a part of the entity (e.g., the #Hobbit).
All movie entities were manually annotated by a graduate student. A total of 911 movie entities were annotated, corresponding to 53 different movies. Note that a single tweet can include zero to many movie entities. Although our corpus is single annotated, it is done by the author of the guidelines such that the quality of our corpus is most likely higher than ones achieved through crowdsourcing. Once we show the feasibility of our approach, we plan to increase the volume of our corpus through crowdsourcing (Finin et al., 2010) and also include more entity types. Our corpus is publicly available: https://github.com/sandeepAshwini/TwitterMovieData.
Figure 1: The architecture of our NER system.
to normalize tweets before attempting candidate identification and entity classification. The followings show our normalization rules. The first rule is general to any social media data whereas the other rules are specific to tweets.
- 1. Hyperlinks, non UTF-8 characters, punctuation marks, and English articles are removed.
- 2. Retweet markers and the following user IDs are removed (e.g, 'RT @eurweb'). Moreover, User IDs are substituted with a generic token (e.g, @JMussehl → #USER#).
- 3. Hashtags that are not syntactically relevant to the context are removed. The relevance is determined by Kaufmann (2010)'s heuristic test. Hashtags that are syntactically relevant to the context are segmented by matching them against tokens in our gazetteer (Section 3.2.1) using dynamic programming.
## 3 Approach
Figure 1 shows the architecture of our three-stage named entity recognition system, which should be general enough to handle most targetable named entity types. Given input data, a single tweet in our case, tokens are first normalized (either removed or substituted with generic tokens; Section 3.1). From the normalized tweet, candidate entities are identified by using a pre-defined gazetteer and heuristics (Section 3.2). Finally, each candidate is classified as either an entity or not using a statistical model trained on various features (Section 3.3).
## 3.1 Normalization
Some tokens in tweets act rather like noise. For instance, hyperlinks, appearing in almost every tweet, usually introduce noise to the NER process and it is hence better to discard them. On the other hand, it is better to segment hashtags consisting of multiple words when they are syntactically relevant to the context in the tweet. Thus, it is preferable
Table 3: An example of our normalization process by each rule in Section 3.1.
| # | Example |
|-----|---------------------------------------------------------------------------------------------------------------------------------------------------------|
| 0 | RT @eurweb The I bus i was on, was on fire and and I saw The Hobbit: 2 with @JMussehl! and it was so #damngood! #Hobbit #busfire http://t.co/M2nYROTpdS |
| 1 | RT @eurweb I bus i was on was on fire and and I saw Hobbit 2 with @JMussehl and it was so #damngood #Hobbit #busfire |
| 2 | I bus i was on was on fire and and I saw Hobbit 2 with #USER# and it was so #damngood #Hobbit #busfire |
| 3 | I bus i was on was on fire and and I saw Hobbit 2 with #USER# and it was so #damn #good |
After running through these rules, tweets are in their normalized forms, where only relevant information is preserved. Normalization helps speed up the candidate identification process by reducing the total number of tokens in the tweet; in our data, it gives a 22.87% reduction in the number of tokens. Moreover, normalization helps to generate more relevant features for the entity classification stage, as much of the noise in the tweet is effectively removed and some sparsity issues are handled. It is possible to perform further normalization (Li et al., 2012), which we will explore in the future.
## 3.2.2 Sequence matching
## 3.2 Candidate identification
From a normalized tweet, candidate entities are identified by matching token sequences in the tweet to entries in a gazetteer. This stage is designed to give high recall while keeping reasonable precision for NER. We separate this stage from entity classification (Section 3.3) because we do not want the statistical model used for entity classification to be retrained every time the gazetteer is updated. This way, candidate identification finds candidates that are likely to be entities using the most up-todate gazetteer while entity classification ascertains whether or not those candidates are valid without being affected by the gazetteer. We find this approach effective for targetable named entities.
## 3.2.1 Gazetteer
We downloaded the entire list of film entities from Freebase (Bollacker et al., 2008) using the Freebase API. 2 This gazetteer can be dynamically updated to maintain coverage for latest movies. Additionally, we collected the release years of all movies in the gazetteer, which are used as features for entity classification (Section 3.3.1).
2 https://developers.google.com/freebase/
For each token in the tweet, we check if there exists any sequence of tokens, beginning from that token, that matches an entry in our gazetteer. If a match is found, this token sequence is added to the list of possible candidates for that tweet. If more than one matches are found, only the longest sequence is added to the list. Some previous studies restricted this matching to a k -number of tokens to reduce complexity (Guo et al., 2013). Instead of limiting the sequence matching to a fixed size, we adapted an affix tree (Weiner, 1973), which tremendously reduced the matching complexity. 3 Thus, such restriction was no longer necessary in our system.
Four types of sequence matching are performed: full title match, main title match, sub-title match, and sequel match, which are inspired by the guidelines in Section 2.2. Although this strategy looks simple, it gives a good recall for movie entity recognition; however, it often over-generates candidates resulting a poor precision. The following shows an example of over-generated candidates, where candidates are shown in bold.
## I do 1 not understand the Pompeii 2 music videos 3
Three candidates are generated from this tweet, 'I do', 'Pompeii', and 'music videos'. Movie titles such as 'I do' (2012) and 'Pompeii' (2014) can be problematic for dictionary-based NER systems that do not facilitate any contextual information because these are general expressions and entities. Our sub-title match allows candidates such as 'music videos' (from 'NBA Live 2001 - The Music Videos', 2000), which is not even a movie but included in our gazetteer. Creating a more finegrained gazetteer (e.g., IMDB) may resolve issues like this, which we are currently looking into.
3 Traversing through the entire gazetteer takes O (log | W | ) , where W is a set of all token types in the gazetteer.
All candidates are automatically tagged with their types (full, main, sub, and sequel), which are later used as features for entity classification. It might be argued that hashtags alone can be used as a source of candidates; that is, we could generate candidates from only sequences including hashtags. However, in our data, only 23.6% of movie entities contain hashtags, and only 26.3% of all hashtags occur as part of valid movie entities. Hence, this approach would fail to identify a significant portion of the movie entities, and also generate too many falsenegative candidate entities.
## 3.3 Entity classification
For each candidate generated from Section 3.2, a statistical model is used to classify whether it is a valid entity or not. Three types of features are used to build the statistical model: orthographic features, n -gram features, and syntactic features. For better generalization, n -gram features are supplemented using word embeddings. It is worth mentioning that our model does not use any lexical features derived from the candidate itself (e.g., 'Hobbit' is not used as a feature), which prevents it from being overfit to any particular set of movies. Our experiments show that this model gives similar performance on movie entities whose corresponding movies are not in training data compared to ones that are.
## 3.3.1 Orthographic features
Orthographic features give information that can be generally applied to any candidate. Our model uses the following orthographic features.
- 1. Capitalization - Whether the candidate is all capitalized, all decapitalized, or only the first character of each token is capitalized.
- 2. Hashtag - Whether or not the candidate includes a hashtag.
- 3. Number of tokens - The total number of tokens in the candidate.
- 4. Title match - Whether the candidate matches a full title, a main title, a sub-title, or a sequel.
- 5. Numerical time difference - The difference between the year of the tweet posted and the year of release of the corresponding movie.
- 6. Categorical time difference -Whether the release date of the corresponding movie is in the past, contemporary, or future with respect to the time of data collection.
From our experiments, we found that orthographic features could be even more useful than the other features when tweets are written in a proper manner (e.g., proper usage of capitalization or hashtags). Section 4.3 gives more detailed analysis of the impact of these features on our system.
## 3.3.2 N -gram features
N -gram features are used to capture the local context of the entities. Before n -gram features are extracted, each candidate is substituted with a generic token, #MOVIE#. Table 4 shows the n -gram feature templates used in our model, where w i stands for the generic token substituted for the candidate, w i -1 and w i +1 stand for tokens prior and next to the generic token, and so on. Note that we also experimented with a bigger window size of tokens, which did not lead to better results in our data.
Table 4: N -gram feature templates. w i,j represents a joined feature between w i and w j .
| N | Template |
|-----|-----------------------------------------------------------------------------------------------|
| 1 | w i ± 1 , w i ± 2 |
| 2 | w i - 2 ,i - 1 , w i - 2 ,i +1 , w i - 2 ,i +2 , w i - 1 ,i +1 , w i - 1 ,i +2 , w i +1 ,i +2 |
## 3.3.3 Syntactic features
Syntactic features are derived from dependency trees automatically generated by the dependency parser in ClearNLP (Choi and Palmer, 2012; Choi and McCallum, 2013). Table 5 shows the syntactic feature templates used for our model.
Table 5: Syntactic feature templates. The f, m, p, and d columns represent a word-form, a lemma, a part-of-speech tag, and a dependency label of the corresponding token, respectively.
| Token | f | m | p | d |
|------------------|--------------|--------------|--------------|--------------|
| Head ( h ) | glyph[check] | glyph[check] | glyph[check] | glyph[check] |
| Grand-head ( g ) | glyph[check] | glyph[check] | glyph[check] | glyph[check] |
| Sibling ( s ) | glyph[check] | glyph[check] | glyph[check] | glyph[check] |
| Dependent ( d ) | glyph[check] | glyph[check] | glyph[check] | glyph[check] |
The generic token, #MOVIE#, is usually recognized as a proper noun by the parser, which is proper behavior if it is a valid entity. We expect that the dependency tree would look more syntactically sound when the generic token is substituted for a valid entity such that features extracted from these trees are distinguishable from features derived from trees that are less syntactically sound.
Figure 2: A dependency tree for a tweet, 'Can we just sit here and sing songs from #MOVIE# all day', where the candidate, Frozen , is substituted with the generic token, #MOVIE#.
Table 6: Features extracted from the tweet in Figure 2 for the candidate, #Frozen .
| Type | Features |
|--------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| Orthographic | Cap:first only, Hashtag:no, Num of tokens:1, Title match:full, Numerical time diff:0, Categorical time diff:contemporary |
| N -gram | w i - 2 :songs, w i - 1 :from, w i +1 :all, w i +2 :day, w i - 2 ,i - 1 :songs from, w i - 2 ,i +1 :songs all, w i - 2 ,i +2 :songs day, w i - 1 ,i +1 :from all, w i - 1 ,i +2 :from day, w i +1 ,i +2 :all day |
| Syntactic | h f :from, h m :from, h p : IN, h d :pobj, g f :sing, g m :sing, g p : VB, g d :dobj ( No sibling or dependent features exist for this candidate .) |
## 3.3.4 Supplementary features
Since our training data is relatively small, n -gram features are supplemented with words (or phrases) similar to them in terms of cosine distance as determined by word embeddings. For each n -gram feature, the top 10 most similar words are found from the embeddings and included as additional features. These supplementary features are weighted by their cosine similarities to the original n -gram features.
Table 7: Examples of the top 5 most similar words extracted from word embeddings.
| songs | awesome | great film |
|-------------------------------------------|-----------------------------------------------|--------------------------------------------|
| soundtrack music clips stunts performance | incredible astounding excellent amazing awful | train-wreck masterpiece flick picture fine |
Since our task focuses on movie entities, we build embeddings using our training data (Section 4.1) combined with two other datasets designed for sentiment analysis on movie reviews (Pang and Lee, 2004; Maas et al., 2011). 4 The supplementary features are added to improve the generalizability of our model. Note that we also tried to supplement
4 We used a publicly available tool, word2vec , for building embeddings (Mikolov et al., 2013).
the n -gram features using synonym relations in WordNet (Fellbaum, 1998), which did not show much impact on our system.
## 4 Experiments
## 4.1 Data split
For our experiments, we divide our corpus into a single training set and two evaluation sets. The first evaluation set includes only entities corresponding to movies in the training set and the second evaluation set excludes all of those entities. This allows us to show whether or not our approach is biased to the set of movie entities in our training set. Since our corpus is not sufficiently large, no development set is separated from the data. We plan to increase the diversity and the volume of our dataset using crowdsourcing in the near future.
## 4.2 Statistical learning
Liblinear L2-regularization, L1-loss support vector classification (Hsieh et al., 2008) is used for building the statistical model for entity classification. We adapt the implementation in ClearNLP, 5 and use their default learning parameters: c = 0.1, e = 0.1, and b = 0. Neither hyper-parameter optimization nor feature selection was done for our model; given these small datasets, we did not find further
5 http://clearnlp.com
Table 8: Entity distributions in our training and evaluation sets. Rows 2 to 6 show the top 5 most frequent movies in each dataset. The 7th row shows the total number of movie entities in each set. The last row shows the total number of movies corresponding to these entities.
| Train | Train | Evaluation 1 | Evaluation 1 | Evaluation 2 | Evaluation 2 |
|------------------|---------|-------------------|----------------|--------------------------|----------------|
| Movie | Count | Movie | Count | Movie | Count |
| Hobbit | 194 | Gravity | 30 | Dallas Buyers Club | 41 |
| Frozen | 107 | Hobbit | 27 | Non Stop | 24 |
| Gravity | 106 | Frozen | 26 | Lego Movie | 21 |
| 12 Years a Slave | 96 | 12 Years a Slave | 17 | Lone Survivor | 20 |
| Son of God | 14 | Lord of the Rings | 4 | Jack Ryan Shadow Recruit | 2 |
| Entities | 667 | Entities | 129 | Entities | 115 |
| Movies | 49 | Movies | 20 | Movies | 8 |
optimization useful since that would bring a higher chance of overfitting the model to this particular data. Once we annotate more data, we will perform more thorough optimization and try to improve our results with different learning algorithms.
## 4.3 Results
Table 9 shows incremental updates of our models. The baseline results are achieved by running only candidate identification (Section 3.2). The results for models 1 to 4 are achieved by running both candidate identification and entity classification using orthographic features (Section 3.3.1), n -gram features (Section 3.3.2), supplementary features (Section 3.3.4), and syntactic features (Section 3.3.3).
As expected, our baseline approach gives very good recall (above 96%) but poor precision (around 14%), due to the fact that our gazetteer provides good coverage while causing a large number of false positives for candidate identification. This implies that the dictionary-based approach itself is not good enough to perform movie entity recognition. On the other hand, interesting results are derived from Model 1, where orthographic features play a critical role for the 2nd evaluation set but are not as effective for the 1st evaluation set. From our experiments, we find that these features do not make any noticeable difference when used individually, but give significant improvement when used together (McNemar, p < 0 0001 . ). 6
The n -gram features used in Model 2 give another significant improvement in both evaluation sets (McNemar, p < 0 0001 . and p < 0 01 . for the 1st and 2nd sets, respectively). The supplementary features used in Model 3 also show improvement
6 No orthographic feature except for the 'number of tokens' gives any improvement when it is used individually.
in both evaluation sets, but not found to be significant by the McNemar test. Finally, the syntactic features used in Model 4 give a slight improvement in recall for the 1st evaluation set but hurt both precision and recall for the 2nd evaluation set. Careful studies with more data need to be conducted to understand why these syntactic features did not show any impact or hurt performance on these datasets. 7
The most interesting aspect about our results is that our final model performs better on the 2nd evaluation set, in which no entity corresponds to any movie in the training set. This gives strong evidence that our system is completely unbiased to the training data. Section 5.1 describes the different natures of the 1st and 2nd evaluation sets in details.
## 5 Discussion
## 5.1 Comparison between two evaluation sets
From our error analysis, we notice that tweets in the 2nd evaluation set are much closer to formal English writing than ones in the 1st evaluation set. This explains why the orthographic features help so much on the 2nd set, but not as much on the 1st set. Furthermore, movie entities present in the 1st set consist of more general terms (e.g,. Gravity , Frozen ) such that although our training data provides more contextual information for entities in the 1st set, they are harder to recognize. Nevertheless, the difference between F1-scores achieved for these evaluation sets is not statistically significant, showing that our system performs robustly for movie entities consisting of general terms.
7 We downloaded Ritter et al. (2011)'s system from their open source project and ran on our data without retraining it, which gave F1-scores of 2.94% and 20.69% for the 1st and 2nd evaluation sets, respectively. These results were excluded from Table 9 because it would have not been fair comparison.
Table 9: Results of movie entity recognition. P, R, and F1 represent precision, recall, and F1-score, respectively. The best result for each category is shown in bold.
| | Evaluation 1 | Evaluation 1 | Evaluation 1 | Evaluation 2 | Evaluation 2 | Evaluation 2 |
|-----------------------------------|----------------|----------------|----------------|----------------|----------------|----------------|
| Model | P | R | F1 | P | R | F1 |
| Baseline: Section 3.2 | 14.45 | 96.10 | 25.13 | 13.82 | 96.52 | 24.18 |
| Model 1: Baseline + Section 3.3.1 | 75.00 | 28.13 | 40.09 | 87.60 | 55.65 | 68.09 |
| Model 2: Model 1 + Section 3.3.2 | 89.01 | 63.28 | 73.97 | 88.64 | 67.83 | 76.85 |
| Model 3: Model 2 + Section 3.3.4 | 85.44 | 68.75 | 76.19 | 84.16 | 73.91 | 78.70 |
| Model 4: Model 3 + Section 3.3.3 | 84.76 | 69.53 | 76.39 | 79.25 | 73.04 | 76.01 |
We also notice that the 2nd evaluation set includes more entities corresponding to movies that are contemporary to the time of our data collection. Moreover, the 1st set includes more entities composed of a single token; 67.44% and 22.61% of entities in the 1st and 2nd evaluation sets are composed of a single token, respectively. Single-token entities are usually harder to recognize because they tend to use more general terms than multi-token entities. All these observations show a good direction of how we should split training and evaluation sets for future research.
## 5.2.2 Errors from entity classification
Most errors in our system are caused in the entity classification stage (83.6% of the total errors). Our final model shows much higher precision than recall, which implies that this model generates more false negatives than false positives. A system with good precision and reasonable recall can be useful for cases where false positives are highly undesirable (e.g., pseudo annotation). More annotated data with careful feature selection and hyper-parameter optimization may help to improve this stage further.
## 6 Related work
It is only recently that natural language processing has been actively applied to social media data. The inherent difficulties of this domain have been well explored (Foster et al., 2011; Dent and Paul, 2011). There has been increasing interest in improving NLP components for this domain. Gimpel et al. (2011) and Owoputi et al. (2013) suggested a new set of part-of-speech tags tailor made for Twitter. Their POS tagger achieved improved accuracies on tweets, compared to other existing taggers.
Named entity recognition is of particular interest in Social Media. Current state-of-the-art systems perform well on formal texts (Lin and Wu, 2009; Ratinov and Roth, 2009; Turian et al., 2010); however, the task becomes much harder in the social media domain. Ritter et al. (2011) adapted many NLP components to Twitter data, including a POS tagger, a shallow parser, and an entity chunker. For movie entities, their dictionary-based approach outperformed their statistical approach, which implies that a new NER system may need to be designed for this type of entities.
Guo et al. (2013) performed entity linking, a task closely related to NER, on tweets. They developed a system to jointly detect and disambiguate entities in tweets, using structural learning techniques. Our system is most similar to theirs, in a way that we
## 5.2 Error analysis
We perform error analysis to identify different types of errors caused by each stage of our system, mainly, candidate identification and entity classification. Table 10 shows an example and the proportion of the total errors that each stage causes.
## 5.2.1 Errors from candidate identification
There are three types of errors caused by candidate identification, which together represent 16.4% of the total errors. First, entities including typos or non-exact tokens could not be identified by our approach. For example, an entity such as Lego Film is not present in our gazetteer although there exists an entity very similar to it, Lego Movie , such that no sequence match is found for this entity. Second, abbreviated entities are not handled by our approach (e.g, LOTR → Lord of the Rings ). Third, the maximal sequence matching in Section 3.2.2 sometimes discards valid entities. For the 3rd example in Table 10, Heat should have been identified as a candidate, whereas our approach identifies in Heat as a candidate instead because the latter yields a longer sequence. We could resolve these issues by allowing partial matches (Guo et al., 2013) or combining our dictionary-based approach with statistical approaches (Li et al., 2012).
Table 10: Examples of different types of errors caused by each stage. The % column shows the proportion of the total errors that each stage causes.
| # | % | Example | Note |
|-----|------|--------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------|
| 1 | 16.4 | NonStop Unseats Lego Film I feel like watching LOTR #USER# looks like albino in Heat | Lego Film should be recognized. LOTR should be recognized. Heat should be identified instead of in Heat . |
| 2 | 83.6 | South Park compares her to a #Hobbit Just got text from Nebraska | #Hobbit is not a movie entity. Nebraska is not a movie entity. |
identify and classify entities. However, while they mainly focus on mention detection, we are concerned with building a system to disambiguate a single entity type in a targeted data stream. Further, while they use several gazetteer based features in their classification stage, we only use a gazetteer to generate candidates, thus allowing us to refresh our gazetteer without having to retrain our model.
Li et al. (2012) developed a system for NER in a targeted Twitter stream. They used dynamic programming to segment each tweet into a set of candidate entities by maximizing what they called a stickiness function . In future work, we aim to incorporate a similar statistical method in our candidate identification stage to reduce our reliance on the gazetteer and improve the precision of this stage. Further, they ranked the validity of candidates by constructing a segment graph, and using a random walk model on this graph.
show the feasibility of our work. We first create a targeted dataset of tweets containing movie entities, which we make publicly available. We then present a new system designed for movie entity recognition in tweets. Our system is evaluated on two datasets, and shows results that are unbiased to our training set. More importantly, our system can be dynamically adjusted as new movie entities are introduced without having to be retrained. Although our approach is developed and evaluated only on movie entities, our framework is general enough to be adapted to other types of targetable named entities.
Liu et al. (2011) presented a system which combines a KNN classifier and a CRF classifier to label each word in a tweet. The model is constantly retrained after a pre-determined number of entities are recognized with high confidence. As mentioned previously, our system achieves very good precision, and hence, can be used to pseudo annotate new data. It would be interesting to evaluate the improvements that can be obtained by using such pseudo-annotated data to occasionally retrain our candidate classification stage.
Han and Baldwin (2011) and Kaufmann (2010) examined lexical and syntactic normalization to reduce noise in Twitter data. We use some of these ideas when we normalize tweets in the first stage in our entity recognition pipeline.
## 7 Conclusion
Weprovide a rationale for performing careful analysis and developing a dedicated system to recognize targetable named entities in social media data. In particular, movie entities in tweets are chosen to
Due to the promising results of this preliminary study, we believe there is great potential for future work. We plan to increase the size of our corpus with more types of targetable named entities, using crowdsourcing and pseudo annotation. Moreover, we will improve our gazetteer with more finegrained sources, and explore statistical approaches for candidate identification. We also aim to identify features that are more generalizable, and evaluate the adaptability of these features for entity classification.
## References
Kurt Bollacker, Colin Evans, Praveen Paritosh, Tim Sturge, and Jamie Taylor. 2008. Freebase: A Collaboratively Created Graph Database For Structuring Human Knowledge. In Proceedings of the 2008 ACM SIGMOD International Conference on Management of Data , SIGMOD'08, pages 1247-1250.
Jinho D. Choi and Andrew McCallum. 2013. Transition-based dependency parsing with selectional branching. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics , ACL'13, pages 1052-1062.
Jinho D. Choi and Martha Palmer. 2012. Fast and Robust Part-of-Speech Tagging Using Dynamic Model Selection. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics , ACL'12, pages 363-367.
Kyle Dent and Sharoda Paul. 2011. Through the Twitter Glass: Detecting Questions in Micro-text. In AAAI Workshop on Analyzing Microtext .
Christiane Fellbaum, editor. 1998. WordNet: An Electronic Lexical Database , volume Language, Speech and Communications. MIT Press.
Tim Finin, Will Murnane, Anand Karandikar, Nicholas Keller, Justin Martineau, and Mark Dredze. 2010. Annotating Named Entities in Twitter Data with Crowdsourcing. In Proceedings of the NAACL:HLT Workshop on Creating Speech and Language Data with Amazon's Mechanical Turk , CSLDAMT'10, pages 80-88.
Jennifer Foster, ¨ zlem O C ¸ etinoglu, Joachim Wagner, Joseph Le Roux, Stephen Hogan, Joakim Nivre, Deirdre Hogan, and Josef van Genabith. 2011. #hardtoparse: POS Tagging and Parsing the Twitterverse. In AAAI Workshop on Analyzing Microtext , pages 20-25.
Abhishek Gattani, Digvijay S. Lamba, Nikesh Garera, Mitul Tiwari, Xiaoyong Chai, Sanjib Das, Sri Subramaniam, Anand Rajaraman, Venky Harinarayan, and AnHai Doan. 2013. Entity Extraction, Linking, Classification, and Tagging for Social Media: A Wikipedia-Based Approach. Proceedings of the VLDB Endowment , 6(11):1126-1137.
Kevin Gimpel, Nathan Schneider, Brendan O'Connor, Dipanjan Das, Daniel Mills, Jacob Eisenstein, Michael Heilman, Dani Yogatama, Jeffrey Flanigan, and Noah A. Smith. 2011. Part-of-speech Tagging for Twitter: Annotation, Features, and Experiments. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies , ACL'11, pages 42-47.
Stephen Guo, Ming-Wei Chang, and Emre Kiciman. 2013. To Link or Not to Link? A Study on Endto-End Tweet Entity Linking. In Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics on Human Language Technology , NAACL:HLT'13, pages 1020-1030.
Bo Han and Timothy Baldwin. 2011. Lexical Normalisation of Short Text Messages: Makn Sens a #Twitter. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies - Volume 1 , HLT '11, pages 368-378. Association for Computational Linguistics.
Cho-Jui Hsieh, Kai-Wei Chang, Chih-Jen Lin, S. Sathiya Keerthi, and S. Sundararajan. 2008. A Dual Coordinate Descent Method for Large-scale Linear SVM. In Proceedings of the 25th international conference on Machine learning , ICML'08, pages 408-415.
M. Kaufmann. 2010. Syntactic normalization of twitter messages. In International Conference on Natural Language Processing, Kharagpur, India.
Chenliang Li, Jianshu Weng, Qi He, Yuxia Yao, Anwitaman Datta, Aixin Sun, and Bu-Sung Lee. 2012. TwiNER: Named Entity Recognition in Targeted Twitter Stream. In Proceedings of the 35th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages 721-730.
Dekang Lin and Xiaoyun Wu. 2009. Phrase Clustering for Discriminative Learning. In Proceedings of the 47th Annual Meeting of the Association for Computational Linguistics , ACL'09, pages 1030-1038.
Xiaohua Liu, Shaodian Zhang, Furu Wei, and Ming Zhou. 2011. Recognizing Named Entities in Tweets. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies - Volume 1 , HLT '11, pages 359-367. Association for Computational Linguistics.
Andrew L. Maas, Raymond E. Daly, Peter T. Pham, Dan Huang, Andrew Y. Ng, and Christopher Potts. 2011. Learning Word Vectors for Sentiment Analysis. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies , ACL:HLT'11, pages 142-150.
Tomas Mikolov, Kai Chen, Greg Corrado, and Jeff Dean. 2013. Efficient Estimation of Word Representations in Vector Space. arXiv:1301.3781 .
Olutobi Owoputi, Brendan O'Connor, Chris Dyer, Kevin Gimpel, Nathan Schneider, and Noah A. Smith. 2013. Improved Part-of-Speech Tagging for Online Conversational Text with Word Clusters. In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , NAACL:HLT'13, pages 380-390.
Bo Pang and Lillian Lee. 2004. A sentimental education: Sentiment analysis using subjectivity summarization based on minimum cuts. In Proceedings of the 42Nd Annual Meeting on Association for Computational Linguistics , ACL'04, pages 271-278.
Lev Ratinov and Dan Roth. 2009. Design Challenges and Misconceptions in Named Entity Recognition. In Proceedings of the Thirteenth Conference on Computational Natural Language Learning , CoNLL'09, pages 147-155.
Alan Ritter, Sam Clark, Mausam, and Oren Etzioni. 2011. Named Entity Recognition in Tweets: An Experimental Study. In Proceedings of the Conference on Empirical Methods in Natural Language Processing , EMNLP'11, pages 1524-1534.
Tom Smith. 2013. Twitter Now The Fastest Growing Social Platform In The World. http://blog.globalwebindex.net/twitter-now-thefastest-growing-social-platform-in-the-world/.
Joseph Turian, Lev-Arie Ratinov, and Yoshua Bengio. 2010. Word Representations: A Simple and General Method for Semi-Supervised Learning. In Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics , ACL'10, pages 384394.
Peter Weiner. 1973. Linear Pattern Matching Algorithms. In Proceedings of the 14th Annual Symposium on Switching and Automata Theory , SWAT'73, pages 1-11.
| null |
[
"Sandeep Ashwini",
"Jinho D. Choi"
] |
2014-08-04T19:31:32+00:00
|
2014-08-04T19:31:32+00:00
|
[
"cs.CL"
] |
Targetable Named Entity Recognition in Social Media
|
We present a novel approach for recognizing what we call targetable named
entities; that is, named entities in a targeted set (e.g, movies, books, TV
shows). Unlike many other NER systems that need to retrain their statistical
models as new entities arrive, our approach does not require such retraining,
which makes it more adaptable for types of entities that are frequently
updated. For this preliminary study, we focus on one entity type, movie title,
using data collected from Twitter. Our system is tested on two evaluation sets,
one including only entities corresponding to movies in our training set, and
the other excluding any of those entities. Our final model shows F1-scores of
76.19% and 78.70% on these evaluation sets, which gives strong evidence that
our approach is completely unbiased to any par- ticular set of entities found
during training.
|
1408.0783v1
|
## Quantum statistics of four-wave mixing by a non-linear resonant microcavity
Y. Sherkunov, 1 David M. Whittaker, 2 Henning Schomerus, 1 and Vladimir Fal'ko 1
1 Physics Department, Lancaster University, Lancaster, LA1 4YB, UK
2 Department of Physics and Astronomy, University of Sheffield, Sheffield S3 7RH, UK
We analyse the correlation and spectral properties of two-photon states resonantly transmitted by a non-linear optical microcavity. We trace the correlation properties of transmitted two-photon states to the decay spectrum of multi-photon resonances in the non-linear microcavity.
Generation and controlled propagation of two-photon states in optical circuits represent a new challenge in quantum optics - the next step from the earlier-developed single-photon sources. Photon pairs have been successfully produced using semiconductor quantum dots [1-5] and by coupling superconducting qubits to high-quality microwave resonators [6-12]. However, the controlled propagation of correlated photon pairs across optical circuits remains an unexplored territory.
Here, we propose a theory describing the transmission of two-photon states in optical circuits where junctions are non-linear optical or microwave resonators [13-16]. In such systems, the interaction between photons inside the cavity both shifts the conditions for their resonant transmission and leads to four-wave-mixing - a redistribution of photon energies in the transmitted pair. We show that the four-wave-mixing by the non-linear cavity results in spectral correlations of the transmitted two-photon states, prescribed by the decay of multi-photon resonances in a non-linear cavity, and we analyse quantum statistics of few-photon states emitted by a non-linear resonator under the action of a weak classical pump.
The key element of the optical circuits discussed in this paper is sketched in Fig. 1. It consists of a nonlinear microcavity, with resonance frequency ω c and decay broadening γ = ( γ L + γ R ) / 2 = ω /Q c ( Q stands for the quality factor) determined by couplings γ L R ( ) to the left(right) waveguides. The interaction of two incoming photons of frequency ω 0 inside the cavity generates two outgoing photons with frequencies ω and ω ′ = 2 ω 0 -ω ( ω , ω ′ ≈ ω 0 ). As a function of the frequency of an outgoing photon ω , the characteristic spectral density of the transmitted photons is illustrated by the colour-scale plot in Fig. 1. The spectral density shows distinctive maxima at ω c and 2 ω 0 -ω c . These statements also apply to a nearly monochromatic pair in a two-photon pulse with a temporal extent of τ glyph[greatermuch] γ -1 . Figure 1 also shows that the four-wave mixing is most efficient either when the incoming photons' frequency resonates with the empty cavity mode, or when the total energy 2 glyph[planckover2pi1] ω 0 of the incoming pair coincides with the energy of the interacting photon pair inside the cavity 2 glyph[planckover2pi1] ( ω c + ) u , where 2 glyph[planckover2pi1] u is the interaction energy. Here, we show that spectral properties of the transmitted photon pair are similar over a broad range of cavity and incoming pulse parameters. The spectral density in Fig. 1 illustrates spectral properties of the res-
Figure 1: Two-photon transmission across a non-linear resonator junction in an optical circuit. Top: Schematic of an optical cavity resonator connecting two waveguides. Bottom: Contour plot of spectral function S R L characterising spectral properties of transmitted two-photon states, as a function of frequencies ω of the outgoing photons and ω 0 of the incident monochromatic photons, for γ L = γ R = γ and non-linear coupling u/γ = 4 . The single-photon transmission peak is indicated by the black line at ω = ω 0 ; maxima of S R L at ω = ω c and ω = 2 ω 0 -ω c are traced using the white dashed lines.
onant four-wave-mixing, where, as we show below, the two-photon spectral function can be factorised into an ω 0 -dependent probability to create a two-photon state in the cavity and the spectral density of the following decay of such a state due to the photon escape into the waveguides.
The evolution of photons transmitted by the non-linear resonator sketched in Fig. 1 is modelled using the Hamiltonian,
each waveguide take the form (see Appendix A):
$$H \ & = \ H _ { 0 } + H _ { u } + H _ { L } + H _ { R }, \quad & ( \\ H _ { 0 } \ & = \ \omega _ { c } \alpha ^ { \dagger } \alpha, \\ H _ { u } \ & = \ u \alpha ^ { \dagger } \alpha ^ { \dagger } \alpha \alpha, \\ H _ { L } \ & = \ \int d x [ \beta _ { L } ^ { \dagger } ( x ) \left ( - i \frac { \partial } { \partial x } \right ) \beta _ { L } ( x ) \\ & + \ \sqrt { \gamma _ { L } } \delta ( x ) ( \beta _ { L } ^ { \dagger } ( x ) \alpha + \alpha ^ { \dagger } \beta _ { L } ( x ) ) ], \\ H _ { R } \ & = \ \int d x [ \beta _ { R } ^ { \dagger } ( x ) \left ( - i \frac { \partial } { \partial x } \right ) \beta _ { R } ( x ) \\ & + \ \sqrt { \gamma _ { R } } \delta ( x ) ( \beta _ { R } ^ { \dagger } ( x ) \alpha + \alpha ^ { \dagger } \beta _ { R } ( x ) ) ], \\ \intertext { e r e } \alpha / \alpha ^ { \dagger } \text{ is the annihilation/creation operator of $t$}.$$
$$& \text{ each waveguide take the form (see Appendix A):} \\ ( 1 ) & \quad a _ { i j } \ = \ c _ { i } c _ { j } - \frac { u \gamma _ { L } \sqrt { \gamma _ { i } \gamma _ { j } } e ^ { i ( \omega _ { 0 } - \omega _ { c } ) - \gamma ] | x _ { 1 } - x _ { 2 } | } } { ( \omega _ { 0 } - \omega _ { c } + i \gamma ) ^ { 2 } ( \omega _ { 0 } - \omega _ { c } - u + i \gamma ) }, \ ( 4 ) \\ c _ { L } \ = \ - \frac { \omega _ { 0 } - \omega _ { c } + i ( \gamma _ { R } - \gamma _ { L } ) / 2 } { \omega _ { 0 } - \omega _ { c } + i \gamma }, \\ x ) & \quad c _ { R } \ = \ \frac { i \sqrt { \gamma _ { L } \gamma _ { R } } } { \omega _ { 0 } - \omega _ { c } + i \gamma }. \\ ( x ) ) ], & \quad \text{$\mathbb{m}$} \ = \ \infty \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \colon \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \text{$\mathbb{m}$} \, \end{cases}$$
where α α / † is the annihilation/creation operator of the resonant cavity mode with energy ω c (here, fundamental constants are set to c = 1 and glyph[planckover2pi1] = 1 ). The non-linear term, H u , in Eq. (1) with coupling constant u , leads to photon-photon interaction, shifting the two-photon state energy by 2 u . This non-linearity is generic for inversion-symmetric media; in particular, it can be the result of the repulsion between exciton-polaritons in a microcavity with a large Rabi frequency [13-16]. The resonant cavity mode is coupled to the photons in the left (L) and right (R) semi-infinite waveguides, described by annihilation operators in the real-space representation, β L R ( ) ( x ) = 1 √ s ∑ k> 0 e ikx β L R ( ) ( k ) ( k is the wave-number and s is the length of the waveguide), and β L R ( ) ( x ) are defined to describe incoming photons for x < 0 and outgoing photon for x > 0 (irrespective of the side of the resonator). The transformation of propagating waves into the cavity mode (and their reflection from the cavity) takes place at x = 0 .
The solution for the two-photon transmission/reflection problem is obtained by studying the evolution of the two-photon states | ψ t ( ) 〉 , governed by the Schrödinger equation, i∂ t | ψ t ( ) 〉 = H ψ t | ( ) 〉 . We match the initial state to the incident photon pair in the left waveguide,
$$| \psi \rangle _ { t \to - \infty } = \frac { e ^ { i \omega _ { 0 } ( x _ { 1 } + x _ { 2 } - 2 t ) } } { \tau \sqrt { 2 } } f ( x _ { 1 }, x _ { 2 }, t ) \beta _ { L } ^ { \dagger } ( x _ { 1 } ) \beta _ { L } ^ { \dagger } ( x _ { 2 } ) | 0 \rangle, ( 2 ) \quad \text{var} \ \text{(see} } \\ \tau \ \text{(see} }$$
where the two-photon pulse envelope f ( x , x 1 2 , t ) has a temporal extent τ and is modelled using the scenarios listed in Table I. The projection of the transmitted/reflected photon pairs onto the outgoing states in the left/right leads ( i = L/R) is described using
$$| \psi \rangle _ { t \rightarrow \infty } = \frac { e ^ { i \omega _ { 0 } ( x _ { 1 } + x _ { 2 } - 2 t ) } } { \tau \sqrt { 2 } } a _ { i j } ( x _ { 1 }, x _ { 2 }, t ) \beta _ { i } ^ { \dagger } ( x _ { 1 } ) \beta _ { j } ^ { \dagger } ( x _ { 2 } ) | 0 \rangle, ( 3 )$$
where, for a monochromatic photon pair ( τ → ∞ ), the amplitudes a LL ( a RR ) of obtaining two photons in the left(right) waveguide and a LR of finding one photon in
Here, c i is a single-photon reflection (L) and transmission (R) amplitude in the circuit with a resonant cavity. The first term in Eq. (4) accounts for independent scattering of photons. The second term describes the correlated two-photon states formed due to the photon-photon interaction inside the cavity. Spectral properties of the amplitudes a ij , calculated for various finite pulse widths, are illustrated in Fig. 2, where we show that the two-photon transmission by the non-linear cavity is strongly affected by the photon-photon interaction, but does not depend much on the detailed spectral shape of the two-photon pulse. The examples in Fig. 2(a,b) demonstrate that the interaction of two photons inside the cavity produces their anti-bunching for ω 0 = ω c , but also generates an additional transmission peak at ω 0 = ω c + u corresponding to the formation of resonant two-photon states inside the cavity. Note that in the limit u → ∞ , the model in Eq. (1) corresponds to the photon blockade regime studied in Ref. [17] for an optical junction consisting of a two-level system, and, in this limit, Eq. (4) reproduces the photon anti-bunching predicted in Ref. [17].
Table I: Incident states
| State | f ( x 1 , x 2 , t ) |
|-----------------------------------------|--------------------------------------------|
| (i) Monochromatic Uncorrelated Gaussian | 1 [( x 1 - t ) 2 +(( x 2 - t ) 2 ] / 2 τ 2 |
To characterise the spectral properties of the photon pair transmitted/reflected by the non-linear cavity into a waveguide i from a pulse arriving over a temporal interval τ (see Table I) in waveguide j , we consider the spectral density of photons detected at a distant position x 0 →∞ in waveguide i , [19]
$$S _ { j } ^ { i } ( \omega ) \equiv \int _ { - \infty } ^ { \infty } \frac { d t } { 2 \pi } e ^ { - i \omega t } \langle \beta _ { i } ^ { \dagger } ( x _ { 0 } ) \beta _ { i } ( x _ { 0 } - t ) \rangle _ { j }.$$
We find that the latter can be written in the form
$$S _ { j } ^ { i } ( \omega ) = \frac { \gamma _$$
where the first term describes the photon-photon correlations introduced by the interaction between two photons simultaneously appearing inside the cavity, whereas
Figure 2: Transmission, | a LL | 2 , reflection, | a RR | 2 , and transmission-reflection, | a LR | 2 amplitudes of two-photon states in various parametric regimes. (a) Non-linear cavity with τγ = 7 ( γ L = γ R = γ ) and u = 4 γ . Frequency dependence of two-photon amplitudes | a ij | 2 at x 1 = x 2 for (i) a monochromatic incident wave, (ii) uncorrelated initial pair of photons, and (iii) a correlated Gaussian initial state (see Table I). (b) Same as (a), but for τγ = 1 . (c) Transmission by a linear cavity ( u = 0 ).
the second term takes into account elastic (uncorrelated) single-photon transfer. In Eq. (5), the factor
$$F ( \omega _ { 0 } ) \ = \ \frac { 2 \gamma ^ {$$
is the probability to form the correlated two-photon state inside the cavity, whereas the factor
$$\tilde { F } ( \omega ) \ = \ \frac { 4 u ^ { 2$$
describes the spectral density of the photons emitted to the left/right waveguide. Finally, single-photon transfer is described by I L L = I R R = 2 | c L | 2 /τ and I R L = I L R = 2 | c R | 2 /τ .
The weak dependence of the correlation properties of scattered photons on the correlation properties of the initial state seen in Fig. 2 indicates that the result in Eq. (5) is applicable to describe correlated photon pairs produced by incident two-photon states with different correlation properties, provided that τγ glyph[greatermuch] 1 . Moreover, the two-photon spectral correlations, described by Eq. (5), emerge even in the case of initial state with different quantum statistics. To demonstrate this, we studied the emission spectrum by a non-linear resonator excited by a coherent state | in 〉 = e -| b | 2 / 2 e b β † ω 0 | 0 〉 incident from one of the waveguides, e.g, j = L . Here, | b | characterizes the amplitude 〈 in | β | in 〉 of the incoming field pulse. In this analysis, we exploit the possibility to relate the spectral function S R L to the correlation function of the cavity resonance mode, as established in Ref. [18],
$$S _ { L } ^ { R } ( \omega ) = \gamma _ { R }$$
where the operators α , α † are used in the interaction representation, and ρ is the reduced density matrix obtained from the full density matrix describing the system by tracing out the degrees of freedom associated with the quantum waveguide modes. The time dependence of the reduced density matrix ρ obeys the Lindblad equation:
$$\begin{array} {$$
where U t , t ( ′ ) is the evolution super operator of the reduced density matrix. When U t, t ( ′ ) is found using perturbation theory (see Appendix B) for a weak nonlinearity, u glyph[lessmuch] γ , this results in the spectral density described by Eq. (5), but rescaled as S R L → | | b 4 S R L . A typical result of the numerical solution of Eq. (7) for a strong non-linearity but weak pumping ( 〈 α α † 〉 glyph[lessmuch] 1 ) is shown in Fig. 3. It indicates that in the case of weak coherent pumping the spectrum of emitted photons is determined by two-photon correlations since its density coincides with that described by Eq. (5).
In addition, we analyse the conditions for the generation of correlated few-photon states with two, three and four photons for a non-linear cavity excited by a classical field. The appearence of such states is quantified using the correlation function
$$g ^ { ( n ) } = \langle ( \beta _ { i } ^ { \dagger } ) ^ { n } \beta _ { i } ^ { n } \rangle / \langle \beta _ { i } ^ { \dagger } \beta _ { i } \rangle ^ { n },$$
where β -operators are taken at coinciding time. By generalising the analysis in Ref. [18], this can be expressed
Figure 3: Generation of few-photon states in a non-linear resonator excited by a classical field. (a) Spectral function S R L ( ω ) (normalized to its maximum value) for a cavity excited by a classical field source for γ L = γ R = γ , ( ω 0 -ω c ) /γ = -4 , u/γ = 4 , √ b/γτ = 0 1 . , ( |〈 α 〉| 2 = 5 9 . × 10 -4 ). Dots represent the results of numerical calculations, and solid line a fit to the spectral density described by Eq. (5). (b) Few-photon correlation function g ( n ) ( ω 0 ) for two-, three-, and four-photon states emitted to the right waveguide by a non-linear optical resonator excited by a classical field arriving from the left waveguide.
in terms of the photon operators inside the cavity,
$$g ^ { ( n ) } = \text{Tr} [ ( \alpha ^ { \dagger } ) ^ { n } ( \alpha ) ^ { n } \rho ] / \text{Tr} [ \alpha ^ { \dagger } \alpha ] ^ { n },$$
which can be determined from Eq. (7). The results of numerical simulations for n = 2 3 4 , , are shown in Fig. 3 b. Here, the second-order correlation function shows antibunching ( g (2) < 1 ) for red-detuned pumping frequencies and bunching ( g (2) > 1 ) for a blue-shifted frequencies, in agreement with [15]. Also, the functions g ( n ) peak at pumping frequencies close to ω 0 = ω c +( n -1) u , which correspond to the resonance conditions for a simultaneous excitation of n photons inside the non-linear cavity.
To summarise, we show that two-photon states resonantly transmitted by a non-linear optical microcavity display spectral correlations, which are almost independent of the correlation properties of the incoming photon pair and can be traced to the decay spectrum of multi- photon resonances in the cavity.
We thank B. Altshuler, J. Bloch, A. Cleland, Y. Pashkin, and M. Skolnick for useful discussions. This work was supported by EPSRC Programme Grant EP/J007544.
## Appendix A: Propagation of two-photon states: Scattering approach
A generic two-photon state | ψ t ( ) 〉 , whose evolution obeys i ∂ ∂t | ψ t ( ) 〉 = H ψ t | ( ) 〉 , can be written as
$$\begin{array} { c c l } \text{obeys} \, i _ { \vec { 0 } t } ^ { \vec { \phi } } | \psi ( t ) \rangle = H | \psi ( t ) \rangle, \, \text{can be written as} \\ \\ | \psi ( t ) \rangle \, = \, 1 / \sqrt { 2 } \left [ \int d x _ { 1 } d x _ { 2 } \psi _ { L L } ( x _ { 1 }, x _ { 2 }, t ) \beta _ { L } ^ { \dagger } ( x _ { 1 } ) \beta _ { L } ^ { \dagger } ( x _ { 2 } ) | 0 \rangle \\ \\ + \, \int d x _ { 1 } d x _ { 2 } \psi _ { R R } ( x _ { 1 }, x _ { 2 }, t ) \beta _ { R } ^ { \dagger } ( x _ { 1 } ) \beta _ { R } ^ { \dagger } ( x _ { 2 } ) | 0 \rangle \right ] \\ \\ + \, \sqrt { 2 } \int d x _ { 1 } d x _ { 2 } \psi _ { L R } ( x _ { 1 }, x _ { 2 }, t ) \beta _ { L } ^ { \dagger } ( x _ { 1 } ) \beta _ { R } ^ { \dagger } ( x _ { 2 } ) | 0 \rangle \\ \\ + \, \int d x \left [ \psi _ { L C } ( x, t ) \beta _ { L } ^ { \dagger } ( x ) \alpha ^ { \dagger } + \psi _ { R C } ( x, t ) \beta _ { R } ^ { \dagger } ( x ) \alpha ^ { \dagger } \right ] | 0 \rangle \\ \\ + \, 1 / \sqrt { 2 } \psi _ { C C } ( t ) \alpha ^ { \dagger } \alpha ^ { \dagger } | 0 \rangle. \quad & ( A 1 ) \\ \\ \text{ Here,} \, | 0 \rangle \, \text{is the vacuum state,} \, \psi _ { i j } \, \text{are the wave functions} \end{array}$$
Here, | 0 〉 is the vacuum state, ψ ij are the wave functions describing photons in the left (right) waveguide ( i, j = L R ( ) ), or in the cavity ( i, j = C ). This leads to the following system of simultaneous equations:
- i ( ∂ x 1 + ∂ x 2 + ∂ t ) ψ ij ( x , x 1 2 , t ) = 1 / √ 2 [ κ δ x i ( 1 ) ψ jC ( x , t 2 )
- + κ δ x j ( 2 ) ψ iC ( x , t 1 )] ,
- i ( ∂ x + ∂ t + iω c ) ψ iC ( x, t ) = √ 2 [ κ i ( δ x ψ ( ) CC ( ) t
glyph[negationslash]
- + ψ ii (0 , x, t )) + κ ψ j ij ( x, 0 , t )] , i = j, √
- i ( ∂ t + 2( i ω c + u )) ψ CC ( ) = t 2 [ κ L ψ LC (0 , t )
- + κ R ψ RC (0 , t )] .
For the incident state | ψ 〉 t →-∞ = 1 τ √ 2 e iω 0 ( x 1 + x 2 -2 ) t f ( x , x 1 2 , t ) β † L ( x 1 ) β † L ( x 2 ) 0 | 〉 (see Eq. (2)) and ψ CC ( t → -∞ ) = ψ LC ( t → -∞ ) = ψ RC ( t → -∞ ) = 0 , we find that
$$- \infty \, = \, 0, \, \text{we find that} \\ \quad \, \text{of nu-} \\ g. \, 3 \, b. \, \begin{array} { c c c } \psi _ { L L } ( x _ { 1 }, x _ { 2 }, t ) & = & \frac { e ^ { i \omega _ { 0 } ( x _ { 1 } + x _ { 2 } - 2 t ) } } { \tau } \, [ f ( x _ { 1 }, x _ { 2 }, t ) \theta ( - x _ { 1 } ) \theta ( - x _ { 2 } ) \\ s \, \text{anti-} \\ \text{encies}, \\ \text{encies}, \end{array} & \quad + & \, c _ { L } ( \theta ( x _ { 1 } ) \theta ( - x _ { 2 } ) + \theta ( - x _ { 1 } ) \theta ( x _ { 2 } ) ) ] \,, \\ \quad \, \text{each at} \, \begin{array} { c } \psi _ { R R } ( x _ { 1 }, x _ { 2 }, t ) & = & \frac { e ^ { i \omega _ { 0 } ( x _ { 1 } + x _ { 2 } - 2 t ) } } { \tau } \, a _ { R R } ( x _ { 1 }, x _ { 2 }, t ) \theta ( x _ { 1 } ) \theta ( x _ { 2 } ), \\ \text{latten-} \\ \text{avity.} \end{array} & \quad \, \psi _ { L R } ( x _ { 1 }, x _ { 2 }, t ) & = & \frac { e ^ { i \omega _ { 0 } ( x _ { 1 } + x _ { 2 } - 2 t ) } } { \tau } \, [ a _ { L R } ( x _ { 1 }, x _ { 2 }, t ) \theta ( x _ { 1 } ) \theta ( x _ { 2 } ) \\ s \, \text{res-} \\ \text{cavity} \end{array} \\ \text{depen-} \quad \, \text{where} \, a _ { L L }, \, a _ { R R } \, \text{and} \, a _ { L R } \, \text{are the amplitudes describing}$$
where a LL , a RR and a LR are the amplitudes describing two reflected photons, two transmitted photons and one reflected and one transmitted photons respectively. In
the case of monochromatic incident wave ( τγ glyph[greatermuch] 1 ) (i), they are given by:
$$\mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \mathbb { m } { \nu } \\ a _ { i j } ( x _ { 1 }, x _ { 2 } ) \ & = \ c _ { i } c _ { j } & & \quad \ u \gamma _ { L } \sqrt { \gamma _ { i } \gamma _ { j } } \exp \{ [ i ( \omega _ { 0 } - \omega _ { c } ) - \gamma ] | x _ { 1 } - x _ { 2 } | \} \\ & \quad - \ \frac { u \gamma _ { L } \sqrt { \gamma _ { i } \gamma _ { j } } \exp \{ [ i ( \omega _ { 0 } - \omega _ { c } - u + i \gamma ) }, \quad X ( 0 ) \ \colon \\ c _ { L } \ & = \ - \frac { \omega _ { 0 } - \omega _ { c } + i ( \gamma _ { R } - \gamma _ { L } ) / 2 } { \omega _ { 0 } - \omega _ { c } + i \gamma }, \\ c _ { R } \ & = \ \frac { i \sqrt { \gamma _ { L } \gamma _ { R } } } { \omega _ { 0 } - \omega _ { c } + i \gamma }.$$
## Appendix B: Classical pumping, weak interaction limit
In the case of weak non-linearity 〈 α α u † 〉 glyph[lessmuch] γ , Eq. (7) can be solved analytically by expanding the density matrix over a coherent basis | a 〉 so that ρ = ∫ P a ( ) | a 〉〈 a d | ( glyph[Rfractur]{ a } ) d ( glyph[Ifractur]{ a } ) ( P -representation) [18] and converting into a Fokker-Planck equation with the help of the operator equivalence relations: αρ → aP a ( ) , α ρ † → ( a ∗ -∂ a ) P a ( ) , ρα → ( a -∂ a ∗ ) P a ( ) , ρα † → a P a ∗ ( ) . Switching to the rotating frame ( a → ae -iω t 0 ), we find:
$$\frac { \partial P } { \partial t } \ = \ - \sum _ { i }$$
$$\text{where} \quad a \quad \ = \quad \left ( \begin{array}$$
$$\begin{array} {$$
$$\begin{array} {$$
is equivalent to the Langevin equation [18]:
$$\partial _ { t } a = B + \zeta ( t ), \text{$$
where ζ i ( ) t is delta-correlated random noise terms so that 〈 ζ i ( ) t ζ j ( t ′ ) 〉 = D δ t ij ( -t ′ ) .
For small fluctuations, we can expand a around the ensemble average 〈 a 〉 , which satisfies the steady-state equation |〈 a 〉| 2 [ γ 2 +( ω 0 -ω c -2 u |〈 a 〉| 2 ) 2 ] = | f | 2 . We introduce new real variables, r and θ , so that a = 〈 a 〉 (1 + r -iθ )
$$\text{and rewrite $Eq.$ (B2) for $X=\binom{r}{\theta}\end{array} \rightarrow \text{$\omega$} \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \codeots \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \. \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \tau \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \}$$
$$\partial _ { t } { \mathbf X } = A { \mathbf X } + \eta ( t ),$$
$$1 ) \text{ (i)}, \quad \text{where } A = \left ( \begin{array} { c } - \gamma & \omega _ { 0 } - \omega _ { c } - 2 | \langle a \rangle | ^ { 2 } u \\ - ( \omega _ { 0 } - \omega _ { c } - 6 | \langle a \rangle | ^ { 2 } u ) & - \gamma \end{array} \right ),$$
$$1 ) \ ( i ), \quad & \text{where } A = \begin{pmatrix} - \gamma & \omega _ { 0 } - \omega _ { c } - 2 | \langle a \rangle | ^ { 2 } u \\ - ( \omega _ { 0 } - \omega _ { c } - 6 | \langle a \rangle | ^ { 2 } u ) & - \gamma \end{pmatrix}, \\ - x _ { 2 } | \} \\ \frac { \eta = \frac { 1 } { 2 } \begin{pmatrix} \zeta _ { 1 } / \langle a \rangle + \zeta _ { 2 } / \langle a ^ { * } \rangle \\ i ( \zeta _ { 1 } / \langle a \rangle - \zeta _ { 2 } / \langle a ^ { * } \rangle ) \end{pmatrix}. \text{ For the initial condition} \\ \frac { - x _ { 2 } | \} } { \gamma ) }, \quad & X ( 0 ) = 0, \, \text{Eq.} \ ( B 3 ) \, \text{has the solution}$$
$$\mathbf X ( t ) = e ^ { A t } \int _ { 0 } ^ { t } e ^ { - A t ^ { \prime } } \eta ( t ^ { \prime } ) d t ^ { \prime },$$
which for the spectral function S R L ( ω ) (Eq. (6)) yields
$$\underset { \text{and} } = \underset { \text{and} } \begin{array} { c } S _ { L } ^ { R } ( \omega ) = | b | ^ { 4 } \left ( I _ { L } ^ { R } \delta ( \omega - \omega _ { 0 } ) + \frac { \gamma _ { L } ^ { 2 } \gamma _ { R } } { \gamma ^ { 3 } } F ( \omega _ { 0 } ) \tilde { F } ( \omega ) \right ), ( \text{B5} ) \end{array}$$
where F ω ( 0 ) and ˜ ( F ω ) are given below Eq. (5).
- [1] R. M. Stevenson, et al. Nature 439 , 179 (2006).
- [2] N. Akopian, et al. Phys. Rev. Lett. 96 , 130501 (2006).
- [3] C. L. Salter, et al. Nature 465 , 594 (2010).
- [4] A. J. Bennett, et al. Nature Phys. 6 , 947 (2010).
- [5] R. M. Stevenson, et al. Phys. Rev. Lett. 108 , 040503 (2012).
- [6] A. Wallraff, et al. Nature 431 , 162 (2004).
- [7] D. I. Schuster, et al. Nature 445 , 515 (2007).
- [8] Max Hofheinz, et al. Nature 454 , 310 (2008).
- [9] Max Hofheinz, et al. Nature 459 , 546 (2009).
- [10] Matteo Mariantoni, et al. Nature Phys. 7 , 287 (2011).
- [11] M. H. Devoret and R. J. Schoelkopf Science 339 , 1169 (2013).
- [12] Yi Yin, et al. Phys. Rev. Lett. 110 , 107001 (2013).
- [13] K. Nozaki, et al. Nature Photon. 6 , 248 (2012).
- [14] H. S. Ngueng, et al. Phys. Rev. Lett. 110 , 236601 (2013).
- [15] A. Verger, C. Ciuti and I. J. Carusotto Phys. Rev. B 73 , 193306 (2006).
- [16] I. Carusotto, et al. Phys. Rev. Lett. 103 , 033601 (2009). CA, 2003).
- [17] J.-T. Shen and S. Fan Phys. Rev. Lett. 98 , 153003 (2007).
- [18] D. F. Walls and G. J. Milburn Quantum Optics (Springer-Verlag, Berlin, 1994).
- [19] We assume a linear dispersion of the wave guide, which implies the following relation for the spatial and temporal variables x = ct , where c is the speed of light .
|
10.1103/PhysRevA.90.033845
|
[
"Y. Sherkunov",
"David M. Whittaker",
"Henning Schomerus",
"Vladimir Fal'ko"
] |
2014-08-04T19:34:55+00:00
|
2014-08-04T19:34:55+00:00
|
[
"quant-ph"
] |
Quantum statistics of four-wave mixing by a non-linear resonant microcavity
|
We analyse the correlation and spectral properties of two-photon states
resonantly transmitted by a non-linear optical microcavity. We trace the
correlation properties of transmitted two-photon states to the decay spectrum
of multi-photon resonances in the non-linear microcavity.
|
1408.0784v2
|
## Blindspot: Indistinguishable Anonymous Communications
Joseph Gardiner and Shishir Nagaraja School of Computer Science University of Birmingham Birmingham, UK, B15 2TT
Email: j.gardiner, [email protected]
Abstract -Communication anonymity is a key requirement for individuals under targeted surveillance. Practical anonymous communications also require indistinguishability - an adversary should be unable to distinguish between anonymised and nonanonymised traffic for a given user. We propose Blindspot, a design for high-latency anonymous communications that offers indistinguishability and unobservability under a (qualified) global active adversary. Blindspot creates anonymous routes between sender-receiver pairs by subliminally encoding messages within the pre-existing communication behaviour of users within a social network. Specifically, the organic image sharing behaviour of users. Thus channel bandwidth depends on the intensity of image sharing behaviour of users along a route. A major challenge we successfully overcome is that routing must be accomplished in the face of significant restrictions channel bandwidth is stochastic . We show that conventional social network routing strategies do not work. To solve this problem, we propose a novel routing algorithm. We evaluate Blindspot using a real-world dataset. We find that it delivers reasonable results for applications requiring low-volume unobservable communication.
In anonymous communication networks, anonymity is provided under the assumption that users solely communicate via anonymous channels. This is unrealistic as widespread adoption of anonymous communications is hindered by usability problems (performance overheads) as well as lack of awareness. Therefore, users of anonymous communication networks also have to communicate over conventional channels. Thus the statistical characteristics of the anonymous channel of a user must be equivalent to (or indistinguishable from) non-anonymous communication channels. Indistinguishability is the property that anonymous and non-anonymous communications are not differentiable by an adversary. Indeed, without indistinguishability, anonymous communication networks have a basic usability problem .
## I. INTRODUCTION
Anonymous communication networks such as Tor and JAP [15], [26] have, in a short period of time, revolutionised communication security landscape on the Internet. These systems and others [11], [10] provide anonymous communications which can defend against traffic analysis attacks. Often, an adversary will try to build a dossier of the most central nodes, and attacking those on the top of the list. Traffic analysis of inter-node communication offers basic tools to collect necessary intelligence in order to plan an attack -i.e an understanding of their communication traffic behaviour -for instance, who they talk to, how often, and the frequency and timing of communication (all of this is contained in metadata, which is collected, and stored, by GCHQ as part of Tempora).
This information is inferred from the traffic data at appropriate vantage points on the Internet. A map of internet backbone links shows obvious positions where large percentages of internet traffic pass through. For example, a majority of the traffic between the USA and Europe passes through Cornwall, where GCHQ also happens to have an listening station at Bude. According to the Snowden revelations, this station captures all traffic passing over ten of these undersea cables as part of the Tempora program. This justifies the need to consider global active adversaries in our threat model.
Current approaches to this challenge take two directions. One line of work is on censorship resistance mechanisms. These works [4], [25], [23], [35], [2] all introduce extra communication endpoints (such as cloud file storage services) into traffic, meaning that there is some form of distinguishable behaviour to monitor. Approaches that make use of protocolmimicking (attempting to hide one type of traffic by making it appear as another) [30], [40], [42] fall victim to incomplete emulation of the protocols that they are attempting to mimic [21].
A second direction is the design of anonymous communication networks. The unlinkability property [32] alone is inadequate for a number of reasons. First, as Danezis et al. pointed out, it does not hide the volume of communications and hence leaks enough information to a global adversary who can compile an ordered list of targets [13]. Second, it does not defend against traffic confirmation attacks [33], [9] where an adversary injects traffic load patterns to determine communicating end points. To combat these problems full unobservability - passive attacker cannot distinguish whether or not a user is communicating - is needed. Systems such as Drac [10] provide unobservable anonymous communications. However, the attacker can still distinguish between the following communicating states of a user: whether or not 'unobservable' anonymous channels are in use, so the protection accorded is still short of what users need.
## A. Motivation
Without indistinguishability, the use of anonymous communications alone could incriminate a user simply by identifying their use. The use of anonymous communication channels can cause escalations such as a targeted intrusion attack on the user or worse, trigger a physical attack on the individual. Anonymity without company [14], can make identification of a user of anonymous communications much easier. For example, in 2013 a student at Harvard University sent an bomb threat through email in order to get his imminent economics exam cancelled [36]. The student used Tor to send an email via the Guerrilla [28] disposable email service, assuming he would be unidentifiable. The investigators simply compiled a list of Tor users filtered by the timing of the threat to identify the student.
## B. Research problem
Credible defences against targeted surveillance attacks require both indistinguishability and unobservability. This places significant restrictions on the statistical characteristics of user traffic - the traffic shape should be independent of the presence of anonymous communications.
To address these requirements, we propose Blindspot , an anonymous communication network that leverages conventional communication channels on social networks. Specifically, the image-exchange behaviour of users. In Blindspot, nodes communicate by broadcasting messages to their neighbours. Messages are steganographically embedded within an image the user uploads. When uploaded by the communicating (sender) node, the message carrying image is available to all its neighbours. Each participating node checks for incoming messages by monitoring images uploaded by its neighbours. It provides indistinguishability by ensuring that the image upload behaviour of participating users remains unchanged and provides probabilistic unobservability through the use of steganography.
To the best of our knowledge, Blindspot is the first system to provide both properties. This may make it a useful building block of some forms of anonymous communications, such as announcing a meeting of a social club (corporate gathering, undercover organisation, dissident organisation, or a protest group).
Blindspot achieves these properties through the use of social trust relationships within the routing process. Unlike previous proposals based on steganographic techniques [4], [25], [23], [35], [2], which introduce additional communication endpoints, Blindspot routes through the pre-existing social network, thus exploiting trust relationships to secure routing. We leverage this in four ways. First, all routing is through preexisting social fabric interconnecting nodes, therefore no new communication endpoints are introduced. Second, it leverages the diffusion properties of the social network to efficiently and anonymously route to the destination. Third, Blindspot does not alter statistical characteristics of conventional user traffic. This is achieved by piggy-backing on existing image sharing behaviour, no extra images are added nor are extra uploads scheduled. Channel traffic characteristics, as observable to the adversary, are not altered. So routing depends wholly on the innate image sharing behaviour of the users. This is a nontrivial challenge. Blindspot operates on a network topology where channel bandwidth is severely constrained. And, the latency is a function of stochastic user behaviour along the entire route. Fourth, social networks can be used to detect Sybil attacks [44], [45], [12], [29] by detecting a small cut that separates honest and sybil groups. Blindspot benefits from Sybil resistance properties of social network topologies. Specifically, this restricts the power of adversaries that seek to inject large numbers of misbehaving participants into the network.
We evaluate Blindspot empirically using real-world social network topologies and real user behaviour. From our evaluation, we find that Blindspot is able to route 90% of messages within a day across diverse routes. We also validate our work on a synthetic social network topology.
The structure of this paper is as follows. First, in Section II we will give a more formal definition of the problem. This will be followed by an overview of the system design (Section III), and then a full evaluation of the systems performance in Section IV and discussion of the security provided by this system in Section V. Finally, we will give an overview of existing systems in Section VI.
## II. PROBLEM DESCRIPTION
Blindspot's goal is to offer high-latency anonymous communications with the following properties: (a) Indistinguishability between (high latency) anonymised and non-anonymised communications of a given user. (b) Unlinkable communication between sender and receiver pairs. (c) Low-delay any-to-any routing in a network with stochastic node behaviour .
In this situation low-delay is used in relative terms. The system will not be able to achieve latencies that match that of systems such as Tor, rather message delivery can be measured in hours or even days. The low-delay property represents the routing algorithm's design that aims to reduce this delay as much as possible.
## A. Low-delay routing in a capacity constrained network
Consider a social network represented by a graph G = ( V, E ) , where nodes are people and edges are trust relationships. Each node v i ∈ V in the graph is associated with a probability distribution over time elapsed between images posted by the corresponding user. The time between image postings is the minimum amount of time that a message can spend sitting on a node. This is called the delay distribution of v i .
A route of length k on the network is a sequences of vertices v , v 1 2 , . . . , v k . A message sent along this route will be influenced by the delay distribution of all the nodes along the route. We assume that a node downloads all messages broadcast by its neighbours whenever it uploads. This assumption is based on the current operational behaviour of many OSNs
(which predicatively cache image posts from neighbours for performance reasons).
The problem of routing in a network with weighted nodes, where each weight is an independent random variable, is an unsolved problem. Note, that graph weights are not generated from a probability distribution function. Instead, each node is associated with a separate PDF which is different (and independent) of all other nodes in the graph.
## B. Indistinguishability
In a nutshell, indistinguishability is defined as the inability to distinguish between anonymous and non-anonymised traffic traces of a specific user. Blindspot leverages photo exchange traffic of users on social networks as the carrier channel. We formally define the Indistinguishability Game (inspired by Cryptography literature), played between the attacker and a communicating user: First, consider photo upload traffic P X 0 generated from a distribution over inter-upload delay times of the user X , when the user is not communicating via Blindspot. Next, consider the photo upload traffic P X 1 when the user is communicating via Blindspot. Now a random i ∈ 0 1 , is generated by a fair coin flip, the defender is given one or more click traffic sets from P i and he/she outputs i ′ . The attacker wins the game if i = i ′ . A communication channel is 'indistinguishable', if no attacker can win the Indistinguishability Game with probability non-negligibly greater than 0 5 . .
## C. Threat model
The Blindspot system attempts to achieve its goals under two types of adversaries. Firstly, the primary threat as considered by this system is the global active adversary. This refers to the OSN provider (for example, Facebook), who we assume has access to all images uploaded to the network and the full underlying social graph. We also assume that the OSN has the capability to modify images in any way that they see fit. The OSN may be colluding (not necessarily knowingly) with an external surveillance agency (for example the NSA or GCHQ).
Secondly, we consider the partially active insider. We assume that the adversary can insert dishonest nodes, either through the introduction of Sybil nodes or by compromising existing users. These nodes can be used to monitor traffic on the network, by collecting any images that pass over them, and has access to the embedded data. The insider can perform tagging attacks on the messages to attempt to remove unlinkability, and can instruct insiders to perform DOS attacks through blackholing (refusing to route messages), or by flooding messages in order to disrupt routing. Blindspot offers partial defences against a subset of active insider attacks, but not all of them. A detailed discussion on the system's resistance and susceptibility to various attacks can be found in Section V.
## D. Intended Use
The system in its earliest form will only support the sending of short, text based messages (around the length of a tweet), including anything that can be encoded into this length of text as hexadecimal or base64 strings. The message data could undergo external compression in order to maximise capacity. For data that is too long to send, multiple messages could be sent provided the contents contain error coding (to mitigate undelivered portions of the data), or the message could consist of a pointer to the data at an external source (for example a link to a file on a cloud storage service).
While short messages may not seem useful, in many cases it is enough in order to communicate key points. For example, in the case of a protest group organising a protest, the organiser simply needs to send a time and location. A whistleblower only needs to send the details of either a dropzone for documents or a location to arrange a meet.
1) Closed Networks: The system supports the ability for users of the system to create smaller, closed networks in order to provide communication solely within a social group. For certain users groups, being able to separate from general traffic and tweak system parameters to suit the situation could provide extra benefits. The use of many smaller, more closed networks has the benefit that there is less risk of congestion due to smaller user numbers, and better trust relationships in order to resist insider attacks. This is discussed more in section III.
## III. SYSTEM DESIGN
The major component of the design of Blindspot is the new probabilistic routing algorithm that is able to successfully route within the confines of the image sharing behaviour on an online social network. To the best of our knowledge, routing across nodes who are communicating at natural rates is an unresolved problem.
## A. Basics
1) Message-secrecy key pair: Each participating node generates a public encryption key and a private decryption key according to a public-key cryptosystem. A pair of communicating nodes (sender and receiver) exchange public encryption keys out-of-band. For example, receivers could publish public keys on an external site or distribute in person in a similar fashion to how PGP keys are distributed today.
2) Neighbourhood key pair: Each node also generates a second encryption key-pair, called the neighbourhood key-pair. The private encryption key is shared with all of the node's neighbours. This allows them to deliver intended messages directly to the receiver, and stop forwarding the message any further.
## B. Routing design
At the core of Blindspot is fully decentralised routing algorithm, based upon a pull-broadcast model. Messages are routed between a sender and receiver without revealing receiver identity to any intermediate node on the route, except the last 'exit' node.
Each node on the social network maintains an input message queue (new incoming messages) and an output message queue (outgoing messages).
TABLE I: Terminology
1) Node Knowledge: We make the assumption that every node on the network has a full knowledge of only those nodes for which it is directly connected, i.e. their 'friends'. This knowledge includes the image upload behaviour of those nodes, and their friend lists. This is the typical case for many social networks. For example, on Facebook (by default) a user can view all activity performed by a friend (posts are by default visible to friends and friends of friends), and can, usually, see the friends list of that friend. The exception to this is if the friend has set extra privacy setting on their account, but this is the small minority of users.
A node has no knowledge of the global network topology, only the 2-hop neighbourhood of which it is the centre node (the node, its friends and its friends' friends). Outside of this 2-hop neighbourhood, the node has zero knowledge of the graph topology or the location and behaviour of other nodes. Each node only has the ability to communicate with (share images with and pull images from) its direct neighbours.
2) Message construction: Each message is of the form K N ,K N r r M r M . To create a new message, the sender encrypts the message with the receiver's message-secrecy public key. This ensure message privacy as only the receiver can recover the actual contents of a message. Next, it generates a unique nonce N r . It then encrypts the the nonce to provide a function for nodes in the receiver's neighbourhood to discard duplicates, and for the receiver to recognise duplicate messages.
The receiver's address (this will vary according to the social network, but will typically be the username) is encrypted with the receiver's neighbourhood key under the Elgamalbased scheme described in section III-B7 using the receiver's neighbourhood public key.
The user agent running on the node now places a newly constructed message and encrypted destination address into the node's output message queue.
Each node on the path therefore only sees the encrypted destination and message. Each node, due to the use of reencryption, will see a different ciphertext for each component, providing unlinkability and resistance to tagging attacks. An important point is that different paths that a message takes result in a different series of ciphertexts. The nodes who possess the receiver's neighbourhood key will be able to identify the destination of messages for them, but will be unable to view the message contents.
3) Output queue processing: To process messages on the output queue, the user-agent waits for the user to upload an image. Note: the same image is not used for each upload. The image selection and uploading is carried out by a human, not the user-agent. The state of the message queue has no impact on the user's decision to post an image.
Once the user selects an image, the user-agent intercepts it and steganographically embeds messages into the image. The precise method of doing this in a reliable manner without interference from OSN compression algorithms is given in Section III-B8 - 'Channel design for unobservability'.
4) Input queue processing and routing strategy: Similarly, when the user reads an image post, as part of their normal browsing behaviour, the image is downloaded to the user's computing device. The user-agent then extracts embedded message within the image and enters them into the input queue. Blindspot then selects a subset of messages from the input queue which are moved to the output queue according to a routing strategy.
Traditionally, the path that is taken between two endpoints is usually taken to be the shortest between them in the number of hops or distance, which is the assumption that many traditional routing and path selection algorithms work to. In this situation, however, the shortest path in terms of hops may not necessarily be usable if not all of the nodes on the path are active (or frequent) uploaders. The state of the network in terms of which nodes are 'active' also changes often, meaning traditional shortest path algorithms are insufficient (again, these usually work on the assumption that nodes are static in behaviour).
We therefore use the inter-upload distributions for each node to aid in the routing decisions, such that the chosen paths represent the lowest waiting time (time between message upload and subsequent upload by neighbour) resulting in a relatively low delay path.
The upload delay distribution of each node is governed by a random variable. Subsequently, the distribution over delivery times, or the message-delay distribution is dependent on this. For any given route, the message-delay distribution is the sum of independent random variables (of nodes) on that route. For a two hop route (s, t, r) where s (sender) is communicating with r (receiver) via t (intermediary), the message delay distribution is given by D = S + T, where S and T are random variables whose PDFs are the inter-upload delay distributions of nodes s and t respectively.
The sum of two random variables is equivalent to the convolution of the respective PDFs, given by the convolution formula:
$$( f * g ) ( t ) = \, \int _ { - \infty } ^ { \infty } f ( \tau ) \, g ( t - \tau ) \, d \tau \, \quad \ \ ( 1 )$$
By applying this formula to every pair of node on the path, the result will be the distribution of the possible total delay of the message. In our system, however, it is not possible to compute this full convolution as part of the routing algorithm as the convolution output (which increases in size by the length of the additional distribution at each convolution) will reveal the number of hops the message has taken if it is passed along with the message (it will also use up a large chunk of capacity). An example convolution output can be found in Figure 1(b).
To solve this problem, we instead use the local knowledge of nodes in order to compute the convolution a small section of each path, without passing any values along with the message.
As we have said, each node knows the image upload distributions of its neighbours (plus itself). This is enough knowledge to compute a two hop section the path (the previous node, the current node, and the next node). For a node with degree d , this equates to d ∗ ( d -1) possible paths (from all neighbours to all neighbours excluding the source).
a) Path selection: A node cannot pull every message that is sees due to the capacity limitations. Therefore nodes need to make some decision on which messages to pull and forward. We use a probabilistic approach for deciding this.
During an upload session (the time period in which the user is making uploads, where 1 or more images will be uploaded) the node will fetch the messages contained in the last 5 uploads made by each neighbour (limited by a TTL) that they have not already processed. Note that uploads can be fetched over time during normal browsing behaviour and the messages contained cached. This means that messages can be processed in the background over time, including the more costly operations such as decryption and re-encryption. It also means that there is not a sudden spike in requests for neighbours images, which could be an indication of the systems use. The messages from all of these uploads is placed in a queue, which is then sorted according to a score function, given by eq. 2, where similarity prev is the percentage of neighbours shared by the current node and the neighbour the messages was received from (the lower this value the better as it allows the message to escape communities), and min cdf ( ) src is the minimum convolution score for all possible exit routes (all neighbours excluding the node from which the message was received) (see section (III-B5) below, 'Convolution Comparison') normalised between 0 and 100, which indicates the future upload prospects of the message, a lower score indicating the greater likelihood of delivery.
The score of message i is given by:
$$s c o r e _ { m _ { i } } = s i m i l a r i t y _ { p r e v } + m i n ( c d f ) _ { s r c } \quad ( 2 ) \stackrel { \mathfrak { m } \circ m } { \int ( \alpha _ { \lambda } \alpha _ { \lambda } } }$$
b) Message expiration: The messages are sorted in ascending order according to this score (eq. 2, lower scores are desirable). The node will then go down this list, and perform a weighted coin toss for each message to decide if it should be pulled - moved from the input queue to the output queue. Each message is selected with a probability 1 /deg src , or discarded with probability 1 -1 /deg src , meaning that the lower the degree of the proceeding node, the more likely the message is to be forwarded. Over time this provides an natural expiration effect, message sprays will thin as time passes.
5) Convolution Comparison: We take the convolutions of the previous node, the current node, and each possible next node (resulting in the deg currentnode -1 distributions). Each of these are then converted into a cumulative density function (CDF), and a score is computed for each. The score measures the steepness of the CDF at all points along it, such that a lower score represents a CDF that climbs (and reaches 1) early. The score is as follows:
$$\sum _ { i = 1 } ^ { l e n g t h ( c d f ) } ( 1 - P _ { i } ) \cdot X _ { i } \quad \quad ( 3 )$$
where P i is the CDF value at position i , X i is the X axis value at position i and length cd f ( ) is the number of values along the X axis. We take 1 -P i to reward CDFs that climb early, as a CDF that reaches 1 sooner will have more of its values multiplied by zero, resulting in a lower score.
A CDF with a lower score represents a PDF that is centred over a low delay, meaning that the path segment represented by it will likely have a low delay associated with it. Three example CDFs and their associated scores can be seen in figure 1.
The score will be larger for longer CDFs. If two CDFs of different lengths need to be compared, the shorter CDF must be extended with the addition of ones to the CDF values (or zeros to the underlying PDF values) to make them of equal length.
6) Sybil Protection: Blindspot does not propose a new Sybil protection mechanism, but enjoys the benefits of existing techniques proposed in this space [44], [12], [45], [29].
7) Tagging attack resistance and unlinkability: To ensure that the input and output messages from each node are bitwise unlinkable and hence resistant to tagging attacks, we make use of universal re-encryption. As proposed by Golle et al [19], universal re-encryption is a scheme that makes use of the ElGamal asymmetric key encryption algorithm to re-encrypt a message without the knowledge of recipient public keys. This is summarised as follows:
Encryption of m using key K : Km : C = [( α , β 0 0 ); ( α , β 1 1 )] = [( my k 0 , g k 0 ); ( y k 1 , g k 1 )] where k , k 0 1 ∈ U Z q and q is a global system parameter, which is 1024 bits in size.
Decryption of a given ciphertext: K -1 ( α , β 0 0 ); ( α , β 1 1 ) : m 0 = α /β 0 x 0 m 1 = α /β 1 x 1 . If m 1 = 1 then the plaintext is m = m 0 .
$$( 2 ) \begin{array} { c c } \stackrel { \cdots } { R e \text{-encryption} } \cdot \\ [ ( \alpha _ { 0 } \alpha _ { 1 } ^ { k _ { 0 } ^ { \prime } }, \beta _ { 0 } \beta _ { 1 } ^ { k _ { 0 } ^ { \prime } } ) ; ( \alpha _ { 1 } ^ { k _ { 1 } ^ { \prime } }, \beta _ { 1 } ^ { k _ { 1 } ^ { \prime } } ) ], \text{ where } k _ { 0 } ^ { \prime }, k _ { 1 } ^ { \prime } \in _ { U } \mathbb { Z } _ { q } \\ [ ( \alpha _ { 0 } \alpha _ { 1 } ^ { k _ { 0 } ^ { \prime } }, \beta _ { 0 } \beta _ { 1 } ^ { k _ { 0 } ^ { \prime } } ) ; ( \alpha _ { 1 } ^ { k _ { 1 } ^ { \prime } }, \beta _ { 1 } ^ { k _ { 1 } ^ { \prime } } ) ], \text{ where } k _ { 0 } ^ { \prime }, k _ { 1 } ^ { \prime } \in _ { U } \mathbb { Z } _ { q } \\ \end{array}$$
At each point along the path a message takes, the forwarding node performs re-encryption on both ciphertexts. Any node on the path can perform this step with no knowledge of the receivers public key or the path that the message has taken/will take. Note that the receiver address and message are re-encrypted individually.
The use of re-encryption provides sender-receiver unlinkability along all paths that the message takes, with no extra key
(a) Top score 398.1211, middle 94056.63 and bottom 226694.9
(b) CDF of convolution of PDFs in (a). Score is 618401.3
Fig. 1: Score output for the CDF of three inter-upload PDFs and convolution output
setup apart from the initial receiver's public key distribution. This is more appropriate in this setting than the traditional onion routing approach, whereas the public keys for nodes on the path need to be collected by the source. Even if the path was known to the source, the key set-up stage would take far too long if key negotiation occurs within the system.
8) Channel design for unobservability: There are two main requirements for the steganography algorithm used. First, and most importantly, the algorithm should be statistically undetectable [5], which means that no statistical test can distinguish between the set of cover objects and the set of stego objects. Secondly, in order to maximise the bandwidth within the system the algorithm also needs to ensure as much capacity is available as possible while maintaining data integrity in the face of compression.
The system only requires that a group of users (such as one social group) agree on one steganographic scheme to use, which provides the trade-off between capacity and security to meet their needs. For particularly sensitive information, using an scheme which provides good security guarantees but a lower capacity will be preferable. For more general users, an higher capacity but lower security scheme could be used.
compression performed by YASS, and q denotes the amount of redundancy provided by the algorithm. The higher the value of q the less data can be inserted into an image, but the more resilient the steganography to error. By default to provide unobservability YASS hides data at a rate of 0.05 bits per nonzero DCT coefficient. Nagaraja et al [31] demonstrated the ability of YASS to withstand uploading to and downloading from Facebook. They found that, assuming a 720px image and a q value of 2, 40280 bits of information could be hidden (when tested on 116 different images). At the other extreme, with a q value of 20, 4028 bits can be embedded. A sensible choice of q = 4 results in 20140 bits of embedding capacity. The YASS algorithm also achieves a low bit error rate (BER, the ration of error message bits to the total number of bits hidden). Using a quality factor of Q = 65 and redundancy parameter of q = 4 results in an average BER rate of 0.1320. The can be reduced to 0.0003 with parameters Q = 65 and q = 20 . These figures can be improved as some images are inherently 'bad' (result in a high BER), by not making use of these images performance can be improved.
An suitable example is the YASS (Yet Another Steganography Scheme) algorithm for JPEG steganography [37]. YASS is an example of minimal distortion embedding [17] [27]. Schemes that employ minimal distortion embedding focus on increasing the embedding efficiency by decreasing the embedding distortion. The YASS algorithm requires two parameters. Q represents the quality factor and dictates the amount of
Fridrich et al [18] showed that both the MM3 [27] and PQt [17] algorithms are also statistically undetectable at an embedding rate of 0.05 bits per non-zero DCT coefficient.
Facebook currently supports images of a maximum size of 2048px (2048*2048) [46]. Flickr also supports images of this size. This increase in image size is of a great benefit to image steganography as the larger an image is the more data can be hidden within it, we have estimated that this increase in size can provide up to 8 times more capacity in an uploaded image than as tested by Nagaraja et al. This gives a effective capacity of 8-10kb (assuming YASS with q = 4 .
a) Steganographic Keys: Each smaller network within the Blind system will be responsible for deciding upon which steganography algorithm is used. All members of the group can be expected to use the same algorithm. Each user will also have to generate an key which the will use when embedding the data. This will need to be given to neighbours, and can be shared along with the neighbourhood keys as described previously. Allowing each user to choose their own key will provide an extra level of security and prevent an insider leaking the key to extract the embedded data across all images. If a key is leaked, it will only affect the images of a single user, and can be replaced.
## C. Global and local networks
As we have mentioned, as the system is fully decentralised it supports the creation of individual instances of the system. Within each instance, users will have to set up keys and decide upon a steganography algorithm. The main benefit of allowing these smaller networks is that it strengthens the trust model within the system. If the system is closed to users within a single social group, it makes it far more difficult for an unknown user to participate (as in a Sybil attack) and fewer users reduces the likelihood of user compromise due to fewer targets.
## IV. RESULTS
To test the effectiveness of the routing algorithm we have produced a simple simulation in the Java language. Two main experiments are performed over two different networks. First, we test the performance of the routing algorithm under increasing congestion. We then test the effectiveness of the algorithm to maintain performance in the face of an active attack, namely the removal of increasing percentages of random and then high degree nodes (an effective black-holing attack).
The simulator takes as input a graph and the individual upload behaviour of each node. We then choose pairs of nodes to communicate, such that a source sends to a destination. The simulator works in a scale of days, with monthly image upload counts for each node split evenly amongst 30 days, for a total of 64 months. Each pair of nodes exchange 1 message per month. To keep the simulator lightweight, we do not apply encryption. After initial experimentation, we set the message TTL to 15 days. As per our channel design, assuming uploaded images are 4096px in size, we allow for 150 messages per image upload. We choose an image size of 4096px as it is (unofficially) accepted by Facebook for image uploads, and most modern smartphones feature a camera capable of taking images of this quality (or greater).
## A. Test Data
The primary dataset that we use is the Flickr dataset from Nagaraja et al [31]. This dataset contains the social graph and monthly image upload behaviour of 7200 users of the Flickr social network. We use this dataset as it is increasingly difficult to apply crawling to social networks such as Facebook that include strict privacy settings and Twitter which has restricted the sharing of tweet data. The difficulty with sites such as Facebook is that, increasingly, profiles are set to private for everyone other than the friends of that person. While this would not impact on the systems ability to function, it prevents automated crawling of the social network. To calculate the inter-upload time for each node, we divide each monthly value by 720, which gives the estimated inter-upload time in hours.
To test the performance of the algorithm on a differently structured network, we also use a network generated using the Barabasi-Albert (BA) model [3], with a parameter of 5. The Barabasi-Albert model produces a scale free network. We generate a network of 7200 nodes and apply the Flickr upload counts to this in a 1:1 mapping. We apply the 1:1 mapping as the resulting correlation between degree and upload count is very similar to that of Flickr.
## B. Effects of Congestion
In a capacity restricted network it is important to measure the effects of congestion. To measure the effects of congestion, we choose increasing number of pairs of nodes to communicates with 1 message per month per pair. The results of this experiment are presented in Figure 2(a).
As can be seen in the figure, even under a high level of congestion, the algorithm achieves a delivery rate of over 85% in both of the test networks. Both networks exhibit a similar degree of degradation under the increasing load. Interestingly, in the Flickr dataset 10 pairs of nodes performed worst that 50 pairs. We believe that this is down to one of these 10 pairs of nodes not having a stable path between them.
In both networks, the average delay for received messages is 1 day. This remains constant as the load increases. What does decrease, however, is the number of duplicate copies of a message that the destination receives. In Flickr for 100 pairs each destination receives on average 27 copies of each message. For 1000 pairs, this reduces to 10. In the BA model network, the number of duplicates reduces from 7.1 to 2.7.
## C. Effects of Black-holing
A possible active attack against this system is for nodes in the system to be removed (either by the social network or insiders). The node does not have to be physically removed from the system, rather they can simply stop participating in the routing protocol. It will take some time for the delay distributions held by neighbours for that node to reflect this change in behaviour, meaning the pull decision, which may include the convolution of the removed node, could be made assuming a neighbour is active when it is not. This will have greatest affect when the node is a key link for many nodes. The goal of this attack is to perform a denial-of-service by breaking the paths between nodes.
We test the impact of this attack by removing an increasing percentage of nodes from the network half way through the routing simulation. By removing the nodes from the graph, this has the same effect as the nodes no longer participating
(a) Effects of congestion
(b) Effect of black-holing both random and high degree nodes in the two test networks
Fig. 2: Performance under congestion and attack
(due to the pull model). We use two strategies for removing nodes, choosing nodes at random, and choosing high degree nodes. We use high degree nodes as even though their is little correlation between upload behaviour and the degree of nodes, high degree nodes are still important for routing as they statistically will view more messages than others. High degree nodes are also likely to be the main connection to the rest of the network for some nodes, meaning their removal may prevent other nodes from communicating.
network for providing a connection point to many nodes. This effect is visible in the sparser Barabasi network; the delivery rate drops at a higher rate before levelling out due to more nodes being disconnected from the network completely.
We use 100 pairs of nodes each communicating 1 message per month. We do not remove nodes that are in one of these pairs. The results of this experiment are shown in Figure 2(b).
As can be seen in the figure, in the face of the random removal of nodes, both networks are very resilient to even 50% of the nodes being removed from the system, losing only 23% of the delivery rate. The exception is for 40% of the Flick network being removed, this has a slightly larger effect. We believe that this is due to the randomness - the 40% of nodes chosen in that test contained a larger percentage of nodes that were key to routing between the chosen pairs (the pairs were kept constant across all test but the nodes to remove were randomised for every test).
Removing high degree nodes proved to be the most effective approach at causing a reduction in network performance, with both networks showing a drop down to 52% and 56% delivery rate when the 50% highest degree nodes were removed. This is expected, as we have mentioned the high degree nodes, while not necessarily being high uploaders, are still important in the
In both networks, and for both tests, the delay remains unchanged at ˜1 day per message (that was delivered). The main noticeable effect, other than the reduction in delivery rate, is the reduction in the number of duplicate messages received by the destination, as much as 15%. As we have already shown there is enough duplication, however, that we can tolerate the random node removal, and the messages that do arrive with the high degree nodes removed still arrive quickly.
## D. Path Consistency
It is important that the paths taken by messages between pairs of nodes maintain their quality over time. We measure the quality of a path using the CDF score as defined in equation (3), applied on the convolution of all nodes' distributions on the path. We compute the score for the first path that results in each message being delivered during the simulation (maximum 64 per pair of nodes). Due to the differing lengths of paths, we extend each path CDF such that all CDFs are of an equal length (by adding 1s). We then compute the Shannon entropy ( H X ( ) = -∑ i P x ( i ) log b P x ( i ) ) of these scores for each pair of nodes, and produce a distribution of these values. Figure 3 provides the score distributions of the pairs with the lowest and highest entropy, and the distribution of entropies across all pairs.
(a) Minimal entropy pair
(b) Maximal entropy pair
Fig. 3: Consistency of path quality
Figure 3(c) shows that a majority of the pairs have an entropy of scores of approximately 0.3. This is a relatively low entropy value, meaning that, over time, the quality of the path does not exhibit much variation. This is indicative of the same paths being used repeatedly if they are available (all nodes uploading).
## V. ANALYSIS
The primary goal of our system is to achieve indistinguishability. We also set out to provide a system that is resilient against certain insider attacks.
applying heavy compression on an image (that even compression resistant steganography algorithms cannot survive) or by applying a transformation to the image in order to invalidate any hidden data. Facebook, for example, already applies heavy compression to all uploaded images. Any further compression will have a visible impact on image quality, as would any image transformations. Obviously, a country can completely block access to the social network itself. This is similar to the issue of a censor blocking access to mixes and relays in anonymous communication networks.
## A. Attacks on Communication Channel
The system as described does not create any extra traffic, or introduce any new communication endpoints - this is the basis for indistinguishability in our system. All communication is achieved using the user behaviour that already 'exists' (the user would still upload images without the hidden channel). Traditional traffic classification methods would not be able to detect unusual behaviour. In addition, the use of good-listing or bad-listing would be ineffective. In countries where internet censorship does not occur, access to social networks is open. In places where social network access is not permitted, there is often a local, closed social network that is in use. The system can easily be transferred amongst social networks.
1) Attacks on Routing: We have shown that, under targeted attacks where insiders are able to remove the top 50% highest degree nodes, the delivery rate can still maintain a level above 50%, with no noticeable impact on latency. Under random attacks -nodes are attacked at random - the effect is a reduction in delivery rate of just a few percent. This is due to the large amount of disjoint paths taken by messages, unless every path is disrupted, delivery can still occur.
2) Denial of Service Attacks: Following our threat model, we assume that the OSN would be able to perform Denial-ofService attacks on the system by destroying the steganographically hidden information. This could be done, for example, by
Unfortunately there is very little that can be done to avoid DOS attacks. The redundancy provided by the routing algorithm helps to prevent local DOS attacks, or in the situation where the adversary only tampers with a subset of images at random. Past this, it is important to make use of a steganography algorithm that provides compression resistance which will prevent the attacker from simply re-compressing images.
3) Attacks on Steganography: The major risk to the unobservability provided by the system is that of steganalysis.
For a person that is already under targeted surveillance, steganalysis could be applied to all images uploaded by that person and their neighbours in an attempt to identify if they are taking part.
The fact that all image payloads are encrypted will hinder steganalysis attempts. Assuming that steganalysis can be applied, the attacker will also have to reliably identify ciphertexts within the image and differentiate them from random data. This will prove difficult if they do not know the ciphertexts that are being sought. Further to this, using keyed steganography algorithms with individual keys will also hinder steganalysis attempts.
As mentioned in section III-B8, there are steganography algorithms that are considered statistically undetectable when used with small payloads. A common mistake when using steganography is the use of large payloads which (a) makes traffic analysis on the images possible and (b) renders the channel vulnerable to compression-based attacks. We have
(c) Distribution of pair entropies
avoided both of these issues in our design.
If a user of the social network is found to be using the system, unlinkability and message confidentiality will be maintained, meaning the user will be reduced to the security level provided by previous anonymity systems such as Tor.
## B. Resistance to Insider Attacks
Our design ensures sender-receiver unlinkability [32] in multiple cases. First, a single compromised node does little damage: there is no sender address included with message blocks, so an attacker node has no knowledge of the source. As the destination address is encrypted (and subsequently universally re-encrypted) with the neighbourhood key of the destination only nodes with the private neighbourhood key of the destination know where it is going. Second, the use of reencryption at each node along the route offers traffic analysis resistance against passive colluding attackers by ensuring bitwise unlinkability between incoming and outgoing message blocks. Finally, resistance against active collusion including replay attacks and tagging attacks is ensured (during reencryption) by raising input ciphertexts to a random number from Zq . Due to this operation, replayed message blocks will not result in identical output message blocks. Similarly, any changes (tagging) to an input block will be blinded during re-encryption, hence tagged inputs will not be recognisable at the output. The inherent protection provided by making use of a social network against Sybil attacks also increases the effort required by an attacker to inject malicious nodes in the first place.
## C. Design Limitations
1) Ciphertext lengths: There is a major downside to using universal re-encryption, in that for every k bits of data, 4 k bits of ciphertext are required to be transmitted. However, this is justified given the high risk of insider attacks. Despite these overheads, Blindspot remains a usable system for lowbandwidth anonymous communications where indistinguishability is required to avoid censorship.
2) Insider key leakage: The requirement of neighbours being able to identify messages destined to neighbours opens the system to a form of insider attack. If a neighbour of a node is compromised, they can leak the decryption key which will enable any node also under the attacker control to identify messages destined for that destination. The message contents, however, are not at risk (as only the recipient has the decryption key).
Making use of key-based steganography goes some way in alleviating this problem. If the steganography is open (meaning no key or a global key is used) the global adversary would be able to capture all images, extract all of the embedded data and decrypt destination addresses. If individual keys are used for steganography which are only shared with neighbours, only other insiders will be able to identify messages as the messages will only be hidden using that key at the node for which they are uploaded. At other points in the path a different steganographic key would be used meaning that the embedded messages are not retrievable.
3) Out-of-band key distribution: Currently, the system requires a large amount of out-of-band key distribution. As the system is fully decentralised, low bandwidth and high latency standard key distribution protocols are not optimal. This problem could in part be solved through the use of identity-based cryptography, although this introduces the need for a key escrow which introduces a communication endpoint.
## VI. BACKGROUND AND RELATED WORK
## A. Unobservable Communications
There have been many attempts at providing unobservable communication on the internet. One common approach is to use steganography in tandem with web-based storage systems to provide methods of communication. Unfortunately, none of these techniques provide anonymity and indistinguishability and hence are of little use in our threat context.
CovertFS [1] is a prototype system for providing a covert file store that stores data (and file structure) in images stored on photo sharing sites. The data can be shared amongst users of the system, with a goal of achieving plausible deniability i.e. a user can deny that they are using the system (which is reliant on the resistance of the steganography algorithm used to steganalysis).
Collage [4] is another system that follows a similar approach. In Collage, data is hidden within media which is uploaded to a user-generated content host. The receiving party then accesses the media by performing some task (such as accessing a URL or searching for specific terms), downloads it and extracts the hidden data. The key point is that the content host is interchangeable - i.e. if one service is blocked another can be used with minimal effort. This is all in full view of the censor, who sees only the transfer of the media file from the uploader to the media server (a legal operation in most states), and the download/access of the file by the receiver (again, a legal, innocent operation in most cases). The censor is assumed to have the ability to collect full traffic information on a target, and disrupt communications. An extension to the system is proposed by Invernizzi et al. [25], in which data is hidden within images on blog posts. The system is different however, as it allows for any blog to be used on any provider, meaning that there are no single entry points (e.g. Flickr or Facebook) to block. The rendezvous point is found using a search engine. The deniability arises from the fact that every minute hundreds of thousands of images are uploaded in blog posts, and so it is infeasible for the censor to either run steganalysis on every image, or block every blog provider.
Collage does not work under our threat model. Under targeted surveillance, the attacker can easily isolate Collage communications and break the user's anonymity. This down to the fact that collage introduces new endpoints into communication (the source and receiver(s) connecting to the intermediary), which will appear in traffic logs (including GCHQ collected Tempora data) and may differ from normal behaviour.
The above systems are high-latency systems designed for one-off transmissions of data. There are also systems that use steganography for lower-latency communications. For example, Huang et al. [23] propose a system for transmitting data by hiding information within VoIP packets. Craig Rowland [35] presents a number of methods for embedding data within TCP/IP packet headers by modifying particular fields (those that are optional, unused and are unlikely to be changed in transit). Ahsan and Kundar [2] also hide data within the headers of TCP/IP packets, but also propose a second method of data hiding by changing the ordering of packets. This can achieve a bandwidth log n ( !) bits (where n is the number of packets, for example 25 packets provides around 87 bits of capacity). This method could be discovered, however, if all packets consistently arrive (or are sent) out of order using simple traffic analysis. Modified TCP/IP headers are also trivial to detect if the header fields contain non standard (default) values.
An alternative to steganography is protocol mimicking, in which attempts are made to make one traffic protocol appear as another. Due to its high detectability rate but high usage rates, Tor is often the subject of work to try and make it less detectable. For example, StegoTorus [42] and SkypeMorph [30] both attempt to hide Tor traffic by making it appear as VoIP (in particular Skype) traffic. The systems attempt to produce protocols that can carry Tor traffic, but exhibit the characteristics of the Skype protocol, including timing and packet headers. CensorSpoofer [40] is designed for obfuscated web browsing by decoupling the upstream and downstream channels of a HTTP session. HTTP requests are sent to the server over a low capacity channel such as email or instant messaging, and the server responds to the client mimicking UDP-based VoIP traffic, by emulating the traffic from a dummy P2P host, more specifically SIP-based VoIP. All three of these systems have been proven broken [21], specifically because they are unable to fully emulate the protocols they mimic, meaning that their usage can be detected by looking for unusual behaviour (for example incorrect error handling or malformed headers).
Houmansadr et al. [22] propose Freewave, a system for converting IP traffic into V oIP traffic for circumventing internet censorship. IP packets are converted into acoustic signals, which are then transmitted to a Freewave server, which converts the VoIP stream back into IP packets and releases them into the uncensored internet, acting as a proxy. The system uses encryption to make sure that the Freewave VoIP packets are indistinguishable from from background packets. Due to the use of Skype supernodes in routing, the source is unlinkable from the server, so even if the censor knows the servers identity, they will not automatically know the identities of end users. The VoIP operator, however, may know who communicates with the Freewave server, so it is assumed that a provider who does not collude with the censor is used (a weak assumption). The authors state that the system is resilient to traffic analysis as the calls follow the same traffic patterns as a normal VoIP call, but there is one small issue in this assumption. If Freewave is used for web-browsing rather than normal speaking, then the frequency and timing of calls may be an indicator of the systems use (if it different from normal usage patterns of users). The call frequency and use pattern for a user who is actually using the service for web browsing rather than calling could be substantially different to that of the user before they started browsing, or before the started using the service for browsing.
## B. Decentralised Routing over Social Networks
Fully distributed anonymous routing precludes the use of design options such as onion routing or its semi-centralised peer-to-peer variants. This is due to their dependency on the availability of a central directory service. Even with a fully distributed directory service, we cannot make conventional onion routing work due to the capacity constraints of each node; deploying such a directory service would almost certainly exhaust the meagre network bandwidth available leaving little opportunity for efficient anonymous communication.
Significant prior art exists on decentralised routing protocols in the context of peer-to-peer adhoc networks.
Gonzalez et al. [20] proposed routing algorithms that leveraged human mobility patterns. Their algorithm was design to handle edges which were available according to an exponential distribution. These findings support those made earlier by Clauset and Eagle [7] on a smaller dataset of 66 individuals. Chaintreau [6] observed that in graph topologies based on human mobility paths of O log n ( ) hops exist. These findings are strongly reminiscent of the characteristics of 'small-world' networks which are thought to closely resemble human social networks [41].
We can also gain some understanding of these networks by several attempts to construct routing protocols. One of the early proposals [39] was epidemic routing which is essentially flooding. Other studies [43] tested the effects of flooding on popular datasets such as the MIT Reality mining dataset . This 1 dataset consists of the mobility patterns of 100 individuals over a nine-month period. The authors found (see Fig. 2 of [43]) that (owing to the periodicity and frequency of human mobility) the message delivery rates were time-invariant; when the time-to-live is between three to five days, 50-60% of the messages were delivered on the average when all pairs of nodes simultaneously communicate.
While epidemic routing(flooding) makes few assumptions about graph structure, it causes congestion for well understood reasons which ultimately results in lower message delivery rates since the network is flooded with duplicates. For these reasons, a number of schemes that use resources more efficiently have been proposed. Spyropoulos et al. [38] produced a routing algorithm in which nodes forward limited copies of the data to nodes that are encountered within the network. If a node is directly connected to the destination, it sends to the packet. This method relies heavily on the mobility of nodes to enable the data packets to reach their destination. This approach requires the use of a decreasing copy count. The copy count would leak a large amount of information about the
source of the message (a node on the path knows how far the source is from them in hops). The lack of mobility also vastly reduces the effectiveness of the algorithm in a static network. Daly and Haar [8] proposed using a centralised path discovery technique to find viable shortest paths. This involves using an O n logn ( 2 ) algorithm to find and route through the most betweenness-central nodes in a graph, using Freeman's [16] definition of betweenness centrality. The authors showed that their scheme performs almost as well as the basic flooding algorithm. Hui et al. [24] found that betweenness centrality and degree centrality are highly correlated and therefore using a local centrality metric to drive route selection can also result in performance (delivery ratios) close to that achieved by flooding. Both of these systems are unsuitable in this system, however, as they both require knowledge of both the destination and the betweenness of nodes with relation to it. This information would break unlinkability.
Mittal et al. [29] propose X-Vine, a system for providing defence against Sybil attacks on distributed-hash table based systems. The proposed design applies a social network overlay to the distributed hash table (based upon Chord). The Chord routing algorithm is used as normal at a high, but rather than route directly between Chord points, data is routed through a subsection of the social network. The paper has a goal of low control data overhead, but it is still far above an acceptable level for our setting. While this level is suitable for networks without the stringent bandwidth limitations of ours, it would use up too much valuable capacity in the lowbandwidth environment.
Blindspot has some similarities to Crowds [34]. Within Crowds, traffic is randomly routed between members of a crowd until one member decides to forward the message to the server. Each member on a path does not know if the preceding member was the originator of the traffic. The routing provided by Blindspot also provides this property; a node is unable to tell if a message it receives or forwards originated from the node it pulled the message from. The major difference, however, is that in Crowds the goal is to communicate with a server that any node can establish a direct connection to (enabling the traffic to reach the destination), whereas in Blindspot a path needs to be found between the source and destination on the underlying social graph.
## VII. CONCLUSION
## A. Future Work
One of the major pieces of future work will be to produce a prototype of Blindspot and to perform a limited beta (for example limited to the author's department). The software will be of the form of a browser plugin that will intercept user's uploads and perform the system operation automatically with minimal user input. There is also scope for creating a version of the software for smartphones due to their increasing use as cameras that upload images directly to social networks. The main goal of the prototype will be to perform a usability study of the system in operation.
One area that needs to be explored in the future is the issue of whether or not users who are participating in the Blindspot (or similar) system may change their uploading behaviour in order to support the system. While this is good for the system's operation, the change in behaviour may be an indication of the systems use. Exploring whether or not users will change their behaviour noticeably while using systems such as Blindspot will be important for any system that aims to achieve unobservability. Finally, currently the system design focuses on the system being present on a single social network. In reality, users are often members of many social networks. The users may behave differently on different networks (a person could well upload more images on their Flickr account that on their Facebook account), but have overlapping social graphs. To this end, the system design could be adapted in such a way that the users entire social graph is used (built from all of their social networks). This could also allow the system to make use of cover texts other than images (including text and video) in order to increase the network capacity.
## B. Conclusion
Evident use of anonymous communication networks can present real-world risks to the lives of users in some countries. It can also allow a censor to wholly block anonymous communications. In this paper, we have proposed Blindspot which is specifically designed to address the needs of protest organisers and others. Its goals are to provide indistinguishability and unobservability against the network infrastructure provider. We have shown that leveraging social behaviour of users is an attractive approach to achieve these goals. In conclusion, our main contribution is a novel decentralised routing protocol that provides high levels of performance, given the constraints, while providing anonymity against a global passive adversary. It leverages trust relationships on a social network to contain node misbehaviour and offers some protection against insider attacks such as blackholing.
## REFERENCES
- [1] J. K. A. Baliga and L. Iftode. A web based covert file system. In Proceedings of the 11th Workshop on Hot Topics in Operating Systems (HotOS XI) , 2007.
- [2] K. Ahsan and D. Kundur. Practical data hiding in tcp/ip. In Proc. Workshop on Multimedia Security at ACM Multimedia '02 , Juan Les Pins, France, 2002.
- [3] R. Albert and A.-L. Barab´ asi. Statistical mechanics of complex networks. Reviews of Modern Physics , 74:47-97, Jan. 2002.
- [4] S. Burnett, N. Feamster, and S. Vempala. Chipping away at censorship firewalls with user-generated content. In Proceedings of the 19th USENIX Conference on Security , USENIX Security'10, pages 29-29, Berkeley, CA, USA, 2010. USENIX Association.
- [5] C. Cachin. An information-theoretic model for steganography. In Information Hiding , volume 1525 of Lecture Notes in Computer Science , pages 306-318. Springer Berlin Heidelberg, 1998.
- [6] A. Chaintreau, A. Mtibaa, L. Massoulie, and C. Diot. The diameter of opportunistic mobile networks. In Proceedings of the 2007 ACM CoNEXT conference , CoNEXT '07, pages 12:1-12:12, New York, NY, USA, 2007. ACM.
- [7] A. Clauset and N. Eagle. Persistence and periodicity in a dynamic proximity network. In Proceedings of the DIMACS Workshop on Computational Methods , DIMACS 2007, 2007.
- [8] E. M. Daly and M. Haahr. Social network analysis for routing in disconnected delay-tolerant manets. In E. Kranakis, E. M. Belding, and E. Modiano, editors, Proceedings of the 8th ACM Interational Symposium on Mobile Ad Hoc Networking and Computing , pages 3240. ACM, 2007.
- [9] G. Danezis. Statistical disclosure attacks: Traffic confirmation in open environments. In Gritzalis, Vimercati, Samarati, and Katsikas, editors, Proceedings of Security and Privacy in the Age of Uncertainty, (SEC2003) , pages 421-426, Athens, May 2003. IFIP TC11, Kluwer.
- [10] G. Danezis, C. Diaz, C. Troncoso, and B. Laurie. Drac: an architecture for anonymous low-volume communications. In Proceedings of the 10th international conference on Privacy enhancing technologies , PETS'10, pages 202-219, Berlin, Heidelberg, 2010. Springer-Verlag.
- [11] G. Danezis, R. Dingledine, and N. Mathewson. Mixminion: Design of a type iii anonymous remailer protocol. In IEEE Symposium on Security and Privacy , pages 2-15, 2003.
- [12] G. Danezis and P. Mittal. Sybilinfer: Detecting Sybil nodes using social networks. In NDSS , 2009.
- [13] G. Danezis and B. Wittneben. The economics of mass surveillance and the questionable value of anonymous communications. In WEIS , 2006.
- [14] R. Dingledine and N. Mathewson. Anonymity loves company: Usability and the network effect. In Proceedings of the Fifth Workshop on the Economics of Information Security (WEIS 2006) , Cambridge, UK, June 2006.
- [15] R. Dingledine, N. Mathewson, and P. Syverson. Tor: The secondgeneration onion router. In Proceedings of the 13th USENIX Security Symposium , Aug. 2004.
- [16] L. C. Freeman. Centrality in social networks: Conceptual clarification. Social Networks , 1:215-239, 1978.
- [17] J. Fridrich, M. Goljan, and D. Soukal. Perturbed quantization steganography with wet paper codes. pages 4-15, 2004.
- [18] J. Fridrich, T. Pevn´ y, and J. Kodovsk´ y. Statistically undetectable jpeg steganography: Dead ends challenges, and opportunities. In Proceedings of the 9th Workshop on Multimedia & Security , MM&#38;Sec '07, pages 3-14, New York, NY, USA, 2007. ACM.
- [19] P. Golle, M. Jakobsson, A. Juels, and P. Syverson. Universal reencryption for mixnets. In IN PROCEEDINGS OF THE 2004 RSA CONFERENCE, CRYPTOGRAPHERS TRACK , pages 163-178. SpringerVerlag, 2002.
- [20] M. C. Gonz´ alez, C. A. H. R., and A.-L. Barab´ asi. Understanding individual human mobility patterns. CoRR , abs/0806.1256, 2008.
- [21] A. Houmansadr, C. Brubaker, and V. Shmatikov. The parrot is dead: Observing unobservable network communications. In Proceedings of the 2013 IEEE Symposium on Security and Privacy , 2013.
- [22] A. Houmansadr, T. Riedl, N. Borisov, and A. Singer. I Want my Voice to be Heard: IP over Voice-over-IP for Unobservable Censorship Circumvention. In The 20 th Annual Network and Distributed System Security Symposium (NDSS) , 2013.
- [23] Y. Huang, B. Xiao, and H. Xiao. Implementation of covert communication based on steganography. In Intelligent Information Hiding and Multimedia Signal Processing, 2008. IIHMSP '08 International Conference on , pages 1512-1515, 2008.
- [24] P. Hui, J. Crowcroft, and E. Yoneki. Bubble rap: social-based forwarding in delay tolerant networks. In Proceedings of the 9th ACM Interational Symposium on Mobile Ad Hoc Networking and Computing , pages 241250, 2008.
- [25] L. Invernizzi, C. Kruegel, and G. Vigna. Message in a bottle: Sailing past censorship. In Proceedings of the 29th Annual Computer Security Applications Conference , ACSAC '13, 2013.
- [26] JAP. Jap anon proxy. http://anon.inf.tu-dresden.de/publications/index en.html.
- [27] Y. Kim, Z. Duric, and D. Richards. Modified matrix encoding technique for minimal distortion steganography. In J. Camenisch, C. Collberg, N. Johnson, and P. Sallee, editors, Information Hiding , volume 4437 of Lecture Notes in Computer Science , pages 314-327. Springer Berlin Heidelberg, 2007.
- [28] G. Mail. https://www.guerrillamail.com/.
- [29] P. Mittal, M. Caesar, and N. Borisov. X-Vine: Secure and pseudonymous routing in DHTs using social networks. In Proceedings of the 19th Annual Network & Distributed System Security Symposium , Feb. 2012.
- [30] H. Mohajeri Moghaddam, B. Li, M. Derakhshani, and I. Goldberg. Skypemorph: protocol obfuscation for tor bridges. In Proceedings of the 2012 ACM conference on Computer and communications security , 2012.
- [31] S. Nagaraja, A. Houmansadr, P. Piyawongwisal, V. Singh, P. Agarwal, and N. Borisov. Stegobot: A covert social network botnet. In Information Hiding , pages 299-313. Springer, 2011.
- [32] A. Pfitzmann and M. Hansen. Anonymity, unobservability, and pseudonymity: A consolidated proposal for terminology. Draft, July 2000.
- [33] M. G. Reed, P. F. Syverson, and D. M. Goldschlag. Anonymous connections and onion routing. IEEE Journal on Selected Areas in Communications , 16(4), 1998.
- [34] M. K. Reiter and A. D. Rubin. Crowds: Anonymity for web transactions. ACM Trans. Inf. Syst. Secur. , 1(1):66-92, Nov. 1998.
- [35] C. H. Rowland. Covert channels in the tcp/ip protocol suite. First Monday , 2(5), 1997.
- [36] R. Sandvick. Harvard student receives f for tor failure while sending 'anonymous' bomb threat, December 2013. http://www.forbes.com/sites/runasandvik/2013/12/18/
- harvard-student-receives-f-for-tor-failure-while-sending-anonymous-bomb-threat/.
- [37] K. Solanki, A. Sarkar, and B. S. Manjunath. Yass: yet another steganographic scheme that resists blind steganalysis. In in 9th Int. Workshop on Info. Hiding , 2007.
- [38] S. T., P. K, and R. C. Spray and wait: An efficient routing scheme for intermittently connected mobile networks. In SIGCOMM '05 . ACM, 2005.
- [39] A. Vahdat and D. Becker. Epidemic routing for partially connected ad hoc networks. Technical report, Duke University, 2000.
- [40] Q. Wang, X. Gong, G. T. Nguyen, A. Houmansadr, and N. Borisov. Censorspoofer: asymmetric communication using ip spoofing for censorshipresistant web browsing. In Proceedings of the 2012 ACM conference on Computer and communications security , 2012.
- [41] D. J. Watts and S. H. Strogatz. Collective dynamics of 'small-world' networks. Nature , 393(6684):440-442, June 1998.
- [42] Z. Weinberg, J. Wang, V. Yegneswaran, L. Briesemeister, S. Cheung, F. Wang, and D. Boneh. Stegotorus: a camouflage proxy for the tor anonymity system. In Proceedings of the 2012 ACM conference on Computer and communications security , 2012.
- [43] G. Williamson, D. Cellai, S. Dobson, and P. Nixon. Self-management of routing on human proximity networks. In Proceedings of the 4th IFIP TC 6 International Workshop on Self-Organizing Systems , IWSOS '09, pages 1-12, Berlin, Heidelberg, 2009. Springer-Verlag.
- [44] H. Yu, P. B. Gibbons, M. Kaminsky, and F. Xiao. Sybillimit: A nearoptimal social network defense against sybil attacks. In Proceedings of the 2008 IEEE Symposium on Security and Privacy , SP '08, pages 3-17, Washington, DC, USA, 2008. IEEE Computer Society.
- [45] H. Yu, M. Kaminsky, P. Gibbons, and A. Flaxman. Sybilguard: Defending against sybil attacks via social networks. In SIGCOMM , 2006.
- [46] M. Zhang. Facebook increases maximum photo size to 2048px, adds lightbox. Web, 9 2010. http://www.petapixel.com/2010/09/30/facebookincreases-maximum-photo-size-to-2048px-adds-lightbox/.
| null |
[
"Joseph Gardiner",
"Shishir Nagaraja"
] |
2014-08-04T19:35:15+00:00
|
2014-08-05T22:26:32+00:00
|
[
"cs.CR"
] |
Blindspot: Indistinguishable Anonymous Communications
|
Communication anonymity is a key requirement for individuals under targeted
surveillance. Practical anonymous communications also require
indistinguishability - an adversary should be unable to distinguish between
anonymised and non-anonymised traffic for a given user. We propose Blindspot, a
design for high-latency anonymous communications that offers
indistinguishability and unobservability under a (qualified) global active
adversary. Blindspot creates anonymous routes between sender-receiver pairs by
subliminally encoding messages within the pre-existing communication behaviour
of users within a social network. Specifically, the organic image sharing
behaviour of users. Thus channel bandwidth depends on the intensity of image
sharing behaviour of users along a route. A major challenge we successfully
overcome is that routing must be accomplished in the face of significant
restrictions - channel bandwidth is stochastic. We show that conventional
social network routing strategies do not work. To solve this problem, we
propose a novel routing algorithm. We evaluate Blindspot using a real-world
dataset. We find that it delivers reasonable results for applications requiring
low-volume unobservable communication.
|
1408.0786v2
|
## Effect of Accelerated Global Expansion on Bending of Light
Mir Emad Aghili, 1, ∗ Brett Bolen, 2, † and Luca Bombelli 1, ‡
1 Department of Physics and Astronomy, The University of Mississippi, University, MS 38677
2 Department of Physics, Grand Valley State University, Allendale, MI 49504
(Dated: May 25, 2016)
## Abstract
In 2007 Rindler and Ishak showed that, contrary to previous claims, the value of the cosmological constant does have an effect on light deflection by a gravitating object in an expanding universe. In their work they considered a Schwarzschild-de Sitter (SdS) spacetime, which has a constant asymptotic expansion rate H 0 . A model with a time-dependent H t ( ) was studied by Kantowski et al., who consider in their 2010 paper a 'Swiss-cheese' model of a Friedmann-Lemaˆ ıtre-RobertsonWalker (FLRW) spacetime with an embedded SdS bubble. In this paper, we generalize the Rindler and Ishak model to time-varying H t ( ) in another way, by considering light bending in a McVittie metric representing a gravitating object in a FLRW cosmological background. We carry out numerical simulations of the propagation of null geodesics in different McVittie spacetimes, in which we keep the values of the distances from the observer to the lensing object and to the source fixed, and vary the form of H t ( ).
PACS numbers: 98.80.Jk, 95.30.Sf
∗ Electronic address: [email protected]
† Electronic address: [email protected]
‡ Electronic address: [email protected]
## I. INTRODUCTION
The effect of the cosmological expansion on the bending of light rays and gravitational lensing in situations where the expansion is driven by a cosmological constant has been studied for more than 30 years, since Islam's 1983 paper [1] on light trajectories in Schwarzschildde Sitter (SdS) spacetime, but the issue has received an increased amount of attention in the past 15 or so years with observations leading to the conclusion that the global rate of expansion is accelerating. Among the recent references on the subject, we refer the reader to the papers by Ishak and Rindler [2, 3], and to the comprehensive discussion by Lebedev and Lake [4], which also contain citations to the rest of the literature. In this paper we will examine light bending in a more general setting, but still representing a nonrotating gravitating object in an asymptotically homogeneous and isotropic background.
glyph[negationslash]
Let us introduce some relevant concepts and model spacetimes. A spatially flat, homogeneous, isotropic cosmological model is described by a Friedmann-Lemaˆ ıtre-Robertson-Walker (FLRW) metric of the form d s 2 = -d t 2 + ( ) a t 2 d glyph[vector] x 2 , where d glyph[vector] x 2 is the line element for Euclidean 3-space, in the time gauge in which t is proper time along a comoving worldline, with spatial scale factor a t ( ). The Hubble parameter H t ( ) := ˙ a/a is usually referred to as the 'expansion rate'; an accelerating expansion rate corresponds to a situation in which ¨ = 0, a and is often quantified by the 'deceleration parameter' q := -¨ aa/a ˙ 2 ≡ -(1 + ˙ H/H 2 ).
In the standard cosmological model, results of observations such as the redshift-luminosity relationship for supernovae are interpreted as indicating that q < 0. The physical reason for this acceleration is not yet well understood, but one simple possibility is the presence of a non-vanishing cosmological constant, and most of the literature on gravitational light bending and the cosmological expansion has focused so far on this possibility. A cosmological constant Λ corresponds to constant values H = H 0 = √ Λ 3 and / q = -1, and the model used to study its effects on light bending is the Schwarzschild-de Sitter spacetime, which in the common 'static' coordinates takes the form of the Kottler metric [5],
$$\mathrm d s ^ { 2 } = - \left ( 1 - \frac { 2 m } { r } - \frac { \Lambda r ^ { 2 } } { 3 } \right ) \mathrm d t ^ { 2 } + \left ( 1 - \frac { 2 m } { r } - \frac { \Lambda r ^ { 2 } } { 3 } \right ) ^ { - 1 } \mathrm d r ^ { 2 } + r ^ { 2 } \mathrm d \Omega ^ { 2 } \,. \quad \quad ( 1. 1 )$$
This line element approaches that of the Schwarzschild metric for a black hole of mass m located at the center of the coordinate system for small r (or Λ = 0), and the de Sitter metric with cosmological constant Λ, written in static coordinates, for large r (or m = 0).
In the presence of matter fields, however, or if the accelerated expansion is due to any reason other than a cosmological constant, we do not expect H t ( ) to be constant. Indeed, in the standard ΛCDM cosmological model the time dependence of H t ( ) is governed by both Λ and different types of matter, and observationally there are indications that H has been varying in time (see, for example, Refs. [6, 7]). Our goal is to extend the work on gravitational light bending at cosmological scales to the case of a time-dependent expansion rate H t ( ) and study the effects of a non-zero ˙ H t ( ).
A model with time-dependent Hubble parameter was analyzed in an 2010 article by Kantowski et al. [8], who used a 'Swiss cheese' model to examine the bending of light in a cosmological setting. In that paper, the authors consider a FLRW cosmology where a spherical region has been replaced by a bubble inside which the metric is of the SdS type. The values of the bubble radius r b , black-hole mass m and cosmological constant Λ are subject to constraints from matching conditions at the bubble boundary; in particular, r b scales in time the same way a t ( ) does. They can then perturbatively calculate the lightlike
geodesics up to orders ( m/r 0 ) 2 and ( m/r 0 )Λ r 2 0 , where r 0 is the distance of closest approach to the central black hole. Using realistic parameter values in a ΛCDM model, the authors found that the combined effects of the additional Λ and matter terms on the bending angles could be as large as a few percent.
Our approach will be to model a gravitating object in an expanding universe with nonconstant H t ( ) by a McVittie metric, and examine the behavior of null geodesics in it. In this model we avoid the need for 'gluing' together two different solutions of Einstein's equation and we can prescribe an arbitrary expansion rate; the price we pay is having an unphysical equation of state for the fluid matter, and the need to integrate the geodesic equation numerically; in this sense, our results are complementary to those of Kantowski et al. The McVittie metric has been recently used in work on the effect of cosmological expansion on gravitational lensing [9], but in that paper H was taken to be constant and, as we will see in the next section, the spacetime is then equivalent to Schwarzschild-de Sitter. In Sec. II we review the McVittie metric, written in two convenient coordinate systems. In Sec. III we summarize previous results on the bending of light in the SdS spacetime. Sec. IV contains our results on light bending in McVittie spacetimes with varying H t ( ), and we finish with some conclusions in Sec. V. We use units in which c = G = 1.
## II. MCVITTIE METRIC
A metric that can be thought of as representing a Schwarzschild black hole embedded in a fluid-filled FLRW spacetime was first derived in 1933 by G.C. McVittie [10]. Surprisingly, despite it being a rather old solution of the Einstein equation, its exact interpretation has been somewhat controversial, and papers have appeared questioning the black-hole interpretation of the metric [11, 12]; recent work [13, 14] has shown, however, that this metric does indeed model a black hole embedded in an expanding cosmology, at least when the background spacetime is spatially flat and the Hubble parameter satisfies H t ( ) > 0. We should point out though that the McVittie metric is not considered as representing a generic solution of this type. Physically, in this situation one would expect the energy density to be inhomogeneous, due to the expansion of the universe at large scales and the attraction of the black hole at small scales, which might cause the black hole to accrete matter as in a cosmological analog of the Vaidya metric [15, 16]. As Kaloper et al. argue, this is not the case; the equation relating the energy density to H t ( ) is exactly the same as for a FLRW cosmology, while the inhomogeneity appears in the expression for the pressure [13].
We use the McVittie metric in this work because it allows us to freely specify the function H t ( ) in an asymptotically FLRW model with a localized gravitating mass m ; since we will consider the light ray as interacting only with the geometry, the form of the matter equation of state will not be directly relevant for us. Of course, that equation of state does affect the local spacetime geometry, but the fact that our local geometry does not have the sharp bubble boundaries of the Swiss-cheese universe makes it arguably more realistic. The McVittie line element can be written down in the diagonal form
$$\mathrm d s ^ { 2 } = - \left ( \frac { 1 - \mu } { 1 + \mu } \right ) ^ { 2 } \mathrm d t ^ { 2 } + ( 1 + \mu ) ^ { 4 } \, a ^ { 2 } ( t ) \, \mathrm d \vec { X } ^ { \, 2 } \, \quad \mu \coloneqq \frac { m } { 2 \, a ( t ) R }$$
where a t ( ) is the asymptotic scale factor; these coordinates are an inhomogeneous-space version of comoving coordinates, in the sense that at large values of | glyph[vector] X | the line element tends to that of a FLRW spacetime with scale factor a t ( ), expressed in comoving coordinates.
↦
Several authors [13, 17, 18] have introduced a coordinate transformation ( t, X glyph[vector] ) → ( t, glyph[vector] x ) to 'Painlev´-Gullstrand type' coordinates defined by e
$$r = ( 1 + \mu ) ^ { 2 } a ( t ) \, R$$
(the coordinates t , θ and φ are unchanged), in which the line element takes the form
$$\mathrm d s ^ { 2 } = - \left ( f ( r ) - H ^ { 2 } ( t ) \, r ^ { 2 } \right ) \mathrm d t ^ { 2 } - \frac { 2 H ( t ) \, r } { \sqrt { f ( r ) } } \, \mathrm d r \, \mathrm d t + \frac { \mathrm d r ^ { 2 } } { f ( r ) } + r ^ { 2 } \, \mathrm d \Omega ^ { 2 }$$
with f ( r ) := 1 -2 m/r and again H t ( ) = ˙ a/a . Thus, r is an area radial coordinate, characterized by the fact that the surface area of a 2-sphere of constant r is 4 πr 2 , and the function H t ( ) represents the asymptotic expansion rate in comoving coordinates, so we still call it the 'Hubble parameter'. Note that in the ADM-type terminology for the 3+1 split of a spacetime metric, the spatial part of this metric has no time dependence and has the same form as in Schwarzschild spacetime, while the cosmological expansion terms have been moved into the shift vector, with non-vanishing component N r = -H t ( ) r/ √ f ( r ).
For us, it is importat to note that the SdS spacetime with cosmological constant Λ is a special case of the McVittie metric with constant H t ( ) = H 0 and Λ = 3 H 2 0 , as one can see by defining the time coordinate transformation
$$\mathrm d t = \mathrm d \bar { t } - \frac { 1 } { 1 - \frac { 2 m } { r } - \frac { 1 } { 3 } \Lambda r ^ { 2 } } \, \sqrt { \frac { \Lambda r ^ { 2 } / 3 } { 1 - 2 m / r } } \, \mathrm d r \,.$$
Using ¯ as time coordinate, the McVittie metric takes the traditional Kottler form of Eq. t (1.1) (see, e.g., Ref. [18]).
In the literature on light deflection in SdS spacetimes authors use the Kottler, diagonal form of the metric, because of its greater simplicity. If H t ( ) is not constant, however, a coordinate transformation of the type (2.4) does not yield a diagonal, 'generalized Kottler' metric with 'time-varying Λ', and we will use instead the form (2.3). We will call (2.3) the (spatially) 'static' form of the McVittie metric, and the corresponding observers 'static', as opposed to the 'comoving' ones of the line element (2.1), in which the spatial metric does not have a static form even when H t ( ) is constant.
## III. NULL GEODESICS AND LIGHT BENDING IN SDS SPACETIMES
In this section we will review some of the definitions and results in the literature on null geodesics in SdS spacetime (see, for example, the papers by Ishak and Rindler [2, 3, 19] and Lebedev & Lake [4]). In view of the fact that we will later generalize those results, however, we will replace Λ by 3 H 2 0 , where H 0 is the constant value of H t ( ), use the line element (2.3), and add some remarks motivated by the more general situation.
The whole McVittie line element is now time-independent, and we can use the static and rotational Killing vector fields ξ = ∂/∂t and η = ∂/∂φ to define two conserved quantities along a geodesic x µ ( λ ) with tangent vector K µ := d x / µ d λ ,
$$e \coloneqq - g _ { \alpha \beta } \, \xi ^ { \alpha } K ^ { \beta } = [ f ( r ) - H _ { 0 } ^ { 2 } r ^ { 2 } ] \, K ^ { t } + \frac { H _ { 0 } \, r } { \sqrt { f ( r ) } } \, K ^ { r } \, \quad \ell \coloneqq g _ { \alpha \beta } \, \eta ^ { \alpha } K ^ { \beta } = r ^ { 2 } \, K ^ { \phi } \, \quad ( 3. 1 )$$
for a geodesic in the equatorial coordinate plane. In other words, the geodesic satisfies
$$K ^ { \phi } = \frac { \ell } { r ^ { 2 } } \, \quad K ^ { t } = \frac { e } { f ( r ) - H _ { 0 } ^ { 2 } \, r ^ { 2 } } - \frac { H _ { 0 } r } { \sqrt { f ( r ) } \, [ f ( r ) - H _ { 0 } ^ { 2 } \, r ^ { 2 } ] } \, K ^ { r } \.$$
Using these equations, the conditions for a null geodesic with g µν K K µ ν = 0 can be reduced to a single differential equation, which in terms of the conserved quantities becomes
$$\frac { 1 } { \ell ^ { 2 } } \left ( \frac { d r } { d \lambda } \right ) ^ { 2 } + \frac { 1 } { r ^ { 2 } } \left ( f ( r ) - H _ { 0 } ^ { 2 } \, r ^ { 2 } \right ) = \frac { 1 } { b ^ { 2 } }$$
where we have defined b := glyph[lscript]/e , the impact parameter. Note that, as expected, if H 0 → 0 this is the same equation one gets for a null geodesic in Schwarzschild spacetime (see, for example, Ref. [20] or most of the references on light bending in SdS spacetime).
The shape of the light trajectory, in the sense of the relationship between the coordinates r and φ , can be obtained using the expression for glyph[lscript] in (3.3), and one finds that
$$\frac { \mathrm d \phi } { \mathrm d r } = \pm \frac { 1 } { r ^ { 2 } } \left [ \frac { 1 } { b ^ { 2 } } + H _ { 0 } ^ { 2 } - \left ( 1 - \frac { 2 m } { r } \right ) \right ] ^ { - 1 / 2 }.$$
Notice that, since these equations do not involve t or the geometry of a constantt hypersurface, we would have obtained the same relationship between r and φ using the form (1.1) for the metric. Eq. (3.4) encodes some aspects of the effect a cosmological constant has on the formal description of light bending near a nonrotating gravitating mass, but one should be careful when discussing any measurable effects on observations. A controversy in this sense arose around the interpretation of (3.4), because although the equation does depend on Λ, it is of the same form as the one obtained in Schwarzschild spacetime if one replaces the impact parameter b in the latter by an 'effective impact parameter' B defined by
$$\frac { 1 } { B ^ { 2 } } \coloneqq \frac { 1 } { b ^ { 2 } } + H _ { 0 } ^ { 2 }$$
and one might conclude that the cosmological constant does not affect observations because it is simply 'absorbed' into the impact parameter, which is not directly measurable. This aspect has been extensively discussed starting with the original paper by Islam [1]. However, the more recent literature has made it clear that the relationship between r and φ only tells us part of the story on how the deflection angle depends on Λ, for various reasons.
One reason is that, as pointed out by Ishak and Rindler [2, 3], from (3.4) alone and without knowing the spatial metric, one can only obtain the value that an angle would have in flat space. Consider for example the angle θ in Fig. 1. This is not the full bending angle, but as the following argument illustrates it is relevant for the calculation of the latter. In Schwarzschild spacetime one can imagine placing the source and observer at infinity and calculate a bending angle that does not refer to specific locations for them and depends only on the impact parameter b , in addition to the lens mass m . Because of the curved geometry of the constantt surfaces, this is not an option in SdS spacetime [2, 3]. The simplest alternative is to set up the coordinates so that the point of closest approach is at φ = π/ 2 and use as measure of total bending the sum α + θ of the angles the light ray makes with the half-lines φ = π and φ = 0 at the intersection points, labeled S and O in Fig. 1. Because the spacetime is static and spherically symmetric, in this situation if
FIG. 1: Definition of the main angles and static-coordinate distances used in the calculation of the light bending angle. The lensing object L is located a distance r L and the light source at a distance r S from the observer O . The full bending angle is α + θ but only θ is measurable by the observer. In SdS spacetime the point of closest approach is at φ = π/ 2 and the figure is symmetric about the vertical axis; in general, this is no longer the case in McVittie spacetime.
the radial positions of S and O are equal, r SL = r L , then so are the angles, α = θ . The 'Euclidean' value θ E = tan -1 ( r d φ/ d ) r O of half of the bending angle can then be easily calculated; neglecting quadratic terms in m , one finds the well-known result (see, e.g., Refs. [2, 3])
$$\theta _ { _ { \text{E} } } \approx \frac { 2 m } { B } \approx \frac { 2 m } { b } + m b H _ { 0 } ^ { 2 } = \frac { 2 m } { b } + \frac { m b \Lambda } { 3 } \.$$
However, a more physically meaningful, covariant value for the angle between two lines takes into account the actual metric on a constantt spatial hypersurface in SdS spacetime. As first derived by Ishak & Rindler [2, 3], this measurable value θ M of the angle θ , to leading order in Λ, is given by an expression with a different H 0 -dependent term,
$$\theta _ { _ { M } } \approx \frac { 2 m } { b } - \frac { \Lambda b ^ { 3 } } { 1 2 m } = \frac { 2 m } { b } - \frac { H _ { 0 } ^ { 2 } b ^ { 3 } } { 4 m } \.$$
Secondly, when discussing observational consequences of H 0 for light bending we need to specify exactly what question we are asking. For example, if we said that Eq. (3.6) implies that the Euclidean bending angle θ E depends on H 0 through the effective parameter B , we would be assuming that the comparison is made between situations in which the impact parameter b is held constant. However, because b is not directly measurable, it may be more meaningful to specify in some other way which situations we are comparing when we change the value of H 0 . We take the point of view that, for each value of H 0 , the geodesic we use to find the angle θ needs to be uniquely determined by a fixed set of values for parameters that can in principle be measured by an observer. As we will see in the next section, our choice will lead to a different conclusion for the dependence of θ E on H 0 .
The third thing to keep in mind is that values of angles are observer-dependent. Thus, the angle θ M in (3.7) is the one that would be measured by a 'static observer', one along whose worldline r and the angular coordinates are constant, but one may want to know what angle would be measured by a 'comoving observer', one along whose worldline R rather than r is constant. This point is discussed in detail by Lebedev & Lake [4]. A general expression for the angle measured by an observer with 4-velocity U µ can be calculated from the dot product between the projections orthogonal to U µ of the vector K µ tangent to the deflected null geodesic and a radial null vector W µ (see Eq. 80 in Ref. [4]),
$$\cos \theta _ { _ { M } } = \frac { K \cdot W } { ( U \cdot K ) ( U \cdot W ) } + 1 \.$$
While we will not go into the details of how this is used to derive explicit expressions for the angles measured by different observers, let us summarize some results we can compare with our numerical ones for McVittie spacetime in the next section. From (3.8) one can derive a relationship between the Euclidean angle θ E and the angle θ M measured by a static observer in SdS spacetime [4],
$$\cdot ^ { \cdot } \tan \theta _ { _ { M } } = \sqrt { f ( r ) - H _ { 0 } ^ { 2 } \, r ^ { 2 } } \, \tan \theta _ { _ { E } } \,.$$
On the other hand, from Eq. (3.3) for null geodesics one obtains the well-known first-order lens equation, valid for the general case of a source at a distance y from the lens-observer axis (see, for example, Refs. [2-4]). The result, using the notation r S = r SL + r L , is
$$y = r _ { _ { S } } \, \theta _ { _ { \mathrm E } } - \frac { 4 m \, r _ { _ { \mathrm S L } } } { r _ { _ { \mathrm L } } \, \theta _ { _ { \mathrm E } } } \.$$
When the source is on the axis, y = 0, if we solve (3.10) for the Euclidean angle we get
$$\theta _ { _ { E } } = \sqrt { \frac { 4 m \, r _ { _ { S L } } } { r _ { _ { L } } \, r _ { _ { S } } } } \ ;$$
this θ E can in turn be plugged into (3.9) and, using the small-angle approximation, for the static-observer measurable angle we get
$$\theta _ { _ { M } } = \sqrt { \left ( 1 - \frac { 2 m } { r _ { _ { L } } } - H _ { 0 } ^ { 2 } \, r _ { _ { L } } ^ { 2 } \right ) \frac { 4 m \, r _ { _ { S L } } } { r _ { _ { L } } \, r _ { _ { S } } } } \.$$
An expression for the comoving-observer measurable angle, which is physically related to the static-observer one by an appropriate aberration factor, was obtained in Refs. [2-4]. For more comments on formulating physically relevant questions about the effect of Λ on the bending angle in SdS spacetime, we refer to the work of Ishak and Rindler [2, 3] or the more recent papers by Lebedev and Lake [4] and by Hammad [21].
## IV. NULL GEODESICS IN MCVITTIE SPACETIME
Let us now consider light bending in a McVittie spacetime with non-constant expansion rate H t ( ). In the coordinates of the 'static' line element (2.3), a light ray in this spacetime obeys the null condition
$$\left [ f ( r ) - r ^ { 2 } H ( t ) ^ { 2 } \right ] ( K ^ { t } ) ^ { 2 } + \frac { 2 r H ( t ) } { \sqrt { f ( r ) } } \left ( K ^ { r } K ^ { t } \right ) - \frac { 1 } { f ( r ) } \left ( K ^ { r } \right ) ^ { 2 } - \frac { \ell ^ { 2 } } { r ^ { 2 } } = 0 \, \quad \ \ ( 4. 1 )$$
where as before the angular momentum glyph[lscript] = r 2 K φ is conserved. Unlike in the earlier SdS case, however, there is no second conserved quantity e , because the spacetime has no timelike Killing vector field in general. Therefore, to find the deflected light ray and determine its tangent vector at the observer's location we must actually solve at least one component of the null geodesic equation in addition to (4.1), as opposed to just using first integrals.
Using the explicit form of the connection coefficients Γ µ αβ for the McVittie metric (2.3), the relevant components of the geodesic equation d K / λ µ d +Γ µ αβ K K α β = 0 become
$$\frac { \text{d} K ^ { t } } { \text{d} \lambda } & + \left ( \frac { r H ( t ) \left [ 2 r H ( t ) ^ { 2 } - f ^ { \prime } ( r ) \right ] } { 2 \sqrt { f ( r ) } } \right ) ( K ^ { t } ) ^ { 2 } + \frac { f ^ { \prime } ( r ) - 2 r H ( t ) ^ { 2 } } { f ( r ) } \, K ^ { r } K ^ { t } \\ & + \frac { H ( t ) } { f ( r ) ^ { 3 / 2 } } \, ( K ^ { r } ) ^ { 2 } + \frac { H ( t ) \, \ell ^ { 2 } } { \sqrt { f ( r ) } \, r ^ { 2 } } = 0 \\ \frac { \text{d} \nu \nu } { \text{r} } \quad \in \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r \, \quad \ e r$$
$$\frac { \ddots \cdot \cdot \cdot } { \ddots \lambda } & + \frac { 2 r H ( t ) ^ { 2 } - f ^ { \prime } ( r ) } { 2 f ( r ) } \left ( K ^ { r } \right ) ^ { 2 } + \frac { r H ( t ) } { \sqrt { f ( r ) } } \left ( f ^ { \prime } ( r ) - 2 r H ( t ) ^ { 2 } \right ) K ^ { t } K ^ { r } + \frac { \ell ^ { 2 } } { r ^ { 3 } } \left ( r ^ { 2 } H ^ { 2 } ( t ) - f ( r ) \right ) \\ & + \frac { 1 } { 2 } \left ( \left ( f ( r ) - r ^ { 2 } H ^ { 2 } ( t ) \right ) ( f ^ { \prime } ( r ) - 2 r H ^ { 2 } ( t ) ) - 2 r \sqrt { f ( r ) } H ^ { \prime } ( t ) \right ) ( K ^ { t } ) ^ { 2 } = 0 \, \\ \cdot \quad \dots \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \end{cases}$$
and our goal is to solve Eqs. (4.2) and (4.3) for K µ = d x / µ d λ , with initial values at the source satisfying the condition (4.1). We do not know how to solve those equations analytically, so we integrated them numerically instead, using the standard fourth-order Runge-Kutta method with approximately 10 6 steps, and a variable integration step which according to standard theory [22] leads to cumulative relative errors due to the Runge-Kutta approximation at most of order 10 -13 for our variables. A further accuracy check was performed by comparing the final value of K t produced by each simulation to that which is demanded by the null normalization condition K K µ µ = 0 using the final values for the other variables; we found relative differences of order 10 -12 , indicating that in practice those errors were dominated by roundoff errors.
Before describing the results of our numerical simulations, we need to discuss how we will determine the effect of the time dependence of H t ( ) on the light bending angle. We will be comparing with each other McVittie metrics with the same mass m and the same value of the Hubble parameter H t ( ) at the time t 0 when the null geodesic leaves the source, but different ˙ H t ( ). Treating H t ( ) as a slowly varying function over the relevant times, we will parameterize it simply by giving the values of H 0 = H t ( 0 ) and A = ˙ ( H t 0 ), or
$$H ( t ) = H _ { 0 } + A \left ( t - t _ { 0 } \right ) \,.$$
This parametrization is a generic one for small A , and it allows us to study the effect of different time rates of change of H t ( ) on the bending angle using a range of values for A motivated by data from cosmological observations, rather than choosing a value tied to a specific cosmological model.
Comparing predictions for the bending angle as A varies, however, raises a conceptual issue. Because McVittie metrics with different values of A are to be thought as entirely different spacetime manifolds, a key point when comparing them is to formulate a criterion for determining which null geodesic in each spacetime is to be used for the comparison. One possibility, mentioned when we discussed SdS metrics, might have been to use geodesics with the same impact parameter b . We take instead the point of view that a meaningful criterion should be based on quantities an observer has access to.
For simplicity, in this paper we will assume that observers are able to measure the distances to the source and lens (which to leading order in m we identify with the values of r S and r L ), in addition to the angle θ M , as well as the fact that the source and lens are aligned, and look for how θ M varies with A when m H , 0 , r S and r L are kept constant. In a McVittie spacetime, fixing a set of values for those parameters determines uniquely a pair of
null geodesics, one on each side of the lens, and for definiteness we will choose the clockwise one, as in Fig. 1. Considering r S and r L as directly measurable is overly simplistic, and considering only cases in which the source, lensing object and observer are aligned limits the generality of the results. We leave it as a goal for future work to remove those limitations, but we should also point out that aligned configurations are the physically most relevant ones; although light bending certainly occurs in more general settings, it is much more likely to be noticed observationally when all objects are nearly aligned, or y = 0 in Eq. (3.10).
When performing the simulations, we start each light ray from the coordinate location r = r SL = r S -r L , φ = π at t = t 0 , and aim it in a trial direction α towards decreasing values of φ . The geodesic equations (4.2) and (4.3) are then integrated numerically until the light ray reaches φ = 0, where the value of r is checked against the chosen r L . If we find that r at φ = 0 is smaller (greater) than the desired value, the simulation is repeated using a larger (smaller) value for α until the final r equals r L , at which point the components of K µ are recorded. The relative tolerance when the value of r is matched with r L is 10 -6 ; this is the largest source of errors in our results for the bending angles, and it is small enough that the corresponding error bars do not show up in Figs. 2-4. Notice that when H t ( ) is time-dependent the spatial projection of a deflected null geodesic is not symmetric about the point of closest approach, and in a sketch of the trajectory of the light ray similar to the SdS one in Fig. 1, even with r SL = r L , the angles α and θ will not be equal and the point of closest approach will not be at φ = π/ 2.
Once the components of K µ at the location of the observer are known, we use (3.8) to calculate the angle θ M measured by an observer with 4-velocity U µ . For static and comoving observers the static-coordinate components of the 4-velocities are
$$U ^ { \mu } _ { \text{stat} } = \left ( \frac { 1 } { \sqrt { f ( r ) - H ^ { 2 } ( t ) \, r ^ { 2 } } }, \, 0, \, 0, \, 0 \right ) \,, \quad U ^ { \mu } _ { \text{com} } = \left ( \frac { 1 } { \sqrt { f ( r ) } }, H ( t ) \, r, \, 0, \, 0 \right ) \,,$$
respectively, and finding a null radial vector is simple; we choose
$$W ^ { \mu } = \left ( 1, \sqrt { f ( r ) } \left [ \sqrt { f ( r ) } + H r \right ], 0, 0 \right ) \,.$$
To specify the metric we choose a single value m = 10 14 M sun for the mass of the lensing object, considered to be representative of situations in which a light ray traveling over cosmological distances is deflected by a galaxy cluster. For H 0 we choose three values in the range from 0 to 70 km/s/Mpc (from no expansion to approximately the currently accepted value). To estimate a realistic range of values for A we use the fact that q = 1 2 (1 + 3 w ), where w is the cosmological equation of state parameter, and we take values consistent with current estimates to be approximately in the range from 1.05 to 1.25 [6]. Based on this and the relationship ˙ H = -(1 + q H ) 0 2 , we choose the range from -1 0 . × 10 -9 to +1 0 . × 10 -9 km/s/Mpc/yr for the values for A . To specify the null geodesic, for the coordinate values of the source-lens and lens-observer distances we choose r SL = r L = 1 Gpc, considered as representative of typical values in a cosmological lensing situation, with a tolerance of 0.1 Mpc in the value of r for the intersection of the geodesic with φ = 0.
The results of our simulations are shown in Figs. 2-4. For comparison with what we said in Sec. III about SdS spacetime, Fig. 2 shows the Euclidean angle θ E = tan -1 ( r K φ /K r ) as a function of A , for various values of H 0 . For A = 0 we recover the SdS situation and the value of θ E agrees with what one finds from Eq. (3.11). In particular, as we see from the plot and as can also be concluded more generally from Eq. (3.6), when we use our criterion
- GLYPH<230> H 0 = 0
- GLYPH<224> H 0 = 35 km s Mpc GLYPH<144> GLYPH<144>
- GLYPH<236> H 0 = 70 km s Mpc GLYPH<144> GLYPH<144>
L
FIG. 2: A plot of the Euclidean angle θ E for an observer at fixed distances r S = 2 Gpc from the source and r L = 1 Gpc from the lensing object (in static coordinates) vs. the acceleration parameter A , for various values of H 0 . For A = 0 the bending angle is independent of H 0 , to a good approximation, and its value agrees with Eq. (3.11).
glyph[negationslash]
for identifying light rays in different spacetimes, based on fixing the values of r S and r L , the Euclidean angle in SdS spacetime does not depend on H 0 , contrary to what one would conclude by fixing b ; when A = 0, however, θ E does depend on H 0 .
Figs. 3 and 4 show the more physically meaningful curved-geometry angle θ M measured by static and comoving observers, respectively, as functions of A for various values of H 0 . It can be checked that for A = 0, in SdS spacetime, the value of the static-observer angle θ M decreases with increasing H 0 , in agreement with (3.12). In order to study the slope of the measured bending angle versus acceleration ( A ) for a comoving observer, let's recalculate cos θ com from (3.8), using the comoving observer velocity U µ com in (4.5) and the null radial vector W µ in (4.6),
$$\cos ( \theta _ { \text{com} } ) = \frac { K ^ { r } - r \sqrt { f ( r ) } \, H ( t ) K ^ { t } } { f ( r ) K ^ { t } } = \frac { 1 } { f ( r ) } \, \frac { K ^ { r } } { K ^ { t } } - \frac { r } { \sqrt { f ( r ) } } \, H ( t ) \.$$
If one further uses a linear polynomial as in (4.4) for H t ( ), the variation of θ com as a function of A , keeping r = r L fixed, can be a calculated from
$$\frac { \partial } { \partial A } \cos ( \theta _ { \text{com} } ) = \frac { 1 } { f ( r ) } \, \frac { \partial } { \partial A } \left ( \frac { K ^ { r } } { K ^ { t } } \right ) - \frac { r } { \sqrt { f ( r ) } } \, t \,.$$
Since our observer is far from the central mass, f ( r ) ≈ 1 and K r ≈ K t , so the second term in (4.8) dominates and the slope of θ com vs. A is positive, as seen in Fig. 4.
L
FIG. 3: A plot of the angle θ M measured by a static observer at fixed distances r S = 2 Gpc from the source and r L = 1 Gpc from the lensing object (in static coordinates) vs. the acceleration parameter A , for various values of H 0 . For each value of A , the arrival angle decreases with H 0 . The direction of change with respect to H 0 is the same as the direction of change with A .
## V. CONCLUSIONS
In this paper we examined the bending of null geodesics in spatially flat McVittie metrics, exact solutions to the Einstein equation which have been shown to represent nonrotating black holes embedded in FLRW background spacetimes. We used a slowly varying Hubble parameter H t ( ) = H 0 + A t ( -t 0 ), and the null geodesics were found numerically for situations in which the source and lens were aligned as seen by the observer. Simulations were run with fixed values for the mass of the central object and distances from the observer to the source and lens, in static coordinates, while we used three different values for H 0 and in each case we looked at how the angle of arrival of null geodesics at the observer's location varied with A ; in the A = 0 case, our results are in agreement with earlier perturbative calculations for light bending in Schwarzschild-de Sitter spacetimes [2-4].
glyph[negationslash]
Each simulation gave us three different values for the angle of arrival. The Euclidean angle (calculated using an auxiliary, fictitious flat spatial metric and therefore not physically measurable, but nevertheless useful for comparison with previous work and as a check on the resuls) is independent of H 0 if there is no acceleration, A = 0, but interestingly our simulations show a small dependence on H 0 if one considers cases with A = 0. The two other types of angles are the ones that would be measured by static and comoving observers, respectively, and both show a dependence on H 0 as well as on A . In this model, using the worldlines of comoving observers as the better approximations to those of physical observers,
FIG. 4: A plot of the angle θ M measured by a comoving observer at fixed distances r S = 2 Gpc from the source and r L = 1 Gpc from the lensing object (in static coordinates) vs. the acceleration parameter A , for various values of H 0 . For each value of A , the arrival angle increases with H 0 . The direction of change with respect to H 0 is the same as the direction of change with A .
and the results from the simulations with H 0 = 70 km/s/Mpc as more representative of the actual expansion rate, we find that over the full range of values we used for A , and with our values for the distances and lens mass the bending angle varies by about 5% around 20 ′′ or so, which agrees in order of magnitude with the results found by Kantowski et al. in the Swiss-cheese model [8]. These values may seem encouraging, but unfortunately the effects we described would most likely be dwarfed by measurement uncertainties and departures of the metrics around actual galaxy clusters from the spherically symmetric ones used here to model them. We view this work as a first step in quantifying the effect of ˙ H on light bending and lensing in a useful way, and an improved lens modeling is one way in which the work will have to be extended before these bending effects can be meaningfully related to measurements.
Two other ways in which our approach can be improved are the fact that we fixed the values of the source and lens coordinates r S and r L as if they were directly measurable, and the fact that we considered only cases in which the source, lensing object and observer are aligned. Obtaining results beyond the latter limitation is essentially straightforward, although one will have to extend our criterion for identifying geodesics in different spacetimes to a more general setting. To replace the criterion based on values of r by a more realistic one based on redshifts, one would have to use the relationship between redshifts and distances, which depends on the cosmological expansion history. Related to this is the fact that, although the simulations themselves and the calculations of the arrival angles are nonperturbative, fixing values of r is equivalent to fixing those of actual spatial distances only
to leading order in m ; and using the same values of H in different spacetimes at the time the geodesic leaves the source is equivalent to using the same values of H when the geodesics arrive at the observer's location also only to leading order.
## REFERENCES
- [1] J.M. Islam, 'The cosmological constant and classical tests of general relativity,' Phys. Lett. A97 , 239 (1983).
- [2] W. Rindler and M. Ishak, 'The contribution of the cosmological constant to the relativistic bending of light revisited,' Phys. Rev. D 76 , 043006 (2007) ( arXiv:0709.2948 ).
- [3] M. Ishak and W. Rindler, 'The relevance of the cosmological constant for lensing,' Gen. Rel. Grav. 42 , 2247 (2010) ( arXiv:1006.0014 ).
- [4] D. Lebedev and K. Lake, 'On the influence of the cosmological constant on trajectories of light and associated measurements in Schwarzschild de Sitter space' ( arXiv:1308.4931 ).
- [5] F. Kottler, ' ¨ ber die physikalischen Grundlagen der Einsteinschen Gravitationstheorie,' U Ann. Phys. 361 , 401 (1918).
- [6] A. Rest et al., 'Cosmological constraints from measurements of type Ia supernovae discovered during the first 1.5 years of the Pan-STARRS1 survey,' Astrophys. J. 795 , 44 (2014) ( arXiv:1310.3828 ).
- [7] I. Odderskov et al., 'On the local variation of the Hubble constant,' JCAP 10(2014)028 ( arXiv:1407.7364 ).
- [8] R. Kantowski, B. Chen and X. Dai, 'Gravitational lensing corrections in flat LambdaCDM cosmology,' Astrophys. J. 718 , 913 (2010) ( arXiv:0909.3308 ).
- [9] O.F. Piattella, 'Lensing in McVittie metric,' Phys. Rev. D 93 : 024020 (2016) ( arXiv:1508.04763 )
- [10] C.G. McVittie, 'The mass-particle in an expanding universe,' MNRAS 93 , 325 (1933).
[11]
B.C. Nolan, 'A point mass in an isotropic universe:
Existence, uniqueness and basic proper- ties,'
Phys. Rev.
D
58
,
064006 (1998) (
gr-qc/9805041
).
- [12] B.C. Nolan, 'A point mass in an isotropic universe. 3. The region R ≤ 2 m ,' Class. Quantum Grav. 16 , 3183 (1999) ( arXiv:gr-qc/9907018 ).
- [13] N. Kaloper, M. Kleban and D. Martin, 'McVittie's legacy: Black holes in an expanding universe,' Phys. Rev. D 81 : 104044 (2010) ( arXiv:1003.4777 ).
- [14] V. Faraoni, A.F. Zambrano Moreno and R. Nandra, 'Making sense of the bizarre behaviour of horizons in the McVittie spacetime,' Phys. Rev. D 85 : 083526 (2012) ( arXiv:1202.0719 ).
- [15] P.C. Vaidya, 'Nonstatic solutions of Einstein's field equations for spheres of fluids radiating energy,' Phys. Rev. 83 , 10 (1951).
- [16] See also E. Poisson, A Relativist's Toolkit: The Mathematics of Black-Hole Mechanics , Cambridge University Press 2004.
- [17] B. Bolen, L. Bombelli and R. Puzio, 'Expansion-induced contribution to the precession of binary orbits,' Class. Quantum Grav. 18 , 1173 (2001) ( gr-qc/0009018 ).
- [18] R. Nandra, A. Lasenby and M. Hobson, 'The effect of an expanding universe on massive objects,' MNRAS 422 , 2945 (2012) ( arXiv:1104.4458 ).
- [19] M. Ishak, W. Rindler and J. Dossett, 'More on lensing by a cosmological constant' Mon. Not. Roy. Astron. Soc. 403 , 2152-2156 (2010) ( arXiv:0810.4956 ).
- [20] J. Hartle, Gravity: An Introduction to Einstein's General Relativity , Addison-Wesley 2002.
- [21] F. Hammad, 'A note on the effect of the cosmological constant on the bending of light,' Mod. Phys. Lett. A 28 , 1350181 (2013) ( arXiv:1309.0263 ).
- [22] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery. Numerical Recipes . Cambridge University Press, 1988.
|
10.1007/s10714-016-2175-6
|
[
"Mir Emad Aghili",
"Brett Bolen",
"Luca Bombelli"
] |
2014-08-04T19:46:56+00:00
|
2016-06-01T14:17:52+00:00
|
[
"gr-qc",
"astro-ph.CO"
] |
Effect of Accelerated Global Expansion on Bending of Light
|
In 2007 Rindler and Ishak showed that, contrary to previous claims, the value
of the cosmological constant does have an effect on light deflection by a
gravitating object in an expanding universe, modeled by a
Schwarzschild-de~Sitter spacetime. In this paper we consider light bending in
the more general situation of a gravitating object in a cosmological background
with varying expansion rate $H(t)$. We calculate numerically the null geodesics
representing light rays deflected by a black hole in an accelerating
Friedmann-Lema\^itre-Robertson-Walker universe, modeled by a McVittie metric.
Keeping the values of the distances from the observer to the lensing object and
to the source fixed, we plot the dependence of the bending angle measured by
two different sets of observers in this spacetime on the rate of change of
$H(t)$.
|
1408.0787v1
|
## Higgs boson production through bb ¯ annihilation at threshold in N LO QCD 3
Taushif Ahmed, a Narayan Rana a and V. Ravindran b
a Regional Centre for Accelerator-based Particle Physics, Harish-Chandra Research Institute, Allahabad, India
b The Institute of Mathematical Sciences, Chennai, India
E-mail: [email protected] , [email protected] , [email protected]
Abstract: We present threshold enhanced N LO QCD corrections to inclusive Higgs 3 production through bottom anti-bottom annihilation at hadron colliders using threshold resummed cross section. The resummed cross section is obtained using factorization properties and Sudakov resummation of the inclusive cross section. We use the recent results on threshold N LO corrections in QCD for Drell-Yan production and three loop QCD cor3 rections to Higgs form factor with bottom anti-bottom quark to achieve this task. This is the first step towards the evaluation of complete N LO result. 3 We have numerically demonstrated the importance of such corrections at the LHC.
Keywords: QCD, Higgs, Threshold corrections
| Contents | Contents | Contents |
|------------|-----------------------|------------|
| 1 | Introduction | 1 |
| 2 | Threshold resummation | 2 |
| 3 | Results | 9 |
## 1 Introduction
The discovery of Higgs boson by ATLAS [1] and CMS [2] collaborations of the Large Hadron Collider (LHC) at CERN has not only shed the light on the dynamics behind the electroweak symmetry breaking but also put the Standard Model (SM) of particle physics on a firmer ground. In the SM, the elementary particles such as quarks, leptons and gauge bosons, Z, W ± acquire their masses through spontaneous symmetry breaking (SSB). The Higgs mechanism provides the framework for SSB. The SM predicts the existence of a Higgs boson whose mass is a parameter of the model. The recent discovery of Higgs boson provides a valuable information on this, namely on its mass which is about 125.5 GeV. The searches for the Higgs boson have been going on for several decades in various experiments. Earlier experiments such as LEP [3] and Tevatron [4] played an important role in the discovery by the LHC collaborations through narrowing down its possible mass range. LEP excluded Higgs boson of mass below 114.4 GeV and their precision electroweak measurements [5] hinted the mass less than 152 GeV at 95% confidence level (CL), while Tevatron excluded Higgs boson of mass in the range 162 -166 GeV at 95% CL.
Higgs bosons are produced dominantly at the LHC via gluon gluon fusion through top quark loop, while the sub-dominant ones are vector boson fusion, associated production of Higgs boson with vector bosons, with top anti-top pairs and also in bottom anti-bottom annihilation. The inclusive productions of Higgs boson in gluon gluon [6], vector boson fusion processes [7] and associated production with vector bosons [8] are known to next to next to leading order (NNLO) accuracy in QCD. Higgs production in bottom anti-bottom annihilation is also known to NNLO accuracy in the variable flavour scheme (VFS) [9, 10], while it is known to NLO in the fixed flavour scheme (FFS) [11]. In the minimal super symmetric standard model (MSSM), the coupling of bottom quarks to Higgs becomes large in the large tan β region, where tan β is the ratio of vacuum expectation values of up and down type Higgs fields. This can enhance contributions from bottom anti-bottom annihilation subprocesses.
While the theoretical predictions of NNLO [6] and next to next to leading log (NNLL) [12] QCD corrections and of two loop electroweak effects [13] played an important role in the Higgs discovery, the theoretical uncertainties resulting from factorization and renormalization scales are not fully under control. Hence, the efforts to go beyond NNLO are
going on intensively. Some of the ingredients to obtain N LO QCD corrections are al3 ready available. For example, quark and gluon form factors [14-17], the mass factorization kernels [18] and the renormalization constant [19] for the effective operator describing the coupling of Higgs boson with the SM fields in the infinite top quark mass limit up to three loop level in dimensional regularization are known for some time. In addition, NNLO soft contributions are known [20] to all orders in glyph[epsilon1] for both DY and Higgs productions using dimensional regularization with space time dimension being d = 4 + glyph[epsilon1] . They were used to obtain the partial N LO threshold effects [21, 22] to Drell-Yan production of di-leptons 3 and inclusive productions of Higgs boson through gluon gluon fusion and in bottom antibottom annihilation. Threshold contribution to the inclusive production cross section is expanded in terms of δ (1 -z ) and D i ( z ) where
$$\mathcal { D } _ { i } ( z ) = \left ( \frac { \ln ( 1 - z ) } { 1 - z } \right ) _ { + }$$
with the scaling parameter z = m /s 2 H ˆ for Higgs and z = m 2 l + l -/s ˆ for DY. Here m H , m l + l -and ˆare mass of the Higgs boson, invariant mass of the di-leptons and center of mass energy s of the partonic reaction responsible for production mechanism respectively. The missing δ (1 -z ) terms for the complete N LO threshold contributions to the Higgs production 3 through gluon gluon fusion are now available due to the seminal work by Anastasiou et al [23] where the relevant soft contributions were obtained from the real radiations at N LO 3 level. In [24], we exploited the universal structure of the soft radiations to obtain the corresponding soft gluon contributions to DY production, which led to the evaluation of missing δ (1 -z ) part of the N LO threshold corrections, later confirmed in [25]. 3 For the Higgs production through bb ¯ annihilation, till date, only partial N 3 LO threshold corrections are known [22]. It was not possible to determine the δ (1 -z ) at N LO due to the lack of 3 information on three loop finite part of bottom anti-bottom higgs form factor in QCD and the soft gluon radiation at N 3 LO level. The recent results on Higgs form factor with bottom anti-bottom by Gehrmann and Kara [26] and on the universal soft distribution obtained for the Drell-Yan production [24] can now be used to obtain δ (1 -z ) part of the threshold N LO contribution. 3 For the soft gluon radiations in the bb ¯ annihilation, the results from [24] can be used as they do not depend on the flavour of the incoming quark states. We have set bottom quark mass to be zero throughout except in the Yukawa coupling.
In the next section, we present the details of the threshold resummation and in section 3 we present our results for threshold N 3 LO QCD contributions to Higgs production through bb ¯ annihilation at hadron colliders and their numerical impact . The numerical impact of threshold enhanced N LO contributions is demonstrated for the LHC energy 3 √ s = 14 TeV by studying the stability of the perturbation theory under factorization and renormalization scales. Finally we give a brief summary of our findings.
## 2 Threshold resummation
The interaction of bottom quarks and Higgs boson is given by the action
$$S _ { I } ^ { b } = - \lambda \, \int d ^ { 4 } x \, \phi ( x ) \overline { \psi } _ { b } ( x ) \psi _ { b } ( x )$$
where ψ b ( x ) denotes the bottom quark field and φ x ( ) the scalar field. λ is the Yukawa coupling given by √ 2 m /ν b , with the bottom quark mass m b and the vacuum expectation value ν ≈ 246 GeV. In MSSM, for the pseudoscalar Higgs boson, we need to replace λφ x ψ ( ) b ( x ψ ) b ( x ) by ˜ ˜ ( λφ x ψ ) b ( x γ ψ ) 5 b ( x ) in the above equation. The MSSM couplings are given by √
$$\tilde { \lambda } = \begin{cases} - \frac { \sqrt { 2 } m _ { b } \sin \alpha } { \nu \cos \beta } \,, \quad & \tilde { \phi } = h \,, \\ \frac { \sqrt { 2 } m _ { b } \cos \alpha } { \nu \cos \beta } \,, \quad & \tilde { \phi } = H \,, \\ \frac { \sqrt { 2 } m _ { b } \tan \beta } { \nu } \,, \quad & \tilde { \phi } = A \end{cases}.$$
respectively. The angle α measures the mixing of weak and mass eigenstates of neutral Higgs bosons. We use VFS scheme throughout, hence except in the Yukawa coupling, m b is taken to be zero like other light quarks in the theory.
The inclusive Higgs production through bottom anti-bottom annihilation can be computed using
$$\sigma ^ { b } ( s, q ^ { 2 } ) = \sigma ^ { ( 0 ) } _ { b \bar { b } } ( \mu _ { R } ^ { 2 } ) \sum _ { a c } \int d x _ { 1 } d x _ { 2 } f _ { a } ( x _ { 1 }, \mu _ { F } ^ { 2 } ) f _ { c } ( x _ { 2 }, \mu _ { F } ^ { 2 } ) \Delta ^ { b } _ { a c } ( \hat { s }, q ^ { 2 }, \mu _ { F } ^ { 2 }, \mu _ { R } ^ { 2 } ), \quad ( 2. 2 )$$
where f a ( x , µ 1 2 F ) and f c ( x , µ 2 2 F ) are parton distribution functions with momentum fractions x 1 and x 2 respectively. µ F is the factorization scale and ˆ = s x x s 1 2 where s (ˆ) is the square s of hadronic (partonic) center of mass energy. The born cross section is given by
$$\sigma _ { b \bar { b } } ^ { ( 0 ) } = \frac { \pi \lambda ^ { 2 } ( \mu _ { R } ^ { 2 } ) } { 1 2 m _ { H } ^ { 2 } } \,.$$
The born normalized partonic subprocesses after mass factorization are denoted by ∆ b ac where initial state partons are a and c . Here, q 2 = m 2 H , µ F results from mass factorization and the renormalization scale µ R is due to UV renormalization. ∆ b ac can be decomposed into two parts denoted by ∆ sv b and ∆ b,R ac .
$$\Delta _ { a c } ^ { b } ( z, q ^ { 2 }, \mu _ { R } ^ { 2 }, \mu _ { F } ^ { 2 } ) = \Delta _ { b } ^ { s v } ( z, q ^ { 2 }, \mu _ { R } ^ { 2 }, \mu _ { F } ^ { 2 } ) + \Delta _ { a c } ^ { b, R } ( z, q ^ { 2 }, \mu _ { R } ^ { 2 }, \mu _ { F } ^ { 2 } )$$
where ∆ sv b contains only distributions such as δ (1 -z ) and D i , often called threshold contributions and the second term contains regular terms in z . Following [22], the threshold resumed cross section at the partonic level in d = 4 + glyph[epsilon1] dimensions is given by
$$\Delta _ { b } ^ { s v } ( z, q ^ { 2 }, \mu _ { R } ^ { 2 }, \mu _ { F } ^ { 2 } ) = \mathcal { C } \exp \left ( \Psi ^ { b } ( z, q ^ { 2 }, \mu _ { R } ^ { 2 }, \mu _ { F } ^ { 2 }, \epsilon ) \right ) \Big | _ { \epsilon = 0 }$$
where the scaling variable z = q 2 /s ˆ and Ψ ( b z, q 2 , µ 2 R , µ 2 F , glyph[epsilon1] ) is a finite distribution. The above resummed expression follows from factorization properties of the inclusive cross section and Sudakov resummation of soft gluons in the QCD amplitudes. The symbol C implies convolution with the following expansion
$$\mathcal { C } e ^ { f ( z ) } = \delta ( 1 - z ) + \frac { 1 } { 1! } f ( z ) + \frac { 1 } { 2! } f ( z ) \otimes f ( z ) + \dots$$
Here ⊗ means Mellin convolution and f ( z ) is a distribution of the kind δ (1 -z ) and D i . We drop all the regular terms in z in the evaluation of threshold contribution ∆ sv b .
In d = 4 + glyph[epsilon1] dimensions, the distribution Ψ b receives contributions from UV renormalization constant Z b for Yukawa coupling λ , the Higgs form factor ˆ F b from bottom anti-bottom, soft gluon distribution Φ b from the real radiations in partonic subprocesses and the mass factorization kernels Γ bb that remove the collinear singularities from the initial state bottom quark states. That is
$$\Psi ^ { b } ( z, q ^ { 2 }, \mu _ { R } ^ { 2 }, \mu _ { F } ^ { 2 }, \epsilon ) = & \left ( \ln \left [ Z ^ { b } ( \hat { a } _ { s }, \mu _ { R } ^ { 2 }, \mu ^ { 2 }, \epsilon ) \right ] ^ { 2 } + \ln \left | \hat { F } ^ { b } ( \hat { a } _ { s }, Q ^ { 2 }, \mu ^ { 2 }, \epsilon ) \right | ^ { 2 } \right ) \delta ( 1 - z ) \quad \ \ ( 2. 7 ) \\ & + 2 \Phi ^ { b } ( \hat { a } _ { s }, q ^ { 2 }, \mu ^ { 2 }, z, \epsilon ) - 2 C \ln \Gamma _ { b b } ( \hat { a } _ { s }, \mu ^ { 2 }, \mu _ { F } ^ { 2 }, z, \epsilon ) \,.$$
While the individual contributions are divergent, the sum is a finite distribution. The scale µ is introduced to define the dimensionless coupling constant ˆ a s = ˆ g / 2 s 16 π 2 in dimensional regularization and Q 2 = -q 2 . The renormalized strong coupling constant a s ( µ 2 R ) is related to bare ˆ a s through strong coupling constant renormalization Z µ ( 2 R ),
$$\hat { a } _ { s } = \left ( \frac { \mu } { \mu _ { R } } \right ) ^ { \epsilon } Z ( \mu _ { R } ^ { 2 } ) S _ { \epsilon } ^ { - 1 } a _ { s } ( \mu _ { R } ^ { 2 } ), \quad S _ { \epsilon } = \exp \left ( ( \gamma _ { E } - \ln 4 \pi ) \frac { \epsilon } { 2 } \right ).$$
where γ E is Euler-Mascheroni constant. In 4 + glyph[epsilon1] dimensions, Z µ ( 2 R ) can be expressed in terms of the coefficients β i of the β function of the strong coupling RG equation. Solving the RGE, we obtain, up to three loop level
$$Z ( \mu _ { R } ^ { 2 } ) = 1 + a _ { s } ( \mu _ { R } ^ { 2 } ) \frac { 2 \beta _ { 0 } } { \epsilon } + a _ { s } ^ { 2 } ( \mu _ { R } ^ { 2 } ) \left ( \frac { 4 \beta _ { 0 } ^ { 2 } } { \epsilon ^ { 2 } } + \frac { \beta _ { 1 } } { \epsilon } \right ) + a _ { s } ^ { 3 } ( \mu _ { R } ^ { 2 } ) \left ( \frac { 8 \beta _ { 0 } ^ { 3 } } { \epsilon ^ { 3 } } + \frac { 1 4 \beta _ { 0 } \beta _ { 1 } } { 3 \epsilon ^ { 2 } } + \frac { 2 \beta _ { 2 } } { 3 \epsilon } \right ). \, ( 2. 1 0 )$$
The coefficients β 0 , β 1 and β 2 are
$$\text{icients } \beta _ { 0 }, \, \beta _ { 1 } \text{ and } \beta _ { 2 } \text{ are } \\ \beta _ { 0 } & = \frac { 1 1 } { 3 } C _ { A } - \frac { 4 } { 3 } T _ { F } n _ { f } \,, \\ \beta _ { 1 } & = \frac { 3 4 } { 3 } C _ { A } ^ { 2 } - 4 T _ { F } n _ { f } C _ { F } - \frac { 2 0 } { 3 } T _ { F } n _ { f } C _ { A } \,, \\ \beta _ { 2 } & = \frac { 2 8 5 7 } { 5 4 } C _ { A } ^ { 3 } - \frac { 1 4 1 5 } { 2 7 } C _ { A } ^ { 2 } T _ { F } n _ { f } + \frac { 1 5 8 } { 2 7 } C _ { A } T _ { F } ^ { 2 } n _ { f } ^ { 2 } \\ & + \frac { 4 4 } { 9 } C _ { F } T _ { F } ^ { 2 } n _ { f } ^ { 2 } - \frac { 2 0 5 } { 9 } C _ { F } C _ { A } T _ { F } n _ { f } + 2 C _ { F } ^ { 2 } T _ { F } n _ { f } \\ > \, S U ( N ) \, \text{QCD color factors are given by}$$
where the SU N ( ) QCD color factors are given by
$$C _ { A } = N, \quad \ \ C _ { F } = \frac { N ^ { 2 } - 1 } { 2 N }, \quad \ \ T _ { F } = \frac { 1 } { 2 }$$
and n f is the number of active flavours.
The UV renormalization constant Z b (ˆ a , µ s 2 R , µ 2 , glyph[epsilon1] ) for Yukawa coupling λ satisfies
$$\mu _ { R } ^ { 2 } \frac { d } { d \mu _ { R } ^ { 2 } } \ln Z ^ { b } ( \hat { a } _ { s }, \mu _ { R } ^ { 2 }, \mu ^ { 2 }, \epsilon ) = \sum _ { i = 1 } ^ { \infty } a _ { s } ^ { i } ( \mu _ { R } ^ { 2 } ) \gamma _ { i - 1 } ^ { b } \,.$$
The above renormalization group equation (RGE) can be solved in 4 + glyph[epsilon1] dimensions to obtain up to O ( a 3 s ) level:
$$Z ^ { b } ( \hat { a } _ { s }, \mu _ { R } ^ { 2 }, \mu ^ { 2 }, \epsilon ) = 1 + \hat { a } _ { s } \left ( \frac { \mu _ { R } ^ { 2 } } { \mu ^ { 2 } } \right ) ^ { \frac { \epsilon } { 2 } } S _ { \epsilon } \left [ \frac { 1 } { \epsilon } \left ( 2 \, \gamma _ { 0 } ^ { b } \right ) \right ] + \hat { a } _ { s } ^ { 2 } \left ( \frac { \mu _ { R } ^ { 2 } } { \mu ^ { 2 } } \right ) ^ { \epsilon } S _ { \epsilon } ^ { 2 } \left [ \frac { 1 } { \epsilon ^ { 2 } } \left ( 2 \, \left ( \gamma _ { 0 } ^ { b } \right ) ^ { 2 } - 2 \, \beta _ { 0 } \, \gamma _ { 0 } ^ { b } \right )$$
$$& + \frac { 1 } { \epsilon } \left ( \gamma _ { 1 } ^ { b } \right ) \right ] + \hat { a } _ { s } ^ { 3 } \left ( \frac { \mu _ { R } ^ { 2 } } { \mu ^ { 2 } } \right ) ^ { 3 \frac { c } { 2 } } S _ { \epsilon } ^ { 3 } \left [ \frac { 1 } { \epsilon ^ { 3 } } \left ( \frac { 4 } { 3 } \, \left ( \gamma _ { 0 } ^ { b } \right ) ^ { 3 } - 4 \, \beta _ { 0 } \, \left ( \gamma _ { 0 } ^ { b } \right ) ^ { 2 } + \frac { 8 } { 3 } \, \beta _ { 0 } ^ { 2 } \, \gamma _ { 0 } ^ { b } \right ) \\ & + \frac { 1 } { \epsilon ^ { 2 } } \left ( 2 \, \gamma _ { 0 } ^ { b } \, \gamma _ { 1 } ^ { b } - \frac { 2 } { 3 } \, \beta _ { 1 } \, \gamma _ { 0 } ^ { b } - \frac { 8 } { 3 } \, \beta _ { 0 } \, \gamma _ { 1 } ^ { b } \right ) + \frac { 1 } { \epsilon } \left ( \frac { 2 } { 3 } \, \gamma _ { 2 } ^ { b } \right ) \right ]$$
where the anomalous dimensions γ b i can be obtained from the quark mass anomalous dimensions [27]
$$\text{missions} \left [ \int \right ] \\ \gamma _ { 0 } ^ { b } & = 3 C _ { F } \,, \\ \gamma _ { 1 } ^ { b } & = \frac { 3 } { 2 } C _ { F } ^ { 2 } + \frac { 9 7 } { 6 } C _ { F } C _ { A } - \frac { 1 0 } { 3 } C _ { F } T _ { F } n _ { f } \,, \\ \gamma _ { 2 } ^ { b } & = \frac { 1 2 9 } { 2 } C _ { F } ^ { 3 } - \frac { 1 2 9 } { 4 } C _ { F } ^ { 2 } C _ { A } + \frac { 1 1 4 1 3 } { 1 0 8 } C _ { F } C _ { A } ^ { 2 } + \left ( \, - 4 6 + 4 8 \zeta _ { 3 } \right ) C _ { F } ^ { 2 } T _ { F } n _ { f } \\ & + \left ( - \frac { 5 5 6 } { 2 7 } - 4 8 \zeta _ { 3 } \right ) C _ { F } C _ { A } T _ { F } n _ { f } - \frac { 1 4 0 } { 2 7 } C _ { F } T _ { F } ^ { 2 } n _ { f } ^ { 2 } \,. \\ \text{TK} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \times \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{K} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad \text{N} \quad$$
The bare form factor ˆ F b (ˆ a , Q , µ s 2 2 , glyph[epsilon1] ) satisfies the following differential equation [28-31]
$$Q ^ { 2 } \frac { d } { d Q ^ { 2 } } \ln \hat { F } ^ { b } = \frac { 1 } { 2 } \left [ K ^ { b } ( \hat { a } _ { s }, \frac { \mu _ { R } ^ { 2 } } { \mu ^ { 2 } }, \epsilon ) + G ^ { b } ( \hat { a } _ { s }, \frac { Q ^ { 2 } } { \mu _ { R } ^ { 2 } }, \frac { \mu _ { R } ^ { 2 } } { \mu ^ { 2 } }, \epsilon ) \right ]$$
where K b contains all the poles in glyph[epsilon1] and G b contains the terms finite in glyph[epsilon1] . Renormalization group invariance of ˆ F b (ˆ a , Q , µ s 2 2 , glyph[epsilon1] ) gives
$$\mu _ { R } ^ { 2 } \frac { d } { d \mu _ { R } ^ { 2 } } K ^ { b } = - \mu _ { R } ^ { 2 } \frac { d } { d \mu _ { R } ^ { 2 } } G ^ { b } = - \sum _ { i = 1 } ^ { \infty } a _ { s } ^ { i } ( \mu _ { R } ^ { 2 } ) A _ { i } ^ { q } \,.$$
where A q i 's are the cusp anomalous dimensions [18, 32] given by
$$\text{where } \alpha _ { i } & s \, \text{are use} \, \text{using} \, \text{in} \, \text{in} \, \text{or} \, \text{given} \, \text{by} \\ A _ { 1 } ^ { q } & = 4 C _ { F } \,, \\ A _ { 2 } ^ { q } & = 8 C _ { F } C _ { A } \left \{ \frac { 6 7 } { 1 8 } - \zeta _ { 2 } \right \} + 8 C _ { F } n _ { f } \left \{ - \frac { 5 } { 9 } \right \}, \\ A _ { 3 } ^ { q } & = 1 6 C _ { F } C _ { A } ^ { 2 } \left \{ \frac { 2 4 5 } { 2 4 } - \frac { 6 7 \zeta _ { 2 } } { 9 } + \frac { 1 1 \zeta _ { 3 } } { 6 } + \frac { 1 1 \zeta _ { 2 } ^ { 2 } } { 5 } \right \} + 1 6 C _ { F } ^ { 2 } n _ { f } \left \{ - \frac { 5 5 } { 2 4 } + 2 \zeta _ { 3 } \right \} \\ & + 1 6 C _ { F } C _ { A } n _ { f } \left \{ - \frac { 2 0 9 } { 1 0 8 } + \frac { 1 0 \zeta _ { 2 } } { 9 } - \frac { 7 \zeta _ { 3 } } { 3 } \right \} + 1 6 C _ { F } n _ { f } ^ { 2 } \left \{ - \frac { 1 } { 2 7 } \right \}. \\ \text{Enanding} \, \text{$u_{2}$ independent part of the solution of RG$ equation for $G^{b}$ as}$$
Expanding µ 2 R independent part of the solution of RG equation for G b as
$$G ^ { b } ( a _ { s } ( Q ^ { 2 } ), 1, \epsilon ) = \sum _ { i = 1 } ^ { \infty } a _ { s } ^ { i } ( Q ^ { 2 } ) G _ { i } ^ { b } ( \epsilon ),$$
one finds that G b i can be decomposed in terms of collinear B q i and soft f q i anomalous dimensions as follows [33-35]
$$G _ { i } ^ { b } ( \epsilon ) = 2 ( B _ { i } ^ { q } - \gamma _ { i } ^ { b } ) + f _ { i } ^ { q } + C _ { i } ^ { b } + \sum _ { k = 1 } ^ { \infty } \epsilon ^ { k } g _ { i } ^ { b, k } \,.$$
The collinear anomalous dimensions B q i [18] are given by
$$& \text{The collinear anomalous dimensions } B _ { i } ^ { q } \left [ 1 8 \right ] \text{are given by} \\ & B _ { 2 } ^ { q } = 3 C _ { F } \,, \\ & B _ { 2 } ^ { q } = \frac { 1 } { 2 } \left ( C _ { F } ^ { 2 } \left \{ 3 - 2 4 \zeta _ { 2 } + 4 8 \zeta _ { 3 } \right \} + C _ { A } C _ { F } \left \{ \frac { 1 7 } { 3 } + \frac { 8 8 } { 3 } \zeta _ { 2 } - 2 4 \zeta _ { 3 } \right \} + n _ { f } T _ { F } C _ { F } \left \{ - \frac { 4 } { 3 } - \frac { 3 2 } { 3 } \zeta _ { 2 } \right \} \right ) \,, \\ & B _ { 3 } ^ { q } = - 1 6 C _ { A } ^ { 2 } C _ { F } \left \{ \frac { 1 } { 8 } \zeta _ { 2 } ^ { 2 } - \frac { 2 8 1 } { 2 7 } \zeta _ { 2 } + \frac { 9 7 } { 9 } \zeta _ { 3 } - \frac { 5 } { 2 } \zeta _ { 5 } + \frac { 1 6 5 7 } { 5 7 6 } \right \} + 1 6 C _ { A } C _ { F } ^ { 2 } \left \{ - \frac { 2 4 7 } { 6 0 } \zeta _ { 2 } ^ { 2 } + \zeta _ { 2 } \zeta _ { 3 } \\ & \quad - \frac { 2 0 5 } { 2 4 } \zeta _ { 2 } + \frac { 2 1 1 } { 1 2 } \zeta _ { 3 } + \frac { 1 5 } { 2 } \zeta _ { 5 } + \frac { 1 5 1 } { 6 4 } \right \} + 1 6 C _ { A } C _ { F } n _ { f } \left \{ \frac { 1 } { 2 0 } \zeta _ { 2 } ^ { 2 } - \frac { 1 6 7 } { 5 4 } \zeta _ { 2 } + \frac { 2 5 } { 1 8 } \zeta _ { 3 } + \frac { 5 } { 4 } \right \} \\ & \quad + 1 6 C _ { F } ^ { 3 } \left \{ \frac { 1 8 } { 5 } \zeta _ { 2 } ^ { 2 } - 2 \zeta _ { 2 } \zeta _ { 3 } + \frac { 9 } { 8 } \zeta _ { 2 } + \frac { 1 7 } { 4 } \zeta _ { 3 } - 1 5 \zeta _ { 5 } + \frac { 2 9 } { 3 2 } \right \} \\ & \quad - 1 6 C _ { F } ^ { 2 } n _ { f } \left \{ - \frac { 2 9 } { 3 0 } \zeta _ { 2 } ^ { 2 } - \frac { 5 } { 1 2 } \zeta _ { 2 } + \frac { 1 7 } { 6 } \zeta _ { 3 } + \frac { 2 3 } { 1 6 } \right \} - 1 6 C _ { F } n _ { f } \left \{ - \frac { 5 } { 2 7 } \zeta _ { 2 } + \frac { 1 } { 9 } \zeta _ { 3 } + \frac { 1 7 } { 1 4 4 } \right \}, \\ & \text{the soft anomalous dimensions } f _ { i } ^ { q } \left [ 3 3 \right ] \text{are}$$
the soft anomalous dimensions f q i [33] are
$$& \text{the soft anomalous dimensions } f _ { i } ^ { q } \left [ 3 3 \right ] \text{ are} \\ & f _ { 1 } ^ { q } = 0 \,, \\ & f _ { 2 } ^ { q } = C _ { A } C _ { F } \left \{ - \frac { 2 2 } { 3 } \zeta _ { 2 } - 2 8 \zeta _ { 3 } + \frac { 8 0 8 } { 2 7 } \right \} + C _ { F } n _ { f } T _ { F } \left \{ \frac { 8 } { 3 } \zeta _ { 2 } - \frac { 2 2 4 } { 2 7 } \right \}, \\ & f _ { 3 } ^ { q } = C _ { A } ^ { 2 } C _ { F } \left \{ \frac { 3 5 } { 5 } \zeta _ { 2 } ^ { 2 } + \frac { 1 7 6 } { 3 } \zeta _ { 2 } \zeta _ { 3 } - \frac { 1 2 6 5 0 } { 8 1 } \zeta _ { 2 } - \frac { 1 3 1 6 } { 3 } \zeta _ { 3 } + 1 9 2 \zeta _ { 5 } + \frac { 1 3 6 7 8 1 } { 7 2 9 } \right \} \\ & \quad + C _ { A } C _ { F } n _ { f } \left \{ - \frac { 9 6 } { 5 } \zeta _ { 2 } ^ { 2 } + \frac { 2 8 2 8 } { 8 1 } \zeta _ { 2 } + \frac { 7 2 8 } { 2 7 } \zeta _ { 3 } - \frac { 1 1 8 4 2 } { 7 2 9 } \right \} \\ & \quad + C _ { F } ^ { 2 } n _ { f } \left \{ \frac { 3 2 } { 5 } \zeta _ { 2 } ^ { 2 } + 4 \zeta _ { 2 } + \frac { 3 0 4 } { 9 } \zeta _ { 3 } - \frac { 1 7 1 1 } { 2 7 } \right \} + C _ { F } n _ { f } ^ { 2 } \left \{ - \frac { 4 0 } { 2 7 } \zeta _ { 2 } + \frac { 1 1 2 } { 2 7 } \zeta _ { 3 } - \frac { 2 0 8 0 } { 7 2 9 } \right \} \quad ( 2. 2 0 ) \\ & \text{and the constants } C _ { i } ^ { p } \text{ are given by}$$
and the constants C b i are given by
$$C _ { 1 } ^ { b } = 0, \quad C _ { 2 } ^ { b } = - 2 \beta _ { 0 } g _ { 1 } ^ { b, 1 }, \quad C _ { 3 } ^ { b } = - 2 \beta _ { 1 } g _ { 1 } ^ { b, 1 } - 2 \beta _ { 0 } ( g _ { 2 } ^ { b, 1 } + 2 \beta _ { 0 } g _ { 1 } ^ { b, 2 } ). \quad ( 2. 2 1 )$$
Since G b 1 ( glyph[epsilon1] ) and G b 2 ( glyph[epsilon1] ) are known to all orders in glyph[epsilon1] and G b 3 ( glyph[epsilon1] ) is known to O ( glyph[epsilon1] 3 ) [26], the coefficients g b,k i for i = 1 2 3 can be readily obtained from , , G glyph[epsilon1] b i ( ). The relevant one loop terms are found to be
$$g _ { 1 } ^ { b, 1 } = C _ { F } ( - 2 + \zeta _ { 2 } ), \quad g _ { 1 } ^ { b, 2 } = C _ { F } ( 2 - \frac { 7 } { 3 } \zeta _ { 3 } ), \quad g _ { 1 } ^ { b, 3 } = C _ { F } ( - 2 + \frac { 1 } { 4 } \zeta _ { 2 } + \frac { 4 7 } { 8 0 } \zeta _ { 2 } ^ { 2 } ) \,, \quad ( 2. 2 2 )$$
the relevant two loop terms [10, 22] are
$$g _ { 2 } ^ { b, 1 } & = C _ { F } n _ { f } \left \{ \frac { 6 1 6 } { 8 1 } + \frac { 1 0 } { 9 } \zeta _ { 2 } - \frac { 8 } { 3 } \zeta _ { 3 } \right \} + C _ { F } C _ { A } \left \{ - \frac { 2 1 2 2 } { 8 1 } - \frac { 1 0 3 } { 9 } \zeta _ { 2 } + \frac { 8 8 } { 5 } \zeta _ { 2 } ^ { 2 } + \frac { 1 5 2 } { 3 } \zeta _ { 3 } \right \} \\ & \quad + C _ { F } ^ { 2 } \left \{ 8 + 3 2 \zeta _ { 2 } - \frac { 8 8 } { 5 } \zeta _ { 2 } ^ { 2 } - 6 0 \zeta _ { 3 } \right \}, \\ g _ { 2 } ^ { b, 2 } & = C _ { F } n _ { f } \left \{ \frac { 7 } { 1 2 } \zeta _ { 2 } ^ { 2 } - \frac { 5 5 } { 2 7 } \zeta _ { 2 } + \frac { 1 3 0 } { 2 7 } \zeta _ { 3 } - \frac { 3 1 0 0 } { 2 4 3 } \right \} + C _ { A } C _ { F } \left \{ - \frac { 3 6 5 } { 2 4 } \zeta _ { 2 } ^ { 2 } + \frac { 8 9 } { 3 } \zeta _ { 2 } \zeta _ { 3 } + \frac { 1 0 7 9 } { 5 4 } \zeta _ { 2 } \\ & \quad - \frac { 2 9 2 3 } { 2 7 } \zeta _ { 3 } - 5 1 \zeta _ { 5 } + \frac { 9 1 4 2 } { 2 4 3 } \right \} + C _ { F } ^ { 2 } \left \{ \frac { 9 6 } { 5 } \zeta _ { 2 } ^ { 2 } - 2 8 \zeta _ { 2 } \zeta _ { 3 } - 4 4 \zeta _ { 2 } + 1 1 6 \zeta _ { 3 } + 1 2 \zeta _ { 5 } - 2 4 \right \}$$
and finally the relevant three loop term [26] is
$$\text{and finally the relevant three loop term [26] is} \\ \frac { b _ { 1 } } { 3 } = & \, C _ { A } ^ { 2 } C _ { F } \left \{ - \frac { 6 1 5 2 } { 6 3 } \, \zeta _ { 2 } ^ { 3 } \, + \frac { 2 7 3 8 } { 9 } \, \zeta _ { 2 } ^ { 2 } \, + \frac { 9 7 6 } { 9 } \, \zeta _ { 2 } \zeta _ { 3 } - \frac { 3 4 2 2 6 3 } { 4 8 6 } \, \zeta _ { 2 } ^ { 2 } \, - \frac { 1 1 3 6 } { 3 } \, \zeta _ { 3 } ^ { 2 } \, + \frac { 1 9 5 8 2 } { 9 } \, \zeta _ { 3 } \\ + \frac { 1 2 2 8 } { 3 } \, \zeta _ { 5 } + \frac { 4 0 9 5 2 6 3 } { 8 7 4 8 } \right \} + C _ { A } C _ { F } ^ { 2 } \left \{ - \frac { 1 5 4 4 8 } { 1 0 5 } \, \zeta _ { 2 } ^ { 3 } - \frac { 3 6 3 4 } { 4 5 } \, \zeta _ { 2 } ^ { 2 } - \frac { 2 5 8 4 } { 3 } \, \zeta _ { 2 } \zeta _ { 3 } + \frac { 1 3 3 5 7 } { 9 } \, \zeta _ { 2 } \\ + 2 9 6 \zeta _ { 3 } ^ { 2 } - \frac { 1 1 5 7 0 } { 9 } \, \zeta _ { 3 } - \frac { 1 9 4 0 } { 3 } \, \zeta _ { 5 } - \frac { 6 1 3 } { 3 } \right \} + C _ { A } C _ { F } n f \left \{ - \frac { 1 0 6 4 } { 4 5 } \, \zeta _ { 2 } ^ { 2 } + \frac { 3 9 2 } { 9 } \, \zeta _ { 2 } \zeta _ { 3 } + \frac { 4 4 5 5 1 } { 2 4 3 } \, \zeta _ { 2 } \\ - \frac { 4 1 5 5 2 } { 8 1 } \, \zeta _ { 3 } - 7 2 \zeta _ { 5 } - \frac { 6 1 1 9 } { 4 3 7 4 } \right \} + C _ { F } ^ { 2 } n f \left \{ 7 7 2 4 5 \, \zeta _ { 2 } ^ { 2 } - \frac { 1 5 2 } { 3 } \, \zeta _ { 2 } \zeta _ { 3 } - \frac { 3 1 7 3 } { 1 8 } \, \zeta _ { 2 } + \frac { 1 5 9 5 6 } { 2 7 } \, \zeta _ { 3 } - \frac { 3 6 8 } { 3 } \, \zeta _ { 5 } \\ + \frac { 3 2 8 9 9 } { 3 2 4 } \right \} + C _ { F } n f _ { f } ^ { 2 } \left \{ - \frac { 4 0 } { 9 } \, \zeta _ { 2 } ^ { 2 } - \frac { 8 9 2 } { 8 1 } \, \zeta _ { 2 } + \frac { 3 2 0 } { 8 1 } \, \zeta _ { 3 } - \frac { 2 7 3 5 2 } { 2 1 8 7 } \right \} + C _ { F } ^ { 2 } \left \{ 2 1 5 8 4 \, \zeta _ { 2 } ^ { 3 } - \frac { 1 6 4 4 } { 1 0 5 } \, \zeta _ { 2 } ^ { 3 } \\ + \frac { 6 2 4 \zeta _ { 2 } \zeta _ { 3 } - 2 7 5 \zeta _ { 2 } + 4 8 \zeta _ { 3 } ^ { 2 } - 2 1 4 2 \zeta _ { 3 } + 1 2 7 2 \zeta _ { 5 } + 6 0 3 } \right \}. \\ \text{The mass factorization kernel } \Gamma _ { b b } ( z, \mu _ { F }, \epsilon ) \text{ removes the collinear singularities resulting from}$$
The mass factorization kernel Γ bb ( z, µ 2 F , glyph[epsilon1] ) removes the collinear singularities resulting from massless partons. It satisfies the following RG equation
$$\mu _ { F } ^ { 2 } \frac { d } { d \mu _ { F } ^ { 2 } } \Gamma _ { b b } ( z, \mu _ { F } ^ { 2 }, \epsilon ) = \frac { 1 } { 2 } \sum _ { c } P _ { b c } \left ( z, \mu _ { F } ^ { 2 } \right ) \otimes \Gamma _ { c b } \left ( z, \mu _ { F } ^ { 2 }, \epsilon \right ) \,,$$
where P bc ( z, µ 2 F ) are Altarelli-Parisi splitting functions. In perturbative QCD, we can expand them as
$$P _ { b c } ( z, \mu _ { F } ^ { 2 } ) = \sum _ { i = 1 } ^ { \infty } a _ { s } ^ { i } ( \mu _ { F } ^ { 2 } ) P _ { b c } ^ { ( i - 1 ) } ( z ).$$
The off diagonal splitting functions are regular as z → 1. On the other hand diagonal ones contain δ (1 -z ) and D 0 as well as regular terms i.e.
$$P _ { b b } ^ { ( i ) } ( z ) = 2 ( B _ { i + 1 } ^ { b } \delta ( 1 - z ) + A _ { i + 1 } ^ { b } \mathcal { D } _ { 0 } ) + P _ { r e g, b b } ^ { ( i ) } ( z ) \,.$$
We find that the regular part of the splitting function, P ( ) i reg,bb , does not contribute to threshold corrections.
The fact that ∆ sv b is finite as glyph[epsilon1] → 0 implies that soft distribution function Φ (ˆ b a , q s 2 , µ 2 , z, glyph[epsilon1] ) also satisfies Sudakov type differential equations [22] namely
$$q ^ { 2 } \frac { d } { d q ^ { 2 } } \Phi ^ { b } = \frac { 1 } { 2 } \left [ \overline { K } ^ { b } ( \hat { a } _ { s }, \frac { \mu _ { R } ^ { 2 } } { \mu ^ { 2 } }, z, \epsilon ) + \overline { G } ^ { b } ( \hat { a } _ { s }, \frac { q ^ { 2 } } { \mu _ { R } ^ { 2 } }, \frac { \mu _ { R } ^ { 2 } } { \mu ^ { 2 } }, z, \epsilon ) \right ].$$
K b and G b take the forms similar to those of K b and G b of the form factors. This guarantees Ψ b is finite order by order in perturbation theory as glyph[epsilon1] → 0. The solution to the above equation is found to be
$$\Phi ^ { b } = \sum _ { i = 1 } ^ { \infty } \hat { a } _ { s } ^ { i } \left ( \frac { q ^ { 2 } ( 1 - z ) ^ { 2 } } { \mu ^ { 2 } } \right ) ^ { i \frac { \epsilon } { 2 } } S _ { \epsilon } ^ { i } \left ( \frac { i \epsilon } { 1 - z } \right ) \hat { \phi } ^ { b, ( i ) } ( \epsilon )$$
where
$$\hat { \phi } ^ { b, ( i ) } ( \epsilon ) = \frac { 1 } { i \epsilon } \left [ \overline { K } ^ { b, ( i ) } ( \epsilon ) + \overline { G } ^ { b, ( i ) } ( \epsilon ) \right ]$$
and the constants K b can be obtained using
$$\mu _ { R } ^ { 2 } \frac { d \overline { K } ^ { b } } { d \mu _ { R } ^ { 2 } } = - \delta ( 1 - z ) \mu _ { R } ^ { 2 } \frac { d K ^ { b } } { d \mu _ { R } ^ { 2 } } \,.$$
This implies that K b, ( ) i ( glyph[epsilon1] ) can be written in terms of A q i and β i . Defining G b i ( glyph[epsilon1] ) through
$$\sum _ { i = 1 } ^ { \infty } \hat { a } _ { s } ^ { i } \left ( \frac { q _ { z } ^ { 2 } } { \mu ^ { 2 } } \right ) ^ { i \frac { \epsilon } { 2 } } S _ { \epsilon } ^ { i } \overline { G } ^ { b, ( i ) } ( \epsilon ) = \sum _ { i = 1 } ^ { \infty } a _ { s } ^ { i } ( q _ { z } ^ { 2 } ) \overline { \mathcal { G } } _ { i } ^ { b } ( \epsilon )$$
where q 2 z = q 2 (1 -z ) 2 and using the fact that ∆ sv b is finite as glyph[epsilon1] → 0, we can express G b i ( glyph[epsilon1] ) as
$$\overline { \mathcal { G } } _ { 1 } ^ { \ b } ( \epsilon ) & = - f _ { 1 } ^ { q } + \sum _ { k = 1 } ^ { \infty } \epsilon ^ { k } \overline { \mathcal { G } } _ { 1 } ^ { \ b, ( k ) } \,, \\ \overline { \mathcal { G } } _ { 2 } ^ { \ b } ( \epsilon ) & = - f _ { 2 } ^ { q } - 2 \beta _ { 0 } \overline { \mathcal { G } } _ { 1 } ^ { \ b, ( 1 ) } + \sum _ { k = 1 } ^ { \infty } \epsilon ^ { k } \overline { \mathcal { G } } _ { 2 } ^ { \ b, ( k ) } \,, \\ \overline { \mathcal { G } } _ { 3 } ^ { \ b } ( \epsilon ) & = - f _ { 3 } ^ { q } - 2 \beta _ { 1 } \overline { \mathcal { G } } _ { 1 } ^ { \ b, ( 1 ) } - 2 \beta _ { 0 } \left ( \overline { \mathcal { G } } _ { 2 } ^ { \ b, ( 1 ) } + 2 \beta _ { 0 } \overline { \mathcal { G } } _ { 1 } ^ { \ b, ( 2 ) } \right ) + \sum _ { k = 1 } ^ { \infty } \epsilon ^ { k } \overline { \mathcal { G } } _ { 3 } ^ { \ b, ( k ) } \,. \\ \tau _ { 1 } \dots \dots \dots \dots \overline { \overline { \overline { \overline { \nu } } } ^ { b, j } \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \tau _ { 1 } \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots.$$
The constants G b,j i arise from the soft part of the partonic reactions. Since the soft part does not depend on the hard process, G b i ( glyph[epsilon1] ) can be directly obtained from G q i ( glyph[epsilon1] ) that contributes to Drell-Yan production [22]
$$\overline { \mathcal { G } } _ { i } ^ { b } ( \epsilon ) = \overline { \mathcal { G } } _ { i } ^ { q } ( \epsilon ) \,.$$
We list the relevant ones that contribute up to N LO level. 3 The terms required for one and two loops [22] are
$$\text{and two loops [22] are} \\ \widetilde { \mathcal { G } } _ { 1 } ^ { q, ( 1 ) } & = C _ { F } ( - 3 \zeta _ { 2 } ) \,, \\ \widetilde { \mathcal { G } } _ { 1 } ^ { q, ( 2 ) } & = C _ { F } ( \frac { 7 } { 3 } \zeta _ { 3 } ) \,, \\ \widetilde { \mathcal { G } } _ { 1 } ^ { q, ( 3 ) } & = C _ { F } ( - \frac { 3 } { 1 6 } \zeta _ { 2 } ^ { 2 } ) \,, \\ \widetilde { \mathcal { G } } _ { 2 } ^ { q, ( 1 ) } & = C _ { F } n _ { f } \left ( - \frac { 3 2 8 } { 8 1 } + \frac { 7 0 } { 9 } \zeta _ { 2 } + \frac { 3 2 } { 3 } \zeta _ { 3 } \right ) + C _ { A } C _ { F } \left ( \frac { 2 4 2 8 } { 8 1 } - \frac { 4 6 9 } { 9 } \zeta _ { 2 } + 4 \zeta _ { 2 } ^ { 2 } - \frac { 1 7 6 } { 3 } \zeta _ { 3 } \right ) \,, \\ \widetilde { \mathcal { G } } _ { 2 } ^ { q, ( 2 ) } & = C _ { A } C _ { F } \left ( \frac { 1 1 } { 4 0 } \zeta _ { 2 } ^ { 2 } - \frac { 2 0 3 } { 3 } \zeta _ { 2 } \zeta _ { 3 } + \frac { 1 4 1 4 } { 2 7 } \zeta _ { 2 } + \frac { 2 0 7 7 } { 2 7 } \zeta _ { 3 } + 4 3 \zeta _ { 5 } - \frac { 7 2 8 8 } { 2 4 3 } \right ) \\ & + C _ { F } n _ { f } \left ( - \frac { 1 } { 2 0 } \zeta _ { 2 } ^ { 2 } - \frac { 1 9 6 } { 2 7 } \zeta _ { 2 } - \frac { 3 1 0 } { 2 7 } \zeta _ { 3 } + \frac { 9 7 6 } { 2 4 3 } \right )$$
and for three loops [24]
$$\text{and for three loops } [ 2 4 ] \\ \overline { \mathcal { G } } _ { 3 } ^ { q, ( 1 ) } & = C _ { F } \Big { \{ } C _ { A } ^ { 2 } \Big { ( } \frac { 1 5 2 } { 6 3 } \ \zeta _ { 2 } ^ { 3 } + \frac { 1 9 6 4 } { 9 } \ \zeta _ { 2 } ^ { 2 } + \frac { 1 1 0 0 0 } { 9 } \ \zeta _ { 2 } \zeta _ { 3 } - \frac { 7 6 5 1 2 7 } { 4 8 6 } \ \zeta _ { 2 } + \frac { 5 3 6 } { 3 } \ \zeta _ { 3 } ^ { 2 } - \frac { 5 9 6 4 8 } { 2 7 } \ \zeta _ { 3 } \\ & - \frac { 1 4 3 0 } { 3 } \ \zeta _ { 5 } + \frac { 7 1 3 5 9 8 1 } { 8 7 4 8 } \Big { ) } + C _ { A } n _ { f } \Big { ( } - \frac { 5 3 2 } { 9 } \ \zeta _ { 2 } ^ { 2 } - \frac { 1 2 0 8 } { 9 } \ \zeta _ { 2 } \zeta _ { 3 } + \frac { 1 0 5 0 5 9 } { 2 4 3 } \ \zeta _ { 2 } + \frac { 4 5 9 5 6 } { 8 1 } \ \zeta _ { 3 } \\ & + \frac { 1 4 8 } { 3 } \ \zeta _ { 5 } - \frac { 7 1 6 5 0 9 } { 4 3 7 4 } \Big { ) } + C _ { F } n _ { f } \Big { ( } \frac { 1 5 2 } { 1 5 } \ \zeta _ { 2 } ^ { 2 } - 8 8 \ \zeta _ { 2 } \zeta _ { 3 } + \frac { 6 0 5 } { 6 } \ \zeta _ { 2 } + \frac { 2 5 3 6 } { 2 7 } \ \zeta _ { 3 } + \frac { 1 1 2 } { 3 } \ \zeta _ { 5 } \\ & - \frac { 4 2 7 2 7 } { 3 2 4 } \Big { ) } + n _ { f } ^ { 2 } \ \Big { ( } \frac { 3 2 } { 9 } \ \zeta _ { 2 } ^ { 2 } - \frac { 1 9 9 6 } { 8 1 } \ \zeta _ { 2 } - \frac { 2 7 2 0 } { 8 1 } \ \zeta _ { 3 } + \frac { 1 1 5 8 4 } { 2 1 8 7 } \Big { ) } \Big { \} }.$$
## 3 Results
Expanding eqn.(2.5) in powers of a s ( µ 2 R ) using eqn(2.6) and performing the convolutions we find
$$\Delta _ { b } ^ { s v } ( z ) = \sum _ { i = 0 } ^ { \infty } a _ { s } ^ { i } ( \mu _ { R } ^ { 2 } ) \Delta _ { b } ^ { s v, ( i ) } ( z, \mu _ { R } ^ { 2 } )$$
where ∆ sv, i ( ) b can be expressed in terms of the distributions δ (1 -z ) and D i :
$$\Delta _ { b } ^ { s v, ( i ) } = \Delta _ { b } ^ { s v, ( i ) } ( \mu _ { R } ^ { 2 } ) | _ { \delta } \delta ( 1 - z ) + \sum _ { j = 0 } ^ { 2 i - 1 } \Delta _ { b } ^ { s v, ( i ) } ( \mu _ { R } ^ { 2 } ) | _ { \mathcal { D } _ { j } } \mathcal { D } _ {$$
The results up to N LO are given below 3
$$\Delta _ { b } \, \hat { \ } n = \Delta _ { b } \, \hat { \ } n \, ( \mu _ { R } ) | \delta 0 ( 1 - z ) \, + \, \sum _ { j = 0 } \Delta _ { b } \, \hat { \ } n \, ( \mu _ { R } ) | D _ { j } \, \mathcal { U } _ { j } \,, \quad ( 3. 2 ) \\ \quad \quad \quad$$
$$& - 4 g _ { 1 } ^ { b _ { 1 } } \left ( f _ { 1 } ^ { b _ { 1 } } \right ) ^ { 2 } + 1 2 g _ { 1 } ^ { b _ { 1 } \overline { g } _ { b } ^ { b _ { 1 } } \overline { g } _ { b } ^ { b _ { 1 } } \overline { g } _ { b } ^ { b _ { 1 } } \overline { g } _ { b } ^ { b _ {$$
In the above equation A b i = A , q i B b i = B q i , f b i = f q i and we have set µ 2 R = µ 2 F = q 2 .
The finite ∆ sv, i ( ) b ( Q 2 ) depend on the anomalous dimensions A b i , B b i , f b i and γ b i , the β function coefficients β i and glyph[epsilon1] expansion coefficients of G glyph[epsilon1] b ( ), g b,i j s and of the corresponding G b ( glyph[epsilon1] ), G b,i j s. The results for the ∆ sv, (1) b and ∆ sv, (2) b agree with those from the exact NLO and NNLO results [10]. At N LO level, only ∆ 3 sv, (3) b | D i 's were known [22] as the terms g b, 2 2 , g b, 1 3 of the form factor and G q, (2) 2 , G q, (1) 3 needed for ∆ sv, (3) b | δ were not available. The recent results for g b, 2 2 and g b, 1 3 from [26], G q, 2 2 from [20] and the N LO 3 G q, (1) 3 from [24] can be used to obtain the missing δ (1 -z ) part namely ∆ sv, (3) b | δ . This completes the evaluation of full threshold N LO contributions in QCD for Higgs production through bottom anti3
bottom annihilation at hadron colliders. Below we present our results up to N LO level 3 after substituting A , B , f , γ i i i i and β i terms as well as the constants g b,i j from the form factors and G b, ( ) i j from the soft distributions function in eqn. (3.3).
$$\begin{array} { l } \text{bottom annihilation at hadron colliders. Below we present our results up to N$^{3}$ LO level} \\ \text{after substituting } A _ { i } B _ { i } f _ { i }, \gamma \text{ and } \beta _ { i } \text{ terms as well as the constants } p _ { i } ^ { * } \text{ from the form} \\ \text{factors and } \beta _ { i } ^ { ( i ) } \text{$$
The numerical impact of our results can be studied using the exact LO, NLO, NNLO
s
Figure 1 . Total cross section for Higgs production in bb ¯ annihilation at various orders in a s as a function of µ R /m H (left panel) and of µ F /m H (right panel) at the LHC with √ s = 14 TeV.
∆ b, ( ) i ac , i = 0 1 2 and the threshold N LO result ∆ , , 3 sv, (3) bb . We have used √ s = 14 TeV for the LHC, the Z boson mass M Z = 91 1876 GeV and Higgs boson mass . m H = 125.5 GeV throughout. The strong coupling constant α s ( µ 2 R ) is evolved using the 4-loop RG equations with α N LO 3 s ( m Z ) = 0 117 and for parton density sets we use MSTW 2008NNLO [36]. The . Yukawa coupling is evolved using 4 loop RG with λ m ( b ) = √ 2 m m /ν b ( b ) and m m b ( b ) = 4 3 . GeV.
The renormalization scale dependence is studied by varying µ R between 0 1 . m H and 10 m H keeping µ F = m / H 4 fixed. For the factorization scale, we have fixed µ R = m H and varied µ F between 0 1 . m H and 10 m H . We find that the perturbation theory behaves better if we include more and more higher order terms (see Fig.1).
To summarize, we have systematically developed a framework to compute threshold contributions in QCD to the production of Higgs boson in bottom anti-bottom annihilation subprocesses at the hadron colliders. Factorization of UV, soft and collinear singularities and exponentiation of their sum allow us to obtain threshold corrections order by order in perturbation theory. Using the recently obtained N LO soft distribution function for 3 Drell-Yan production and the three loop Higgs form factor with bottom anti-bottom quarks, we have obtained threshold N LO corrections to Higgs production through bottom anti3 bottom annihilation. We have also studied the stability of our result under renormalization and factorization scales.
## Acknowledgments
We sincerely thank T. Gehrmann for constant encouragement and fruitful discussions. We also thank M. K. Mandal for useful discussions. The work of TA and NR has been partially supported by funding from RECAPP, Department of Atomic Energy, Govt. of India.
## References
- [1] G. Aad et al. [ATLAS Collaboration], Phys. Lett. B 716 (2012) 1 .
- [2] S. Chatrchyan et al. [CMS Collaboration], Phys. Lett. B 716 (2012) 30 .
- [3] R. Barate et al. [ LEP Working Group for Higgs boson searches and ALEPH and DELPHI and L3 and OPAL Collaborations ], Phys. Lett. B565 (2003) 61.
- [4] CDF and D0 Collaborations, Phys. Rev. Lett. 104 (2010) 061802.
- [5] ALEPH,CDF,D0, DELPHI,L3,OPAL,SLD Collaborations, CERN PH-EP-2010-095,(2010).
- [6] H. M. Georgi, S. L. Glashow, M. E. Machacek, and D. V. Nanopoulos, Phys. Rev. Lett. 40 (1978) 692 ; A. Djouadi, M. Spira, and P. M. Zerwas, Phys. Lett. B 264 (1991) 440 ; S. Dawson, Nucl. Phys. B359 (1991) 283 ; M. Spira, A. Djouadi, D. Graudenz, and P. M. Zerwas, Nucl. Phys. B453 (1995) 17 ; S. Catani, D. de Florian and M. Grazzini, JHEP 0105 (2001) 025 ; R. V. Harlander and W. B. Kilgore, Phys. Rev. D 64 (2001) 013015 , Phys. Rev. Lett. 88 (2002) 201801 ; C. Anastasiou and K. Melnikov, Nucl. Phys. B646 (2002) 220 ; V. Ravindran, J. Smith, and W. L. van Neerven, Nucl. Phys. B665 (2003) 325.
- [7] P. Bolzoni, F. Maltoni, S. Moch, M. Zaro, Phys. Rev. Lett. 105 (2010) 011801
- [8] T. Han and S. Willenbrock, Phys. Lett. B 273 (1991) 167; O. Brein, A. Djouadi and R. Harlander, Phys. Lett. B579 (2004) 149.
- [9] D. A. Dicus and S. Willenbrock, Phys. Rev. D 39 (1989) 751 ; D. Dicus, T. Stelzer, Z. Sullivan and S. Willenbrock, Phys. Rev. D 59 (1999) 094016 ; F. Maltoni, Z. Sullivan and S. Willenbrock, Phys. Rev. D 67 (2003) 093005 ; F. I. Olness and W. -K. Tung, Nucl. Phys. B 308 (1988) 813 ; J. F. Gunion, H. E. Haber, F. E. Paige, W. -K. Tung and S. S. D. Willenbrock, Nucl. Phys. B 294 (1987) 621 .
- [10] R. V. Harlander and W. B. Kilgore, Phys. Rev. D 68 (2003) 013001 .
- [11] L. Reina and S. Dawson, Phys. Rev. Lett. 87 (2001) 201804 ; W. Beenakker, S. Dittmaier, M. Kramer, B. Plumper, M. Spira and P. M. Zerwas, Phys. Rev. Lett. 87 (2001) 201805 ; S. Dawson, L. H. Orr, L. Reina and D. Wackeroth, Phys. Rev. D 67 (2003) 071503 ; W. Beenakker, S. Dittmaier, M. Kramer, B. Plumper, M. Spira and P. M. Zerwas, Nucl. Phys. B 653 (2003) 151 ; R. Raitio and W. W. Wada, Phys. Rev. D 19 (1979) 941 ; Z. Kunszt, Nucl. Phys. B 247 (1984) 339.
- [12] S. Catani, D. de Florian, M. Grazzini, and P. Nason, JHEP 0307 (2003) 028.
- [13] U. Aglietti, R. Bonciani, G. Degrassi and A. Vicini, Phys. Lett. B 595 (2004) 432 ; S. Actis, G. Passarino, C. Sturm and S. Uccirati, Phys. Lett. B 670 (2008) 12 . G. Altarelli, R. K. Ellis and G. Martinelli, Nucl. Phys. B 143 , 521 (1978) [Erratum-ibid. B 146 , 544 (1978)] ; T. Matsuura and W. L. van Neerven, Z. Phys. C 38 , 623 (1988) ; T. Matsuura, S. C. van der Marck and W. L. van Neerven, Nucl. Phys. B 319 , 570 (1989) ; R. Hamberg, W. L. van Neerven and T. Matsuura, Nucl. Phys. B 359 , 343 (1991) [Erratum-ibid. B 644 , 403 (2002)] .
- [14] S. Moch, J. A. M. Vermaseren and A. Vogt, JHEP 0508 (2005) 049 ; Phys. Lett. B 625 (2005) 245.
- [15] P. A. Baikov, K. G. Chetyrkin, A. V. Smirnov, V. A. Smirnov and M. Steinhauser, Phys. Rev. Lett. 102 (2009) 212002 .
- [16] T. Gehrmann, E. W. N. Glover, T. Huber, N. Ikizlerli and C. Studerus, JHEP 1006 (2010) 094 .
- [17] T. Gehrmann, T. Huber and D. Maitre, Phys. Lett. B 622 (2005) 295 .
- [18] S. Moch, J. A. M. Vermaseren and A. Vogt, Nucl. Phys. B 688 , 101 (2004) ; Nucl. Phys. B 691 , 129 (2004).
- [19] K. G. Chetyrkin, B. A. Kniehl and M. Steinhauser, Nucl. Phys. B 510 , 61 (1998) .
- [20] D. de Florian and J. Mazzitelli, JHEP 1212 (2012) 088 .
- [21] S. Moch and A. Vogt, Phys. Lett. B 631 (2005) 48 ; E. Laenen and L. Magnea, Phys. Lett. B 632 (2006) 270 ; A. Idilbi, X. -d. Ji, J. -P. Ma and F. Yuan, Phys. Rev. D 73 (2006) 077501 .
- [22] V. Ravindran, Nucl. Phys. B 746 (2006) 58 ; Nucl. Phys. B 752 (2006) 173 .
- [23] C. Anastasiou, C. Duhr, F. Dulat, E. Furlan, T. Gehrmann, F. Herzog and B. Mistlberger, arXiv:1403.4616 [hep-ph].
- [24] T. Ahmed, M. Mahakhud, N. Rana and V. Ravindran, arXiv:1404.0366 [hep-ph].
- [25] Y. Li, A. von Manteuffel, R. M. Schabinger and H. X. Zhu, arXiv:1404.5839 [hep-ph]; S. Catani, L. Cieri, D. de Florian, G. Ferrera and M. Grazzini, arXiv:1405.4827 [hep-ph].
- [26] T. Gehrmann and D. Kara, arXiv:1407.8114 [hep-ph].
- [27] T. van Ritbergen, J. A. M. Vermaseren and S. A. Larin, Phys. Lett. B 400 (1997) 379 ; M. Czakon, Nucl. Phys. B 710 (2005) 485 .
- [28] V. V. Sudakov, Sov. Phys. JETP 3 (1956) 65 [Zh. Eksp. Teor. Fiz. 30 (1956) 87 ].
- [29] A. H. Mueller, Phys. Rev. D 20 (1979) 2037 .
- [30] J. C. Collins, Phys. Rev. D 22 (1980) 1478 .
- [31] A. Sen, Phys. Rev. D 24 (1981) 3281 .
- [32] S. Catani and L. Trentadue, Nucl. Phys. B 327 (1989) 323 ; Nucl. Phys. B 353 (1991) 183 ; A. Vogt, Phys. Lett. B 497 (2001) 228 ;
- [33] V. Ravindran, J. Smith and W. L. van Neerven, Nucl. Phys. B 704 (2005) 332 ;
- [34] T. Becher and M. Neubert, Phys. Rev. Lett. 102 (2009) 162001 ;
- [35] E. Gardi and L. Magnea, JHEP 0903 (2009) 079 .
- [36] A. D. Martin, W. J. Stirling, R. S. Thorne and G. Watt, Eur. Phys. J. C 63 (2009) 189 .
|
10.1007/JHEP10(2014)139
|
[
"Taushif Ahmed",
"Narayan Rana",
"V. Ravindran"
] |
2014-08-04T19:49:52+00:00
|
2014-08-04T19:49:52+00:00
|
[
"hep-ph",
"hep-ex"
] |
Higgs boson production through $b \bar b$ annihilation at threshold in N$^3$LO QCD
|
We present threshold enhanced N$^3$LO QCD corrections to inclusive Higgs
production through bottom anti-bottom annihilation at hadron colliders using
threshold resummed cross section. The resummed cross section is obtained using
factorization properties and Sudakov resummation of the inclusive cross
section. We use the recent results on threshold N$^3$LO corrections in QCD for
Drell-Yan production and three loop QCD corrections to Higgs form factor with
bottom anti-bottom quark to achieve this task. This is the first step towards
the evaluation of complete N$^3$LO result. We have numerically demonstrated the
importance of such corrections at the LHC.
|
1408.0789v1
|
## The energy source and dynamics of infrared luminous galaxy ESO 148-IG002
## Sarah K. Leslie 1 /star , Jeffrey A. Rich , Lisa J. Kewley 2 1 3 , , Michael A. Dopita 1 3 4 , , ,
- 1 Research School of Astronomy and Astrophysics, Australian National University, Cotter Road, Weston, ACT 2611, Australia
- 2 Carnegie Observatories, 813 Santa Barbara Street, Pasadena, California, 91101 USA
- 3 Institute for Astronomy, University of Hawaii, 2680 Woodlawn Drive, Honolulu, HI 96822
- 4 Astronomy Department, Faculty of Science, King Abdulaziz University, PO Box 80203, Jeddah, Saudi Arabia
5 August 2014
## ABSTRACT
ESO 148-IG002 represents a transformative stage of galaxy evolution, containing two galaxies at close separation which are currently coalescing into a single galaxy. We present integral field data of this galaxy from the ANU Wide Field Spectrograph (WiFeS). We analyse our integral field data using optical line ratio maps and velocity maps. We apply active galactic nucleus (AGN), star-burst and shock models to investigate the relative contribution from star-formation, shock excitation and AGN activity to the optical emission in this key merger stage. We find that ESO 148-IG002 has a flat metallicity gradient, consistent with a recent gas inflow. We separate the line emission maps into a star forming region with low velocity dispersion that spatially covers the whole system as well as a southern high velocity dispersion region with a coherent velocity pattern which could either be rotation or an AGN-driven outflow, showing little evidence for pure star formation. We show that the two overlapping galaxies can be separated using kinematic information, demonstrating the power of moderate spectral resolution integral field spectroscopy.
Key words: galaxies: evolution- galaxies: individual (ESO 148-IG002).
## 1 INTRODUCTION
Understanding the processes involved in galaxy formation and development is a pressing problem in modern astrophysics. Understanding galaxy mergers is key to understanding the formation of elliptical galaxies (Toomre 1977).
galaxy evolution both locally and at higher redshift (Martin 2005; Rupke et al. 2005; Rich et al. 2011).
Ultraluminous ( L IR > 10 12 L /circledot ) and Luminous ( L IR > 10 10 L /circledot ) Infrared Galaxies (U/LIRGs) emit more energy in infrared (5-1000 µ m) than at all other wavelengths combined (Sanders & Mirabel 1996). U/LIRGs are more common at higher redshifts than locally and by z=1 they form the dominant component of the IR luminosity function (Elbaz et al. 2002). The majority of LIRGs are formed by strong interactions/mergers of gas rich spirals (Sanders et al. 1988). The enormous gas concentrations involved facilitate phenomena such as powerful starbursts with accompanying galactic winds, and the feeding of Active Galactic Nuclei (AGNs) which could contribute to the infrared luminosity (Sanders & Mirabel 1996; Lonsdale et al. 2006). Resolving the detailed ionization structure, kinematics and power sources of local U/LIRGs will further our understanding of
As galaxies collide, large quantities of the gas in the disk of each galaxy propagate towards the central regions (Barnes & Hernquist 1996). Gravitational forces between the two galaxies produce tidal tails and disrupted morphology. Tidally induced gas motions and outflows from galactic winds become increasingly common as the merger progresses (e.g., Hopkins et al. 2013). Shocks induced by such large-scale gas flows can influence the emission line gas (Armus et al. 1989; Heckman et al. 1990; Colina et al. 2005; Zakamska 2010). This may contaminate line ratios used to determine metallicity, star formation rate (SFR), and power source (Rich et al. 2011).
Yuan et al. (2010) explains how the composite merger is a critical stage for studying the spectral evolution of two galaxies coalescing. In particular, composite galaxies which are in the process of forming one system, form a bridge between pure starburst and Seyfert galaxies. Shock excitation and its effects on emission line spectra are complex; for example, shock excitation can exhibit extended lowionization narrow emission-line region (LINER) emission characteristics, or shocks can mimic starburst-AGN composite activity (e.g Rich et al. 2010, 2011). Integral field
Figure 1. Hubble Space Telescope broad I band image of ESO 148-IG002 taken with the F814w filter on the WFPC2 instrument. The combined field of view of our two WiFeS pointings is overlaid in red and 15" shown for scale. The blue rectangle shows the region with the sufficiently high signal-to-noise for our analyses, and hence only this section is shown in subsequent maps.
spectroscopy (IFS) is a powerful tool for separating various spatial and kinematic components associated with different power sources (for example Monreal-Ibero et al. 2006; Harrison et al. 2012; Cano-Díaz et al. 2012; Harrison et al. 2014; Förster Schreiber et al. 2014). In this paper we investigate the power sources in ESO 148-IG002. ESO 148-IG002 has a redshift of z=0.045, and is is a late stage ULIRG merger between two disk galaxies of similar size with L IR ∼ = 2 × 10 12 L /circledot (Johansson & Bergvall 1988). The system still contains the nuclei of the original galaxies separated by 5 kpc in projection, and there is evidence that the bright southern nucleus contains an AGN based on [NeV], [OIV], and [NeII] fine-structure emission line diagnostics (Petric et al. 2011) and X-ray colour criterion (Iwasawa et al. 2011).
We discuss the observations and data in Section 2. In Section 3, we present the results of our observations by analysing line ratio maps and emission line diagnostic diagrams. We use the distribution of velocity dispersions to separate the galaxy's ionizing sources and discuss rotation maps in Section 4. In Section 5 we employ shock, AGN and HII region models to determine the roles that shocks and AGN activity play in this system. We also investigate the metallicity properties of this galaxy. In Section 6 we discuss the implications of our results. Finally, we give our conclusions in Section 7.
Throughout this paper we adopt the cosmological parameters H 0 = 70 km s -1 Mpc -1 , Ω λ = 0 72 . , and Ω M = 0 28 . , based on the five-year WMAP results (Hinshaw et al. 2009) and consistent with the Armus et al. (2009) summary of the GOALS sample. With this cosmology at the redshift of ESO 148-IG002, 1" corresponds to 880 pc.
## 2 ESO 148-IG002; OBSERVATIONS, DATA REDUCTION AND LINE FITTING
## 2.1 ESO 148-IG002 (IRAS 23128-5919)
In appearance, this strongly interacting system is very similar to the nearby NGC 4038/4039 ('Antennae') pair. Pre- vious studies of ESO 148-IG002 (with IRAS designation IRAS 23128-5919) show that the galaxy has two nuclei separated by ∼ 4 5 . " (Zenner & Lenzen 1993; Duc et al. 1997), surrounded by bright, possibly star forming knots. The system has two faint tidal tails curling in opposite directions, indicating there was originally two dynamically independent galaxies. Bergvall & Johansson (1985), and Johansson & Bergvall (1988) found LINER and starburst features in the nuclear spectra. They suggested that the nuclear emission-line ratios are consistent with shocks associated with a star burst driven galactic wind. Further studies of infrared mergers by Lípari et al. (2003), agree with this interpretation. Wolf-Rayet (WR) features have also been detected in the southern nucleus by Lípari et al. (2003). Johansson & Bergvall (1988) found a broad intensity enhancement around λ = 4650 Å which they interpret as NIII λ 4641 and HeII λ 4686 emission from WR stars of type N. The nucleus of the southern galaxy, although radio quiet, is relatively bright and pointlike, and is suggested to host an active nucleus (Petric et al. 2011; Iwasawa et al. 2011; Bushouse et al. 2002; Stierwalt et al. 2013).
VLT-SINFONI integral field spectroscopic observations have been taken of ESO 138-IG002 by Piqueras López et al. (2012) in the H (1.45-1.85 µ m) and K (1.95-2.45 µ m) bands, covering the central ∼ 7-11 kpc. From Pa α equivalent width measurements, they find that the northern nucleus is dominated by star formation, whereas in the southern nucleus, the presence of an AGN is supported by strong compact [SiVI] emission. We compare our kinematic results with the findings of Piqueras López et al. (2012) in Section 4. Rodríguez-Zaurín et al. (2011) use the VLT/VIMOS instrument to study H α in LIRGS and ULIRGS. They are unable to detect ESO 148-IG002's tidal tails in the ionized gas emission and report that the optical spectrum shows a mix between LINER, Sy2 and HII-like features.
We analyse integral field unit (IFU) data of ESO 148IG002 to elucidate the relative fraction of the power sources (starburst, AGN, shocks) and nature of shock activity in this galaxy.
## 2.2 Observations and Data Reduction
Our data for ESO 148-IG002 were taken with the Wide Field Spectrograph (WiFes) at the Mount Stromlo and Siding Springs Observatory 2.3m telescope. WiFeS is a dual beam, image slicing IFU described in detail by Dopita et al. (2007, 2010). Our data consist of separate blue ( ∼ 3500-5800 Å) and red ( ∼ 5500-7000 Å) spectra with a resolution of R=3000 for the blue and R=7000 for the red. This resolution corresponds to a velocity resolution of 100 km s -1 at H β and 40 km s -1 at H α . Two pointings with WiFeS were taken on July 28 and August 14 and 15 2009, with total exposure times of 3500s and 4000s respectively. The region overlapping the two pointings would therefore have an exposure time of 7500s. The seeing was on average 1.25" for our observations. The WiFeS field of view compared to ESO 148-IG002 is shown in Figure 1. The IFU field consists of 25 × 1" wide slitlets, each of which is 38" long. The spatial pixel is 0.5" along the slitlet axis and 1.0" in the spectral direction. Post reduction, data were binned 2 pixels in the y-direction in order to increase the signal to noise and produce a final resolution of 1" × 1".
The data were reduced and flux calibrated using the Flux Standard Feige 110 (a white dwarf star) and the WiFeS pipeline. The WiFeS pipeline uses IRAF routines adapted from the Gemini NIFS data reduction package and is briefly described in Dopita et al. (2010). Each observation was reduced into a data cube using the process described in Rich et al. (2011). ESO 148-IG002 was observed using Nod and Shuffle (N & S) mode allowing sky subtraction to be performed using the N & S data. The sky was observed for the same amount of time as the object. The data has bias frames subtracted and any residual bias level is accounted for with a fit to unexposed regions of the detector. The mapping of the slitlets required for spatial calibration is carried out by placing a thin wire in the filter wheel whilst illuminating the slitlet array with a continuum lamp to define the centre of each slitlet. CuAr and NeAr arc lamps are used to wavelength calibrate the blue and red spectra respectively. Observations of a featureless white dwarf taken at similar air mass were used to remove telluric absorption features from the red data cubes.
## 2.3 Spectral Fitting
To fit and remove stellar continuum from each spectrum, we used an automated fitting routine, UHSPECFIT, which is described in Rupke et al. (2010b) and Rich et al. (2010). The software was based on the fitting routines of Moustakas & Kennicutt (2006), which fits a linear combination of stellar templates to a galaxy spectrum. To fit our relatively high resolution data, we used population synthesis models from González Delgado et al. (2005) as our stellar continuum templates. After the subtraction of the fitted continuum, lines in the resulting emission spectra are fit simultaneously using a one- or two-component Gaussian, depending on the goodness of the fit. The emission line fits are carried out in the same manner as Rich et al. (2011). All spectra were fit using both one- and two-component Gaussians. We adopt the fit with the lowest reduced χ 2 , however, properties of surrounding spaxels were also used to decide which fit to use. In addition, all fits were checked by eye, to ensure they were reasonable. We determined that ∼ 200 spaxels required a two-component Gaussian fit, compared to ∼ 140 spaxels whose emission lines were fit with a single component. Each emission line component fit has a corresponding redshift, flux peak and width. The instrumental resolution of 46 km s -1 FWHM (full width at half maximum) is subtracted in quadrature from the line fit FWHM to compensate for instrumental broadening and converted to a σ for the purpose of our analysis. The fitting routines used to fit both the continuum and emission lines made use of the MPFIT package, which performs a least-squares analysis using the Levenberg-Marquardt algorithm (Markwardt 2009). Example fits are provided in Figure 2, along with an example of a typical spectrum across the entire spectral range.
Errors in the parameters used in fitting Gaussian functions to the emission lines are calculated by the MPFIT package, using the variance spectrum as input. To analyse multiple component fits, we minimise the χ 2 for all lines simultaneously. As such all lines ([OII]3726, H β , [OIII]4959, [OIII]5007, [OI]6300, [NII]6548, H , [NII]6583, α [SII] 6716. [SII]6731) have the same velocity and velocity dispersion.
Figure 2. The top two panels show an example of raw data for a single spectrum across the whole spectral range. The bottom panel shows example two-component fits for the H α and [NII] 6583 lines. The flux of the [NII] 6548 component is fixed to the [NII] 6583 flux (ratio 1:3; Osterbrock & Ferland 2006). We do not plot the [NII]6548 line components so the other fitted components can be seen clearly. The right hand spectrum is from the northern nuclear region, whilst the left is from an off-nuclear region east of the southern nucleus. The dashed red lines show the two Gaussian functions which together give the overall emission fit. The raw data are in black and the continuum plus emission fit is in red. Both components are often well defined in one emission line but not in another. Where the amplitude of an individual component is not more than five standard deviations above the noise, it is not considered in subsequent analysis.
Assuming that all the gas contained within a given spaxel is undergoing the same processes, we would expect each emission line in the spectrum to have the same velocity dispersion. By fitting all lines simultaneously, we decrease the overall uncertainty in the velocity dispersion and maintain the ability utilise the multiple component fits.
As our analysis of this kinematically complex system relies on a careful decomposition of multiple velocity dispersions, it is crucial that more than one Gaussian can be used in the emission line fit.
## 3 EMISSION LINE GAS PROPERTIES
## 3.1 Line Ratio Maps
We examine the maps of the ratios of the total flux of [NII] λ 6583, [SII] λλ 6716,6731 and [OI] λ 6300 to H α in Figure 3 as a first step towards understanding the processes at work in ESO 148-IG002. Where two Gaussians are used to fit an individual emission line, the total flux is taken to be the sum of the two component fluxes. These ratios are sensitive to metallicity as well as ionization parameter and
have the advantage that they do not require reddening corrections. Line ratio values corresponding to each spaxel are also plotted on standard diagnostic diagrams in Figure 4. In general, the weaker the line ratio, the greater the contribution of pure star formation to the ionizing spectrum (Kewley et al. 2006). In Figure 3 we also show the ratio of [OIII] λ 5007 to H β , used in all diagnostic diagrams of Figure 4. We represent the errors in the line ratio values by mapping the uncertainty in each spaxel on a logarithmic scale.
The [NII]/H α ratio is sensitive to metallicity (Kewley & Dopita 2002; Denicoló et al. 2002; Pettini & Pagel 2004; Kewley & Ellison 2008). The higher [NII]/H α ratio in the southern corner of the galaxy could thus correspond to a higher metallicity clump, however the AGN in the southern nucleus (bottom white cross in Figure 3) may also contribute to the [NII]/H α ratio. The other two ratios, [SII]/H α and [OI]/H α , are more sensitive to a hard radiation field from AGN or shocks. Higher ratios in the outer regions, such as in these two maps, usually correspond to shocked regions (e.g. Rich et al. 2010; Monreal-Ibero et al. 2010; Sharp & Bland-Hawthorn 2010; Rich et al. 2011. However, in this case there are only ∼ 2 spaxels detected in the outer regions with elevated [SII]/H α and [OI]/H α ratios, so we cannot conclude whether shocks are affecting the area.
## 3.2 Diagnostic Diagrams
To better understand the contributing power sources in this galaxy, we use line ratio diagnostic diagrams. Diagnostics using the [NII]/H α , [SII]/H α and [OI]/H α against [OIII]/H β ratios were first employed by Baldwin et al. (1981) and Veilleux & Osterbrock (1987) to distinguish the likely ionizing source of emission line gas in galaxies. Kewley et al. (2001) used a combination of stellar population synthesis models and self-consistent photoionization models to determine a theoretical 'maximum starburst line' on the diagrams which indicates the theoretical upper limit given by pure stellar photoionization models. The diagrams have been subsequently updated by Kauffmann et al. (2003) and Kewley et al. (2006) to include empirical lines dividing pure star-forming from Seyfert-HII composite galaxies and Seyferts from LINER galaxies respectively.
We show the diagnostic diagrams for ESO 148-IG002 in Figure 4. We include the emission line fluxes from each spaxel with S/N > 5 in all relevant lines. This procedure allows us to classify the dominant energy source in each spaxel. In all three diagrams the lower left hand section of the plot traces photoionization by HII regions. The solid curve traces the upper theoretical limit to the pure HII region contribution measured by Kewley et al. (2001). The observed dashed line in the [NII] diagnostic provides an empirical upper limit to the pure HII region sequence of Sloan Digital Sky Survey galaxies measured by Kauffmann et al. (2003). The region lying between these two lines represents objects with a composite spectrum which is a mix of HII region emission and a stronger ionizing source. LINER-like emission lies to the right-hand side of the diagrams whilst contribution from a Seyfert AGN will push the spaxels upwards on all three diagrams. If the galaxy is influenced by shocks, the line ratios can be moved towards the LINER region (Rich et al. 2010,
2011). From the [NII] diagnostic plot it is clear that there are very few purely star forming regions in this galaxy.
The large number of SDSS galaxies ( ∼ 45,000), allowed Kewley et al. (2006) to separate two clear branches on the [SII]/H α and [OI]/H α diagnostic diagrams, empirically deriving a boundary between Seyfert 2s and LINERs seen as the dashed lines on these diagrams. LINERs have a harder ionizing radiation field and lower ionization parameter than Seyfert galaxies, making the [SII]/H α and [OI]/H α diagrams ideal for separating Seyferts and LINERS. The [SII] and [OI] emission lines are produced in the partially ionized zone at the edge of the nebula. For hard radiation fields, this zone is large and extended. The [NII]/H α ratio only weakly depends on the hardness of the radiation field, is much more dependent on the metallicity of the nebular gas and thus cannot be used to separate Seyfert and LINER galaxies. The [SII] and [OI] diagnostic diagrams in Figure 4 indicate that the composite HII regions seen in the [NII] diagram could be influenced by AGN activity as the spaxels are spread upwards towards the Seyfert 2 branch.
ESO 148-IG002 has a bimodal distribution of [OIII]/H β line ratios, which we have visually separated in the top panel of Figure 4. The location of the spaxels with high and low [OIII]/H β ratios on the BPT diagram, combined with their physical location in the galaxy, could suggest that the northern nucleus has a higher metallicity than the southern nucleus.
## 4 KINEMATIC PROPERTIES
The power source of the emission line ratios is unclear from the diagnostic diagrams alone. In this section we try to better understand the dynamics as well as the different power sources of ESO 148-IG002. Here we make most use of the two Gaussian component fits. As all emission lines in a spectrum were fit simultaneously with one or two Gaussian components, all lines per spaxel have the same velocities, and velocity dispersions, which are used in this section. In data with high spectral resolution, one can analyse the distribution of velocity dispersions to separate the galaxy's ionizing sources.
## 4.1 Velocity Dispersion
Velocity dispersion, σ , is the result of the superposition of many line profiles, each of which has been Doppler shifted and broadened because of the gas motions within the galaxy. The velocity dispersions used for our analysis have had the instrumental resolution ( ∼ 40 km s -1 FWHM in the red) subtracted in quadrature to account for instrumental broadening. In a complex galaxy such as ESO 148-IG 002, a large proportion of the gas may be affected by more than one source of ionization. In this section we focus on the velocity dispersion of individual emission line components, determined from the two component fits, as a useful way to probe the processes influencing each spaxel. HII regions correspond to a low velocity dispersion, typically of a few tens of km s -1 (Epinat et al. 2010), while slow shocks associated with galactic winds have velocity dispersions > ∼ 100 km s -1 (Rich et al. 2010, 2011), and the presence of an AGN can produce even broader lines (Wilson & Heckman 1985). In
(a) HST
(e) HST
(i) HST
(f) H α error
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
(j) [SII]/H α
-17.0 -16.5 -16.0 -15.5 -15.0 -14.5 (c) [NII]
(g) [NII] error
(k) [OI]/H α
(h) [NII]/H α error
(l) [OIII]/H β
Figure 3. Emission line ratio maps using total fluxes (the sum of both Gaussians if a double fit is made). All scale bars show the log of the value indicated by the captions. The top images show the log of the total emission of [NII] and H α (in units of ergs s -1 cm -2 Å -1 ) of ESO 148-IG002, and the ratio of the two. In both the H α and [NII] maps, two peaks in the emission can be seen. This most likely corresponds to the two nuclei from the original galaxies (as determined by the peaks in the HST image), shown by white crosses on the maps. Underneath the images their corresponding relative error maps are also shown on a logarithmic scale. Images (j), (k), and (l) are the line ratio maps corresponding to the diagnostic diagrams for log([SII]/H α ), log([OI]/H α ) and log([OIII]/H β ). The left most panels are the HST image of ESO 148-IG002 shown in Figure 1, with the same dimensions and orientation as the other maps, for reference. The white circle on the bottom right corner of the HST image is an estimate of the seeing for our WiFeS observations.
Figure 5 we show the velocity dispersion distribution from every component fit by our routine with S/N > 5. The spectrum from every spaxel was fit with either one or two Gaussian curves with each emission line in a spectrum having the same fixed dispersion values (one or two depending how many components were needed), and all components (for every spaxel) are shown.
We establish a cutoff of σ = 155 km s -1 between the lowσ HII region emission and higher σ emission by shocks and/or AGN, as there is a local minimum in the velocity dispersion distribution at this value (Figure 5). The choice of appropriate cutoffs was also influenced by the velocity dispersion maps, shown in Figure 6. The fact that the different components occupy different regions indicates that
- (b) Location of spaxels
Figure 4. Emission line ratio diagnostic diagrams of each individual fitted emission line. Here total fluxes for each line are used (as in Figure 3). Black solid curves form an upper limit for star-forming galaxies as derived by Kewley et al. (2001). The dashed line on the [NII] diagram is the empirical Kauffmann et al. (2003) boundary below which galaxies are classified as star-forming. The dashed lines on the [SII] and [OI] diagrams were derived by Kewley et al. (2006) to empirically separate Seyfert 2 galaxies and LINERs. In the leftmost panel, NLAGN represents narrow emission-line AGN (Seyfert 2 plus LINERs); Comp represents starburst-AGN composites. We apply a minimum S/N cut of 5 in every diagnostic line. Typical errors are 0.009 dex for log([NII]/H α ) and log([SII]/H α ), 0.016 dex for log([OI]/H α ) and 0.012 for log([OIII]/H β ). The spaxels in each diagram reflect a mix of ionization sources. The spaxels are distributed in two clumps in [OIII]/H β as shown in the histogram of the top right panel. The division between the two clumps is shown across the top panel by the dotted line, marking the minimum in the bimodal [OIII]/H β distribution. The points are also colour coded based on this division. The spatial position of the clumps are shown in the bottom right panel, with the red and blue colors matching those from the BPT diagram. This could indicate that gas from the northern nucleus (blue) has a higher metallicity.
they may be influenced by different phenomena. In some cases a spaxel has more than one component in the same velocity dispersion bin. In this case, to create the velocity dispersion maps, the σ values were averaged (weighted by the component's H α flux) over the two components within the same bin. The high velocity dispersion gas is located to the South of the system, where the AGN lies. However, the peak velocity dispersion is offset from the nucleus. This is consistent with the idea that the high velocity dispersion gas is powered in part by an AGN. As shocks tend to correspond to a larger velocity dispersion than that seen on the outskirts of this galaxy, it is unlikely that shocks are contributing significantly to the emission spectrum.
## 4.2 Velocity Maps
Using the separate Gaussian components we are able to calculate the recessional velocity for each of the two velocity dispersion groups. We adopted a systemic redshift of z=0.04460 (Lauberts & Valentijn 1989). The velocity maps are shown in Figure 7. A velocity shear can be seen in the map of the second, broader component ( σ > 155 km s -1 ), centred over the southern nucleus. The low velocity dispersion component, which likely corresponds to star forming regions based on its velocity dispersion, does not show rotation. However, it is possible that this component could be a face-on rotating structure, in which case we would not see a velocity shear.
Studying P α in ESO 148-IG002, Piqueras López et al. (2012) similarly find low velocity dispersions ( ∼ 65 km s -1 ) in the northern nucleus and high velocity dispersion ( ∼ 190 km s -1 ) in the southern nucleus. They observe red and
(c)
σ >
155
km s
-
1
Figure 6. Velocity dispersions in km s -1 of the two components, with grouping determined by the cut off. Left to right corresponds to increasing velocity dispersion. For spatial comparison the HST image has been aligned and placed in the leftmost panel. Where a spaxel contains two components fit which fall in the same velocity dispersion bin, we show a flux weighted average. Spaxels which do not appear in the σ > 155 km s -1 figure are the cases where either only one component is fit, and that component is narrow, or two components are fit and both are narrow. The low velocity dispersion gas is extended over the whole galaxy, whereas the high velocity dispersion gas is only found in the southern region.
rotating component seen to the south, could be the disk of the progenitor spiral, containing an AGN. Due to the large spatial coverage of this rotating component ( ∼ 15 kpc) it is difficult to compare this component to other studies such as U et al. (2013) who observe rotating gas < 1 kpc from an AGN. However, larger samples of (U)LIRGs (Medling et al. 2014) do see evidence of both small nuclear disks (r ∼ few hundred pc) and larger disks (r > 1 kpc), which they also interpret as the progenitor galactic disk. The velocity broadening is likely due to the presence of the AGN. However, beam smearing could also cause an increased velocity dispersion. [OIII] is commonly used to measure the size of narrow-line regions (NLRs) in AGNs, typically giving extents of 1-5 kpc (Bennert et al. 2006; Davies et al. 2014). It is unlikely that the AGN at the center of the southern nucleus is able to cause the high line ratios observed ([OIII]/H β ) at 15 kpc without help from star formation or shocks.
Figure 5. Velocity dispersion distribution from every individual Gaussian component fit. The distribution has a peak around ∼ 100 km s -1 . Pure star-forming regions have velocity dispersions of ∼ 40 km s -1 (Epinat et al. 2010). Therefore it is likely that spaxels with velocity dispersions less than the local minimum at σ < 155 km s -1 are likely to be dominated by star formation, but also have contribution from shock or AGN. The extended distribution out to higher velocity dispersions could represent an increasing contribution from AGN activity.
blue wings in the P α and H 1-OS(1) line 2 profiles which they suggested forms a cone-like structure, centered on the AGN and extending ∼ 3 -4 kpc north-east and south-west. Piqueras López et al. (2012) also report a velocity gradient of ∆ v ∼ 140 km s -1 around the northern nucleus in a northsouth direction, a feature which can also be seen in our Figure 6b. We show, by decomposing emission lines into two Gaussian components, that the velocity shear observed in the south, extends out to ∼ 15 kpc in the optical.
We propose two possible explanations for the velocity and velocity dispersion fields of ESO 148-IG002. Firstly, the
Secondly, the velocity shear, which was previously interpreted as rotation, could represent an AGN bipolar outflow, similar to that seen in Davis et al. (2012) and suggested in Piqueras López et al. (2012). Harrison et al. (2012) find broad [OIII] emission lines ( σ = 300-600 kms -1 ) in highredshift(z=1.4-3.4) galaxies containing AGN. The broad emission line regions extend across 4-15 kpc and have high velocity offsets from the systemic redshift ( ≈ 850 km s -1 ) and are attributed to galaxy-scale AGN-driven winds. ESO 148-IG002's broad velocity dispersion gas has a more modest velocity offset ( ∼ 350 km s -1 ) than the outflows in the Harrison et al. (2012) sample, however, the high velocity dispersion, large spatial extent, and velocity shear are consistent with the AGN-driven outflow scenario.
By comparing the ionized gas kinematics with the stellar velocity field, Davis et al. (2012) was able to conclude that their high velocity dispersion component was due to shocked outflow, as the axis of the gas' velocity gradient was offset from the stellar rotation axis. Information on the stellar kinematics of ESO 148-IG002's could help differentiate between the two scenarios proposed here.
For the purposes of clarity, we refer to the kinematically distinct broad component as 'rotating' in subsequent anal-
ysis, though the reader should remember that its coherent structure may either be rotation or outflows.
Fitting a disk model to the rotating component could also help decide whether the velocity shear is due to rotation or outflow. Disk fitting and kinemetry analysis (Krajnović et al. 2006) will be subject of future work, with a larger sample of galaxies, allowing us to better understand the kinematic properties of complex galaxy systems.
## 5 ANALYSIS
In the previous section we separated the emission line fluxes into two components, one with velocity dispersions σ < 155 km s -1 , and the other with velocity dispersions higher than σ > 155 km s -1 . The distribution of flux into the two components is illustrated in Figure 8, which maps the H α flux in the low and high velocity dispersion components. Low velocity dispersion gas is present throughout the system, whereas the high velocity dispersion gas is centred on the southern nucleus, where the AGN lies. In this section we use the emission line ratios of the separate components to analyse two potentially distinct power sources.
## 5.1 Diagnostic Diagrams
In Figure 9 the spaxels on the optical diagnostic diagrams are colour coded by velocity dispersion. Line emission from each Gaussian component of every spectrum with S/N > 5 is considered here. If the two different colours occupy separate positions on the diagnostic diagrams, then this separation would imply that different ionizing sources are responsible for each velocity component. Separating the emission lines in a spaxel in up to two components results in a greater number of data points than shown in Figure 4. When two or more processes are working to power the emission spectrum of the gas, using a line's total flux for analysis, as in Figure 4, works to average the effects that the different processes may have. Separating the flux into two separate components, based on their velocity dispersion, allows us to analyse two potentially different power sources. There are fewer high velocity dispersion components present in the [SII]/H α diagnostic diagram and only one in the [OI]/H α diagram. The lack of high velocity dispersion spaxels is a result of applying a signal to noise cut on each component fit.
Figure 9 shows that the broad component is dominated by spaxels in the Seyfert and composite region of the diagnostic diagram, while a larger fraction of the narrow component lies in the HII-region portion of the diagnostic diagram. The low velocity dispersion components spread upwards predominantly to the AGN regions in both the [SII] and [OI] diagnostic diagrams indicating that even the low velocity dispersion gas could be affected by the AGN and/or shocks.
## 5.2 Metallicity Gradient
To find the metallicity (given as 12+log(O/H)) of ESO 148IG002, we use the method of Kobulnicky & Kewley (2004), which uses the stellar evolution and photoionization grids from Kewley & Dopita (2002) to produce an (updated) analytic prescription for estimating oxygen abundances using the traditional strong emission line ratio R 23 , where R 23 = ([ OII λ ] 3727 [ / OIII λλ ] 4959 5007) , /Hβ . The R 23 calibration is sensitive to the ionization state of the gas, particularly for low metallicities. The ionization state of the gas is characterised by the ionization parameter (the number of hydrogen-ionizing photons passing through a unit area per second, divided by the hydrogen density of the gas), which is typically derived using the [OIII]/[OII] ratio. However, this ratio is also sensitive to metallicity, so Kobulnicky & Kewley (2004) suggest an iterative approach to derive a consistent ionization and metallicity solution. We first determine whether the spaxels lie on the upper or lower R 23 branch using the [NII]/[OII] ratio and calculate an initial ionization parameter assuming a nominal lower branch [12+log(O/H)=8.2] or upper branch [12+log(O/H)=8.7] metallicity using:
$$\dots \quad & \bullet \quad \sim \\ \text{ereas} \quad & \log ( q ) = ( 3 2. 8 1 - 1. 1 5 3 y ^ { 2 } + [ 1 2 + \log ( O / H ) ] ( - 3. 3 9 6 \\ \text{hern} \quad & \quad - 0. 0 2 5 y + 0. 1 4 4 4 y ^ { 2 } ) ( 4. 6 0 3 - 0. 3 1 1 9 y - 0. 1 6 3 y ^ { 2 } \\ \vdots \text{two} \quad & \quad + [ 1 2 + \log ( O / H ) ] ( - 0. 4 8 + 0. 0 2 7 1 y + 0. 0 2 0 3 7 y ^ { 2 } ) ) ^ { - 1 } \quad ( 1 )$$
where y = log([ OIII λ ] 5007 [ / OII λ ] 3727) . A spaxel is determined to lie on the lower R 23 branch if log([ NII / OII ] [ ]) < -1 2 . , and on the upper branch if log([ NII / OII ] [ ]) > -1 2 . . The initial ionization parameter is then used to derive a metallicity estimate, using
$$\text{sion} \quad 1 2 + \log ( O / H ) _ { l o w e r } = 9. 4 0 + 4. 6 5 x - 3. 1 7 x ^ { 2 } \\ \text{with} \quad - \log ( q ) ( 0. 2 7 2 + 0. 5 4 7 x - 0. 5 1 3 x ^ { 2 } ) \quad \quad ( 2 )$$
where x = log( R 23 ) , if the spaxel lies on the lower branch, and
$$\text{at} \quad 1 2 + \log ( O / H ) _ { u p p e r } = 9. 7 2 - 0. 7 7 7 x - 0. 9 5 1 x ^ { 2 } - 0. 0 7 2 x ^ { 3 } \\ \text{v} \text{ in} \quad - 0. 8 1 1 x ^ { 4 } - \log ( q ) ( 0. 0 7 3 7 - 0. 0 7 1 3 x - 0. 1 4 1 x ^ { 2 } \\ \text{over} \quad + 0. 0 3 7 3 x ^ { 3 } - 0. 0 5 8 x ^ { 4 } ), \ \ ( 3 )$$
if the spaxel lies on the upper branch. Equations 1 and 2 (or 3) are iterated until 12 + log( O/H ) converges. To find the metallicity of spaxels dominated by star-formation, emission line fluxes from the low velocity dispersion component, with S/N greater than 5 in all relevant lines, lying to the left of the empirical line from Kauffmann et al. (2003) (seen in Figure 9) were used, and an average metallicity of log ( O H ) + 12 = 9 09 . ( ± 0 03 . dex) was obtained. Metallicity as a function of radius is plotted in Figure 10. As there are two nuclei in ESO 148-IG002, we have separated the spaxels into two groups, a northern and southern, using the spatial cut-off given by the boundary of the rotating broad component gas (Figure 8). Spaxels belonging to the northern nucleus were determined to be spaxels north of the broad component seen in Section 4.2, and as such the distance to these spaxels is given from the northern nucleus. Distances to spaxels south of this spatial boundary are calculated from the position of the southern nucleus. The gas metallicity of the star forming regions remains constant around the mean value as a function of nuclear distance, with a scatter of ∼ 0 2 . dex.
Normal spiral galaxies have negative metallicity gradients consistent with central enrichment from generations of star formation (Henry & Worthey 1999). Figure 10 indicates that ESO 148-IG002 has a flat metallicity gradient. Flat
(b) Velocity(km s -1 ): σ < 155 km s -1
(c) Velocity(km s -1 ): σ > 155 km s -1
Figure 7. Maps of velocity for the two components determined by our cutoff. From left to right the images correspond to the velocity of spaxels with velocity dispersions of: σ < 155 km s -1 and σ > 155 km s -1 . Where a spaxel contains two components fit which fall in the same velocity dispersion bin, we show a flux weighted average velocity. For spatial comparison the HST image has been aligned and placed in the leftmost panel. The high velocity dispersion gas located to the south shows a velocity shear which we interpret either rotation or outflow.
(a) H α σ < : 155 km/s
(b) H α error: σ < 155 km/s
(c) H α σ > : 155 km/s
(d) H α error: σ > 155 km/s
Figure 8. Maps of the log of the H α emission (in units of ergs s -1 cm -2 Å -1 ) of the two Gaussian components determined by our velocity dispersion cutoff and their corresponding uncertainty maps. From left to right the images correspond to the velocity of spaxels with velocity dispersions of: σ < 155 km s -1 and σ > 155 km s -1 . The distribution of the two components match with the HST nuclei with the southern component peak consistent with the AGN position. Horizontal white lines on the flux maps mark the division between the north and south nuclei from which we calculate distances in Section 5.2.
Figure 10. Metallicity versus distance from corresponding nucleus. Metallicities in red lie to the north of the system. For these metallicities, the distance has been measured from the northern nucleus. Blue metallicities lie to the south of the galaxy and the distance to them is calculated from the southern nucleus. Uncertainties in the distance from the nucleus are derived from the average seeing of our observations. No gradient in the metallicity as a function of radius is evident.
metallicity gradients can be produced by merger-induced gas inflows (Kewley et al. 2010; Rupke et al. 2010; Torrey et al. 2012). The gradient in ESO 148-IG002 is flatter than any mergers in Kewley et al. (2010) and may indicate very recent gas infall.
## 5.3 AGN, Shock and HII models
To determine the effects (if any) of shock excitation, we employ slow shock models, which were introduced and described in Farage et al. (2010) and Rich et al. (2010). Slow shock models with velocities consistent with our observed line widths, (100-200 km s -1 ) were generated using an updated version of Mappings III code, originally introduced in Sutherland & Dopita (1993). We also compare HII region models generated using Starburst99 (Leitherer et al. 1999) and Mappings III (Kewley et al. 2001; Levesque et al. 2010). We show mixing sequences of star formation and shock excitation in Figure 11 for a metallicity of 8.69 (solar), calculated by varying the fractional contribution of the HII and shock models on the line ratios. Higher metallicity models would lie lower on the diagrams and are inconsistent with the position of ESO 148-IG002. Not all spaxels are covered by the shock-mixing models shown in Figure 11. Therefore
Figure 9. As Figure 4, but with separate Gaussian components color-coded based on their velocity dispersions. Separating the emission lines in a spaxel in up to two components results in a greater number of data points than shown in Figure 4. Spaxels with σ less than 155 km s -1 are plotted in blue, and the highest velocity dispersions, σ greater than 155 km s -1 are coloured red. Every line used has S/N > 5 . The broad components are mostly in the Seyfert region on these diagrams and the narrow component spaxels are spread from the star forming region upwards, indicating a larger contribution from star formation in the low velocity dispersion gas.
although some spaxels may be influenced by shocks, it is likely that (a mixing between star formation and) another phenomenon is causing the high line ratios seen.
We also employ a dusty, radiation pressure-dominated photoionization model first introduced in Dopita et al. (2002), and updated in Groves et al. (2004). These models provide a self consistent explanation for the emission from narrow-line regions of AGN. From the family of models available, we explored those with hydrogen number density of 10 3 cm -3 and a metallicity of 2Z /circledot to match the properties of ESO 148-IG002. A simple power law represents the spectral energy ( ν ) distribution of the ionizing source, with
$$F _ { \nu } \in \nu ^ { \alpha }, \nu _ { m i n } < \nu < \nu _ { m a x }.$$
and ν min = 5eV and ν max = 1000eV. Four values of the power-law index, α are shown in Figure 12; -1.2,-1.4,-1.7, and -2.0 as these indices encompass the range of indices seen in AGN locally (Groves et al. 2004). The ionization parameter, log U 0 in the model shown varies between 0.0 and -2.3 dex. We also construct a starburst-AGN mixing sequence, by varying the fractional contribution of the HII and AGN models on the line ratios in linear space. The mixing sequence shown in Figure 12 results from applying this method on a single pure HII region and pure twice solar metallicity AGN region indicating changes in the fraction of starburst to AGN of 10 % in linear space.
Neither model encompasses all spaxels from ESO 148IG002. Recall from Figure 9 that almost all of the spaxels on the [OI] diagnostic are of lower velocity dispersion. Low velocity dispersion spaxels lie on both the shocked mixing sequence and the AGN mixing sequence. Spaxels in the higher velocity dispersion group are closer to the AGN models, which indicates that this high σ component is influenced more strongly by the AGN in the southern nucleus, over which those spaxels lie.
To determine whether the starburst-AGN or starburstshock models better describes our emission line data, we perform the statistical Kolmogorov-Smirnov (KS) test. The KS test is a nonparametric test of the equality of two continuous distributions and can be used to determine the likelihood that a sample distribution was drawn from reference distribution (one-sided KS test), or to compare two sample distributions (two-sided KS test). We compare the values of ESO 148-IG002's [OIII]/H β and [SII]/H α ratios with the values estimated from the above mixing sequences using a two-dimensional, two-sided KS test. Although the KS test is only strictly defined for one dimensional probability distributions, Press et al. (2002) describes how an analogous test can be designed and implemented when the distribution depends on two variables. We use the combination of [OIII]/H β and [SII]/H α ratios for this test as the [SII]/H α ratio is more sensitive to shocks than [NII]/H α , and is of higher signal to noise in our data than the [OI]/H α . Table 1. lists the p values calculated using the two sided KS test. A p value close to one means that there is a high probability that the two samples originated from the same parent distribution. We find that gas with σ < 155 km s -1 can be better described by the AGN mixing sequence, than with the shock mixing sequence used, with p values 0.73 and 0.59 respectively. The high velocity dispersion gas with σ > 155 km s -1 is also better described by our AGN mixing model, with p =0.63 as compared with p =0.48 for the shock mixing model. As previous studies have indicated the presence of an AGN in the southern nucleus as well as star formation, it is not surprising that both the line ratios and velocity dispersions of ESO 148-IG02 show a mixture of star formation and AGN activity. Both the low and high velocity dispersion components are better described by the model AGN mixing sequence than the shock mixing sequence. It is likely that the high velocity dispersion component traces gas strongly
Table 1. The results of performing the KS test which compared the ratios of [SII]/H α and [OIII]/H β for each spaxel, with a model star forming-shock mixing sequence, and a model star formingAGN mixing sequence. Values of p closer to one imply that the underlying distributions are more likely the same, i.e that the model is preferred. We find that the AGN mixing sequence better describes our data than the shock mixing sequence.
| p value | shock | AGN |
|--------------|---------|-------|
| σ < 155 km/s | 0.587 | 0.729 |
| σ > 155 km/s | 0.481 | 0.628 |
influenced by the AGN, whilst the low velocity dispersion gas is more strongly influenced by star formation with AGN activity having a small effect.
Figure 11 suggests that more theoretical work is needed to accurately model the [SII] emission-line fluxes in merging galaxies. Unlike the [NII]/Ha and [OI]/Ha mixing sequences, the [SII]/Ha line ratios are not well-fit by the mixing sequence. The difficulty in modeling the [SII]/Ha lines in galaxies is well known (e.g., Levesque et al. 2010), and has been attributed to the theoretical shape of the EUV radiation field.
## 6 DISCUSSION
Observations by Rich et al. (2012) and Kewley et al. (2010), and simulations (Rupke et al. 2010) show that there is a strong relationship between metallicity gradients and the gas dynamics in galaxy interactions and mergers. As a merger progresses the metallicity gradient is flattened. This flattening reflects the effects of gas redistribution over the disks, including less enriched gas being torqued towards the centre and the growth of tidal tails that carry metals out to large radius. In the previous section we found the metallicity gradient of ESO 148-IG002 is flatter than any of the sample galaxies in Kewley et al. (2010). Based on our analysis, we present the following picture of ESO 148-IG002:
Gas from the northern galaxy has been spread out across the whole system as a result of the merger. This constitutes the low velocity dispersion component ( σ < 155 km s -1 ) which covers ESO 148-IG002. This area is most likely to be dominated by star formation as indicated from the optical diagnostic line ratios. The mixing of the gas due to the gravitational forces acting between the two galaxies is likely to be responsible for the flat metallicity gradient observed. The southern galaxy seems to have remained dynamically distinct from the northern galaxy, either maintaining a rotation that is not seen in the gas with lower velocity dispersion, or being the result of an AGN-driven outflow. The southern nucleus contains an AGN which is the most likely cause for the high velocity dispersion in the surrounding gas ( σ > 155 km s -1 ). The few spaxels that lie near the southern nucleus have velocity dispersions of ∼ 600 km s -1 , consistent with an AGN. It is not possible to entirely rule out shocks from the galaxy. From the line ratios it is possible that at least some of the galaxy is influenced by shock excitation.
Considering this interesting scenario and the flat metallicity gradient of the galaxy, we suggest that we are observing ESO 148-IG002 in the middle of (or just after) major gas rearrangement related to its merger, which makes it an exciting candidate for future studies of star formation and AGN fuelling.
## 7 CONCLUSION
Using WiFeS wide integral field spectroscopy of the ULIRG ESO 148-IG 02 we found that:
- · The distribution of gas velocity dispersion is bimodal, indicating a combination of power sources.
- · The emission line ratios indicate composite starburst and AGN activity, with a starburst-AGN mixing model better able to explain the data than a starburst-shock mixing model (AGN mixing model produced p values of 0.73 and 0.63 for the low and high velocity dispersion gas respectively, compared to p values of 0.59 and 0.48 for shock mixing models when tested using the KS statistic).
- · The HII regions, dominated by star formation and given by the lowerσ component, have a mean metallicity of 9.09 ± 0.03.
- · The galaxy has a flat metallicity gradient as a result of the merging process.
- · The high σ gas associated with the southern nucleus has a coherent velocity pattern which could either be rotation or an AGN-driven outflow.
It is possible that we have caught this merger as the gas from the northern galaxy is being mixed throughout the two galaxies. If this is the case, we have shown that it is possible to disentangle two galaxies in a merging system based entirely on its kinematic analysis.
ESO 148-IG002 provides an exciting test bed for future work on the potential triggering of starbursts and AGN by merger induced gas flows.
## ACKNOWLEDGMENTS
L.K. gratefully acknowledges the support of an ARC Future Fellowship and ARC Discovery Project grant DP130103925. We would like to thank A. Medling for discussing the interpretation of the kinematic components with us. This research has made use of NASA's Astrophysics Data System Bibliographic Services and the NASA/IPAC Extragalactic Database (NED). The authors are very grateful to two anonymous referees whose comments and suggestions greatly improved the clarity of this paper.
## References
Armus, L., Heckman, T. M., & Miley, G. K. 1989, ApJ, 347, 727 Armus, L., Mazzarella, J. M., Evans, A. S., et al. 2009, PASP, 121, 559 Baldwin, J. A., Phillips, M. M., & Terlevich, R. 1981, PASP, 93, 5 Barnes, J. E. & Hernquist, L. 1996, ApJ, 471, 115 Bennert, N., Jungwiert, B., Komossa, S., Haas, M., & Chini, R. 2006, A&A, 456, 953 Bergvall, N. & Johansson, L. 1985, A&A, 149, 475
Figure 11. As Figure 9, but with spaxels corresponding to the high velocity dispersion gas in blue, and the low velocity dispersion gas in black. Overplotted are HII region and shock models, with varying ionization parameter (6.5-8.0), and shock speed (100-200 km s -1 ) for solar metallicity (8.69). In between, we show a mixing sequence from pure HII region to pure shock excitation. Increasing shock velocity or decreasing ionization parameter decreases the [OIII]/H β ratio and increases the other ratios. The shock mixing models do cover some of the data but are unable to explain all of ESO 148-IG002's line ratios. Parts of the shock mixing models fall on the same region as AGN mixing models (Figure 12), as they both start from the HII region model. Thus even though some spaxels are consistent with our shock models, we can not conclude that the gas is shocked.
Figure 12. As Figure 9, but with spaxels corresponding to the high velocity dispersion gas in blue, and the low velocity dispersion gas in black. Included are the twice solar metallicity AGN models from Groves et al. (2004). The power-law index α values (black grid lines) represented are -1.2, -1.4, -1.7, and -2.0. Ionization parameter ( log U 0 ) values represented include 0.0, -0.6, -1.3, -2.0, and -2.3. A mixing sequence between a pure HII region and a pure twice solar metallicity AGN region is created, with each line representing a change in the fraction of starburst to AGN of 10 % in linear space (dotted line). A combination of AGN and HII region models better describe our line ratios than the shock models of Figure 11.
| Bushouse, H. A., Borne, K. D., Colina, L., et al. 2002, ApJS, 138, 1 |
|------------------------------------------------------------------------------------------------------------------|
| Cano-Díaz, M., Maiolino, R., Marconi, A., et al. 2012, A&A, 537, L8 |
| Colina, L., Arribas, S., & Monreal-Ibero, A. 2005, ApJ, 621, 725 |
| Davies, R. L., Rich, J. A., Kewley, L. J., & Dopita, M. A. 2014, MNRAS, 439, 3835 |
| Davis, T. A., Krajnović, D., McDermid, R. M., et al. 2012, MNRAS, 426, 1574 |
| Denicoló, G., Terlevich, R., & Terlevich, E. 2002, MNRAS, 330, 69 |
| Dopita, M., Hart, J., McGregor, P., et al. 2007, Ap&SS, 310, 255 |
| Dopita, M., Rhee, J., Farage, C., et al. 2010, Ap&SS, 327, 245 |
| Dopita, M. A., Groves, B. A., Sutherland, R. S., Binette, L., & Cecil, G. 2002, ApJ, 572, 753 |
| Duc, P.-A., Mirabel, I. F., & Maza, J. 1997, A&AS, 124, 533 |
| Elbaz, D., Cesarsky, C. J., Chanial, P., et al. 2002, A&A, 384, 848 |
| Epinat, B., Amram, P., Balkowski, C., & Marcelin, M. 2010, MNRAS, 401, 2113 |
| Farage, C. L., McGregor, P. J., Dopita, M. A., & Bicknell, G. V. 2010, ApJ, 724, 267 |
| Förster Schreiber, N. M., Genzel, R., Newman, S. F., et al. 2014, ApJ, 787, 38 |
| González Delgado, R. M., Cerviño, M., Martins, L. P., Lei- therer, C., & Hauschildt, P. H. 2005, MNRAS, 357, 945 |
| Groves, B. A., Dopita, M. A., & Sutherland, R. S. 2004, ApJS, 153, 9 |
| Harrison, C. M., Alexander, D. M., Mullaney, J. R., & Swinbank, A. M. 2014, MNRAS, 441, 3306 |
| Harrison, C. M., Alexander, D. M., Swinbank, A. M., et al. 2012, MNRAS, 426, 1073 |
| Heckman, T. M., Armus, L., & Miley, G. K. 1990, ApJS, 74, 833 |
| G., Weiland, J. L., Hill, R. S., et al. 2009, |
| 432, 2639 |
| K., Sanders, D. B., Teng, S. H., et al. 2011, 529, A106 |
| G., Heckman, T. M., Tremonti, C., et al. |
| 2003, MNRAS, 346, 1055 |
| L. J., Dopita, M. A., Sutherland, R. S., C. A., & Trevena, J. 2001, ApJ, 556, 121 |
| L. J. & Ellison, S. L. 2008, ApJ, 681, 1183 |
| L. J., Groves, B., Kauffmann, G., & Heckman, 2006, MNRAS, 372, 961 |
| L. J., Rupke, D., Zahid, H. J., Geller, M. J., Barton, E. J. 2010, ApJL, 721, L48 |
| D., Cappellari, M., de Zeeuw, P. T., & Y. 2006, MNRAS, 366, 787 A. & Valentijn, E. A. 1989, The surface |
| photom- |
| etry catalogue of the ESO-Uppsala galaxies |
| Lauberts, |
| H. A. & Kewley, L. J. 2004, ApJ, 617, 240 |
| Hinshaw, 180, 225 P. F., Kereš, D., & Murray, N. 2013, MNRAS, |
| L. & Bergvall, N. 1988, A&A, 192, 81 |
| R. B. C. & Worthey, G. 1999, PASP, 111, 919 |
| Henry, ApJS, |
| Hopkins, |
| Iwasawa, A&A, |
| Johansson, Kauffmann, Kewley, L. J. & Dopita, M. A. 2002, ApJS, 142, 35 |
| Kewley, |
| Heisler, Kewley, Kewley, |
| T. Kewley, Kobulnicky, |
| Krajnović, |
| & |
| Copin, |
| Leitherer, C., Schaerer, D., Goldader, J. D., et al. 1999, ApJS, 123, 3 |
|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| Levesque, E. M., Kewley, L. J., Berger, E., & Zahid, H. J. 2010, AJ, 140, 1557 |
| Lípari, S., Terlevich, R., Díaz, R. J., et al. 2003, MNRAS, 340, 289 |
| Lonsdale, C. J., Farrah, D., & Smith, H. E. 2006, Ultralu- minous Infrared Galaxies, ed. J. W. Mason, 285 |
| Markwardt, C. B. 2009, in Astronomical Society of the Pa- cific Conference Series, Vol. 411, Astronomical Data Anal- ysis Software and Systems XVIII, ed. D. A. Bohlender, D. Durand, & P. Dowler, 251 |
| Martin, C. L. 2005, ApJ, 621, 227 Medling, A. M., U, V., Guedes, J., et al. 2014, ApJ, 784, 70 |
| Monreal-Ibero, A., Arribas, S., & Colina, L. 2006, ApJ, 637, 138 |
| Monreal-Ibero, A., Vílchez, J. M., Walsh, J. R., & Muñoz- Tuñón, C. 2010, A&A, 517, A27 |
| Moustakas, J. & Kennicutt, Jr., R. C. 2006, ApJS, 164, 81 Osterbrock, D. E. & Ferland, G. J. 2006, Astrophysics of |
| gaseous nebulae and active galactic nuclei |
| Petric, A. O., Armus, L., Howell, J., et al. 2011, ApJ, 730, 28 |
| Pettini, M. & Pagel, B. E. J. 2004, MNRAS, 348, L59 Piqueras López, J., Colina, L., Arribas, S., Alonso-Herrero, |
| Press, W. H., Teukolsky, S. A., Vetterling, W. T., & Flan- nery, B. P. 2002, Numerical recipes in C++ : the art of scientific computing |
| Rich, J. A., Dopita, M. A., Kewley, L. J., &Rupke, D. S. N. 2010, ApJ, 721, 505 |
| Rich, J. A., Kewley, L. J., & Dopita, M. A. 2011, ApJ, 734, 87 |
| Rich, J. A., Torrey, P., Kewley, L. J., Dopita, M. A., & Rupke, D. S. N. 2012, ApJ, 753, 5 |
| Rodríguez-Zaurín, J., Arribas, S., Monreal-Ibero, A., et al. 2011, A&A, 527, A60 |
| Rupke, D. S., Veilleux, S., & Sanders, D. B. 2005, ApJS, 160, 115 |
| Rupke, D. S. N., Kewley, L. J., &Barnes, J. E. 2010, ApJL, 710, L156 |
| Rupke, D. S. N., Kewley, L. J., & Chien, L.-H. 2010b, ApJ, 723, 1255 |
| Sanders, D. B. & Mirabel, I. F. 1996, ARA&A, 34, 749 Sanders, D. B., Soifer, B. T., Elias, J. H., et al. 1988, ApJ, 325, 74 |
| Sharp, R. G. & Bland-Hawthorn, J. 2010, ApJ, 711, 818 Stierwalt, S., Armus, L., Surace, J. A., et al. 2013, ApJS, |
| 206, 1 Sutherland, R. S. & Dopita, M. A. 1993, ApJS, 88, 253 |
| Toomre, A. 1977, in Evolution of Galaxies and Stellar Pop- ulations, ed. B. M. Tinsley & R. B. G. Larson, D. Camp- bell, 401 |
| Torrey, P., Cox, T. J., Kewley, L., & Hernquist, L. 2012, ApJ, 746, 108 U, V., Medling, A., Sanders, D., et al. 2013, ApJ, 775, 115 |
| Veilleux, S. & Osterbrock, D. E. 1987, ApJS, 63, 295 Wilson, A. S. & Heckman, T. M. 1985, in |
| Astrophysics of Active Galaxies and Quasi-Stellar Objects, ed. J. |
| S. Miller, 39-109 |
## 14 S. Leslie et al.
```
Yuan, T.-T., Kewley, L. J., & Sanders, D. B. 2010, ApJ,
709, 884
Zakamska, N. L. 2010, Nature, 465, 60
7
9
```
|
10.1093/mnras/stu1547
|
[
"Sarah Leslie",
"Jeffrey Rich",
"Lisa Kewley",
"Michael Dopita"
] |
2014-08-04T19:59:23+00:00
|
2014-08-04T19:59:23+00:00
|
[
"astro-ph.GA"
] |
The energy source and dynamics of infrared luminous galaxy ESO 148-IG002
|
ESO 148-IG002 represents a transformative stage of galaxy evolution,
containing two galaxies at close separation which are currently coalescing into
a single galaxy. We present integral field data of this galaxy from the ANU
Wide Field Spectrograph (WiFeS). We analyse our integral field data using
optical line ratio maps and velocity maps. We apply active galactic nucleus
(AGN), star-burst and shock models to investigate the relative contribution
from star-formation, shock excitation and AGN activity to the optical emission
in this key merger stage. We find that ESO 148-IG002 has a flat metallicity
gradient, consistent with a recent gas inflow. We separate the line emission
maps into a star forming region with low velocity dispersion that spatially
covers the whole system as well as a southern high velocity dispersion region
with a coherent velocity pattern which could either be rotation or an
AGN-driven outflow, showing little evidence for pure star formation. We show
that the two overlapping galaxies can be separated using kinematic information,
demonstrating the power of moderate spectral resolution integral field
spectroscopy.
|
1408.0791v2
|
Nuclear Physics A
Nuclear Physics A 00 (2014) 1-4
## Highp T Direct Photon Azimuthal Correlation Measurements
Ahmed M. Hamed (for the STAR Collaboration) 1
University of Mississippi, Oxford, USA Texas A & MUniversity, College Station, USA
## Abstract
The azimuthal correlations of direct photons ( γ dir ) with high transverse momentum ( p T ), produced at mid-rapidity ( | η γ dir | < 1) in Au + Au collisions at center-of-mass energy √ s NN = 200 GeV, are measured and compared to those of neutral pions ( π 0 ) in the same kinematic range. The measured azimuthal elliptic anisotropy of direct photon, v γ dir 2 ( p T ), at high p T (8 < p γ dir T < 20 GeV / c ) is found to be smaller than that of π 0 and consistent with zero when using the forward detectors (2 4 . < η < | | 4 0) in reconstructing . the event plane. The associated charged hadron spectra recoiled from γ dir show more suppression than those recoiled from π 0 ( I γ dir -h ± < I π 0 -h ± ) in the new measured kinematic range 12 < p γ dir ,π 0 < 24 GeV / c and 3 < p assoc < 24 GeV / c .
AA AA T T
Keywords: Electromagnetic probes, high- p T direct photons, STAR
## 1. Introduction
A major goal of measurements at the Relativistic Heavy Ion Collider (RHIC) is to quantify the properties of the QCD matter created in heavy-ion collisions at high energy [1]. Unlike quarks and gluons, photons do not fragment into hadrons and can be directly observed as a final state particle. Furthermore, due to their negligible coupling to the QCD matter in contrast to hadrons, direct photons are considered as a calibrated probe for the QCD medium.
The previous measurements at RHIC indicate unexpected finite values of azimuthal elliptic anisotropy parameter v 2 of charged hadrons at highp T [2]. The measured v 2 at highp T is beyond the applicability of hydrodynamic models, and the path-length dependence of jet quenching is the only proposed explanation of v 2 at highp T [3]. The v γ dir 2 measurement would provide a gauge for the energy loss at highp T .
The highp T γ dir sample unbiased spatial distribution of the hard scattering vertices in the QCD medium [4], in contrast to hadrons which su ff er from the geometric biases. Therefore, a comparison between the spectra of the awayside charged hadrons associated with γ dir vs. π 0 can provide a benchmark for the energy loss and its dependence on the path-length. Although the previous measurements have indicated similar level and pattern of suppression for the away-side of γ dir and π 0 [5], the current work explores a softer region in the fragmentation functions where a more significant di ff erence is expected [6].
## 2. Analysis and Results
## 2.1. Electromagnetic neutral clusters
The STAR detector is well suited for measuring azimuthal angular correlations due to the large coverage in pseudorapidity and full coverage in azimuth ( φ ). While the Barrel Electromagnetic Calorimeter (BEMC) [7] measures
1 A list of members of the STAR Collaboration and acknowledgements can be found at the end of this issue.
the electromagnetic energy with high resolution, the Barrel Shower Maximum Detector (BSMD) provides fine spatial resolution and enhances the rejection power for the hadrons. The Time Projection Chamber (TPC: | η < | 1) [8] identifies charged particles, measures their momenta, and allows for a charged-particle veto cut with the BEMC matching. The Forward Time Projection Chamber (FTPC: 2 4 . < η < | | 4 0) [9] is used to measure the charged particles' momenta . and to reconstruct the event plane angle. Using the BEMC to select events ( i.e. 'trigger') with highp T γ , the STAR experiment collected an integrated luminosity of 23 p b -1 of p + p collisions in 2009 and 973 µ b -1 of Au + Au collisions in 2011. In this analysis, events having a primary vertex within ± 55 cm of the center of the TPC along the beamline in Au + Au and ± 80 cm in p + p are selected. In addition, each event must have at least one electromagnetic cluster with ET > 8 GeV for the event plane correlation analysis and ET > 12 GeV for the charged hadron correlation analysis. More than 97% of these clusters have deposited energy greater than 0.5 GeV in each layer of the BSMD. A trigger tower is rejected if it has a track with p > 3 0 GeV . / c pointing to it, which reduces the number of the electromagnetic clusters by only ∼ 7%.
## 2.2. v 2 of neutral particles
The v 2 is determined using the standard method [10]:
$$v _ { _ { 2 } } ( p _ { _ { r } } ) = \langle \langle \cos 2 ( \phi _ { p _ { r } } - \psi _ { _ { \text{EP} } } ) \rangle, \\ \dots \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot$$
where the brackets denote statistical averaging over particles and events, φ p T is the azimuthal angle of the neutral particle with certain value of p T , and ψ EP is the azimuthal angle of the event plane. The event plane is reconstructed from charged particles, within the detector acceptance, with p T < 2 GeV / c , and determined by
$$\psi _ { \text{EP} } = \frac { 1 } { 2 } \tan ^ { - 1 } ( \frac { \sum _ { i } \sin ( 2 \phi _ { i } ) } { \sum _ { i } \cos ( 2 \phi _ { i } ) } ),$$
where φ i are the azimuthal angles of all the particles used to define the event plane. In this analysis, the charged-track quality criteria are similar to those used in previous STAR analyses [11]. The event plane is measured using two di ff erent detectors in their pseudorapidity coverage: 1) using all the selected tracks inside the TPC, and 2) using all tracks inside the FTPC in order to reduce the 'non-flow' contributions (azimuthal correlations not related to the event plane). Since the event plane is only an approximation to the true reaction plane, the observed correlation is divided by the event plane resolution. The event plane resolution is estimated using the sub-event method in which the full event is divided up randomly into two sub-events as described in [10]. Biases due to the non-uniform acceptance of the detector are removed according to the method in [12].
## 2.3. Azimuthal correlations of a neutral trigger particle with charged hadrons
The azimuthal correlations of a neutral trigger particle with charged hadrons, measured as the number of associated particles per neutral cluster per ∆ φ ('correlation functions'), are used in both p + p and Au + Au collisions to determine the (jet) associated particle yields in the near- ( ∆ φ ∼ 0) and away-sides ( ∆ φ ∼ π ). The near- and away-side yields, Y n and Y a , of associated particles per trigger are extracted by integrating the (1 / Ntrig ) dN d / ( ∆ φ ) distributions over | ∆ φ | ≤ 0.63 and | ∆ φ -π | ≤ 0.63, respectively. The yield is corrected for the tracking e ffi ciency of charged particles as a function of event multiplicity.
## 2.4. Transverse shower profile analysis
A crucial part of the analysis is to discriminate between showers from γ dir and two close γ 's from highp T π 0 symmetric decays. At p π 0 T ∼ 8 GeV / c , the angular separation between the two γ 's resulting from a π 0 decay is small, but a π 0 shower is generally broader than a single γ shower. The BSMD is capable of 2 γ / 1 γ separation up to p π 0 T ∼ 24 GeV / c due to its high granularity ( ∆ η ∼ 0 007, . ∆ φ ∼ 0 007). . The shower shape is quantified as the cluster energy, measured by the BEMC, normalized by the position-weighted energy moment, measured by the BSMDstrips [5]. The shower profile cuts were tuned to obtain a nearly γ dir -free ( π 0 rich ) sample and a sample rich in γ dir ( γ rich ). Since the shower-shape analysis is only e ff ective for rejecting two close γ showers, the γ rich sample contains a mixture of direct photons and contamination from fragmentation photons ( γ f rag ) and photons from asymmetric hadron ( π 0 and η ) decays.
The v γ rich 2 and v π 0 2 are measured as discussed in section 2.2 and the away (near)-side yields of associated particles per γ rich and π 0 rich triggers ( Y a n ( ) γ rich + h and Y a n ( ) π 0 rich + h ) are measured as discussed in section 2.3.
## 2.5. v 2 of direct photons
Assuming zero near-side yield for γ dir triggers and a sample of π 0 rich free of γ dir , the v γ dir 2 is given by:
$$v _ { 2 } ^ { \gamma _ { d r r } } = \frac { v _ { 2 } ^ { \gamma _ { r i c h } } - \mathcal { R } v _ { 2 } ^ { \pi ^ { 0 } _ { r i c h } } } { 1 - \mathcal { R } }, \\. \hat { \ }. \hat { \ }. \hat { \ }. \hat { \ }. \hat { \ }. \hat { \ }. \hat { \ }. \hat { \ }. \hat { \ }. \hat { \ }. \hat { \ }. \hat { \ }. \hat { \ }. \hat { \ }.$$
where R = N rich π 0 N rich γ , and the numbers of π 0 rich and γ rich triggers are represented by N rich π 0 and N rich γ respectively. Although the R quantity approximates all background triggers in the γ rich sample to the measured π 0 rich triggers, all background to γ dir is subtracted assuming that all background triggers have the same correlation function as the π 0 rich sample [5]. The value of R is measured in [5] and found to be ∼ 30% in central Au + Au. In Eq. 3 all background sources for γ dir are assumed to have the same v 2 as the measured π 0 .
T
T
T
Figure 1. v 2 of π 0 , and γ dir in 10-40% centrality of Au + Au collisions at √ s NN = 200 GeV using TPC. Boxes show the systematic errors.
Figure 2. v 2 of π 0 , and γ dir in 10-40% centrality of Au + Au collisions at √ s NN = 200 GeV using FTPC. Boxes show the systematic errors.
Figure 3. The z T dependence of I AA for γ dir and π 0 triggers in Au + Au 0-10% centrality. Boxes show the systematic errors.
Figures 1 and 2 show the v π 0 2 and v γ dir 2 for (8 < p γ dir T < 20 GeV / c ) from Au + Au data set of Run 2011 using the TPC ( | η | < 1) and FTPC (2 4 . < η < | | 4 0). . The results of Fig. 1 are consistent with those from di ff erent STAR data set of Run 2007 [13], and the results of Fig. 2 agree with other measurements [14, 15]. While using the FTPC in determining the event plane (Fig. 2) the v γ dir 2 is consistent with zero. Assuming the dominant source of direct photons is prompt hard production, the zero value implies no remaining bias in the event-plane determination. Accordingly, the measured value of v π 0 2 would be the e ff ect of path-length dependent energy loss.
## 2.6. Extraction of γ dir associated yields
Assuming zero near-side yield for γ dir triggers and a sample of π 0 rich free of γ dir , the away-side yield of hadrons correlated with the γ dir is extracted as
$$\Lambda _ { d i r } \, \stackrel { \sim } { \sim } \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \delta \, \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \ \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ | \eta _ _ { \eta _ { \eta _ { \eta _ { \eta _ _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ { \eta _ _ { \eta _ { \eta _ { \eta _ _ { \eta _ { \eta _ { \eta _ { \ | _ { \eta _ { \eta _ _ { \eta _ { \eta _ _ { \eta _ { \eta _ { \ | _ { \eta _ { \ | _ { \eta _ { \eta _ _ { \eta _ _ { \eta _ _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ - \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { | _ { \ | _ { | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ - \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ { \ | _ | _ { \ | _ { \ | _ { \ | _ | _ { \ | _ | _ | _ { \ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ | _ _ | _ | _ | _ | _ | _ | _ _ | _ | _ _ | _ | _ | _ _ | _ | _ | _ _ | _ | _ _ | _ | _........................................................................................................................................................................................................$$
Here, Y a n ( ) γ rich + h and Y a n ( ) π 0 rich + h are the away (near)-side yields of associated particles per γ rich and π 0 rich triggers, respectively. The ratio R is equivalent to the fraction of 'background' triggers in the γ rich trigger sample, and N dir γ and N rich γ are the
numbers of γ dir and γ rich triggers, respectively. The value of R is found to be ∼ 55% in p + p and decreases to ∼ 30% in central Au + Au with little dependence on p trig T . All background to γ dir is subtracted with the assumption that the background triggers have the same correlation function as the π 0 rich sample.
In order to quantify the away-side suppression, we calculate the quantity I AA , which is defined as the ratio of the integrated yield of the away-side associated particles per trigger particle in Au + Au to that in p + p collisions. The values of I γ dir -h ± AA and I π 0 -h ± AA , as shown in Fig. 3, are z T ( z T = p assoc T / p trig T ) independent in agreement with results of [5] where the recoiled parton from γ dir and π 0 experience constant fractional energy loss in the QCD medium. It is also observed that the charged hadron spectra recoiled from γ dir show unexpectedly more suppression than those recoiled from π 0 ( I γ dir -h ± AA < I π 0 -h ± AA ) within the covered kinematics range 12 < p γ dir ,π 0 T < 24 GeV / c and 3 < p assoc T < 24 GeV / c .
## 3. Conclusions
The STAR experiment has reported the first v γ dir 2 at highp T (8 < p γ dir T < 20 GeV / c ), and explored new kinematic range (12 < p γ dir ,π 0 T < 24 GeV / c ) and (3 < p assoc T < 24 GeV / c ) for I AA measurements of γ dir -h correlations at √ s NN = 200 GeV. Using the mid-rapidity detectors in determining the event plane, the measured value of v γ dir 2 is non-zero, and is probably due to biases in the event-plane determination. Using the forward detectors in determining the event plane could eliminate remaining biases, and the measured v γ dir 2 is consistent with zero. The zero value of v γ dir 2 suggests a negligible contribution of jet-medium photons [17], and negligible e ff ects of γ f rag [16] on the v γ dir 2 over the covered kinematics range. The measured finite value of v π 0 2 , using the forward detectors in determining the event plane, is apparently due to the path-length dependence of energy loss. The γ dir -h correlation results indicate that the associated charged hadron spectra recoiled from γ dir show more suppression than those recoiled from π 0 ( I γ dir -h ± AA < I π 0 -h ± AA ) within the covered kinematic range, in contrast to the theoretical predictions [18]. The disagreement with the theoretical expectations may indicate that the lost energy is distributed to lower p T of the associated particles in the case of a γ dir trigger than a π 0 trigger. To further test this, one must explore the region of low p assoc T and z T .
## References
- [1] STAR Collaboration, J. Adams et al , Nucl. Phys. A 757 , 102 (2005).
- [2] STAR Collaboration, J. Adams et al , Phys. Rev. Lett. 93 , 252301 (2004).
- [3] E. V. Shuryak, Phys. Rev. C 66 , 027902 (2002).
- [4] X.-N.Wang, Z. Huang and I. Sarcevic, Phys. Rev. Lett. 77 , 231 (1996).
- [5] STAR Collaboration, B. I. Abelev et al , Phys. Rev. C 82 , 034909 (2010).
- [6] T. Renk, Phys. Rev. C 74 , 034906 (2006).
- [7] M. Beddo et al , Nucl. Instrum. Meth. A 499 , 725 (2003).
- [8] M. Anderson et al , Nucl. Instrum. Meth. A 499 , 659 (2003).
- [9] K. H. Ackermann et al , Nucl. Instrum. Meth. A 499 , 713 (2003).
- [10] A. M. Poskanzer and S. A. Voloshin , Phys. Rev. C 58 , 1671 (1998).
- [11] STAR Collaboration, B. I. Abelev et al , Phys. Rev. C 77 , 054901 (2008).
- [12] E877 Collaboration, J. Barrette et al , Phys. Rev. C 56 , 3254 (1997).
- [13] A. Hamed (STAR Collaboration) J. Phys: Conf. Ser. 270 , 012010 (2011).
- [14] D. Lohner (Alice Collaboration) J. Phys.: Conf. Ser. 446 , 012028 (2013).
- [15] PHENIX Collaboration, A. Adare et al , Phys. Rev. Lett. 109 , 122302 (2012).
- [16] B. G. Zakharov, JETP Lett. 80 , 1 (2004).
- [17] R. J. Fries, B. M¨ller, and D. K. Srivastava, Phys. Rev. Lett. u 90 , 132301 (2003).
- [18] H. Zhang et al , Nucl. Phys. A 830 , 443c (2009).
|
10.1016/j.nuclphysa.2014.08.091
|
[
"Ahmed M. Hamed"
] |
2014-08-04T04:51:09+00:00
|
2014-08-27T00:42:32+00:00
|
[
"nucl-ex"
] |
High-$p_{_{T}}$ Direct Photon Azimuthal Correlation Measurements
|
The azimuthal correlations of direct photons ($\gamma_{_{dir}}$) with high
transverse momentum ($p_{_{T}}$), produced at mid-rapidity
($|\eta^{\gamma_{_{dir}}}|<1$) in Au+Au collisions at center-of-mass energy
$\sqrt{s_{_{NN}}}=200$~GeV, are measured and compared to those of neutral pions
($\pi^{0}$) in the same kinematic range. The measured azimuthal elliptic
anisotropy of direct photon, $v_{_{2}}^{\gamma_{_{dir}}}(p_{_{T}})$, at high
$p_{_{T}}$ ($8< p_{_{T}}^{\gamma_{_{dir}}}<20$~GeV/$c$) is found to be smaller
than that of $\pi^{0}$ and consistent with zero when using the forward
detectors ($2.4 <|\eta|< 4.0$) in reconstructing the event plane. The
associated charged hadron spectra recoiled from $\gamma_{_{dir}}$ show more
suppression than those recoiled from $\pi^{0}$
$(I_{AA}^{\gamma_{_{dir}}-h^{\pm}} < I_{AA}^{\pi^{0}-h^{\pm}})$ in the new
measured kinematic range $12< p_{_{T}}^{\gamma_{_{dir}},\pi^{0}}<24$~GeV/$c$
and $3< p_{_{T}}^{assoc}<24$~GeV/$c$.
|
1408.0792v2
|
## Probing anomalous ttZ ¯ interactions with rare meson decays
Joachim Brod , Admir Greljo , Emmanuel Stamou , Patipan Uttayarat a ∗ b † c ‡ a,d §
- a Department of Physics, University of Cincinnati, Cincinnati, Ohio 45221, USA
b Joˇ zef Stefan Institute, Jamova 39, 1000 Ljubljana, Slovenia
- c Department of Particle Physics and Astrophysics, Weizmann Institute of Science, Rehovot 76100, Israel
d Department of Physics, Srinakharinwirot University, Bangkok 10110, Thailand
## Abstract
Anomalous couplings of the Z boson to top quarks are only marginally constrained by direct searches and are still sensitive to new particle dynamics at the TeV scale. Employing an effective field theory approach we consider the dimension-six operators which generate deviations from the standard-model vector and axial-vector interactions. We show that rare B and K meson decays together with electroweak precision observables provide strong constraints on these couplings. We also consider constraints from t-channel single-top production.
## 1 Introduction
In the case of no direct observation of new particles at the Large Hadron Collider (LHC), precision measurements of the Standard Model (SM) interactions provide an alternative route to discover New Physics (NP). The measurement of the ttZ ¯ production cross section at the LHC [1,2] offers a direct test of anomalous ttZ ¯ couplings. In a recent publication, Schulze and R¨ntsch presented the calculation and analysis of o ttZ ¯ production at next-toleading order in QCD [3]. They used current and future-projected LHC data to constrain anomalous vector and axial-vector couplings of the Z boson to top quarks.
∗ [email protected]
‡ [email protected]
† [email protected]
§ [email protected]
In this work we study constraints on anomalous ttZ ¯ couplings arising from electroweak precision observables (EWPO), rare B and K meson decays, and t-channel single-top production. The ttZ ¯ couplings contribute to the rate of rare meson decays and to EWPO via radiative corrections. Thus, in the absence of contributions to these observables from other sources, rare meson decays and EWPO offer strong bounds on anomalous ttZ ¯ couplings. In fact, we show that the indirect constraints exceed the direct bounds, even after the projected high-luminosity upgrade of the LHC [3].
We focus on the rare decays B s → µ + µ -, K + → π + νν ¯ and K L → π νν 0 ¯. These decays proceed via Z -penguin diagrams 1 which, in the SM, are dominated by the top-quark loop. This makes these processes especially sensitive to anomalous ttZ ¯ couplings. For this reason, the current observation of the B s → µ + µ -rate already competes with constraints from EWPO and the expected future improvements in the measurements of rare meson decay rates are especially encouraging as they may provide the first evidence for anomalous ttZ ¯ interactions.
The use of rare decays to obtain bounds on anomalous Z couplings has already been discussed in Refs. [4,5]. The authors derived constraints on bbZ ¯ couplings by relating them to flavor-changing couplings within the minimal flavor violation (MFV) framework. In this work, we consider constraints arising from operator mixing, induced by electroweak loops, within an effective field theory (EFT) approach. This strategy has already been proposed in Refs. [6,7] for flavor-changing neutral currents (FCNCs) involving the top quark, as well as for constraining anomalous Wtb couplings [8,9].
The remainder of the paper is structured as follows: In Sec. 2 we define the appropriate EFT approach to study possible deviations in ttZ ¯ couplings. In Sec. 3 we revisit the main constraints from EWPO and t-channel single-top production, and present our analysis of rare meson decays. In Sec. 4 we discuss the numerical results and conclude.
## 2 Effective field theory
Our starting point is the SM augmented by gauge-invariant dimension-six operators constructed out of the SM fields. This may be thought of as the effective field theory after integrating out NP at some scale Λ significantly larger than the electroweak scale. We write the effective Lagrangian as
$$\mathcal { L } ^ { \text{eff} } = \mathcal { L } ^ { \text{SM} } + \sum _ { k } \frac { 1 } { \Lambda ^ { 2 } } C _ { k } Q _ { k } \,.$$
We are interested in the subset of operators that induce modifications of the vector and axial-vector ttZ ¯ couplings at tree-level. They will have the main impact on direct searches for anomalous ttZ ¯ couplings. If we assume no correlations between Wilson coefficients, a
1 The photon penguin does not contribute to B s → µ + µ -due to conservation of the vector current.
complete set of such operators is given by (cf. [10,11]):
$$\text{sun operators is given by (cfi. [1U,11])} \colon \\ Q _ { \phi q, 3 3 } ^ { ( 3 ) } & \equiv ( \phi ^ { \dagger } i \stackrel { \leftrightarrow } { D }$$
These operators contain the Higgs doublet φ , the left-handed third-generation quark doublet Q L, 3 , and the right-handed top quark t R . Moreover, σ a are the Pauli matrices and D µ is the SM gauge-covariant derivative and we defined
$$( \phi ^ { \dagger } i \stackrel { \leftrightarrow } { D _ { \mu } } \phi ) & = i \phi ^ { \dagger } ( D _ { \mu } \phi ) - i ( D _ { \mu } \phi ) ^ {$$
so that the operators are manifestly Hermitian. Therefore, all Wilson coefficients considered in this work are real.
In order to include in Eq. (2.2) all operators that induce ttZ ¯ at tree-level, we have chosen the basis in which the up-type quark Yukawa matrix is diagonal. Accordingly, we have
$$Q _ { L, 3 } \equiv \left [ \begin{array} { c } t _ { L } \\ \sum _ { j } V _ { 3 j } d _ { L, j } \end{array} \right ] \,. \quad \quad \quad$$
The fields t L R ( ) , d L,j are mass eigenstates and V is the Cabibbo-Kobayashi-Maskawa (CKM) matrix. A generic NP model can generate FCNC transitions in the up-quark sector; to take these effects into account we have to consider additional operators involving the first and second quark generation. However, in the restricted, but phenomenologically well-motivated, framework of MFV [12] these operators are suppressed with respect to those in Eq. (2.2) by elements of the CKM matrix. Thus, the resulting bounds on ttZ ¯ couplings are negligible. Moreover, in the limit that only the top-quark Yukawa is non-vanishing in the MFV spurion counting, such additional operators are absent. For simplicity, we will use this approximation in the following, but we will comment on the effect of keeping a large bottom-quark Yukawa in Sec. 4. This discussion will cover MFV models with down-type quark alignment, i.e. models in which, at tree-level, only up-type quark FCNCs are generated by NP.
In our analysis, only the operators in Eq. (2.2) receive non-zero initial conditions at the scale Λ. However, electroweak corrections and corrections involving the SM top-quark Yukawa coupling y t will induce mixing into the following additional operators relevant for our analysis:
$$\times \psi \quad & \quad \sim \quad \sim \quad \cdot \\ & Q _ { \phi q, i i } ^ { ( 3 ) } & \equiv ( \phi ^ { \dagger } i \stackrel { \leftrightarrow } { D _$$
where i = 1 2 and , j = 1 2 3. , , We can safely neglect the remaining Yukawa couplings in the mixing calculation. The operators in Eq. (2.5) will lead to additional bounds on the anomalous ttZ ¯ couplings.
To study the effects of these operators on rare decays we need to know their Wilson coefficients at the electroweak scale where we match to the five-flavor effective theory. The running of the Wilson coefficients is given by the renormalization group equations (RGE). The matching is most conveniently performed in the broken phase where the Higgs field obtains its vacuum-expectation value, v . The relevant part of the effective Lagrangian in the broken phase reads
$$\mathcal { L } ^ { \text{eff} } \supset \mathcal { L } ^ { \text{SM} } + \frac { v ^ { 2 } } { \Lambda ^ { 2 } } ( \mathcal { L } _ { Z } ^ { \text{diag} } + \mathcal { L } _ { Z } ^ { \text{FCNC} } + \mathcal { L } _ { W } ) \,.$$
Using the conventions of Ref. [13], the flavor-diagonal neutral-current part of the Lagrangian is given by
$$\rangle \quad \rangle \quad \rangle \quad \rangle \quad \rangle \\ \quad \mathcal { L } _ { Z } ^ { \text{diag} } = \frac { e } { 2 s _ { w } c _ { w } } Z _ { \mu } \left [ & ( C ^ { ( 3 ) } _ { \phi q, 3 3 } - C ^ { ( 1 ) } _ { \phi q, 3 3 } ) \, \bar { t } _ { L } \gamma ^ { \mu } t _ { L } - C _ { \phi u, 3 3 } \, \bar { t } _ { R } \gamma ^ { \mu } t _ { R } \\ & - \sum _ { i = 1, 2, 3 } V ^ { * } _ { i 3 } V _ { i 3 } ( C ^ { ( 3 ) } _ { \phi q, i i } + C ^ { ( 1 ) } _ { \phi q, i i } ) \, \bar { b } _ { L } \gamma ^ { \mu } b _ { L } \right ].$$
This Lagrangian induces a shift of the SM Z -boson couplings to third-generation quarks. We follow the conventions in Ref. [14] and parametrize this shift by
$$\delta g _ { L } ^ { t } & = \frac { v ^ { 2 } } { 2 \Lambda ^ { 2 } } ( C _ { \phi q, 3 3 } ^ { ( 3 ) } - C _ { \phi q, 3 3 } ^ { ( 1 ) } ) \,, \quad \delta g _ { R } ^ { t } = - \frac { v ^ { 2 } } { 2 \Lambda ^ { 2 } } C _ { \phi u, 3 3 } \,, \\ \delta g _ { L } ^ { b } & = - \frac { v ^ { 2 } } { 2 \Lambda ^ { 2 } } \sum _ { i = 1, 2, 3 } V _ { i 3 } ^ { * } V _ { i 3 } ( C _ { \phi q, i i } ^ { ( 3 ) } + C _ { \phi q, i i } ^ { ( 1 ) } ) \,.$$
The operators Q (1) φq, 33 and Q φu, 33 involve only neutral quark currents, whereas Q (3) φq, 33 will necessarily induce deviations also in the effective W couplings. These charged-current contributions include
$$\mathcal { L } _ { W } = \frac { e } { \sqrt { 2 } s _ { w } } C _ { \phi q, 3 3 } ^ { ( 3 ) } \, \sum _ { i } \, V _ { 3 i } \, W _ { \mu } ^ { + } \, \bar { t } _ { L } \gamma ^ { \mu } d _ { L, i } + \text{h.c.} \,.$$
The FCNC contributions relevant for us involve only down-type quarks and are given by glyph[negationslash]
$$\mathcal { L } _ { Z } ^ { \text{FCNC} } = - \frac { e } { 2 s _ { w } c _ { w } } \left ( C _ { \phi q, 3 3 } ^ { ( 3 ) } + C _ { \phi q, 3 3 } ^ { ( 1 ) } \right ) \sum _ { i \neq j } V _ { 3 i } ^ { * } V _ { 3 j } \, Z _ { \mu } \, \bar { d } _ { L, i } \gamma ^ { \mu } d _ { L, j } \,.$$
Note that Eq. (2.10) also receives contributions from the operators in Eq. (2.5), but these contributions cancel via the unitarity of the CKM matrix (see Sec. 3.3) and we do not include them here. Finally, we remark that the operators in Eq. (2.2) do not induce modifications of couplings involving the Higgs boson.
Constraints from flavor physics and EWPO suggest the absence of both tree-level downtype FCNC transitions and modifications of the bbZ ¯ couplings if NP enters at the TeV scale. Thus, we take C (3) φq, 33 + C (1) φq, 33 = 0 at the scale Λ. This relation can be realized in models with additional vector-like quarks [15]. Moreover, these models do not generate four-fermion operators at tree level. Under these assumptions, rare meson decays and the effective bbZ couplings are sensitive to anomalous ttZ couplings generated by the operators in Eq. (2.2).
## 3 Phenomenology
## 3.1 t-channel single top-quark production
The operator Q (3) φq, 33 induces a tree-level correction to the Wtb coupling given by Eq. (2.9). Both the ATLAS and the CMS collaborations have recently measured the t-channel single top-quark production cross section [16, 17], which constitutes a direct measurement of the Wtb coupling. In particular, they report the measured cross-section normalized with respect to the SM expectation, √ σ t-ch ( ) t /σ theo t-ch ( ) t ≡ R σ . The measured values reported by the two collaborations are
$$R _ { \sigma } = \begin{cases} 0. 9 7 \ \pm 0. 1 0 & [ \text{ATLAS} ] \\ 0. 9 9 8 \pm 0. 0 4 1 & [ \text{CMS} ] \\.. \. \. \. \. \.. \.. \.. \.. \.. \.. \.. \.. \end{cases} \quad ( 3. 1 )$$
where we have combined the experimental and theoretical errors in quadrature. In our analysis this ratio is given by
$$R _ { \sigma } = 1 + v ^ { 2 } C _ { \phi q, 3 3 } ^ { ( 3 ) } / \Lambda ^ { 2 } \,.$$
Combining the measurements, we find
$$v ^ { 2 } C _ { \phi q, 3 3 } ^ { ( 3 ) } / \Lambda ^ { 2 } = - 0. 0 0 6 \pm 0. 0 3 8 \,.$$
## 3.2 Electroweak precision observables
We saw in Sec. 2 that it is phenomenologically reasonable to assume C (3) φq, 33 + C (1) φq, 33 = 0 at the scale Λ. In particular, this implies δg b L (Λ) = 0. Operator mixing will reintroduce a non-zero δg b L at the electroweak scale. The relevant parts of the RGE are
$$\text{non-zero} \, \delta g _ { L } ^ { ( 1 ) } & \text{ at the electroweak scale. The relevant parts of the RGE are} \\ \mu \frac { d } { d \mu } C _ { \phi q, 3 3 } ^ { ( 3 ) } & = \frac { y _ { L } ^ { 2 } } { 1 6 \pi ^ { 2 } } \left ( 8 C _ { \phi q, 3 3 } ^ { ( 3 ) } - 3 C _ { \phi q, 3 3 } ^ { ( 1 ) } \right ) - \frac { g _ { L } ^ { 2 } } { 1 6 \pi ^ { 2 } } \frac { 1 1 } { 3 } C _ { \phi q, 3 3 } ^ { ( 3 ) } \,, \\ \mu \frac { d } { d \mu } C _ { \phi q, 3 3 } ^ { ( 1 ) } & = \frac { y _ { L } ^ { 2 } } { 1 6 \pi ^ { 2 } } \left ( - 9 C _ { \phi q, 3 3 } ^ { ( 3 ) } + 1 0 C _ { \phi q, 3 3 } ^ { ( 1 ) } - C _ { \phi u } \right ) + \frac { g _ { L } ^ { 2 } } { 1 6 \pi ^ { 2 } } \frac { 1 } { 9 } \left ( 5 C _ { \phi q, 3 3 } ^ { ( 1 ) } + 4 C _ { \phi u } \right ) \,, \\ \mu \frac { d } { d \mu } C _ { \phi q, 1 1 } ^ { ( 3 ) } & = \mu \frac { d } { d \mu } C _ { \phi q, 2 2 } ^ { ( 3 ) } = \frac { g _ { L } ^ { 2 } } { 1 6 \pi ^ { 2 } } 2 C _ { \phi q, 3 3 } ^ { ( 3 ) } \,, \\ \mu \frac { d } { d \mu } C _ { \phi q, 1 1 } ^ { ( 1 ) } & = \mu \frac { d } { d \mu } C _ { \phi q, 2 2 } ^ { ( 1 ) } = \frac { g _ { L } ^ { 2 } } { 1 6 \pi ^ { 2 } } \frac { 2 } { 9 } ( C _ { \phi q, 3 3 } ^ { ( 1 ) } + 2 C _ { \phi u } ) \,,$$
Figure 1: Diagrams that induce non-zero mixing proportional to gauge couplings of Q (3) φq, 33 into itself. All other diagrams cancel among themselves or against contributions from the field renormalization. Note that there are additional contributions from penguin-type insertions. As a check, we have calculated the mixing both for external states including a B gauge boson (left) and for external states not including a B gauge boson (right) and find the same result.
with y t , g 1 , and g 2 the SM top-Yukawa coupling and the electroweak gauge couplings in the conventions of Ref. [13], respectively. The terms proportional to the top-quark Yukawa coupling have been presented in Ref. [18] and those proportional to the gauge couplings in Ref. [19]. However, our result for the latter differs from the one of Ref. [19] in the last term of the first line in Eq. (3.4). We have calculated the mixing in a general R ξ gauge and find a gauge-independent result. While we agree with Ref. [19] on the terms arising from penguin-type insertions of the operators, we find an additional contribution for the mixing of Q (3) φq, 33 into itself. It arises from the diagrams shown in Figure 1. An additional non-trivial check of our result can be found in the next section . 2
Solving the RGE in Eqs. (3.4), we determine the Wilson coefficients at the electroweak scale:
$$\text{scale} \colon \\ C ^ { ( 3 ) } _ { \phi _ { q }, 3 3 } ( \mu _ { W } ) = C ^ { ( 3 ) } _ { \phi _ { q }, 3 3 } ( \Lambda ) + \left \{ \frac { y _ { t } ^ { 2 } } { 1 6 \pi ^ { 2 } } \left [ 8 C ^ { ( 3 ) } _ { \phi _ { q }, 3 3 } ( \Lambda ) - 3 C ^ { ( 1 ) } _ { \phi _ { q }, 3 3 } ( \Lambda ) \right ] - \frac { g _ { 2 } ^ { 2 } } { 1 6 \pi ^ { 2 } } \frac { 1 1 } { 3 } C ^ { ( 3 ) } _ { \phi _ { q }, 3 3 } ( \Lambda ) \right \} \log \frac { \mu _ { W } } { \Lambda } \,, \\ C ^ { ( 1 ) } _ { \phi _ { q }, 3 3 } ( \mu _ { W } ) = C ^ { ( 1 ) } _ { \phi _ { q }, 3 3 } ( \Lambda ) + \left \{ \frac { y _ { t } ^ { 2 } } { 1 6 \pi ^ { 2 } } \left [ 1 0 C ^ { ( 1 ) } _ { \phi _ { q }, 3 3 } ( \Lambda ) - 9 C ^ { ( 3 ) } _ { \phi _ { q }, 3 3 } ( \Lambda ) - C _ { \phi u } ( \Lambda ) \right ] \\ & + \frac { g _ { 1 } ^ { 2 } } { 1 6 \pi ^ { 2 } } \frac { 1 } { 9 } \left [ 5 C ^ { ( 1 ) } _ { \phi _ { q }, 3 3 } ( \Lambda ) + 4 C _ { \phi u } ( \Lambda ) \right ] \right \} \log \frac { \mu _ { W } } { \Lambda } \,, \\ \cdots \quad \cdots \quad \cdots$$
$$\begin{matrix} 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & 1 \\ \end{matrix}$$
$$C _ { \phi q, 1 1 } ^ { ( 3 ) } ( \mu _ { W } ) = C _ { \phi q, 2 2 } ^ { ( 3 ) } ( \mu _ { W } ) = \frac { g _ { 2 } ^ { 2 } } { 8 \pi ^ { 2 } } C _ { \phi q, 3 3 } ^ { ( 3 ) } ( \Lambda ) \log \frac { \mu _ { W } } { \Lambda } \,,$$
$$\left | = \frac { y _ { 2 } } { 8 \pi ^ { 2 } } C _ { \phi q, 3 3 } ^ { ( 3 ) } ( \Lambda ) \log \frac { \mu _ { W } } { \Lambda } \,,$$
$$C _ { \phi q, 1 1 } ^ { ( 1 ) } ( \mu _ { W } ) = C _ { \phi q, 2 2 } ^ { ( 1 ) } ( \mu _ { W } ) = \frac { g _ { 1 } ^ { 2 } } { 1 6 \pi ^ { 2 } } \frac { 2 } { 9 } \left [ C _ { \phi q, 3 3 } ^ { ( 1 ) } ( \Lambda ) + 2 C _ { \phi u } ( \Lambda ) \right ] \, \log \frac { \mu _ { W } } { \Lambda } \,.$$
$$( 3. 5 )$$
We used the leading-log approximation to relate the Wilson coefficient at the scale Λ to
2 The authors of Ref. [19] confirmed that this contribution was accidentally omitted in their result; however, a corresponding term appears in their anomalous dimension of Q (3) φl .
the electroweak scale µ W . Using Eq. (2.8), we find
$$\delta g _ { L } ^ { b } = - \frac { v ^ { 2 } } { 2 \Lambda ^ { 2 } } \frac { \alpha } { 4 \pi } \left \{ V _ { 3 3 } ^ { * } & V _ { 3 3 } \left [ \frac { x _ { t } } { 2 s _ { w } ^ { 2 } } \left ( 8 C _ { \phi, 3 3 } ^ { ( 1 ) } ( \Lambda ) - C _ { \phi u } ( \Lambda ) \right ) + \frac { 1 7 c _ { w } ^ { 2 } + s _ { w } ^ { 2 } } { 3 s _ { w } ^ { 2 } c _ { w } ^ { 2 } } C _ { \phi q, 3 3 } ^ { ( 1 ) } ( \Lambda ) \right ] \\ & + \left [ \frac { 2 s _ { w } ^ { 2 } - 1 8 c _ { w } ^ { 2 } } { 9 s _ { w } ^ { 2 } c _ { w } ^ { 2 } } C _ { \phi q, 3 3 } ^ { ( 1 ) } ( \Lambda ) + \frac { 4 } { 9 c _ { w } ^ { 2 } } C _ { \phi u } ( \Lambda ) \right ] \right \} \log \frac { \mu _ { W } } { \Lambda } \,. \\ & \quad \ w \, \quad \ w \,,$$
Here, we defined x t ≡ m /M 2 t 2 W and used the relation C (3) φq, 33 (Λ) + C (1) φq, 33 (Λ) = 0.
The above corrections lead to a shift in the parameter glyph[epsilon1] b , defined in Ref. [20], given by δglyph[epsilon1] b = δg b L . The authors of Ref. [3] have compared their direct constraints on anomalous ttZ ¯ couplings with the indirect constraint derived from δglyph[epsilon1] b . They use the expression for δglyph[epsilon1] b given in Ref. [21] while using the same effective operators as in our work. However, the calculation in Ref. [21] has been carried out in a different framework, namely, the nonlinearly realized electroweak chiral Lagrangian [22]. Their result therefore does not agree with ours, and the expression for δglyph[epsilon1] b in Ref. [3] should be replaced by the one in Eq. (3.6).
Quantum corrections also induce the mixing of the operators in Eq. (2.2) into Q φD in Eq. (2.5). This operator is tightly constrained by EWPO, since it leads to the universal oblique T parameter [23-25]. The mixing is given by [18,19]
$$1 6 \pi ^ { 2 } \mu \frac { d } { d \mu } \, C _ { \phi D } = \frac { 8 } { 3 } g _ { 1 } ^ { 2 } \left ( C _ { \phi q, 3 3 } ^ { ( 1 ) } + 2 C _ { \phi u, 3 3 } \right ) + 2 4 y _ { t } ^ { 2 } \left ( C _ { \phi q, 3 3 } ^ { ( 1 ) } - C _ { \phi u, 3 3 } \right ) \,.$$
Accordingly, we obtain the following expression for the T parameter (cf. Ref. [26])
$$T & = - \frac { v ^ { 2 } } { 2 \alpha \Lambda ^ { 2 } } C _ { \phi D } \\ & = - \frac { v ^ { 2 } } { \Lambda ^ { 2 } } \left [ \frac { 1 } { 3 \pi c _ { w } ^ { 2 } } \left ( C _ { \phi q, 3 3 } ^ { ( 1 ) } + 2 C _ { \phi u, 3 3 } \right ) + \frac { 3 x _ { t } } { 2 \pi s _ { w } ^ { 2 } } \left ( C _ { \phi q, 3 3 } ^ { ( 1 ) } - C _ { \phi u, 3 3 } \right ) \right ] \log \frac { \mu _ { W } } { \Lambda } \,.$$
We checked that the T parameter obtained via the above RGE analysis agrees with a direct computation of the vacuum-polarization diagrams.
The T parameter is directly related to the quantity glyph[epsilon1] 1 , defined in Ref. [27], via glyph[epsilon1] 1 ≡ αT . The term proportional to x t in Eq. (3.8) can be deduced from the corresponding result for δglyph[epsilon1] 1 in Ref. [28]; we agree with that result. However, our result disagrees with that quoted in Ref. [3], where the contributions from the modified charged current have not been included.
Note that there are no contributions to the S parameter within our setup because the operators in Eq.(2.2) do not mix into Q φWB ≡ ( φ σ φ W † a ) a µν B µν . Vertex corrections could, however, lead to an indirect contribution to S (see e.g. Ref. [29]). This is illustrated, for instance, by the correlation of δglyph[epsilon1] 3 and δglyph[epsilon1] b [30]. We neglect these effects for simplicity.
$$\mu \frac { d } { d \mu } C _ { \phi l, 1 1 } ^ { ( 3 ) } = 2 \frac { g _ { 2 } ^ { 2 } } { 1 6 \pi ^ { 2 } } C _ { \phi q, 3 3 } ^ { ( 3 ) } \,.$$
On the other hand, the operators in Eq.(2.2) mix into Q (3) φl, 11 ≡ ( φ i D † ↔ a µ φ )( ¯ L L, 1 γ µ σ L a L, 1 ) according to
This will induce a shift in the Z coupling to electrons, defined by
$$\delta g _ { L } ^ { e } & = - \frac { v ^ { 2 } } { 2 \Lambda ^ { 2 } } ( C _ { \phi l, 1 1 } ^ { ( 3 ) } + C _ { \phi l, 1 1 } ^ { ( 1 ) } ) \,.$$
Solving Eq. (3.9) we find
$$\delta g _ { L } ^ { e } = - \frac { v ^ { 2 } } { \Lambda ^ { 2 } } \frac { \alpha } { 4 \pi s _ { w } ^ { 2 } } C _ { \phi q, 3 3 } ^ { ( 3 ) } ( \Lambda ) \log \frac { \mu _ { W } } { \Lambda } \,.$$
Comparing with LEP data [31] we see that the resulting bound on C (3) φq, 33 is weaker than that from δg b L and we do not consider this constraint further. Similar considerations apply also for the corresponding µ and τ couplings. A more comprehensive analysis taking all of these effects into account might be worthwhile.
## 3.3 Rare meson decays
We now advocate the use of rare meson decays to constrain anomalous ttZ ¯ couplings. Recall that in this work we assume C (3) φq, 33 + C (1) φq, 33 = 0, and thus the absence of tree-level FCNC transitions at the scale Λ. Operator mixing reintroduces these transitions at lower scales. We calculate the running of the Wilson coefficients from Λ to the electroweak scale, where we match onto the five-flavor effective theory. We then compute the modifications of the rare meson decay rates, which allows us to bound the Wilson coefficients at the scale Λ. We focus here on the processes B ( s,d ) → µ + µ -and K → πνν ¯. The reason is that all these decays are dominated by the Z -penguin within the SM and are thus the best candidates to constrain anomalous ttZ ¯ couplings.
The part of the RGE relevant for the rare meson decays is the same as in Eq. (3.4), with the addition
$$\mu \frac { d } { d \mu } C _ { l q } ^ { ( 3 ) } = - \frac { g _ { 2 } ^ { 2 } } { 1 6 \pi ^ { 2 } } \frac { 1 } { 3 } C _ { \phi q, 3 3 } ^ { ( 3 ) } \,, \quad \mu \frac { d } { d \mu } C _ { l q } ^ { ( 1 ) } = \frac { g _ { 1 } ^ { 2 } } { 1 6 \pi ^ { 2 } } \frac { 1 } { 3 } C _ { \phi q, 3 3 } ^ { ( 1 ) } \,.$$
As a non-trivial check of our calculation we reproduce the logarithmic part of the loop functions f ( j ) νν ¯ given in Ref. [9]. Solving the equations (3.12), we find the following expressions for the Wilson coefficients at the electroweak scale:
$$C _ { l q } ^ { ( 3 ) } ( \mu _ { W } ) & = C _ { l q } ^ { ( 3 ) } ( \Lambda ) + \frac { 1 } { 3 } C _ { \phi q, 3 3 } ^ { ( 3 ) } ( \Lambda ) \frac { g _ { 2 } ^ { 2 } } { 1 6 \pi ^ { 2 } } \log \frac { \mu _ { W } } { \Lambda } \,, \\ C _ { l q } ^ { ( 1 ) } ( \mu _ { W } ) & = C _ { l q } ^ { ( 1 ) } ( \Lambda ) - \frac { 1 } { 3 } C _ { \phi q, 3 3 } ^ { ( 1 ) } ( \Lambda ) \frac { g _ { 1 } ^ { 2 } } { 1 6 \pi ^ { 2 } } \log \frac { \mu _ { W } } { \Lambda } \,. \\. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad.. \quad........................................................................................................................................................................................................$$
The other relevant Wilson coefficients are given in Eq. (3.5). Note, however, that now some of the contributions to the FCNC transitions in Eq. (3.5) cancel because of the unitarity of the CKM matrix. In particular, C (1) φq,ii and C (3) φq,ii for i = 1 2 do not appear in the final , expressions.
The rate of the rare decay B s → µ + µ -was measured by the CMS [32] and LHCb [33] collaborations at the LHC and it was found to be consistent with the SM expectations [34]. The experiments are expected to reach the sensitivity of observing the SM B d → µ + µ -rate in the next LHC run. Within the SM and the restricted set of operators we consider, a single operator Q A ≡ ( ¯ bγ µ γ q 5 )(¯ µγ µ 5 ) with q = d, s mediates the decays to a very good approximation. The rates depend on a single hadronic quantity, the decay constant f B q of the meson. The average time-integrated branching ratios then read [34]:
$$\overline { \mathcal { B } } _ { q \mu } \, = \, \frac { | N | ^ { 2 } M _ { B _ { q } } ^ { 3 } f _ { B _ { q } } ^ { 2 } } { 8 \pi \, \Gamma _ { H } ^ { q } } \, \beta _ { q \mu } \, r _ { q \mu } ^ { 2 } \, | C _ { A } ( \mu _ { b } ) | ^ { 2 } \, + \, \mathcal { O } ( \alpha _ { e m } ) \,,$$
with M B q the mass of the B meson, r qµ = 2 m /M µ B q , β qµ = √ 1 -r 2 qµ , N = V ∗ tb V tq G M 2 F 2 W /π 2 and Γ q H the total width of the heavy mass eigenstate in B B q -q mixing. In general, the timeintegrated rates depend on the details of B B q -q mixing [35], but Eq. (3.14) holds to a very good approximation for both the B s and the B d system within the SM. C A ( µ b ) is the Wilson coefficient of Q A . It incorporates effects from heavy NP to the SM contribution and is evolved to the scale of the B -meson, µ b ∼ m b . The SM part includes perturbative next-to-leading-order (NLO) QCD [36-38] as well as recently calculated next-to-next-toleading (NNLO) QCD [39] and NLO electroweak [40] corrections, which we all include in our analysis. However, at leading order, we have that C A = -2( Y 0 ( x t ) + δY NP ) where Y 0 ( x t ) is the SM loop function and δY NP is the additional contribution from NP.
Similarly, the branching ratio for K + → π + νν ¯ can be written as [41-43]
$$\text{Br} \left ( K ^ { + } & \to \pi ^ { + } \nu \bar { \nu } ( \gamma ) \right ) \\ & = \kappa _ { + } ( 1 + \Delta _ { \text{EM} } ) \left [ \left ( \frac { \text{Im} \lambda _ { t } } { \lambda ^ { 5 } } X _ { t } \right ) ^ { 2 } + \left ( \frac { \text{Re} \lambda _ { c } } { \lambda } \left ( P _ { c } + \delta P _ { c, u } \right ) + \frac { \text{Re} \lambda _ { t } } { \lambda ^ { 5 } } X _ { t } \right ) ^ { 2 } \right ], \ \ ( 3. 1 5 )$$
where λ ≡ | V us | and λ i ≡ V ∗ is V id . The leading contribution is contained in the function X t which comprises the top-quark loops. It is known to NLO in QCD and electroweak interactions [36, 38, 42, 44]. The parameter P c describes the short-distance contribution of the charm quark and has been calculated including NNLO QCD [45-47] and NLO electroweak corrections [48]. Other long-distance contributions are contained in the parameter δP c,u [43]. The hadronic matrix element of the low-energy operator is parameterized by κ + . This parameter is extracted from K glyph[lscript] 3 data and known precisely, including long-distance QED radiative corrections (∆ EM ), and NLO and partially NNLO corrections in chiral perturbation theory [49,50].
The branching ratio of the CP -violating neutral mode involves only the top-quark contribution and can be written as 3
$$\text{Br} \left ( K _ { L } \rightarrow \pi ^ { 0 } \nu \bar { \nu } \right ) = \kappa _ { L } \left ( \frac { \text{Im} \lambda _ { t } } { \lambda ^ { 5 } } X _ { t } \right ) ^ { 2 } \,.$$
3 In fact, in our numerics we take into account also the (small) contribution from indirect CP violation, proportional to glyph[epsilon1] K - see Ref. [51].
| Observable | Value | Ref. |
|----------------------------|------------------------------------|--------|
| T | 0 . 08 ± 0 . 07 | [14] |
| δg b L | 0 . 0016 ± 0 . 0015 | [14] |
| Br( B s → µ + µ - ) [CMS] | (3 . 0 +1 . 0 - 0 . 9 ) × 10 - 9 | [32] |
| Br( B s → µ + µ - ) [LHCb] | (2 . 9 +1 . 1 - 1 . 0 ) × 10 - 9 | [33] |
| Br( K + π + νν ¯) | (1 . 73 +1 . 15 - 1 . 05 ) 10 - 10 | [54] |
→
×
Table 1: Numerical input values for our fit.
Again, the hadronic matrix element can be extracted from the K l 3 decays, parametrized here by κ L [49].
As in the B q → µ + µ -decays, we include all NP effects as additional contributions to the SM top-quark function X t = X SM t + δX NP . At the order we consider, all NP effects originate from the modifications of the ttZ ¯ coupling. Thus, these are the same for B q → µ + µ -and K → πνν ¯ decays
$$\delta Y ^ { \text{NP} } = \delta X ^ { \text{NP} } = \frac { x _ { t } } { 8 } \left ( C _ { \phi u } - \frac { 1 2 + 8 x _ { t } } { x _ { t } } C _ { \phi q, 3 3 } ^ { ( 1 ) } \right ) \frac { v ^ { 2 } } { \Lambda ^ { 2 } } \log \frac { \mu _ { W } } { \Lambda } \,,$$
where x t = m /M 2 t 2 W and we used again the relation C (3) φq, 33 = -C (1) φq, 33 at the scale Λ.
We conclude this section by comparing our work to existing results in the literature. Rare meson decays have been used to constrain FCNC Zqt couplings, where q = u, c , in Ref. [6, 7], and to constrain anomalous Wtb couplings in Ref. [9]. The corrections to the Wilson coefficients presented in the above publications amount to one-loop threshold corrections at the electroweak scale. These corrections are scheme dependent [52, 53]; the scheme dependence would cancel only when performing the two-loop running of the Wilson coefficients in the effective theory above the electroweak scale. On the other hand, the logarithmic dependence of the corrections on the scale Λ is scheme independent (since the leading-order anomalous dimensions are scheme independent). By effectively choosing a questionably low matching scale, of the order of the W -boson mass, these terms are rendered numerically insignificant in the above articles.
## 4 Numerics and discussion
In this section we present the constraints on anomalous ttZ ¯ couplings derived from the observables discussed above. We show the individual constraints and perform a combined fit using current experimental data. In addition, we show the impact of future precision measurements of the rare B and K decay branching ratios.
Our main results are summarized in Fig. 2. In the left panel we show the individual 68% CL regions resulting from the measurement of the T parameter, δg b L , Br( K + → π + νν ¯),
Figure 2: The preferred regions at 68% and 95% CL from our combined fit to EWPO and rare decays are shown as the dark-gray and light-gray ellipses, respectively. The colored bands show the 68% CL constraints from the individual observables. The star denotes the SM value.
and Br( B s → µ + µ -). In addition, we show the region compatible with the measurements in Table 1 at 68% and 95% CL.
We find that the branching ratio of B s → µ + µ -and the T parameter currently lead to the most stringent constraints. In particular, the combination of the two leads to a strong bound on both Wilson coefficients C (1) φq, 33 log( µ W / Λ) v / 2 Λ 2 and C φu, 33 log( µ W / Λ) v / 2 Λ , of 2 the order of a few percent. We note that t-channel single-top production leads to the bound -0 032 . < v C 2 (1) φq, 33 / Λ 2 < 0 044 (cf. Eq. (3.3)). . This is weaker than the indirect bounds and leaves C φu, 33 completely unconstrained.
In the future, we expect improvements in the measurement of Br( B s → µ + µ -), with a final uncertainty of ∼ 5% [55]. In addition, various experiments plan to measure the branching ratios of the rare K decays with high precision. The NA62 experiment at CERN aims at a final precision of ∼ 10% for the charged mode, which could be improved to ∼ 3% by an experiment at Fermilab [56]. The KOTO experiment aims at a similar precision for the neutral mode. On the other hand, the bounds from EWPO are mainly obtained from fits to LEP data and we do not expect any significant improvement within the next few years. In the right panel of Fig. 2 we show our future projections. As an illustration we assume a branching ratio measurement of all three rare decay modes with the SM central values and a precision of 5%. We keep the current constraints from the EWPO, but note that these bounds could be improved at future e + e -colliders [57].
The indirect constraints on the anomalous ttZ ¯ couplings are much stronger than the constraints from direct searches, i.e. from tt ¯ + Z production, even after a high-luminosity upgrade of the LHC. For instance, the authors of Ref. [3] give the bounds -0 04 . <
v / 2 Λ 2 C (1) φq, 33 < 0 19 and . -0 13 . < v / 2 Λ 2 C φu, 33 < 0 32, assuming 3000 fb . -1 of data. However, one has to keep in mind that indirect constraints rely on a set of assumptions. In this work we assumed i) only C (3) φq, 33 , C (1) φq, 33 , and C φu, 33 receive non-zero initial conditions at the scale Λ; ii) C (3) φq, 33 + C (1) φq, 33 = 0 at Λ; iii) only the top-quark Yukawa coupling is non-vanishing.
Assumption i) is compatible only with NP models with non-trivial flavor structure, while assumption ii) can be motivated by explicit models [15]. A simple way to deviate from assumption iii) is to consider models with a large enhancement of the bottom-quark Yukawa coupling; a generic example is a two Higgs-doublet model with large tan β . The large bottom-Yukawa coupling will induce flavor off-diagonal versions of the operators in Eq. (2.2) and Eq. (2.5), via the MFV counting. These off-diagonal operators lead to additional contributions to FCNC top decays and D 0 -D 0 mixing. In order to relate these observables to ttZ ¯ couplings, we assume MFV. Thus the resulting constraints on anomalous ttZ ¯ couplings are suppressed by CKM-matrix elements. As an illustrative example consider an extreme case where the bottom-Yukawa coupling is much larger than the top-Yukawa coupling. In this case, we have C (3) φq, 23 ∼ λ C 2 (3) φq, 33 etc., where λ ≡ | V us | ≈ 0 22 . is the Wolfenstein parameter. Then D 0 -D 0 mixing is suppressed by λ 10 ≈ 10 -7 and thus completely negligible. Also, the branching ratio for t → cZ is
$$\text{Br} ( t \to c Z ) \simeq \frac { \lambda ^ { 4 } v ^ { 4 } } { \Lambda ^ { 4 } } \left [ \left ( C _ { \phi q, 3 3 } ^ { ( 3 ) } - C _ { \phi q, 3 3 } ^ { ( 1 ) } \right ) ^ { 2 } + C _ { \phi u, 3 3 } ^ { 2 } \right ] \,.$$
Using the present bound Br( t → cZ ) < 0 05% given by the CMS collaboration [58] we see . that the resulting bounds are not competitive with bounds from EWPO and rare B K / decays.
Note that the off-diagonal operators will also lead to additional contributions to rare B K / decays and anomalous bbZ ¯ couplings. The generalization of our assumption ii) can be used to eliminate such contribution from these off-diagonal operators [15].
More generally, all rare decays in the down sector which receive a Z -penguin contribution can be used to obtain bounds on anomalous ttZ ¯ couplings with our method. Suitable decays which will be measured in the future include B d → µ + µ -[59] and B → Kνν ¯ [60]. It would be interesting to allow for complex Wilson coefficients of the operators in Eq. (2.2) and study their effect on CP violation in rare meson decays.
To conclude, in this work we studied the effects of dimension-six operators, generating anomalous vector and axial-vector ttZ ¯ couplings at tree-level, on precision observables. In particular, we advocate the use of rare K and B meson decays to obtain strong constraints on ttZ ¯ couplings.
## Acknowledgements
We thank Jernej Kamenik and Jure Zupan for useful comments on the manuscript, Yosef Nir and Gilad Perez for insightful discussions, and Marco Ciuchini, Alex Pomarol and
Luca Silvestrini for helpful correspondence concerning the S parameter. This work was supported in part by National Science Foundation Grant No. PHYS-1066293 and the hospitality of the Aspen Center for Physics. J.B. thanks the Perimeter Institute for Theoretical Physics for their hospitality. Research at Perimeter Institute is supported by the Government of Canada through Industry Canada and by the Province of Ontario through the Ministry of Economic Development & Innovation. The work of J.B. is supported in part by the U.S. National Science Foundation under CAREER Grant PHY-1151392. The work of P.U. is supported in part by the DOE grant DE- SC0011784. A.G. acknowledges the support by the Slovenian Research Agency (ARRS).
## References
- [1] 'Search for ttZ ¯ production in the three lepton final state with 4 7 fb . -1 of √ s = 7 TeV pp collision data collected by the ATLAS detector'. Technical Report ATLASCONF-2012-126, CERN, Geneva, 2012.
- [2] S. Chatrchyan et al. 'Measurement of associated production of vector bosons and top quark-antiquark pairs at √ s = 7 TeV'. Phys.Rev.Lett. 110 :172002, 2013. doi:10.1103/PhysRevLett.110.172002. e-print:1303.3239.
- [3] R. R¨ ontsch and M. Schulze. 'Constraining couplings of the top quarks to the Z boson in tt ¯ + Z production at the LHC' 2014. e-print:1404.1005.
- [4] U. Haisch and A. Weiler. 'Determining the Sign of the Z -Penguin Amplitude'. Phys.Rev. D76 :074027, 2007. doi:10.1103/PhysRevD.76.074027. e-print:0706.2054.
- [5] D. Guadagnoli and G. Isidori. 'BR( B s → µ + µ -) as an electroweak precision test'. Phys.Lett. B724 :63, 2013. doi:10.1016/j.physletb.2013.05.054. e-print:1302.3909.
- [6] X.-Q. Li, Y.-D. Yang and X.-B. Yuan. 'Anomalous tqZ Coupling Effects in Rare B- and K-Meson Decays'. JHEP 1203 :018, 2012. doi:10.1007/JHEP03(2012)018. e-print:1112.2674.
- [7] H. Gong, Y.-D. Yang and X.-B. Yuan. 'Constraints on anomalous tcZ coupling from ¯ B → ¯ K µ µ ∗ + -and B s → µ + µ -decays'. JHEP 1305 :062, 2013. doi:10.1007/JHEP05(2013)062. e-print:1301.7535.
- [8] J. Drobnak, S. Fajfer and J. F. Kamenik. 'Interplay of t → bW Decay and B q Meson Mixing in Minimal Flavor Violating Models'. Phys.Lett. B701 :234, 2011. doi:10.1016/j.physletb.2011.05.052. e-print:1102.4347.
- [9] J. Drobnak, S. Fajfer and J. F. Kamenik. 'Probing anomalous tWb interactions with rare B decays'. Nucl.Phys. B855 :82, 2012. doi:10.1016/j.nuclphysb.2011.10.004. e-print:1109.2357.
- [10] J. Aguilar-Saavedra. 'A Minimal set of top anomalous couplings'. Nucl.Phys. B812 :181, 2009. doi:10.1016/j.nuclphysb.2008.12.012. e-print:0811.3842.
- [11] B. Grzadkowski, M. Iskrzynski, M. Misiak and J. Rosiek. 'Dimension-Six Terms in the Standard Model Lagrangian'. JHEP 1010 :085, 2010. doi:10.1007/JHEP10(2010)085. e-print:1008.4884.
- [12] G. D'Ambrosio, G. F. Giudice, G. Isidori and A. Strumia. 'Minimal flavour violation: An effective field theory approach'. Nucl. Phys. B645 :155, 2002. doi:10.1016/S05503213(02)00836-2. e-print:hep-ph/0207036.
| [13] | A. Denner. 'Techniques for calculation of electroweak radiative corrections at the one loop level and results for W physics at LEP-200'. Fortschr. Phys. 41 :307, 1993. e-print:0709.1075. |
|--------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [14] | M. Ciuchini, E. Franco, S. Mishima and L. Silvestrini. 'Electroweak Precision Ob- servables, New Physics and the Nature of a 126 GeV Higgs Boson'. JHEP 1308 :106, 2013. doi:10.1007/JHEP08(2013)106. e-print:1306.4644. |
| [15] | F. del Aguila, M. Perez-Victoria and J. Santiago. 'Observable contributions of new exotic quarks to quark mixing'. JHEP 0009 :011, 2000. doi:10.1088/1126- 6708/2000/09/011. e-print:hep-ph/0007316. |
| [16] | 'Measurement of the Inclusive and Fiducial Cross-Section of Single Top-Quark t - Channel Events in pp Collisions at √ s = 8 TeV'. Technical Report ATLAS-CONF- 2014-007, CERN, Geneva, 2014. |
| [17] | V. Khachatryan et al. 'Measurement of the t-channel single-top-quark production cross section and of the | V tb | CKM matrix element in pp collisions at √ s = 8 TeV' 2014. e-print:1403.7366. |
| [18] | E. E. Jenkins, A. V. Manohar and M. Trott. 'Renormalization Group Evolution of the Standard Model Dimension Six Operators II: Yukawa Dependence'. JHEP 1401 :035, 2014. doi:10.1007/JHEP01(2014)035. e-print:1310.4838. |
| [19] | R. Alonso, E. E. Jenkins, A. V. Manohar and M. Trott. 'Renormalization Group Evo- lution of the Standard Model Dimension Six Operators III: Gauge Coupling Depen- dence and Phenomenology'. JHEP 1404 :159, 2014. doi:10.1007/JHEP04(2014)159. e-print:1312.2014. |
| [20] | G. Altarelli, R. Barbieri and S. Jadach. 'Toward a model independent analysis of electroweak data'. Nucl.Phys. B369 :3, 1992. doi:10.1016/0550-3213(92)90376-M. |
| [21] | F. Larios, M. Perez and C. Yuan. 'Analysis of tbW and ttZ couplings from CLEO and LEP / SLC data'. Phys.Lett. B457 :334, 1999. doi:10.1016/S0370-2693(99)00541-9. e-print:hep-ph/9903394. |
| [22] | F. Larios, E. Malkawi and C. Yuan. 'Probing the electroweak symmetry breaking sector with the top quark' 1997. e-print:hep-ph/9704288. |
| [23] | M. E. Peskin and T. Takeuchi. 'A New constraint on a strongly interacting Higgs sector'. Phys.Rev.Lett. 65 :964, 1990. doi:10.1103/PhysRevLett.65.964. |
| [24] | M. E. Peskin and T. Takeuchi. 'Estimation of oblique electroweak corrections'. Phys.Rev. D46 :381, 1992. doi:10.1103/PhysRevD.46.381. |
| [25] | B. Grinstein and M. B. Wise. 'Operator analysis for precision electroweak physics'. Phys.Lett. B265 :326, 1991. doi:10.1016/0370-2693(91)90061-T. |
| [26] | W. Skiba. 'TASI Lectures on Effective Field Theory and Precision Electroweak Mea- surements' 2010. e-print:1006.2142. |
|--------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [27] | G. Altarelli, R. Barbieri and F. Caravaglios. 'Nonstandard analysis of electroweak precision data'. Nucl.Phys. B405 :3, 1993. doi:10.1016/0550-3213(93)90424-N. |
| [28] | E. Malkawi and C. Yuan. 'A Global analysis of the top quark couplings to gauge bosons'. Phys.Rev. D50 :4462, 1994. doi:10.1103/PhysRevD.50.4462. e-print:hep- ph/9405322. |
| [29] | R. S. Gupta, A. Pomarol and F. Riva. 'BSM Primary Effects' 2014. e-print:1405.0181. |
| [30] | M. Ciuchini, E. Franco, S. Mishima, M. Pierini, L. Reina et al. 'Update of the electroweak precision fit, interplay with Higgs-boson signal strengths and model- independent constraints on new physics' 2014. e-print:1410.6940. |
| [31] | S. Schael et al. 'Precision electroweak measurements on the Z resonance'. Phys.Rept. 427 :257, 2006. doi:10.1016/j.physrep.2005.12.006. e-print:hep-ex/0509008. |
| [32] | S. Chatrchyan et al. 'Measurement of the B s → µ + µ - branching fraction and search for B 0 → µ + µ - with the CMS Experiment'. Phys.Rev.Lett. 111 :101804, 2013. doi:10.1103/PhysRevLett.111.101804. e-print:1307.5025. |
| [33] | R. Aaij et al. 'Measurement of the B 0 s → µ + µ - branching fraction and search for B 0 → µ + µ - decays at the LHCb experiment'. Phys.Rev.Lett. 111 :101805, 2013. doi:10.1103/PhysRevLett.111.101805. e-print:1307.5024. |
| [34] | C. Bobeth, M. Gorbahn, T. Hermann, M. Misiak, E. Stamou et al. ' B s,d → glyph[lscript] + glyph[lscript] - in the Standard Model with Reduced Theoretical Uncertainty'. Phys.Rev.Lett. 112 :101801, 2014. doi:10.1103/PhysRevLett.112.101801. e-print:1311.0903. |
| [35] | K. De Bruyn, R. Fleischer, R. Knegjens, P. Koppenburg, M. Merk et al. 'Probing New Physics via the B 0 s → µ + µ - Effective Lifetime'. Phys.Rev.Lett. 109 :041801, 2012. doi:10.1103/PhysRevLett.109.041801. e-print:1204.1737. |
| [36] | G. Buchalla and A. J. Buras. 'QCD corrections to the ¯ s dZ vertex for arbitrary top quark mass'. Nucl. Phys. B398 :285, 1993. doi:10.1016/0550-3213(93)90110-B. |
| [37] | G. Buchalla and A. J. Buras. 'QCD corrections to rare K and B decays for arbitrary top quark mass'. Nucl.Phys. B400 :225, 1993. doi:10.1016/0550-3213(93)90405-E. |
| [38] | M. Misiak and J. Urban. 'QCD corrections to FCNC decays mediated by Z-penguins and W-boxes'. Phys. Lett. B451 :161, 1999. doi:10.1016/S0370-2693(99)00150-1. e- print:hep-ph/9901278. |
| [39] | T. Hermann, M. Misiak and M. Steinhauser. 'Three-loop QCD corrections to B s → µ + µ - '. JHEP 1312 :097, 2013. doi:10.1007/JHEP12(2013)097. e-print:1311.1347. |
- [40] C. Bobeth, M. Gorbahn and E. Stamou. 'Electroweak Corrections to B s,d → glyph[lscript] + glyph[lscript] -'. Phys.Rev. D89 :034023, 2014. doi:10.1103/PhysRevD.89.034023. e-print:1311.1348.
- [41] G. Buchalla and A. J. Buras. 'The rare decays K + → π + νν ¯ and K L → µ + µ -beyond leading logarithms'. Nucl. Phys. B412 :106, 1994. doi:10.1016/0550-3213(94)90496-0. e-print:hep-ph/9308272.
- [42] G. Buchalla and A. J. Buras. 'The rare decays K → πνν ¯, B → Xνν ¯ and B → l + l -: An update'. Nucl. Phys. B548 :309, 1999. doi:10.1016/S0550-3213(99)00149-2. eprint:hep-ph/9901288.
- [43] G. Isidori, F. Mescia and C. Smith. 'Light-quark loops in K → πνν ¯'. Nucl. Phys. B718 :319, 2005. doi:10.1016/j.nuclphysb.2005.04.008. e-print:hep-ph/0503107.
- [44] J. Brod, M. Gorbahn and E. Stamou. 'Two-Loop Electroweak Corrections for the K → πνν ¯ Decays'. Phys.Rev. D83 :034030, 2011. doi:10.1103/PhysRevD.83.034030. e-print:1009.0947.
- [45] M. Gorbahn and U. Haisch. 'Effective Hamiltonian for non-leptonic | ∆ F | = 1 decays at NNLO in QCD'. Nucl. Phys. B713 :291, 2005. doi:10.1016/j.nuclphysb.2005.01.047. e-print:hep-ph/0411071.
- [46] A. J. Buras, M. Gorbahn, U. Haisch and U. Nierste. 'The rare decay K + → π + νν ¯ at the next-to-next- to-leading order in QCD'. Phys. Rev. Lett. 95 :261805, 2005. doi:10.1103/PhysRevLett.95.261805. e-print:hep-ph/0508165.
- [47] A. J. Buras, M. Gorbahn, U. Haisch and U. Nierste. 'Charm quark contribution to K + → π + νν ¯ at next-to-next-to-leading order'. JHEP 11 :002, 2006. e-print:hepph/0603079.
- [48] J. Brod and M. Gorbahn. 'Electroweak Corrections to the Charm Quark Contribution to K + → π + νν ¯'. Phys. Rev. D78 :034006, 2008. doi:10.1103/PhysRevD.78.034006. e-print:0805.4119.
- [49] F. Mescia and C. Smith. 'Improved estimates of rare K decay matrix-elements from K glyph[lscript] 3 decays'. Phys. Rev. D76 :034017, 2007. doi:10.1103/PhysRevD.76.034017. eprint:0705.2025.
- [50] J. Bijnens and K. Ghorbani. 'Isospin breaking in Kπ vector form-factors for the weak and rare decays K glyph[lscript] 3 , K → πνν ¯ and K → πglyph[lscript] + glyph[lscript] -' 2007. e-print:0711.0148.
- [51] G. Buchalla and A. J. Buras. ' K → πνν ¯ and high precision determinations of the CKM matrix'. Phys. Rev. D54 :6782, 1996. doi:10.1103/PhysRevD.54.6782. eprint:hep-ph/9607447.
- [52] A. J. Buras, M. Jamin, M. Lautenbacher and P. H. Weisz. 'Effective Hamiltonians for Delta S = 1 and Delta B = 1 nonleptonic decays beyond the leading logarithmic approximation'. Nucl.Phys. B370 :69, 1992. doi:10.1016/0550-3213(92)90345-C.
- [53] M. Ciuchini, E. Franco, G. Martinelli and L. Reina. 'The Delta S = 1 effective Hamiltonian including next-to-leading order QCD and QED corrections'. Nucl.Phys. B415 :403, 1994. doi:10.1016/0550-3213(94)90118-X. e-print:hep-ph/9304257.
- [54] A. V. Artamonov et al. 'New measurement of the K + → π + νν ¯ branching ratio'. Phys. Rev. Lett. 101 :191802, 2008. doi:10.1103/PhysRevLett.101.191802. e-print:0808.2459.
- [55] R. Aaij et al. 'Implications of LHCb measurements and future prospects'. Eur.Phys.J. C73 :2373, 2013. doi:10.1140/epjc/s10052-013-2373-2. e-print:1208.3355.
- [56] A. S. Kronfeld, R. S. Tschirhart, U. Al-Binni, W. Altmannshofer, C. Ankenbrandt et al. 'Project X: Physics Opportunities' 2013. e-print:1306.5009.
- [57] M. Bicer et al. 'First Look at the Physics Case of TLEP'. JHEP 1401 :164, 2014. doi:10.1007/JHEP01(2014)164. e-print:1308.6176.
- [58] S. Chatrchyan et al. 'Search for flavor-changing neutral currents in top-quark decays t to Zq in pp collisions at √ s = 8 TeV'. Phys.Rev.Lett. 112 :171802, 2014. doi:10.1103/PhysRevLett.112.171802. e-print:1312.4194.
- [59] J. Hewett, H. Weerts, R. Brock, J. Butler, B. Casey et al. 'Fundamental Physics at the Intensity Frontier' 2012. e-print:1205.2671.
- [60] T. Aushev, W. Bartel, A. Bondar, J. Brodzicka, T. Browder et al. 'Physics at Super B Factory' 2010. e-print:1002.5012.
|
10.1007/JHEP02(2015)141
|
[
"Joachim Brod",
"Admir Greljo",
"Emmanuel Stamou",
"Patipan Uttayarat"
] |
2014-08-04T20:00:00+00:00
|
2015-03-01T06:50:54+00:00
|
[
"hep-ph"
] |
Probing anomalous $t\bar t Z$ interactions with rare meson decays
|
Anomalous couplings of the $Z$ boson to top quarks are only marginally
constrained by direct searches and are still sensitive to new particle dynamics
at the TeV scale. Employing an effective field theory approach we consider the
dimension-six operators which generate deviations from the standard-model
vector and axial-vector interactions. We show that rare $B$ and $K$ meson
decays together with electroweak precision observables provide strong
constraints on these couplings. We also consider constraints from t-channel
single-top production.
|
1408.0793v1
|
## THE MYSTERY OF SPECTRAL BREAKS: LYMAN CONTINUUM ABSORPTION BY PHOTON-PHOTON PAIR PRODUCTION IN THE FERMI GEV SPECTRA OF BRIGHT BLAZARS
Boris E. Stern 1,2 and Juri Poutanen 3,4
1 Institute for Nuclear Research, Russian Academy of Sciences, Prospekt 60-letiya Oktyabrya 7a, Moscow 117312, Russia; [email protected]
ApJ, in press
2 Astro Space Center, Lebedev Physical Institute, Russian Academy of Sciences, Profsoyuznaya 84/32, Moscow 117997, Russia 3 Tuorla Observatory, University of Turku, V¨is¨l¨ntie 20, FI-21500 Piikki¨, Finland; [email protected] a a a o 4 Astronomy Division, Department of Physics, PO Box 3000, FI-90014 University of Oulu, Finland (Dated: Received 2013 December 19; accepted 2014 August 3)
## ABSTRACT
We reanalyze Fermi /LAT gamma-ray spectra of bright blazars with a higher photon statistics than in previous works and with new Pass 7 data representation. In the spectra of the brightest blazar 3C 454.3 and possibly of 4C +21.35 we detect breaks at ∼ 5 GeV (in the rest frame) associated with the photon-photon pair production absorption by He ii Lyman continuum (LyC). We also detect confident breaks at ∼ 20 GeV associated with hydrogen LyC both in the individual spectra and in the stacked redshift-corrected spectrum of several bright blazars. The detected breaks in the stacked spectra univocally prove that they are associated with atomic ultraviolet emission features of the quasar broad-line region (BLR). The dominance of the absorption by hydrogen Ly complex over He ii , rather small detected optical depth, and the break energy consistent with the head-on collisions with LyC photons imply that the gamma-ray emission site is located within the BLR, but most of the BLR emission comes from a flat disk-like structure producing little opacity. Alternatively, the LyC emission region size might be larger than the BLR size measured from reverberation mapping, and/or the γ -ray emitting region is extended. These solutions would resolve a long-standing issue how the multi-hundred GeV photons can escape from the emission zone without being absorbed by softer photons.
Subject headings: black hole physics - BL Lacertae objects: general - galaxies: active - galaxies: jets - gamma rays: general - quasars: emission lines
## 1. INTRODUCTION
The spectra of bright blazars obtained by Fermi Gamma-ray Space Telescope ( Fermi ) Large Area Telescope (LAT) showed clear deviations from a powerlaw shape (Abdo et al. 2009, 2010). There spectra could not be described by smooth functions such as exponentially cutoff power-law or a log-parabola (lognormal distribution), but were found to be better described by a broken power-law. The derived break energies lying in the 1-10 GeV energy range (Abdo et al. 2010; Poutanen & Stern 2010; Harris et al. 2012) were rather stable (Ackermann et al. 2010; Abdo et al. 2011; Stern & Poutanen 2011). Those breaks seemed puzzling: the hypothesis that the break is caused by photon-photon annihilation through e ± pair production was considered and rejected by Abdo et al. (2010), who argued that 'to produce a break in the 1-10 GeV, the photon field should have an energy peaking in the 0.05-0.5 keV range, which excludes the broad-line region peaking in the UV.' The conclusion that such large energies of the target photons were required was based on an erroneous assumption that the break energy correspond to the maximum cross-section for pair production.
Poutanen & Stern (2010) suggested that the breaks should actually appear at the energies close to the threshold for a corresponding reaction, where the opacity has a sharp rise. The observed breaks at a few GeV correspond then well (correcting for the redshift) to the Lyman recombination continuum (LyC) and Ly α emission of ionized He. They also showed that the inner part of the BLR can provide sufficient flux of He ii Lyman lines and LyC to provide enough opacity for GeV photons and to produce a spectral break. Poutanen & Stern (2010) further demonstrated that the data for a number of bright blazars are well described by a power-law spectrum modified by the absorption within the BLR. The fits with this model were acceptable and the reduction in χ 2 (compared to a simple power-law model) was very significant. Similar χ 2 could also be achieved with the broken power-law model which, however, does not have any physical basis.
A high significance of the spectral breaks partially results from a null hypothesis for the underlying spectrum, which is assumed to be a power-law. However, a typical blazar has a curved spectrum extending over many orders in energy and peaking in the MeV-GeV range. This implies that the spectrum in the Fermi energy band should be slightly concave and the power-law null hypothesis gives an overestimated significance of the break. As a more realistic null hypothesis one can take a lognormal distribution (log-parabola), which is the simplest way to introduce a curvature in logarithmic coordinates.
Stern & Poutanen (2011) studied in details the spectrum of the exceptionally bright flat-spectrum radio quasar (FSRQ) 3C 454.3 and indeed found that the spectrum below the absorption break significantly differs from the power-law and can be well described by a log-parabola. With this null hypothesis the statistical significance of any break is lower than with a power-law hypothesis. However, the absorption break in the time-
Figure 1. Acceptance (i.e. effective area integrated over the field of view) of Fermi /LAT. The solid red curve gives the acceptance for Pass 7 version P7\_CLEAN\_V6 . The blue curves are for Pass 6 version P6\_V3\_DIFFUSE : the dashed upper curve is the total acceptance for two detectors, the dotted curve is for the front detector and the dot-dashed curve is for the back detector. The acceptance of the front detector has a hump starting at ∼ 3 GeV and peaking at 30 GeV, which enhances the He ii absorption feature or even mimics
it. integrated spectrum of 3C 454.3 was still highly significant and its energy coincided with the predicted one from He ii LyC absorption.
While the main attention of the cited papers has been payed to the GeV breaks, Poutanen & Stern (2010) and Stern & Poutanen (2011) also revealed hydrogen LyC breaks at ∼ 20 GeV, which were less significant and less impressive because of a lower photon statistics in that energy range. Actually, it is obvious that in most cases H Ly radiation should produce a stronger absorption feature that He ii Ly emission. That is why it is important to revisit the spectral analysis of bright blazars with a higher statistics that was accumulated since previous works. Another, more important reason to revisit previous results is a slight but significant difference in the blazar spectra obtained with the new Pass 7 version of the Fermi /LAT data and detector response and those obtained with the older Pass 6 version. The new spectra look smoother and this is a reason to suspect that the sharp breaks at a few GeV were the artifacts of the Pass 6 response function - partially or completely.
## 2. DATA AND THEIR ANALYSIS
We use Fermi /LAT photon data for 1740 days (from 2008 August 6 till 2013 May 12) using the new Pass 7 event classification, selecting photons of clean class, and imposing the cut on the zenith angle at θ < 105 . ◦ We used P7\_CLEAN\_V6 response function. The diffuse background was calculated using the background model elaborated by the Fermi team.
It should be noted that the Pass 7 response function significantly differs from the Pass 6 one in at least two aspects:
- 1. it has a wider point spread function,
- 2. it has a different effective area function (see Fig. 1).
The second fact is of crucial importance for the analysis of the GeV energy breaks because the Pass 6 response has
Figure 2. Central part of spectra of 3C454.3 using response functions for Pass 6 version P6\_V3\_DIFFUSE (black crosses) and Pass 7 version P7\_CLEAN\_V6 (red circles and crosses).
a hump starting at 4-5 GeV, which introduces a break in the photon spectrum at energies close to the He ii LyC absorption (see Fig. 2 for comparison of the spectra of 3C 454.3 obtained with Passes 6 and 7). The break is not very strong: it changes the index of the power-law by ∼ 0.1. If one then fits the resulting spectrum with a broken power law, one obtains a significant break. We believe that the break in the Pass 6 effective area is not real, because there is no clear reason for existence of such a feature at this energy. Moreover, it looks strange that the hump appears only in the effective area of the front detector. The Pass 7 version of the effective area does not have such a feature and looks more natural.
The energy-dependent exposure function was calculated using the spacecraft pointing history:
$$\begin{array} { c } \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \\ \text{Exposure} ( E, \Delta T ) = \int _ { \Delta T } S ( E, \theta ( t ) ) \, d t, \quad ( 1 ) \\ \text{where} \, \Lambda T \, \text{is the time interval} \, \text{of interest} \, \text{$A$ is the an}$} \end{array}.$$
where ∆ T is the time interval of interest, θ is the angle between the detector axis and the direction to the object and S E,θ ( ) is the detector effective area at energy E . We accumulated counts in the circle centered at the source location with the energy-dependent radius r E ( ) = min { r 90 , 4 ◦ } , where r 90 is the radius of 90% event containment, which was calculated with MonteCarlo integration of the point-spread function assuming the isotropic distribution of the exposure angle. Then number of counts in each energy bin was corrected to the containment factor for r E ( ).
In order to reveal the absorption features in the blazar emission, we analyzed the spectra of individual bright FSRQs as well as the stacked spectra of various samples. We have selected a sample of 15 brightest objects from the 2nd Fermi catalog (Nolan et al. 2012) using the following criteria:
- 1. the total number of counts above 1 GeV after background subtraction is above 1800,
- 2. the signal/background ratio exceeds 2,
- 3. known redshift (according to the 2nd Fermi catalog),
Table 1 The brightest GeV blazars
| Object | Group a | Redshift | Dates b |
|-------------------------|-----------|------------|-----------|
| 3C 454.3 | | 0.859 | 0-1740 |
| 4C +55.17 | 1, 2 | 0.896 | 0-1740 |
| PKS 0537 - 441 | 1, 2 | 0.892 | 0-1740 |
| PKS 2326 - 502 | 1 | 0.518 | 600-1740 |
| 4C +21.35 (PKS 1222+21) | 1, 2 | 0.433 | 350-1100 |
| PKS B1424 - 418 | 1, 2 | 1.522 | 1200-1740 |
| PKS 0426 - 380 | 1, 2 | 1.111 | 0-1740 |
| PKS 0454 - 234 | 1, 2 | 1.003 | 0-1740 |
| PKS 0727 - 11 | 1 | 1.591 | 0- 800 |
| PKS 1510 - 08 | 1, 2 | 0.360 | 0-1740 |
| 3C 279 | 1 | 0.536 | 0-1300 |
| PKS 1502+106 | 1, 2 | 1.893 | 0- 500 |
| B2 1520+31 | 1 | 1.484 | 0-1400 |
| PKS 0235+164 | 1 | 0.940 | 0- 400 |
| 4C +38.41 | 1 | 1.813 | 0-1740 |
| BL Lacs | BL Lacs | BL Lacs | BL Lacs |
| Mrk 421 | 3 | 0.030 | 0-1740 |
| 3C 66A | 3 | 0.444 | 0-1740 |
| S5 0716+714 | 3 | 0.310 | 0-1740 |
| PKS 2155 - 304 | 3 | 0.117 | 0-1740 |
a Group memberships.
b The start and the end of the observation measured from MJD 54684 (2008 August 6).
- 4. classification as a FSRQ, or as a low-synchrotron peak BL Lac if its redshift exceeds 0.5 (which would mean that the latter is probably a misclassified FSRQ),
- 5. there is no strong source confusion.
The brightest blazar 3C 454.3 was excluded from the stacking analysis and studied only individually because of its exceptional brightness, which is comparable to the total signal from other selected objects. The remaining 14 FSRQs bright above 1 GeV constitute Group 1, where we hoped to reveal the absorption break at ∼ 5 GeV associated with He ii LyC emission (Poutanen & Stern 2010).
We also selected Group 2 of eight blazars with the highest number of counts above 5 GeV (more than 200 after background subtraction). This sample was more promising to study H LyC absorption break at ∼ 20 GeV. All such blazars also belong to Group 1 despite the selection was independent.
To optimize the signal-to-noise ratio, we have selected for each blazar a time interval, when its flux substantially exceeded the background. Limits of these time intervals are given in Table 1. In order to prepare the stacked spectra, we first derived individual spectra using the energy bins with the width of 0.1 in decimal logarithm. The bin edges were adjusted in such a way that they are the same in the object rest frame. Each spectrum was blue-shifted by a factor (1 + z ). We summed up the obtained spectra with their absolute normalization, so that the spectra of the brighter objects have a larger contribution. Such approach optimizes relative statistical errors. The blazar spectra were modeled with a lognormal function with the superimposed absorption by the BLR emission (see Sect. 3). We use the opacity computed for different ionization parameters ξ of the BLR as described in Poutanen & Stern (2010). We also checked a simpler monochromatic absorber model (H and He ii LyC), which is less realistic but helps to estimate separate contributions of H and He emission.
We have also constructed two comparison spectra, where breaks are not expected, to make sure that the detection of the breaks is not an artifact of the detector response. The first one is the stacked redshift-corrected spectrum of four bright BL Lacs (Group 3 in Table 1). The second one is the 'empty' sky spectrum. The sky was sliced into 72 000 bins of equal area (1 74 . × 10 -4 str) and the photons were collected from the bins where number of photons above 100 MeV is less that 300, these bins constitute about 0.226 fraction of the sky. The main contribution to this spectrum is from the high-energy protons producing pions and a smaller contribution of unresolved BL Lacs.
Statistical errors were treated as Gaussian, except in a few bins at higher energies, where the number of photons is low, we use Poisson likelihood adding -2 ln P n,µ ( ) to χ 2 (here n is the number of counts in the bin and µ is the prediction of the model). The number of such bins is small and the meaning of χ 2 is not significantly affected. For the minimization we use the standard code minuit from the CERN library.
## 3. GAMMA-RAY OPACITY
Gamma-ray photons emitted presumably by the relativistic jet emanating from the black hole are strongly beamed. They propagate through the radiation field made by the accretion disk, the BLR and the dusty torus are potentially can be absorbed by photon-photon pair production. The disk radiation moves nearly parallel to the photon beam and therefore does not interact efficiently. The infrared photons from the dust absorb radiation mostly in the TeV range. Thus, the most important source of opacity is the optical/UV (nearly) isotropic radiation from the BLR.
The opacity depends on the BLR spectrum, which is computed using spectral synthesis code xstar (version 2.2, Kallman & Bautista 2001) as described by Poutanen & Stern (2010). We repeat here the basic assumption for completeness. The ionizing spectrum of a quasar is taken as a sum of the standard multi-color accretion disk plus a power law of total luminosity 10% extending to 100 keV (Laor et al. 1997). The BLR clouds are assumed to be simple slabs of constant gas density and a clear view to the ionizing source. The hydrogen column density was fixed at N H = 10 23 cm -2 and the BLR spectra were computed for different ionization parameters ξ = L/ r n ( 2 H ) varying from 10 0 5 . to 10 2 5 . . If one assumes a dependence of the cloud density on distance from the black hole n H = 10 10 r -1 18 , then ξ = 10 L 47 r -1 18 and our ionization range would correspond to the distance interval between 1 and 0.01 pc to the ionizing source for a quasar luminosity L = 10 47 erg s -1 1 . The scaling and the estimated distances are very approximate and are model-dependent.
If the BLR were to emit only one line at energy E 0 , the optical depth for a γ -ray photon of energy E through the region of size R filled with isotropic soft photon field of column density N ph would be τ γγ ( E,E 0 ) = σ γγ ( s ) τ T /σ T = N ph σ γγ ( s ), where
$$\tau _ { \text{T} } = N _ { \text{ph} } \sigma _ { \text{T} } = \frac { L _ { \text{line} } \sigma _ { \text{T} } } { 4 \pi R c E _ { 0 } } = 1 1 0 \frac { L _ { \text{line}, 4 5 } } { R _ { 1 8 } } \frac { 1 0 \ e V } { E _ { 0 } }, \quad ( 2 ) \\ ^ { 1 } \text{We defined } Q = 1 0 ^ { x } Q _ { x } \text{ in cgs units.}$$
1 We defined Q = 10 Q x in cgs units.
Figure 3. Examples of the photon spectrum transmitted through the BLR of various ionizations and optical depths. The incident spectrum (dashed black line) is taken as a power-law of photon index Γ = 2. The total photon column density corresponds to τ T = 10 in all cases. Transmission function exp( -τ γγ ( E )) for different log ξ is shown by different lines: 0.5 (red long-dashed), 1.0 (green dot-dashed), 1.5 (blue short-dashed), 2.0 (pink dotted), 2.5 (black solid).
σ T is the Thomson cross-section, L line is the line luminosity, s = EE / m c 0 ( e 2 ) 2 and σ γγ is the angleaveraged cross-section for photon-photon pair production (see, e.g. Gould & Schr´ eder 1967; Zdziarski 1988). Note, that σ γγ has threshold s = 1 (i.e. at E thr = 19 2 GeV ( . / E / 0 13 6 eV)) and has a peak of about . σ T / 5 at E ≈ 3 5 . E thr .
The BLR spectrum, of course, contains many lines and recombination continua. One can introduce the crosssection weighted with the photon distribution:
$$\cos \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \varepsilon.$$
where N ph = ∫ N ph ( E 0 ) dE 0 . The spectrum transmitted through the BLR is attenuated as ∝ exp( -τ γγ ( E )), where τ γγ ( E ) = τ T σ γγ ( E /σ ) T , and τ T is computed using Equation (2) replacing L line by L BLR and E 0 by the mean photon energy of the BLR:
$$\Gamma \text{www.com/} \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \infty \begin{array} { c } ( 4 ) & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & && & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &
& & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & \\ & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & - & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & ; & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & } & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & \ & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & + & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &. & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & _ & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & 0 & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & = & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &, & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & b & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & ^ & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &$$
As an illustration we present the results of absorption of a power-law spectrum by the BLR of different ionizations in Fig. 3 fixing the total BLR photon column density at N ph = 1 5 . × 10 25 cm -2 , which corresponds to τ T = 10. For the considered τ T , the flux drops at most by a factor of 3-4.5 depending on ξ , corresponding to the maximum optical depth of about 1.1-1.5. Note that the transmitted spectrum in the range from 30 GeV to 1 TeV has nearly the same slope as the intrinsic one at larger ξ , because the opacity is nearly constant in this range. The opacity drops at energies above 1 TeV and the spectrum recovers. We see that the He ii LyC breaks at 5 GeV are more pronounced at high ionizations log ξ > 1 5, . while the H LyC breaks are seen at any log ξ . This allowed Poutanen & Stern (2010) to introduce a simpler doubleabsorber model for γ -ray opacity, where the BLR spectrum is replaced by the strongest emission features of H and He ii LyC. For low ionization, one can even consider only a single absorber due to the H LyC.
Figure 4. The redshift-corrected Fermi /LAT spectra of individual bright blazars and their best-fit model of the log-normal distribution with absorption by the BLR (with log ξ = 1 5). . The dashed lines show the same log-normal distributions without absorption.
Figure 5. Same as Fig. 4, but for the stacked rest-frame spectra for the two samples of blazars from Table 1 for 1740 days of Fermi observations.
## 4. RESULTS
## 4.1. Detection of GeV breaks
The results of the spectral fits for 3C 454.3, all objects of Group 2 and for the stacked spectra are presented in Table 2 and some of them are shown in Figs. 4 and 5. The best-fit model for all objects except 3C 454.3 and 4C +21.35 is that of the BLR emission with lower ionization degree log ξ = 1.5. In this ionization state, the contribution by He ii absorption is small and one can see from Table 2 that the double absorber model H+He ii LyC does not improve the fits with respect to the single H LyC absorber. This means that in most spectra there is no sign of He ii LyC absorption. The exceptions are 3C 454.3 and 4C +21.35 (PKS 1222+21) where the presence of He ii absorption is detected at ∼ 3 σ level. The best-fit model for absorber in these sources is BLR emission with log ξ = 2 5. . In the stacked spectra of both groups, the situation is similar: the addition of
Table 2 Spectral Properties of Blazars
| Object | lognorm 2 e | lognorm+H LyC a | lognorm+H LyC a | lognorm+H&He ii LyC b | lognorm+H&He ii LyC b | lognorm+log ξ =1.5 c | lognorm+log ξ =1.5 c | lognorm+log ξ =2.5 d | lognorm+log ξ =2.5 d | |
|-------------------|---------------|-------------------|--------------------|-------------------------|-------------------------|------------------------|------------------------|------------------------|------------------------|----------------|
| | χ /dof | χ 2 /dof e | τ H f | χ 2 /dof e | τ He ii g | χ 2 /dof e | τ T | χ 2 /dof e | τ T | Significance h |
| 3C 454.3 | 55.0/21 | 38.3/20 | 4.4 ± 1.0 | 28.1/19 | 0.94 ± 0.3 | 29.6/20 | 14.0 ± 4.2 | 25.8/19 | 8.8 ± 1.7 | 5.5 σ |
| PKS B1424 - 418 | 23.0/23 | 19.0/22 | 2.0 ± 1.0 | 19.0/19 | < 0.3 | 18.0/20 | 6.1 ± 2.9 | 23.8/20 | < 4.2 | · · · |
| PKS 0426 - 380 | 42.3/23 | 27.5/22 | 3.8 ± 0.7 | 27.5/21 | < 0.4 | 22.9/22 | 9.6 ± 1.7 | 36.4/22 | 6.3 ± 2.6 | 4.5 σ |
| PKS 1502+106 | 30.5/21 | 22.5/20 | 3.2 ± 1.2 | 22.5/19 | < 0.17 | 21.1/20 | 9.0 ± 3.3 | 30.2/20 | 1.5 ± 1.3 | 3 σ |
| PKS 0537 - 441 | 46.0/23 | 34.1/22 | 3.2 +1 . 5 - 1 . 0 | 34.1/21 | < 1.4 | 29.3/22 | 9.1 ± 1.5 | 40.6/22 | 5.5 ± 2.6 | 4 σ |
| PKS 0454 - 234 | 35.7/23 | 28.7/22 | 3.7 ± 1.5 | 27.5/21 | < 0.4 | 28.2/22 | 10.6 ± 4.5 | 34.1/22 | 4.2 ± 3.4 | 2.5 σ |
| 4C +21.35 | 35.7/22 | 33.1/21 | 3.3 ± 2.3 | 25.5/20 | 1.8 +1 . 0 - 0 . 5 | 33.4/21 | 25 +14 - 21 | 23.8/21 | 11.2 ± 2.5 | 3.5 σ |
| PKS 1510 - 08 | 20.0/21 | 20.0/20 | < 0.5 | 20.0/19 | < 0.5 | 20.0/20 | < 1.4 | 20.0/20 | < 0.9 | · · · |
| 4C +55.17 | 66.0/21 | 57.9/20 | 3.1 ± 1.2 | 57.0/19 | < 0.8 | 57.0/20 | 7.9 ± 2.7 | 58.0/20 | 5.6 ± 1.2 | · · · |
| Group 1 | 44.0/23 | 36.8/22 | 1.0 ± 0.35 | 36.4/21 | < 0.05 | 30.2/22 | 3.4 ± 1.0 | 44.0/22 | < 1.2 | 3.5 σ |
| Group 2 | 65.6/23 | 42.6/22 | 2.0 ± 0.4 | 42.6/21 | < 0.1 | 31.6/22 | 6.2 ± 1.1 | 52.9/22 | 2.9 ± 1.2 | 6 σ |
| Group 3 (BL Lacs) | 35.4/22 | 33.7/21 | 0.37 ± 0.3 | 33.7/20 | < 0.1 | . . . | . . . | . . . | . . . | . . . |
- a The lognormal distribution with a single H LyC absorber.
- c The lognormal distribution with absorption provided by the BLR spectrum with ionization parameter log ξ = 1 5 (see Sect. 3 and Poutanen & Stern . 2010).
- b The lognormal distribution with a double absorber by H and He ii LyC.
- d Same as case c, but for log ξ = 2 5. .
- e The number of degrees of freedom (dof) differs because the spectra are cut at the first bin with negative flux.
- f Optical depth τ T due to H LyC only.
- h The significance of χ 2 reduction of the best-fit model with respect to the fits with lognormal function.
- g Optical depth τ T due to He ii LyC only.
## He ii does not change χ 2 significantly.
The typical optical depth, τ T , for the best-fit BLR emission model (mostly with log ξ = 1 5) was measured . to between 4 and 20. This corresponds to the maximum optical depth of about 0.4-2.2 (see blue dashed line in Fig. 3) and the flux reduction at ∼ 100 GeV by a factor of 1.5-9. To estimate the absorption optical depth that is contributed by H and He ii emission only, one can consider corresponding optical depths from single- or double-absorber models (see Table 2, columns 4 and 6).
case of PKS 0426 -380. This is much above possible systematic errors like uncertainties in the response function.
Spectra of the six from the nine brightest (above 5 GeV) blazars demonstrate clear absorption breaks dominated by the H LyC absorption. The significance of these breaks ranges from 2.5 σ to 5.5 σ . The typical optical depth by H LyC only is τ H ∼ 2-4, which can be converted directly to the column density of LyC (plus Ly α ) photons on the line of sight to the γ -ray emitting region N ph HLyC , = τ H /σ T ≈ (3 -6) × 10 24 . The stacked spectrum of Group 2 (which does not include 3C 454.3) has 6 σ significance of the break. This is the most confident and the most conservative demonstration that the H LyC absorption in bright blazars is a very typical phenomenon. Previously Poutanen & Stern (2010), Stern & Poutanen (2011), and recently Tanaka et al. (2013) revealed this absorption in individual sources. However, in first and in the last papers a less conservative assumption of a power law null hypothesis was used.
As for the GeV breaks associated with He ii absorption (now with a more accurate Pass 7 effective area), they get a status of a rare phenomenon. There are indications of such breaks in two objects: 3C 454.3 (3 σ ) and 4C +21.35 (almost 3 σ ), see Fig. 4.
## 4.2. Significance of the breaks
Detection of the breaks due to H LyC absorption has high significance. The amplitude of the spectral deviation from the null hypothesis is factor of 2 in the case of the stacked spectrum (Group 2) and factor of 3 in the
The breaks due to He ii LyC are much weaker than was claimed before (Ackermann et al. 2010; Abdo et al. 2011; Stern & Poutanen 2011) mostly because of the difference between Pass 6 and Pass 7 detector response functions. A sharp rise in the Pass 6 effective area (see Fig. 1) made the spectral break observed at 2-3 GeV sharper and more significant. However, we believe that the breaks observed in two objects, 3C 454.3 and 4C +21.35, are real. First, now there is no any feature in the LAT response function at the corresponding energy. Second, deviation of the spectra in the rest-frame range 5-20 GeV from the extrapolation of the lognormal function fitted to the data below 5 GeV reach at least 50% (see Fig. 4), much above the uncertainty in the LAT response. It should be noted that the 3 σ significance level is quite serious in this case, because the detection of the effect does not include 'hidden trials' like thresholds adjustments and sample manipulation. This significance is also very conservative as it was measured for the lognormal null hypothesis, not a power-law. For 3C 454.3, we do not expect systematical errors associated with the background subtraction or source confusion due to exceptionally high γ -ray brightness. 4C +21.35 is weaker and there could be some systematics, e.g., an underestimated soft background: the soft part of the spectrum could be lower and the spectrum could then be fitted with a narrower lognormal distribution without absorption. Therefore, the He ii absorption break in 4C +21.35 needs further studies.
In order to prove the presence of the GeV breaks in FSRQ, we checked whether the breaks appear also in the comparison spectra (see Fig. 6), where they are not expected. We see that the stacked BL Lac spectrum (Group 3) is well described by a power-law (with some deviations because of imperfectness of the response function) without any breaks (except possibly at 100 GeV). The upper limits on the opacity due to H or He LyC are significantly below the detected opacity in the bright
Figure 6. Spectral energy distribution of the 'empty' sky is shown by red open squares. The stacked spectrum of four brightest BL Lacs (Group 3: Mrk 421, 3C 66A, S5 0716+714 and PKS 2155 -304) is shown by blue circles. Neither of the spectra show any signs of absorption by H or He LyC in the range 2-20 GeV.
blazars (see Table 2). The 'empty' sky spectrum also does not show any signs of the breaks in the 2-20 GeV range. Thus, inaccuracies in the detector response are unlikely to affect our conclusions.
We also made a simple test whether a different null hypothesis for the underlying spectrum would remove the necessity of the breaks. Instead of the log-normal function, we assumed, following the referee's suggestion, a biquadratic function EF E ( ) ∝ exp[ -A ln 2 ( E/E peak ) -B ln 4 ( E/E peak )], which has the same number of free parameters as our log-normal distribution with BLR absorption. This ad hoc function gives a slightly better fit with χ / 2 dof = 20 19 for 3C 454.3, but a much worse / fit with χ / 2 dof = 41 / 22 for the Group 2 spectrum (see Table 2). Interestingly, a fit for 3C 454.3 gives A ≈ 0, so that the spectral curvature is fully determined by the quartic term; what is the physical meaning of such a model is a mystery to us.
## 5. DISCUSSION AND SUMMARY
Our main results can be formulated as follows.
- · We find that the 20 GeV breaks due to H LyC are ubiquitous. They are statistically significant in the majority of the bright blazars as well as in the stacked redshift-corrected spectra.
- · The 5 GeV breaks due to He ii LyC remain significant only in two objects.
- · A more complicated function describing the underlying spectrum can change the significance of the break existence. An ad hoc biquadratic function of ln E gives a slightly better fit that the physically justified lognormal function with the BLR absorption for 3C 454.3, but a much worse fit for the stacked blazar spectrum. Thus this model does not eliminate the need for the break.
- · Breaks are not seen in the stacked redshiftcorrected BL Lac spectrum or the spectrum of the 'empty' sky.
- · The presence of the breaks associated with absorption by UV photons implies that at least some fraction of the γ -rays are produced within the BLR.
The presence of He ii LyC absorption in 3C 454.3 is not surprising. This object is exceptional in all its components: the γ -ray emission from the jet reaches luminosities in excess of 2 × 10 50 erg s -1 (Abdo et al. 2011), the accretion disk emits L d ∼ 10 47 erg s -1 (Bonnoli et al. 2011), the BLR 3 × 10 45 erg s -1 (Pian et al. 2005), and luminosity in Ly α only is L Ly α ∼ 10 45 erg s -1 (Wills et al. 1995). Here we can expect that the ionization degree is high and the photon-photon optical depth is substantial.
The second object that shows He ii LyC absorption is 4C +21.35. That case is important because this FSRQ has been detected during a flare in the 70-400 GeV range by MAGIC (Aleksi´ c et al. 2011) and the coexistence of the absorption break and VHE emission is difficult to understand in a single emission zone scenario, because the multi-hundred GeV photons would have trouble escaping from the BLR. The γ -ray luminosity of 4C +21.35 is smaller than in 3C 454.3, 10 48 erg s -1 during flares (Tanaka et al. 2011), but the accretion disk is almost as luminous with L d ≈ 5 × 10 46 erg s -1 (Tavecchio et al. 2011). On the other hand, the BLR luminosity is significantly smaller 5 × 10 44 erg s -1 (Wang et al. 2004; Fan et al. 2006; Tanaka et al. 2011) with Ly α producing ∼ 10 44 erg s -1 , i.e. ten times less than in 3C 454.3. A much lower BLR luminosity implies that the opacity for 100 GeV photons is smaller, because BLR size scales roughly as R BLR ≈ 10 18 L 1 / 2 d 47 , cm (Kaspi et al. 2007; Bentz et al. 2009) and the opacity as τ T ∝ L line /L 1 / 2 d (see Equation (2) and Poutanen & Stern 2010). Thus in 4C +21.35, the optical depth through the BLR is expected to be 5-7 times smaller than in 3C 454.3. This might be the reason why VHE emission is detected in 4C +21.35, but not in a much stronger source 3C 454.3.
For a given object having both the measurement of the line luminosity and the estimation of the BLR size, we can obtain an expected value for the opacity τ T using Equation (2), i.e. assuming that the BLR emission is isotropic and γ -rays have to penetrate through the whole BLR. For 3C 454.3 with L d 47 , ∼ 1 and L Ly α, 45 ∼ 1, we get the opacity from H LyC/Ly α of τ H ∼ 100, while for 4C +21.35 we have τ H ≈ 15. We see, however, for both objects the observed value is τ H ∼ 2-5 (Table 2), much below the expectation. What are the possible solutions for this discrepancy? We can propose at least two solutions:
- · The size of the H LyC/Ly α emission region is larger than that measured from reverberation mapping using C iv (Kaspi et al. 2007) and H β lines (Bentz et al. 2009). We note that no Ly α variability was detected in any of the quasars analyzed by Kaspi et al. (2007) supporting this picture. This would immediately reduce the expected γ -ray opacity, which scales inversely proportionally with the size.
- · Alternatively, if the BLR is flat (Shields 1978; Decarli et al. 2011), i.e. elongated alone the accretion disk, the γ -ray opacity is much reduced (e.g. Lei & Wang 2014). In this case, the threshold energy, which depends on the maximum interaction
angle between the BLR photons and the γ -rays, should be a factor of two larger (i.e. ∼ 40 GeV instead of ∼ 20 GeV) contradicting the data. However, if in addition a few BLR clouds are situated along the jet axis outside the γ -ray emitting region, they would produce enough photons to collide head-on with the γ -rays to make a break at 20 GeV.
- · Finally, the γ -ray emitting region can be extended and its location can change depending on the luminosity, with less GeV absorption during the strong flares (Stern & Poutanen 2011; Pacciani et al. 2014). The VHE emission does not need to be produced in exactly the same place as the GeV emission. This can explain the fact that we see both GeV breaks as well as > 100 GeV emission in 4C +21.35 (but not necessarily at the same time).
How then these models compares with the results on He ii absorption? Photoionization models predict that luminosity in He ii LyC/Ly α is typically ∼ 10% of the hydrogen one (Tavecchio & Ghisellini 2008; Poutanen & Stern 2010). Thus, taking the same BLR size and geometry and using four times larger photon energy gives τ He ∼ τ H / 40. In that case, the He ii absorption would be negligible. On the other hand, reverberation mapping shows that He ii lines are produced closer to the black hole that e.g. H β (Peterson & Wandel 1999), implying a higher photon density inside the zone of complete He ionization and a larger τ He . However, if the γ -ray emitting region is slightly outside of that region the opacity is reduced. Again in this situation the break energy should be higher, but the quality of the data do not allow us to reject a hypothesis that the He ii break energy is actually two times larger than the fiducial (rest-frame) 5 GeV value. Thus we do not see an obvious contradiction between the fact that the γ -rays are produced outside (or at the edge) of He ii LyC emitting region and the presence of the break.
In general, the situation with the H LyC absorption strongly dominating over He ii LyC absorption seems more natural than the picture presented by Abdo et al. (2010) and Poutanen & Stern (2010), where a few GeV breaks look more prominent than those at ∼ 20 GeV. Now the statement of Poutanen & Stern (2010) that the jet emission takes place in the inner (higher ionization) regions of BLRs should be modified: the gamma-ray emission site lies within the normal H ii BLR region. This fact does not change the main astrophysical implications of photon-photon absorption of the jet gammaray emission, this particularly implies that the jet is already accelerated within a parsec distance from the black hole and therefore the Blandford-Znajek mechanism (Blandford & Znajek 1977; Komissarov et al. 2007) is responsible for the jet launching and the BLR dense photon field can provide conditions for energy dissipation via photon breeding (Stern & Poutanen 2006, 2008).
This research was supported by the Academy of Finland grant 268740 and the Magnus Ehrnrooth foundation. The research made use of public data obtained from the Fermi Science Support Center.
## REFERENCES
- Abdo, A. A., Ackermann, M., Ajello, M., et al. 2009, ApJ, 699,
- 817 Abdo, A. A., Ackermann, M., Ajello, M., et al. 2010, ApJ, 710, 1271 Abdo, A. A., Ackermann, M., Ajello, M., et al. 2011, ApJ, 733, L26 Ackermann, M., Ajello, M., Baldini, L., et al. 2010, ApJ, 721, 1383 Aleksi´, J., Antonelli, L. A., Antoranz, P., et al. 2011, ApJ, 730, c L8 Bentz, M. C., Walsh, J. L., Barth, A. J., et al. 2009, ApJ, 705, 199 Blandford, R. D. & Znajek, R. L. 1977, MNRAS, 179, 433 Bonnoli, G., Ghisellini, G., Foschini, L., Tavecchio, F., & Ghirlanda, G. 2011, MNRAS, 410, 368 Decarli, R., Dotti, M., & Treves, A. 2011, MNRAS, 413, 39 Fan, Z., Cao, X., & Gu, M. 2006, ApJ, 646, 8 Gould, R. J. & Schr´der, G. P. 1967, Physical Review, 155, 1404 e Harris, J., Daniel, M. K., & Chadwick, P. M. 2012, ApJ, 761, 2 Kallman, T. & Bautista, M. 2001, ApJS, 133, 221 Kaspi, S., Brandt, W. N., Maoz, D., et al. 2007, ApJ, 659, 997 Komissarov, S. S., Barkov, M. V., Vlahakis, N., & K¨nigl, A. o 2007, MNRAS, 380, 51 Laor, A., Fiore, F., Elvis, M., Wilkes, B. J., & McDowell, J. C. 1997, ApJ, 477, 93 Lei, M. & Wang, J. 2014, PASJ, 66, 7 Nolan, P. L., Abdo, A. A., Ackermann, M., et al. 2012, ApJS, 199, 31 Pacciani, L., Tavecchio, F., Donnarumma, I., et al. 2014, 790, 45 Peterson, B. M. & Wandel, A. 1999, ApJ, 521, L95 Pian, E., Falomo, R., & Treves, A. 2005, MNRAS, 361, 919 Poutanen, J. & Stern, B. 2010, ApJ, 717, L118 Shields, G. A. 1978, in Proc. of the Pittsburgh Conference on BL Lac Objects, ed. A. M. Wolfe (Pittsburgh: University of Pittsburgh), 257 Stern, B. E. & Poutanen, J. 2006, MNRAS, 372, 1217 Stern, B. E. & Poutanen, J. 2008, MNRAS, 383, 1695 Stern, B. E. & Poutanen, J. 2011, MNRAS, 417, L11 Tanaka, Y. T., Cheung, C. C., Inoue, Y., et al. 2013, ApJ, 777, L18 Tanaka, Y. T., Stawarz, /suppress L., Thompson, D. J., et al., 2011, ApJ, 733, 19 Tavecchio, F., Becerra-Gonzalez, J., Ghisellini, G., et al. 2011, A&A, 534, A86 Tavecchio, F. & Ghisellini, G. 2008, MNRAS, 386, 945 Wang, J.-M., Luo, B., & Ho, L. C. 2004, ApJ, 615, L9
- Wills, B. J., Thompson, K. L., Han, M., et al. 1995, ApJ, 447, 139 Zdziarski, A. A. 1988, ApJ, 335, 786
|
10.1088/0004-637X/794/1/8
|
[
"Boris E. Stern",
"Juri Poutanen"
] |
2014-08-04T20:00:01+00:00
|
2014-08-04T20:00:01+00:00
|
[
"astro-ph.HE",
"astro-ph.GA"
] |
The mystery of spectral breaks: Lyman continuum absorption by photon-photon pair production in the Fermi GeV spectra of bright blazars
|
We reanalyze Fermi/LAT gamma-ray spectra of bright blazars with a higher
photon statistics than in previous works and with new Pass 7 data
representation. In the spectra of the brightest blazar 3C 454.3 and possibly of
4C +21.35 we detect breaks at 5 GeV (in the rest frame) associated with the
photon-photon pair production absorption by He II Lyman continuum (LyC). We
also detect confident breaks at 20 GeV associated with hydrogen LyC both in the
individual spectra and in the stacked redshift-corrected spectrum of several
bright blazars. The detected breaks in the stacked spectra univocally prove
that they are associated with atomic ultraviolet emission features of the
quasar broad-line region (BLR). The dominance of the absorption by hydrogen Ly
complex over He II, rather small detected optical depth, and the break energy
consistent with the head-on collisions with LyC photons imply that the
gamma-ray emission site is located within the BLR, but most of the BLR emission
comes from a flat disk-like structure producing little opacity. Alternatively,
the LyC emission region size might be larger than the BLR size measured from
reverberation mapping, and/or the gamma-ray emitting region is extended. These
solutions would resolve a long-standing issue how the multi-hundred GeV photons
can escape from the emission zone without being absorbed by softer photons.
|
1408.0794v1
|
## An Enigmatic Pointlike Feature within the HD 169142 Transitional Disk
Beth A. Biller 1 2 , , Jared Males 3 , ∗ , Timothy Rodigas , Katie Morzinski 4 3 , ∗ , Laird M. Close , Attila 3 Juh´ asz 5 , Katherine B. Follette 3 , Sylvestre Lacour , Myriam Benisty , Aurora Sicilia-Aguilar 6 7 8 9 , , Philip M. Hinz , Alycia Weinberger , Thomas Henning , J¨rg-Uwe Pott , Micka¨l Bonnefoy 3 4 2 o 2 e 10 , Rainer K¨ ohler 2
[email protected]
## ABSTRACT
We report the detection of a faint pointlike feature possibly related to ongoing planetformation in the disk of the transition disk star HD 169142. The pointlike feature has a ∆mag(L) ∼ 6.4, at a separation of ∼ 0.11 ′′ and PA ∼ 0 . ◦ Given its lack of an H or K S counterpart despite its relative brightness, this candidate cannot be explained by purely photospheric emission and must be a disk feature heated by an as yet unknown source. Its extremely red colors make it highly unlikely to be a background object, but future multi-wavelength followup is necessary for confirmation and characterization of this feature. 1 2
1 Institute for Astronomy, University of Edinburgh, Blackford Hill, Edinburgh, UK
2 Max-Planck-Institut f¨r Astronomie, K¨nigstuhl 17, 69117 Heidelberg, Germany u o
3 Steward Observatory, University of Arizona, 933, N. Cherry Ave, Tucson, AZ, 85719, USA
* NASA Sagan Fellow
4 Department of Terrestrial Magnetism, Carnegie Institution of Washington, 5241 Broad Branch Road, NW Washington, DC 20015, USA
5 Leiden Observatory, Leiden University, P.O. Box 9513, 2300 RA Leiden, The Netherlands
6 LESIA, CNRS/UMR-8109, Observatoire de Paris, UPMC, Universit´ e Paris Diderot, 5 place Jules Janssen, 92195 Meudon, France
7 UJF-Grenoble 1 / CNRS-INSU, Institut de Plan´ etologie et d'Astrophysique de Grenoble (IPAG) UMR 5274, Grenoble, F-38041, France
8 Departamento de F´ ısica Te´rica, Facultad de Ciencias, Universidad Aut´noma de Madrid, 28049 Cantoblanco, o o Madrid, Spain
9 SUPA, School of Physics and Astronomy, University of St Andrews, North Haugh, St Andrews KY16 9SS, UK
10 Universit´ Grenoble Alpes, IPAG, F-38000 Grenoble, France. CNRS, IPAG, F-38000 Grenoble, France e
1 Based on observations made with ESO Telescopes at the La Silla Paranal Observatory under programme ID 091.C-0572.
2 This letter includes data gathered with the 6.5 Magellan Telescopes located at Las Campanas Observatory (LCO),
Subject headings:
## 1. Introduction
Transition disks trace a key step in the formation of planetary systems, intermediate between gas-rich protoplanetary disks and debris disks, where primordial gas is cleared away, leaving only remnant dust. These disks are observationally identified by weak mid-IR emission (at ∼ 15 µ m) relative to the Taurus median spectral energy distribution (Najita et al. 2007, i.e. the median SED of primordial disks in the young ( < 2 Myr) Taurus star-forming region). While numerous physical processes may be responsible for the depletion of gas and dust in transition disks, cleared gaps in particular may be an indicator of a planet or brown dwarf companion in the midst of formation. Thus, these disk systems have been key targets for direct imaging searches for planets.
Transition disks tend to be found in young star-forming regions > 100 pc from Earth. Only recently have imaging techniques been available to probe the inner portions of these disks, where planets are likely to form. In the last few years, there have been a number of notable discoveries of companions in transition disks with cleared gaps. Via the interferometric sparse-aperture masking (henceforth SAM) technique (Lacour et al. 2011), Kraus & Ireland (2012) found LkCa 15b, a candidate protoplanet embedded within the disk of a ∼ 2 Myr solar analog. Biller et al. (2012) detected a 0.1-0.4 M /circledot companion in the disk around the Herbig Ae/Be star HD 142527. This binary companion was recently confirmed and shown to be accreting via direct imaging (Close et al. 2014). HD 142527 has a particularly wide gap in its disk ( > 10-120 AU, Fukugawa et al. 2006, Rameau et al. 2012, Casassus et al. 2012), which may be explained by a stellar companion on an eccentric orbit. Finally, Quanz et al. (2013a) reported an as yet unconfirmed candidate protoplanet detection in the disk of the Herbig Ae/Be star HD 100546.
The Herbig Ae/Be star HD 169142 possesses a nearly face-on transition disk, and has been studied in detail both spectroscopically (Meeus et al. 2001, van Boekel et al. 2005), and through imaging (Habart et al. 2006, Kuhn et al. 2001, Hales et al. 2006, Grady et al. 2007, Fukugawa et al. 2010, Honda et al. 2012, Mari˜as et al. n 2011). Recently, Quanz et al. (2013b) detected a well-resolved annular gap from 40 - 70 AU via polarized light imaging. Thus, like LkCa 15, HD 142527, and HD 100546, it is an excellent candidate to possess a planetary mass or brown dwarf companion in the midst of formation.
Chile
## 2. Stellar Parameters
Adopted stellar parameters are presented in Table 1. In general, we adopt the stellar parameters from Table 1 of Quanz et al. 2013b. An accurate age estimate is particularly important for estimating candidate companion properties; HD 169142 has been assigned a fairly wide range of ages. Guimar˜ aes et al. 2006 assign a comparatively young age of 1-5 Myr based in HD 169142's HR diagram position, while Blondel & Djie 2006 claim a similar mass and age as β Pic for HD 169142. Grady et al. 2007 find a comoving companion 9.3' separated from HD 169142. This object is a 130 mas separation weak-line T Tauri binary, thus they assign an age of 6 +6 -3 Myr for the entire system. For our analysis here, we thus adopt an age range of 3-12 Myr. Sylvester et al. 1996 derive a photometric distance of 145 pc to HD 169142, based on an A5 spectral type; Blondell & Djie 2006 update this to 151 pc based on a spectral type of A7V. As derived spectral type will strongly affect the photometric distance, we adopt a distance range of 145 ± 10 pc here to account for any uncertainty in spectral typing.
## 3. Observations and Data Reduction
## 3.1. July 2013 VLT NACO Vortex Observations
First epoch observations were conducted using the novel annular groove phase mask (henceforth AGPM) vector vortex coronagraph (Mawet et al. 2013) with the NACO camera at the VLT (Lenzen et al. 2003; Rousset et al. 2003). The AGPM coronagraph uses an annular groove phase mask to redirect on-axis starlight out of the pupil (Mawet et al. 2005). In this manner, the AGPM coronagraph enables L' band contrasts of ∆mag > 7.5 mag at inner working angles down to 0.09 ′′ (Mawet et al. 2013). HD 169142 was observed in L' from 02:47 UT to 04:54 UT on 14 July 2013, covering nearly an hour both before and after transit. The derotator was turned off to enable azimuthal differential imaging (ADI) techniques. Seeing varied from 1-2 ′′ over the observation. We used a base exposure time of 0.25 s with 120 coadds and obtained a total of 1 hour 15 minutes on-sky exposure time. Sky frames were interspersed after every 20 frames and 0.05 s unsaturated exposures were taken at the beginning of the sequence. Care was taken to keep the primary star centered under the AGPM. Observation details are presented in Table 2.
Data were flat-fielded and bad-pixel corrected using dome flats and darks taken as part of ESO standard calibrations. Numerous dust spots are apparent on the AGPM, thus, slight misalignments between science and sky frames can produce significant cosmetic errors. Thus, we carefully selected the sky frame which minimized cosmetic errors for each science frame. Eleven science frames (out of 126 total science frames) which showed significant cosmetic errors even after sky subtraction were discarded. Each frame was centroided on the raw data using the idl routine mpfit2dpeak.
After basic data reduction and centroiding on the center of the coronagraphic mask using the IDL mpfit2dpeak routine, we analyze the data using 3 independent principal component analysis
(PCA) pipelines (following the algorithms of Soummer et al. 2012, Amara & Quanz 2012). All three pipelines yielded similar results - we report here the results using the publically available pipeline of Dimitri Mawet (http://www.sc.eso.org/ ∼ dmawet/DimitriMawet/IDL PCA pipeline.html).
## 3.2. April 2014 Magellan AO Observations
Followup observations were conducted using the 585 actuator 1000 Hz adaptive secondary AO System (MagAO) at the 6.5m Magellan Clay Telescope (Close et al. 2013). With 378 corrected modes, the MagAO system is one of the highest sampled AO systems on a large telescope ( > 5 m) and has demonstrated high spatial resolutions (20-30 mas) in the visible (as short as λ =0.6 µ m, Close et al. 2013). MagAO feeds both a visible (VisAO, 0.5-1 µ m) and infrared camera (Clio2, 1-5 µ m) simultaneously.
We obtained a total of 5 datasets for HD 169142 with MagAO from 8-15 April 2014. Observing details are presented in Table 2. Four of these datasets were obtained with the Clio2 Narrow camera including two H band observations, one K S band observation, and one 3.9 µ m 5% filter observation. We also obtained a deep z' dataset using the VisAO camera. The derotator was turned off to enable angular differential imaging (ADI) techniques. All MagAO data was reduced in the standard manner (flat-field correction, sky-subtraction, bad-pixel correction). The H, K S , and z' datasets remained unsaturated in the core, thus we align our images using the unsaturated core. We then process our data using the independent PCA-based pipelines of CoIs Rodigas and Males (see e.g. Rodigas et al. 2014 and Males et al. 2014). Both pipelines yielded similar results.
At 3.9 µ m, we acquired data for HD 169142 as well as for a PSF star of similar magnitude. We achieved very high Strehl ratios for both stars in dry, photometric conditions. Seeing remained around 0.4-0.6 ′′ for the entire night. After image registration, we built a PSF from the nearby star images and subtracted the PSF from each of our MagAO/Clio2 images of HD 169142, then rotated to place North up/East left. Our best subtraction is for a PSF scaling of 115%, which removes the Airy rings quite well. Dark pixels in the core are due to saturation/non-linearity.
## 4. Results
In July 2013, we detected a faint, pointlike feature in our L' vortex dataset at ∼ 0 ◦ PA, with a separation of ∼ 0.11'. This pointlike feature was independently imaged in June 2013 by Reggiani et al. 2014. Multiband followup observations were conducted using Magellan AO in April 2014. Images from all epochs are presented in Figs. 1 and 2. Astrometry and photometry are presented in Table 1. The correct rotation to place north up and east left was verified at each epoch. For the July 2013 vortex data, a dataset on HD 142527 was acquired on the same run. The bright HD 142527 disk (see e.g. Fukagawa et al. 2006, Rameau et al. 2012) is correctly derotated by the pipeline. For the April 2014 MagAO data, the correct derotation has been validated using images
of the Trapezium cluster.
## 4.1. July 2013 VLT NACO Vortex Observations
L' photometry and astrometry was derived for the pointlike feature by inserting scaled PSF images into the raw data and selecting the fluxes and positions that best match the actual feature. Best astrometry and photometry is presented in Table 1. Assuming a point source, our best subtraction yielded ∆mag=6.4 ± 0.2 for this feature, at a separation of 0.11 ± 0.03 ′′ and PA of 0 ± 14 . ◦ Given that the inner working angle of the vortex coronagraph is 0.09 ′′ (Mawet et al. 2013), this pointlike feature is marginally resolved at best.
To estimate the S/N level of our detection, we calculated the mean and standard deviation in a circular aperture with diameter of λ /D centered on the pointlike feature and in the 6 additional resolution elements available at this separation and then calculated S/N from equation 9 of Mawet et al. (2014) (appropriate for speckle-dominated regions):
$$S / N = \frac { \bar { x _ { 1 } } - \bar { x _ { 2 } } } { s _ { 2 } \sqrt { 1 + \frac { 1 } { n _ { 2 } } } }$$
where ¯ is the mean within the aperture containing the feature, x 1 ¯ is the mean of the remaining x 2 6 resolution elements, s 2 is the standard deviation of the remaining 6 resolution elements, and n 2 is the number of resolution elements not containing the feature. We find S/N ∼ 6 for the detected feature, thus this detection is unlikely to be a false positive - a result which is further bolstered by the independent detection of the same feature by Reggiani et al. 2014.
Initially assuming purely photospheric emission, we estimate mass using Monte Carlo methods to account for the range of possible ages and uncertainties in photometry for this system. An ensemble of 10 6 possible ages are drawn from a uniform distribution running from 3-12 Myr. An ensemble of 10 6 absolute magnitudes are simulated, assuming Gaussian errors on photometry and distance (0.3 mag error in ∆mag, 10 pc error in distance). We then interpolate with age and single band absolute magnitude to find the best mass for the companion from the DUSTY and COND models of Chabrier et al. (2000) and Baraffe et al. (2002). Mass range histograms are presented in Fig. 2. Adopting a 3-12 Myr age range at a distance of 145 pc and the DUSTY models, this candidate would correspond to a relatively high mass 60-80 M Jup brown dwarf.
## 4.2. April 2014 Magellan AO Observations
No counterpart to the L' detection was found in the April 2014 MagAO followup. Our noncoronagraphic 3.9 µ mfollowup imaging was too shallow to retrieve the pointlike feature (∆mag=5.6 at 0.11'). However, a 60-80 M Jup object at these ages should have been quite bright in the near-IR
(H band absolute magnitude < 7) and have been easily detected given our H and K s sensitivity. Indeed, our sensitivity is sufficient to detect a ∼ 7-15 M Jup object at this separation and age range. Such a high-mass companion should open up a very wide gap in the disk, inconsistent with the ring at similar radii observed polarimetrically by Quanz et al. 2013b. The lack of a near-IR counterpart to the L' detection signifies that we are not observing the photosphere of a substellar object here but may be a potential disk feature its extremely red colors indicate that it is extraordinarily unlikely to be a background.
We performed radiative transfer calculations to investigate whether or not a passively heated disk 'feature' can reproduce the observed point-like source. We used the best fit disk model parameters from Maaskant et al. 2013 to describe an axisymmetric disk. Then to simulate a clump we increased the pressure scale height in the disk. The scale height was increased as a 2D Gaussian centered on the observed point source with an FWHM of 7 AU and a peak perturbation of 30%. We used 0.1 µ m size astronomical silicates for the dust opacity to get an upper limit on the scattered light perturbations. Larger grains would scatter preferentially forward, perpendicular to the line of sight, which would decrease the contrast between the clump and the central star. Then we measured the contrast on the calculated L' images. We find that a 'dust clump' at the position of our candidate would have a contrast of ∼ 17 mag (1 × 10 -7 ) between the star and the clump, much fainter than the detected candidate.
This result is not surprising, given the distance of the pointlike feature from the star. Assuming we observe only thermal dust emission from an irradiated clump, if the clump peaks at L', it would have T ∼ 850 K. Thus, adopting this as the maximum temperature of a potential dust clump, an irradiated clump with this temperature would be at a distance of the order of 0.6 AU (considering temperature and radius of the primary from Quanz et al. 2013b), inconsistent with the > 10 AU separation of this detection. Tentatively, we consider this candidate to be a disk feature heated by an as yet unknown mechanism.
We also marginally detected a candidate companion in H and Ks with a separation of ∼ 0.18' and a PA of ∼ 33 , but at low significance (S/N=3-5, using the same procedure as for the pointlike ◦ L' feature). These data were taken at two different nods, with very different looking PSFs in the two nods, limiting the precision of our photometry to ± 0.5 mag. Thus, while this point source was detected in two different bands on 3 nights (with S/N ∼ 5 in the first dataset, but only S/N ∼ 3 in the additional datasets), it requires confirmation at an additional epoch, with a more stable PSF, to verify common proper motion and rule out imaging and reduction artifacts. If real and purely photospheric, this candidate would be a 8-15 M Jup planet / substellar object (DUSTY and COND models). No corresponding L' counterpart was found in the July 2013 L' vortex dataset. If purely photospheric, the L' counterpart to this object is expected to have an absolute magnitude > 8.5 mag (DUSTY or COND models), corresponding to an apparent magnitude of > 14.3 mag. This is considerably below our achieved contrast for this dataset at this separation (limiting magnitude of ∼ 12.7, from insertion and retrieval of PSF images into the raw data). We also do not see a counterpart to this point source in our deep MagAO/VisAO z' dataset, down to a contrast of 10 -4 .
If real, the very red colors of this object are inconsistent with background stars with spectral types earlier than mid to late M.
## 5. Discussion
The known geometry of this disk limits the potential masses of companions associated with imaged asymmetries. Quanz et al. 2013b image an inner dust ring peaking at 25 AU and an annular gap from 40-70 AU. The L' detection lies within the 25 AU dust ring and may be connected to ongoing companion formation in the disk. If associated with a massive object, it may have had some role in clearing out the region inside this ring. Our tentative H/K detection does not lie within the 40-70 AU gap but is coincident with the 25 AU dust ring. This suggests that any substellar/planetary counterpart to this tentative detection cannot be very massive, as a massive object ( /greaterorsimilar 0.55 M Jup should completely clear away dust from its immediate environment (Kley & Nelson, 2012). If massive objects are associated with these detections, they would be some of the closest companions imaged to date. Among confirmed directly imaged exoplanets, only β Pic b, HR 8799d, and HR 8799e have comparably small projected separations (Lagrange et al. 2010, Marois et al. 2008, 2010).
Some disk asymmetries detected via Sparse Aperture Masking (SAM) have been described due to forward scattering off an inclined disk (e.g. Cieza et al. 2013, Olofsson et al. 2013). However, HD 169142 is nearly face on (Quanz et al. 2013b), so forward scattering cannot have produced any observable structure in this case.
Our L' detection is also not well-described by dust features heated only by the central star; it is simply too far from the star to be heated enough to produce the observed emission. However, bright disk-related features have been observed in other transition disks. For instance, Kraus et al. 2013 detect bright asymmetries via sparse aperture masking in H, K', and L' band within the 15-40 AU inner gap in the pre-transitional disk V1247 Orionis. These asymmetries are not well-fit by a companion model; these authors attribute these features to spiral features or accretion streams due to the interaction of the dust disk with the substellar bodies inside the gap responsible for its clearing. Kraus & Ireland 2012 detect a protoplanet candidate (also via sparse aperture masking) around LkCa 15. While their detection is well fit by a single point source at K, the L detection appears extended and not entirely coincident with the K detection. They interpret the K detection as the protoplanet itself and the L detection as circumplanetary material potentially shocked and heated by jets from the protoplanet.
In the case of HD 169142, the possible energy source for the observed L' disk feature remains unknown. As previously noted, the lack of a H/K counterpart rules out photospheric emission and the L' detection is too strong to be described by passive heating from the star. If due to jets or accretion onto a forming planetary companion, we would expect a H α counterpart to the L' detection (Close et al. 2014). However, Follette et al. in prep have observed HD 169142 with
MagAO in H α and do not detect an H α counterpart to the L' disk feature.
The L' detection may be related to emission line regions inside the HD 169142 disk. Habart et al. 2006 found strong 3.3 µ m PAH emission (from a C-H stretching feature of neutral PAHs) in the HD 169142 disk out to 0.3' (using high-resolution AO-supported NACO long-slit spectroscopy and along a slit oriented north to south, i.e. intersecting our detection). Maaskant et al. 2014 also find strong PAH emission (however, primarily in ionized features) in the HD 169142 disk, likely stemming from the inner disk gap. A weak blue leak (on order 1%) in the NACO L filter might allow us sensitivity to PAH features in the disk. If our L' feature is indeed due to PAH emission, future 3.3 µ m follow-up observations should easily retrieve this feature.
## 6. Conclusions
We report the detection of a faint pointlike structure in the HD 169142 transition disk, also independently detected by Reggiani et al. 2014. This structure has ∆mag(L) ∼ 6.4, at a separation of ∼ 0.11 ′′ and PA of ∼ 0 . ◦ Given its lack of an H or K S counterpart despite its relative brightness, this object cannot be due to the photosphere of a substellar or planetary mass companion and must instead be a disk feature. However, the observed L' detection is too strong to be described by passive heating from the star and ongoing accretion is ruled out by the MagAO H α non-detection of Follette et al. in prep. While the location of the L' pointlike feature right within the disk gap strongly suggests it may be connected to ongoing planet-formation (i.e. whatever process cleared out the gap), the energy source fueling this feature without producing a corresponding near-IR counterpart still remains mysterious. Hopefully, future multi-wavelength followup observations will elucidate the source of this feature - and perhaps in the process refine our understanding of ongoing planet-formation in this disk.
MagAO was constructed with NSF MRI, TSIP, and ATI awards.
## REFERENCES
Amara, A., & Quanz, S. P. 2012, MNRAS, 427, 948 Baraffe, I., Chabrier, G., Allard, F., & Hauschildt, P. H. 2002, A&A, 382, 563 Biller, B., Lacour, S., Juh´sz, A., et al. 2012, ApJ, 753, L38 a Blondel, P. F. C., & Djie, H. R. E. T. A. 2006, A&A, 456, 1045 Casassus, S., Perez, M., S., Jord´n, A., et al. 2012, ApJ, 754, L31 a
Chabrier, G., Baraffe, I., Allard, F., & Hauschildt, P. 2000, ApJ, 542, 464
| Cieza, L. A., Lacour, S., Schreiber, M. R., et al. 2013, ApJ, 762, L12 |
|--------------------------------------------------------------------------------------------------|
| Close, L. M., Males, J. R., Morzinski, K., et al. 2013, ApJ, 774, 94 |
| Close, L. M., Follette, K. B., Males, J. R., et al. 2014, ApJ, 781, L30 |
| Fukagawa, M., Tamura, M., Itoh, Y., et al. 2006, ApJ, 636, L153 |
| Fukagawa, M., Tamura, M., Itoh, Y., et al. 2010, PASJ, 62, 347 |
| Grady, C. A., Schneider, G., Hamaguchi, K., et al. 2007, ApJ, 665, 1391 |
| Guimar˜es, a M. M., Alencar, S. H. P., Corradi, W. J. B., & Vieira, S. L. A. 2006, A&A, 457, 581 |
| Habart, E., Natta, A., Testi, L., & Carbillet, M. 2006, A&A, 449, 1067 |
| Hales, A. S., Gledhill, T. M., Barlow, M. J., & Lowe, K. T. E. 2006, MNRAS, 365, 1348 |
| Høg, E., Fabricius, C., Makarov, V. V., et al. 2000, A&A, 357, 367 |
| Honda, M., Maaskant, K., Okamoto, Y. K., et al. 2012, ApJ, 752, 143 |
| Kley, W., & Nelson, R. P. 2012, ARA&A, 50, 211 |
| Kraus, A. L., & Ireland, M. J. 2012, ApJ, 745, 5 |
| Kraus, S., Ireland, M. J., Sitko, M. L., et al. 2013, ApJ, 768, 80 |
| Kuhn, J. R., Potter, D., & Parise, B. 2001, ApJ, 553, L189 |
| Lacour, S., Tuthill, P., Amico, P., et al. 2011, A&A, 532, A72 |
| Lagrange, A.-M., Bonnefoy, M., Chauvin, G., et al. 2010, Science, 329, 57 |
| Lenzen, R., Hartung, M., Brandner, W., et al. 2003, Proc. SPIE, 4841, 944 |
| Maaskant, K. M., Honda, M., Waters, L. B. F. M., et al. 2013, A&A, 555, A64 |
| Maaskant, K. M., Min, M., Waters, L. B. F. M., & Tielens, A. G. G. M. 2014, A&A, 563, A78 |
| Males, J. R., Close, L. M., Morzinski, K. M., et al. 2014, ApJ, 786, 32 |
| Malfait, K., Bogaert, E., & Waelkens, C. 1998, A&A, 331, 211 |
| Mari˜as, n N., Telesco, C. M., Fisher, R. S., & Packham, C. 2011, ApJ, 737, 57 |
| Marois, C., Macintosh, B., Barman, T., et al. 2008, Science, 322, 1348 |
| Marois, C., Zuckerman, B., Konopacky, Q. M., Macintosh, B., & Barman, T. 2010, Nature, 1080 |
```
1
2 3 4 0 6 8 7 9 10 5 11 19 17 18 16
```
van Boekel, R., Min, M., Waters, L. B. F. M., et al. 2005, A&A, 437, 189
This preprint was prepared with the AAS LT E X macros v5.2. A
Table 1. Properties of the HD 169142 System
| | Primary | Disk Feature |
|-----------------------------------|-----------------------|---------------------------|
| Distance | 145 ± 10 a b | 145 ± 10 a b |
| Age | 3-12 Myr | 3-12 Myr |
| Proper Motion ( µ α , µ δ ) | (-2.1 ± 1.5, - 40.2 ± | 1.5) mas/yr c |
| Separation: 14 July 2013 UT | · · · | 0.11 ± 0.03 ′′ ( ∼ 16 AU) |
| Position Angle: 14 July 2013 UT ′ | · · · | 0 ± 14 ◦ |
| ∆ L | · · · | 6.4 ± 0.3 |
| H (mag) | 6.91 d | · · · |
| Ks (mag) | 6.41 d | · · · |
| L ′ (mag) | 5.64 e | 12.0 ± 0.3 |
| M L ′ (mag) | -0.2 | 6.2 ± 0.3 |
| Spectral type | A7V f | · · · |
| Estimated Mass | 1.65 M /circledot f | · · · |
a Sylvester et al. (1996), but errors derived here.
b Grady et al. (2007)
c Høg et al. (2000)
d From 2MASS
e Malfait et al. (1998)
f Blondel & Djie (2006)
Table 2. Observation log
| Telescope/Instrument | UT Date | Band | Exposure Time | Field Rotation |
|------------------------|---------------|---------|-----------------|------------------|
| VLT NACO Vortex | 14 July 2013 | L' | 75 min | 135.2 |
| MagAO CLIO-2 Narrow | 9 April 2014 | H | 46 min | 178.5 |
| MagAO CLIO-2 Narrow | 10 April 2014 | H | 30.4 min | 172.9 |
| MagAO CLIO-2 Narrow | 12 April 2014 | K S | 70 min | 180.1 |
| MagAO CLIO-2 Narrow | 15 April 2014 | 3.9 µ m | 60 min | 164.1 |
| MagAO VisAO Narrow | 15 April 2014 | z p | 82.4 | 176.8 |
Fig. 1.Left: L' July 2013 NACO Vortex image. We find a pointlike feature in the disk with ∆mag=6.4 ± 0.2, at a separation of 0.11 ± 0.03 ′′ and PA of 0 ± 14 . ◦ Right: 3.9 µ m April 2014 PSFsubtracted MagAO CLIO-2 image. Dark pixels in the core are due to saturation/non-linearity. The April 2014 MagAO image reached shallower contrasts than the original vortex image, hence the faint point-like feature is not retrieved. Our noncoronagraphic dataset was too shallow to retrieve the candidate (∆mag=5.6 at 0.11').
Fig. 2.Top left and right : H band April 2014 MagAO CLIO-2 images over two consecutive nights, bottom left: K band April 2014 MagAO CLIO-2 image, bottom right: z' band April 2014 MagAO VisAO image. All images have been reduced with PCA. The position of the L' pointlike feature is circled.
Fig. 3.- Mass estimate histogram for the pointlike feature observed in L'. If due to photospheric emission from a companion, this object would have mass > 40 M Jup and an easily detected H and Ks counterpart.
|
10.1088/2041-8205/792/1/L22
|
[
"Beth A. Biller",
"Jared Males",
"Timothy Rodigas",
"Katie Morzinski",
"Laird M. Close",
"Attila Juhász",
"Katherine B. Follette",
"Sylvestre Lacour",
"Myriam Benisty",
"Aurora Sicilia-Aguilar",
"Philip M. Hinz",
"Alycia Weinberger",
"Thomas Henning",
"Jörg-Uwe Pott",
"Mickaël Bonnefoy",
"Rainer Köhler"
] |
2014-08-04T20:00:01+00:00
|
2014-08-04T20:00:01+00:00
|
[
"astro-ph.EP",
"astro-ph.SR"
] |
An Enigmatic Pointlike Feature within the HD 169142 Transitional Disk
|
We report the detection of a faint pointlike feature possibly related to
ongoing planet-formation in the disk of the transition disk star HD 169142. The
pointlike feature has a $\Delta$mag(L)$\sim$6.4, at a separation of $\sim$0.11"
and PA$\sim$0$^{\circ}$. Given its lack of an H or K$_{S}$ counterpart despite
its relative brightness, this candidate cannot be explained by purely
photospheric emission and must be a disk feature heated by an as yet unknown
source. Its extremely red colors make it highly unlikely to be a background
object, but future multi-wavelength followup is necessary for confirmation and
characterization of this feature.
|
1408.0797v2
|
## MICROLENS MASSES FROM 1-D PARALLAXES AND HELIOCENTRIC PROPER MOTIONS
Andrew Gould
Department of Astronomy Ohio State University, 140 W. 18th Ave., Columbus, OH 43210, USA [email protected]
Received -; accepted -
Abstract: One-dimensional (1-D) microlens parallaxes can be combined with heliocentric lens-source relative proper motion measurements to derive the lens mass and distance, as suggested by Ghosh et al. (2004). Here I present the first mathematical anlysis of this procedure, which I show can be represented as a quadratic equation. Hence, it is formally subject to a two-fold degeneracy. I show that this degeneracy can be broken in many cases using the relatively crude 2-D parallax information that is often available for microlensing events. I also develop an explicit formula for the region of parameter space where it is more difficult to break this degeneracy. Although no mass/distance measurements have yet been made using this technique, it is likely to become quite common over the next decade.
Key words: astrometry - gravitational microlensing - planets - stars: fundamental parameters (masses)
## 1. INTRODUCTION
Even at the dawn of microlensing experiments 20 years ago, it was already realized that vastly more microlensing events will yield one-dimensional (1-D) microlens parallaxes than full (2-D) parallaxes (Gould et al., 1994). However, these 1-D parallaxes remain largely unexploited.
The amplitude of the microlens parallax vector π E is the ratio of the lens-source relative parallax π rel to the angular Einstein radius θ E , while its direction is set by the lens-source relative proper motion µ ,
$$\pi _ { \text{E} } = \frac { \pi _ { \text{rel} } } { \theta _ { \text{E} } } \, \frac { \mu } { \mu }. \quad \ \ \ \ ( 1 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
The numerator of the first ratio quantifies the apparent angular lens-source displacement as the observer changes position, while the denominator translates this angular scale into the scale of microlensing phenomena. The second term appears because the evolution of microlensing effects due to parallax depends on the direction of lens-source relative motion. See Figure 1 of Gould & Horne (2013).
Note that from the definition of θ E ,
$$\theta _ { \text{E} } ^ { 2 } \equiv \kappa M \pi _ { \text{rel} }, \quad \kappa \equiv \frac { 4 G } { c ^ { 2 } \text{AU} } \simeq 8. 1 \frac { \max } { M _ { \odot } }, \quad ( 2 ) \, \begin{array} { c } \text{Snos.} \\ \text{Mura} \\ \text{the n} \\. \end{array}.$$
measurement of both θ E and π E directly leads to determinations of the lens mass and relative parallax
$$M = \frac { \theta _ { \text{E} } } { \kappa \pi _ { \text{E} } }, \quad \pi _ { \text{rel} } = \theta _ { \text{E} } \pi _ { \text{E} }. \quad \quad ( 3 ) \quad ( 2 0 0 \, \\ \max = 0$$
The reason that 1-D parallaxes are easier to measure is that, to the extent that µ is parallel to the direction of
Earth's acceleration (projected on the sky at the peak of the event) ˆ n a , it causes the event evolution to speed up or slow down as it progresses, thereby inducing an asymmetric distortion on the lightcurve. Since microlensing lightcurves are otherwise intrinsically symmetric, such distortions are easily measured. By contrast, to the extent that µ is perpendicular to ˆ n a , it gives rise to a symmetric distortion, which is much more difficult to disentangle from other symmetric parameters. Hence, we define
$$\pi _ { E, \| } \equiv \pi _ { E, \text{geo} } \cdot \hat { \mathbf n } _ { a } \quad \quad \quad ( 4 )$$
and π E , ⊥ = ( π 2 E -π 2 E , ‖ ) 1 / 2 (see Gould 2004 for sign conventions).
A very important notational point for the current work is that π E is subscripted with 'geo' (to indicate geocentric frame at the peak of the event) whereas π E is not. This is because the magnitude of this vector does not depend on the frame of reference, but the direction does (due to explicit dependence on µ /µ in Equation (1)).
Despite their predicted ubiquity, 1-D parallaxes are almost never reported in the literature. Among the handful of exceptions (Park et al., 2004; Ghosh et al., 2004; Jiang et al., 2004; Dong et al., 2009; Muraki et al., 2011), the first three reports are due to the novelty of the phenomenon and the last two are due to the fact that the actual value of π E , ‖ was so large that π E (and so M ) could be reasonably estimated despite the poor measurement of π E , ⊥ .
Nevertheless, as first pointed out by Ghosh et al. (2004), 1-D microlens parallaxes could yield microlens mass (and distance) measurements simply by making a late-time measurement of the heliocentric proper motion. Stated in the most naive way, knowing one component ( π E , ‖ ) of the vector π E (from the microlensing event) and its direction µ /µ (from late-time astrome-
try), one can extract the amplitude of the full vector π E . Then knowing the Einstein timescale t E (again from the event) and the amplitude of proper motion µ , one can determine the angular Einstein radius θ E = µt E .
The problem with this naive reasoning, as already recognized by Ghosh et al. (2004), is that π E , ‖ and t E are derived in the geocentric frame, whereas µ is measured in the heliocentric frame. The relation between these frames, as noted by Ghosh et al. (2004) but in a form closer to that preferred by Janczak et al. (2010), is
$$\mu _ { \text{hel} } = \mu _ { \text{geo} } + \mu _ { \oplus } \pi _ { \text{rel} } = \frac { \theta _ { \text{E} } } { t _ { \text{E} } } \frac { \pi _ { \text{E,geo} } } { \pi _ { \text{E} } } + \mu _ { \oplus } \pi _ { \text{E} } \theta _ { \text{E} }, \quad ( 5 ) \quad \text{Note t}$$
where µ ⊕ ≡ v ⊕ ⊥ , / AU and v ⊕ ⊥ , is the transverse velocity of Earth in the frame of the Sun at the peak of the event, projected on the plane of the sky.
However, in the intervening 10 years, there has been essentially no work aimed at understanding Equation (5). For example, it has not even been recognized that this equation leads to a quadratic equation in π E , ⊥ , which therefore has two (or zero) real solutions.
Nevertheless, many changes in the observational landscape are leading to radical improvements for the prospects of implementing the original suggestion of Ghosh et al. (2004). First is the simple fact that the massive search for events made possible by the inauguration of the Optical Gravitational Lensing Experiment (OGLE-III) survey in 2002 are now more than 10 years old, implying that sources and lenses have already substantially separated. Second, the Giant Magellan Telescope (GMT) is already under construction and likely to be operational within another 10 years. When it is, its FWHM in J band will be about 11 mas, meaning that sources and lenses separated by just 2 FWHM (22 mas) will be eligible for very good µ hel measurements. Third, microlensing event detections have already increased dramatically since 2011 with the inauguration of the OGLE-IV survey and are likely to accelerate further in 2015 with the inauguration of the new Korea Microlensing Telescope Network (KMTNet) of 1.6m telescopes with 4 deg 2 cameras on three continents. Since typical lens-source proper motions are µ ∼ 4 mas yr -1 , many of these events will be accessible to GMT. Finally, it is quite plausible that new space-based wide-field survey telescopes, like Euclid and WFIRST will survey essentially the entire bulge microlensing field at roughly the same epoch as GMT first light. Since these have much smaller apertures than GMT, their FWHM will be much larger (roughly 140 mas for Euclid and 110 for WFIRST ). However, their greater point-spread-function stability will permit proper-motion measurements at 1 FWHM (rather than 2), and their wider (i.e., systematic) coverage will permit a systematic search for relatively high proper-motion archival events over a very large field.
Of course, once such data are obtained, the analysis will proceed by simultaneous fitting of the microlensing and astrometric data. However, a proper analytic investigation of Equation (5) is still very important for understanding what can be learned from such observations, which is the central motivation for taking them.
## 2. QUADRATIC FORM
Simply taking the ratio of the two components (perpendicular and parallel to ˆ n a ) of µ hel (Equation (5)) yields the tangent of the angle φ of this vector relative to ˆ n a ,
$$\text{orm} \quad \text{tan} \, \phi \equiv \frac { \mu _ { \text{hel}, \perp } } { \mu _ { \text{hel}, \| } } = \frac { \pi _ { \text{E}, \perp } + \theta _ { \oplus, \perp } \pi _ { \text{E} } ^ { 2 } } { \pi _ { \text{E}, \| } + \theta _ { \oplus, \| } \pi _ { \text{E} } ^ { 2 } } ; \quad \theta _ { \oplus } \equiv \mu _ { \oplus } t _ { \text{E} }. \ \ ( 6 )$$
Note that θ ⊕ = 0 23( . v ⊕ ⊥ , / 20 kms -1 )( t E / 20 day). Then writing π 2 E = π 2 E , ‖ + π 2 E , ⊥ and rearranging terms yields
$$A \pi _ { E, \perp } ^ { 2 } - \pi _ { E, \perp } + C = 0,$$
where
$$\underset { \tau _ { E, \perp }, } { \overset { \dots } { \i z e d } } \quad A \equiv \theta _ { \oplus, \| } \tan \phi - \theta _ { \oplus, \perp } ; \ \ C \equiv A \pi _ { E, \| } ^ { 2 } + \pi _ { E, \| } \tan \phi, \ \ ( 8 )$$
which has the solutions
$$\pi _ { \text{E}, \perp } = \frac { 1 \pm \sqrt { 1 - 4 A C } } { 2 A }. \text{ \quad \ \ } ( 9 )$$
## 2.1. Limit of 4 AC /lessmuch 1
In the limit 4 AC /lessmuch 1, the two solutions of Equation (9) can be approximated as
$$\pi _ { \text{E,} \perp } = C ( 1 + A C + \dots ) \quad ( 4 A C \ll 1 ) \quad ( 1 0 )$$
and
$$\mathbb { I } \ ( 2 2 \\ \text{ents.} _ { \text{ly in} } \quad \pi _ { E, \perp } = \frac { 1 } { A } - C + \dots \quad ( 4 A C \ll 1, \ \text{ alternate} ). \ ( 1 1 )$$
Note that both solutions are perfectly valid. The leading term in the first solution corresponds to ignoring the first term in Equation (7) and yields 'small' π E , ⊥ , while the leading term in the second solution corresponds to ignoring the last term in Equation (7) and yields 'large' π E , ⊥ .
In practice, there will be some information about π E , ⊥ from the microlensing event, which may well be adequate to break this degeneracy. The important point is, however, that these two solutions are likely to correspond to distinct minima, so that they will not both automatically be probed by simple downhill minimization: they must both be explicitly checked. This situation is similar to the 'jerk-parallax' degenerate solutions, which also constitute the roots of quadratic equation in π E , ⊥ . See Equation (20) of Gould (2004). In fact, prior to the discovery of this degeneracy, Alcock et al. (2001) found one of the two solutions by downhill minimization. Gould (2004) then found the second solution from the symmetries of the quadratic equation, which turned out to have equally good χ 2 . Subsequent astrometric measurements by Drake et al. (2004) showed that the second solution was correct (Gould et al., 2004).
## 2.2. General Case
Equation (9) can be reformulated to eliminate ' C ' in favor of direct observables
$$\pi _ { E, \perp } = \frac { 1 \pm \sqrt { \sec ^ { 2 } \phi - ( 2 A \pi _ { E, \| } + \tan \phi ) ^ { 2 } } } { 2 A }. \quad ( 1 2 ) \ \ \text{critic} \\ \text{that is}$$
Thus, if 2 Aπ E , ‖ ∼ sec φ -tan φ , then the argument under the radical (discriminant) is close to zero and there is a danger that the error ellipses from the two solutions merge and/or cannot be distinguished by the microlensing data. Thus checking for this approximate equality is an important diagnostic.
If the source lies on the ecliptic, then θ ⊕ ⊥ , = 0. Because microlensing fields lie close to the ecliptic, this can often be a useful approximation. In this case
$$\pi _ { E, \perp } & = \frac { \cot \phi \pm \sqrt { \csc ^ { 2 } \phi - ( 1 + 2 \theta _ { \oplus, \| } \pi _ { E, \| } ) ^ { 2 } } } { 2 \theta _ { \oplus, \| } } ; \quad ( \theta _ { \oplus, \perp } \equiv 0 ). \quad \text{$current$} \text{$top$} \\ \text{In the form it is clear that as $ok$ tends toward $+\pi/9$} \quad \text{$\colon$} \colon.$$
In this form, it is clear that as φ tends toward ± π/ 2, the discriminant will very likely be small (and hence the solution prone to degeneracy).
Another key point is that even if π E , ‖ = 0, it may still be possible to measure π E , ⊥ (and so π E ) from the direction of µ hel . This would not be possible from the naive perspective outlined in Section 1. Explicitly,
$$\pi _ { E, \perp } = \frac { 1 } { A } = \frac { 1 } { \theta _ { \oplus, \| } \tan \phi - \theta _ { \oplus, \perp } } \quad ( \pi _ { E, \| } = 0 ). \quad ( 1 4 ) \quad \text{as $M^{1}$} \\ \text{...} \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \end{cases}$$
This is important, particularly if there are widefield high-resolution survey data (from e.g., Euclid or WFIRST ) for which special observations of 'lowprobability' (i.e., π E , ‖ ∼ 0) targets are not required. However, events that have high priority (such as those with planetary events) with π E , ‖ ∼ 0 could be targeted for individual observations to measure µ hel , particularly if the planetary-event lightcurve indicated a high geocentric scalar proper motion µ geo (Henderson et al., 2014).
## 3. ERROR ANALYSIS
While v ⊕ ⊥ , is known exactly, none of the quantities entering Equation (7) are known exactly. In particular, θ ⊕ = v ⊕ ⊥ , t E / AU, and t E is a measured quantity from the event. Similarly, π E , ‖ is measured from the event, while tan φ comes from the proper motion measurement. However, in most cases, π E , ‖ will be measured with substantially worse precision than the other quantities. Hence, it is useful to approximate the others as 'known perfectly' and ask how the uncertainty in π E , ⊥ depends on the measurement of π E , ‖ . To determine this, I differentiate Equation (7) and find
$$\frac { \delta \pi _ { \text{E,} \perp } } { \delta \pi _ { \text{E,} \| } } = \frac { \tan \phi + 2 A \pi _ { \text{E,} \| } } { 1 - 2 A \pi _ { \text{E,} \perp } }. \quad \quad ( 1 5 ) \quad \frac { \text{row} } { \frac { \text{two} } { \text{I} \text{It} } } \\. \quad \. \. \. \. \. \. \. \. \. \ \ f i c i$$
Clearly the appearance of π E , ⊥ in the denominator of the right hand side reflects a fundamental shortcoming of making a linearized analysis of an intrinsically nonlinear problem. Nevertheless, this expression points to the possibility of a strong degeneracy if π E , ⊥ ∼ 1 / 2 A . Note from Equation (9) that this corresponds to the discriminant (1 -4 AC ) being close to zero 1 . Recall from Section 2.2 that this is the same region of solution space that is potentially most sensitive to the discrete degeneracy. Therefore, measurements of π E , ‖ ∼ 1 / 2 A are particularly problematic. Note, however, from comparison with Equation (14) that this degeneracy specifically does not apply to the π E , ⊥ = 0 case.
## 4. INFORMATION FROM DIRECT IMAGING
Of course, if the lens is separately imaged from the source, it is possible in principle to make a photometric estimate of its mass and distance. However, such measurements face a number of challenges. First, the photometry will most often be done in the near-infrared, a spectral region for which reddening is highly degenerate with intrinsic temperature. Second, even if the lens is in the Galactic bulge (and so behind essentially all the dust, whose extinction properties can then be measured from nearby clump stars, e.g., Nataf et al. 2013), even main-sequence stars can differ in mass by several tens of percent at fixed color. Finally, and most fundamentally, most stars are in binaries, and for a significant minority of cases it will be the lower-mass component that gives rise to the microlensing event (because event rate scales as M 1 / 2 ), while the more massive component will be seen by direct imaging (because light is a high power of mass).
Note that while in high-resolution imaging of microlensing lens/source pairs carried out to date it has often been possible to rule out binary companions over most of parameter space, this will not be possible in the imaging of more generic events in the future. For example, Batista et al. (2014) were able to rule out companions closer than 11 mas. This corresponded to physical separations less than 80 AU. However, the basis for this limit was that the event (MOA-2011-BLG-293) had very high magnification, making it very sensitive to binary companions. By contrast, typical events with 1-D parallaxes will not be high-magnification, implying that microlensing-based constraints on companions will be very weak.
Thus, overall, mass/distance determinations from ( π E , ‖ , µ hel ) will be more accurate than those derived from photometry, and in a substantial minority of cases there will be significant disagreement due to lensing by secondary components of binaries. It is true that in these cases the initial µ hel measurement will be in error due to an implicit assumption that the primary and the source were initially aligned, whereas actually the alignment was with the unseen (or barely seen) secondary. However, once these cases are identified, they can be rectified by a second epoch of imaging, which would directly
1 It is a generic property of quadratic equations that if the coefficients depend on some quantity q , then the ratio of the error in the solution to the error in q diverges as the discriminant approaches zero
measure the proper motion of the primary. This will differ from the proper motion of the secondary (i.e., the lens) due to orbital motion, but usually by an amount that is very small compared to the lens-source proper motion itself.
In brief, photometric mass/distance estimates can be an important check on estimates derived from a combination of 1-D parallaxes and heliocentric proper motions, but they will generally be less precise and less accurate.
## 5. CONCLUSION
Determinations of microlens masses from the combination of 1-D microlens parallax ( π E , ‖ ) and heliocentric proper motion ( µ hel ) are likely to become quite common over the next decade. I have shown that solutions derived from such data are in the form of a quadratic equation and therefore have an intrinsic two-fold degeneracy. This degeneracy may be broken by microlensing data, which weakly constrain π E , ⊥ even when they are unable to measure it precisely. The degeneracy is most severe when the discriminant of the quadratic equation is near zero. In this case, each solution separately has large errors, so that the two solutions may merge together. Photometric mass/distance estimates can be an important check on mass/distance determinations based on 1-D parallaxes and heliocentric proper motions, but are overall less accurate and less precise.
## ACKNOWLEDGMENTS
I thank Radek Poleski for stimulating discussions. This work was supported by NSF grant AST 1103471 and NASA grant NNX12AB99G.
## REFERENCES
Alcock, C., Allsman, R.A., Alves, D.R. et al. 2001, Direct detection of a microlens in the Milky Way, Nature, 414, 617
Batista, V., Beaulieu, J.-P., Gould, A., Bennett, D.P., Yee, J.C., Fukui, A., Sumi, T., & Udalski, A. 2014, MOA-2011-BLG-293Lb: First Microlensing Planet Possibly in the Habitable Zone, ApJ, 780, 54
Dong, S., Gould, A., Udalski, A., et al. 2009, OGLE2005-BLG-071Lb, the Most Massive M Dwarf Planetary Companion?, ApJ, 695, 970
Drake, A.J., Cook, K.H., & Keller, S.C. 2004, Resolving the Nature of the Large Magellanic Cloud Microlensing Event MACHO-LMC-5, ApJ, 607, L29
Ghosh, H., DePoy, D.L., Gal-Yam, A. et al. 2004, Potential Direct Single-Star Mass Measurement, ApJ, 615, 450
Gould, A. 2000, A Natural Formalism for Microlensing, ApJ, 542, 785
- Gould, A. 2004, Resolution of the MACHO-LMC-5 Puzzle: The Jerk-Parallax Microlens Degeneracy, ApJ, 606, 319,
Gould, A, Bennett, D.P., & Alves, D.R. 2004, The Mass of the MACHO-LMC-5 Lens Star, ApJ, 614, 404
Gould, A. & Horne, K. 2013, Kepler-like Multi-plexing for Mass Production of Microlens Parallaxes ApJL,
779, L28 Gould, A., Miralda-Escud´ e, J. & Bahcall, J.N. 1994, Microlensing Events: Thin Disk, Thick Disk, or Halo?, ApJ, 423, L105 Henderson, C.B., Park, H., Sumi, T. et al. 2014, Candidate Gravitational Microlensing Events for Future Direct Lens Imaging, ApJ, 794, 71 Janczak, J., Fukui, A., Dong, S., et al. 2010, Sub-Saturn Planet MOA-2008-BLG-310Lb: Likely to be in the Galactic Bulge, ApJ, 711, 731 Jiang, G., DePoy, D.L., Gal-Yam, A. et al. 2004, OGLE2003-BLG-238: Microlensing Mass Estimate of an Isolated Star, ApJ, 617, 1307 Muraki, Y., Han, C., Bennett, D.P., et al. 2011, Discovery and Mass Measurements of a Cold, 10 Earth Mass Planet and Its Host Star ApJ, 741, 22 Nataf, D.M., Gould, A., Fouqu´, P. et al. 2013, Reddene ing and Extinction toward the Galactic Bulge from OGLE-III: The Inner Milky Way's RV 2.5 Extinction Curve, ApJ, 769, 88 Park B.-G., DePoy, D.L., Gaudi, B.S. et al. 2004, MOA
2003-BLG-37: A Bulge Jerk-Parallax Microlens Degeneracy, ApJ, 609, 166
|
10.5303/JKAS.2014.47.6.215
|
[
"Andrew Gould"
] |
2014-08-04T20:00:05+00:00
|
2014-09-25T23:03:38+00:00
|
[
"astro-ph.SR",
"astro-ph.EP"
] |
Microlens Masses from 1-D Parallaxes and Heliocentric Proper Motions
|
One-dimensional (1-D) microlens parallaxes can be combined with heliocentric
lens-source relative proper motion measurements to derive the lens mass and
distance, as suggested by Ghosh et al. (2004). Here I present the first
mathematical anlysis of this procedure, which I show can be represented as a
quadratic equation. Hence, it is formally subject to a two-fold degeneracy. I
show that this degeneracy can be broken in many cases using the relatively
crude 2-D parallax information that is often available for microlensing events.
I also develop an explicit formula for the region of parameter space where it
is more difficult to break this degeneracy. Although no mass/distance
measurements have yet been made using this technique, it is likely to become
quite common over the next decade.
|
1408.0798v2
|
## Determination of the Higgs CP mixing angle in the tau decay channels at the LHC including the Drell-Yan background
## Stefan Berge , Werner Bernreuther 1 2 and Sebastian Kirchner 3
Institut für Theoretische Physik, RWTH Aachen University, 52056 Aachen, Germany
## Abstract
We investigate how precisely the CP nature of the 125 GeV Higgs boson resonance h can be unraveled at the LHC in its decays to t pairs, h → -t t + . We use a method which allows to determine the scalarpseudoscalar Higgs mixing angle ft in this decay mode. This mixing angle can be extracted from the distribution of a signed angle, denoted by j ∗ CP , which we analyze for the major charged-prong t decays. For definiteness, we consider Higgs-boson production by gluon fusion at NLO QCD. We take into account also the irreducible background from Drell-Yan production, Z ∗ / g ∗ → tt , at NLO QCD. We compute, for the signal and background reactions, angular and energy correlations of the charged prongs and analyze which type of cuts suppress the Drell-Yan background. An important feature of this background is that its contribution to the distribution of our observable j ∗ CP is a flat line, also at NLO QCD. By separating the Drell-Yan t events into two different sets, two different non-trivial j ∗ CP distributions are obtained. Based on this observation we propose to use these sets for calibation purposes. By Monte Carlo simulation we study also the effect of measurement uncertainties on this distribution. We estimate that the Higgs mixing angle ft can be determined with our method to a precision of D ft glyph[similarequal] 14 ◦ ( 5 ◦ ) at the high luminosity LHC (14 TeV) with an integrated luminosity of 500 fb -1 (3 ab -1 ).
PACS numbers: 11.30Er, 12.60.Fr, 14.80.Bn Keywords: Higgs bosons, tau leptons, parity, Z boson, spin correlations, CP violation
1 [email protected]
2 [email protected]
3 [email protected]
## I. INTRODUCTION
The Large Hadron Collider (LHC) has had its first major triumph with the discovery of a new, electrically neutral boson h with mass mh glyph[similarequal] 125 GeV by the ATLAS and CMS experiments [1, 2]. According to present experimental knowledge on the spin and parity of h [3, 4] and its couplings to gauge bosons and quarks and leptons [3-11], the properties of this boson agree with those of the Higgs boson predicted by the Standard Model (SM) of particle physics. In particular, the decay of h to a pair of t leptons was recently established [8, 10].
Nevertheless, much more experimental analysis is required for completely unraveling the properties of this resonance. In particular, although the LHC data strongly prefer that h is a J P = 0 + state, it is not yet excluded with high probability that h has a pseudoscalar component. How the spin and CP quantum number of a neutral Higgs-like boson can be pinned down at a hadron collider or at a future (linear) e + e -collider has been investigated in numerous papers, including [12-74].
In this paper we elaborate on a method for determining the CP nature of a Higgs-like resonance at the LHC in its decays to t leptons which has been developed in a series of papers [48, 51, 55, 66] both for Higgs production at the LHC and in e + e -collisions. Our approach is based on the distribution of a signed angle j ∗ CP between the decay planes of the charged-prong decays t -→ -a and t + → ′ a + in the a - ′ a + zero-momentum frame. We apply this method to the 125 GeV resonance h . We assume that h is a mixture of a CP-even and CP-odd state with the CP-odd admixture being smaller than the CP-even one. This assumption is in accord with the analysis of present data, cf. for instance [75-79]. We investigate also the contribution of the irreducible background Z ∗ / g ∗ → tt to the j ∗ CP distribution. While the signal contribution to this distribution shows a characteristic dependence on cos ( j ∗ CP -2 ft ) , where the angle ft describes the mixing of the scalar and pseudoscalar Higgs component which couple to t leptons (see Sec. II), we find that the background contribution is flat. We point out how the background events Z ∗ / g ∗ → tt , which are numerous at the LHC, can be used by experiments for calibrating the measurements of the distribution of the angle j ∗ CP . Moreover, we consider the two-dimensional helicity-angle and energy distributions of the charged prongs from t ∓ decay, which differ for h → tt and Z ∗ / g ∗ → tt because of the different spins of the bosons, and analyze whether appropriate cuts can enhance the signal-to-background ratio. Furthermore, we estimate the precision with which the scalar-pseudoscalar mixing angle ft might be measured at the LHC (14 TeV).
The paper is organized as follows. In the next section we briefly describe the t -decay modes and decay density matrices which are used in this analysis. In Sec. III and IV we exhibit the helicity angle distributions of the charged prongs from h → tt decay at the LHC. We recall the definition of the angle j ∗ CP and its distribution in h → tt with subsequent decays to charged prongs, which allows to determine the CP-mixing angle ft . In Sec. V we analyze the helicity and azimuthal angle distributions of the charged prongs for Drell-Yan production of t -t + at the LHC. In particular, we elaborate on the distribution of the angle j ∗ CP in
Z ∗ / g ∗ → tt → p -p + . The discussion in this section is based on the leading-order distributions. We have computed these distributions also at next-to-leading order (NLO) QCD. The results are given in Sec. VI and are compared with the corresponding distributions for inclusive Higgs production at the LHC, which we computed at NLO QCD by including differential Higgs boson-distributions obtained with the computer code MCFM [80] into our Monte Carlo program. We analyze how appropriate cuts on the polar angle distributions of the charged prongs, respectively associated cuts on their energies can reduce the irreducible background. Moreover, we study the impact of measurement uncertainties on these distributions by Monte Carlo simulation. Finally we estimate the precision with which the scalarpseudoscalar mixing angle ft might be measured at the LHC (14 TeV). We summarize in Sec. VII.
## II. HIGGS-BOSON PRODUCTION AND DECAY TO tt
Our method to determine the CP nature of the 125 GeV resonance h in its tt decays, which will be described in Sec. IV, can be applied to any h production mode, but for definiteness, we consider h production at the LHC by gluon gluon fusion,
$$p p \rightarrow h + X$$
We consider the decay mode h → -t t + with subsequent decays
$$h \rightarrow \tau ^ { - } \tau ^ { + } \rightarrow a ^ { - } a ^ { \prime + } + X \,,$$
where a ± , a ′± ∈ { e ± , m ± , p ± , a L T , , ± 1 } and X denotes neutrinos and p 0 . We take into account the main 1- and 3-charged prong t decay modes:
$$\tau \, & \to \, l + \nu _ { l } + \nu _ { \tau }, \\ - \, & \cdot \, - \cdot \,.$$
$$\tau \, & \to \, \rho + \nu _ { \tau } \to \pi + \pi ^ { 0 } + \nu _ { \tau }, \\ \tilde { \ } _ { \ } & \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \tilde { \ } _ { \ } = \tilde { \ } _ { \ } = 0 \, \cdot \, \cdot \, \cdot \, \cdot$$
$$\tau \, & \to \, \pi + \nu _ { \tau }, \\ \tilde { - } \, & \cdot \, & \cdot \, \cdot \, \cdot \, \cdot \, - \, \cdot \, - 0 \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cd.$$
$$\tau \, & \to \, a _ { 1 } + \nu _ { \tau } \to \pi + 2 \pi ^ { 0 } + \nu _ { \tau }, \\ \tilde { - } \, & \,. \, & L. T \, \cdot \,. \,. \,. \,. \, \tilde { \sim } \, - \pm \,. \, - \mp \,. \,. \,. \, & \, \tilde { \omega }$$
$$\tau \, \to \, a _ { 1 } ^ { L, T } + \nu _ { \tau } \to 2 \pi ^ { \pm } + \pi ^ { \mp } + \nu _ { \tau } \,.$$
We call the decay mode (7) also '1-prong', because the 4-momentum of a ± 1 can be obtained from the measured 4-momenta of the 3 charged pions. The longitudinal ( L ) and transverse ( T ) helicity states of the a 1 resonance can be separated by using known kinematic distributions [81-84].
The dynamics of the above t decays is, to the precision relevant for our purposes, known Standard Model physics. The interaction of a Higgs boson h of arbitrary CP nature to t leptons is described by the Yukawa Lagrangian
$$\mathcal { L } _ { Y } = - ( \sqrt { 2 } G _ { F } ) ^ { 1 / 2 } m _ { \tau } ( a _ { \tau } \bar { \tau } \tau + b _ { \tau } \bar { \tau } i \gamma _ { 5 } \tau ) \, h \,,$$
where GF denotes the Fermi constant and a t , b t are the reduced dimensionless t Yukawa coupling constants. Instead of (8), we use in the following the equivalent parameterization
$$\mathcal { L } _ { Y } = - g _ { \tau } \left ( \cos \phi _ { \tau } \bar { \tau } \tau + \sin \phi _ { \tau } \bar { \tau } i \gamma _ { 5 } \tau \right ) h \,,$$
where g t is the effective strength of the t -Yukawa interaction and ft describes the degree of mixing of the scalar and pseudoscalar Higgs component which couple to t leptons.
$$g _ { \tau } = ( \sqrt { 2 } G _ { F } ) ^ { 1 / 2 } m _ { \tau } \sqrt { a _ { \tau } ^ { 2 } + b _ { \tau } ^ { 2 } } \,, \quad \tan \phi _ { \tau } = \frac { b _ { \tau } } { a _ { \tau } } \,.$$
As to the mixing angle ft , we remark the following. It is in general not universal, but specific to the t -Yukawa interaction. The reduced Yukawa couplings af , bf to quarks and leptons f are model-dependent. As an example one may consider type-II two-Higgs doublet extensions of the Standard Model, where the SU 2 ( ) Higgs doublet F 2 is coupled to the right-chiral u -type quarks and the other doublet F 1 is coupled to right-chiral d -type quarks and charged leptons. Referring to the model described for instance in [85] one obtains in this case that af and likewise bf are identical for d -type quarks and charged leptons, while they differ in general for u -type quarks. Defining tan f t = bt / at , where at , bt are the reduced Yukawa couplings of the top quark, one gets tan ft = tan a tan b tan f t , where a denotes the mixing angle of the two CP-even neutral components of the two Higgs doublet fields and tan b = v 2 / v 1 is the ratio of the vacuum expectation values of the Higgs doublets F 2 and F 1. For notational simplicity we call ft the Higgs mixing angle.
glyph[negationslash]
For the SM Higgs boson, which is CP-even, one has g t = ( √ 2 GF ) 1 / 2 m t and ft = 0. The ATLAS and CMS results on the 125 GeV resonance h exclude that it is a pure pseudoscalar. In the following we investigate how precisely a possible pseudoscalar component of h , i.e. sin ft = 0, can be determined in its t decays at the LHC by means of the observables defined below.
The observables that we use [14, 28, 48, 51, 55, 66] are based on t -spin correlations. The charged prongs, i.e., the charged lepton l = e , m in (3), the charged pion in (4) - (6), and the a L T , 1 serve as t -spin analyzers. The normalized distributions of polarized t decays to a p ∓ , a charged lepton l = e , m , a charged r or a 1, and to a charged pion via r and a 1 decay are, in the t rest frame, of the form
$$\Gamma _ { a } ^ { - 1 } d \Gamma \left ( \tau ^ { \mp } ( \hat { \mathfrak { s } } ^ { \mp } ) \rightarrow a ^ { \mp } ( q ^ { \mp } ) + X \right ) \, = \, n ( E _ { \mp } ) \left [ 1 \pm b ( E _ { \mp } ) \, \hat { \mathfrak { s } } ^ { \mp } \cdot \hat { \mathfrak { q } } ^ { \mp } \right ] d E _ { \mp } \frac { d \Omega _ { \mp } } { 4 \pi } \,. \quad ( 1 1 )$$
Here, ˆ s ∓ denote the normalized spin vectors of the t ∓ and E ∓ and ˆ q ∓ are the energies and directions of flight of a ∓ = l ∓ , p ∓ in the respective t rest frame. The spectral functions n and b are given in [55]. The function b E ( ∓ ) encodes the t -spin analyzing power of the particle a ∓ . The t -spin analyzing power is maximal for the direct decays to pions, t ∓ → p ∓ , and for t ∓ → a L T , , ∓ 1 . (We recall that the t -spin analyzing power of a L -1 and a T -1 is + 1 and -1, respectively.) For the other decays, the t -spin analyzing power of l ∓ and p ∓ depends on the energy of these particles. It can be optimized by judiciously chosen energy cuts.
## III. DISTRIBUTIONS FOR HIGGS PRODUCTION AND DECAY TO tt AT THE LHC
The hadronic differential cross section d s for Higgs production at the LHC is given as a convolution of parton distribution functions and the partonic differential cross section d ˆ s i j for the production of h by partons i and j and subsequent h decay. For the decays (2) of h to t leptons d ˆ s i j factorizes into a product of the squared h production and decay matrix elements, as long as one neglects higher order electroweak corrections that connect the production and t -decay stage of h . The 125 GeV resonance h is narrow, G h < 4 2 . G SM h at 95% CL [86] where G SM h = 4 29 MeV (see, for instance [87]). . Therefore we can use the narrow width approximation for h . In the following we are interested in the angular correlations of the charged prongs a + and a ′-in the decays (2). The characteristic features of these correlations depend only on the CP nature of h , but not on the details of its production. Therefore we exhibit these correlations for the case of inclusive Higgs production i j → h + X → -t t + + X → - ′ a a + + X (which is dominated by gluon fusion). The structure of these correlations applies also to other processes, for instance h + jet production or h production by vector boson fusion.
We choose a right-handed coordinate frame where the t -direction of flight ˆ k in the t -t + zero-momentum frame ( tt ZMF) defines the z axis. The t ± rest frames are connected with the tt ZMF by rotation-free Lorentz boosts. In the formula (13) below, q ∓ = ∠ ( ˆ k ˆ q , ∓ ) are the polar angles of a -and a ′ + , where ˆ q ∓ are the directions of flight of a -and a ′ + in the t ∓ rest frame, respectively, and
$$\varphi = \phi _ { - } - \phi _ { + } \,, \quad 0 \leq \varphi \leq 2 \pi \,,$$
is the difference of their azimuthal angles. With (9) and (11) we obtain for the differential partonic cross section at leading order:
$$\cdot \quad & \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\quad d \hat { \sigma } _ { i j } \ = \ & \frac { g _ { \tau } ^ { 2 } m _ { h } } { 1 2 8 \pi ^ { 3 } s \Gamma _ { h } } \overline { \sum } | M ( i j \to h ) | ^ { 2 } \text{Br} _ { \tau ^ { - \to a ^ { - } } } \text{Br} _ { \tau ^ { + \to d ^ { \prime } + } } & & ( 1 3 ) \\ & \quad \times d \Omega _ { \tau } d E _ { - } d \Omega _ { - } d E _ { + } d \Omega _ { + } n ( E _ { + } ) n ( E _ { - } ) \\ & \quad \times [ 1 + b ( E _ { + } ) b ( E _ { - } ) ( \cos \theta _ { + } \cos \theta _ { - } - \sin \theta _ { + } \sin \theta _ { - } \cos ( \varphi - 2 \phi _ { \tau } ) ) ] \. \\ \quad \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\ \text{More...\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\} \,m,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,
\quad \, \text{Are...\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, \ \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\}\,m,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, \, \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, \,
\, \, \text{is the\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\text{contain\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\
\, \, \text{is the\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\ \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\\} \,m,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\\,\,\ \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, \,\,\, \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,
\, \text{Are...$$
Here, √ s is the partonic center-of-mass energy, mh is the mass of h , and ft is the Higgs mixing angle defined in (9). Moreover, we have put in (13) the t velocity bt in the h restframe equal to one.
The angular correlations in (13) reflect the t spin correlations induced in the decay of h . Integrating Eq. (13) with respect to d W t d cos q -d cos q + , the differential partonic cross section
takes the form
$$d \hat { \sigma } _ { i j } \, = \, & \frac { g _ { \tau } ^ { 2 } m _ { h } } { 8 \pi ^ { 2 } s \Gamma _ { h } } \overline { \sum } | M ( i j \to h ) | ^ { 2 } \, \text{Br} _ { \tau ^ { - } \to d ^ { - } } \text{Br} _ { \tau ^ { + } \to d ^ { \prime } + } ( 1 4 ) \\ & \times d \varphi \, d E _ { - } d E _ { + } n ( E _ { + } ) n ( E _ { - } ) \\ & \times \left [ 1 - b ( E _ { + } ) b ( E _ { - } ) \, \frac { \pi ^ { 2 } } { 1 6 } \cos ( \varphi - 2 \phi _ { \tau } ) \right ]. \\. \, \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \,.$$
Also this distribution encodes the CP nature of h . It was shown in [51, 66] that the difference j of the azimuthal angles, which is equal to the angle between the signed normal vectors of the t -→ -a and t → ′ a + decay planes, can actually be measured in the zero-momentum frame of the charged prongs a -and a ′ + . This has the big advantage that the t ∓ momenta need not be reconstructed in experiments.
## IV. OBSERVABLES
Our method to determine the CP nature of a spin-zero resonance h in its decays (2) has been described in detail in [51, 55, 66]. We recall here its salient features. It requires the measurement of the 4-momenta of the charged prongs a -, a ′ + and their impact parameter vectors n ∓ in the laboratory frame. The corresponding unit vectors are denoted by ˆ n ∓ . The 4-vectors n m ∓ =( 0 , ˆ n ∓ ) are boosted into the a - ′ a + ZMFandthe spatial parts of the resulting 4vectors n ∗ m ∓ are decomposed into their normalized components ˆ n ∗∓ || and ˆ n ∗∓ ⊥ which are parallel and perpendicular to the respective 3-momentum of a -and a ′ + . With the 'unsigned' normal vectors ˆ n ∗∓ || one determines the 'unsigned' angle j ∗ between the t -→ -a and t → ′ a + decay planes in the a - ′ a + ZMF:
$$\varphi ^ { * } = \arccos ( \hat { n } _ { \perp } ^ { * + } \cdot \hat { n } _ { \perp } ^ { * - } ) \,, \quad 0 \leq \varphi ^ { * } \leq \pi \,.$$
The simultaneous measurement of (15) and of the CP -odd and T -odd triple correlation
$$\mathcal { O } _ { C P } ^ { * } = \hat { \mathbf p } _ { - } ^ { * } \cdot ( \hat { \mathbf n } _ { \perp } ^ { * + } \times \hat { \mathbf n } _ { \perp } ^ { * - } ) \,,$$
where ˆ p ∗ -is the normalized a ′-momentum in the a - ′ a + ZMF, allows to determine a 'signed' angle (in the range 0 to 2 p ) between the t -→ -a and t → ′ a + decay planes in the a - ′ a + ZMF, which is denoted by j ∗ CP , by the following prescription:
$$\varphi _ { C P } ^ { * } = \begin{cases} \varphi ^ { * } \quad i f \ \mathcal { O } _ { C P } ^ { * } \geq 0 \,, \\ 2 \pi - \varphi ^ { * } \ i f \ \mathcal { O } _ { C P } ^ { * } < 0 \,. \end{cases}$$
The distribution of (17) is given by (14) with j → j ∗ CP . In terms of this angle, the triple correlation (16) is given by sin j ∗ CP .
The distribution of (17) allows for an unambiguous determination of the CP nature of h , that is, of the Higgs mixing angle ft . For illustration, the distribution of j ∗ CP is shown in Fig. 1
Figure 1: Normalized j ∗ CP distribution for the case where both t -and t + decay to pn . The blue dashed line is the distribution for a CP -even Higgs boson ( ft = 0) and the black long-dash dotted line corresponds to a CP mixture with ft = -p 4 . In addition, the distribution for a CP -odd Higgs boson ( ft = ± p / 2, black dotted line) is also shown. The solid red line is the distribution due to the Z ∗ / g ∗ → t + t -background, cf. Sec. V.
for the decay mode t -t + → p + p -+ 2 n for a CP-even and CP-odd Higgs boson and a CP mixture.
By fitting the function f = u cos ( j ∗ CP -2 ft ) + v to the measured differential distributions of j ∗ CP for the respective final states aa ′ , one can determine the mixing angle ft . The function f is subject to the constraint GLYPH<1> 2 p 0 d j ∗ CP f = 2 p v = s aa ′ , where s aa ′ is the h -production cross section including the respective decay branching fractions. For a certain final state aa ′ the estimate of the statistical uncertainty of j ∗ CP depends on the values of the associated parameters u and v. Because u and v describe the unnormalized distributions, it is more convenient to compare the different channels with a normalized quantity. The following asymmetry turns out to suit this purpose [66]:
$$A ^ { a a ^ { \prime } } \ = \frac { 1 } { \sigma _ { a a ^ { \prime } } } \, \int _ { 0 } ^ { 2 \pi } d \varphi _ { C P } ^ { * } \left \{ d \sigma _ { a a ^ { \prime } } ( u \cos ( \varphi _ { C P } ^ { * } - 2 \phi _ { \tau } ) > 0 ) - d \sigma _ { a a ^ { \prime } } ( u \cos ( \varphi _ { C P } ^ { * } - 2 \phi _ { \tau } ) < 0 ) \right \} \\ = \, \frac { - 4 u } { 2 \pi \nu } \,.$$
This asymmetry can also be computed from (13), (14). In the absence of cuts, one obtains
$$A ^ { a a ^ { \prime } } = \frac { \pi } { 8 } \frac { \int d E _ { a ^ { \prime } } + d E _ { a ^ { - } } n ( E _ { a ^ { \prime } } + ) n ( E _ { a ^ { - } } ) b ( E _ { a ^ { \prime } } + ) b ( E _ { a ^ { - } } ) } { \int d E _ { a ^ { \prime } } + d E _ { a ^ { - } } n ( E _ { a ^ { \prime } } + ) n ( E _ { a ^ { - } } ) } \,. \\ \intertext { 1 0 \searrow \, } \, \text{$two$ that the values of $d a^{\prime}$ are independent of the mixing and $\theta$} \, \text{but do}$$
Eqs. (18), (19) show that the values of A aa ′ are independent of the mixing angle ft but do depend on the product of the t -spin analyzing powers of a and a ′ . The larger A aa ′ the smaller the statistical error D ft in this decay channel, for a given number of events. The t -spin analyzing power, and thus A aa ′ , is maximal for the direct decays t ∓ → p ∓ and for t ∓ → a L T , ∓ 1 . The t -spin analyzing power of the charged lepton in t ∓ → ∓ l and of the charged pion
from t ∓ → r ∓ and t ∓ → ∓ a 1 can be enhanced by applying an appropriate cut on the energy of the lepton and the pion, respectively [55, 66].
The background from Drell-Yan processes to Higgs production at the LHC affects the respective distribution d s / d j ∗ CP . As will be shown below, this background contribution to the j ∗ CP distribution is flat for all charged prongs a a , ′ if integrated over the full phase space of the final states from the t ∓ decays. Of course, this contribution decreases the height of the normalized distribution and thus the magnitude of the asymmetry (18). In the next sections we investigate which cuts may be used to significantly suppress this background.
## V. DRELL-YAN PRODUCTION OF t -t +
Background reactions to the h → -t t + signal include production of QCD multijets, tt ¯, single top, W + jets, Z ∗ / g ∗ + jets, WW WZ , , and ZZ . The Drell-Yan process Z ∗ / g ∗ → -t t + is an essentially irreducible background to Higgs production by the reaction (1). Because the mass of h is relatively close to the mass of the Z boson, an appropriate cut on the tau-pair invariant mass M tt > M cut suppresses the photon contribution, but not the contributions from Z and the Z g interference term to the squared Drell-Yan matrix element.
Because our method of determining the CP nature of h uses the distribution (14) in the a - ′ a + ZMF, we need the corresponding distribution for Drell-Yan production of t pairs. The spin correlations of the t pairs produced by the intermediate vector bosons and the subsequent angular correlations between a -and a ′ + differ from the correlations (13) induced by h decay. In particular, unlike in h decay 4 , the t ∓ samples are longitudinally polarized to some degree due to the parity-violating couplings of the Z boson. In order to exhibit these features we consider Drell-Yan production of t pairs and their subsequent decays to charged prongs a -and a ′ + to lowest order in the SM couplings. The corresponding parton reaction is
$$q + \bar { q } \to \gamma ^ { * }, Z ^ { * } \to \tau ^ { - } + \tau ^ { + } \to a ^ { - } + a ^ { \prime + } + X \,.$$
For the partonic differential cross section which is analogous to (13) we obtain 5 with (11), neglecting terms of order m t / √ s :
$$d \hat { \sigma } _ { D Y } ^ { ( 0 ) } = \frac { 1 } { 5 7 6 \pi ^ { 3 } } \text{Br} _ { \tau ^ { - \to d } } \text{Br} _ { \tau ^ { + \to d ^ { + } } } d \cos \theta _ { - } d \cos \theta _ { + } d \phi _ { - } d \phi _ { + } d E _ { - } d E _ { + } F ( E _ { i }, \theta _ { i }, \phi _ { i } ) \,, \quad ( 2 1 )$$
4 If one takes into account higher order electroweak corrections in h → tt , longitudinal t ∓ polarizations are also induced in t pair production by h decay [28], which are however too small to be of relevance here.
5 We use the matrix elements given in [88], adapted to the reactions (20).
where i = ± and
$$F = & \quad n ( E _ { - } ) n ( E _ { + } ) \sum _ { B _ { 1 }, B _ { 2 } = Z, \gamma } a ( B _ { 1 }, B _ { 2 } ) \\ & \times \left \{ v _ { \tau } ^ { B _ { 1 } } v _ { \tau } ^ { B _ { 2 } } \left [ 1 - b ( E _ { + } ) b ( E _ { - } ) \left ( \cos \theta _ { + } \cos \theta _ { - } + \frac { 1 } { 2 } \sin \theta _ { + } \sin \theta _ { - } \cos ( \phi _ { + } + \phi _ { - } ) \right ) \right ] \\ & \quad + \ a _ { \tau } ^ { B _ { 1 } } a _ { \tau } ^ { B _ { 2 } } \left [ 1 - b ( E _ { + } ) b ( E _ { - } ) \left ( \cos \theta _ { + } \cos \theta _ { - } - \frac { 1 } { 2 } \sin \theta _ { + } \sin \theta _ { - } \cos ( \phi _ { + } + \phi _ { - } ) \right ) \right ] \\ & \quad + \left ( a _ { \tau } ^ { B _ { 1 } } v _ { \tau } ^ { B _ { 2 } } + a _ { \tau } ^ { B _ { 2 } } v _ { \tau } ^ { B _ { 1 } } \right ) ( b ( E _ { + } ) \cos \theta _ { + } - b ( E _ { - } ) \cos \theta _ { - } ) \right \}. \\ \vdots \dots \dots \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \ \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \, \Delta \cdots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots$$
The angles q ± , f ± are the polar and azimuthal angles of the a -and a ′ + in the tt ZMF, where the direction of the t -momentum is chosen to be the z -axis, and the momentum of the initial quark is located in the x z , -plane. Furthermore,
$$v _ { f } ^ { \gamma } = Q _ { f } e \quad ( e > 0 ) \,, \quad \quad \ a _ { f } ^ { \gamma } = 0 \,,$$
$$v _ { f } ^ { Z } = e \, \frac { T _ { 3 f } - 2 Q _ { f } s _ { \theta _ { W } } ^ { 2 } } { 2 s _ { \theta _ { W } } c _ { \theta _ { W } } } \,, \quad \ a _ { f } ^ { Z } = e \, \frac { T _ { 3 f } } { 2 s _ { \theta _ { W } } c _ { \theta _ { W } } } \,,$$
and
$$a ( B _ { 1 }, B _ { 2 } ) = s \, \frac { v _ { q } ^ { B _ { 1 } } v _ { q } ^ { B _ { 2 } } + a _ { q } ^ { B _ { 1 } } a _ { q } ^ { B _ { 2 } } } { D ( B _ { 1 } ) D ^ { * } ( B _ { 2 } ) } \,, \quad D ( B ) = s - m _ { B } ^ { 2 } + i m _ { B } \Gamma _ { B } \,. \quad \ \quad ( 2 5 )$$
Eq. (22) shows that the angular correlations, which are characteristic for the t spin correlations induced by an intermediate spin-1 boson with vector and axial vector couplings, differ from those in (13). The last term in (22) signifies the polarization of the t ± samples. Substituting f -= j + + f in (22), where j is defined in Eq. (12), and integrating (21) with respect to f + from 0 to 2 p , the terms proportional to cos ( f + + f -) in (22) vanish. That is, the resulting hadronic distribution d s ( 0 ) DY / dE + dE -d cos q + d cos q -d j is independent of j for any final state a - ′ a + . This is displayed, for Z ∗ / g ∗ → tt → pp , in Fig. 1 by the solid red line.
Wefind it instructive to investigate this feature in more detail. As a result we obtain a proposal for calibrating the distribution of j respectively j ∗ CP with Drell-Yan tt events, see below. For definiteness, we choose the charged prongs a -, a + = p -, p + in the following discussion. Eq. (22) shows that the Z ∗ / g ∗ contribution to the j distribution is flat only if (22) is integrated over the full 2 p range of f + (or alternatively of f -). The j distribution will deviate from a flat line if the phase space of one of the pions is restricted. For instance, if one demands the p -momentum to lie in the plane defined by f -= 0, the contribution of, for instance, the pure photon exchange to the j distribution (which follows from the first line in the curly bracket of Eq. (22)) is proportional to 1 -c cos j . This distribution and, therefore, the distribution of j ∗ CP differs from a flat line.
How can this be probed experimentally? We define a variable cos ˜ a -by
$$\cos \tilde { \alpha } _ { - } \, = \, \left | \frac { \hat { \mathbf e } _ { \mathbf z } \times \hat { \mathbf k } _ { L - } } { \left | \hat { \mathbf e } _ { \mathbf z } \times \hat { \mathbf k } _ { L - } \right | } \cdot \frac { \hat { \mathbf p } _ { L - } \times \hat { \mathbf k } _ { L - } } { \left | \hat { \mathbf p } _ { L - } \times \hat { \mathbf k } _ { L - } \right | } \right |$$
-
τ
Figure 2: Drell-Yan production of t -t + and subsequent decays to pions via photon exchange, pp → g ∗ → -t t + → p + p -+ 2 n . Left: The un-normalized distribution of j ∗ CP for events with p -'nearly coplanar' ( a -< p / 4, red solid line) and events with p -'nearly perpendicular' ( a -> p / 4, dashed blue line) to the q t production plane. The dotted black line is half of the sum of the two distributions. The cuts M tt ≥ 80 GeV and | hp ± | ≤ 1 were used. Right: Spin configuration for qq ¯ → ∗ → g t -t + events where the t -t + are produced orthogonal to the beam direction.
which allows to classify the Z ∗ / g ∗ → t -t + → p -p + events into events where the p -is 'nearly coplanar' and 'nearly perpendicular' to the q t production plane in the laboratory frame. In (26) the unit vectors ˆ k L -and ˆ p L -are the t -and p -directions of flight in the laboratory frame and ˆ e z points along the direction of one of the proton beams. The range of ˜ a -is 0 ≤ ˜ a - ≤ p / 2. Events with p -being 'nearly coplanar' ('nearly perpendicular') are defined by demanding ˜ a -< p / 4 ( ˜ a -> p / 4). In order to define a discriminating variable in terms of measurable quantities, we use the impact parameter vector ˆ n -(cf. Sec. IV) instead of ˆ k L -and replace (26) by
This variable is nearly identical to (26), i.e., events with p -being 'nearly coplanar' ('nearly perpendicular') are in the following defined by requiring a -< p / 4 ( a -> p / 4).
$$\cos \alpha _ { - } \, = \, \left | \frac { \hat { \mathbf e } _ { z } \times \hat { \mathbf p } _ { L - } } { | \hat { \mathbf e } _ { z } \times \hat { \mathbf p } _ { L - } | } \cdot \frac { \hat { \mathbf n } _ { - } \times \hat { \mathbf p } _ { L - } } { | \hat { \mathbf n } _ { - } \times \hat { \mathbf p } _ { L - } | } \right | \,. \quad \\ \intertext { n a r l v i d a t i c a l t o } \omega \, \left \{ \frac { 1 } { 2 } \, \text{as a sunt} \, \text{with} \, \pi ^ { - } \, \text{hina} \, \lq n a r l v \, \text{container} ^ { 2 } \, \lq n a r l v \,.$$
Let us first consider Drell-Yan production of t -t + and subsequent decays to pions via photon exchange, pp → ∗ → g t -t + → p + p -+ 2 n . The j ∗ CP distributions (computed with the first line in the curly bracket of (22)) are shown in Fig. 2, left plot, for events with p -'nearly coplanar' ( a -< p / 4, red solid line) and events with p -'nearly perpendicular' ( a -> p / 4, dashed blue line) to the q t production plane.
The distribution for events with p -'nearly coplanar' is enhanced for j ∗ CP ∼ p which corresponds to p -and p + being (nearly) antiparallel. In order to understand this let us consider qq ¯ → ∗ → g t -t + where, for illustration, the t pair is emitted perpendicular to the incoming quark direction. The t pair is produced in a s-wave and the resulting spin configuration for this type of events is shown in the right plot of Fig. 2, i.e., the t -and t + spin projections
onto the quark axis are parallel. The t ± decay distributions (11) tell us that the p -( p + ) are then preferentially emitted in (opposite to) the direction of the t -( t + ) spin, which means that the j ∗ CP distribution is enhanced for f - -f + ∼ p .
On the other hand if the p -is emitted 'nearly perpendicular' to the q t production plane ( a -> p / 4), the j ∗ CP distribution is enhanced at j ∗ CP ∼ 0 and 2 p , cf. the left plot of Fig. 2. Again this can be understood from the right plot of of Fig. 2 and the t ± decay distributions (11). The projection of the spin of g ∗ and thus the projection of the total t -t + spin onto the axis orthogonal to the q t production plane is zero, i.e., the t -and t + spins are predominantly anticorrelated with respect to this axis ('up-down' and 'down-up'). Therefore, the momenta of the p -and p + are preferentially parallel in this case. The j ∗ CP distributions for the two sets of events ( a -< p / 4 and a -> p / 4) add up exactly to a flat line as stated above and already shown in Fig. 1.
Let us now consider (20) with the intermediate Z boson. In view of the analysis of Sec. VI below, we apply a cut on the t -pair invariant mass M tt ≥ M cut ( M cut glyph[greaterorsimilar] 80 GeV). Then (20) is dominated by Z -boson exchange. The strengths of the vector and axial vector couplings of the t leptons imply that the differential cross section (21) is dominated by the second line of the curly bracket in (22), i.e., by t -pair production through the axial vector current, which corresponds to p-wave production of t -t + . The resulting t spin correlations differ from those induced by g ∗ exchange discussed above. In the case of axial vector production the t -and t + spin projections onto the quark axis are predominantly anticorrelated (excluding the forward and backward regions), while the t -and t + spin projections onto the axis orthogonal to the q t production plane are predominantly correlated. Therefore, in the case of axial vector production, the j ∗ CP distributions for events with p -emitted 'nearly coplanar' and 'nearly perpendicular', respectively, are opposite to the corresponding distributions for g ∗ exchange shown in Fig. 2.
The left plot of Fig. 3 shows the corresponding un-normalized j ∗ CP distributions computed with the complete tree-level differential cross section (21), (22). The cuts M tt ≥ 80 GeV and | hp ± | ≤ 1 on the pseudo-rapidity of the pions were applied. The shape of these distributions reflect the outcome of the discussion made in the previous paragraph 6 . Again, the two distributions add up to a flat line.
Contrary to the case of an intermediate Higgs boson, the tree-level distributions (22) will be affected by higher-order QCD corrections to (20) because of the correlations of the t spins with the initial-state parton momenta. We have computed the respective differential distributions for
$$\ p p \to Z ^ { * }, \gamma ^ { * } + X \to \tau ^ { - } + \tau ^ { + } + X \to a ^ { - } + a ^ { \prime + } + X$$
6 As to the signal reaction pp → → -h t t + → p -p + 2 n we remark the following. For a Higgs boson of any CP nature, the un-normalized j ∗ CP distributions have the same shape for events with p -being nearly coplanar and nearly perpendicular.
Figure 3: Drell-Yan production of t -t + and subsequent decays to pions via Z ∗ / g ∗ exchange, pp → Z ∗ / g ∗ → t -t + → p + p -+ 2 n . The un-normalized distribution of j ∗ CP for events with p -'nearly coplanar' ( a -< p / 4, red solid line) and events with p -'nearly perpendicular' ( a -> p / 4, dashed blue line) to the q t production plane. The dotted black line is half of the sum of the two distributions. The cuts M tt ≥ 80 GeV and | hp ± | ≤ 1 were used. Left plot, LO QCD. Right plot, NLO QCD.
at next-to-leading order (NLO) in the QCD coupling a s , taking the t correlations in the virtual and real corrections into account. We calculated the respective t spin density matrices for qq ¯ → ∗ Z , g ∗ ( g ) → -t t + ( g ) and gq q ( ¯ ) → ∗ Z , g ∗ q q ( ¯ ) → -t t + q q ( ¯ ) at order a s . The soft and collinear divergences are treated with Catani-Seymour dipole subtraction [89] with collinear factorization in the MS scheme.
As to the NLO QCD j ∗ CP distributions for Z ∗ / g ∗ → t -t + → p -p + + X for events with a -< p / 4 and a -> p / 4: they are displayed in the right plot Fig. 3. The comparison with the LO distributions shows that the order a s QCD corrections amount to about 12% and the shapes of these NLO distributions remain essentially the same as the LO distributions.
The results shown in Fig. 3 suggest that one may use the Drell-Yan events pp → Z ∗ / g ∗ → tt → ppnn ¯ , which are quite abundant at the LHC, for experimentally calibrating and validating the j ∗ CP distribution(s) before this observable is used to determine the CP nature of the Higgs boson h .
The NLO QCD polar angle and energy distributions of the charged prongs aa ′ , which can be used for background suppression, will be discussed in the next section.
## VI. NUMERICAL RESULTS FOR THE LHC
In this section we consider Higgs production (1) and decay into t pairs (2) at the LHC for a collider center-of-mass energy √ S = 14 TeV. We analyze how the differences between the Higgs-boson induced and Z ∗ / g ∗ induced t spin correlations, respectively the differences between the resulting aa ′ angular and energy correlations can be used to reduce the Z ∗ / g ∗ background. At the end of this section we estimate the precision D ft with which the Higgs mixing angle may be determined in the h → tt decay mode at the LHC.
cos q-
Figure 4: LHC, √ S = 14 TeV, M tt ≥ 100 GeV. Production of h + X (left) of arbitrary CP nature and of Z ∗ / g ∗ + X (right) with subsequent decay of the respective boson to t -t + → p + p -+ 2 n . The plots show the distributions s -1 d s / d cos q -d cos q + without cuts on the pion momenta.
As long as no cuts on the final state particles/jets are applied, the normalized distributions (13) for the signal reaction (1) hold also when higher order QCD corrections are taken into account. In order to obtain NLO QCD distributions with cuts for (1), we compute the twodimensional distributions of the Higgs boson transverse momentum and rapidity by means of the computer code MCFM [80, 90, 91]. We include h decay into t pairs with t spin correlations and their subsequent decays into charged prongs by Monte Carlo simulation. We put mh = 125 GeV and assume the gg → h amplitude to be the same as in the SM, and we take G h = G SM h = 4 29 MeV. For estimating signal-to-background ratios we take the K . factor for inclusive Higgs production, K NNLO NLO / = 1 35 into account, calculated with the . computer code HNNLO [92-94]. We calculated the angular distributions analogous to (21) and corresponding energy distributions for the background reactions (28) also to NLO QCD (cf. above). For the numerical evaluation we used the parton distribution functions CT10 [95] with a s ( mZ ) = 0 1180 and two-loop running in . a s . Moreover, we used a ( MZ ) = 1 128 89 / . and the weak mixing angle sin 2 q W = 0 2228. As a default value for the renormalization and . factorization scale m we have set m = m R = m F = mh .
The ATLAS and CMS experiments, which reported evidence for h → tt [8, 10], have reconstructed the tt invariant mass with methods described in [96] and [10], respectively. The Higgs-boson signal appears as an enhancement of d s / dM tt in a mass window around M tt = mh over the background which is mostly due to Z ∗ / g ∗ → tt . If not stated otherwise, we apply in the following sections a cut M tt > 100 GeV which strongly suppresses the background from Z ∗ / g ∗ → tt .
## A. t + t -spin correlations and subsequent polar angle and energy correlations
We analyze the distributions d s / d cos q -d cos q + and associated energy distributions for the signal and background reactions, where the helicity angles q ∓ of a -and a ′ + are defined as in Sec. III. In the left and right plot of Fig. 4 the normalized distributions s -1 d s / d cos q -d cos q + at NLO QCD are shown for the signal and background reaction for the t -t + → p -p + + 2 n decay mode. No cuts on the pion transverse momentum p T p or the pion rapidity are applied. Solid grey contour lines denote constant values. The normalized signal distribution is given by ( 1 + cos q -cos q +) / 4, cf. Eq. (13), and becomes maximal for cos q -cos q + → 1. The background distribution contains terms proportional to 1 -cos q -cos q + due to the Z ∗ / g ∗ induced t spin correlations and, in addition, terms linear in cos q ∓ which signify a longitudinal t ∓ polarization generated by the parity-violating Z tt coupling. The background can be reduced by applying a cut on the product of cosines, cos q -cos q + > -cmin ( cmin > 0 , where ) cmin should be different for cos q - → ± 1, in view of the unsymmetric background distribution. These cuts should not be too hard because that part of the signal cross section which is sensitive to the Higgs CP mixing angle ft is proportional to sin q -sin q + , cf. Eq. (13).
Both distributions in Fig. 4 are affected if a pT cut on the pion momenta is applied. A p T p cut mostly removes events in the vicinity of cos q -= -1 and cos q + = + 1, because in these phase-space regions the pion is emitted opposite to the corresponding t direction of flight and, therefore, its energy in the tt ZMF is small.
The uncertainties due to scale variations of these NLO distributions are as follows. The normalized signal distribution in Fig. 4, left, is identical to the normalized LO distribution because production and decay of the Higgs boson factorizes. In order to estimate the scale uncertainty of the normalized Z ∗ / g ∗ distribution in Fig. 4, right, we vary the scale m between m = mh / 2 and m = 2 mh and calculate the deviation [ s -1 d s m ( = mh ) -s -1 d s m ( )] / s -1 d s m ( = mh ) for each value of cos q -and cos q + . Apart from a small region in the lower left and upper right corner of Fig. 4, right, the maximal deviation of each point from the respective value of 1 / s · d s m ( = mh ) is small, about ± 3%.
Cuts on cos q ∓ may be unrealistic because, at the LHC, the reconstruction of the t rest frames is complicated, even for hadronic t decays. This is because the partonic center-of-mass energy is unknown for a certain event and the decay of each tt pair involves at least two neutrinos. However, one can approximately reconstruct the tt ZMF with a fitting procedure [96] and determine the pion energies in this frame, denoted by ˜ E ∓ in the following. The energies ˜ E ∓ are related to cos q ∓ by boosts. In Fig. 5 the NLO QCD distributions s -1 d s / dE ˜ -dE ˜ + are displayed for the signal and background reaction. Solid grey contour lines denote constant values. The normalized signal distribution in Fig. 5, left, shows that h → tt events decay preferably into one pion with a large energy and one pion with a small energy in the tt ZMF. On the other hand, the right plot of Fig. 5 shows that in the case of the Z ∗ / g ∗ → tt background, events where both pion energies are small are strongly enhanced.
GLYPH<142>
GLYPH<142>
Figure 5: LHC, √ S = 14 TeV, M tt ≥ 100 GeV. Production of h + X (left) of arbitrary CP nature and of Z ∗ / g ∗ + X (right) with subsequent decay of the respective boson to t -t + → p + p -+ 2 n . The plots show the distributions s -1 d s / dE ˜ -dE ˜ + without cuts on the pion momenta.
If one applies a cut on the transverse momenta of the p ∓ , for instance, p T p ≥ 20 GeV, the number of events with small ˜ E ∓ are reduced. At LO QCD this cut removes all events with ˜ E ∓ < 20 GeV because the transverse momentum of the h Z , ∗ , g ∗ boson is zero. At NLO QCD this is lifted to some extent because the finite transverse momentum of the respective boson results in p T p glyph[greaterorsimilar] 20 GeV even if ˜ E ∓ < 20 GeV in the tt ZMF.
The normalized distributions of Fig. 5 suggest the application of cuts on the energies ˜ E ∓ in order to enhance the signal-to-background ratio. This ratio is enhanced by rejecting events where both ˜ E + and ˜ E -are smaller than, for instance, 20 GeV, or where both energies are larger than 45 GeV. However, one should not reject regions including ˜ E + ∼ ˜ E + ∼ 30 GeV which corresponds to sin q -sin q + ∼ 1. Here the sensitivity to j ∗ CP and therefore to the mixing angle ft is largest (cf. Eq. (13)).
## B. The distribution of j ∗ CP for h → tt
## 1. Direct t + t -→ p + + p -+ 2 n decay
The normalized j ∗ CP distribution without cuts is shown in Fig. 1 for h → t + t -→ p + p -+ 2 n . If no cuts on the final-state particles are applied, the distribution is the same also for other Higgs production modes, or if higher order QCD corrections are included. The distribution will change if kinematical cuts like p T p cuts are applied, because the j distribution results from the term proportional to sin q + sin q -in Eq. (13). Enhancing the region cos q + cos q - ∼ 0 of Fig. 4, left, increases the asymmetry defined in Eq. (18). The dependence of the normalized j ∗ CP distribution at NLO QCD on cuts on p T p and on ˜ E ∓ is displayed in Fig. 6, left, for a Higgs mixing angle ft = -p / 4. The solid black line shows the distri-
Figure 6: Left: pp → → -h t t + → p + p -+ 2 n , where h is assumed to be a CP mixture with mixing angle ft = -p / 4. Dependence of the distribution s -1 d s / d j ∗ CP on different kinematical cuts. Right: h → t + t -→ r + r -+ 2 n for a CP-even Higgs boson. Dependence of the distribution s -1 d s / d j ∗ CP on cuts on the energies ˜ E ∓ of the charged pions. The horizontal lines in both plots are drawn to guide the eye.
bution without cuts. The corresponding asymmetry is A pp = 39 3%. If a cut . p T p ≥ 20 GeV is applied the asymmetry, associated with the distribution shown by the dashed black line, increases to A NLO pp = 49 5% ( . A LO pp = 50 2%). . For the Higgs-boson production mode (1) the Higgs-boson transverse momentum is, on average, small. Therefore the p T p cut removes events with E tt ± glyph[lessorsimilar] 20 GeV in the distribution displayed in Fig. 5, left. For these events the value of sin q ± is small for at least one of the pions. Therefore the value of the product sin q + sin q -is on average rather large for the remaining events. This is why the asymmetry A is increased by this cut. If, in addition, events with large transverse pion momenta are rejected by selecting, for instance, events with 40GeV ≥ p T p ≥ 20 GeV, the asymmetry is further enhanced to 56% ( A LO pp = 59 5%). The corresponding . j ∗ CP distribution is given by the dotted black line in Fig. 6. Cuts on the pion energies will also change this distribution and the resulting asymmetry. The j ∗ CP distribution shown by dashed red line in Fig. 6, left, results from applying the cuts p T p ≥ 20 GeV and ˜ E ± ≤ 40 GeV. The corresponding asymmetry is 61 2% ( . A LO pp = 61 8%). .
Notice that all these cuts do not change the location of the maximum of the j ∗ CP distribution. Furthermore, cuts on the pseudo-rapidity of the pions hp do not change the normalized j ∗ CP distributions displayed in Fig. 6.
## 2. Other t decay modes
At the end of Sec. VI A, cuts on the energies ˜ E ∓ of the charged pions were suggested for the direct t → p decays, in order to enhance the signal-to-background ratio and the asymmetry (18). If the Higgs mixing angle ft is to be determined from other t decay modes with the method described in Sec. IV, e.g. from t + t -→ r + r -+ 2 n and subsequent r ± → p ± + p 0
decay, these cuts on ˜ E ∓ can, however, not be used for background suppression in this case. This is because the t -spin analyzing power of the charged pion from r decay is energydependent, cf. for instance Fig. 4 in [55], where this analyzing power is shown as a function of the pion energy in the t rest frame. This energy is related by a boost to the p ± energy ˜ E ± in the tt ZMF. Dividing the ˜ E -, ˜ E + phase space into four regions, two with ˜ E ± ≤ 30 GeV, ˜ E ± ≥ 30 GeV and two, where one energy is smaller than and the other one larger than 30 GeV, we show in Fig. 6, right, the resulting normalized distributions of the angle j ∗ CP . The distribution given by the red solid line, which has the largest asymmetry (18), results from events where both energies ˜ E ∓ ≥ 30 GeV. For events with ˜ E ∓ ≤ 30 GeV the distribution is almost flat (dashed black line). For events with ˜ E - ≥ 30 GeV and ˜ E + ≤ 30 GeV (dot-dashed blue line) the resulting asymmetry is also quite small. Furthermore, the j ∗ CP distribution is shifted in this case by an angle p with respect to the solid red line. This is because for t → r → p decay, the function b E ( ) in (11), which encodes the t -spin analyzing power of the charged pion for this decay mode, is negative for ˜ E ≤ 30 GeV. The asymmetry (18) is largest for events with ˜ E ∓ ≥ 30 GeV because if ˜ E ± are large, the pion energies in the respective t rest frames are also large on average. In this energy range the t -spin analyzing power of the charged pion from r decay is large (and positive).
## 3. Impact of measurement uncertainties
The normalized j ∗ CP distributions are affected by measurement uncertainties, in particular by the uncertainties associated with the measurements of the directions ˆ n ∓ of the impact parameters of the charged prongs a -, a ′ + (cf. Sec. IV). In order to assess the effect of these uncertainties on the distributions j ∗ CP for the various tt decay modes with Monte Carlo methods, we have 'smeared' the relevant quantities with a Gaussian distribution function GLYPH<181> exp ( -( X / s ) 2 / 2 . ) Here X denotes the generated quantity (coordinate in position space, momentum component, energy) and s its expected standard deviation.
The primary vertex (PV), i.e., the Higgs-boson production/decay vertex is varied along and transverse to the beam axis with s PV z = 20 m m and s PV tr = 10 m m , respectively. In the following, we discuss the effect of smearing in some detail for the t -t + → p -p + decay mode. The intersection point of the impact parameter vector n ∓ with the respective track of the charged pion p ∓ is varied by s p tr = 10 m m within a circle transverse to the direction of the pion momentum. Furthermore we assume the angular resolution of the charged p ∓ track at its intersection point with n ∓ to be distributed with s p q = 1 mrad, and the resolution of the p ± energy is taken to be D E p / E p = 5%. With these values, suggested in [97, 98], we arrive at the conclusion that a rather precise measurement of the Higgs mixing angle ft is possible at the LHC, see below.
First, we determine the average length 〈| n ∓|〉 of the impact parameter in the laboratory frame. We use an exponential decay law for the t leptons from h -decay with an average t decay length of 〈 c tt 〉 = 87 m m . If one assumes that the pion is emitted, in the t rest fame, trans-
Figure 7: LHC ( √ S = 14 TeV), pp → h → t -t + → p + p -+ 2 n with cuts p T p ≥ 20 GeV, | hp | ≤ 2 5. . The normalized j ∗ CP distribution, taking measurement uncertainties into account, for different minimum cuts on the length of the impact parameters n ± . The left and right plots refer to a CP-odd and CP-even Higgs boson, respectively. The horizontal lines in both plots are drawn to guide the eye.
versely to the t direction of flight, then we obtain 〈| n ∓|〉 = 44 m m . This estimate indicates also the magnitude of the resolution which must be experimentally achieved both for the primary vertex and the tracks of the pions.
Taking the smearing of the various quantities into account, with standard deviations as specified above, the resulting effects on the j ∗ CP distribution are shown in Fig. 7, left, for the tt → pp decay of a CP-odd Higgs boson and Fig. 7, right, for a CP-even Higgs boson. The black solid lines show the distributions without any smearing. The dotted blue lines include the effect of smearing using the parameters given above. The asymmetry (18) is then strongly reduced from A NLO pp = 49 5% to 18% ( . A LO pp = 18 2%) in the case of a CP-even Higgs boson and . to 37% ( A LO pp = 37 4%) for a CP-odd Higgs boson. . These asymmetries can be enhanced by taking into account only events with impact parameter lengths n ± above a certain minimum value. For the cuts n ± ≥ 20 m m n , ± ≥ 30 m m , and n ± ≥ 40 m m , the resulting j ∗ CP distributions are displayed in Fig. 7. The associated asymmetry A NLO pp is 25 7% . , 32 3% 38 6% in the case . , . of a CP-even Higgs boson and 45 2% . , 49 8% 53 3% for a CP-odd Higgs boson. Our Monte . , . Carlo simulations indicate that the value n min of the minimum cut should be of the same size as the largest value from the set { s PV z , s PV tr , s p tr } . Of course, the number of events is reduced by a cut on n ± . It is important to notice that for a CP-even or a CP-odd Higgs boson, the position of the maximum of the j ∗ CP distribution, whose true value is at j ∗ CP = 0, respectively at j ∗ CP = p , is neither affected by the smearing procedure nor by a cut on the impact parameters.
Next we apply the same smearing procedure and cuts also to a Higgs boson h being a CP mixture with ft = -p / 4. The resulting j ∗ CP distributions are shown in the left plot of Fig. 8. The asymmetry (18), whose NLO QCD value for this decay mode is A NLO pp = 49 5% . ( A LO pp = 50 2%) is reduced to 29% ( . A LO pp = 29 3%) by the measurement uncertainties as speci-. fied above. By applying a minimum cut on both impact parameters n ± the asymmetry can be
Figure 8: LHC ( √ S = 14 TeV), pp → h + X , for a CP mixture h with mixing angle ft = -p / 4. Left: h → -t t + → p + p -with p T p ≥ 20 GeV, | hp | ≤ 2 5. The normalized . j ∗ CP distribution, taking measurement uncertainties into account, for different minimum cuts on the length of the impact parameters n ± . Right: pp → h → t -t + → r -r + → p -p + . The normalized j ∗ CP distribution for events with ˜ E ± ≥ 30 GeV in the tt ZMF, without and with smearing. The horizontal lines in both plots are drawn to guide the eye.
enhanced to 36 7% . , 42% 46 6% for , . n ± ≥ 20 m m , 30 m m , 40 m m . More importantly, however, the position of the maximum of the j ∗ CP distribution turns out to depend on the smearing parameters and on the cut on n ± . In the case of no smearing the position of the maximum is at j ∗ CPmax , = p / 2 = 1 57 . . With smearing and no cut on n ± the location of the maximum moves to j ∗ CPmax , = 1 24 . . For n ± ≥ 20 m m , 30 m m , 40 m m the maximum is at j ∗ CPmax , = 1 3 . , 1 36 . , 1 4. . The reason for this shift of the maximum is the smearing of the primary vertex. For larger values of s PV z or s PV tr the reconstructed PV moves further away from the two tracks of p ± . Therefore, the angle between the two impact parameters becomes smaller. This leads to an enhancement of the j ∗ CP distribution near j ∗ CP ∼ 0 and j ∗ CP ∼ 2 p . In the case of smearing, p T p cuts affect also the position of the maximum of the j ∗ CP distribution. Because the Higgs mixing angle ft is determined from the difference between the position j ∗ CP , max of the maximum of the measured distribution and j ∗ CP = p (cf. Fig. 1) it is crucial to understand the measurement uncertainties. As will be shown in the next section, the measurement of the j ∗ CP distribution for Drell-Yan t -pair production can be used to get a handle on these uncertainties.
This shift of the maximum of the j ∗ CP distribution, which occurs for a CP mixture, can become even larger for decay modes such as rr or r a 1 if additional cuts on ˜ E ± are applied. As an example, we consider the h → -t t + → r -r + → p -p + decay channel for a CP mixture h with ft = -p / 4. We apply the cuts ˜ E ± ≥ 30 GeV in order to obtain a large asymmetry (18), cf. Sec. VI B 2. The solid black curve in Fig. 8, right, shows the associated normalized j ∗ CP distribution without smearing. Taking measurement uncertainties into account one obtains the distribution given by the blue dashed line. Its maximum is shifted from j ∗ CPmax , = p / 2 to j ∗ CPmax , glyph[similarequal] 1. Notice that the smeared distribution is raised at j ∗ CP ∼ 0 and j ∗ CP ∼ 2 p and
Figure 9: Left: pp → Z ∗ / g ∗ → t -t + → p + p -with p T p ≥ 20 GeV and n ± ≥ 20 m m . Normalized smeared j ∗ CP distribution for two different values of s PV z . The dotted black line is the prediction without measurement uncertainties. Right: decays h Z , ∗ / g ∗ → t -t + → r -r + → p -p + , taking h to be a CP mixture with ft = -p / 4. Normalized (un)smeared j ∗ CP distributions for events with ˜ E ± ≥ 20 GeV.
lowered at j ∗ CP ∼ p as compared to the unsmeared distribution. This is due to the smearing of the PV. The PV uncertainty causes the same effect on the j ∗ CP distributions of the other rr → pp event categories, e.g. for events with ˜ E - ≥ 30 GeV and ˜ E + ≤ 30 GeV, and of other tt → aa ′ decay modes. At this point we notice that an increase of the uncertainties of the other parameters discussed above makes the j ∗ CP distributions flatter.
## C. The j ∗ CP distribution for Drell-Yan production of t pairs
In this section we investigate the impact of measurement uncertainties on the j ∗ CP distribution for the Drell-Yan production of tt . If not stated otherwise we use, as above, for the smeared distributions the parameters s PV z = 20 m m , s PV tr = 10 m m , s p tr = 10 m m , s p q = 1 mrad and D E p / E p = 5%. Furthermore, we apply the cut n ± ≥ 20 m m in the computation of the distributions of this section. For definiteness, we consider the normalized j ∗ CP distribution for the pp → Z ∗ / g ∗ → -t t + → p + p -+ 2 n production mode. Without smearing the distribution is given by the dotted black flat line in Fig. 9, left, while the effect of the PV uncertainty, simulated with s PV z = 20 m m and with s PV z = 40 m m , results in the distribution shown by the dashed and solid blue line, respectively. The shape of these curves can be understood as follows. If the measurement uncertainty of the PV becomes larger, the distance of the reconstructed PV to each of the tracks of the charged pions p ± increases. This results in a smaller angle between the two reconstructed impact parameter vectors. This, in turn, enhances the region of j ∗ CP ∼ 0 and j ∗ CP ∼ 2 p in the j ∗ CP distribution. On the other hand we found that larger values of s p tr , s p q and D E p decrease the curvature of the smeared distribution.
An important result of our simulation of the smeared normalized j ∗ CP distributions for h → tt and Z ∗ / g ∗ → tt is that they are both enhanced (for h of arbitrary CP nature) at j ∗ CP ∼ 0 and
Figure 10: pp → Z ∗ / g ∗ → -t t + → p + p -+ 2 n . (Un)smeared j ∗ CP distributions for events with p -being nearly coplanar ( a -< p / 4) and events with p -being nearly perpendicular ( a -> p / 4).
j ∗ CP ∼ 2 p as compared to the respective unsmeared distribution - an effect which is due to the finite experimental resolution of the primary vertex. Based on this result we suggest the following procedure to obtain a j ∗ CP distribution for the signal reactions h → tt → aa ′ , with which these distortions can be eliminated to a large extent. We assume that a clean data sample of Drell-Yan t pair events can be recorded at the LHC. One measures the normalized j ∗ CP distribution for this sample and subtracts it from the distribution measured with the tt events in the signal region, M tt ∼ 125 GeV.
We exemplify this proposal for h Z , ∗ / g ∗ → -t t + → r -r + → p -p + , taking h to be a CP mixture with ft = -p / 4. Fig. 9, right, shows the unsmeared and smeared j ∗ CP distributions for h decay and the smeared distribution for Z ∗ / g ∗ decay. Subtracting the latter distribution from the smeared distribution for h decay and adding the flat line s -1 d s / d j ∗ CP = 1 / ( 2 p ) , one obtains the solid blue curve. The maximum of this corrected signal distribution is at j ∗ , max CP = p / 2 where it should be. This illustrates that with this procedure, one gets rid of the distortions caused by the PV measurement uncertainties to a large extent.
Finally, we compute for pp → Z ∗ / g ∗ → -t t + → p + p -+ 2 n the smeared j ∗ CP distribution for events with p -being nearly coplanar ( a -< p / 4) and events with p -being nearly perpendicular ( a -> p / 4), cf. Sec. V. The distribution for a -> p / 4 gets significantly distorted by the smearing, as shown in Fig. 10. As already discussed in Sec. V we propose to measure these two distributions as a means to calibrate the signal distribution.
## D. Estimate of D ft
In this section we estimate the statistical uncertainty D ft with which the mixing angle ft can be determined from the smeared j ∗ CP distributions of the h → tt decay modes, taking into account the t decay channels (3) - (7). As discussed above, the asymmetry A aa ′ in Eq. (18) is a measure of the statistical uncertainty D ft for each decay channel aa ′ , for a given number
Table I: Asymmetries and signal reduction for the hadron-hadron, lepton-hadron, and lepton-lepton decay modes. The estimate of "events/fb" includes background events.
| tt decay channel | A S [ % ] | S S + B | A S + B [ % ] | s n ± ≥ 20 m m / s | events / f b |
|--------------------|-------------|-----------|-----------------|----------------------|----------------|
| had-had | 13 . 2 | 0 . 5 | 6 . 6 | 0 . 58 | 1 . 16 |
| lep-had | 9 . 0 | 0 . 5 | 4 . 5 | 0 . 63 | 1 . 26 |
| lep-lep | 7 . 0 | 1 / 3 | 2 . 3 | 0 . 61 | 1.22 |
| combined | | | 4 . 85 | 0 . 61 | 3 . 66 |
## of events 7 .
For each t -decay mode the asymmetry (18) is calculated by generating the smeared j ∗ CP distribution of the Higgs-boson signal and of the Z ∗ / g ∗ background with our Monte Carlo program, using the smearing parameters s PV z = 20 m m , s PV tr = 10 m m , s p tr = 10 m m , s p q = 1 mrad, D E p / E p = 5%, the cut n ± ≥ 20 m m on the length of the impact parameters and the cut M tt ≥ 100 GeV. Furthermore, for the leptonic decay modes we apply the charged-lepton cuts pT l , ≥ 20 GeV and | h l | ≤ 2 5. For the hadronic decay modes the cuts . pT , t ≥ 20 GeV and | ht | ≤ 2 5 are used, which approximate roughly corresponding cuts on the hadronic . t jets used in experiments. As discussed in the last section, we correct the normalized Higgs-boson j ∗ CP distribution, for each decay channel, by subtracting the normalized j ∗ , Z g CP distribution and adding the flat distribution 1 / ( 2 p ) . The resulting distribution is then reweighted in order that it is properly normalized. From this distribution we calculate the signal asymmetry AS . It is given for the final-state event categories 'hadron-hadron' (had-had), 'lepton-hadron' (lephad), and 'lepton-lepton' (lep-lep) in column 2 of Table I. The asymmetry for signal plus background is then obtained by AS + B = AS × S / ( S + ) B .
With these values of the asymmetry AS + B and number of events/fb for the different event categories introduced above, we estimate the statistical uncertainty D ft in the following way [38].
In order to estimate the number of events including the background that may be available for j ∗ CP measurements at the LHC (14 TeV), we assume for the hadron-hadron and lepton-hadron decay channels the ratio ratio S / B = 1 and 8 S + B = 2events / f b [8]. For the lepton-lepton decay modes we assume S / B = 1 / 2 and S + B = 2events / f b . The resulting ratios S / ( S + ) B and asymmetries AS + B are given in column 3 and 4 of Table I. Next we calculate the factor Rn = s n ±≥ 20 m m / s by which the respective signal cross section is reduced by a cut on the impact parameters. These factors are given in column 5 of Table I. The number of events/fb is then given by ( S + ) B × Rn , cf. column 6 of Table I.
7 The analysis of the various decays h → tt → charged-prongs made here is analogous to our investigation in [66], where Higgs-boson production in e + e -collisions by e + e -→ Zh was considered. In the present analysis, the cuts on the energies of the pions and charged leptons are made in the Higgs-boson rest frame.
8 We have extracted these numbers from the results of the ATLAS experiment for the hadron-hadron and lepton-hadron channels in h → tt at the LHC (8 TeV) [8]. The actual number of events at the LHC (14 TeV) will be higher - however, the S / B ratio will be worse.
Figure 11: LHC ( √ S = 14 TeV). Estimated statistical uncertainty D ft of the Higgs mixing angle ft as a function of the number of events (signal + background). The horizontal lines display the asymmetries AS + B of the three event categories given in Table I. The yellow dot on the solid black line is not shown, it corresponds to 11000 events.
We choose some value of the Higgs mixing angle, for example ft = -p / 4, and generate the corresponding differential j ∗ CP distribution using 20 bins between 0 and 2 p . We then fit this distribution with the function u cos ( j ∗ CP -2 ft )+ v . This is repeated a sufficiently large number of times ( ∼ 1000 times). In this way we obtain a distribution of the values of ft extracted from these fits. This ft distribution is fitted with a Gaussian, and we take its width as our estimate of the statistical uncertainty D ft . The result of this procedure is shown in Fig. 11. The grey contour lines display D ft as a function of AS + B and the number of events. The horizontal lines are the asymmetries AS + B for the three event categories and their combination. Assuming that at the LHC (14 TeV) an integrated luminosity of 150 f b -1 , 500 f b -1 , and 3 ab -1 will be achieved - the two latter numbers are goals for the high-luminosity LHC upgrade [99] - the resulting event numbers are sketched in as black, red, and yellow dots, respectively. The yellow dot on the line for the combined asymmetry, which corresponds to 11000 events, is not shown. For these luminosities our estimate of the statistical uncertainty D ft which can be achieved by using the combination of the three event categories is 27 , ◦ 14 3 , and 5 1 . ◦ . ◦ , respectively.
## VII. SUMMARY
We have investigated how precisely the CP nature of the 125 GeV Higgs boson resonance h can be determined at the LHC (14 TeV) in its decay to t leptons. As to the subsequent t decays, we have taken into account all the major decay modes (3) - (7). Our method for determining the Higgs mixing angle ft , which parameterizes the ratio of the reduced pseudoscalar and scalar Higgst Yukawa couplings, is based on the distribution of the angle j ∗ CP defined in (17). This distribution can be measured for all charged-prong t decays without having to
reconstruct the t ∓ rest frames. For definiteness, we have considered inclusive Higgs-boson production by gluon fusion. The irreducible background from Drell-Yan production of t pairs was analyzed in detail, in particular its contribution to the j ∗ CP distribution. We have studied by Monte Carlo simulation how measurement uncertainties affect the signal and background contributions to this distribution. Based on this study we devised a procedure for obtaining a corrected distribution of this angle. This procedure eliminates to a large extent the distortions due to the measurement uncertainty of the Higgs production vertex. Moreover, we made a proposal how to use the j ∗ CP distribution of Drell-Yan t -pair events for calibrating the experimental uncertainties. Taking the background from Z ∗ / g ∗ → tt and measurement uncertainties by Monte Carlo simulation into account, we found that at the LHC (14 TeV), respectively at the LHC-upgrade, with an integrated luminosity of 150 f b -1 , 500 f b -1 , and 3 ab -1 , the Higgs mixing angle ft can be determined with a precision D ft of 27 ◦ , 14 3 . ◦ , and 5 1 . . ◦
## Acknowledgments
We wish to thank the members of the M tt working group of the Helmholtz Alliance 'Physics at the Terascale' for discussions. The work of W.B. is supported by B.M.B.F., contract 05H12PAE, and S.K. is supported by Deutsche Forschungsgemeinschaft through Graduiertenkolleg GRK 1675.
- [1] G. Aad et al. [ATLAS Collaboration], Phys. Lett. B 716 (2012) 1 [arXiv:1207.7214 [hep-ex]].
- [2] S. Chatrchyan et al. [CMS Collaboration], Phys. Lett. B 716 (2012) 30 [arXiv:1207.7235 [hepex]].
- [3] S. Chatrchyan et al. [CMS Collaboration], Phys. Rev. Lett. 110 (2013) 081803 [arXiv:1212.6639 [hep-ex]].
- [4] G. Aad et al. [ATLAS Collaboration], Phys. Lett. B 726 (2013) 120 [arXiv:1307.1432 [hep-ex]].
- [5] S. Chatrchyan et al. [CMS Collaboration], JHEP 06 (2013) 081 [arXiv:1303.4571 [hep-ex]].
- [6] G. Aad et al. [ATLAS Collaboration], Phys. Lett. B 726 (2013) 88 [arXiv:1307.1427 [hep-ex]].
- [7] S. Chatrchyan et al. [CMS Collaboration], arXiv:1312.5353 [hep-ex].
- [8] ATLAS Collaboration, conference note ATLAS-CONF-2013-108.
- [9] S. Chatrchyan et al. [CMS Collaboration], arXiv:1401.6527 [hep-ex].
- [10] S. Chatrchyan et al. [CMS Collaboration], arXiv:1401.5041 [hep-ex].
- [11] ATLAS Collaboration, conference note ATLAS-CONF-2014-009.
- [12] J. R. Dell'Aquila and C. A. Nelson, Phys. Rev. D 33 (1986) 93.
- [13] J. R. Dell'Aquila and C. A. Nelson, Nucl. Phys. B 320 (1989) 61.
- [14] W. Bernreuther and A. Brandenburg, Phys. Lett. B 314 (1993) 104.
- [15] W. Bernreuther and A. Brandenburg, Phys. Rev. D 49 (1994) 4481 [hep-ph/9312210].
- [16] A. Soni and R. M. Xu, Phys. Rev. D 48 (1993) 5259 [hep-ph/9301225].
- [17] D. Chang, W. -Y. Keung and I. Phillips, Phys. Rev. D 48 (1993) 3225 [hep-ph/9303226].
- [18] V. D. Barger, K. Cheung, A. Djouadi, B. A. Kniehl and P. M. Zerwas, Phys. Rev. D 49 (1994) 79 [hep-ph/9306270].
- [19] M. Krämer, J. H. Kühn, M. L. Stong and P. M. Zerwas, Z. Phys. C 64 (1994) 21 [hepph/9404280].
- [20] K. Hagiwara and M. L. Stong, Z. Phys. C 62 (1994) 99 [hep-ph/9309248].
- [21] T. Arens, U. D. J. Gieseler and L. M. Sehgal, Phys. Lett. B 339 (1994) 127 [hep-ph/9408316].
- [22] A. Skjold and P. Osland, Phys. Lett. B 329 (1994) 305 [hep-ph/9402358].
- [23] T. Arens and L. M. Sehgal, Z. Phys. C 66 (1995) 89 [hep-ph/9409396].
- [24] A. Skjold and P. Osland, Nucl. Phys. B 453 (1995) 3 [hep-ph/9502283].
- [25] S. Bar-Shalom, D. Atwood, G. Eilam, R. R. Mendel and A. Soni, Phys. Rev. D 53 (1996) 1162 [hep-ph/9508314].
- [26] B. Grzadkowski and J. F. Gunion, Phys. Lett. B 350 (1995) 218 [hep-ph/9501339].
- [27] J. F. Gunion, B. Grzadkowski and X. -G. He, Phys. Rev. Lett. 77 (1996) 5172 [hep-ph/9605326].
- [28] W. Bernreuther, A. Brandenburg and M. Flesch, Phys. Rev. D 56 (1997) 90 [hep-ph/9701347].
- [29] S. Bar-Shalom, D. Atwood and A. Soni, Phys. Lett. B 419 (1998) 340 [hep-ph/9707284].
- [30] B. Grzadkowski, J. F. Gunion and J. Kalinowski, Phys. Rev. D 60 (1999) 075011 [hepph/9902308].
- [31] K. Hagiwara, S. Ishihara, J. Kamoshita and B. A. Kniehl, Eur. Phys. J. C 14 (2000) 457 [hepph/0002043].
- [32] T. Han and J. Jiang, Phys. Rev. D 63 (2001) 096007 [hep-ph/0011271].
- [33] T. Plehn, D. L. Rainwater and D. Zeppenfeld, Phys. Rev. Lett. 88 (2002) 051801 [hepph/0105325].
- [34] S. Y. Choi, D. J. Miller, M. M. Mühlleitner and P. M. Zerwas, Phys. Lett. B 553 (2003) 61 [hep-ph/0210077].
- [35] G. R. Bower, T. Pierzchala, Z. Was and M. Worek, Phys. Lett. B 543 (2002) 227 [hepph/0204292].
- [36] K. Desch, Z. Was and M. Worek, Eur. Phys. J. C 29 (2003) 491 [hep-ph/0302046].
- [37] E. Asakawa and K. Hagiwara, Eur. Phys. J. C 31 (2003) 351 [hep-ph/0305323].
- [38] K. Desch, A. Imhof, Z. Was and M. Worek, Phys. Lett. B 579 (2004) 157 [hep-ph/0307331].
- [39] C. P. Buszello, I. Fleck, P. Marquard and J. J. van der Bij, Eur. Phys. J. C 32 (2004) 209 [hepph/0212396].
- [40] R. M. Godbole, S. Kraml, M. Krawczyk, D. J. Miller, P. Niezurawski and A. F. Zarnecki, hepph/0404024.
- [41] A. Rougé, Phys. Lett. B 619 (2005) 43 [hep-ex/0505014].
- [42] S. S. Biswal, R. M. Godbole, R. K. Singh and D. Choudhury, Phys. Rev. D 73 (2006) 035001
[Erratum-ibid. D 74 (2006) 039904] [hep-ph/0509070].
- [43] J. R. Ellis, J. S. Lee and A. Pilaftsis, Phys. Rev. D 72 (2005) 095006 [hep-ph/0507046].
- [44] E. Accomando et al. , hep-ph/0608079.
- [45] R. M. Godbole, D. J. Miller and M. M. Mühlleitner, JHEP 0712 (2007) 031 [arXiv:0708.0458 [hep-ph]].
- [46] P. S. Bhupal Dev, A. Djouadi, R. M. Godbole, M. Mühlleitner and S. D. Rindani, Phys. Rev. Lett. 100 (2008) 051801 [arXiv:0707.2878 [hep-ph]].
- [47] G. Klamke and D. Zeppenfeld, JHEP 0704 (2007) 052 [hep-ph/0703202 [HEP-PH]].
- [48] S. Berge, W. Bernreuther and J. Ziethe, Phys. Rev. Lett. 100 (2008) 171605 [arXiv:0801.2297 [hep-ph]].
- [49] S. Dutta, K. Hagiwara and Y. Matsumoto, Phys. Rev. D 78 (2008) 115016 [arXiv:0808.0477 [hep-ph]].
- [50] S. S. Biswal, D. Choudhury, R. M. Godbole and Mamta, Phys. Rev. D 79 (2009) 035012 [arXiv:0809.0202 [hep-ph]].
- [51] S. Berge and W. Bernreuther, Phys. Lett. B 671 (2009) 470 [arXiv:0812.1910 [hep-ph]].
- [52] M. Reinhard, CP violation in the Higgs sector with a next-generation detector at the ILC , PhD thesis, LLR - École Polytechnique, November 2009; H. Videau, Measuring the CP state of the Higgs through its decay H to tau tau , IWLC2010 (International Workshop on Linear Colliders 2010) at CERN, https://espace.cern.ch/LC2010/default.aspx
- [53] A. De Rujula, J. Lykken, M. Pierini, C. Rogan and M. Spiropulu, Phys. Rev. D 82 (2010) 013003 [arXiv:1001.5300 [hep-ph]].
- [54] N. D. Christensen, T. Han and Y. Li, Phys. Lett. B 693 (2010) 28 [arXiv:1005.5393 [hep-ph]].
- [55] S. Berge, W. Bernreuther, B. Niepelt and H. Spiesberger, Phys. Rev. D 84 (2011) 116003 [arXiv:1108.0670 [hep-ph]].
- [56] R. M. Godbole, C. Hangst, M. Mühlleitner, S. D. Rindani and P. Sharma, Eur. Phys. J. C 71 (2011) 1681 [arXiv:1103.5404 [hep-ph]].
- [57] S. S. Biswal, R. M. Godbole, B. Mellado and S. Raychaudhuri, Phys. Rev. Lett. 109 (2012) 261801 [arXiv:1203.6285 [hep-ph]].
- [58] J. Ellis, D. S. Hwang, V. Sanz and T. You, JHEP 1211 (2012) 134 [arXiv:1208.6002 [hep-ph]].
- [59] S. Banerjee, J. Kalinowski, W. Kotlarski, T. Przedzinski and Z. Was, Eur. Phys. J. C 73 (2013) 2313 [arXiv:1212.2873 [hep-ph]].
- [60] K. Hagiwara, T. Li, K. Mawatari and J. Nakamura, Eur. Phys. J. C 73 (2013) 2489 [arXiv:1212.6247 [hep-ph]].
- [61] A. Djouadi, R. M. Godbole, B. Mellado and K. Mohan, Phys. Lett. B 723 (2013) 307 [arXiv:1301.4965 [hep-ph]].
- [62] C. Englert, D. Goncalves, G. Nail and M. Spannowsky, Phys. Rev. D 88 (2013) 013016 [arXiv:1304.0033 [hep-ph]].
- [63] B. Ananthanarayan, S. K. Garg, J. Lahiri and P. Poulose, Phys. Rev. D 87 (2013) 114002
- [arXiv:1304.4414 [hep-ph]].
- [64] J. Frank, M. Rauch and D. Zeppenfeld, arXiv:1305.1883 [hep-ph].
- [65] R. Harnik, A. Martin, T. Okui, R. Primulando and F. Yu, Phys. Rev. D 88 (2013) 7, 076009 [arXiv:1308.1094 [hep-ph]].
- [66] S. Berge, W. Bernreuther and H. Spiesberger, Phys. Lett. B 727 (2013) 488 [arXiv:1308.2674 [hep-ph]].
- [67] F. Bishara, Y. Grossman, R. Harnik, D. J. Robinson, J. Shu and J. Zupan, JHEP 1404 (2014) 084 [arXiv:1312.2955 [hep-ph]].
- [68] J. Ellis, D. S. Hwang, K. Sakurai and M. Takeuchi, JHEP 1404 , 004 (2014) [arXiv:1312.5736 [hep-ph]].
- [69] A. Kaczmarska, J. Piatlicki, T. Przedzinski, E. Richter-Was and Z. Was, arXiv:1402.2068 [hepph].
- [70] T. Przedzinski, E. Richter-Was and Z. Was, arXiv:1406.1647 [hep-ph].
- [71] Y. Chen, A. Falkowski, I. Low and R. Vega-Morales, arXiv:1405.6723 [hep-ph].
- [72] J. Kopp and M. Nardecchia, arXiv:1406.5303 [hep-ph].
- [73] M. J. Dolan, P. Harris, M. Jankowiak and M. Spannowsky, arXiv:1406.3322 [hep-ph].
- [74] F. Demartin, F. Maltoni, K. Mawatari, B. Page and M. Zaro, arXiv:1407.5089 [hep-ph].
- [75] A. Djouadi and G. Moreau, Eur. Phys. J. C 73 (2013) 2512 [arXiv:1303.6591 [hep-ph]].
- [76] P. Bechtle, S. Heinemeyer, O. Stal, T. Stefaniak and G. Weiglein, Eur. Phys. J. C 74 (2014) 2711 [arXiv:1305.1933 [hep-ph]].
- [77] J. Brod, U. Haisch and J. Zupan, JHEP 1311 (2013) 180 [arXiv:1310.1385 [hep-ph], arXiv:1310.1385].
- [78] P. Bechtle, S. Heinemeyer, O. Stal, T. Stefaniak and G. Weiglein, arXiv:1403.1582 [hep-ph].
- [79] S. Inoue, M. J. Ramsey-Musolf and Y. Zhang, Phys. Rev. D 89 (2014) 115023 [arXiv:1403.4257 [hep-ph]].
- [80] J. M. Campbell and R. K. Ellis, Nucl. Phys. Proc. Suppl. 205-206 (2010) 10 [arXiv:1007.3492 [hep-ph]]; J. M. Campbell, R. K. Ellis, C. Williams, http://mcfm.fnal.gov
- [81] A. Rougé, Z. Phys. C 48 (1990) 75.
- [82] M. Davier, L. Duflot, F. Le Diberder and A. Rougé, Phys. Lett. B 306 (1993) 411.
- [83] J. H. Kühn, Phys. Rev. D 52 (1995) 3128.
- [84] A. Stahl, Springer Tracts Mod. Phys. 160 (2000) 1.
- [85] W. Bernreuther, T. Schröder and T. N. Pham, Phys. Lett. B 279 (1992) 389.
- [86] CMS Collaboration, CMS-PAS-HIG-14-002.
- [87] S. Dittmaier et al. [LHC Higgs Cross Section Working Group Collaboration], arXiv:1101.0593 [hep-ph].
- [88] W. Bernreuther, O. Nachtmann and P. Overmann, Phys. Rev. D 48 (1993) 78.
- [89] S. Catani and M. H. Seymour, Nucl. Phys. B 485 (1997) 291 [Erratum-ibid. B 510 (1998) 503] [hep-ph/9605323].
- [90] J. M. Campbell, R. K. Ellis and G. Zanderighi, JHEP 0610 (2006) 028 [hep-ph/0608194].
- [91] J. M. Campbell, R. K. Ellis and C. Williams, Phys. Rev. D 81 (2010) 074023 [arXiv:1001.4495 [hep-ph]].
- [92] S. Catani and M. Grazzini, Phys. Rev. Lett. 98 (2007) 222002 [hep-ph/0703012].
- [93] M. Grazzini, JHEP 0802 (2008) 043 [arXiv:0801.3232 [hep-ph]].
- [94] M. Grazzini and H. Sargsyan, JHEP 1309 (2013) 129 [arXiv:1306.4581 [hep-ph]].
- [95] H. -L. Lai, M. Guzzi, J. Huston, Z. Li, P. M. Nadolsky, J. Pumplin and C. -P. Yuan, Phys. Rev. D 82 (2010) 074024 [arXiv:1007.2241 [hep-ph]].
- [96] A. Elagin, P. Murat, A. Pranko and A. Safonov, Nucl. Instrum. Meth. A 654 (2011) 481 [arXiv:1012.4686 [hep-ex]].
- [97] S. Gennai et al., [CMS collaboration], Eur. Phys. J. C46 , S01 (2006) 1.
- [98] F. Tarrade [ATLAS Collaboration], Nucl. Phys. Proc. Suppl. 169 (2007) 357.
- [99] HL-LHC: High Luminosity Large Hadron Collider,
- http://hilumilhc.web.cern.ch/HiLumiLHC/index.html
|
10.1140/epjc/s10052-014-3164-0
|
[
"Stefan Berge",
"Werner Bernreuther",
"Sebastian Kirchner"
] |
2014-08-04T20:00:14+00:00
|
2014-12-10T14:32:58+00:00
|
[
"hep-ph",
"hep-ex"
] |
Determination of the Higgs CP mixing angle in the tau decay channels at the LHC including the Drell-Yan background
|
We investigate how precisely the CP nature of the 125 GeV Higgs boson
resonance h can be unraveled at the LHC in its decays to tau pairs. We use a
method which allows to determine the scalar-pseudoscalar Higgs mixing angle in
this decay mode. This mixing angle can be extracted from the distribution of a
signed angle, which we analyze for the major charged-prong tau decays. For
definiteness, we consider Higgs-boson production by gluon fusion at NLO QCD. We
take into account also the irreducible background from Drell-Yan production at
NLO QCD. We compute, for the signal and background reactions, angular and
energy correlations of the charged prongs and analyze which type of cuts
suppress the Drell-Yan background. An important feature of this background is
that its contribution to the distribution of our observable is a flat line,
also at NLO QCD. By separating the Drell-Yan events into two different sets,
two different non-trivial distributions are obtained. Based on this observation
we propose to use these sets for calibation purposes. By Monte Carlo simulation
we study also the effect of measurement uncertainties on this distribution. We
estimate that the Higgs mixing angle can be determined with our method to a
precision of 14 degree (5 degree) at the high luminosity LHC (14 TeV) with an
integrated luminosity of 500 inverse fb (3 inverse ab).
|
1408.0799v1
|
## Mapping the Cosmic Web with the largest all-sky surveys
Maciej Bilicki 1 2 , , John A. Peacock , Thomas H. Jarrett , 3 1 Michelle E. Cluver 1 and Louise Steward 1
1 Department of Astronomy, University of Cape Town, South Africa email: maciek(at)ast.uct.ac.za
2 Kepler Institute of Astronomy, University of Zielona Góra, Poland
3 Institute for Astronomy, University of Edinburgh, United Kingdom
Abstract. Our view of the low-redshift Cosmic Web has been revolutionized by galaxy redshift surveys such as 6dFGS, SDSS and 2MRS. However, the trade-off between depth and angular coverage limits a systematic three-dimensional account of the entire sky beyond the Local Volume ( z < 0 05 . ). In order to reliably map the Universe to cosmologically significant depths over the full celestial sphere, one must draw on multiwavelength datasets and state-of-the-art photometric redshift techniques. We have undertaken a dedicated program of cross-matching the largest photometric all-sky surveys - 2MASS, WISE and SuperCOSMOS - to obtain accurate redshift estimates of millions of galaxies. The first outcome of these efforts - the 2MASS Photometric Redshift catalog (2MPZ, Bilicki et al. 2014a) - has been publicly released and includes almost 1 million galaxies with a mean redshift of z = 0 08 . . Here we summarize how this catalog was constructed and how using the WISE mid-infrared sample together with SuperCOSMOS optical data allows us to push to redshift shells of z ∼ 0 2 . -0 3 . on unprecedented angular scales. Our catalogs, with ∼ 20 million sources in total, provide access to cosmological volumes crucial for studies of local galaxy flows (clustering dipole, bulk flow) and cross-correlations with the cosmic microwave background such as the integrated Sachs-Wolfe effect or lensing studies.
Keywords. catalogs, surveys, galaxies: distances and redshifts, techniques: photometric, methods: data analysis, (cosmology:) large-scale structure of universe, cosmology: observations
## 1. The need for all-sky galaxy surveys in three dimensions
A complete picture of the Cosmic Web we live in would only be known if we could observe the entire extragalactic sky ( 4 π sr) up to the surface of last scattering and with full redshift information. This is of course observationally unachievable, in part due to our Galaxy creating the so-called Zone of Avoidance (ZoA) and the instrumental limitations which force us to choose between wide angular coverage and the depth of a redshift survey. There is, however, a reason and need to observe as much of the sky as possible in three dimensions, as only this enables the framework for comprehensively testing cosmological models. For instance, cosmic microwave background (CMB) observations provide a consistent picture of an early Universe which was very homogeneous and isotropic; such an assumption is also applied to the late-time Universe when modeling it with a simple Friedman-Lemaˆ ıtre metric. However, validity of this Copernican Principle (CP) needs to be verified observationally, and this is only possible if we observe the deep cosmos in all possible directions, and at various redshifts: cosmological tests of the CP require access to the entire celestial sphere in 3D. For instance, analyses of the CMB data have brought to light several 'anomalies' such as the quadrupole-octopole alignment or low observed variance of the CMB signal (Planck collaboration XXIII 2013). A fundamental question is whether these anomalies are confirmed as today's anisotropy and/or inhomogeneity in
the galaxy distribution; some studies indicate this is not the case (Hirata 2009; Pullen & Hirata 2010). The issue, however, requires more scrutiny using more comprehensive low-redshift catalogs. One possible manifestation of late-time violation of the CP would be high-amplitude large-scale flows of galaxies and indeed claims of their existence have been made (Kashlinsky et al. 2008; Watkins et al. 2009), although several independent analyses show otherwise (e.g. Nusser & Davis 2011; Turnbull et al. 2012). Also the issue of which structures, and on what scales, contribute to the pull on the Local Group of galaxies has still not been settled; despite some claims of the convergence of the 'clustering dipole' at distances ∼ 100 Mpc (Rowan-Robinson et al. 2000; Erdogdu et al. 2006), other analyses suggest influence on our motion from much larger depths (Kocevski & Ebeling 2006; Bilicki et al. 2011; see also Nusser et al. 2014). Finally, several other important cosmological signals - such as the integrated Sachs-Wolfe effect (ISW), CMB lensing on the large-scale structure (LSS) or baryon acoustic oscillations (BAO) - do not require full-sky coverage, but are only detectable with sufficiently wide-angle galaxy surveys probing large volumes.
State of the art in all-sky surveys of galaxies. Presently, the largest all-sky catalog of extended sources is the Two Micron All Sky Survey Extended Source Catalog (2MASS XSC, Jarrett et al. 2000). It contains approx. 1.5 million galaxies, of which 1 million are within the completeness limit of the survey, K < s 13 9 . mag. This catalog is however only photometric, and the largest all-sky 3D spectroscopic dataset is its subsample, the 2MASS Redshift Survey of 44,000 galaxies (2MRS, Huchra et al. 2012), with median redshift of only 〈 z 〉 = 0 03 . . There have been attempts to expand 2MRS by adding redshifts of 2MASS galaxies available from other surveys such as 6dFGS or SDSS, which resulted in the the 2M++ compilation by Lavaux & Hudson (2011) of 70,000 galaxies; the depth of such a sample is however non-uniform and limited to what is available from 2MRS on the parts of the sky not covered by other redshift surveys.
In addition to 2MASS, there exist two other major photometric all-sky datasets, namely the Wide-field Infrared Survey Explorer (WISE, Wright et al. 2010) and optical SuperCOSMOS (SCOS for short, Hambly et al. 2001). These samples include hundreds of millions sources, mostly galaxies at high latitudes. WISE mid-infrared data were gathered by an orbiting telescope, which makes them free of such issues as seeing or atmospheric glow. WISE currently provides only a point-source catalog with no 'XSC' release; however, its limited angular resolution will not allow to resolve sources beyond the depths already probed by 2MASS (Cluver et al. 2014). SCOS on the other hand was produced from digitized and calibrated photographic plate data, which are of limited accuracy compared to modern CCD-based surveys. Despite such shortcomings, these datasets will remain the largest all-sky astronomical catalogs at least until Gaia, which will however be dominated by stars. We have thus decided to take advantage of the information contained in 2MASS, WISE and SCOS, focusing on their usefulness for extragalactic science. Specifically, we have undertaken a dedicated program of combining them into multi-wavalength galaxy datasets and adding the third dimension through the methodology of photometric redshifts.
## 2. 2MASS Photometric Redshift catalog (2MPZ)
At the moment, about 30% of 2MASS galaxies have spectroscopic redshifts from other surveys, such as 2MRS, 2dFGRS, 6dFGS or SDSS. This percentage is likely to improve in the coming years, notably in the southern hemisphere with the planned TAIPAN survey, we cannot, however, hope for complete spectroscopic coverage of all the 2MASS galaxies in
Figure 1. All-sky Aitoff projection of 1 million galaxies in the 2MASS Photometric Redshift catalog (2MPZ) in Galactic coordinates, color-coded by photometric redshifts.
the coming years. Presently, the only way to obtain 3D information for this sample - even if of limited accuracy - is to estimate redshifts based on other information, such as fluxes,. This is the widely used technique of photometric redshifts (photo-z's), which so far has not been very popular for lowz samples because of limited availability of multiwavelength data. The situation has though changed mostly due to WISE, which together with SCOS allowed us to add the third dimension to the 2MASS XSC sample. For that purpose we used 2MASS J , H and K s near-IR bands, together with mid-IR WISE ( W 1 , 3.4 µ m and W 2 , 4.6 µ m) and optical B,R,I from SCOS. This gave us 8-band coverage for almost 95% of 2MASS galaxies (with incompleteness mostly at low Galactic latitudes), and the availability of comprehensive spectroscopic subsamples allowed to apply empirical (machine-learning) methodology to calculate photo-z's. We employed an artificial neural network algorithm (ANNz, Collister & Lahav 2004), trained on a representative spec-z subsample of 350,000 galaxies. As a result, we have produced the 2MASS Photometric Redshift catalog (2MPZ, Bilicki et al. 2014a), containing 940,000 galaxies with 〈 z 〉 = 0 07 . , covering most of the sky. This dataset was publicly released in December 2013 and is available for download from http://surveys.roe.ac.uk/ssa/TWOMPZ . The photometric redshifts are well constrained: 1 σ scatter in δz is σ δz = 0 015 . , typical photo-z error is 13% and there are only 3% outliers beyond 3 σ δz . Fig. 1 presents the all-sky distribution of 2MPZ sources, color-coded by photometric redshifts. Despite the tendency of the photoz's to dilute radial information, the 3D cosmic web is evident, with major structures such as the Shapley Superconcentration prominent in the maps. Note that by using infraredselected galaxies we are able to probe deep into the ZoA, although photo-z's at very low latitudes are of limited accuracy due to uncertainties in extinction and large star density compromising the photometric multiwavelength data.
First cosmological results from 2MPZ. The 2MPZ catalog has been already applied to several cosmological tests, of which some have been published, and others are in preparation. Appleby & Shafieloo (2014) used it to test the degree of isotropy in the Local Universe through the K s -band luminosity function (LF), while Xu et al. (2014) showed that the sample is appropriate for galaxy cluster identification. In a forthcoming paper (Steward et al. 2014, in prep.) we will present the constraints on the ISW obtained by
cross-correlating 2MPZ with Planck CMB data. Working in redshift shells (cf. Francis & Peacock 2010, FP10), we have found only mild preference for the ISW over no signal: less than 2 σ (odds 3.5:1). This is an improvement over FP10 (who had odds 1.5:1), possible thanks to our much more accurate photo-z's, the lack of significant detection is however hardly surprising. 2MASS is too shallow to probe the full ISW signal, and the constraints would not be much more significant even if we had full spectroscopic coverage.
Other applications in progress of the 2MPZ data include a detailed study of the nearIR LF at z ∼ 0 (cf. Branchini et al. 2012) from the largest sample to date (Feix et al. , in prep.), as well as constraining the bulk flow in a sphere of ∼ 300 Mpc/ h from LF variations (cf. Nusser et al. 2011; Feix et al. 2014 and Feix et al. in this volume). The catalog is now also analyzed in terms of its applicability to various clustering-related analyses: looking for the transition to homogeneity through the angular correlation function (cf. Alonso et al. 2014) or more general attempts to recover the 3D power spectrum and correlation function from photometric redshifts, setting the ground for upcoming and future surveys such as DES or Euclid. Many more applications of the catalog are of course expected.
## 3. Beyond 2MASS: 20 million galaxies from WISE × SuperCOSMOS
The depth of 2MPZ is limited by the shallowest of the three cross-matched samples, i.e. 2MASS. Both SCOS and WISE probe the Cosmic Web to much larger distances: the former is about 3 times deeper than 2MASS, while the latter reaches even farther. In addition, both contain a significant number of quasars, some at high redshifts (especially WISE). By pairing up WISE with SCOS we were thus able to access sources and redshift shells not available with 2MPZ. Here the sample is limited by the depth and accuracy of the optical data: we adopted B < 21 and R < 19 5 . (AB) as the reliability limits, requiring also source detections in both these bands, while WISE was preselected to have measurements in W 1 and W 2 . Cross-matching the two samples at | b Gal | > 10 ◦ gives 170 million sources, however such a catalog is dominated by stars at the bright end and at low latitudes, so further cuts were needed to clean up the sample. Using Galaxy And Mass Assembly (GAMA, Driver et al. 2009) and SDSS DR10 (Ahn et al. 2014) spectroscopic data, we defined simple color cuts to purify the sample of stars and highz quasars, which together with SCOS morphological information gave us about 20 million galaxies over > 3 π sr. The quasars present in our sample, which were treated as contamination here, will be a subject of a future study; we estimate that there might be even 2 million of them in the WISE × SCOS cross-match. A forthcoming paper (Krakowski et al. 2014) will present a more sophisticated method of source classification in our catalog, namely with the use of Support Vector Machines (e.g. Solarz et al. 2012; Małek et al. 2013) trained on SDSS DR10 spectroscopic data.
The next step was to calculate photometric redshifts for the galaxies in the sample. We used five photometric bands for that purpose: optical B,R and infrared W ,W ,W 1 2 3 . As in 2MPZ, we took the empirical approach (ANNz, Collister & Lahav 2004) but here the training sets were the most recent GAMA and SDSS DR10 data. The former of the two spectroscopic catalogs is crucial here, as it is complete in 3 equatorial fields to r < 19 8 . , which makes it deeper than our sample, hence representative for photo-z calibration. The resulting photometric redshifts have a median 〈 z 〉 ∼ 0 2 . , the redshift distribution is however broad and the sample probes the LSS reliably up to z ∼ 0 35 . , with some information even at z glyph[greaterorsimilar] 0 4 . . Normalized 1 σ scatter of photo-z errors is σ δz = 0 03 . , median error is 12% and there are 3% outliers over 3 σ δz . With the present catalog this is the best performance attainable, as there is no other all-sky data of sufficient depth that could bring additional bands to our full sample. Further details regarding the
Figure 2. The Cosmic Web 2.5 Gyr ago: the large-scale structure of the Universe at z = 0 2 . . The plot shows 1.7 million galaxies from the WISE × SCOS photometric redshift catalog in a shell of 0 19 . < z phot < . 0 21 , in Aitoff projection in Supergalactic coordinates. The gaps are due to our Galaxy and the LMC obscuring the view.
WISE × SCOS photometric redshift catalog construction will be provided in a forthcoming paper (Bilicki et al. 2014b, in prep.). Fig. 2 shows an example of what the WISE × SCOS photo-z sample gives access to: the Cosmic Web as it was 2.5 Gyr ago. Pictured here are 1.7 million galaxies in a redshift shell of 0 19 . < z phot < 0 21 . ; note that there are more sources in this narrow shell than in the entire 2MASS galaxy catalog.
Possible applications of the WISE × SuperCOSMOS photo-z catalog. The availability of the deep, almost all-sky WISE × SCOS data opens new possibilities of LSS analysis, as this is the first 3D galaxy catalog which probes the redshifts of z = 0 2 . ∼ 0 3 . on 75% of the sky. In particular, it will be suitable for similar studies as done with 2MPZ, but in more distant redshift shells and in much larger volumes. Most generally, we expect it to allow for testing the Copernican Principle of isotropy and homogeneity of the Universe up to z ∼ 0 4 . . It should help constrain the contributions to the bulk flows and pull on the Local Group from scales of > 1 Gpc, if there are any. It is also very promising in terms of various correlation analyses: it should provide a higher S/N of ISW than possible with 2MPZ; it can also be cross-correlated with the CMB lensing signal to constrain e.g. non-Gaussianity (Giannantonio & Percival 2014). Some other possible applications include those planned with such major photometric redshift surveys as DES, for instance measuring angular BAOs in redshifts shells to constrain expansion history (see e.g. Blake & Bridle 2005 for an early discussion of such applications). This is certainly not an exhaustive list; as we are planning to publicly release the final photometric redshift catalog, the community will be able to apply many other ideas to these data.
...and beyond: quasars and deep WISE data. Except for the WISE × SCOS z ∼ 0 2 . galaxy catalog described here, we are also planning to compile a quasar photo-z sample based on the same data. Our present estimates show that we should be able to identify about 2 million quasars on > π 3 sr (Krakowski et al. 2014, in prep.); it remains to verify how accurate quasar photo-z's will be possible with the limited photometric
information we have. In addition, as far as WISE itself is concerned, its all-sky 5 σ limit is at ∼ 17 mag (Vega) in the W 1 band, which makes it useful to probe additional redshift layers beyond those available through the cross-match with SCOS. WISE itself is also a great AGN/quasar repository as already shown by several authors. All these properties give great promise for future cosmological studies that we are envisaging to undertake.
Acknowledgments. We thank the Wide Field Astronomy Unit at the IfA, Edinburgh for archiving the 2MPZ catalog, which can be accessed at http://surveys.roe.ac.uk/ssa/TWOMPZ . We made use of data products from the Two Micron All Sky Survey, Wide-field Infrared Survey Explorer, SuperCOSMOS Science Archive, Sloan Digital Sky Survey and Galaxy And Mass Assembly catalogs. Special thanks to Mark Taylor for his wonderful TOPCAT software, http:// www.starlink.ac.uk/topcat/ . The financial assistance of the South African National Research Foundation (NRF) towards this research is hereby acknowledged. MB was partially supported by the Polish National Science Center under contract #UMO-2012/07/D/ST9/02785.
## References
Ahn, C.P.; Alexandroff, R.; Allende Prieto, C.; Anders, F.; et al. 2014, ApJS , 211, 2
Alonso, D.; Bueno Belloso, A.; S´nchez, F.J.; Garc´ ıa-Bellido, J.; a et al. 2014, MNRAS , 440, 1
- Appleby, S.; Shafieloo, A. 2014, arXiv:1405.4595
Bilicki, M.; Chodorowski, M.; Jarrett, T.; Mamon, G.A. 2011, ApJ , 741, 1
Bilicki, M.; Jarrett, T.H.; Peacock, J.A.; Cluver, M.E.; Steward, L. 2014, ApJS , 210, 9
- Bilicki, M.; Peacock, J.A.; Jarrett, T.H.; Maddox, N.; Steward, L.; Cluver, M.E. 2014, in prep.
Blake, C.; Bridle, S. 2005, MNRAS , 363, 4
Branchini, E.; Davis, M.; Nusser, A. 2012, MNRAS , 424, 1
Cluver, M.E.; Jarrett, T.H.; Hopkins, A.M.; Driver, S.P.; Liske, J.; et al. 2014, ApJ , 782, 2 Collister, A.A.; Lahav, O. 2004, PASP , 116, 818
Driver, S.P.; Norberg, P.; Baldry, I.K.; Bamford, S.P.; Hopkins, A.M.; et al. 2009, A&G , 50, 5
Erdogdu, P.; Huchra, J.P.; Lahav, O.; Colless, M.; Cutri, R.M.;
et al.
2006,
MNRAS
,
368, 4
Feix, M.; Nusser, A.; Branchini, E. 2014, arXiv:1405.6710
Francis, C.L.; Peacock, J.A. 2010, MNRAS , 406, 1
Giannantonio, T.; Percival, W.J. 2014, MNRAS , 441, 1
Hambly, N.C.; MacGillivray, H.T.; Read, M.A.; Tritton, S.B.; et al. 2001, MNRAS , 326, 4
Hirata, C.M. 2009, JCAP , 09, 011
Huchra, J.P.; Macri, L.M.; Masters, K.L.; Jarrett, T.H.; Berlind, P.; et al. 2012, ApJS , 199, 2
Jarrett, T.H.; Chester, T.; Cutri, R.; Schneider, S.; Skrutskie, M.; Huchra, J.P. 2000, AJ , 119, 5
Kashlinsky, A.; Atrio-Barandela, F.; Kocevski, D.; Ebeling, H. 2008, ApJ , 686, 2
Kocevski, D.D.; Ebeling, H. 2006, ApJ , 645, 2
Krakowski, T.; et al. 2014, in prep.
Lavaux, G.; Hudson, M.J. 2011, MNRAS , 416, 4
Małek, K.; Solarz, A.; Pollo, A.; Fritz, A.; Garilli, B.; Scodeggio, M.; et al. 2013, A&A , 557, A16
- Nusser, A.; Branchini, E.; Davis, M. 2011, ApJ , 735, 2
- Nusser, A.; Davis, M. 2011, ApJ , 736, 2
- Nusser, A.; Davis, M.; Branchini, E. 2014, ApJ , 788, 2
Planck Collaboration: Ade, P.A.R.; Aghanim, N.; et al. 2013, arXiv:1303.5083
Pullen, A.R.; Hirata, C.M. 2010, JCAP , 05, 027
Rowan-Robinson, M.; Sharpe, J.; Oliver, S.J.; Keeble, O.; et al. 2000, MNRAS , 314, 2
Solarz, A.; Pollo, A.; Takeuchi, T.T.; Pępiak, A.; Matsuhara, H.; et al. 2012, A&A , 541, A50
Steward, L.; Bilicki, M.; Peacock, J.A.; Jarrett, T.H.; Cluver, M.E. 2014, in prep.
Turnbull, S.J.; Hudson, M.J.; Feldman, H.A.; Hicken, M.; et al. 2012, MNRAS , 420, 1
Watkins, R.; Feldman, H.A.; Hudson, M.J. 2009, MNRAS , 392, 2
Wright, E.L.; Eisenhardt, P.R.M.; Mainzer, A.K.; Ressler, M.E.; et al. 2010, AJ , 140, 6
Xu, W.W.; Wen, Z.L.; Han, J.L. 2014, arXiv:1406.0943
|
10.1017/S1743921316009753
|
[
"Maciej Bilicki",
"John A. Peacock",
"Thomas H. Jarrett",
"Michelle E. Cluver",
"Louise Steward"
] |
2014-08-04T20:00:14+00:00
|
2014-08-04T20:00:14+00:00
|
[
"astro-ph.CO",
"astro-ph.IM"
] |
Mapping the Cosmic Web with the largest all-sky surveys
|
Our view of the low-redshift Cosmic Web has been revolutionized by galaxy
redshift surveys such as 6dFGS, SDSS and 2MRS. However, the trade-off between
depth and angular coverage limits a systematic three-dimensional account of the
entire sky beyond the Local Volume (z<0.05). In order to reliably map the
Universe to cosmologically significant depths over the full celestial sphere,
one must draw on multiwavelength datasets and state-of-the-art photometric
redshift techniques. We have undertaken a dedicated program of cross-matching
the largest photometric all-sky surveys -- 2MASS, WISE and SuperCOSMOS -- to
obtain accurate redshift estimates of millions of galaxies. The first outcome
of these efforts -- the 2MASS Photometric Redshift catalog (2MPZ, Bilicki et
al. 2014a) -- has been publicly released and includes almost 1 million galaxies
with a mean redshift of z=0.08. Here we summarize how this catalog was
constructed and how using the WISE mid-infrared sample together with
SuperCOSMOS optical data allows us to push to redshift shells of z~0.2--0.3 on
unprecedented angular scales. Our catalogs, with ~20 million sources in total,
provide access to cosmological volumes crucial for studies of local galaxy
flows (clustering dipole, bulk flow) and cross-correlations with the cosmic
microwave background such as the integrated Sachs-Wolfe effect or lensing
studies.
|
1408.0800v2
|
## Viscous Driving of Global Oscillations in Accretion Discs Around Black Holes
## Ryan Miranda, 1 Jiˇ ´ Hor´k rı a 2 and Dong Lai 1
1 Department of Astronomy, Cornell University, Ithaca, NY 14853, USA
2 Astronomical Institute of the Academy of Sciences, Boˇn´ ı II 1401/1a, 141-31 Praha 4, Czech Republic c
14 April 2015
## ABSTRACT
We examine the role played by viscosity in the excitation of global oscillation modes (both axisymmetric and non-axisymmetric) in accretion discs around black holes using two-dimensional hydrodynamic simulations. The turbulent viscosity is modeled by the α -ansatz, with different equations of state. We consider both discs with transonic radial inflows across the innermost stable circular orbit, and stationary discs truncated by a reflecting wall at their inner edge, representing a magnetosphere. In transonic discs, viscosity can excite several types of global oscillation modes. These modes are either axisymmetric with frequencies close to multiples of the maximum radial epicyclic frequency κ max , non-axisymmetric with frequencies close to multiples of of the innermost stable orbit frequency Ω ISCO , or hybrid modes whose frequencies are linear combinations of these two frequencies. Small values of the viscosity parameter α primarily produce non-axisymmetric modes, while axisymmetric modes become dominant for large α . The excitation of these modes may be related to an instability of the sonic point, at which the radial infall speed is equal to the sound speed of the gas. In discs with a reflective inner boundary, we explore the effect of viscosity on trapped p -modes which are intrinsically overstable due to the corotation resonance effect. The effect of viscosity is either to reduce the growth rates of these modes, or to completely suppress them and excite a new class of higher frequency modes. The latter requires that the dynamic viscosity scales positively with the disc surface density, indicating that it is a result of the classic viscous overstability effect.
Key words: accretion, accretion discs - hydrodynamics - instabilities - waves.
## 1 INTRODUCTION
High-frequency quasi-periodic oscillations (HFQPOs) in Xray flux have been observed in a number of black hole X-ray binaries since the 1990s. These low-amplitude ( ∼ 1%) variabilities occur in the so-called intermediate state or 'steep power law' state, and have frequencies of 40 -450 Hz, comparable to the orbital frequency the innermost stable circular orbit (ISCO) of a ∼ 10 M glyph[circledot] black hole. This indicates that the physical mechanism behind them is closely related to the dynamics of the innermost regions of black hole accretion discs (for reviews, see Remillard & McClintock 2006 and Belloni et al. 2012). Several classes of models for HFQPOs have been proposed (see Lai & Tsang 2009, Lai et al. 2013 for a review), including the orbital motion of hot spots in the disc (Stella et al. 1999; Schnittman & Bertschinger 2004; Wellons et al. 2014), nonlinear resonances (Abramowicz & Klu´ zniak 2001; Abramowicz et al. 2007), and oscillations of finite accretion tori (Rezzola et al. 2003; Blaes et al. 2006). A large class of models are based on relativistic discoseismol- ogy, in which the HFQPO frequencies are identified as those of global oscillations modes of the inner accretion discs (see Kato 2001 or Ortega-Rodr´ ıguez et al. 2008 for reviews of disc oscillations, Lai et al. 2013 for their connection to HFQPOs). Two possible types of oscillations are inertial-gravity modes (or g -modes) with vertical structure and inertial-acoustic p -modes with no vertical structure. While the g -modes exhibit a unique self-trapping property, there is some indication that they may be destroyed by subthermal magnetic fields (Fu & Lai 2009), comparable to those produced by saturation of the magnetorotational instability, or be destroyed by turbulence (Reynolds & Miller 2009). Therefore it is unclear whether or not g -modes can exist in real discs. However, p -modes are only weakly affected by magnetic fields (Fu & Lai 2011), and their lack of vertical structure makes them insensitive to turbulence.
The types of oscillations which are possible in the disc depend on the properties of its inner edge. Either the gas freely flows into the central black hole, or there exists an inner edge which is impermeable to radial motions, for exam-
ple due to the presence of a magnetosphere (e.g. BisnovatyiKogan & Ruzmaikin 1974, 1976; Igumenshchev et al. 2003; Rothstein & Lovelace 2008; McKinney et al. 2012). In the latter case, there exist 'trapped' p -modes which can propagate between the reflective inner boundary and their inner Lindblad resonances. A trapped p -mode can be overstable if it can draw negative energy from the background flow through the corotation resonance. This is possible if the radial derivative of vortensity (vorticity |∇ × v | divided by surface density) is positive at the corotation radius, which, in the absence of sharp changes in surface density, is not possible in Newtonian discs, but can be achieved with general relativistic (GR) effects (see Tsang & Lai 2008; Lai & Tsang 2009; Hor´k & Lai 2013). This 'corotational instaa bility' has been demonstrated in 2D simulations of inviscid discs (Fu & Lai 2013).
One of the goals of this paper is to examine the effect of viscosity on trapped p -modes. In particular, we are interested in whether viscosity suppresses or enhances their growth. The latter may occur as the result of viscous overstability. We model the disc turbulent viscosity by the standard α -ansatz, with different equations of state. The classic theory of viscous overstability describes an axisymmetric instability in which viscosity injects energy drawn from the disc shear into growing oscillations (Kato 1978; Schmit & Tscharnuter 1995; Schmidt et al. 2001). This effect requires viscous forces to act in phase with oscillations, which is achieved if the dynamic viscosity η scales sufficiently steeply with surface density Σ, i.e., if the parameter A = dln η/ dlnΣ is larger than unity. Viscous overstability has been demonstrated in both hydrodynamic (Latter & Ogilvie 2010) and N -body simulations (Rein & Latter 2013). Extensions of the original axisymmetric theory show that a similar effect can drive non-axisymmetric oscillations (Papaloizou & Lin 1988; Lyubarskij et al. 1994). Previous analytical calculations, based on local analysis (see Section 2), indicate that viscosity can excite disc p -modes due to viscous forces acting in phase with the oscillation (e.g. Ortega-Rodr´ ıguez & Wagoner 2000; Kato 2001). However, local analysis cannot determine the damping or growth rate of global modes.
In the absence of a reflective boundary, the inner edge of the disc is free-flowing, with gas interior to the ISCO plunging into the central black hole, forming a transonic radial flow. Kato et al. (1988a) found that the sonic point of such flow is unstable to axisymmetric perturbations for sufficiently large viscosities, a mechanism which is distinct from the standard viscous overstability. One-dimensional simulations of transonic discs have shown that viscosity can drive global oscillations at the maximum epicyclic frequency κ max (Milsom & Taam 1996; Mao et al. 2009). Two-dimensional and three-dimensional simulations have shown global oscillations at the same frequency (O'Neill et al. 2009), as well as at multiples of innermost stable orbital frequency Ω ISCO (Chan 2009). Another goal of this paper is to study the conditions for producing axisymmetric and various non-axisymmetric modes in transonic discs.
The outline of this paper is as follows. In Section 2, we review the theory of viscous overstability, deriving the formula for the growth rate of local non-axisymmetric perturbations due to viscosity. We also provide estimates of the global growth rates of trapped p -modes under a pseudoNewtonian potential. The setup for our numerical experi- ments is described in Section 3. In Section 4, we present the results of simulations of overstable oscillations in a transonic disc with a free inner boundary. We classify several types of global oscillation modes produced in this flow, as well their dependence on the sound speed and viscosity parameter of the disc. In Section 5 we examine trapped p -modes in a disc with a reflecting boundary, first reproducing semi-analytic results for an inviscid disc, then investigating the effect of viscosity as described by the theory of viscous overstability. As part of this discussion (Section 5.3), we present an unexpected result on overstable modes driven purely by viscosity in the absence of GR effects, whose frequencies are higher than trapped p -modes. We summarize our results and discuss their implications in Section 6.
## 2 THEORY OF VISCOUS INSTABILITY
The role of viscosity for the stability of thin accretion discs has been studied extensively for several decades. As shown by Kato (1978) in his pioneering work, viscosity can affect the disc stability through both thermal (affecting the energy balance) and dynamical processes (by changing the angular momentum balance). The former process is analogous to so called ' glyph[epsilon1] -mechanism' in stellar pulsations. In that case, the instability arises as a consequence of an additional increase in the pressure restoring force, due to viscous heat generation in the compressed phase of the oscillations. In the later case, the instability comes from the mechanical work that is done by the azimuthal component of the viscous force on the fluid elements during the oscillations. To explore this mechanism, it is sufficient to consider barotropic fluid, which we adopt in our work.
We consider a thin, two-dimensional disc described in polar coordinates ( r, φ ) by velocity u = ( u , u r φ ), surface density Σ and height integrated pressure P which obey the continuity and Navier-Stokes equations
$$\frac { \partial \Sigma } { \partial t } + \nabla \cdot ( \Sigma u ) = 0, \quad \quad \quad$$
$$\frac { \partial u } { \partial t } + \left ( u \cdot \nabla \right ) u = - \frac { 1 } { \Sigma$$
where σ is the viscous stress tensor,
$$\sigma = \eta \left [ \nabla v + v \nabla - \frac { 2 } { 3 } \left ( \nabla \cdot$$
Φ = Φ( ) is the gravitational potential and r η is the height integrated dynamic viscosity coefficient that describes momentum transport due to turbulent motion of the fluid. In this section, it is sufficient to assume that both P and η are functions of the surface density only, P = P (Σ) and η = η (Σ). The particular forms of these functions will be specified later, when the numerical simulations are carried out.
## 2.1 Subsonic Part of Disc and Trapped Modes
In absence of the viscosity ( η = 0), the equations (1) and (2) admit a stationary axisymmetric solution, that describes a purely rotating flow with the squared angular velocity
$$\Omega ^ { 2 } = \Omega _ { K } ^ { 2 } + \frac { c _ { s } ^ { 2 } } { r ^ { 2 }$$
c 0000 RAS, MNRAS 000 , 000-000
Figure 1. Regions of local viscous stability and instability in the propagation diagram of axisymmetric (left) and non-axisymmetric p-waves. The 'evanescent' regions correspond to the cases where D > 0, the 'locally damped' and 'locally unstable' regions correspond to positive or negative ω v , respectively.
where Ω K is the Keplerian angular frequency and c s = (d P/ dΣ) 1 / 2 is the sound speed. Its contribution to the rotational velocity is negligible in thin discs giving nearly Keplerian rotation, Ω ≈ Ω . K
The perturbations to this equilibrium, whose time and azimuthal dependence is of the form of exp(i mφ -i ωt ), obey the wave equation (e.g. Lai & Tsang 2009),
$$\hat { L } ^ { 0 } \delta h & = \left [ \frac { d ^ { 2 } } { d r ^ { 2 } } - \left ($$
for the enthalpy perturbation δh = δP/ Σ. Here ω and m are the angular frequency and azimuthal wavenumber of the perturbation, ˜ = ω ω -m Ω, D = κ 2 -˜ ω 2 and
$$\kappa ^ { 2 } = r \frac { d \Omega ^ { 2 } } { d r } + 4 \Omega ^ { 2 } \quad \quad \ ( 6 ) \quad \ v _ { d }$$
is the squared radial epicyclic frequency. The perturbation of the flow velocity δ u are given as
$$\delta u _ { r } = \frac { i } { D } \left [ \tilde { \omega } \frac { \mathrm d } { \mathrm d r } - \frac { 2 m \Omega } { r } \right ] \delta h, \quad \delta u _ { \phi } = \frac { 1 } { D } \left [ \frac { \kappa ^ { 2 } } { 2 \Omega } \frac { \mathrm d } { \mathrm d r } - \frac { m \tilde { \omega } } { r } \right ] \delta h. \quad \quad \hat { L } \\ \tau _ { \mu, \mu } \, - \, \dots \, \dots \, \hat { r } \, 0 \, \colon - \, \dots \, \dots \, \pi \, \colon \, \alpha \, - \, \hat { r } \, \colon \, \alpha \, \tau _ { \mu, \mu } \, - \, \hat { r } \, \colon \, \alpha \, \tau _ { \mu, \mu } \, - \, \hat { r } \, \colon \, \alpha \, \tau _ { \mu, \mu } \, - \, \hat { r } \, \colon \, \alpha \, \tau _ { \mu, \mu } \, - \, \hat { r } \, \colon \, \alpha \, \tau _ { \mu, \mu } \, - \, \hat { r } \, \colon \, \alpha \, \tau _ { \mu, \mu } \, - \, \hat { r } \, \colon \, \alpha \, \tau _ { \mu, \mu } \, - \, \hat{ r } \, \colon \, \alpha \, \tau _ { \mu, \mu } \, - \, \hat { r } \, \colon \, \alpha \, \tau _ { \mu, \mu } \, - \, \hat { r } \, \colon \, \alpha \, \tau _ { \mu, \mu } \, - \, \hat { r } \, \colon \, \alpha \, \tau _ { \mu, \mu } \, - \, \hat { r } \, \colon \, \alpha \, \tau _ { \mu, \mu } \, - \, \hat { r } \, \colon \, \alpha \, \tau _ { \mu, \mu } \, - \, \hat { r } \, \colon.$$
The operator ˆ L 0 is singular when D → 0 or ˜ ω → 0. The first corresponds to the Lindblad resonances and the second to the corotation resonance. While the Lindblad resonances are not real singularities and represent turning points of the waves, the waves may be absorbed at the corotation resonance, which may lead to the instability or damping of the oscillations of the disc.
Introducing the viscosity changes both the equlibrium state and the dynamics of the oscillations. The angular momentum transport causes a slow radial inflow of the matter towards the central black hole in the stationary case. The radial velocity u r of this inflow can be found by expanding the azimuthal component of equation (2) up to the first order in η ,
$$u _ { r } = \frac { 2 \Omega } { r ^ { 2 } \kappa ^ { 2 } \Sigma } \frac { d } { d r } \left ( r ^ { 3 } \eta \frac { d \Omega } { d r } \right ). \quad \quad ( 8 )$$
This expansion however breaks down close to ISCO, where κ → 0 and the flow becomes transonic. The stability of this region of the disc is reviewed later in this section and numerically examined in section 4.
In addition to a small change in the mean flow, the viscosity affects the oscillations through the perturbation of the viscous force δ ( ∇ · σ ) that may do a positive or negative work on the waves. Up to the first order in viscosity, the perturbations of the disc are governed by the equation (see Appendix A)
$$\left ( \hat { L } ^ { 0 } + L _ { \nu } ^ { 1 } + L _ { \text{in} } ^ { 1 } \right ) \delta h = 0,$$
where ˆ L 0 is the differential operator of the inviscid problem defined in equation (5) and ˆ L 1 in and ˆ L 1 v are its first-order corrections due to the radial inflow in the stationary case and the action of the viscous force:
$$\cdot \quad \hat { L } _ { \text{in} } ^ { 1 } = \text{i} u _ { r } \frac { \tilde { \omega } } { D } \left [ \left ( 1 + \frac { \kappa ^ { 2 } } { \tilde { \omega } ^ { 2 } } \right ) \frac { d ^ { 3 } } { d r ^ { 3 } } - \frac { D ^ { 2 } } { \tilde { \omega } ^ { 2 } c _ { s } ^ { 2 } } \frac { d } { d r } \right ], \quad \quad ( 1 0 )$$
$$\hat { L } _ { \nu } ^ { 1 } = - \mathrm i \nu \frac { \tilde { \omega } } { D } \left [ \left ( \frac { 4 } { 3 } + \frac { \kappa ^ { 2 } } { \tilde { \omega } ^ { 2 } } \right ) \frac { d ^ { 4 } } { d r ^ { 4 } } - 2 q A \frac { D } { \tilde { \omega } ^ { 2 } } \frac { \mu } { c _ { s } ^ { 2 } } \frac { d ^ { 2 } } { d r ^ { 2 } } \right ], \quad ( 1 1 )$$
where ν = η/ Σ is the kinematic viscosity.
By substituting the local WKBJ ansatz, δh ∝ exp(i kr ) we may recover the formula of Kato (1978) for the local growth rate of the oscillations due to the action of the viscous force,
$$\delta \omega _ { v } ( r ) = - \mathrm i \nu k ^ { 2 } \left [ \frac { 2 } { 3 } + \frac { \Omega ^ { 2 } } { \tilde { \omega } ^ { 2 } } \left ( \frac { \kappa ^ { 2 } } { 2 \Omega ^ { 2 } } - q A \right ) \right ], \quad ( 1 2 )$$
where q = -dlnΩ dln / r and A = dln η/ dlnΣ. Because q > 0, the last term reduces the damping and in principle it can lead to an instability. A necessary condition for
mode growth [Im ( δω v ) > 0] is positive A , i.e. the shear viscosity coefficient η increases with increasing density Σ on a timescale shorter than the oscillation period ( ∼ 1 Ω). For a / given oscillation frequency, we may separate the regions of local instability or damping of the p -waves in the propagation diagram. This is done in Figure 1.
A similar analysis in the case of the operator ˆ L in gives the correction due to the radial inflow in the disc,
$$\delta \omega _ { \text{lin} } ( r ) = - k u _ { r }. \quad \quad ( 1 3 ) \quad \stackrel { \tilde { \cdot } } { i }$$
The meaning of this result is straightforward; it is the Doppler shift between the stationary observer at a fixed radius and an observer comoving with the fluid. Hence, the inflow only slightly changes the real part of the frequency of the oscillations without affecting their stability. The total frequency change is the sum of the two contributions, δω r ( ) = δω v ( r ) + δω in ( r ).
Similarly, we may derive a global change of the eigenfrequencies of the p -modes trapped between two boundaries. A simple calculation presented in Appendix B leads to the expressions
$$\delta \omega _ { \nu, \text{global} } & = \frac { \int \delta \omega _ { \nu } ( r ) w ( r ) d r } { \int w ( r ) d r }, \quad \delta \omega _ { \text{in,global} } = \frac { \int \delta \omega _ { \text{in} } ( r ) w ( r ) d r } { \int w ( r ) d r }, \quad \text{their g} \\ \dots \dots$$
where
$$w ( r ) = \frac { 1 } { c _ { s } } \frac { \tilde { \omega } } { \sqrt { - D } }. \quad \quad ( 1 5 ) \quad \ \ 3 \\ \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
Hence, the effects of viscosity and radial inflow are just the average of the local rates of equations (12) and (13) with the weight function w r ( ).
## 2.2 Transonic flow
The studies of the stability of the transonic regions of accretion discs were initiated by the work of Kato et al (1988a). By perturbing equations (1) and (2) around a transonic stationary solution describing an isothermal accretion disc, the authors identified two types of unstable axisymmetric perturbations. The first type is essentially a generalization of the inertial-acoustic p -waves discussed in the previous subsection to the case of a nonzero radial flow velocity. These modes are trapped between the disc inner edge (corresponding to the sonic radius) and the radius of the inner Lindblad resonance. Consequently, the frequency of these waves is always smaller than κ max , we note however, that this limit is valid only in the case of the axisymmetric perturbations. According to Kato et al. (1988a), as the sonic radius acts only as a partial reflector for the waves, the instability appears only for a sufficiently high viscosity when the viscous driving overcomes the leakage of wave energy through the sonic radius.
The other type of instability appears only in the case of a transonic flow and represents a standing wave pattern localized around the sonic radius that grows exponentially with time. Kato et al. (1988a) assumed so-called conventional or ' αp '-type viscosity, in which the rφ -component of the viscous stress tensor is directly proportional to the pressure, σ rφ = αp (contrary to our 'diffusive' prescription for the viscosity, that relates pressure to the shear viscosity coefficient η ). They found that the instability occurs only for high enough values of α , when α glyph[greaterorequalslant] u / ′ c Ω , c with u ′ c and Ω c being a radial velocity gradient and orbital frequency of the flow at the sonic radius. They also noted that the same condition is satisfied when the flow changes from being subsonic to supersonic by passing through the nodal critical point. They speculated that the appearance of this second type of instability is directly related to the topology of the flow at the sonic radius. This idea was supported in subsequent studies: Kato et al. (1988b) relaxed the assumption of isothermal flow by including the energy balance between viscous heating and radiative cooling and found that the instability criterion coincides with the one for the nodal topology of the sonic point. Later on, Kato et al (1993) examined a stability of the isothermal accretion flows with the diffusive form of the viscosity similar to our work. In that case the sonic point is always of the saddle type and consequently the authors found that it is stable against this type of perturbation.
Based on these results we expect that possible instabilities of our flow arise only due to propagating acoustic waves mentioned in the beginning of this subsection. In the following sections, we will examine numerically the conditions for their growth as well as the spatial structure of the unstable modes.
## 3 NUMERICAL SETUP
As in the previous section, we consider a thin disc described by velocity u = ( u , u r φ ), surface density Σ and pressure P which obey the Navier-Stokes equations. It is subject to the 'pseudo-Newtonian' (Paczy´ski & Wiita 1980) gravitan tional potential
$$\Phi = - \frac { G M } { r - r _ { s } }, \quad \quad \quad ( 1 6 )$$
where M is the mass of the central black hole and r s = 2 GM/c 2 is its Schwarzschild radius. This potential mimics GR effects, having a Keplerian orbital frequency and radial epicyclic frequency given by
$$\Omega _ { K } = \sqrt { \frac { G M } { r ^ { 3 } } } \left ( \frac { r } { r - r _ { s } } \right ), \ \kappa = \Omega _ { K } \sqrt { \frac { r - 3 r _ { s } } { r - r _ { s } } }. \quad ( 1 7 )$$
There is an innermost stable circular orbit (ISCO) defined by κ 2 ( r ISCO ) = 0, and a radius at which the epicyclic frequency peaks, κ r ( max ) = κ max = ( 9 -5 √ 3 ) Ω ISCO , which are located at r ISCO = 3 r s and r max = ( 2 + √ 3 ) r s , respectively.
We adopt numerical units such that
$$r _ { I S C O } = \Omega _ { I S C O } = 1, \text{ \quad \ \ } ( 1 8 )$$
and define an 'orbit' as one orbital period at r ISCO (equal to 2 π in numerical units). We use a polytropic equation of state P = K Σ , which has the corresponding sound speed Γ c s = √ Γ P/ Σ. We define the parameter c s0 = c s ( r ISCO ), whose value we will refer to rather than that of the constant K , to which it is directly related. We prescribe the kinematic viscosity of the fluid as
$$\nu = \alpha c _ { s } H, \text{ \quad \ \ } ( 1 9 )$$
where H = c / s Ω K is the disc scale height (Shakura & Sunyaev 1973). Note that ν is proportional to c 2 s , so that
ν (Σ) ∝ Σ Γ -1 or η (Σ) ∝ Σ , giving Γ A = dln η/ dlnΣ = Γ. Therefore each simulation is defined by three parameters: c s0 , Γ and α .
The outer boundary is always located at r = 4, at which we fix Σ and u φ at their initial values and impose ∂u /∂r r = 0. Adjacent to the outer boundary, we implement a wave damping zone of width 1 3 (i.e., / between r = 11 3 and / r = 4) in which a damping force (per unit mass), given by
$$f _ { \text{damp} } = - \frac { \mathfrak { u } - \mathfrak { u } _ { 0 } } { \tau } R \left ( r \right ) \text{ \quad \ \ } \text{(2)}$$
is applied (de Val-Borro et al. 2006). Here u 0 = ( Ω r 0 , 0), τ is a damping timescale equal to the orbital period at the outer boundary, and R r ( ) is a parabolic function, which is equal to unity at r = 4 and reduces to zero at r = 11 3. The / damping zone minimizes reflection of waves from the outer boundary and mimics an outgoing wave boundary condition used in linear analysis of disc modes (e.g., Lai & Tsang 2009). We simulate two distinct physical setups which correspond to two different inner boundary locations and their corresponding boundary conditions, choosing either a free boundary at r = 2 3 (Section 4) or a reflecting boundary / at r = 1 (Section 5). The details and significance of these boundary conditions are discussed in more detail in their corresponding sections. The initial conditions of the simulations always have u r = 0 everywhere, but the initial surface density profile Σ( ) varies in the different sections of this r paper. The orbital velocity u φ is always chosen to be initially in centrifugal balance (equation 4) given the surface density profile.
The Navier-Stokes equations are solved using the PLUTO code (Mignone et al. 2007) with third-order RungeKutta time stepping, parabolic reconstruction and a Roe solver on static polar grid with uniform spacing in both directions. The standard resolution of the simulations is N r × N φ = 1024 × 256, so that the radial direction has a spatial resolution of at least 300 per unit r (this varies slightly with the location of the inner boundary), and modes with m glyph[lessorequalslant] 4 are captured with at least 64 zones per azimuthal wavelength. The sensitivity of our numerical results to the chosen resolution is addressed seperately for each major results section (Sections 4 and 5).
## 4 OVERSTABLE GLOBAL OSCILLATIONS IN TRANSONIC DISCS
For our transonic disc simulations, the inner boundary is placed at r in = 2 3, at which an outflow boundary condition / given by
$$\frac { \partial \Sigma } { \partial r } = \frac { \partial \mathbf u } { \partial r } = 0 \quad \quad \ \ ( 2 1 ) \quad \ \ \ \ ;$$
is imposed. We choose as our initial condition
$$\Sigma \in \frac { r ^ { - \frac { 1 } { 2 } } } { r - r _ { 8 } } \left [ 1 - \left ( \frac { r } { r _ { \text{in} } } \right ) ^ { - \frac { 1 } { 2 } } \right ], \quad \quad ( 2 2 )$$
which is an approximate steady-state surface density profile consistent with constant ˙ M and vanishing torque at r in , but it is not a solution for the transonic flow region interior to r ISCO . From these initial conditions, the disc is evolved for 100 orbits. We wait for it to settle into a quasi-steady
Figure 2. Surface density at t = 100 for runs with significant time variability. The parameters corresponding to the alphabetical labels are given in Table 1. Various features can be seen, including four-armed, two-armed and one-armed spirals, as well as axisymmetric rings. The top two panels are zoomed in relative to the other panels to show finer detail of the non-axisymmetric structure close to the inner edge.
state before analyzing its variabilities. The parameters for various runs and their corresponding key results are summarized in Table 1. Run ( a ), which has the longest viscous timescale, is the slowest to reach a quasi-steady state, taking ∼ 60 orbits (however, this is much less than the viscous timescale). Therefore we perform all analysis in the last 30 orbits of each run, at which point all have reached a quasisteady state. Since we ignore the initial transient phase, we do not attempt to characterize the growth process or linear behavior of any observed oscillations. Instead, we focus on the properties of unstable oscillations which have saturated at some (often large) amplitudes and which may have become significantly nonlinear.
We use the following formalism to analyze the variabilities of the quasi-steady, nonlinear phase of the simulations. We define the power spectrum | ˜ u | 2 , where
$$\tilde { u } \left ( r, \omega \right ) = \frac { 1 } { 2 \pi } \int _ { t _ { 1 } } ^ { t _ { 2 } } u _ { r } \left ( r, 0, t \right ) e ^ { - i \omega t } \mathrm d t \quad \ \ ( 2 3 )$$
is the Fourier transformed radial velocity and ( t 1 , t 2 ) = (70 100). , The power spectrum gives a measure of the strength of the oscillations, along with their frequencies and locations in the disc where they are most visible. We determine the azimuthal number m associated with each fre-
-
Table 1. Summary of transonic disc simulations. The first three columns give the name of the run and the value of parameters c s0 and α . The next two columns give the numerically measured location of the sonic point r c and its dimensionless stability quantity u / ′ c Ω . c The last three columns indicate whether or not overstable oscillations of the three types described in Section 4.1 are present. For the κ max type, we also indicate whether or not they are linear or nonlinear, and if the m Ω ISCO type are present, we indicate which azimuthal number m is dominant in the power spectrum.
| Label | c s0 | α | r c | u ′ c / Ω c | κ max | m Ω ISCO | m Ω ISCO ± κ max |
|---------|--------|--------|--------|---------------|-----------|------------|--------------------|
| (a) | 0 . 01 | 0 . 05 | 0 . 99 | 0 . 150 | No | No | No |
| (b) | 0 . 01 | 0 . 10 | 1 . 02 | 0 . 120 | No | 4 | No |
| (c) | 0 . 01 | 0 . 25 | 1 . 12 | 0 . 087 | Linear | 2 | Yes |
| (d) | 0 . 01 | 0 . 50 | 1 . 30 | 0 . 082 | Nonlinear | 1 | Yes |
| (e) | 0 . 02 | 0 . 10 | 1 . 01 | 0 . 170 | No | No | No |
| (f) | 0 . 02 | 0 . 25 | 1 . 17 | 0 . 109 | No | 1 | Yes |
| (g) | 0 . 02 | 0 . 50 | 1 . 69 | 0 . 116 | Nonlinear | No | No |
| (h) | 0 . 05 | 0 . 25 | 1 . 12 | 0 . 197 | No | No | No |
| (i) | 0 . 05 | 0 . 50 | 1 . 57 | 0 . 446 | Nonlinear | No | No |
Figure 3. Space-time diagrams of surface density Σ at φ = 0, demonstrating the structure of the spatial variabilities and temporal variabilities of the oscillations. Time is shown relative to the reference time t 0 = 90. We can see that waves with frequencies close to multiple of Ω ISCO (if they are present) do not propagate at radii much larger than r ISCO , beyond which only axisymmetric oscillations with frequency approximately κ max propagate.
quency by decomposing the radial velocity into components proportional to exp ( imφ -iωt ). Since, as we will demonstrate, the oscillations are global (their frequencies are coherent across a range of radii), we average this amplitude
)
(
|
|
Figure 4. The power spectrum | ˜ u r | 2 over a range of disc radii. Global modes with power in discrete frequencies which are coherent across a range of r are present in all cases. They can be seperated into modes with frequencies close to multiples of Ω ISCO (and their nonlinear splittings) with power concentrated near r ISCO , and those with frequencies close to multiples of κ max with power concentrated at larger radii. These correspond respectively to the spirals and axisymmetric rings seen in Figure 2.
over all r . The resulting quantity,
$$\dots \, \dots \, \dots \, \dots \, \cdots \, \cdots \, \nu \, \dots \, \nu \,, \\ \tilde { u } _ { m } \left ( \omega \right ) & = \frac { 1 } { 4 \pi ^ { 2 } \left ( r _ { \text{out} } - r _ { \text{in} } \right ) } \\ \vdots & \quad \times \int _ { r _ { \text{in} } } ^ { r _ { \text{out} } } \int _ { t _ { 1 } } ^ { t _ { 2 } } \int _ { 0 } ^ { 2 \pi } u _ { r } \left ( r, \phi, t \right ) e ^ { - \left ( i m \phi - i \omega t \right ) } d \phi d t d r, \\ \vdots & \quad \text{(24)} \\ \ell \quad \text{species the global power spectrum delineated by $m$. The}$$
specifies the global power spectrum delineated by m . The
c
0000 RAS, MNRAS
000
,
000-000
Figure 5. Power spectrum delineated by azimuthal number m , given by the quantity | ˜ u m | 2 defined in eq. (24).
spatial and temporal variability and propagation properties of the oscillations are examined by plotting the surface density Σ at a fixed azimuthal angle ( φ = 0) as a function of radius r and time t .
For this section we restrict the equation of state to only the isothermal case, Γ = 1, so that each run is characterized only by the parameters c s0 and α . The parameters of the nine numerical runs (see Table 1) were chosen because they lead to a variety of behaviors, producing overstable oscillations of several types (to be described below) which are present or absent in various combinations. For each sound speed c s0 , there is a threshold value of the viscosity parameter α below which no overstable oscillations are observed within the duration of the run (100 orbits). Three of our runs, ( a ), ( e ) and ( h ), fall into this regime, leaving six runs which show significant variabilities. The subsequent discussion focuses only on these six interesting cases.
Figure 2 shows snapshots of the disc surface density Σ at t = 100 for the runs that exhibit overstable oscillations. Figure 3 shows Σ at φ = 0 as function of time and radius. Together they illustrate the spatial structure and wave propagation properties of the modes. In runs ( ), b ( c ), ( d ) and ( ), i non-axisymmetric waves with frequencies approximately multiples of Ω ISCO propagate at small radii ( r glyph[lessorsimilar] 1 2 for . c s0 = 0 01, . r glyph[lessorsimilar] 1 7 for . c s0 = 0 02). In runs ( ), . c ( d ), ( g ) and ( ), axisymmetric waves with lower frequencies i propagate at larger radii. In cases in which both types of waves are present, there is a merging or winding-up of nonaxisymmetric waves into axisymmetric ones at intermediate radii.
Figure 4 depicts the power spectra, for a range of radial locations in the disc, of the runs which exhibit overstable oscillations. In all cases, power is sharply concentrated in discrete frequencies and present across a large range of r , manifesting as narrow vertical strips in Figure 4. We therefore interpret these as frequencies of coherent global modes, rather than local oscillations whose properties vary with r . Since they are global, with frequencies independent of location, we can justify the radial integration in equation (24), which allows us to analyze their relative strengths globally, rather than locally. The m -delineated power spectra are shown in Figure 5. Note that the absolute scale of the quantity | ˜ u m | 2 does not have a direct physical meaning, as it depends on the radial integration range (which we choose as the entire computational domain, whose size is arbitrary). However the relative amplitudes, both between different values of m and across multiple panels in Figure 5, give a meaningful comparison of the power in the various modes. From these we see that run ( ), ( b c ), ( d ) and ( f ) are dominated by m = 4, m = 2, m = 1 and m = 1 modes, respectively. Runs ( g ) and ( ) have very little power in non-axisymmetric modes, with i the majority of their power in axisymmetric ( m = 0) modes, whose corresponding | ˜ u m | is not shown.
## 4.1 Classification of Oscillations
We classify the observed oscillation modes into three types. First, there is a ' κ max ' type, which is axisymmetric ( m = 0) and has a frequency approximately equal to (for small c s0 ) or somewhat smaller than (for large c s0 ) the maximum epicyclic frequency, κ max = 0 34. Under various conditions, this mode . is either linear or nonlinear. In the former case, its wave function is smooth and has a small amplitude, and only the fundamental frequency is present in the power spectrum. In the latter case, the wave function is sharp and large in amplitude, and many overtones (integer multiples of κ max ) are present in the power spectrum. This behavior is analogous to that of a nonlinear oscillator with natural frequency κ max . As the amplitude of the oscillation becomes large, nonlinear effects introduce resonances at multiples of the natural frequency, and oscillations at these frequencies can also be driven to large amplitudes. The result is the excitation of a series of harmonics of the frequency κ max . In the most extreme examples of this nonlinearity [e.g. runs ( d ) and ( g )], the power density (see Figure 4) in the first three to four overtones is comparable to (less than, but within an order of magnitude of) that of the fundamental frequency, and up to ten overtone frequencies can be identified [run ( g )].
The second type of mode is the ' m Ω ISCO ' type, which is non-axisymmetric with azimuthal number m> 0 and frequency ω ≈ m Ω ISCO . These always appear in a series of successive m , and therefore represent a harmonic series of the fundamental frequency (similar to the nonlinear effect seen in the κ max modes), which is simply Ω ISCO . However, the m with the largest power is not always 1, we find cases in which it is 2 or 4. This manifests as a sharp, nonlinear m -armed spiral in the disc.
The third type of mode is the ' m Ω ISCO ± κ max ' type, which has azimuthal number m> 0 and frequency approxi-
mately m Ω ISCO ± glyph[epsilon1] , where glyph[epsilon1] is close to, but sometimes less than, κ max . These are prominently present for moderate values of α when c s0 is small [runs ( c ) and ( d )], moderately present for a similar α at a larger c s0 [run ( f )] and entirely absent for large α or large c s0 [runs ( g ) and ( )]. We suggest i that these modes are a result of a splitting of the m Ω ISCO modes due to a nonlinear coupling with the κ max modes, as these are the most fundamental modes associated with the two 'special' locations in the disc, r ISCO and r max .
## 4.2 Dependence on Parameters
From our runs, we can extrapolate the dependence on the parameters c s0 and α in determining power spectrum of overstable modes. For a given sound speed c s0 , we imagine slowly increasing the value of α , starting from a very small value. A sufficiently small α produces no overstable oscillations, leading to a steady-state axisymmetric flow with no variabilities. As α increases, there is a threshold value at which m Ω ISCO type modes begin to appear. This threshold value of α is larger for larger c s0 . For example, the threshold is α glyph[greaterorsimilar] 0 05 for . c s0 = 0 01 and . α glyph[greaterorsimilar] 0 1 for . c s0 = 0 02. Once this . threshold is reached, the dominant m Ω ISCO mode initially involves large m , such as m = 4 seen in run ( ). As b α is further increased, the dominant m of these modes becomes smaller, for example m = 2 in run ( c ) and m = 1 in run ( d ). As α increases further, beyond a second threshold value, axisymmetric κ max modes become present, while the m Ω ISCO modes may still be present with comparable power, as in run ( d ), or may be suppressed, such that only κ max modes are present, as in runs ( g ) and ( ). In between the two threshold i values of α , it is possible for the m Ω ISCO ± κ max modes to be present, sometimes in the absence of strong κ max modes, such as in runs ( ) c and ( f ), probably due to a nonlinear coupling between the two other types of modes.
As reviewed in Section 2, Kato et al. (1988a) derived a criterion a for a viscous pulsational instability at the sonic point r c , defined by u r ( r c ) = c s , of a transonic flow. According to this criterion, the sonic point is unstable if α > u / ′ c Ω , c where u c and Ω c are the radial velocity and angular velocity at the critical point and the prime denotes a radial derivative. Note that these quantities themselves depend on α (and c s0 ) in a self-consistent solution of the flow. While this instability criterion may not be exactly applicable in our simulations (due to our 'diffusive' rather than ' αP ' viscosity prescription), we find it is still a useful diagnostic. Using azimuthally-averaged and time-averaged profiles of u r and u φ , we measure the location of the sonic point r c and the value of u / ′ c Ω c for each run, which are presented in Table 1. From these we can see that the Kato et al. criterion successfully indicates whether or not overstable oscillations are found, with the exception of run ( b ), for which a weak m = 4 mode is found even though the criterion is not strictly satisfied.
The results of our simulations of transonic discs can be most directly compared to those of Chan (2009), who also examined the role of viscosity in height-integrated discs (with a more general equation of state), and found evidence for viscous excitation of global axisymmetric and nonaxisymmetric p -modes of the types found in our simulations. Our work extends Chan's analysis by delineating the different regimes (ranges of c s0 and α ) in which particular combi-
Figure 6. Surface density at t = 100 and power spectrum of resolution test runs ( c ) half and ( ) c double (compare to Figures 2 and 5).
nations of these modes are excited, and highlights the possibility of extreme nonlinearities and the existence of the mixed m Ω ISCO ± κ max type modes. The driving of the κ max modes at large α is consistent with the 1D (height-integrated and axisymmetric) simulations of Milsom & Taam (1996) and Mao et al. (2009), as well as the 2D (axisymmetric) simulations of O'Neill et al. Note that the aforementioned studies are unable to capture the global spiral modes seen in our simulations due to their assumptions of axisymmetry.
## 4.3 Resolution Study
In order to determine the robustness and dependence on grid resolution of the features seen in our transonic disc simulations, we perform run ( c ) with both half the standard resolution in both dimensions, N r × N φ = 512 × 128, denoted ( ) c half and double the standard resolution N r × N φ = 2048 × 512, denoted ( ) c double . The surface density at t = 100 and the m -delineated power spectra for these runs are shown in Figure 6. It can be seen from these that the runs are remarkably similar to one another, as well as to the standard resolution run, therefore we conclude that the observed features are not sensitive to resolution and that the standard resolution provides sufficient convergence. Since run ( c ) has the largest Reynolds number Re = α -1 c -2 s0 of the runs which exhibit all three types of oscillations, we conclude that our other runs are at least as well resolved as ( c ) under the standard resolution.
Table 2. Summary of trapped p -mode simulations. From left to right, the columns give the equation of state ( c s0 and Γ), azimuthal mode number m , the semi-analytic mode frequency ω r and growth rate ω i (zero viscosity), the numerically-calculated mode frequency, and the numerically-calculated growth rate for α = 0 and several non-zero values of α . The blank entries correspond to runs for which p -modes were not observed due to a second type of instability, as described in Section 5.3.
| | m | ( ω r , ω i ) semi - analytic | ω r | ω i ,α =0 | ω i ,α =0 . 01 | ω i ,α =0 . 05 |
|---------------------------|-----|---------------------------------|--------|-------------|------------------|------------------|
| c s0 = 0 . 02 , Γ = 1 | 2 | 1 . 78 , 0 . 0422 | 1 . 78 | 0 . 0453 | 0 . 0314 | 0 . 0266 |
| c s0 = 0 . 02 , Γ = 1 | 3 | 2 . 69 , 0 . 0528 | 2 . 69 | 0 . 0579 | 0 . 0426 | 0 . 0235 |
| c s0 = 0 . 02 , Γ = 1 | 4 | 3 . 62 , 0 . 0607 | 3 . 63 | 0 . 0646 | 0 . 0478 | 0 . 0205 |
| c s0 = 0 . 02 , Γ = 4 / 3 | 2 | 1 . 78 , 0 . 0408 | 1 . 78 | 0 . 0434 | - | - |
| c s0 = 0 . 02 , Γ = 4 / 3 | 3 | 2 . 69 , 0 . 0512 | 2 . 69 | 0 . 0550 | - | - |
| c s0 = 0 . 02 , Γ = 4 / 3 | 4 | 3 . 62 , 0 . 0591 | 3 . 63 | 0 . 0611 | - | - |
| c s0 = 0 . 05 , Γ = 1 | 2 | 1 . 62 , 0 . 0588 | 1 . 62 | 0 . 0600 | 0 . 0526 | 0 . 0211 |
| c s0 = 0 . 05 , Γ = 1 | 3 | 2 . 48 , 0 . 0716 | 2 . 47 | 0 . 0716 | 0 . 0642 | 0 . 0376 |
| c s0 = 0 . 05 , Γ = 1 | 4 | 3 . 36 , 0 . 0800 | 3 . 36 | 0 . 0776 | 0 . 0704 | 0 . 0479 |
| c s0 = 0 . 05 , Γ = 4 / 3 | 2 | 1 . 62 , 0 . 0553 | 1 . 61 | 0 . 0552 | - | - |
| c s0 = 0 . 05 , Γ = 4 / 3 | 3 | 2 . 48 , 0 . 0682 | 2 . 47 | 0 . 0665 | - | - |
| c s0 = 0 . 05 , Γ = 4 / 3 | 4 | 3 . 36 , 0 . 0763 | 3 . 36 | 0 . 0724 | - | - |
Figure 7. Evolution of the m = 3 and m = 4 components of radial velocity [ | u m | , in units of c s0 = c s ( r ISCO )] at r = 1 05 . for a typical run, showing exponential growth of trapped p -mode amplitudes, for zero and non-zero α . The dashed and dotted lines show the exponential curves used to fit the mode growth rates. The reduction in growth rate and reduced saturation amplitude due to viscosity can be seen.
## 5 OVERSTABLE GLOBAL OSCILLATIONS IN TRUNCATED DISCS
We now investigate the behavior of trapped p -modes in a disc truncated at its inner edge by an impermeable wall. As noted before, such an inner disc edge mimics a magnetospheric boundary. We place the inner boundary at r ISCO and impose a reflecting boundary condition with u r = 0 (imposed by anti-symmetrizing u r across the boundary), and with all other variables fixed at their equilibrium values. In general, viscous forces induce a radial drift, which would lead to accumulation of mass at the inner boundary. We avoid this by initializing the surface density with a profile such that there is no radial drift (equation 8), given by
$$\Sigma \left ( r \right ) \in \left [ r ^ { - 2 } \left ( \frac { r - r _ { s } } { 3 r - r _ { s } } \right ) \right ] ^ { 1 / \Gamma }. \quad \quad ( 2 5 ) \quad B ;$$
Figure 8. Power spectra delineated by azimuthal number m in the linear phase (solid curves) and nonlinear phase (dashed curves) for trapped p -modes with zero (left) and non-zero (right) α . The linear phase power spectra have been multiplied by 10 10 and 10 4 (left and right, respectively) in order to compare them to the nonlinear phase power spectra.
This ensures that the azimuthally-averaged surface density does not change significantly as long as the perturbations to this equilibrium remain small (in principle this may be violated when the perturbation amplitudes become large and the modes are nonlinear). This initial surface density is modified by small ( δ Σ Σ / glyph[lessorsimilar] 10 -6 ) random perturbations which serve as seeds for the growth of p -modes. They lead to the growth of many modes with many different values of m , the frequencies and growth rates of which can be measured independently from a single run (the accuracy of this approach is examined in Section 5.4).
Each mode is assumed to have a complex frequency ω = ω r + iω i where ω r and ω i (both real) are the mode frequency and growth rate. The amplitude of radial velocity perturbations with azimuthal number m at radius r are determined numerically by computing
$$u _ { m } \left ( r, t \right ) = \frac { 1 } { 2 \pi } \int _ { 0 } ^ { 2 \pi } u _ { r } \left ( r, \phi, t \right ) e ^ { - i m \phi } d \phi. \quad \left ( 2 6 \right )$$
By fitting a straight line to log | u m | in the linear growth
c
= 0 05 Γ = 1
.
,
Figure 9. Surface density perturbations near the end of the linear phase (left) and in the nonlinear phase (right) with zero (top) and non-zero (bottom) α , showing the qualitative difference in saturation of trapped p -modes between the two cases.
phase, we can determine the growth rate of each mode. The frequencies are determined using a power spectrum delineated by azimuthal number m (equation 24). Some examples of analysis using these tools will be shown in the subsequent discussion of the results of the simulations.
Figure 7 shows the evolution of | u m | (for m = 3 and m = 4, evaluated at a representative radius) for a pair of typical runs. Initially there is a linear growth phase lasting 20 -40 orbits, in which the amplitudes grow exponentially. The fitted curves used to measure the growth rates are shown. After the departure of the mode amplitudes from simple exponential growth, we refer to the rest of the evolution as the nonlinear phase. Example power spectra are shown in Figure 8, showing the mode frequencies in both the linear and nonlinear phases.
## 5.1 Inviscid Case
We first perform simulations of an inviscid disc ( α = 0). Since the complex frequencies of the trapped p -modes can be compared to linear theory (Lai & Tsang 2009), these serve as test cases for verifying the robustness and accuracy of the simulations before we investigate the effects of viscosity (see Fu & Lai 2013 for a more detailed analysis of numerical simulations of trapped p -modes in inviscid discs). We choose four equations of state, with c s0 = 0 02 or . c s0 = 0 05 and . Γ = 1 or Γ = 4 / 3. The main results, numerically measured frequencies and growth rates of the modes, are presented in Table 2. For runs with c s0 = 0 02, the numerical growth . rates agree with those computed from linear theory to better than 10%, while for c s0 = 0 05 they agree to better than 5%, . with some cases in even better agreement. In all cases the numerical mode frequencies agree with linear theory to better than 1%, and these frequencies differ very little between the linear and nonlinear phases (see Figure 8). See Section 5.4 for a resolution test demonstrating that these results are converged. Having accurately captured the properties of the inviscid trapped p -modes, we move on to exploring the effects of viscosity in the next subsection.
## 5.2 Viscous Effects
We carry out simulations using the same parameters described above but now with non-zero viscosity, choosing α = 0 01 and . α = 0 05. Our analysis will focus on the mod-. ification of mode growth and saturation properties due to viscous effects. We separate this discussion into two parts based on the value of the polytropic index, as the effects of viscosity are different for Γ = 1 and Γ = 4 / 3.
The effect of viscosity on the mode growth rates are summarized in Table 2. We find qualitatively different results depending on the value of the polytropic index Γ. For Γ = 1, the growth rates of all modes are reduced compared to the inviscid case. For c s0 = 0 05, the difference in growth . rate due to viscosity δω ν = ω i -ω ( α =0) i , is approximately proportional to α , as expected theoretically (equation 12). This is not the case for c s0 = 0 02, for which such an ex-. trapolation of ω i based on the α = 0 01 case would imply . a negative value of ω i for α = 0 05 (and thus the modes . would not be overstable), which is not what we find. More importantly, the linear analysis of viscous effects predicts enhanced growth rates, rather than the diminished ones measured numerically. This indicates that the estimate of the viscous effect on the growth of global modes (equation 14) is inaccurate, perhaps due to dynamics of the corotation region which are neglected in equation (14).
The saturation behavior of the the modes in the nonlinear phase is also affected by viscosity. Figure 7 shows that while in the inviscid case, the various u m continue to grow slowly after the linear phase before saturating at larger amplitudes, with viscosity they decrease or level off after the linear phase, remaining somewhat steady at a relatively small amplitude. This effect can also be seen in Figure 9, which shows that when viscosity is included, the disc perturbations remain remarkably similar to the linear phase in appearance during the nonlinear phase, unlike the inviscid case for which the nonlinear phase looks qualitatively different than the linear phase. Therefore it appears that viscous forces ultimately suppress the mode amplitudes enough to prevent any significant nonlinear effects from occurring.
Our simulations with reflective inner boundary yield unexpected results for Γ = 4 / 3. Unlike the case of Γ = 1 for which the overstable p -modes behave similarly with nonzero α , albeit with reduced growth rates, when Γ = 4 / 3 we find no trace of the trapped p -modes. Instead, the power spectrum for these runs shows oscillations with frequencies approximately equal to m Ω ISCO . For these runs, the evolution of | u m | does not show clean exponential growth as the p -modes do, but rather a sharp initial rise to a significant amplitude followed by slow incoherent growth. The amplitude of these oscillations eventually becomes unphysically large with radial velocities much greater than c s0 . As the behavior of these unexpected modes is drastically different from the original p -modes, we forego detailed analysis of these runs. This highly unstable unexpected behavior may be the result of the combined effects of overstability due to GR (resulting from the smallness of κ/ Ω and increased q ) and due to viscosity (resulting from large A ), both of which enhance the growth rate as given in eq. (12). Since we are
Figure 10. Evolution of | u m | at r = 1 05 for Newtonian disc with . reflective inner boundary. The m = 1 and m = 2 components are shown, along with exponential fits used to determine growth rates, for two values of α . The growth rates increase with increasing α , in qualitative agreement with the theory of viscous overstability.
)
(
Figure 12. Power spectra for lowest m boundary modes in linear and nonlinear phases. In the linear phase, each mode corresponds to a single sharp frequency peak close to m Ω ISCO , while in the nonlinear phase these peaks are broader and exhibit splittings, but retain powers concentrated in frequencies similar to those found in the linear phase.
)
0
)
s
(
c
r
u
in
r
(
y
|
|
Figure 11. Linear phase power spectrum for viscosity-driven boundary modes in Newtonian disc.
interested in the viscous effect, in Section 5.3 we will isolate it by modifying the numerical setup to simulate a disc with a pure Newtonian 1 /r potential, to see if this behavior persists in a simple way that allows for a concrete analysis.
## 5.3 Viscosity-Driven Boundary Modes in Newtonian Potential
We wish to determine whether or not the growth of the new modes which overwhelm the p -modes is due to a purely viscous effect which can operate in the absence of GR effects. To this end, we perform simulations using the same setup as described at the beginning of this section, but replace the pseudo-Newtonian gravitational potential with a pure Newtonian potential, Φ = -GM/r . Under this potential there is no longer an ISCO, so the location of the inner boundary no longer has any physical significance, and we simply denote it as r in . The corresponding boundary condition, as in the beginning of this section, is reflective with
Figure 13. Radial velocity of boundary modes in Newtonian discs in the linear and nonlinear phase.
u r = 0. We choose four sets of parameters, using c s0 = 0 1 . and (Γ , α ) = (1 , 0) , (1 0 01) , . , (4 / 3 0) , , (4 / , 3 0 01). Due to the . larger sound speed used, and therefore larger radial wavelengths, we reduce our resolution to N r × N φ = 512 × 128.
We observe no growing oscillations for any of these runs except for (Γ , α ) = (4 / 3 0 01), confirming that the anoma-, . lous modes observing in Section 5.2 can be attributed purely to viscous effects (that is, requires α > 0 and Γ > 1) and do not depend on GR effects. We therefore performed an additional run with (Γ , α ) = (4 3 0 02) to probe their de-/ , . pendence on α . Figure 10 shows the evolution of the lowest m mode amplitudes (represented by | u m | at r = 1 05), show-. ing their exponential growth at a rate which increases with α (see Table 3 for complete list of frequencies and growth rates), as well as their relatively flat saturation amplitude
Table 3. Frequencies and growth rates of boundary modes.
| | m | ω r | ω i |
|------------|-----|--------|----------|
| α = 0 . 01 | 1 | 1 . 05 | 0 . 0389 |
| α = 0 . 01 | 2 | 2 . 00 | 0 . 0471 |
| α = 0 . 01 | 3 | 2 . 99 | 0 . 0488 |
| α = 0 . 01 | 4 | 3 . 98 | 0 . 0488 |
| α = 0 . 02 | 1 | 1 . 01 | 0 . 0519 |
| α = 0 . 02 | 2 | 2 . 02 | 0 . 0692 |
| α = 0 . 02 | 3 | 3 . 01 | 0 . 0743 |
| α = 0 . 02 | 3 | 4 . 00 | 0 . 0744 |
Table 4. Frequency and growth rate of m = 2 trapped p -mode for ( c s0 , Γ) = (0 02 1) with initial . , m = 2 perturbation for half, standard and double resolution runs.
| α | ( ω r , ω i ) half | ( ω r , ω i ) standard | ( ω r , ω i ) double |
|--------|----------------------|--------------------------|------------------------|
| 0 | 1 . 79 , 0 . 0396 | 1 . 78 , 0 . 0419 | 1 . 78 , 0 . 0423 |
| 0 . 01 | 1 . 77 , 0 . 0301 | 1 . 76 , 0 . 0315 | 1 . 76 , 0 . 0317 |
consistent with the the viscous effect on mode saturation seen in Section 5.2. Figure 11 shows the power spectrum for these modes, indicating that they are global, with power in discrete frequencies across a range of radii. This justifies the use of eq. (24) to produce a radially integrated m -delineated power spectrum, as shown in Figure 12. From this we see that in the nonlinear phase, the modes retain approximately the same frequencies as in the linear phase, but with broader power spectra peaks and the presence of other sub-dominant peaks at new frequencies. The morphology of the two phases is shown in Figure 13. Note that the linear phase bears some resemblance to that of the p -modes (compare to Figure 9), while the nonlinear phase appears very different, having a more complex appearance.
These modes may be related to the interface modes studied by Tsang & Lai (2009b) and Fu & Lai (2012), who considered a Keplerian disc in contact with a uniformly rotating cylinder (for example, a magnetosphere), which is similar to the setup we consider. Their results indicate that modes with frequencies close to m Ω in can be overstable due to the corotation resonance effect. In our case, they may not be intrinsically overstable but are only driven by viscous forces which can act in phase with the oscillations, by having a sufficiently large A parameter. A qualitative description of this phenomenon is as follows. In our simulation, the disc is initialized with a surface density for which viscous forces induce zero radial drift. However, a small perturbation to this equilibrium leads to a non-zero u r , which leads to a radial oscillation at the local epicyclic frequency (which is simply equal to the orbital frequency Ω in a Newtonian potential). When material close to the inner solid boundary is pushed inward due to this oscillation, it is compressed, leading to a stronger viscous forces, which act as an additional restoring force (in addition to gravity which is producing the epicyclic motion). In this way, the restoring force increases with the amplitude of the oscillation, leading to overstability.
## 5.4 Resolution Study
As in Section 4.3, we perform two resolution test runs with half (512 × 128) and double (2048 × 512) the standard resolution, for the case ( c s0 , Γ , α ) = (0 02 1 0). . , , These runs also differ from the standard run in that the initial density perturbations are proportional to cos (2 φ ), so that the m = 2 mode is isolated. This allows us to test the accuracy of measuring mode frequencies and growth rates for modes of multiple m from an initially random perturbation, as is done in our standard runs. The results are shown in Table 4. We see that with increasing resolution, the numerically measured growth rate converges on the result from linear theory, agreeing to better than 1% for the double resolution run. Additionally, we see that a standard resolution run with a pure m = 2 initial perturbation reproduces the growth rate of this mode more accurately than a run with random perturbations, as used in Section 5.2. However, the difference is small enough to justify the use of the random perturbation approach, which allows us to capture multiple modes from single runs. We note that since this resolution test was performed for our smallest sound speed, for which the modes have the smallest radial wavelengths, these tests provide a lower limit on our accuracy, as we expect the longer wavelength modes associated with our larger sound speed to be better resolved by the same resolution. We also perform the same test using ( c s0 , Γ , α ) = (0 02 1 0) to ensure that vis-. , , cous effects are not sensitive to resolution. For this case we achieve similarly excellent convergence of growth rate with increasing resolution. Additionally, the discrepancy between the run with random perturbations and the one with m = 2 perturbations is smaller than the inviscid case.
## 6 DISCUSSION
We have carried out viscous hydrodynamical simulations of black hole (BH) accretion discs to determine how viscosity may drive global oscillation modes in the disc. Analytical considerations (see Section 2) indicate that, depending on the viscosity law (particularly how viscosity scales with density) and the background disc flow properties, various global oscillations with different azimuthal wavenumbers may be excited. We considered two types of accretion flows. The first involves discs with transonic radial inflow across the innermost stable circular orbit (ISCO) of the BH, and the second involves discs truncated at an inner edge and with a free outer boundary at large distance.
Our simulations on transonic discs (Section 4) follow previous works in one dimension (Milsom & Taam 1996) and in two or three dimensions (O'Neill et al. 2009; Chan 2009). We show that, depending on the viscosity parameter α , global oscillations with different azimuthal wavenumbers m , and with frequencies close to κ max , m Ω ISCO or their linear combinations (where κ max and Ω ISCO are the maximum radial epicyclic frequency and the orbital frequency at the ISCO, respectively) are excited. In general, as α increases above a critical value α crit1 ( ∼ 0 1 . -0 25 . for isothermal discs), non-axisymmetric modes (typically m = 4) first develop, then axisymmetric modes develop above a second critical value α crit2 ( ∼ 0 5). . We propose that these modes are the result of viscous overstability, possibly combined with an effect related to instability of the sonic point, although the theory of the latter is not strictly applicable in our simulations.
Our simulations on truncated discs (Section 5) extend previous semi-analytic works on overstable inertial-acoustic modes ( -modes) p driven by corotation resonance in nonmagnetic discs (Lai & Tsang 2009; Tsang & Lai 2008, 2009b; Horak & Lai 2013) and magnetic discs (Fu & Lai 2009; 2011), on magnetosphere-disc interface modes (Tsang & Lai 2009a; Fu & Lai 2012), and numerical studies of nonlinear p -modes (Fu & Lai 2013). We find that the growth rates of overstable p -modes are reduced by viscosity in isothermal discs, while for other equations of state (and corresponding viscosity laws), they are suppressed due to the excitation of a
different class of modes with frequencies ω ≈ m Ω , which in are related to magnetosphere-disc interface modes.
Obviously, our 2D hydrodynamical disc models do not capture various complexities (e.g., magnetic field, turbulence and radiation) associated with real BH accretion discs. Much progress in numerical General Relativistic Magnetohydrodynamic (GRMHD) simulations of BH accretion discs has been made in the past decade, but much work remains to understand the complex phenomenology of BH X-ray binaries (e.g., Remillard & McClintock 2006; Done et al. 2007; Belloni et al. 2012). Several recent simulations have revealed quasi-periodic variabilities of various fluid variables, but the connection of these variabilities to the observed HFQPOs is far from clear (e.g., Henisey et al. 2009; ONeill et al. 2011; Dolence et al. 2012; McKinney et al. 2012; Shcherbakov & McKinney 2013). Our simulations, although based on highly simplified disc models, demonstrate that under appropriate conditions, various global oscillation modes can grow to nonlinear amplitudes with well-defined frequencies. We emphasize that our height-integrated treatment cannot capture modes with dependence on vertical structure ( g -modes and c -modes), which may be present in 3D simulations. The properties and excitation of the p -modes considered in this paper may also be modified in a fully 3D approach.
It is of interest to compare the frequencies of our simulated modes to those of observed HFQPOs in X-ray binaries (Remillard & McClintock 2006; Belloni et al. 2012), particulary the three systems whose BH spins are constrained using the continuum-fitting method (see Narayan & McClintock 2012 and references therein). These considerations indicate that the frequencies of the HFQPOs are smaller than the ISCO frequency, therefore they are unlikely to be explained by any non-axisymmetric modes with ω glyph[greaterorsimilar] Ω ISCO . However, this does not rule out the axisymmetric κ max modes which have lower frequencies. We note that we have not addressed the mechanism by which disc oscillations may manifest as variabilities in X-ray flux, which affects the observability (or lack of observability) of any oscillations that occur in the disc.
Although our simulations cannot be directly compared with observations, some tentative conclusions can be drawn. The thermal (high-soft) state of BH X-ray binaries may be approximately described by the transonic disc model (e.g. Done et al. 2007). Our simulations show that such a disc does not allow trapping of p -modes, but can excite global κ max -waves or m Ω ISCO -waves when the viscosity parameter α is sufficiently large. Since no significant HFQPOs are observed in the thermal state of X-ray binaries, our simulations provide an upper limit on the effective disc viscosity parameter of α glyph[lessorsimilar] 0 5. .
The physical nature of the intermediate state (or 'steep power law state') of BH X-ray binaries, when HFQPOs are observed, is currently uncertain (Done et al. 2007; Oda et al. 2010). Since episodic jets are produced in this state, large-scale magnetic fields likely play an important role (e.g., Tagger & Varniere 2006; Yuan et al. 2009; McKinney et al. 2012). When magnetic fields advect inwards in the accretion disc and accumulate around the BH (e.g. BisnovatyiKogan & Ruzmaikin 1974, 1976; Igumenshchev et al. 2003; Rothstein & Lovelace 2008), a magnetosphere may form. The truncated disc model studied in this paper may mimic the disc outside the magnetosphere. Our calculations show that overstable p -modes (driven by corotation resonance) can be trapped in the inner disc, but their growth rates are typically reduced by disc viscosity. More importantly, with increasing α and more general equation of state, a new series of global waves can be excited at the disc boundary (see Tsang & Lai 2009a; Fu & Lai 2012; McKinney et al. 2012).
## ACKNOWLEDGMENTS
We thank David Tsang for useful discussion. This work has been supported in part by NSF grants AST-1008245, 1211061, and NASA grant NNX12AF85G. J.H. acknowledges support from grants LH 14049 and M100031202.
## REFERENCES
Abramowicz, M. A., Klu´ zniak, W., 2001, A&A, 374, L19 Abramowicz, M. A., Klu´ zniak, W., Bursa, M., Hor´k, J., Rebusco, a
- P., T¨r¨k, G., 2007, RevMexAA, 27, 8 o o
- Belloni, T. M., Sanna, A., M´ndez, M., 2012, MNRAS, 426, 1701 e Bisnovatyi-Kogan, G. S., Ruzmaikin, A. A., 1974, Ap&SS, 28, 45 Bisnovatyi-Kogan, G. S., Ruzmaikin, A. A., 1976, Ap&SS, 42, 401 Blaes, O. M., Arras, P., Fragile, P. C., 2006, MNRAS, 369, 1235 Chan, C.-K., 2009, ApJ, 704, 68
- Dolence, J. C., Gammie, C. F., Shiokawa, H., Noble, S. C., 2012, ApJ, 746, L10
- Done, C., Gierli´ski, M., Kubota, A., 2007, A&AR, 15, 1 n
- Fu, W., Lai, D., 2009, ApJ, 690, 1386
- Fu, W., Lai, D., 2011, MNRAS, 410, 399
- Fu, W., Lai, D., 2012, MNRAS, 423, 831
- Fu, W., Lai, D., 2013, MNRAS, 431, 3697
- Henisey, K. B., Blaes, O. M., Fragile, P. C., Ferreira, B. T., 2009, ApJ, 706, 705
- Hor´ ak, J., Lai, D., 2013, MNRAS, 434, 2761
- Igumenshchev, I. V., Narayan, R., Abramowicz, M. A., 2003, ApJ, 592, 1042
- Kato, S., 1978, MNRAS, 185, 629
- Kato, S., Honma, F. & Matsumoto, R. 1988a, MNRAS, 231, 37
- Kato, S., Honma, F. & Matsumoto, R. 1988b, PASJ, 40, 709
- Kato, S., Wu, X.-B., Yang, L.-T. & Yang, Z.-L. 1993, MNRAS,
## 260, 317
- Kato, S., 2001, PASJ, 53, 1
- Lai, D., Tsang, D., 2009, MNRAS, 393, 979
- Lai, D., Fu, W., Tsang, D., Hor´k, J., Yu, C., 2013, IAUS, 290, a 57
- Latter, H. N., Ogilvie, G. I., 2010, Icarus, 210, 318
- Lyubarskij, Y. E., Postnov, K. A., Prokhorov, M. E., 1994, MNRAS, 266, 583
- Mao, S. A., Psaltis, D., Milsom, J. A., 2009, ApJ, 703, 717
- McKinney, J. C., Tchekhovskoy, A., Blandford, R. D., 2012, MNRAS, 423, 3083
- Mignone, A., Bodo, G., Massaglia, S., Matsakos, T., Tesileanu, O., Zanni, C., Ferrari, A., 2007, ApJS, 170, 228
- Milsom, J. A., Taam, R. E., 1996, MNRAS, 283, 919
- Narayan, R., McClintock, J. E., 2012, MNRAS, 419, L69
- O'Neill, S. M., Reynolds, C. S., Miller, M. C., 2009, ApJ, 693, 1100
- O'Neill, S. M., Reynolds, C. S., Miller, M. C., Sorathia, K. A., 2011, ApJ, 736, 107
- Oda, H., Machida, M., Nakamura, K. E., Matsumoto, R., 2010, ApJ, 712, 639
- Ortega-Rodr´ ıguez, M., Silbergleit, A., Wagoner, R., 2008, GApFD, 102, 75
- Ortega-Rodr´ ıguez, M., Wagoner, R., 2000, ApJ, 537, 922
- Paczy´ski, B., Witta, P. J., 1980, A&A, 88, 23 n
Papaloizou, J. C. B., Lin, D. N. C., 1988, ApJ, 331, 838
Rein, H., Latter, H. N., 2013, MNRAS, 431, 145 Remillard, R. A., McClintock, J. E., 2006, ARA&A, 44, 49 Reynolds, C. S., Miller, M. C., 2009, ApJ, 692, 869 Rezzolla, L., Yoshida, S., Maccarone, T. J., Zanotti, O., 2003, MNRAS, 344, L37 Rothstein, D. M., Lovelace, R. V. E., 2008, ApJ, 677, 1221 Schmidt, J., Salo, H., Spahn, F., Petzschmann, O., 2001, Icarus,
153, 316
Schmit, U., Tscharnuter, W. M., 1995, Icarus, 115, 304 Schnittman, J. D., Bertschinger, E., 2004, ApJ, 606, 1098 Shakura, N. I., Sunyaev, R. A., 1973, A&A, 24, 337 Shcherbakov, R. V., McKinney, J. C., 2013, ApJL, 774, L22 Stella, L., Vietri, M., Morsink, S. M., 1999, ApJ, 524, L63 Tagger, M., Varni`re, P., 2006, ApJ, 652, 1457 e Tsang, D., Lai, D., 2008, MNRAS, 387, 446 Tsang, D., Lai, D., 2009a, MNRAS, 396, 589 Tsang, D., Lai, D., 2009b, MNRAS, 400, 470 de Val-Borro, M. et al., 2006, MNRAS, 370, 529
Wellons, S., Zhu, Y., Psaltis, D., Narayan, R., McClintock, J. E., 2014, ApJ, 785, 142
Yuan, F., Lin, J., Wu, K., Ho, L. C. 2009, MNRAS, 395, 2183
## APPENDIX A: WAVE EQUATION UP TO FIRST ORDER IN VISCOSITY AND RADIAL INFALL VELOCITY
In this appendix we summarize the procedure of deriving the equations (9)-(11). The linearly perturbed continuity equation (1) and the Navier-Stokes equation (2) in the polar coordinates ( r, φ ) read
$$- \, i \tilde { \omega } \delta \Sigma + \frac { 1 } { r } \partial _ { r } \left ( r \Sigma \delta u _ { r } \right ) + \frac { \text{im} } { r } \Sigma \delta u _ { \phi } = - \, \frac { 1 } { r } \partial _ { r } \left ( r u ^ { r } \delta \Sigma \right ) \equiv \delta \dot { \Sigma } \, \quad \text{Fina} \\ ( \text{A1} ) \quad \text{cont}$$
$$- \mathrm i \tilde { \omega } \delta u _ { r } - 2 \Omega \delta u _ { \phi } + \delta h _ {, r } = - \partial _ { r } \left ( u _ { r } \delta u _ { r } \right ) + \delta \left ( \frac { 1 } { \Sigma } \mathcal { F } _ { r } ^ { \text{visc} } \right ) \equiv \delta \mathring { u } _ { r }, \quad \text{tions} \\ ( A 2 ) \quad \text{ of sh} \\.$$
$$- i \tilde { \omega } \delta u _ { \phi } + \frac { \kappa ^ { 2 } } { 2 \Omega } \delta u _ { r } + \frac { i m } { r } \delta h = - \frac { u _ { r } } { r } \partial _ { r } \left ( r \delta u _ { \phi } \right ) + \delta \left ( \frac { 1 } { \Sigma } \mathcal { F } _ { \phi } ^ { \text{visc} } \right ) \equiv \delta \dot { u } _ { \phi }, \quad \text{regions} \\ \tau u \, - \, \dots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \colon$$
The equations were arranged in such a way that the coefficients on the LHS do not depend on the viscosity, while the RHS terms, denoted by δ ˙ Σ, δu ˙ r and δu ˙ φ , are at least linear in the viscosity η . The first RHS term in each equation describes the effect of the radial inflow in the stationary case, the second terms in the last two equations are contributions due to the perturbation of the viscous force F visc = ∇ · σ .
In what follows we would like to express the RHS terms δu ˙ r , δu ˙ φ and δ ˙ Σ in terms of the velocity perturbation δu r , δu φ and enthalpy δh . To simplify the analysis we will further work in the leading order of the WKBJ approximation. Therefore, we assume that the radial wavelength λ r = 2 π/k r of the perturbations is much smaller than the radius r . More specifically, we assume that k r r = O ( r/H ), where H is the disc semi-thickness. In thin discs, this assumption is satisfied everywhere except for small regions close to the Lindblad and corotation resonances. In the leading order in the ratio r/H , we obtain
$$\delta \dot { \Sigma } \ \approx \ - u _ { r } \frac { \Sigma } { c _ { s } ^ { 2 } } \frac { d } { d r } \delta h _ {, r } \quad \ \ ( A 4 )$$
$$2 0 0 3, \quad \delta \dot { u } _ { r } \ \approx \ \left [ - u _ { r } \frac { d } { d r } + \frac { 4 } { 3 } \nu \frac { d ^ { 2 } } { d r ^ { 2 } } \right ] \delta u _ { r },$$
$$\colon \text{aru}, \quad \delta \dot { u } _ { \phi } \ \approx \ \left [ - u _ { r } \frac { d } { d r } + \nu \frac { d ^ { 2 } } { d r } \right ] \dot { \delta u } _ { \phi } - \nu q A \frac { \Omega } { c _ { 8 } ^ { 2 } } \frac { d } { d r } \delta h. \quad \ \ ( A 6 )$$
In all three equations, the first terms are proportional to u r and describe the effects of the radial inflow. The second terms in equation (A5) and (A6) are the contributions of the perturbed visous force due to velocity field connected to the wave. Finally, the third term of equation (A6) results from the change of the viscous force due to perturbation of the dynamic viscosity coefficient η .
By substituting expressions (A5) and (A6) into the equations (A2) and (A3) and expanding the velocity perturbation δ u in powers of ν ,
$$\mathfrak { d } u _ { r } = \mathfrak { d } u _ { r } ^ { 0 } + \nu \mathfrak { d } u _ { r } ^ { 1 } + \dots, \ \mathfrak { d } u _ { \phi } = \mathfrak { d } u _ { \phi } ^ { 0 } + \nu \mathfrak { d } u _ { \phi } ^ { 1 } + \dots, \ ( A 7 )$$
we may solve perturbativelly these equations for δ u in terms of the enthalpy perturbation δh . Up to the first order in ν (and u r ), we obtain
$$& ( \text{and} \, u _ { r } ), \, \text{we obtain} \\ & \delta u _ { r } \ = \ \frac { i } { D } \left ( \tilde { \omega } \frac { d } { d r } - \frac { 2 m \Omega } { r } \right ) \delta h - \frac { u _ { r } } { D ^ { 2 } } \left ( \kappa ^ { 2 } + \tilde { \omega } ^ { 2 } \right ) \frac { d ^ { 2 } } { d r ^ { 2 } } \delta h + \\ & \frac { \nu } { D ^ { 2 } } \left [ \left ( \kappa ^ { 2 } + \frac { 4 } { 3 } \tilde { \omega } ^ { 2 } \right ) \frac { d ^ { 3 } } { d r ^ { 3 } } - 2 q A \Omega ^ { 2 } \frac { D } { c _ { s } ^ { 2 } } \frac { d } { d r } \right ] \delta h + \mathcal { O } ( \nu ^ { 2 } ), \\ \text{iving} \\ \text{unity} \\ \text{qu} \\ \text{qu} \\ \text{per} \\ \text{all} \\ \text{$\lambda$} \\ \Phi$} ^ { 2 } & \text{Finally, by substituting these expressions into the perturbed}$$
Finally, by substituting these expressions into the perturbed continuity equation (A1), we recover the desired equation (9) with operators ˆ L 0 and ˆ L 1 in and ˆ L 1 v . given by equations (5) and (10) and (11). We remind the reader that these equations are valid only in the leading order in both ( H/r )- ratio, radial drift velocity u r and viscosity ν under the assumpion of short radial wavelength (as compared to the radial coordinate r ). Hence, in principle, it cannot be applied in the regions of the Lindblad and corotation resonances. A proper description of those effects would require keeping also the singular terms in equations (A4)-(A8).
## APPENDIX B: GROWTH RATES OF MODES
We would like to find the global change of the eigenfrequencies of p -modes trapped in the inner disc. In the following, we will show that it is given by some proper radial average of the Kato (1978) local growth rate over the propagation region. Introducing ψ = S -1 / 2 δh with S = D/ r ( Σ), equation (9) takes the form
$$\left [ \frac { d ^ { 2 } } { d r ^ { 2 } } - V ( r ) \right ] \psi + S ^ { - 1 / 2 } \hat { L } ^ { 1 } \left ( S ^ { 1 / 2 } \psi \right ) = 0, \quad ( B 1 )$$
where
$$\overset { \dots } { \text{lad} } \quad V ( r ) = \frac { D } { c _ { 8 } ^ { 2 } } + \frac { 2 m \Omega } { r \tilde { \omega } } \frac { d } { d r } \ln \left ( \frac { \Omega \Sigma } { D } \right ) + S ^ { 1 / 2 } \frac { d ^ { 2 } } { d r ^ { 2 } } S ^ { - 1 / 2 } + \frac { m ^ { 2 } } { r ^ { 2 } }, \ ( B 2 )$$
and ˆ L 1 is either ˆ L 1 v , or ˆ L 1 in or their sum. Assuming that the region of interest does not contain the resonances, the potential V ( r ) can be approximated by just the first term because the others are by factor of order ( H/r ) 2 smaller. If we express ψ and D as ψ = ψ 0 + ψ 1 and D = D 0 + D 1 (where ψ 1 and D 1 are of order ν , note that D 1 = -2˜ ω ω 0 1 ), we obtain in the zero order equation,
$$\left [ \frac { d ^ { 2 } } { d r ^ { 2 } } - \frac { D _ { 0 } } { c _ { s } ^ { 2 } } \right ] \psi _ { 0 } = 0, \text{ \quad \quad } \text{(B3)}$$
and the first order equation,
$$\left [ \frac { d ^ { 2 } } { d r ^ { 2 } } - \frac { D _ { 0 } } { c _ { s } ^ { 2 } } \right ] \psi _ { 1 } - \frac { D _ { 1 } } { c _ { s } ^ { 2 } } \psi _ { 0 } + S ^ { - 1 / 2 } \hat { L } ^ { 1 } \left ( S ^ { 1 / 2 } \psi _ { 0 } \right ) = 0. \ \ ( B 4 )$$
Multiplying by ψ ∗ 0 and integrating between radii r 1 and r 2 (trapping range of the mode), we obtain
$$\vec { \ m r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r l } & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & \vec { \ m r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r rr r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r rr r r rr r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r rr r r r l r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r c r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r rs r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r rrs r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R K R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R L R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R r R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R R r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r tr r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r s r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r r
r & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & & &$$
The integral on the RHS can be written as (after two integrations by parts)
$$\left [ \psi _ { 1 } \frac { d \psi _ { 0 } ^ { * } } { d r } - \psi _ { 0 } ^ { * } \frac { d \psi _ { 1 } } { d r } \right ] _ { r _ { 1 } } ^ { r _ { 2 } } - \int _ { r _ { 1 } } ^ { r _ { 2 } } \psi _ { 1 } \left [ \frac { d ^ { 2 } } { d r ^ { 2 } } - \frac { D _ { 0 } } { c _ { 8 } ^ { 2 } } \right ] \psi _ { 0 } ^ { * } d r. \ \ ( B 6 )$$
Assuming that ω 0 is real, the first integral vanishes. The second term is a factor O ( H/r ) smaller than the other integrals in equation (B5). Then, neglecting I , we obtain
$$\omega _ { 1 } = - \frac { \int \psi _ { 0 } ^ { * } \left ( S ^ { - 1 / 2 } \hat { L } ^ { 1 } S ^ { 1 / 2 } \psi _ { 0 } \right ) \mathrm d r } { 2 \int \left ( \tilde { \omega } _ { 0 } / c _ { 8 } ^ { 2 } \right ) | \psi _ { 0 } | ^ { 2 } \mathrm d r }. \quad \text{ (B7)}$$
An approximate WKBJ solution of equation (B3), valid up to the second order, is
$$\ddot { \psi } _ { 0 } & = \frac { 1 } { \sqrt { k } } \left [ \mathcal { A } _ { - } \exp \left ( - \mathfrak { i } \int ^ { r } k d r \right ) + \mathcal { A } _ { + } \exp \left ( \mathfrak { i } \int ^ { r } k d r \right ) \right ], \\ k & = \sqrt { - D } / c _ { s }, \\ \text{for which } \left | \mathfrak { d } \right | ^ { 2 } \, = \, \left ( \mathfrak { l } \, \mathcal { A } \, \left | ^ { 2 } + \mathcal { A } \right | ^ { 2 } \right \rangle / k \, \text{Favalatin} \, \text{the two} }$$
for which | ψ 0 | 2 = ( |A -| 2 + |A -| 2 ) /k . Evaluating the two integrals in equation (B7) for this solution, we find that
$$\omega _ { 1 } = \frac { \int \delta \omega ( r ) w ( r ) d r } { \int w ( r ) d r }, \ \ w ( r ) = \frac { 1 } { c _ { s } } \frac { \tilde { \omega } _ { 0 } } { \sqrt { - D } }. \quad ( { \mathbf B } 9 )$$
Hence, the global change of the frequency of the mode is just the averaged local changes δω v ( r ) or δω in ( r ) of equations (12) and (13) with the weight function w r ( ).
Although expression (B9) does not apply close to the Lindblad resonances (LRs), we believe that the first-order LRs can be included as well, because the surrounding regions give negligible contributions to both integrals [ w r ( ) is integrable close to the first-order LRs]. Therefore, the integrals can be extended to the propagation regions of the oscillations. The lower integral however diverges for the secondorder LR (at maxima of Ω + κ/m ) and more sophisticated analysis is needed in those cases.
|
10.1093/mnras/stu2122
|
[
"Ryan Miranda",
"Jiri Horak",
"Dong Lai"
] |
2014-08-04T20:00:15+00:00
|
2015-04-13T18:38:50+00:00
|
[
"astro-ph.HE"
] |
Viscous Driving of Global Oscillations in Accretion Discs Around Black Holes
|
We examine the role played by viscosity in the excitation of global
oscillation modes (both axisymmetric and non-axisymmetric) in accretion discs
around black holes using two-dimensional hydrodynamic simulations. The
turbulent viscosity is modeled by the $\alpha$-ansatz, with different equations
of state. We consider both discs with transonic radial inflows across the
innermost stable circular orbit, and stationary discs truncated by a reflecting
wall at their inner edge, representing a magnetosphere. In transonic discs,
viscosity can excite several types of global oscillation modes. These modes are
either axisymmetric with frequencies close to multiples of the maximum radial
epicyclic frequency $\kappa_\mathrm{max}$, non-axisymmetric with frequencies
close to multiples of of the innermost stable orbit frequency
$\Omega_\mathrm{ISCO}$, or hybrid modes whose frequencies are linear
combinations of these two frequencies. Small values of the viscosity parameter
$\alpha$ primarily produce non-axisymmetric modes, while axisymmetric modes
become dominant for large $\alpha$. The excitation of these modes may be
related to an instability of the sonic point, at which the radial infall speed
is equal to the sound speed of the gas. In discs with a reflective inner
boundary, we explore the effect of viscosity on trapped $p$-modes which are
intrinsically overstable due to the corotation resonance effect. The effect of
viscosity is either to reduce the growth rates of these modes, or to completely
suppress them and excite a new class of higher frequency modes. The latter
requires that the dynamic viscosity scales positively with the disc surface
density, indicating that it is a result of the classic viscous overstability
effect.
|
1408.0803v1
|
## Determining Majorana Nature of Neutrino from Nucleon Decays and n -¯ oscillations n
K.S. Babu a ∗ and Rabindra N. Mohapatra b †
a Department of Physics, Oklahoma State University, Stillwater, Oklahoma 74078, USA
b Maryland Center for Fundamental Physics and Department of Physics University of Maryland, College Park, MD 20742, USA
## Abstract
We show that discovery of baryon number violation in two processes with at least one obeying the selection rule ∆( B -L ) = ± 2 can determine the Majorana character of neutrinos. Thus observing p → e + π 0 and n → e -π 0 decays, or p → e + π 0 and n -¯ oscillations, or n n → e -π + and n -¯ oscillations would establish that neutrinos n are Majorana particles. We discuss this in a model-independent effective operator approach.
An outstanding question in particle physics today is whether neutrinos are their own antiparticles (Majorana fermions) or not. This can be settled most directly by the detection of a nonzero signal in searches for neutrinoless double beta decay [1], which would establish the Majorana character of the electron type neutrino. If the neutrino masses exhibit a normal and non-degenerate mass hierarchy, however, the prospects are bleak for observing this decay in the near future as it would require experiments with much higher sensitivity than currently available. It may therefore be of interest to search for alternative ways to probe the Majorana character of neutrinos, which plays a pivotal role in the understanding of the physics behind neutrino masses.
In this Letter we suggest that discovery of baryon number violation in two processes, with at least one obeying the selection rule ∆( B -L ) = ± 2 for the change in the number of baryons minus leptons, can provide such an alternative. SuperKamiokane and other deep
∗ Email: [email protected]
† Email: [email protected]
underground experiments on the horizon are sensitive to nucleon decays of the type p → e + π 0 which obeys the selection rule ∆( B -L ) = 0, as well as n → e -π + , obeying ∆( B -L ) = -2 selection rule [2]. These experiments are also sensitive to matter disintegration caused by neutron-antineutron ( n -¯) transition in nuclear matter, a ∆( n B -L ) = -2 process. As we shall see, discovery of any two of these (or certain related) processes would establish the Majorana character of the neutrino, which is intrinsically a ∆( B -L ) = ± 2 effect. To arrive at this conclusion we rely on a model-independent effective operator analysis. This connection between baryon number violation and Majorana character of the neutrino should provide added impetus to higher sensitivity nucleon decay searches at the proposed LBNF facility [3] and the HyperKamiokande experiment [4]. This should also provide added motivation for a high sensitivity search for n -¯ oscillation with free neutrons which is under consideration n at the European Spallation Source (ESS) currently [5].
We focus on the three baryon number violating processes listed below:
$$( i ) \ p \to e ^ { + } \pi ^ { 0 } ; \quad ( i i ) \ n \to e ^ { - } \pi ^ { + } ; \quad ( i i i ) \ n - \bar { n } \text{ oscillations }.$$
Process ( ) obeys ∆( i B -L ) = 0, while processes ( ii ) and ( iii ) obey ∆( B -L ) = -2. To establish the Majorana nature of the neutrino from here, at least two of these processes will have to be observed. While discovery of nucleon decay in itself will be profound, once a certain decay mode of the nucleon is seen, it is perhaps not unreasonable to expect additional decay modes as well. This is what we assume in getting to our proof. But before proceeding, we note that the electron and the positron in the final states of Eq. (1) can be replaced by µ ± . The discovery of p → µ + π 0 and n → µ -π + will establish the Majorana character of the muon type neutrino, something not possible in neutrinoless double beta decay owing to kinematic limitations. Furthermore, the pions Eq. (1) may be replaced by other mesons, including kaon, ρ , ω etc, in which case our proof for the Majorana nature of neutrinos will still go through, aided by the dressing of diagrams by the weak W ± gauge boson which can change the strangeness quantum number. Thus the more general baryon number violating processes relevant for establishing Majorana nature of neutrinos are p → glyph[lscript] + M 0 ; n → glyph[lscript] -M + ; and n -¯ oscillations, with n glyph[lscript] being e or µ and M being any light meson.
In Fig. 1 we depict the Feynman diagrams responsible for the B -violating processes listed in Eq. (1). In a model-independent approach, we can characterize the p → e + π 0 decay by the dimension six effective Lagrangian [6]
$$\mathcal { L } ^ { \text{eff} } ( p \rightarrow e ^ { + } \pi ^ { 0 } ) \ = \ & \ \frac { 1 } { \Lambda _ { p } ^ { 2 } } \left [ c _ { 1 } \left ( u _ { R } d _ { R } \right ) ( u _ { L } e _ { L } ) + c _ { 2 } \left ( u _ { L } d _ { L } \right ) ( u _ { R } e _ { R } ) \\ & + c _ { \left \{ 3, 4 \right \} } \left ( u _ { L } d _ { L } \right ) \left ( ( u _ { L } e _ { L } ) + c _ { 5 } \left ( u _ { R } d _ { R } \right ) ( u _ { R } e _ { R } ) \right ]$$
Figure 1: Diagrams responsible for (a) p → e + π 0 decay; (b) n → e -π + decay; and (c) n -¯ n oscillations.
suppressed by two powers of the cutoff Λ . p Here color contraction, which is unique, should be understood. Similarly the decay n → e -π + arises from the dimension seven effective Lagrangian given by [7,8]
$$& \mathcal { L } ^ { \text{eff} } ( n \to e ^ { - } \pi ^ { + } ) = \frac { \langle H ^ { 0 } \rangle } { \Lambda _ { n } ^ { 3 } } \left [ \hat { c } _ { 5 } \left ( d _ { R } d _ { R } \right ) ( d _ { L } e ^ { c } _ { L } ) + \hat { c } _ { 6 } \left ( d _ { R } d _ { R } \right ) ( d ^ { c } _ { L } e _ { L } ) ^ { * } \right ] \\ & + \quad \frac { 1 } { \Lambda _ { n } ^ { 3 } } \left [ \hat { c } _ { 7 } \left ( \bar { c } _ { L } \gamma _ { \mu } d _ { L } \right ) ( d ^ { c } _ { L } \partial ^ { \mu } d ^ { c } _ { L } ) ^ { * } + \hat { c } _ { 8 } \left ( d ^ { c } _ { L } \partial _ { \mu } e _ { L } \right ) ^ { * } ( \bar { d } ^ { c } _ { L } \gamma ^ { \mu } d _ { L } ) + \hat { c } _ { 9 } \left ( d ^ { c } _ { L } \partial _ { \mu } d ^ { c } _ { L } ) ^ { * } ( \bar { d } ^ { c } _ { L } \gamma ^ { \mu } e ^ { c } _ { L } ) \right ].$$
Note that these Lagrangian terms involve a vacuum expectation value of the Standard Model Higgs field, 〈 H 0 〉 glyph[similarequal] 174 GeV, or a derivative which would yield a light fermion mass in the decay amplitude, and thus are suppressed by three powers of a cutoff scale Λ n . (For applications of such d = 7 terms to nucleon decay, baryogenesis and collider signals see Ref. [8, 9].) Neutron-antineutron oscillations arise from the dimension nine effective Lagrangian suppressed by five powers of a cutoff scale Λ nn ¯ and is given by
$$\mathcal { L } ^ { \text{eff} } ( n - \bar { n } ) = \frac { 1 } { \Lambda _ { n \bar { n } } ^ { 5 } } \left [ c _ { 1 } ^ { \prime } \left ( u _ { R } d _ { R } \right ) ( u _ { R } d _ { R } ) ( d _ { R } d _ { R } ) + \dots \right ].$$
Here there are a total of eighteen terms [10] which obey four constraint equations [11]. For brevity we have not displayed them all. These terms are all similar to the term shown in Eq. (4), but differ in their chiral structure, Lorentz contraction as well as color contraction. For our purpose an illustrative term is sufficient.
-
Figure 2: Neutrinoless double beta decay diagrams resulting from combining (a) p → e + π 0 and n → e -π + decays; (b) p → e + π 0 decay and n -¯ oscillation, and (c) n n → e -π + decay and n -¯ oscillation. n
The SuperKamiokande experiment has set a limit on the inverse decay rate for p → e + π 0 , τ ( p → e + π 0 ) > 1 4 . × 10 34 yrs [12]. This leads to a lower limit of Λ p > 4 9 . × 10 15 GeV, where we used the most recent lattice evaluation of the nucleon matrix element, | α H | = 0 012 GeV . . 3 The current best limit on n → e -π 0 lifetime is τ ( n → e -π 0 ) > 6 5 . × 10 31 yrs from IMB experiment [13], which results in the limit Λ n > 6 6 . × 10 10 GeV. It would be interesting to see if SuperK can improve this limit with data already collected. SuperKamiokande has also searched for matter disintegration caused by n -¯ transition inside nuclear matter, and has n set a limit of τ > 1 89 . × 10 32 yrs [14], which would correspond to Λ nn ¯ > 10 5 GeV, with a larger uncertainty resulting from the relevant nuclear matrix element. An experiment carried out at ILL with free neutrons has set a limit on the n -¯ oscillation time of n τ ( n -¯) n > 0 86 . × 10 8 sec [15], which would also yield comparable limit, Λ nn ¯ > 10 5 GeV.
Let us now suppose that the decays p → e + π 0 and n → e -π + are both observed experimentally. By combining the responsible diagrams of Fig. 1 (a) and Fig. 1 (b), or equivalently the effective Lagrangians of Eqs. (2) and (3), we can construct a diagram shown in Fig. 2 (a) for neutrinoless double beta decay. This is the canonical method of determining the Majorana character of the neutrino. While the effective neutrinoless double beta decay lifetime arising from this diagram would be extremely long, well beyond the experimental reach, observation of p → e + π 0 and n → e -π + would nevertheless imply an upper limit on the double beta decay lifetime, which is sufficient to establish the Majorana character of the neutrino.
We can estimate the lifetime for neutrinoless double beta decay ( ββ 0 ν ) arising from Fig. 2 (a) as follows. In the amplitude for the standard neutrino mass mediated ββ 0 ν diagram, replace G 2 F by Λ -2 p 〈 H 0 〉 Λ -3 n , and m / q eff ν 〈 2 〉 by 1 / 〈 q 〉 with 〈 q 〉 ≈ 100 MeV being the average Fermi momentum of the nucleon. Ignoring any differences in the nuclear matrix element, and
Figure 3: Neutrino Majoran mass induced through any diagram that causes neutrinoless double beta decay.
using m ββ < 0 5 eV corresponding to the lifetime limit of . τ > 2 1 . × 10 25 yrs from GERDA experiment [16], we obtain an upper limit on the ββ 0 ν lifetime arising from Fig. 2 (a) to be
$$\tau _ { \beta \beta } < \frac { \tau ( p \to e ^ { + } \pi ^ { 0 } ) } { 1. 4 \times 1 0 ^ { 3 4 } \ y r } \times \frac { \tau ( n \to e ^ { - } \pi ^ { + } ) } { 6. 5 \times 1 0 ^ { 3 1 } \ y r } \times 1 0 ^ { 1 1 3 } \ y r \.$$
While this lifetime is extremely long, it is finite, provided that τ ( p → e + π 0 ) and τ ( n → e -π + ) are observed and thus finite. No matter how long-lived, it is the finiteness of τ ββ that is required to establish the Majorana character of the neutrino.
The diagram of Fig. 2 (a) would generate a small Majorana neutrino mass through the loop diagram of Fig. 3. This diagram exists in all models where ββ 0 ν occurs [17]. The box at the center of the diagram stands for the fundamental vertex leading to ββ 0 ν . The induced neutrino mass, while extremely tiny, is nonzero, which is sufficient to establish the Majorana character of the neutrino. Evaluation of Fig. 3 naively would be fraught with divergencies, but this is not the case once the box is opened up. (For a general discussion of evaluation of this diagram see Ref. [18].) We focus on a well motivated and realistic scenario where both p → e + π 0 and n → e -π + decays are potentially near the current experimental limit, viz., non-supersymmetric SO (10) grand unified theories with an intermediate scale [8]. The diagram for ββ 0 ν of Fig. 2 (a) can arise in these models through the diagrams shown in Fig. 4. Fig. 4 (a) shows these processes mediated by SO (10) gauge bosons, and Fig. 4 (b) by scalar boson belonging to the 120 of Higgs. In Fig. 4 (b) if the scalars fields have masses of order 10 11 GeV, which may be identified as the intermediate symmetry breaking scale of SO (10), then n → e -π + decay may be in the observable range. p → e + π 0 decay is mediated by the X gauge bosons, and may also be within reach. If we insert one of the figures of Fig. 4 in place of the box of Fig. 3, we obtain finite and small neutrino Majorana masses. We estimate the induced mass to be
$$m _ { \nu } \sim \frac { g ^ { 4 } } { ( 1 6 \pi ^ { 2 } ) ^ { 4 } } \, \frac { m _ { e } ^ { 2 } \, m _ { d } ^ { 2 } \, m _ { u } \, \langle H ^ { 0 } \rangle } { \Lambda _ { p } ^ { 2 } \, \Lambda _ { n } ^ { 3 } } \.$$
u
u c d c
d
∆
VQ
c
Vu c
H
(a)
Figure 4: Diagrams inducing neutrinoless double beta decay in SO (10) models by combining p → e + π 0 and n → e -π + decays. (a) Exchange of vector bosons of SO (10), and (b) exchange of vector boson and scalars belonging to 120 of SO (10). The quantum numbers of these fields are X (3 , 2 , -5 6) / , V Q (3 , 2 1 , / 6) , V u c (3 ∗ , 1 , -2 / 3) , ¯(3 η ∗ , 1 , -2 / 3) and ¯(3 χ ∗ , 2 , -7 / 6) under Standard Model gauge symmetry. Here H (1 , 2 1 , / 2) is the SM Higgs field, and ∆ c is the Higgs field responsible for B -L breaking [8].
Numerically, m ν ∼ 5 × 10 -86 GeV, if Λ p and Λ n are at their current experimental limits. We stress again that it is not the numerical value of m ν that is important here, but the fact that m ν is nonzero. Obviously, neutrino oscillation phenomenology would require other sources of Majorana mass generation, which is present in SO (10) models, in the form of the seesaw mechanism.
Now let us turn to the case where p → e + π 0 and n -¯ oscillations are observed. n In this case, the diagram shown in Fig. 2 (b) would lead to neutrinoless double beta decay. Note that in the bottom half of this diagram, a n -¯ vertex has been combined with a n p → e + π 0 vertex to generate effectively a n → e -π + vertex. Thus this case would lead to identical results for the Majorana nature of the neutrino as in the case of p → e + π 0 and n → e -π + discussed earlier. Due to the appearance of two loops, the lifetime for ββ 0 ν arising from Fig. 2 (b) would be about four orders of magnitude longer than the one quoted in Eq. (5). The induced neutrino Majorana mass would be smaller, of order 10 -90 GeV.
The third scenario assumes that n → e -π + decay and n -¯ oscillations are observed. n Note that both of these processes change B -L by two units. One can construct a ∆( B -L ) = 0 effective operator by combining n → e -π + decay diagram with the conjugate of n -¯ n oscillation diagram, as shown in Fig. 2 (c). This diagram shows how neutrinoless double beta decay occurs in this case. A simple way of generating observable n -¯ oscillations is n through the exchange of color sextet scalars which are present in gauge theories based on SU (2) L × SU (2) R × SU (4) c [19] as well as in SO (10). As noted already, the existence of color
X
d
d c e c
e c
Figure 5: Leading diagram for neutrinoless double beta decay in a model with observable n → e -π + decay and n -¯ oscillations. n Here ∆ ud and ∆ dd are color sextet scalars.
triplet scalars ¯ and η ¯ from the 120 Higgs of χ SO (10) at an intermediate scale of 10 11 GeV or lower could lead to observable n → e -π + decay. When these two sources are combined, we find the dominant diagram for neutrinoless double beta decay to be the one shown in Fig. 5. This diagram is structurally identical to the two-loop neutrino mass generation diagram of the model of Ref. [20], which has been studied in detail. Using the results of Ref. [21], we obtain the neutrinoless double beta decay amplitude to be
$$A _ { \beta \beta } ^ { 0 \nu } \simeq \frac { \hat { I } } { ( 1 6 \pi ^ { 2 } ) ^ { 2 } } \frac { M _ { d d } ^ { 2 } M _ { u d } ^ { 2 } } { \Lambda _ { n \bar { n } } ^ { 5 } } \times \frac { \langle H ^ { 0 } \rangle ^ { 2 } } { \Lambda _ { n } ^ { 6 } }$$
where ˆ is a dimensionless integral of order unity [21]. I Using the present lower limits on Λ n and Λ nn ¯ , we obtain the lifetime for neutrinoless double beta decay to be of order 10 131 yrs in this case. Thus we see that in all three cases the Majorana character of the neutrino will be established.
We conclude by making an observation that connects baryon number violation and neutrino Majorana mass through the non-perturbative instanton/sphaleron configurations of weak interactions. These solutions lead to an effective operator involving the twelve Standard Model doublet fermions QQQQQQQQQLLL , with an exponentially suppressed coefficient [22] (at zero temperature) which is very difficult to observe at colliders. Here Q and L are the quark and lepton fields. When expanded, after rotations to bring the mass matrices to diagonal forms, this term would contain terms which look like uddudduudeνν . We can rewrite this as a product of three parts: [ uddudd ] · [ uude ] · [ νν ]. Note that the first part is the piece that contributes to n -¯ oscillations, the second part is relevant to n p → e + π 0 decay, and the last part leads to Majorana mass for the neutrinos. One can represent this in terms
Figure 6: ' B -L Triangle' explains how discovering n -¯ oscillations and proton decay n would impliy that neutrinos are Majorana fermions
a triangle shown in Fig. 6. Thus, if p → e + π 0 and n -¯ oscillations are observed, the inn stanton/sphaleron configuration would imply that neutrino has a Majorana mass. This time it is a direct induction of Majorana mass rather than neutrinoless double beta decay unlike the cases discussed earlier. We realize that the weak instanton/sphaleron effects have not been directly observed and in that sense, this connection is not based on purely experimental observations. Nonetheless we believe that this is an interesting theoretical connection.
The work of KSB is supported in part by the US Department of Energy Grant No. desc0010108 and RNM is supported in part by the National Science Foundation Grant No. PHY-1315155. We thank Saki Khan for graphical help.
## References
- [1] For recent reviews see: J. D. Vergados, H. Ejiri and F. Simkovic, Rept. Prog. Phys. 75 , 106301 (2012) [arXiv:1205.0649 [hep-ph]]; W. Rodejohann, J. Phys. G 39 , 124008 (2012) [arXiv:1206.2560 [hep-ph]].
- [2] For a recent review see: K. S. Babu, E. Kearns, U. Al-Binni, S. Banerjee, D. V. Baxter, Z. Berezhiani, M. Bergevin and S. Bhattacharya et al. , 'Working Group Report: Baryon Number Violation,' arXiv:1311.5285 [hep-ph].
- [3] C. Adams et al. [LBNE Collaboration], 'The Long-Baseline Neutrino Experiment: Exploring Fundamental Symmetries of the Universe,' arXiv:1307.7335 [hep-ex].
- [4] K. Abe, T. Abe, H. Aihara, Y. Fukuda, Y. Hayato, K. Huang, A. K. Ichikawa and M. Ikeda et al. , 'Letter of Intent: The Hyper-Kamiokande Experiment - Detector Design and Physics Potential -,' arXiv:1109.3262 [hep-ex].
- [5] See proceedings of the nn-bar at ESS Workshop, CERN, July 2014: https://indico.esss.lu.se/indico/conferenceDisplay.py?confId=171
- [6] S. Weinberg, Phys. Rev. Lett. 43 , 1566 (1979); F. Wilczek and A. Zee, Phys. Rev. Lett. 43 , 1571 (1979); L. F. Abbott and M. B. Wise, Phys. Rev. D 22 , 2208 (1980).
- [7] S. Weinberg, Phys. Rev. D 22 , 1694 (1980); H. A. Weldon and A. Zee, Nucl. Phys. B 173 , 269 (1980).
- [8] K. S. Babu and R. N. Mohapatra, Phys. Rev. Lett. 109 , 091803 (2012) [arXiv:1207.5771 [hep-ph]]; Phys. Rev. D 86 , 035018 (2012) [arXiv:1203.5544 [hep-ph]].
- [9] S. M. Barr and X. Calmet, Phys. Rev. D 86 , 116010 (2012) [arXiv:1203.5694 [hep-ph]]; I. Baldes, N. F. Bell and R. R. Volkas, Phys. Rev. D 84 , 115019 (2011) [arXiv:1110.4450 [hep-ph]]; A. de Gouvea, J. Herrero-Garcia and A. Kobach, Phys. Rev. D 90 , 016011 (2014) [arXiv:1404.4057 [hep-ph]].
- [10] S. Rao and R. Shrock, Phys. Lett. B 116 , 238 (1982).
- [11] W. E. Caswell, J. Milutinovic and G. Senjanovic, Phys. Lett. B 122 , 373 (1983).
- [12] E. Kearns, Talk presented at the ISOUP Symposium, Asilomar, CA, May 2013.
- [13] S. Seidel et al. , IMB Collaboration, Phys. Rev. Lett. 61 , 2522 (1988).
- [14] K. Abe et al. [Super-Kamiokande Collaboration], arXiv:1109.4227 [hep-ex].
- [15] M. Baldo-Ceolin, P. Benetti, T. Bitter, F. Bobisut, E. Calligarich, R. Dolfini, D. Dubbers and P. El-Muzeini et al. , Z. Phys. C 63 , 409 (1994).
- [16] M. Agostini et al. [GERDA Collaboration], Phys. Rev. Lett. 111 , no. 12, 122503 (2013) [arXiv:1307.4720 [nucl-ex]].
- [17] J. Schechter and J. W. F. Valle, Phys. Rev. D 23 , 1666 (1981).
- [18] M. Duerr, M. Lindner and A. Merle, JHEP 1106 , 091 (2011) [arXiv:1105.0901 [hep-ph]].
- [19] R. N. Mohapatra and R. E. Marshak, Phys. Rev. Lett. 44 , 1316 (1980) [Erratum-ibid. 44 , 1643 (1980)].
- [20] A. Zee, Nucl. Phys. B 264 , 99 (1986); K. S. Babu, Phys. Lett. B 203 , 132 (1988).
- [21] K. S. Babu and C. Macesanu, Phys. Rev. D 67 , 073010 (2003) [hep-ph/0212058].
- [22] G. 't Hooft, Phys. Rev. Lett. 37 , 8 (1976).
|
10.1103/PhysRevD.91.013008
|
[
"K. S. Babu",
"Rabindra N. Mohapatra"
] |
2014-08-04T20:02:27+00:00
|
2014-08-04T20:02:27+00:00
|
[
"hep-ph",
"hep-ex"
] |
Determining Majorana Nature of Neutrino from Nucleon Decays and n-nbar oscillations
|
We show that discovery of baryon number violation in two processes with at
least one obeying the selection rule \Delta (B-L) = \pm 2 can determine the
Majorana character of neutrinos. Thus observing p \to e^+ \pi^0 and n \to e^-
\pi^0 decays, or p \to e^+ \pi^0 and n-nbar oscillations, or n \to e^- \pi^+
and n-nbar oscillations would establish that neutrinos are Majorana particles.
We discuss this in a model-independent effective operator approach.
|
1408.0804v1
|
## Excitonic Quasiparticles in a Spin-Orbit Mott Insulator
Jungho Kim, 1 M. Daghofer, 2 A. H. Said, 1 T. Gog, 1 J. van den Brink, 2 G. Khaliullin, 3 B. J. Kim 3 4 , , ∗ 1 Advanced Photon Source, Argonne National Laboratory, Argonne, Illinois 60439, USA 2 Institute for Theoretical Solid Sate Physics, IFW Dresden, Helmholtzstr. 20, 01069 Dresden, Germany 3 Max Planck Institute for Solid State Research, Heisenbergstraße 1, D-70569 Stuttgart, Germany and 4 Materials Science Division, Argonne National Laboratory, Argonne, IL 60439, USA
In condensed matter systems, out of a large number of interacting degrees of freedom emerge weakly coupled particles, in terms of which most physical properties are described. For example, Landau quasiparticles 1 (QP) determine all electronic properties of a normal metal. The lack of identification of such QPs is a major barrier for understanding myriad exotic properties of correlated electrons, such as unconventional superconductivity 2 and non-Fermi liquid behaviors. 3 Here, we report the observation of a composite particle in a Mott insulator Sr IrO 2 4 an exciton dressed with magnons-that propagates with the canonical characteristics of a QP: a finite QP residue and a lifetime longer than the hopping time scale. The dynamics of this chargeneutral bosonic excitation mirrors the fundamental process of the analogous one-hole propagation in the background of ordered spins, 4 for which a well-defined QP has never been observed. The much narrower linewidth of the exciton reveals the same intrinsic dynamics that is obscured for the hole and is intimately related to the mechanism of high temperature superconductivity.
The dynamics of a single hole doped into a Mott insulator is one of the unresolved fundamental issues in the physics of high temperature superconductivity in the cuprates. 4-6 In one picture, the single hole forms a coherent quasiparticle (QP) propagating in a medium of ordered spins, which strongly renormalizes the dispersion relations and the mutual interactions of the holes. This spin-polaron picture, supported by numerical approaches, such as quantum Monte-Carlo, 7 exact diagonalization 8 and self-consistent Born approximation (SCBA), 5 naturally connects to the pairing mechanism of pre-existing QPs glued by retarded bosons, analogous to Bardeen-Cooper-Schrieffer mechanism for conventional superconductivity. In a contrasting picture, the phenomenological absence of QPs in angle-resolved photoemission (ARPES) 9-11 and scanning tunneling spectra 12 is argued as due to orthogonality catastrophe, 13 spin-charge separation, 14 and/or localization effects, 15 reflecting the unconventional nature of two-dimensional, strongly interacting fermions, which may support more exotic mechanisms of superconductivity such as the resonating valence bond theory proposed by Anderson. 16
Motivated by the recent discovery of a new pseudospin1/2 Heisenberg antiferromagnet on a square lattice, Sr IrO , 2 4 17-22 we demonstrate in this letter a novel experi-
FIG. 1. Hole vs. exciton propagation in an antiferromagnetic background. a , a moving hole (blue square) creates a string of excited spins. Black wavy lines indicate pairs of 'misaligned' spins. b , an analogous exciton hopping.
mental approach using resonant inelastic x-ray scattering (RIXS), which is a rapidly evolving tool especially well suited for such 5d transition-metal oxides, to address this longstanding problem. Based on its remarkable similarity to superconducting cuprates in structural, 22 electronic, 17 and magnetic aspects, 20,21 superconductivity has been predicted in Sr 2 IrO . 4 23 Although the electron correlation strength in Sr IrO 2 4 has been under much debate, 24-28 questioning the validity of classifying this compound as a Mott insulator, Sr 2 IrO 4 shares the same phenomenology that the hole spectral function as measured by ARPES lacks a legitimate QP. 17 This similarity suggests a common origin for the absence of a well-defined QP in the ARPES spectra for two distinct but similar classes of materials. As we shall see, however, Sr IrO 2 4 supports a well-defined QP in another excitation channel, which RIXS is sensitive to and reflects the same dynamics as that of a hole, offering a novel route to studying elementary excitations in correlated oxides.
In an earlier RIXS study of Sr 2 IrO , dispersive d-d ex4 citations (or excitons) across the spin-orbit coupling split levels (see Fig. 2a) have been identified and it has been shown that their dispersions can be understood in close analogy to the single hole problem. 20 Figure 1 depicts the
a
b
FIG. 2. RIXS spectra of Sr IrO . a 2 4 , the spin-orbital level scheme of the three Kramers pairs and their orbital shapes in Sr IrO . 2 4 The spin-orbit entangled nature of these quantum states is illustrated with colors; orange (blue) represent spin up (down) projection. b , Spectrum at q =( π /2, π /2) measured at three different x-ray incident angle φ . A, B, and C denote the energy positions of the three Kramers pairs illustrated in a . The scattering angle 2 θ was kept within 5 degrees from 90 . ◦
hole-vs.-exciton analogy: a foreign object injected into a quantum Heisenberg antiferromagnet, be it a hole or an exciton, creates a string of flipped spins along its hopping path. This analogy, in principle, suggests that the dynamics of a particle moving in a magnetic medium can also be studied using the exciton. However, the energy resolution of RIXS used in the earlier study ( ≈ 130 meV) was insufficient to resolve the dispersion and the intrinsic linewidth of the two exciton modes associated with the two pairs of Kramers doublets in the j=3/2 manifold (Fig. 2a). In this Letter, we exploit the different orbital symmetries of the two exciton modes to selectively probe each mode, and with the much improved energy resolution ( ≈ 30 meV) offered by RIXS after recent developments, reveal their full dynamics.
Figure 2b shows the RIXS spectra measured at three different incident angles φ of x-ray while fixing momentum transfer at q =( π /2, π /2). The two exciton modes, labeled as peaks B and C, show strong modulations in intensity as a function of through the change in the incident and outgoing x-ray polarizations relative to the sample surface and thereby the RIXS matrix elements. In particular, the peak B is strongly enhanced (completely suppressed) by tuning φ to normal (grazing) incidence geometry. This strong matrix element effect enables se-
FIG. 3. Selective mapping of the two exciton modes and their comparison to SCBA calculations. a , Image plot of RIXS spectra measured along high symmetry lines in the normal and grazing incidence geometry. b , SCBA calculations using the parameters λ SO = 382 meV, ∆ BC = 137 meV, t 1 = J 1 /2 =30 meV, t 2 = t 3 = 7.6 meV, J 1 = 60 meV, J 2 = -20 meV, and J 3 = 15 meV. t 1 2 3 , , ( J 1 2 3 , , ) denote first, second, and third nearest neighbor hoppings (magnetic couplings). λ SO and ∆ BC are defined in fig. 2. The spectral functions obtained by SCBA calculations are convoluted with a Lorentzian function with 5 meV width.
lective mapping of B and C modes, as shown in Fig. 3a. While both B and C modes display rather similar dispersions, the B mode has much narrower linewidth (see Fig. 4c) partly due to the fact that it has lower excitation energy and thus has less phase space to decay into. With the two exciton modes disentangled, our high-resolution measurement yields for the B mode a bandwidth of 112 meV. This finding is consistent with the expectation that the bandwidth is on the scale of a few times the antiferromagnetic exchange coupling J of about 60 meV. 20 The global topology of the dispersions with minimum at q =( π /2, π /2) and maximum at the point precisely matches that measured for a hole in cuprates by ARPES, 9 which strongly supports the holevs.-exciton analogy.
We compare the experimental data to the spectral function of the effective t J -model calculated within the SCBA (Fig. 3b). For details, see Supplementary Information and Ref. 20. The calculation yields two modes, one of which is predominantly from the Kramers doublet with quantum numbers | J=3/2, J = z ± 3/2 , 〉 while the other has mostly | J=3/2, J = z ± 1/2 〉 character. Through an
FIG. 4. Exciton dynamics in Sr IrO . 2 4 a , RIXS spectrum at q =( π /2, π /2) (black open circles). The blue shaded peak corresponds to the exciton QP peak. The spectrum was fitted (red solid curve) using a Lorenzian lineshape for the exciton QP peak and Gaussian lineshapes for all other peaks (red dashed lines). The low energy features relative to exciton QP consist of elastic, single- and double-magnon peaks, and high-energy features the sum of C mode and background due to electron-hole continuum and incoherent part of B mode. b , Stack plot of the image plot in Fig. 2a, left panel. Oval symbols mark the energy position of the QP. In addition to the QP, a small peak with the dispersion minimum at the Γ point at E ≈ 0.37 eV is observed. A similar peak has been observed in a related material Na IrO 2 3 and attributed to a bound state at the edge of the particle-hole continuum. 42 c , Energy, width, and intensity of the QP peaks along high symmetry lines extracted from peak fitting as exemplified in a .
explicit calculation of the RIXS matrix elements, we find that their dependences on φ are such that the J z = ± 1/2 states, with more in-plane components, are higher in energy (in the hole picture) than the J z = ± 3/2 states. This level scheme is opposite to expectations based on perfectly cubic or c-axis elongated oxygen octahedra, but agrees with the results from non-resonant inelastic xray scattering, 29 electron spin resonance 30 and quantum chemistry calculations. 31 Taking a crystal field splitting ∆ BC of 137 meV and fixing all other parameters to values inferred from independent studies, the effective t J -model reproduces the gross features of the experimen-
FIG. 5. Asymmetric exciton QP intensity. a , Image plot of RIXS spectra along (0,0)-( π π , ) direction. b , Stack plot of the image plot in a . Oval symbols mark the energy position of the QP. c , Intensity of the QP peaks along (0,0)-( π π , ) direction.
tal spectra. Details are improved by including further neighboring hoppings, which are expected to be significant from sizable further neighbor magnetic couplings. With the parameters given in the Fig. 3b caption, we find an excellent agreement with the data in terms of the polarization dependence and the dispersion relations, confirming that the exciton dynamics is essentially captured by the effective t J -model.
Having justified the hole-vs.-exciton analogy, we now bring to light the key observation from the exciton spectra. The energy distribution curves, measured in the normal incidence geometry to highlight the B mode, reveal a very sharp exciton peak, most prominent at q =( π /2, π /2) (Fig. 4a) and resolved throughout most part of the Brillouin zone (Fig. 4b). We use a phenomenological Lorentzian lineshape to fit the spectra to extract the peak energy, width, and intensity, which are summarized in Fig. 4c. Remarkably, the peak width is as narrow as ≈ 50 meV (of which 30 meV is contributed by the experimental resolution), much narrower than that of the sharpest peak ( ∼ 200 meV) in the hole spectral function measured by ARPES for the same material. 17,32 The peak width is also much smaller than its total bandwidth ( ≈ 112 meV), which establishes the exciton as a propagating mode in a solid, or a QP. More importantly, our observation of an excitonic QP establishes hard evidence that a particle can propagate coherently through a quantum antiferromagnet. This raises a fundamental question: why is a QP absent for the single hole excitation?
Thus far we have focused on the similarity between the dynamics of a hole and an exciton. Let us now discuss some important differences. A hole is a charge monopole
and its sudden creation in the ARPES process leads to deformation of the surrounding ionic oxygen cage, which results in a strong hole-lattice coupling detrimental for the hole propagation. Further, a hole interacts with charged impurities always present and poorly screened in an insulator. Both of these effects have been shown to strongly damp or wash out sharp QPs, 33-35 thereby significantly redistributing the hole spectral function. In fact, most likely for these reasons, for a quasi-onedimensional system Sr CuO 2 3 in which the phenomenon of spin-charge separation is established, 36 ARPES measures much-broadened spectra 37,38 of the theoretically predicted sharp edge-singularity in the exactly soluble model. By contrast, a charge-neutral exciton with a quadrupole moment should couple much more weakly to the lattice and is not subject to long-range Coulomb forces due to impurities. Thus, an exciton avoids these 'side effects' and reveals the intrinsic dynamics of the t -J model that have remained elusive for the past several decades.
Our central message is that the unprecedentedly narrow linewidth of the QP allows a direct access into the nature of a QP living in the background of ordered spins, and thus that it allows direct verification of a class of theories that predict a finite quasiparticle residue in the Mott insulating phase, which has thus far remained inconclusive in the apparent absence of a QP in the hole channel. For instance, the observed exciton intensity can be compared with that calculated in theoretical models on a quantitative level. Figures 5a and 5b show the exciton dispersion along (0,0)-( π π , ) direction. The dispersion is symmetric with respect to ( π /2, π /2) point on the mag-
## References
- 1 Landau, L. D. The theory of a Fermi liquid. Sov. Phys. JETP 3 , 920-925 (1957).
- 2 Norman, M. R. Unconventional superconductivity, Prepreint at http://arxiv.org/pdf/1302.3176v1
- 3 Stewart, G. R. Non-Fermi-liquid behavior in dand felectron metals, Rev. Mod. Phys. 73 , 797-848 (2001).
- 4 Lee, P. A., Nagaosa, N., Wen, X.-G. Doping a Mott insulator: Physics of high-temperature superconductivity. Rev. Mod. Phys. 78 , 17-85 (2001).
- 5 Schmitt-Rink, S., Varma, C. M., Ruckenstein, A. E. Spectral function of holes in a quantum antiferromagnet. Phys. Rev. Lett. 60 , 2793 (1988).
- 6 Kane, C. L, Lee, P. A., Read, N. Motion of a single hole in a quantum antiferromagnet. Phys. Rev. B 39 , 6880 (1989).
- 7 Duffy, D., Nazarenko, A., Haas, S., Moreo, A., Riera, J., Dagotto, E. Hole-doping evolution of the quasiparticle band in models of strongly correlated electrons for the high-Tc cuprates. Phys. Rev. B 56 , 5597 (1997).
- 8 Eder, R., Ohta, Y., Sawatzky, G. A. Doping-dependent quasiparticle band structure in cuprate superconductors. Phys. Rev. B 55 , R3414 (1996).
- 9 Wells, B. O., Shen, Z. -X., Matsuura, A., King, D. M., Kast-
netic zone boundary, but the intensities inside and outside of the magnetic zone differ significantly (Fig. 5c); the exciton QP displays a steep drop in intensity as ( π /2, π /2) is crossed. This is a generic feature predicted for a hole in a t J -model within a SCBA approach 39 and indirectly inferred from ARPES measurements on cuprates, 40 which possibly accounts for the Fermi arc observed in the doped case as due to the strongly suppressed shadow band intensity. Although some deviations from the cuprate physics might have been expected on general grounds due to fundamentally different nature of a hole and an exciton, and material specific differences between Sr IrO 2 4 and cuprates, the striking agreement down to a level of fine details with theories constructed for hole dynamics in cuprates demonstrates excellent parallel between the low-energy physics of Sr 2 IrO 4 and the cuprates.
Methods Single crystals of Sr 2 IrO 4 were grown by the flux method. The sample was mounted in a displex closed-cycle cryostat and measured at 15 K. The RIXS measurements were performed using the MERIX spectrometer at the 30-ID beamline 41 of the Advanced Photon Source. X-rays were monochromatized to a bandwidth of 15 meV, and focused to have a beam size of 45(H) × 30(V) µ m . 2 A horizontal scattering geometry was used with the incident photon polarization. A Si (844) diced spherical analyzer with 4 inch radius and a position-sensitive silicon microstrip detector were used in the Rowland geometry. The overall energy resolution of the MERIX spectrometer at the Ir L 3 edge was 30 meV, as determined from the full-width-half-maximum of the elastic peak.
- ner, M. A., Greven, M., Birgeneau, R. J. E versus k relations and many body effects in the model insulating copper oxide Sr CuO Cl . Phys. Rev. Lett. 2 2 2 74 , 964967 (1995).
- 10 Tohyama, T., Maekawa S. Angle-resolved photoemission in high T C cuprates from theoretical viewpoints. Supercond. Sci. Technol. 13 , R17R32 (2000).
- 11 Shen, K. M., Ronning, F., Lu, D. H., Lee, W. S., Ingle, N. J. C., Meevasana, W., Baumberger, F., Damascelli, A., Armitage, N. P., Miller, L. L., Kohsaka, Y., Azuma, M., Takano, M., Takagi, H., Shen, Z.-X. Missing quasiparticles and the chemical potential puzzle in the doping evolution of the cuprate superconductors. Phys. Rev. Lett. 93 , 267002 (2004).
- 12 Ye, C., Cai, P, Yu, R., Zhou, X., Ruan, W., Liu, Q., Jin, C., Wang, Y. Visualizing the atomic-scale electronic structure of the Ca 2 CuO Cl 2 2 Mott insulator. Nat. Commun. 4 . 1365 (2013).
- 13 Anderson, P. W. Luttinger-liquid behavior of the normal metallic state of the 2D Hubbard model. Phys. Rev. Lett. 64 , 1839 (1990).
- 14 Laughlin, R. B. Evidence for Quasiparticle Decay in Photoemission from Underdoped Cuprates. Phys. Rev. Lett. 79 , 1726 (1997).
- 15 Sheng, D. N., Chen, Y. C., Weng, Z. Y. Phase string ef-
fect in a doped antiferromagnet. Phys. Rev. Lett. 77 , 5102 (1996).
- 16
- Anderson, P. W., Lee, P. A., Randeria, M., Rice, T. M., Trivedi, N, Zhang, F. C. The physics behind hightemperature superconducting cuprates: the 'plain vanilla' version of RVB. J. Phys.: Condens. Matter 16 , R755 (2004). 17 Kim, B. J., Jin, H., Moon, S. J., Kim, J.-Y, Park, B.-G., Leem, C. S., Yu, J., Noh, T.W., Kim, C., Oh, S.-J., Park, J.H., Durairaj, V., Cao, G., Rotenberg, E. Novel J eff =1/2 Mott state induced by relativistic spin-orbit coupling in Sr 2 IrO . Phys. Rev. Lett. 4 101 , 076402 (2008).
- 18 Jackeli, G., Khaliullin, G. Mott insulators in the strong spin-orbit coupling limit: from Heisenberg to a quantum compass and Kitaev models. Phys. Rev. Lett. 102 , 017205 (2009).
- 19 Kim, B. J., Ohsumi, H., Komesu, T., Sakai, S., Morita, T., Takagi, H., Arima, T. Phase-sensitive observation of a spin-orbital Mott state in Sr 2 IrO . Science 4 323 , 1329-1332 (2009).
- 20 Kim, J., Casa, D., Upton, M. H., Gog, T., Kim, Y.-J., Mitchell, J. F., van Veenendaal, M., Daghofer, M., van den Brink, J., Khaliullin, G., Kim, B. J. Magnetic excitation spectra of Sr 2 IrO 4 probed by resonant inelastic x-ray scattering: Establishing links to cuprate superconductors. Phys. Rev. Lett. 108 , 177003 (2012).
- 21 Fujiyama, S., Ohsumi, H., Komesu, T., Matsuno, J., Kim, B. J., Takata, M., Arima, T., Takagi, H., Two-dimensional Heisenberg behavior of J eff =1/2 isospins in the paramagnetic state of the spin-orbital Mott insulator Sr 2 IrO , Phys. 4 Rev. Lett. 108 , 247212 (2012).
- 22 Crawford, M. K., Subramanian, M. A., Harlow, R. L., Fernandez-Baca, J. A., Wang, Z. R., Johnston, D. C. Structural and magnetic studies of Sr IrO . Phys. Rev. B 2 4 49 , 9198-9201 (1994).
- 23 Wang, F., Senthil, T. Twisted Hubbard model for Sr IrO : 2 4 Magnetism and possible high temperature superconductivity. Phys. Rev. Lett. 106 , 136402 (2011).
- 24 Martins, C., Aichhorn, M., Vaugier, L., Biermann, S. Reduced effective spin-orbital degeneracy and spin-orbital ordering in paramagnetic transition-metal oxides: Sr IrO 2 4 versus Sr 2 RhO . Phys. Rev. Lett. 4 107 , 266404 (2011).
- 25 Arita, R., Kune, J., Kozhevnikov, A. V., Eguiluz, A. G., Imada, M. Ab initio studies on the interplay between spinorbit interaction and coulomb correlation in Sr IrO 2 4 and Ba IrO . Phys. Rev. Lett. 2 4 108 , 086403 (2012).
- 26 Hsieh, D., Mahmood, F., Torchinsky, D. H., Cao, G., Gedik, N. Observation of a metal-to-insulator transition with both Mott-Hubbard and Slater characteristics in Sr IrO 2 4 from time-resolved photocarrier dynamics. Phys. Rev. B 86 , 035128 (2012).
- 27 Li, Q., Cao, G., Okamoto, S., Yi, J., Lin W., Sales, B. C., Yan, J., Arita, R., Kunes, J., Kozhevnikov, A. V., Eguiluz, A. G., Imada, M., Gai, Z., Pan, M., Mandrus, D. G. Microscopic and spectroscopic evidence for a Slater metal-Insulator transition in Sr IrO . 2 4 Preprint at http://arxiv.org/abs/1303.7265.
- 28 Yamasaki, A. et al. Bulk nature of layered perovskite iridates beyond the Mott scenario: An approach from bulk sensitive photoemission study. Phys. Rev. B 89 , 121111(R) (2014).
- 29 Fujiyama, S., Ohsumi, H., Ohashi, K., Hirai, D., Kim, B. J., Arima, T., Takata, M., Takagi, H. Spin and orbital contributions to magnetically ordered moments in 5d layered perovskite Sr 2 IrO . Phys. Rev. Lett. 4 112 , 016405 (2014).
- 30 Bahr, S., Alfonsov, A., Jackeli, G., Khaliullin, G., Matsumoto, A., Takayama, T., Takagi, H., Bchner, B., Kataev, V. Low energy magnetic excitations in the spin-orbital Mott insulator Sr IrO . 2 4 Preprint at http://arxiv.org/abs/1312.4847.
- 31 Katukuri, V. M., Stoll, H., van den Brink J., Hozoi, L. Ab initio determination of excitation energies and magnetic couplings in correlated quasi-two-dimensional iridates. Phys. Rev. B 85 , 220402(R) (2012).
- 32 Wang, Q., Cao. Y, Waugh, J. A., Park, S. R., Qi, T. F., Korneta, O. B., Cao, G., Dessau, D. S. Dimensionalitycontrolled Mott transition and correlation effects in singlelayer and bilayer perovskite iridates. Phys. Rev. B 87 , 245109 (2013).
- 33 Mishchenko, A. S., Nagaosa, N. Electron-phonon coupling and a polaron in the t J -model: from weak to the strong coupling regime. Phys. Rev. Lett. 93 ,036402 (2004).
- 34 Chen, W., Khaliullin, G., Sushkov, O. P. Coulomb disorder effects on angle-resolved photoemission and nuclear quadrupole resonance spectra in cuprates. Phys. Rev. B 80 , 094519 (2009).
- 35 King, P. D. C., Takayama, T., Tamai, A., Rozbicki, E., Walker, S. M., Shi, M., Patthey, L., Moore, R. G., Lu, D., Shen, K. M., Takagi, H., Baumberger, F. Spectroscopic evidence for polaronic behaviour of the strong spin-orbit insulator Sr 3 Ir 2 O . Phys. Rev. B 7 87 , 241106(R) (2013).
- 36 Zaliznyak, I. A., Woo, H., Perring, T. G., Broholm, C. L., Frost, C. D., Takagi, H. Spinons in the strongly correlated copper oxide chains in SrCuO . Phys. Rev. Lett. 2 93 , 087202 (2004).
- 37 Kim, C. Matsuura, A. Y., Shen, Z.-X., Motoyama, N., Eisaki, H., Uchida, S., Tohyama, T., Maekawa, S. Observation of spincharge separation in one-dimensional SrCuO . 2 Phys. Rev. Lett. 77 , 40544057 (1996).
- 38 Kim, B. J., Koh, H., Rotenberg, E., Oh, S.-J., Eisaki, H., Motoyama, N., Uchida, S., Tohyama, T., Maekawa, S., Shen, Z.-X., Kim, C. Distinct spinon and holon dispersions in photoemission spectral functions from one-dimensional SrCuO . Nat. Phys. 2 2 , 397401 (2006).
- 39 Sushkov, O. P., Sawatzky, G. A., Eder, R., Eskes, H. Hole photoproduction in insulating copper oxide. Phys. Rev. B 56 , 11769 (1997).
- 40 LaRosa, S., Vobornik, I., Zwick, F., Berger, H., Grioni, M., Margaritondo, G., Kelley, R. J., Onellion M., Chubukov, A. Electronic structure of CuO 2 planes: From insulator to superconductor. Phys. Rev. B 56 , R525 (1997).
- 41 Gog, T., Casa, D. M., Said, A. H., Upton, M. H., Kim, J., Kuzmenko, I., Huang, X., Khachatryan, R. Spherical analyzers and monochromators for resonant inelastic hard X-ray scattering: a compilation of crystals and reflections. J. Synchrotron Rad. 20 74-79 (2013).
- 42 Gretarsson, H., Clancy, J. P., Liu, X., Hill, J. P., Bozin, E., Singh, Y., Manni, S., Gegenwart, P., Kim, J., Said, A. H., Casa, D., Gog, T., Upton, M. H., Kim, H.-S., Yu, J., Katukuri, V. M., Hozoi, L., van den Brink, J., Kim, Y.J. Crystal-field splitting and correlation effect on the electronic structure of A 2 IrO . Phys. Rev. Lett. 3 110 , 076402 (2013).
## Supplementary Note 1
The effective t -J model and the calculation of exciton spectra within self-consistent Born approximation
The theoretical modeling is based on the observation that propagation of an orbital excitation in a Mott insulator can be mapped onto a single hole moving in the system, 1,2 as illustrated in Fig. 1. We extend here the approach used in Ref. 1 by employing a more accurate technique, the self-consistent Born approximation (SCBA). 3 In contrast to the second-order perturbation theory used previously, 1 which is valid for exciton hopping small compared to magnetic interactions, it contains diagrams to all orders of the hopping, i.e, in the coupling between exciton motion and the magnetic background. Together with the much increased experimental resolution, the use of the SCBA rather than perturbation theory allows a more quantitative comparison of theory and experiment. We find that a good description of the data can be obtained with spin-orbit coupling λ = 382 meV, crystal-field splitting ∆ = -188 meV (in the notation of Ref. 4), magnetic couplings fitted to the magnon dispersion 1 and by allowing for longer-range hopping of the spin-orbit excitation. As mentioned in the main text, these parameter values are well supported by other experiments and ab initio methods - with the exception of the longer-range hoppings, which can only be investigated via precise measurement of the exciton propagation, as done here.
The SCBA is a diagrammatic approach that describes the antiferromagnetic (AFM) background with linear spin-wave theory. The diagrams describing coupling of the moving hole or exciton to the (pseudo-)spin background are those without crossing magnon lines, an approximation that has been shown to be valid as the leading crossing diagrams drop out for symmetry reasons. 5 The method has been extensively used to investigate hole motion in an AFM and has been used to discuss propagation of an orbital excitation 6 . In contrast to secondorder perturbation theory, it allows to describe incoherent spectral weight as well as coherent quasi-particle motion. Nevertheless, it underestimates quantum fluctuations of the magnetic background and higher-order vertex corrections may become more relevant for a hopping of a similar size as exchange coupling.
The A, B and C modes all mix spin and orbital of the 5 d wave functions, depending on spin-orbit coupling and crystal-field splitting ∆ between the xy and xz/yz orbitals. The low-energy A states are then given by | A σ 〉 = sin θ τ | = 0; σ 〉 -cos θ τ | = σ ; - 〉 σ , where σ = ± 1 denotes the pseudospin and τ = 0 , ± 1 the orbital angular momentum and θ is given by tan 2 θ = 2 √ 2 λ/ λ ( -2∆). 4 The higher-energy modes B and C, which we focus on
FIG. S1. Theoretical spectra for a scattering geoemtry with (a) φ =90 ◦ and (b) φ =0 . ◦ In (c) and (d), the weight for the two doublets B and C is shown, i.e. -I G BB in (c) and -I G CC in (d). The parameters used were λ = 382 meV, ∆ = -188 meV, W 1 = J / 1 2 = 30 meV, W 2 = W 3 = 7 6 meV, . J 1 = 60 meV, J 2 = -20 meV, and J 3 = 15 meV. Imaginary part used in the SCBA was δ = 5 meV.
here, are given by
$$| B _ { \sigma } \rangle = | \tau = \sigma ; \sigma \rangle, \quad \text{and}$$
$$| C _ { \sigma } \rangle = \cos \theta \, | \tau = 0 ; \sigma \rangle + \sin \theta \, | \tau = \sigma ; - \sigma \rangle \. \quad ( 2 )$$
The energy cost of exciting a hole from the A into the B or C doublets is given by
$$E ^ { \text{C} } & = \frac { 3 \lambda } { 2 } \sqrt { 1 - \frac { 4 \Delta } { 9 \lambda } + \frac { 4 \Delta ^ { 2 } } { 9 \lambda ^ { 2 } } } \\ E ^ { \text{B} } & = \frac { \Delta } { 2 } + \frac { 3 \lambda } { 4 } + \frac { 3 \lambda } { 4 } \sqrt { 1 - \frac { 4 \Delta } { 9 \lambda } + \frac { 4 \Delta ^ { 2 } } { 9 \lambda ^ { 2 } } } \. \quad ( 3 ) \\ \intertext { T, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dcls, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \delta, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dts, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dct, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dts, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dcl, \dots, \dots, \dots, \dots, \dots, \dots, \dots, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl. \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl{ \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl, \dcl}$$
The splitting between A, B and C modes observed in the RIXS data can be reproduced by taking spin-orbit coupling λ = 382 meV and a crystal-field splitting ∆ = -188 meV, yielding θ ≈ 0 15 . π . The sign of ∆ is here chosen as in Ref. 4, i.e., it is positive for elongated octahedra and negative for shortened ones, if one considers a simple point-charge model. However, we find that in order to reproduce the order of the B and C modes as seen in experiment, and consistent with first-principle calculations 7 and other experiments , ∆ 8 < 0 has to be chosen for Sr 2 IrO . In the notation used in the main text, 4 the resulting splitting E C -E B = ∆ BC = 137 4 meV be-. tween B and C levels has the sign opposite to that of ∆, i.e., it is here positive.
As discussed in Ref. 1, the orbital excitation moves via a superexchange process with the matrix element
$$\overset { \cdot } { \slash y z } ^ { \cdot } & \quad \bullet \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\overset { \cdot } { \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\underset { \substack { \pm 1 \\ \text{$\pm$} } } } \longrightarrow \quad \begin{array} { c } \circ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \\ \text{$\mathbb{ } } \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \\ \sl u l a r \\ \underset { \substack { \mathbb{ } } } \circ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \\ \mathbb{ } } \cdot \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \. \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ } \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ m \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ & \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ < \ \text{$\mathbb{ } } \quad \textbb{} } \quad \textbb{} } \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \mathbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \\ \mathbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{\} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad\textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{ } \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{} \quad \textbb{$$
along the x and y directions, respectively. t α,β are here the direct hopping elements for the three Kramersdoublets A, B, and C. U ′ is the Coulomb repulsion between two electrons on the same site, but in different orbitals. Since SU (2) symmetry pseudospin excitations of the A sector indicate small Hund's rule coupling, it is not expected to differ much from the intraorbital repulsion U and we take here U ′ = U . The overall factor scaling the propagation is then W 1 = J 2 ≈ 30 meV. To distinguish them more clearly from the bare hopping elements t referring to electron motion, effective exciton hoppings are denoted by W in this supplementary material, following earlier work. 1 The bare hopping elements entering (4) depend on the composition of the states as
$$t _ { \text{AA} } & = \frac { 3 } { 4 } - \frac { \cos 2 \theta } { 4 }, \quad t _ { \text{BB} } = \frac { 1 } { 2 } \quad \quad ( 5 ) \quad \text{where} \\ t _ { \text{CC} } & = \frac { 3 } { 4 } + \frac { \cos 2 \theta } { 4 }, \quad t _ { \text{BC} } = \frac { \sin \theta } { 2 } \. \quad \quad \ \text{$num at n$} \\ \cdot \., \quad \. \. \. \. \. \. \. \. \. \. \. \. \. \.$$
As the excitonic hopping is a superexchange process, it scales with the magnetic exchange J , see (4). In Sr 2 IrO , 4 nearest-neighbor exchange is here taken as J = 60 meV. Longer-range couplings are also present and sizable with ≈ 15 -20 meV, analogously, we introduce longer-range exciton hoppings W βγ 2 and W βγ 3 for second and third neighbors. For simplicity, they are assumed to be identical for the B and C doublet with W BC 2 / 3 = 0, leaving us with two parameters W BB 2 = W CC 2 = W 2 and W BB 3 = W CC 3 = W 3 . They are adjusted to fit the quasiparticle dispersion, like longer-range hoppings in the case of ARPES. Exciton hopping and magnetic background together are thus described by an effective t J -model, where both hopping and magnetic exchange are included up to third neighbors. In order to make the analogy to cuprates more transparent, W 1 , W 2 and W 3 are denoted by t 1 , t 2 , and t 3 in the main text. One difference to the t J -model used to describe hole motion in an antiferromagnet is the orbital index of the exciton. Moreover, hopping W and magnetic coupling J are of the same order of magnitude, as they share the same origin, superexchange.
In order to treat this problem, we follow a route extensively used to describe hole propagation in the t J -model, where the AFM order is treated in linear spin-wave approximation. For the case with only nearest-neighbor terms and for a single excited-orbital mode, it has been used to discuss propagation of an orbital excitation 6 . The Hamiltonian is Fourier-transformed into momentum space and a Bogoliubov transformation is applied to the magnetic part. The magnetic energy becomes
$$H _ { \text{mag} } = \sum _ { \mathbf q } \omega _ { \mathbf q } a _ { \mathbf q } ^ { \dagger } a _ { \mathbf q } & & ( 6 ) & & \text{with}$$
where a q ( a † q ) annihilates(creates) a magnon with momentum q , with a magnon energy
$$\omega _ { \mathbf q } = \sqrt { A _ { \mathbf q } ^ { 2 } - B _ { \mathbf q } ^ { 2 } }$$
and coefficients
$$\underset { \cdot \,. } { \text{aime} } = & \, \ A _ { \mathbf q } = 2 ( J _ { 1 } - J _ { 2 } - J _ { 3 } + J _ { 2 } \cos q _ { x } \cos q _ { y } ) + J _ { 3 } ( \cos 2 q _ { x } + \cos 2 q _ { y } ), \\ \underset { \cdot \,. } { \text{iferent} } \quad & \, \ B _ { \mathbf q } = J _ { 1 } ( \cos q _ { x } + \cos q _ { y } ) \,.$$
The magnetic couplings are chosen as J 1 = 60 meV, J 2 = -20 meV, and J 3 = 15 meV, in order to describe the 'magnon' excitations of the pseudospin-1 / 2 checkerboard order. 1 Nearest-neighbor hopping moves the exciton, i.e., the hole in the AFM background from one sublattice of the AFM order to the other and thereby couples to the magnetic background. It can be written as
$$\text{ints} _ { \text{as} } H _ { \text{NN} } = - \frac { z W _ { 1 } } { \sqrt { N } } \sum _ { \text{k,q,\alpha,\beta} } \left [ M _ { \text{k,q} } ^ { \alpha \beta } X _ { \alpha \text{k} } ^ { \dagger } X _ { \beta \text{k-q} } a _ { \text{q} } + h. c. \right ] \,, \ \ ( 9 )$$
where N is the number of sites and z = 4 the coordination number. X † α k and X β k create and annihilate the exciton at momentum k and orbital index α . The orbital and direction dependence of Eq. (4) was absorbed into the orbital-dependent vertex
$$M _ { \mathbf k, \mathbf q } ^ { \alpha \beta } = | \tau ^ { \alpha \beta } | ( u _ { \mathbf q } \gamma _ { \mathbf k - \mathbf q } ^ { \alpha \beta } + v _ { \mathbf q } \gamma _ { \mathbf k } ^ { \alpha \beta } ) \quad \quad ( 1 0 )$$
via
$$\widetilde { \text{hird} } _ { \text{len} ^ { - } } \quad \gamma _ { \text{k} } ^ { \alpha \beta } = \frac { 1 } { 2 } \left ( \begin{array} { c c } ( \cos k _ { x } + \cos k _ { y } ) & ( \cos k _ { x } - \cos k _ { y } ) \\ ( \cos k _ { x } - \cos k _ { y } ) & ( \cos k _ { x } + \cos k _ { y } ) \end{array} \right ) \, ; \ ( 1 1 )$$
Bogoliubov factors u q and v q are given by the relations
$$\underset { \text{ind} } { \text{asi} } \underset { \text{ind} } { \text{asi} } \quad u _ { \text{q} } = \frac { 1 } { \sqrt { 2 } } \sqrt { \frac { \text{A} _ { \text{q} } } { \omega _ { \text{q} } } + 1 }, \ \ v _ { \text{q} } = - \frac { \text{sgnB} _ { \text{q} } } { \sqrt { 2 } } \sqrt { \frac { \text{A} _ { \text{q} } } { \omega _ { \text{q} } } - 1 } \. \ \ ( 1 2 )$$
Second- and third-neighbor hoppings W 2 and W 3 , in contrast, allow the exciton to move within one sublattice of the AFM background and consequently do not disturb the the magnetic order. Together with the onsite energies E B and E C reflecting the energy cost of exciting a hole from the A into the B or C doublets, see (3), they add a 'free' non-interacting exciton Hamiltonian: 9
$$\frac { \mathfrak { a }, \ s u - } { \times \text{xten} - } \quad H _ { 0 } = - \frac { 1 } { \sqrt { N } } \sum _ { \mathfrak { q }, \alpha } \epsilon _ { \mathfrak { q } } ^ { \alpha } X _ { \alpha \mathfrak { q } } ^ { \dagger } X _ { \alpha \mathfrak { q } } \quad \text{with}$$
$$\det \dots \quad \epsilon _ { \mathbf q } ^ { \alpha } = E ^ { \alpha } - 4 W _ { 2 } \cos \mathbf q _ { x } \cos \mathbf q _ { y } - 2 W _ { 3 } ( \cos 2 \mathbf q _ { x } + \cos 2 \mathbf q _ { y } ).$$
The full Hamiltonian H 0 + H NN + H mag is then treated using the self-consistent Born approximation, 3 a diagrammatic approach where the self energy is expressed as
$$\text{enum} \\ \text{to the} } \sum _ { \substack { \Sigma ^ { \alpha \beta } ( k, \omega ) = - \frac { z ^ { 2 } W ^ { 2 } } { N } \sum _ { \mathbf q, \gamma, \gamma ^ { \prime } } M _ { k, \mathbf q } ^ { \alpha \gamma } G ^ { \gamma \gamma ^ { \prime } } ( k - \mathbf q, \omega - \omega _ { \mathbf q } ) M _ { k, \mathbf q } ^ { \gamma ^ { \prime } \beta } }, \\ ( 6 ) }$$
with the interacting Green's function conveniently expressed via its inverse matrix G -1 as glyph[negationslash]
$$\left \{ G ^ { - 1 } ( { \mathbf q }, \omega ) \right \} ^ { \alpha \alpha } & = \omega + i \delta - \epsilon _ { \mathbf q } ^ { \alpha } - \Sigma ^ { \alpha \alpha } ( { \mathbf q }, \omega ), \quad ( 1 5 ) \\ \left \{ G ^ { - 1 } ( { \mathbf$$
The small imaginary part iδ was here chosen as δ = 5 meV. Many of the diagrams neglected in thin noncrossing approximation drop out for the square-lattice N´ eel state, however, it underestimates quantum fluctuations of the background.
The resonant inelastic x-ray scattering (RIXS) spectra can be simulated from the spectral density -I G αβ ( q , ω ) with proper consideration of the factors arising from the scattering geometry. The RIXS intensity can be expressed as
$$I = - | \beta | ^ { 2 } \mathcal { I } G ^ { \text{BB} } ( \mathbf q, \omega ) - | \gamma | ^ { 2 } \mathcal { I } G ^ { \text{CC} } ( \mathbf q, \omega ) \quad ( 1 6 ) \\ - \beta ^ { * } \gamma \mathcal { I } G ^ { \text{BC} } ( \math$$
where coefficients β and γ are obtained in a similar manner as in Ref. 1, extended here to account for the deviations from the cubic limit. They depend on incoming and outgoing polarizations glyph[epsilon1] and glyph[epsilon1] ′ as well as on the spin of the excited particle. As the spin does not enter the calculation of the spectral weight and the outgoing polarization glyph[epsilon1] ′ is not analyzed, we average over both. The matrix elements are given by
$$\beta _ { \sigma, \sigma } = \frac { 1 } { 2 } \cos \theta _ { 0 } \cos ( \theta - \theta _ { 0 } ) ( Q _ { 2 } + i \sigma T _ { z } ), \text{ \quad \ \ } ( 1 7 )$$
$$\beta _ { \sigma, - \sigma } & = \frac { \varepsilon } { 2 \sqrt { 2 } } \sin \theta _ { 0 } \cos ( \theta - \theta _ { 0 } ) \left [ - \sigma ( T _ { y } + P _ { y } ) + i ( T _ { x } - P _ { x } ) \right ], \\ & \quad \quad \quad$$
$$\gamma _ { \sigma, \sigma } & = \frac { \sin ( 2 \theta _ { 0 } - 2 \theta ) } { 4 \sqrt { 3 } } ( \sqrt { 2 } Q _ { 1 } + Q _ { 3 } ) + \frac { \sqrt { 3 } \sin 2 \theta } { 4 } Q _ { 3 } & \quad \text{The aI} \\ & \quad + i \$$
$$\gamma _ { \sigma, - \sigma } = \frac { 1 } { 2 \sqrt { 2 } } \left [ \sigma ( T _ { y } - \cos 2 \theta P _ { y } ) + i ( T _ { x } + \cos 2 \theta P _ { x } ) \right ] \begin{array} { c } \frac { \text{and} \colon \text{$\mu$} } { \text{the B}$} \\ \text{cel out} \\ ( 2 0 ) \end{array}$$
## Supplementary References
- 1 J. Kim, D. Casa, M. Upton, T. Gog, Y.-J. Kim, J. Mitchell, M. Van Veenendaal, M. Daghofer, J. van den Brink, G. Khaliullin, and B. Kim, Phys. Rev. Lett. 108 , 177003 (2012).
- 2 K. Wohlfeld, M. Daghofer, S. Nishimoto, G. Khaliullin, and J. van den Brink, Phys. Rev. Lett. 107 , 147201 (2011).
- 3 S. Schmitt-Rink, C. Varma, and A. Ruckenstein, Phys. Rev. Lett. 60 , 2793 (1988).
- 4 G. Jackeli and G. Khaliullin, Phys. Rev. Lett. 102 , 017205
where θ 0 is the 'mixing angle' in the cubic limit, which applies to the oxygen levels that are not affected by octahedral distortions and is given by tan 2 θ 0 = 2 √ 2. Further,
$$\vec { P } = \epsilon ^ { \prime } \times \epsilon$$
$$Q _ { 1 } = \sqrt { 2 / 3 } \epsilon \cdot \epsilon ^ { \prime }$$
$$Q _ { 2 } = \epsilon _ { y } ^ { \prime } \epsilon _ { y } - \epsilon _ { x } ^ { \prime } \epsilon _ { x },$$
$$Q _ { 3 } = \frac { \mu } { \sqrt { 3 } } ( \vec { \epsilon ^ { \prime } } \cdot \vec { \epsilon } - 3 \epsilon ^ { \prime } _ { z } \epsilon _ { z } ), \quad \quad ( 2 4 )$$
$$T _ { x } = \epsilon _ { y } ^ { \prime } \epsilon _ { z } + \epsilon _ { z } ^ { \prime } \epsilon _ { y },$$
$$T _ { y } = \epsilon _ { x } ^ { \prime } \epsilon _ { z } + \epsilon _ { z } ^ { \prime } \epsilon _ { x },$$
$$T _ { z } = \epsilon _ { y } ^ { \prime } \epsilon _ { x } + \epsilon _ { x } ^ { \prime } \epsilon _ { y }.$$
We approximate the normal incidence scattering geometry in Fig. 2b by taking φ =90 . The resulting spectra are ◦ shown in Supplementary Fig. 1a. A comparison to the orbital-resolved spectra shown in Supplementary Figs. 1c and 1d reveals that the spectral weight is dominated by the B doublet. For the φ =0 ◦ geometry the β -factors cancel out and the resulting spectrum, shown in Supplementary Fig. 1b, shows only the higher-energy C mode.
(2009).
- 5 Z. Liu and E. Manousakis, Phys. Rev. B 45, 2425 (1992).
- 6 K. Wohlfeld, M. Daghofer, G. Khaliullin, and J. van den Brink, J. Phys.: Conf. Ser. 391 , 012168 (2012).
- 7 V. M. Katukuri, H. Stoll, J. van den Brink J., and L. Hozoi, Phys. Rev. B 85 , 220402(R) (2012).
- 8 S. Fujiyama et al. , arXiv:1308.0923; S. Bahr et al. , private communication.
- 9 J. Baglyph[suppress]a, l A. M. Ole´ s, and J. Zaanen, Phys. Rev. B 52 , 4597 (1995).
|
10.1038/ncomms5453
|
[
"Jungho Kim",
"M. Daghofer",
"A. H. Said",
"T. Gog",
"J. van den Brink",
"G. Khaliullin",
"B. J. Kim"
] |
2014-08-04T20:02:30+00:00
|
2014-08-04T20:02:30+00:00
|
[
"cond-mat.str-el"
] |
Excitonic quasiparticles in a spin-orbit Mott insulator
|
In condensed matter systems, out of a large number of interacting degrees of
freedom emerge weakly coupled particles, in terms of which most physical
properties are described. For example, Landau quasiparticles (QP) determine all
electronic properties of a normal metal. The lack of identification of such QPs
is major barrier for understanding myriad exotic properties of correlated
electrons, such as unconventional superconductivity and non-Fermi liquid
behaviours. Here, we report the observation of a composite particle in a Mott
insulator Sr2IrO4---and exciton dressed with magnons---that propagates with the
canonical characteristics of a QP: a finite QP residue and a lifetime longer
than the hopping time scale. The dynamics of this charge-neutral bosonic
excitation mirrors the fundamental process of the analogous one-hole
propagation in the background of ordered spins, for which a well-defined QP has
never been observed. The much narrower linewidth of the exciton reveals the
same intrinsic dynamics that is obscured for the hole and is intimately related
to the mechanism of high temperature superconductivity.
|
1408.0805v3
|
Electron. J. Probab. 20 (2015), no. 32, 1-16.
ISSN: 1083-6489 DOI: 10.1214/EJP.v20-3881
## Subcritical contact process seen from the edge: Convergence to quasi-equilibrium
Enrique Andjel * François Ezanno † Pablo Groisman ‡ Leonardo T. Rolla ‡
## Abstract
The subcritical contact process seen from the rightmost infected site has no invariant measures. We prove that nevertheless it converges in distribution to a quasi-stationary measure supported on finite configurations.
Keywords: Interacting random processes; statistical mechanics models; percolation theory. AMS MSC 2010: 60K35; 82C22.
Submitted to EJP on October 22, 2014, final version accepted on March 20, 2015.
## 1 Introduction
The contact process is a stochastic model for the spread of an infection among the members of a population. Individuals are identified with points of a lattice ( Z in our case) and the process evolves according to the following rules. An infected individual will infect each of its neighbors at rate λ > 0 , and recover at rate 1 . This evolution defines an interacting particle system whose state at time t is a subset η t ⊆ Z , or equivalently an element η t ∈ { 0 1 , } Z . We interpret that individual x ∈ Z is infected at time t if η t ( x ) = 1 , and is otherwise healthy.
The contact process is one of the simplest particle systems that exhibits a phase transition. There exists a critical value 0 < λ c < ∞ such that the probability that a single individual infects infinitely many others is zero when λ < λ c and is positive when λ > λ c . See [12] for the precise definition of the model.
For A ⊆ Z , let ( η A t ) t glyph[greaterorequalslant] 0 denote the process starting from η 0 = A . When A is random and has distribution µ , we denote the process by η µ t . We also write η x t when A = { x } . Let
$$\Sigma = \{ A \subseteq \mathbb { Z } \, \colon A \cap \mathbb { N } \, \text{is finite} \} \quad \text{and} \quad \Sigma ^ { ^ { * } } = \{ A \in \Sigma \, \colon A \cap - \mathbb { N } \, \text{is infinite} \}.$$
Notice that both Σ and Σ ∗ are invariant for the contact process dynamics. For A ∈ Σ , the contact process seen from the rightmost point is the Markov process on Σ defined by
$$\zeta _ { t } ^ { A } = \eta _ { t } ^ { A } - \max \eta _ { t } ^ { A }$$
* Université d'Aix-Marseille, visting IMPA supported by CNPq
† Lycée Roland Garros
‡ Universidad de Buenos Aires
glyph[negationslash]
if η A t = ∅ and ζ A t = ∅ otherwise. In fact, defining
Σ = 0 { A ∈ Σ : max A = 0 or A = ∅} and Σ ∗ 0 = { A ∈ Σ : 0 A is infinite } , the state-space of the process ( ζ t ) t is Σ 0 , and the subset Σ ∗ 0 is invariant.
Durrett [6] proved the existence of an invariant measure for ζ t when λ glyph[greaterorequalslant] λ c on Σ ∗ 0 . In the supercritical phase, Galves and Presutti [9] proved that the invariant measure µ is unique for each λ , and that ζ A t converges in distribution to µ for any A ∈ Σ ∗ 0 . Kuczek [11] provided an alternative proof and showed an invariance principle for the position of the rightmost infected site. Uniqueness of µ and convergence in distribution was extended to the critical case by Cox, Durrett, and Schinazi [5]. While some of these results were stated for the contact process and some others for planar oriented percolation, the arguments in [9, 11, 5] can be translated effortlessly from one model to the other, which is not always the case.
The behavior in the subcritical phase is quite different. Schonmann [14] showed that in this phase, planar oriented percolation seen from its rightmost point does not have any invariant measure on Σ ∗ 0 . This result was extended to the contact process by Andjel, Schinazi, and Schonmann [2].
In this paper we show that, despite non-existence of stationary measures, the subcritical contact process seen from the rightmost point does converge in distribution. The limiting measure is quasi-stationary and is supported on configurations that contain finitely many infected sites.
This extends an analogous result for subcritical planar oriented percolation [1]. The proof in [1] used quite heavily that in the discrete setting the speed of the propagation of the infection is bounded by 1 almost surely. Since this does not hold for the contact process, there is no simple adaptation of that proof to this case. The difficulty is mostly due to the fact that unlikely events may have considerable influence when we observe an event of small probability.
Hereafter we assume that 0 < λ < λ c is fixed.
For ζ 0 ∼ µ , define τ µ = inf { t glyph[greaterorequalslant] 0 : ζ µ t = ∅} . Define τ A and τ x analogously. We say that µ is a quasi-stationary distribution on Σ 0 if for every t > 0 the law of ζ µ t satisfies
$$\mathcal { L } ( \zeta _ { t } ^ { \mu } \, | \, \tau ^ { \mu } > t ) = \mu.$$
By the Markov property, the above identity implies that L ( τ µ | τ µ > t ) = L ( τ µ + ) t , so if µ is quasi-stationary, τ µ is exponentially distributed with some parameter α = 1 E [ τ µ ] glyph[greaterorequalslant] 0 . We say that ν is minimal if E [ τ ν ] is minimal among all quasi-stationary distributions. Notice that stationary is a particular case of quasi-stationary with α = 0 . See [16, 4, 13] for an introduction on this topic.
Proposition 1.1. The subcritical contact process seen from the rightmost point ( ζ t ) t glyph[greaterorequalslant] 0 has a unique minimal quasi-stationary distribution ν . This measure ν is supported on finite configurations. Moreover it satisfies the Yaglom limit
$$\mathcal { L } ( \zeta _ { t } ^ { A } \, | \, \tau ^ { A } > t ) \to \nu \, \text{as} \, t \to \infty,$$
for any finite configuration A ⊆ Z .
An analogous result was obtained by Ferrari, Kesten, and Martínez [8] for a class of probabilistic automata that includes planar oriented percolation. The main step of their proof is to show that the transition matrix is R -positive with left eigenvector ν summable. In our proof we show that the contact process observed at discrete times falls in that class, and then apply standard theory of α -positive continuous-time Markov chains to obtain the Yaglom limit. In Section 3 we state and prove a more general version of the above proposition, valid on Z d .
We finally state our main result.
Theorem 1.2. For every infinite initial configuration A ⊆ -N , the subcritical contact process seen from the rightmost point ζ A t converges in distribution to ν as t →∞ .
Theorem 1.2 is proved in Section 2 using Proposition 1.1. A natural attempt to get Theorem 1.2 would be to consider the rightmost site x ∈ -N whose infection survives up to time t , and simply apply Proposition 1.1 to the set ζ x t of sites infected by x at time t . However, the choice of x as the first surviving site brings more information than simply ' τ x > t '. We define a sort of renewal space-time point in order to handle this extra information, and finally show that such point exists with high probability.
Some of the main arguments in this paper come from the second author's thesis [7].
## 2 The set infected by an infinite configuration
In this section we prove Theorem 1.2. Subsection 2.1 describes the graphical construction of the contact process, and in Subsection 2.2 we recall the FKG and BK inequalities for this construction.
In Subsection 2.3 we introduce the definitions of a good space-time point, and a break point , for fixed time t . The presence of a break point neutralizes the inconvenient information mentioned at the end of Section 1, provided that all points nearby are good. Choosing some constants correctly, it turns out that most points are good, even when considering rare regions such as those where an infection happens to survive until time t . We conclude this subsection with the proof of Theorem 1.2, assuming existence of break points.
In Subsection 2.4 we prove that a break point can be found with high probability as t →∞ . To that end we use again geometric properties of good points and the exponential decay of subcritical contact processes.
## 2.1 Graphical construction
Define L = Z + {± 1 / 3 } and let U be a Poisson point process in R 2 with intensity given by ( ∑ y ∈ Z δ y + ∑ y ∈ L λδ y ) × d t. Notice that U ⊆ ( Z ∪ L ) × R . Let (Ω , F , P ) be the underlying probability space. For x ∈ Z we write U x,x ± 1 = U ∩ { ( x ± 1 / 3 } × R ) and U x = U ∩ { } × ( x R ) .
Given two space-time points ( y, s ) and ( x, t ) , we define a path from ( y, s ) to ( x, t ) as a finite sequence ( x , t 0 0 ) , . . . , ( x , t k k ) with x 0 = x , x k = y , s = t 0 glyph[lessorequalslant] t 1 glyph[lessorequalslant] . . . glyph[lessorequalslant] t k = t with the following property. For each i = 1 , . . . , k , the i -th segment [( x i -1 , t i -1 ) , ( x , t i i )] is either vertical , that is, x i = x i -1 , or horizontal , that is, | x i -x i -1 | = 1 and t i = t i -1 . Horizontal segments are also referred to as jumps . If all horizontal segments satisfy t i = t i -1 ∈ U x i -1 ,x i then such path is also called a λ -path . If, in addition, all vertical segments satisfy ( t i -1 , t i ] ∩ U x i = ∅ we call it an open path from ( y, s ) to ( x, t ) .
The existence of an open path from ( y, s ) to ( x, t ) is denoted by ( y, s ) glyph[squiggleright] ( x, t ) . Also for two sets of the plane C D , we use { C glyph[squiggleright] D } = { ( y, s ) glyph[squiggleright] ( x, t ) for some ( y, s ) ∈ C, ( x, t ) ∈ D } . We denote by L t the set Z ×{ } ⊆ t R 2 .
For A ∈ Σ 0 , we define η A s,t ∈ Σ and ζ A s,t ∈ Σ 0 by
$$\eta _ { s, t } ^ { A } = \{ x \colon ( A \times \{ s \} ) \sim \tilde { \sim } \, ( x, t ) \}, \quad \zeta _ { s, t } ^ { A } = \eta _ { s, t } ^ { A } - \max \eta _ { s, t } ^ { A } \, \text{ \quad \quad (2.1)}$$
glyph[negationslash]
if η A s,t = ∅ and ζ A s,t = ∅ otherwise. When s = 0 we omit it. We use ( η t ) and ( ζ t ) for the processes defined by (2.1), that is ( η t ) is a contact process with parameter λ and ( ζ t ) is this process as seen from the rightmost infected site. Both of them are Markov. Note that if A is finite, the same holds for η A t and ζ A t for every t glyph[greaterorequalslant] 0 with total probability. Also note that ∅ is absorbing for both processes. When A is a singleton { y } we write η y t and ζ y t .
## 2.2 FKG and BK inequalities
We use ω for a configuration of points in R 2 and ω δ , ω λ for its restrictions to Z × R and L × R respectively. We write ω glyph[greaterorequalslant] ω ′ if ω ′ λ ⊆ ω λ and ω δ ⊆ ω ′ δ .
.
We slightly abuse the notation and identify a set of configurations Q with ( U -1 Q ) ⊆ Ω A minor topological technicality needs to be mentioned. Consider the space of locally finite configurations with the Skorohod topology: two configurations are close if they have the same number of points in a large space-time box and the position of the points are approximately the same. In the sequel we assume that all events considered have zero-probability boundaries under this topology. The important fact is that events of the form { E glyph[squiggleright] F } are measurable and satisfy this condition, as long as E and F are bounded Borel subsets of Z × R . See [3, Sect. 2.1] for a proof and precise definitions.
Definition 2.1. A set of configurations Q is increasing if ω glyph[greaterorequalslant] ω ′ ∈ Q implies ω ∈ Q .
Theorem (FKG Inequality) . If Q 1 and Q 2 are increasing, then P ( Q 1 ∩ Q 2 ) glyph[greaterorequalslant] P ( Q 1 ) P ( Q 2 ) .
Definition 2.2. Let D denote a Borel subset of R 2 . For a given configuration ω , we say that Q occurs on D if ω ′ ∈ Q for any configuration ω ′ such that ω ′ ∩ D = ω ∩ D .
Definition 2.3. We way that Q 1 and Q 2 occur disjointly if there exist disjoint sets D 1 and D 2 such that Q 1 occurs on D 1 and Q 2 occurs on D 2 . This event is denoted by Q 1 2 Q 2 .
For the proofs, see for instance [3, Sect. 2.2].
Theorem (BK Inequality) . If Q 1 and Q 2 are increasing, and depend only on the configuration ω within a bounded domain, then P ( Q 1 2 Q 2 ) glyph[lessorequalslant] P ( Q 1 ) P ( Q 2 ) .
## 2.3 Good points and break points
The definitions below are parametrized by t > 0 and β > 0 , but we omit it in the notation. We write βt as a short for glyph[ceilingleft] βt glyph[ceilingright] . For simplicity we assume through this whole section that the initial configuration A ∈ Σ ∗ 0 is fixed.
Definition 2.4 (Good point) . We say that ( z, s ) is good , an event denoted by G s z , if every λ -path starting at ( z, s ) makes less than βt jumps during [ s, s + ] t . We also denote ˆ G s z := G s z ∩ G s z +2 βt . The time s is omitted when s = 0 . We will say that ( z, s ) is ( β, t ) -good when we need to make β and t explicit.
The above definition is helpful in this continuous-time context, as a way to recover independence of connectivity at distant regions. As an example, observe that even though the events { 0 glyph[squiggleright] L t } and { (2 βt, 0) glyph[squiggleright] L t } are not independent, they are conditionally independent given ˆ G 0 . Hereafter we write 0 for the space-time point (0 , 0) . Moreover, each point is typically good, so that conditioning on a large set of points to be good has negligible impact on the underlying distribution.
glyph[negationslash]
Definition 2.5 (Break point) . We say that the space-time point ( y, s ) is a break point if, for every w ∈ ( y, y +2 βt ] ∩ Z , L 0 glyph[squiggleright] ( w,s ) .
Let X := max { x ∈ A : ( x, 0) glyph[squiggleright] L t } denote the first site whose infection survives up to time t , and let Γ : [0 , t ] → Z given by
$$\Gamma ( s ) \coloneqq \max \{ y \colon ( X, 0 ) \sim \left ( y, s \right ) \sim L _ { t } \}$$
denote the 'rightmost path' from ( X, 0) to L t . Take
$$R \coloneqq \inf \{ s \in [ 0, t ] \, \colon ( s, \Gamma ( s ) ) \, \text{is a break point} \}$$
as the time of the first break point in Γ , and let Y := Γ( R ) , see Figure 1. Finally take
$$\mathcal { A } \coloneqq \{ x \leqslant 0 \, \colon A \times \{ 0 \} \sim ( Y + x, R ) \}$$
Figure 1: The set A is represented by squared points, and X ∈ A is its rightmost point connected to L t . Among all paths from A ×{ } 0 to L t , the curve Γ (in bold) is the rightmost one. The space-time point ( Y, R ) ∈ Γ is a break point, i.e., no points in L 0 are connected to the horizontal segment with width 2 βt lying to the right of ( Y, R ) . It is also the first point of Γ with such property.
as the set of sites infected at time R , lying to the left of Γ , seen from Γ .
Here is a sketch on how the above objects will be used to prove Theorem 1.2. We want to compare L ( ζ A t ) and ν . The main property of break points is that
$$\mathcal { L } \left ( \zeta _ { t } ^ { A } \, \Big | \, R = s, Y = y, \hat { G } _ { s } ^ { y } \, \right ) = \mathcal { L } \left ( \zeta _ { t - s } ^ { A } \, \Big | \, 0 \sim L _ { t - s }, G _ { 0 } \, \Big ) \,,$$
which will be explained with Figure 2. Another important property is that, with high probability, R < t 2 , so one can think of t -s as being a large number. Using this and BK inequality we can show that
$$\mathcal { L } \left ( \zeta _ { t - s } ^ { \mathcal { A } } \left | 0 \sim L _ { t - s }, G _ { 0 } \right ) \approx \mathcal { L } \left ( \zeta _ { t - s } ^ { 0 } \left | 0 \sim L _ { t - s } \right ),$$
and the latter converges to ν by Proposition 1.1.
In the sequel we state these two properties.
Proposition 2.6. If β is large enough, then P ( R glyph[lessorequalslant] t 2 ) → 1 as t →∞ .
Proposition 2.6 is proved in Section 2.4. The first lemma below describes a regular conditional distribution 1 of ζ A t given A , Y and R , on the event that some points are good.
Lemma 2.7. For any s ∈ [0 , t ] , y ∈ Z , and A ′ ∈ Σ ∗ 0 ,
$$\mathcal { L } \left ( \zeta _ { t } ^ { A } \, | \, \mathcal { A } = A ^ { \prime }, Y = y, R = s, \hat { G } _ { y } ^ { s } \right ) = \mathcal { L } \left ( \zeta _ { t - s } ^ { A ^ { \prime } } \, | \, 0 \sim L _ { t - s }, G _ { 0 } \right ).$$
In the sequel we state two lemmas that will fill the remaining technical gaps. We then prove Theorem 1.2 and finally the lemmas.
Lemma 2.8. If β is large enough, then, as t →∞ ,
$$\frac { \mathbb { P } \left ( G _ { X }, \hat { G } _ { Y } ^ { R } \right ) \to 1 \text{ and } \sup _ { s \in [ 0, t ] } \sup _ { A \in \Sigma _ { 0 } } \left \| \mathcal { L } \left ( \zeta _ { s } ^ { A } \left | 0 \sim L _ { s }, G _ { 0 } \right ) - \mathcal { L } \left ( \zeta _ { s } ^ { A } \left | 0 \sim L _ { s } \right ) \right \| _ { T V } \to 0.$$
1 The random elements considered in this paper are a graphical construction U and sometimes a random initial condition η 0 , both given by locally finite subsets of an Euclidean space. Therefore we can assume that (Ω , F ) is a Polish space, and as a consequence regular conditional probabilities exist. In particular, conditioning on events such as { R = s } , { Γ = γ } , etc. is well defined.
Lemma 2.9. As t →∞ ,
$$\sup _ { A \in \Sigma _ { 0 } } \left \| \mathcal { L } \left ( \zeta _ { t } ^ { A } \cap \left [ - 2 t, 0 \right ] \left | \, 0 \, \sim \, L _ { t } \right ) - \mathcal { L } \left ( \zeta _ { t } ^ { 0 } \cap \left [ - 2 t, 0 \right ] \left | \, 0 \, \sim \, L _ { t } \right ) \right \| _ { T V } \to 0.$$
Theorem 1.2 can now be proved using the preceding results.
Proof of Theorem 1.2. For a signed measure µ on { 0 1 , } Z , we use ‖ µ ‖ = ‖ µ ‖ TV to denote the total variation norm. Denote H s y := ˆ G s y ∩ { Y = y, R = s } . Given A ∈ Σ ∗ 0 ,
$$\lim _ { t \to \infty } \psi \left \| \mathcal { L } \left ( \zeta _ { t } ^ { A } \cap [ - t, 0 ] \right ) - \nu \right \| \leqslant$$
$$\leqslant \lim \sup _ { t \to \infty } \sup _ { \frac { t } { 2 } \leqslant r \leqslant t } \left \| \mathcal { L } \left ( \zeta _ { r } ^ { 0 } \cap \left [ - t, 0 \right ] \left | \mathbf 0 \, \stackrel { \sim } { \sim } L _ { r } \right ) - \nu \right \|.$$
$$\text{the total variation norm. Denote } H _ { y } ^ { s } \coloneqq \hat { G } _ { y } ^ { s } \cap \{ Y = y, R = s \}. \, \text{Given } \hat { A } \in \Sigma _ { 0 } ^ { s }, \\ \limsup \| \mathcal { L } \left ( \zeta _ { t } ^ { A } \cap [ - t, 0 ] \right ) - \nu \| \leqslant \\ t \rightarrow \infty \, \right ) \, \text{P} \left ( \hat { G } _ { y } ^ { R }, R < \frac { 1 } { 2 } \right ) \, \| \mathcal { L } \left ( \zeta _ { t } ^ { A } \cap [ - t, 0 ] \, | \, \hat { G } _ { y } ^ { R }, R < \frac { 1 } { 2 } \right ) - \nu \| + \left [ 1 - \text{P} \left ( \hat { G } _ { y } ^ { R }, R < \frac { 1 } { 2 } \right ) \, \right ] \\ = \limsup \, \left \| \int _ { Z \times [ 0, \frac { 1 } { 2 } ] } \left [ \mathcal { L } \left ( \zeta _ { t } ^ { A } \cap [ - t, 0 ] \, | \, H _ { y } ^ { s } \right ) - \nu \right ] \, \text{D} \left ( Y = y, R = s \, | \, \hat { G } _ { y } ^ { R }, R < \frac { 1 } { 2 } \right ) \, \right \| \\ \leqslant \limsup \, \sup \, \sup \, \sup \, \left \| \mathcal { L } \left ( \zeta _ { t } ^ { A } \cap [ - t, 0 ] \, | \, H _ { y } ^ { s } \right ) - \nu \right \| \\ = \limsup \, \sup \, \sup \, \sup \, \subseteqslant \frac { 1 } { 2 } \, y \in \Sigma _ { 0 } \left \| \int _ { Z \times [ 0, \frac { 1 } { 2 } ] } \left [ \mathcal { L } \left ( \zeta _ { t } ^ { A } \cap [ - t, 0 ] \, | \, H _ { y } ^ { s }, A = A ^ { \prime } \right ) - \nu \right ] \, \text{D} \left ( A = A ^ { \prime } | H _ { y } ^ { s } \right ) \right \| \\ = \limsup \, \sup \, \sup \, \subseteqslant \frac { 1 } { 2 } \, y \in \Sigma _ { 0 } \left \| \mathcal { L } \left ( \zeta _ { t } ^ { A ^ { \prime } } - 2 \cap [ - t, 0 ] \, | \, 0 \longrightarrow \, L _ { t - s } \right ) - \nu \right \| \\ \leqslant \limsup \, \sup \, \sup \, \sup \, \subseteqslant \frac { 1 } { 2 } \, \left \| \mathcal { L } \left ( \zeta _ { t } ^ { 0 } \cap [ - t, 0 ] \, | \, 0 \longrightarrow \, L _ { r } \right ) - \nu \right \|.$$
On the first equality we used Proposition 2.6 and Lemma 2.8. On the third equality we used Lemma 2.7. The last two inequalities are due to Lemma 2.8 and Lemma 2.9, respectively. The last lim sup vanishes by Proposition 1.1.
We proceed to prove the previous lemmas.
Proof of Lemma 2.7. Consider the regions
$$E _ { y, s } ^ { + } = ( - \infty, y + \beta t ] \times ( s, 2 t ], \quad E _ { y, s } ^ { - } = \mathbb { R } \times [ 0, 2 t ] \, \sum E _ { y, s } ^ { + },$$
and the random variables X y,s = max { x ∈ A : ( x, 0) glyph[squiggleright] ( y, s ) } , the first site whose infection reaches ( y, s ) and Γ y,s : [0 , s ] → Z given by
$$\Gamma _ { y, s } ( u ) = \max \{ z \, \colon ( X _ { y, s }, 0 ) \sim \left ( z, u \right ) \sim \left ( y, s \right ) \},$$
the rightmost path from ( X y,s , 0) to ( y, s ) . Before continuing with the proof, the reader may see Figure 2 to have a glance of the argument.
For a configuration η ⊆ Z we use the following convention: η · 1 = η and η · 0 = ∅ . We have
$$\mathcal { L } \left ( \zeta _ { t } ^ { A } \, | \, \mathcal { A } = A ^ { \prime }, Y = y, R = s, \hat { G } _ { y } ^ { s } \right ) = \mathcal { L } \left ( \zeta _ { s, t } ^ { A ^ { \prime } + y } \cdot \mathbf 1 ( G _ { y } ^ { s } ) \, | \, \mathcal { A } = A ^ { \prime }, ( y, s ) \sim L _ { t }, H, \hat { G } _ { y } ^ { s } \right ),$$
where glyph[negationslash]
$$H = & \{ ( A \times \{ 0 \} ) \stackrel { \sim } { \sim } ( y, s ) \} \cap \{ L _ { 0 } \not \to ( y, y + 2 \beta t ] \times \{ s \} \} \\ & \cap \bigcap _ { u \in [ 0, s ) } \{ [ X _ { y, s }, + \infty ) \times \{ 0 \} \stackrel { \sim } { \sim } ( \Gamma _ { y, s } ( u ), \Gamma _ { y, s } ( u ) + 2 \beta t ] \times \{ u \} \} \\ & \cap \{ \eta _ { s } ^ { A } \cap ( y + 2 \beta t, + \infty ) \times \{ s \} \not \to L _ { t } \}.$$
A
Figure 2: The event A = A , Y ′ = y, R = s, G ˆ s y . This event is split in two parts. The first part depends on the region E + y,s , in light gray, and consists of the occurrence of both G s y and { ( y, s ) glyph[squiggleright] L t } . A consequence of the former is that the rightmost λ -path starting at ( y, s ) , depicted by a dash-dotted line, does not reach distance βt by time t , and the latter is represented by a solid arrowed line from ( y, s ) to L t . The second part depends on the region E -y,s , in dark gray, and consists of the occurrence of both H ∩ G s y +2 βt , as well as an open path from A × { } 0 to each point in y + A ′ . The latter is depicted by solid arrowed lines from A at time 0 to y + A ′ at time s . The event H breaks down to the following conditions being satisfied. First, there are open paths from A to ( y, s ) , of which the rightmost one Γ y,s ( ) · is depicted in a thick arrowed line. Second, there is no open path from L 0 to the segment of size 2 βt to the right of ( y, s ) , and moreover s is the first time with this property, i.e., there are open paths from L 0 to Γ y,s ( u ) + z u ( ) for some 0 glyph[lessorequalslant] z u ( ) glyph[lessorequalslant] 2 βt for all u < s . Third, there are no connections from A ×{ } 0 to L t to the right of Γ y,s . Even though the third condition might not depend only on the region E -y,s , it is the case when G s y +2 βt occurs, since it implies that the leftmost path starting from ( y +2 βt, s ) , also depicted by a dash-dotted line, does not reach distance βt by time t . Finally, E + y,s and E -y,s are disjoint, and under the occurrence of G s y the configuration ζ A t = ζ A ′ s,t depends only on E + y,s . Therefore only the second part ( G s y ∩ { ( y, s ) glyph[squiggleright] L t } ) influences its distribution.
Observe that H ∩ {A = A ′ } ∩ G s y +2 βt depends on U ∩ E -y,s . Since ζ A ′ s,t · 1 ( G s y ) depends on U ∩ E + y,s , we have
$$& \mathcal { L } \left ( \zeta _ { t } ^ { A } \left | \mathcal { A } = A ^ { \prime }, Y = y, R = s, \hat { G } _ { y } ^ { s } \right ) = \mathcal { L } \left ( \zeta _ { s, t } ^ { A ^ { \prime } + y } \cdot 1 ( G _ { y } ^ { s } ) \right | ( y, s ) \sim L _ { t }, G _ { y } ^ { s } \right ) = \\ & \mathcal { L } \left ( \zeta _ { s, t } ^ { A ^ { \prime } + y } \left | ( y, s ) \sim L _ { t }, G _ { y } ^ { s } \right ) = \mathcal { L } \left ( \zeta _ { t - s } ^ { A ^ { \prime } } \left | \mathbf 0 \sim L _ { t - s }, G _ { 0 } \right ),$$
by translation invariance.
Proof of Lemma 2.8. The two limits hold for similar reasons. First notice that the probability that ( y, 0) glyph[squiggleright] L t by a straight vertical path is e -t and that these events are independent over y .
Let x k ( ) denote the k -th point of A from the right. By independence, P [ X < x k ( )] glyph[lessorequalslant] (1 -e -t ) k glyph[lessorequalslant] e -ke -t and writing x r ( ) = x ( glyph[floorleft] r glyph[floorright] ) we have P ( X < x e ( 2 t ) ) → 0 . Observe that if ( y, 0) is (2 t, β 2 ) -good then ( y, s ) is ( t, β ) -good for every s ∈ [0 , t ] . So we can pick a ρ > 3 and a β according to Lemma 2.10 below to obtain P ( ˆ G s y for all s ∈ [0 , t ] ) > 1 -2 e -ρt for large enough t . Hence,
$$\mathbb { P } ( \hat { G } _ { y } ^ { s } \text{ for all } y \in A \cap [ x ( e ^ { 2 t } ), 0 ] \text{ and } s \in [ 0, t ] ) \rightarrow 1,$$
and the first limit holds.
Finally, P ( G c 0 ) glyph[lessorequalslant] e -ρt glyph[lessmuch] e -t glyph[lessorequalslant] e -s glyph[lessorequalslant] P ( 0 glyph[squiggleright] L s ) , and therefore
$$\lim _ { t \to \infty } \inf _ { s \leqslant t } \mathbb { P } ( G _ { 0 } \, | \, 0 \sim L _ { s } ) = 1,$$
proving the second limit.
Lemma 2.10. For every ρ < ∞ , one can choose β < ∞ such that
$$\mathbb { P } ( 0 \text{ is good} ) > 1 - e ^ { - \rho t } \text{ for large enough } t.$$
Proof. Given the Poisson process U , the λ -paths starting at 0 can be constructed by choosing at each jump mark whether or not to follow that arrow. This way each finite path is associated to a finite binary sequence a ∈ { 0 1 , } n for some n ∈ N .
The λ -path corresponding to a finite sequence a makes | a | := ∑ n i =1 a i jumps. Such path is performed in time T a , whose distribution is that of the sum of n independent exponential random variables with parameter 2 λ .
Since
$$\mathbb { P } ( T _ { a } \leqslant t ) = e ^ { - 2 \lambda t } \sum _ { k = n } ^ { \infty } \frac { ( 2 \lambda t ) ^ { k } } { k! } \leqslant \frac { ( 2 \lambda t ) ^ { n } } { n! } \\ \theta \quad. \quad. \quad. \quad. \quad.$$
for every a ∈ { 0 1 , } n , choosing β > max 12 { λe, ρ } , we have by Stirling's approximation
$$\L _ { n } ( 2 \lambda t ) ^ { n }$$
$$\mathbb { P } ( 0 \text{ is not good} ) \leqslant \sum _ { | a | \geqslant \beta t } \mathbb { P } ( T _ { a } \leqslant t ) \leqslant \sum _ { n \geqslant \beta t } \sum _ { a \in \{ 0, 1 \} ^ { n } } \mathbb { P } ( T _ { a } \leqslant t ) = \sum _ { n \geqslant \beta t } 2 ^ { n } \frac { ( 2 \lambda t ) ^ { n } } { n! } \leqslant \\ \leqslant \sum _ { n \geqslant \beta t } ( 4 \lambda t ) ^ { n } \left ( \frac { 3 } { n } \right ) ^ { n } \leqslant \sum _ { n \geqslant \beta t } \left ( \frac { 1 2 \lambda t } { 1 2 \lambda t e } \right ) ^ { n } \leqslant \frac { 1 } { 1 - e ^ { - 1 } } e ^ { - \beta t } \leqslant e ^ { - \rho t } \\ \sim \sum _ { n \geqslant \beta t } \left ( 4 \lambda t ) ^ { n } \left ( \frac { 1 } { n } \right ) ^ { n } \leqslant \sum _ { n \geqslant \beta t } \left ( \frac { 1 2 \lambda t } { 1 2 \lambda t e } \right ) ^ { n } \leqslant \frac { 1 } { 1 - e ^ { - 1 } } e ^ { - \beta t } \leqslant e ^ { - \rho t } \\ \sim \quad \end{array}$$
for all t large enough.
glyph[negationslash]
Proof of Lemma 2.9. By monotonicity on A ′ ∈ Σ 0 , it suffices to show that P ( H | 0 glyph[squiggleright] L t ) → 0 as t →∞ , where H = { ζ A ′ t ( x ) = ζ 0 t ( x ) for some x ∈ -[ 2 t, 0] } and A ′ = -N 0 .
glyph[negationslash]
Assume that the events { 0 glyph[squiggleright] L t } and G 0 occur (the value of β will be fixed later). Then the rightmost point Y of η t satisfies -βt < Y < βt . Therefore, if H also occurs then ( A ′ ×{ } 0 ) glyph[squiggleright] ( z, t ) and 0 glyph[squiggleright] ( z, t ) for some z ∈ -[ 2 t -βt, βt ] , which in turn implies that ( A ′ ×{ } 0 ) glyph[squiggleright] ( z, t ) and 0 glyph[squiggleright] L t by disjoint paths. Let E n = [ -n, n ] × [0 , t ] . Using the BK inequality,
$$\lambda \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \rangle \\ \text{inequality}, \\ \mathbb { P } ( H \cap G _ { 0 } \, | \, 0 \, \longrightarrow \, L _ { t } ) & \leq \sum _ { z } \mathbb { P } \left ( ( A ^ { \prime } \times \{ 0 \} ) \, \longrightarrow \, ( z, t ) \, \square \, 0 \, \longrightarrow \, L _ { t } \, | \, 0 \, \longrightarrow \, L _ { t } \right ) \\ & = \sum _ { z } \lim _ { n } \mathbb { P } \left ( ( A ^ { \prime } \times \{ 0 \} ) \, \longrightarrow \, ( z, t ) \, \square \, 0 \, \longrightarrow \, L _ { t } \, | \, \sin \, E _ { n } \right ) \frac { \mathbb { P } \left ( 0 \, \longrightarrow \, L _ { t } \, \sin \, E _ { n } \right ) } { \mathbb { P } \left ( 0 \, \longrightarrow \, L _ { t } \right ) } \\ & = \sum _ { z } \mathbb { P } \left ( ( A ^ { \prime } \times \{ 0 \} ) \, \longrightarrow \, ( z, t ) \right ) \leq \, ( 2 \beta t + 2 t ) e ^ { - \alpha t }. \\ \text{Choosing} \, \beta \, \text{large enough} \, \text{so that} \, ( 2. 2 ) \, \text{holds, we get the desired limit}.$$
Choosing β large enough so that (2.2) holds, we get the desired limit.
## 2.4 Existence of break points
In this section we prove Proposition 2.6. To this end we show that there must be several time intervals where the path Γ is reasonably smooth, so that a break point is produced on each such piece with positive probability.
Definition 2.11 (Favorable time intervals) . Let γ be a path in the time interval [0 , t ] and let β > 0 . We say that a time interval [ s -√ t, s ) ⊆ [0 , t ] is favorable for path γ if for any u ∈ [ s -√ t, s ) the number of jumps of γ during [ u, s ) is at most 4 β s | -u | .
Lemma 2.12. Let γ be a path in the time interval [0 , t ] with at most βt jumps. Then there are at least √ t 4 -1 disjoint favorable intervals for γ contained in [0 , t 2 ] .
To prove this lemma we will seek favorable intervals in a top-down fashion, and use the fact that the existence of a non-favorable interval requires too many jumps. Notice that, on the event that X is good, any λ -path starting from ( X, 0) makes less than βt jumps, and in particular we can apply this lemma to the path Γ .
Define the sets C t = ( { x, 0) : x = 1 2 , , . . . , 2 βt } and
$$D _ { t } = \left \{ ( x, - u ) \, \colon x = \lfloor 4 \beta u \rfloor, u \in [ 0, \sqrt { t } ] \right \} \cup \left \{ ( x, - \sqrt { t } ) \, \colon x \geqslant 4 \beta \sqrt { t } \right \},$$
shown in Figure 3. The following fact is a direct consequence of exponential decay for the subcritical contact process. It will be proved in the end of this section for convenience.
Figure 3: Sets C t and D t
Lemma 2.13. For any β , let p β := sup t glyph[greaterorequalslant] 1 P ( D t glyph[squiggleright] C t ) . Then p β < 1 .
The key step in proving Proposition 2.6 is to observe that X is good with high probability, so that one can apply Lemma 2.12 combined with the following fact.
Lemma 2.14. If a path γ in the time interval [0 , t ] has at least k disjoint favorable intervals in [0 , t 2 ] , then P ( R glyph[lessorequalslant] t 2 ∣ ∣ Γ = γ ) glyph[greaterorequalslant] 1 -p k β for all t glyph[greaterorequalslant] 1 .
glyph[negationslash]
glyph[negationslash]
In order to prove the above lemma, we will attach a copy of C t and D t to disjoint pieces of Γ corresponding to a favorable time interval, and observe that, in order to find a break point, it suffices to have D t glyph[squiggleright] C t , see Figure 6. Knowing the path Γ gives negative information about connectivity properties to the right of Γ itself, which by the FKG inequality will increase the probability of the event D t glyph[squiggleright] C t .
We are ready to prove Proposition 2.6.
Proof of Proposition 2.6. Let H denote the event that Γ makes less than βt By Lemmas 2.12 and 2.14,
$$\mathbb { P } \left ( R \leqslant \frac { t } { 2 } \right ) \geqslant \mathbb { E } \left [ \mathbb { P } \left ( R \leqslant \frac { t } { 2 }, H \left | \, \mathbb { P } \right ) \right ] \geqslant \left ( 1 - p _ { \beta } \frac { \sqrt { 4 } - 1 } { 4 } - 1 \right ) \cdot \mathbb { P } ( H ).$$
jumps.
By Lemma 2.8, we can choose β large enough so that P ( H ) glyph[greaterorequalslant] P ( G X ) → 1 as t →∞ , and the result follows.
We finish this section with the proof of the lemmas above.
Proof of Lemma 2.12. We split time interval [0 , t 2 ) into a collection of favorable and non-favorable intervals as follows. Let t 0 = t 2 and let
$$v = \inf \left \{ u \geqslant 0 \, \colon \gamma \, \text{has more than } 4 \beta u \, \text{jumps in time interval } \left [ t _ { 0 } - u, t _ { 0 } \right ) \right \}.$$
If v > √ t , the interval [ t 0 -√ t, t 0 ) is favorable and we let t 1 = t 0 -√ t . If not, we declare the interval [ t 0 -v, t 0 ) non-favorable, and let t 1 = t 0 -v . We then continue with t 1 playing the role of t 0 , to find I 2 = [ t 2 , t 1 ) which may be favorable or non-favorable, and so on. This algorithm is performed until we reach a t i < √ t .
Let L denote the sum of the lengths of the non-favorable intervals among I 1 , . . . , I i . Note that a non-favorable interval of length glyph[lscript] has at least 4 βglyph[lscript] jumps, so 4 βL glyph[lessorequalslant] βt and therefore L glyph[lessorequalslant] t 4 . Hence the sum of the lengths of the favorable intervals among I 1 , . . . , I i is at least t 2 -√ t -t 4 and there must be at least √ t 4 -1 favorable intervals among the I j 's.
Proof of Lemma 2.14. Let D γ be the closed set given by the union of the horizontal and vertical segments of γ . Then ( R × [0 , t ]) \ D γ has two components: D + γ to the right and D -γ to the left.
We note that { Γ = γ } = { γ is open } ∩ H c γ , where
$$H _ { \gamma } = \left \{ L _ { 0 } \stackrel { \sim } { \sim } L _ { t } \, \text{in} \, D _ { \gamma } ^ { + } \right \} \cup \left \{ L _ { 0 } \stackrel { \sim } { \sim } \, \gamma \, \text{in} \, D _ { \gamma } ^ { + } \right \} \cup \left \{ \gamma \stackrel { \sim } { \sim } \, \gamma \, \text{in} \, D _ { \gamma } ^ { + } \right \}.$$
Here the last event means that there is an open path starting and ending at different points of γ , whose existence is determined by the configuration ω ∩ D + γ , see Figure 4.
Figure 4: The event H γ occurs if a path such as these four is open
On the other hand, { Γ = γ } ∩ { ( y, s ) is a break point } = Γ = { γ } ∩ J c ,
$$J _ { y, s, \gamma } = \left \{ \gamma \sim \left ( y, s \right ) + C _ { t } \text{ in } D _ { \gamma } ^ { + } \right \} \cup \left \{ L _ { 0 } \sim \left ( y, s \right ) + C _ { t } \text{ in } D _ { \gamma } ^ { + } \right \},$$
$$& \nolimits, s \right ) \text{ is a break point} \right \} = \left \{ \Gamma = \gamma \right \} \cap J _ { y, s, \gamma } ^ { c }, \text{ where} \\ & C _ { t } \text{ in } D _ { \gamma } ^ { + } \right \} \cup \left \{ L _ { 0 } \stackrel { \sim } { \sim } \left ( y, s \right ) + C _ { t } \text{ in } D _ { \gamma } ^ { + } \right \},$$
see Figure 5.
Now the event { γ is open } depends on U ∩ D γ and the events H γ and J y,s,γ depend on U ∩ D + γ . Since D γ and D + γ are disjoint,
$$\Gamma \left ( R \leq \frac { t } { 2 } \left | \Gamma = \gamma \right ) & = \Gamma \left ( J _ { y, s, \gamma } ^ { c } \text{ for some } s \in [ 0, \frac { t } { 2 } ], y = \Gamma ( s ) \left | \Gamma = \gamma \right ) \\ & = \Gamma \left ( J _ { y, s, \gamma } ^ { c } \text{ for some } s \in [ 0, \frac { t } { 2 } ], y = \gamma ( s ) \left | \Gamma = \gamma \right ) \\ & = \Gamma \left ( J _ { y, s, \gamma } ^ { c } \text{ for some } s \in [ 0, \frac { t } { 2 } ], y = \gamma ( s ) \left | \left \{ \gamma \text{ is open } \right \} \cap H _ { \gamma } ^ { c } \right ).$$
Figure 5: The event J y,s,γ occurs if a path such as these two is open
Finally, applying the FKG inequality to the last line,
$$\mathbb { P } \left ( R \leqslant \frac { t } { 2 } \left | \, \Gamma = \gamma \right ) \geqslant \mathbb { P } \left ( J _ { y, s, \gamma } ^ { c } \text{ for some } s \in [ 0, \frac { t } { 2 } ], y = \gamma ( s ) \right ).$$
From now on we drop the subindex γ from J . Let √ t glyph[lessorequalslant] t 1 < t 2 < · · · < t k glyph[lessorequalslant] t 2 be such that t j glyph[greaterorequalslant] t j -1 + √ t and [ t j -√ t, t j ) is a favorable interval for γ . Write z j = ( t j , γ ( t j )) . By definition of favorable interval and of the set D t , we have that z j + D t ⊆ D + γ ∪ D γ . On the other hand, if J z j occurs then z j + D t glyph[squiggleright] z j + C t , see Figure 6.
Figure 6: The event J z j implies that z j + D t glyph[squiggleright] z j + C t
Since these events depend on U ∩ ( R × ( t j -√ t, t j ] ) , which are disjoint as j goes from 1 to k , we have that
$$\mathbb { P } \left ( R \leqslant \frac { t } { 2 } \, | \, \Gamma = \gamma \right ) \geqslant 1 - \mathbb { P } \left ( z _ { j } + D _ { t } \, \stackrel { \sim } { \sim } \, z _ { j } + C _ { t } \, \text{for all } j \right ) = 1 - \mathbb { P } \left ( D _ { t } \, \stackrel { \sim } { \sim } \, C _ { t } \right ) ^ { k } \geqslant 1 - p _ { \beta } ^ { k }$$
$$\mathbb { P } \left ( R \leqslant \frac { t } { 2 } \left | 1 ^ { \ell } = \gamma \right ) \geqslant 1 - \mathbb { P } \left ( z _ { j } + D _ { t } \stackrel { \sim } { \sim } z _ { j } + C _ { t } \mathfrak { f i } \\ \text{by Lemma 2.13, which finishes the proof.}$$
Proof of Lemma 2.13. This lemma is a simple consequence of Lemma 2.15 below. We give a full proof for convenience. If D t glyph[squiggleright] C t then either ( x, -u ) glyph[squiggleright] C t for some x ∈ Z and u = √ t , or ( x, -u ) glyph[squiggleright] L + 0 for some x = 0 1 2 , , , . . . and u glyph[greaterorequalslant] x/ β 4 , where L + 0 = { 1 2 3 , , , . . . } × { 0 } . Using (2.3) and summing over y ∈ C t , the probability of the first event is bounded by 2 βte -α √ t . For the second event, using FKG inequality and
Lemma 2.15 below we get glyph[negationslash]
$$q _ { \beta } \coloneqq \mathbb { P } \left ( ( x, - u ) \not \sim L _ { 0 } ^ { + } \text{ for any } x = 0, 1, 2, \dots \text{ and } u \geqslant x / 4 \beta \right ) \geqslant \\ \geqslant \prod _ { \pi } \mathbb { P } \left ( ( x, - u ) \not \sim L _ { 0 } ^ { + } \text{ for any } u \geqslant x / 4 \beta \right ) > 0$$
glyph[negationslash]
$$x$$
does not depend on t . By the FKG inequality, p β glyph[lessorequalslant] sup t glyph[greaterorequalslant] 1 2 βte -α √ t (1 -q β ) < 1 .
Lemma 2.15. For large enough t , P ( L 0 glyph[squiggleright] (0 , s ) for some s glyph[greaterorequalslant] t ) glyph[lessorequalslant] e -αt/ 2 .
Proof. On the one hand the existence of a QSD ν , Proposition 1.1, implies
$$\mathbb { P } ( L _ { 0 } \stackrel { \sim } { \longleftrightarrow } ( 0, t ) ) = \mathbb { P } ( \mathbf 0 \stackrel { \sim } { \longleftrightarrow } L _ { t } ) = \mathbb { P } ( \tau ^ { 0 } > t ) \leqslant \mathbb { P } ( \tau ^ { \nu } > t ) = e ^ { - \alpha t }. \quad \quad ( 2. 3 )$$
On the other hand
$$\mathbb { P } \left ( L _ { 0 } \stackrel { \sim } { \longrightarrow } \left ( 0, t + 1 \right ) \right ) \geq \mathbb { P } \left ( L _ { 0 } \stackrel { \sim } { \longrightarrow } \left ( 0, s \right ) \text{ for some } s \in [ t, t + 1 ] \right ) \mathbb { P } \left ( U ^ { x } \cap [ t, t + 1 ] = \emptyset \right )$$
$$\text{and } \mathbb { P } \left ( U ^ { x } \cap [ t, t + 1 ] = \emptyset \right ) = e ^ { - 1 }, \text{whence}$$
$$\mathbb { P } \left ( L _ { 0 } \stackrel { \sim } { \sim } \left ( 0, s \right ) \text{ for some } s \geqslant t \right ) & \leqslant \sum _ { n } \mathbb { P } \left ( L _ { 0 } \stackrel { \sim } { \sim } \left ( 0, s \right ) \text{ for some } s \in \left [ t + n, t + n + 1 \right ] \right ) \\ & \leqslant e \cdot \sum _ { n } \mathbb { P } \left ( L _ { 0 } \stackrel { \sim } { \sim } \left ( 0, t + n + 1 \right ) \right ) \leqslant e ^ { - \alpha t / 2 }$$
for t large enough.
## 3 Yaglom limit for the set infected by a single site
In this section we prove Proposition 1.1, building upon Chapter 3 of the second author's PhD thesis [7].
We start by recalling some properties of jump processes on countable spaces which are almost-surely absorbed but positive recurrent when conditioned on non-absorption, known as R -positive or α -positive processes. In the sequel we define the finite contact process modulo translations, extending to Z d the concept of 'seen from the edge'. We then discretize time appropriately to obtain some moment control using exponential decay, showing that it satisfies some probabilistic criteria for R -positiveness which ultimately implies the desired result.
## 3.1 Positive recurrence of conditioned processes
Let Λ be a countable set and consider a Markov jump process ( ζ t ) t glyph[greaterorequalslant] 0 on Λ ∪ {∅} such that Λ is an irreducible class and ∅ is an absorbing state which is reached almost-surely. The sub-Markovian transition kernel restricted to Λ is written as P t ( A,A ′ ) = P ( ζ A t = A ′ ) , a matrix doubly-indexed by Λ and continuously parametrized by t .
A measure µ on Λ is seen as a row vector, and a real function f as a column vector, so that µP f t = E f ( ζ µ t ) . With this notation, µ is quasi-stationary if and only if µP t = e -α µ t ( ) µ . By [10, Theorem 1] there exists α > 0 with the property that t -1 log P t ( A,A ′ ) →-α as t →∞ for every A,A ′ ∈ Λ . The semi-group ( P t ) is said to be α -positive if
$$\lim _ { t \to \infty } e ^ { \alpha t } P _ { t } ( A, A ) > 0.$$
In this case, by [10, Theorem 4] there exist a measure ν and a positive function h , both unique modulo a multiplicative constant, such that
$$\nu P _ { t } = e ^ { - \alpha t } \nu, \quad P _ { t } h = e ^ { - \alpha t } h.$$
Moreover, νh < ∞ . If in addition ν is summable, then it can be normalized to become a probability measure on Λ , and the Yaglom limit follows from the result below.
Theorem 3.1. If an irreducible sub-Markovian standard semi-group ( P t ) t glyph[greaterorequalslant] 0 on a countable space Λ is α -positive with summable normalized left-eigenvector ν , then
$$\lim _ { t \to \infty } \frac { P _ { t } ( A, A ^ { \prime } ) } { P _ { t } ( A, \Lambda ) } = \nu ( A ^ { \prime } ), \ \forall A, A ^ { \prime } \in \Lambda.$$
Proof. We reprove this classical result [15, 17] for the reader's convenience. Let 1 denote the unit column vector, and choose ν and h so that
$$P _ { t } h = e ^ { - \alpha t } h, \ \nu P _ { t } = e ^ { - \alpha t } \nu, \ \nu \mathbf 1 = 1, \ \nu h = 1.$$
Let H denote the diagonal matrix corresponding to h . The h -transform of P t is
$$Q _ { t } = e ^ { \alpha t } H ^ { - 1 } P _ { t } H.$$
Since νHQ t = νH Q , t 1 = 1 and ( Q t ) t is a multiplicative semi-group, it defines a Markov process on Λ with invariant measure νH .
glyph[negationslash]
It follows from the α -positiveness of ( P t ) t that Q t → 0 , thus it is positive recurrent and hence Q t → 1 νH as t → ∞ . Therefore, e αt P t → hν as t → ∞ . Summing over the second coordinate we have e αt P t 1 → h. That is,
$$e ^ { \alpha t } P _ { t } ( A, A ^ { \prime } ) \to h ( A ) \nu ( A ^ { \prime } ) \text{ \ and \ } e ^ { \alpha t } P _ { t } ( A, \Lambda ) \to h ( A ).$$
Therefore we get
$$\frac { P _ { t } ( A, A ^ { \prime } ) } { P _ { t } ( A, \Lambda ) } \rightarrow \frac { h ( A ) \nu ( A ^ { \prime } ) } { h ( A ) } = \nu ( A ^ { \prime } ).$$
It remains to justify that summation over the second coordinate preserves the limit. Since e αt νP t = ν we get for every t glyph[greaterorequalslant] 0 and A ′ ∈ Λ
$$e ^ { \alpha t } P _ { t } ( A, A ^ { \prime } ) = \frac { e ^ { \alpha t } \nu ( A ) P _ { t } ( A, A ^ { \prime } ) } { \nu ( A ) } \leqslant \frac { \nu ( A ^ { \prime } ) } { \nu ( A ) },$$
which is summable over A ′ . The limit thus follows by dominated convergence.
## 3.2 Finite contact process modulo translations
For the contact process on Z d in arbitrary dimension d glyph[greaterorequalslant] 1 , the concept of 'seen from the edge' is generalized by considering the process modulo translations. We say that two configurations η and η ′ in the space { A ⊆ Z d : A is non-empty and finite } are equivalent if η = η ′ + y for some y ∈ Z d . Let Λ denote the quotient space resulting from this equivalence relation. We will denote by ζ the equivalence class of a configuration η , or indistinguishably any representant of such class when there is no confusion.
Since the evolution rules of the contact process are translation-invariant, the process ( ζ t ) t glyph[greaterorequalslant] 0 obtained by projecting ( η t ) t glyph[greaterorequalslant] 0 onto Λ ∪ {∅} is a homogeneous Markov process with values on Λ ∪{∅} . Moreover, the subset Λ is an irreducible class, and the absorbing state ∅ is almost-surely reached if λ < λ c . We call ( ζ t ) t glyph[greaterorequalslant] 0 the contact process modulo translations .
For d = 1 this is the same as taking Λ = { A ⊆ -N 0 : A is finite and 0 ∈ A } . Therefore, Proposition 1.1 is the specialization to d = 1 of the next result.
Proposition 3.2. Let ( ζ t ) t glyph[greaterorequalslant] 0 denote the contact process modulo translations on Z d with subcritical infection parameter λ . This process has a unique minimal quasi-stationary distribution ν . Moreover the Yaglom limit L ( ζ A t | τ A > t ) → ν as t → ∞ holds for any finite non-empty initial configuration A .
Let ( ξ n n ) denote an irreducible, aperiodic, discrete-time Markov chain on the statespace Λ ∪ {∅} , with transition matrix p ( · , · ) such that the absorbing time τ A = inf { n : ξ A n = ∅} is a.s. finite. As for continuous-time chains discussed above, there is R such that p n ( A,A ) = R -n + ( o n ) , and we say that p is R -positive if lim sup R p n n ( A,A ) > 0 . The proof of Proposition 3.2 will be based on the following criteria for R -positiveness.
Theorem 3.3 ([8, Theorem 1]) . Suppose that there exist a subset Λ ′ ⊆ Λ , a configuration A ′ ∈ Λ ′ , some ρ < R -1 , and positive constants M and ε such that
(H1) For all A ∈ Λ ′ and n glyph[greaterorequalslant] 0 , P ( τ A > n ; ξ A 1 , . . . , ξ A n / ∈ Λ ) ′ glyph[lessorequalslant] Mρ n ;
(H2) For all A ∈ Λ ′ and n glyph[greaterorequalslant] 0 , P ( τ A > n ) glyph[lessorequalslant] M P ( τ A ′ > n ) ;
(H3) For all A ∈ Λ ′ , P ( ξ A n = A ′ for some n glyph[lessorequalslant] M ) glyph[greaterorequalslant] ε .
Then p is R -positive and its left eigenvector ν is summable.
The following proposition provides a set of configurations Λ ′ and an appropriate time discretization that satisfy the above criteria. It is analogous to Theorem 2 in [8], but since the range of interaction of the contact process is infinite for any positive period of time, we cannot apply the latter directly. We give a simpler proof instead.
Proposition 3.4. For a subcritical contact process ( η t ) t glyph[greaterorequalslant] 0 on Z d , there exists a time ψ > 0 with the following property. For every ρ > 0 , one can find constants K and M such that
$$\mathbb { P } \left ( | \eta _ { \psi } ^ { A } | > K, \dots, | \eta _ { n \psi } ^ { A } | > K \right ) \leqslant M \rho ^ { n }$$
for all A ⊆ Z d with 1 glyph[lessorequalslant] | A | glyph[lessorequalslant] K and n glyph[greaterorequalslant] 1 .
Proof. A consequence of the exponential decay for the set of points infected from the origin is that ∫ ∞ 0 [ E | η 0 t | q ]d t < ∞ for every q > 0 [3, (1.13)-(1.14)]. Therefore we can choose a time ψ > 0 such that E | η 0 ψ | < 1 , and such that E | η 0 ψ | q < ∞ for every q .
We claim that | η A t | is stochastically bounded by the sum of | A | independent copies of | η 0 t | , which we denote by η ( ) i t , i = 1 , . . . , | A | . The proof of this fact is standard and will be omitted. By the Law of Large Numbers in L q ,
$$\frac { | \eta _ { \psi } ^ { A } | } { | A | } \ \stackrel { \text{st} } { \leqslant } \ \frac { 1 } { | A | } \sum _ { i = 1 } ^ { | A | } | \eta _ { \psi } ^ { ( i ) } | \stackrel { L ^ { q } } { \longrightarrow } \ \mathbb { E } | \eta _ { \psi } ^ { 0 } | \text{ as } | A | \rightarrow \infty.$$
Let ρ > 0 . Choosing q large so that ( E | η 0 ψ | ) q < ρ , there exist C and K such that
$$\mathbb { E } \frac { | \eta _ { \psi } ^ { A } | ^ { q } } { | A | ^ { q } } \leqslant C \text{ for all } A \in \Lambda \quad \text{and moreover} \quad \mathbb { E } \frac { | \eta _ { \psi } ^ { A } | ^ { q } } { | A | ^ { q } } \leqslant \rho \text{ whenever } | A | > K. \quad ( 3. 2 )$$
Writing ξ n for η nψ , for any A ∈ Λ with | A | glyph[lessorequalslant] K ,
$$1$$
$$\text{ Writing } \xi _ { n } \text{ for } \eta _ { n \psi }, \text{ for any } A \in \Lambda \text{ with } | A | \leq K, \\ \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{\, } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text` } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text
} \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text { } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text1 } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text. \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{} \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{
} \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{\{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \end{
\text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ }
\text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{\{} \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \end{ \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{\} \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text
} \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \end\text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{$$
We have used (3.2) n times here. Writing M = C ρ the result follows.
Finally we prove Proposition 3.2 using the previous results.
Proof of Proposition 3.2. Let ψ be given by Proposition 3.4. We now consider the discrete-time chain given by ξ n = ζ nψ , n = 0 1 2 , , , . . . , which has decay rate R = e αψ . Choose ρ < R -1 . By Proposition 3.4, there are K and M such that (3.1) holds, which implies (H1) with Λ = ′ { A ∈ Λ : | A | glyph[lessorequalslant] K } .
Take A ′ = 0 { } and observe that P ( τ A > n ) glyph[lessorequalslant] | A | · P ( τ 0 > n ) , which implies (H2). Finally, (H3) follows from
$$\mathbb { P } ( \zeta _ { \psi } ^ { A } = \{ 0 \} ) \geqslant e ^ { - \psi } e ^ { - 2 d \lambda | A | \psi } ( 1 - e ^ { - \psi } ) ^ { | A | - 1 } \geqslant \varepsilon, \text{ \ for every } A \in \Lambda ^ { \prime },$$
$$\text{where} \, \varepsilon = [ e ^ { \psi + 2 d \lambda \psi } ( 1 - e ^ { - \psi } ) ] ^ { K } > 0.$$
By Theorem 3.3, the matrix P ψ is R -positive with summable left-eigenvector ν . Therefore the semi-group ( P t ) t glyph[greaterorequalslant] 0 is α -positive with the same left-eigenvector. By Theorem 3.1 P t ( A,A ′ ) P t ( A, Λ) → ν A ( ′ ) , ∀ A,A ′ ∈ Λ , proving Proposition 3.2.
## Acknowledgments
We thank Santiago Saglietti for helpful suggestions.
## References
- [1] E. D. Andjel, Convergence in distribution for subcritical 2d oriented percolation seen from its rightmost point , Ann. Probab. 42 (2014), 1285-1296. MR-3189072
- [2] E. D. Andjel, R. B. Schinazi, and R. H. Schonmann, Edge processes of one-dimensional stochastic growth models , Ann. Inst. H. Poincaré Probab. Statist. 26 (1990), 489-506. MR1066090
- [3] C. Bezuidenhout and G. Grimmett, Exponential decay for subcritical contact and percolation processes , The Annals of Probability 19 (1991), no. 3, 984-1009. MR-1112404
- [4] P . Collet, S. Martínez, and J. S. Martín, Quasi-stationary distributions , Probability and its Applications (New York), Springer, Heidelberg, 2013, Markov chains, diffusions and dynamical systems. MR-2986807
- [5] J. T . Cox, R. Durrett, and R. Schinazi, The critical contact process seen from the right edge , Probab. Theory Related Fields 87 (1991), 325-332. MR-1084333
- [6] R. Durrett, Oriented percolation in two dimensions , Ann. Probab. 12 (1984), 999-1040. MR-0757768
- [7] F. Ezanno, Systèmes de particules en interaction et modèles de déposition aléatoire , Ph.D. thesis, Université d'Aix Marseille, 2012.
- [8] P . A. Ferrari, H. Kesten, and S. Martínez, R -positivity, quasi-stationary distributions and ratio limit theorems for a class of probabilistic automata , Ann. Appl. Probab. 6 (1996), 577-616. MR-1398060
- [9] A. Galves and E. Presutti, Edge fluctuations for the one-dimensional supercritical contact process , Ann. Probab. 15 (1987), 1131-1145. MR-0893919
- [10] J. F . C. Kingman, The exponential decay of Markov transition probabilities , Proc. London Math. Soc. (3) 13 (1963), 337-358. MR-0152014
- [11] T. Kuczek, The central limit theorem for the right edge of supercritical oriented percolation , Ann. Probab. 17 (1989), 1322-1332. MR-1048929
- [12] T. M. Liggett, Interacting particle systems , Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 276, Springer-Verlag, New York, 1985. MR-0776231
- [13] S. Méléard and D. Villemonais, Quasi-stationary distributions and population processes , Probab. Surv. 9 (2012), 340-410. MR-2994898
EJP 20 (2015), paper 32.
## Subcritical contact process seen from the edge
- [14] R. H. Schonmann, Absence of a stationary distribution for the edge process of subcritical oriented percolation in two dimensions , Ann. Probab. 15 (1987), 1146-1147. MR-0893920
- [15] E. Seneta and D. Vere-Jones, On quasi-stationary distributions in discrete-time Markov chains with a denumerable infinity of states , J. Appl. Probability 3 (1966), 403-434. MR-0207047
- [16] E. A. van Doorn and P . K. Pollett, Quasi-stationary distributions for discrete-state models , European J. Oper. Res. 230 (2013), 1-14. MR-3063313
- [17] D. Vere-Jones, Some limit theorems for evanescent processes , Austral. J. Statist. 11 (1969), 67-78. MR-0263165
|
10.1214/EJP.v20-3881
|
[
"Enrique Andjel",
"François Ezanno",
"Pablo Groisman",
"Leonardo T. Rolla"
] |
2014-08-04T20:09:02+00:00
|
2015-12-17T00:17:55+00:00
|
[
"math.PR"
] |
Subcritical Contact Process Seen from the Edge: Convergence to Quasi-Equilibrium
|
The subcritical contact process seen from the rightmost infected site has no
invariant measures. We prove that nevertheless it converges in distribution to
a quasi-stationary measure supported on finite configurations.
|
1408.0806v2
|
## EUROPEAN ORGANIZATION FOR NUCLEAR RESEARCH (CERN)
Search for pair production of third-generation scalar leptoquarks and top squarks in proton-proton collisions at √ s = 8 TeV
The CMS Collaboration ∗
## Abstract
A search for pair production of third-generation scalar leptoquarks and supersymmetric top quark partners, top squarks, in final states involving tau leptons and bottom quarks is presented. The search uses events from a data sample of proton-proton collisions corresponding to an integrated luminosity of 19.7 fb -1 , collected with the CMS detector at the LHC with √ s = 8 TeV. The number of observed events is found to be in agreement with the expected standard model background. Third-generation scalar leptoquarks with masses below 740 GeV are excluded at 95% confidence level, assuming a 100% branching fraction for the leptoquark decay to a tau lepton and a bottom quark. In addition, this mass limit applies directly to top squarks decaying via an R-parity violating coupling l ′ 333 . The search also considers a similar signature from top squarks undergoing a chargino-mediated decay involving the R-parity violating coupling l ′ 3 jk . Each top squark decays to a tau lepton, a bottom quark, and two light quarks. Top squarks in this model with masses below 580 GeV are excluded at 95% confidence level. The constraint on the leptoquark mass is the most stringent to date, and this is the first search for top squarks decaying via l ′ 3 jk .
Published in Physics Letters B as doi:10.1016/j.physletb.2014.10.063.
## 1 Introduction
Many extensions of the standard model (SM) predict new scalar or vector bosons, called leptoquarks (LQ), which carry non-zero lepton and baryon numbers, as well as color and fractional electric charge. Examples of such SM extensions include SU(5) grand unification [1], Pati-Salam SU(4) [2], composite models [3], superstrings [4], and technicolor models [5]. Leptoquarks decay into a quark and a lepton, with a model-dependent branching fraction for each possible decay. Experimental limits on flavor-changing neutral currents and other rare processes suggest that searches should focus on leptoquarks that couple to quarks and leptons within the same SM generation, for leptoquark masses accessible to current colliders [3, 6].
The dominant pair production mechanisms for leptoquarks at the CERN LHC would be gluongluon fusion and quark-antiquark annihilation via quantum chromodynamic (QCD) couplings. The cross sections for these processes depend only on the leptoquark mass for scalar leptoquarks. In this Letter, a search with the CMS detector for third-generation scalar leptoquarks, each decaying to a tau lepton and a bottom quark, is presented.
Similar signatures arising from supersymmetric models are also covered by this search. Supersymmetry (SUSY) [7, 8] is an attractive extension of the SM because it can resolve the hierarchy problem without unnatural fine-tuning, if the masses of the supersymmetric partner of the top quark (top squark) and the supersymmetric partners of the Higgs boson (higgsinos) are not too large [9, 10]. In many natural SUSY models the top squark and the higgsinos are substantially lighter than the other scalar SUSY particles. This light top squark scenario can be realized in both R-parity conserving (RPC) and R-parity violating (RPV) SUSY models, where R-parity is a new quantum number [11] that distinguishes SM and SUSY particles. In the context of an RPC decay of the top squark, the presence of an undetected particle (the lightest SUSY particle) is expected to generate a signature with large missing transverse momentum. If R-parity is violated, however, SUSY particles can decay into final states containing only SM particles. The RPV terms in the superpotential are:
$$W \ni \frac { 1 } { 2 } \lambda _ { i j k } L _ { i } L _ { j } E _ { k } ^ { c } + \lambda _ { i j k } ^ { \prime } L _ { i } Q _ { j } D _ { k } ^ { c } + \frac { 1 } { 2 } \lambda _ { i j k } ^ { \prime \prime } U _ { i } ^ { c } D _ { j } ^ { c } D _ { k } ^ { c } + \mu _ { i } L _ { i } H _ { u }$$
where W is the superpotential; L is the lepton doublet superfield; E is the lepton singlet superfield; Q is the quark doublet superfield; U and D are the quark singlet superfields; Hu is the Higgs doublet superfield that couples to up-type quarks; l l , ′ , and l ′′ are coupling constants; and , i j , and k are generation indices.
At the LHC, top squarks (t) would be directly pair-produced via strong interactions. ˜ In this search, two different decay channels of directly produced top squarks are considered. Both scenarios relate to simplified models in which all of the other SUSY particles have masses too large to participate in the interactions. In the first case we study the two-body lepton number violating decay t ˜ → t b [11] with a coupling constant l ′ 333 allowed by the trilinear RPV operators. The final-state signature and kinematic distributions of such a signal are identical to those from the pair production of third-generation scalar leptoquarks. When the masses of the supersymmetric partners of the gluon and quarks, excluding the top squark, are large, the top squark pair production cross section is the same as that of the third generation LQ. Thus, the results of the leptoquark search can be directly interpreted in the context of RPV top squarks.
In some natural SUSY models [12], if the higgsinos ( ˜ c 0 , ˜ c ± ) are lighter than the top squark, or if the RPV couplings that allow direct decays to SM particles are sufficiently small, the top squark decay may preferentially proceed via superpartners. In the second part of the search we focus on a scenario in which the dominant RPC decay of the top squark is t ˜ → ˜ c ± b. This
requires the mass splitting between the top squark and the chargino to be less than the mass of the top quark, so it is chosen to be 100 GeV. The chargino is assumed to be a pure higgsino and to be nearly degenerate in mass with the neutralino. We consider the case when ˜ c ± → ˜ nt ± → qq t ± . The decay of the sneutrino occurs according to an RPV operator with a coupling constant l ′ 3 jk , where the cases j , k = 1, 2 are considered. Such signal models can only be probed by searches that do not require large missing transverse momentum, as the other decay of the chargino, ˜ c ± → nt ˜ , does not contribute to scenarios involving the l ′ 3 jk coupling because of chiral suppression. From such a signal process, we expect events with two tau leptons, two jets originating from hadronization of the bottom quarks, and at least four additional jets.
In this Letter, the search for scalar leptoquarks and top squarks decaying through the coupling l ′ 333 is referred to as the leptoquark search. The search for the chargino-mediated decay of top squarks involving the l ′ 3 jk coupling is referred to as the top squark search. The data sample used in this search has been recorded with the CMS detector in proton-proton collisions at a center-of-mass energy of 8 TeV and corresponds to an integrated luminosity of 19.7 fb -1 . One of the tau leptons in the final state is required to decay leptonically: t → glyph[lscript] ¯ n nt glyph[lscript] , where glyph[lscript] can be either an electron or a muon, denoted as a light lepton. The other tau lepton is required to decay to hadrons ( t h): t → hadrons + nt . These decays result in two possible final states labeled below as e t h and mt h, or collectively glyph[lscript] t h when the lepton flavor is unimportant. The leptoquark search is performed in a mass range from 200 to 1000 GeV using a sample of events containing one light lepton, a hadronically decaying tau lepton, and at least two jets, with at least one of the jets identified as originating from bottom quark hadronization (b-tagged). The top squark search is performed in a mass range from 200 to 800 GeV using a sample of events containing one light lepton, a hadronically decaying tau lepton, and at least five jets, with at least one of the jets b-tagged.
No evidence for third-generation leptoquarks or top squarks decaying to tau leptons and bottom quarks has been found in previous searches [13, 14]. The most stringent lower limit to date on the mass of a scalar third generation leptoquark decaying to a tau lepton and a bottom quark with a 100% branching fraction is about 530 GeV from both the CMS and ATLAS experiments. This Letter also presents the first search for the chargino-mediated decay of the top squark through the RPV coupling l ′ 3 jk .
## 2 The CMS detector
The central feature of the CMS apparatus is a superconducting solenoid, of 6 m internal diameter, providing a field of 3.8 T. Within the field volume are several subdetectors. A silicon pixel and strip tracker allows the reconstruction of the trajectories of charged particles within the pseudorapidity range | h | < 2.5. The calorimetry system consists of a lead tungstate crystal electromagnetic calorimeter (ECAL) and a brass and scintillator hadron calorimeter, and measures particle energy depositions for | h | < 3. The CMS detector also has extensive forward calorimetry (2.8 < | h | < 5.2). Muons are measured in gas-ionization detectors embedded in the steel flux-return yoke of the magnet. Collision events are selected using a two-tiered trigger system [15]. A more detailed description of the CMS detector, together with a definition of the coordinate system used and the relevant kinematic variables, can be found in Ref. [16].
## 3 Object and event selection
Candidate LQ or top squark events were collected using a set of triggers requiring the presence of either an electron or a muon with transverse momentum ( p T) above a threshold of 27 or 24 GeV, respectively.
Both electrons and muons are required to be reconstructed within the range | h | < 2.1 and to have p T > 30 GeV. Electrons, reconstructed using information from the ECAL and the tracker, are required to have an electromagnetic shower shape consistent with that of an electron, and an energy deposition in ECAL that is compatible with the track reconstructed in the tracker. Muons are required to be reconstructed by both the tracker and the muon spectrometer. A particle-flow (PF) technique [17-19] is used for the reconstruction of hadronically decaying tau lepton candidates. In the PF approach, information from all subdetectors is combined to reconstruct and identify all final-state particles produced in the collision. The particles are classified as either charged hadrons, neutral hadrons, electrons, muons, or photons. These particles are used with the 'hadron plus strips' algorithm [20] to identify t h objects. Hadronically decaying tau leptons with one or three charged pions and up to two neutral pions are reconstructed. The reconstructed t h is required to have visible p T > 50 GeV and | h | < 2.3. Electrons, muons, and tau leptons are required to be isolated from other reconstructed particles. The identified electron (muon) and t h are required to originate from the same vertex and be separated by D R = √ ( D h ) 2 +( D f ) 2 > 0.5. The light lepton and the t h are also required to have opposite electric charge. Events are vetoed if another light lepton is found, passing the kinematic, identification, and isolation criteria described above, that has an opposite electric charge from the selected light lepton.
Jets are reconstructed using the antik T algorithm [21, 22] with a size parameter 0.5 using particle candidates reconstructed with the PF technique. Jet energies are corrected by subtracting the average contribution from particles coming from other proton-proton collisions in the same beam crossing (pileup) and by applying a jet energy calibration, determined empirically [23]. Jets are required to be within | h | < 2.4, have p T > 30 GeV, and be separated from both the light lepton and the t h by D R > 0.5. The minimum jet p T requirement eliminates most jets from pileup interactions. Jets are b-tagged using the combined secondary vertex algorithm with the loose operating point [24]. In the leptoquark search, the b-tagged jet with the highest p T is selected, and then the remaining jet with the highest p T is selected whether or not it is b-tagged. In the top squark search, the b-tagged jet with the highest p T is selected, and then the remaining four jets with the highest p T are selected whether or not they are b-tagged.
To discriminate between signal and background in the leptoquark search, the mass of the t h and a jet, denoted M ( t h, jet ) , is required to be greater than 250 GeV. There are two possible pairings of the t h with the two required jets. The pairing is chosen to minimize the difference between the mass of the t h and one jet and the mass of the light lepton and another jet. According to a simulation, the correct pairing is selected in approximately 70% of events.
The S T distribution after the final selection is used to extract the limits on both the leptoquark and top squark signal scenarios, where S T is defined as the scalar sum of the p T of the light lepton, the t h, and the two jets (five jets) for the leptoquark search (top squark search).
## 4 Background and signal models
Several SM processes can mimic the final-state signatures expected from leptoquark or top squark pair production and decay. For this analysis, the backgrounds are divided into three
groups, which are denoted as tt irreducible, major reducible, and other. The tt irreducible background comes from the pair production of top quarks (tt) when both the light lepton and t h are genuine, each produced from the decay of a W boson. In this case, the light lepton can originate either directly from the W boson decay or from a decay chain W → tnt → glyph[lscript] n ntnt glyph[lscript] . The major reducible background consists of events in which a quark or gluon jet is misidentified as a t h. The processes contributing to the major reducible background are associated production of a W or Z boson with jets, and tt. Additionally, a small contribution from the QCD multijet process is included, in which both the light lepton and the t h are misidentified jets. The third group, other backgrounds, consists of processes that make small contributions and may contain either genuine or misidentified tau leptons. This includes the diboson and single-top-quark processes, the tt and Z+jets processes when a light lepton is misidentified as a t h, and the Z+jets process when the Z boson decays to a pair of tau leptons. The other backgrounds are estimated from the simulation described below, while the tt irreducible and major reducible backgrounds are estimated using observed data. The major reducible and other backgrounds include events with both genuine and misidentified light leptons.
The PYTHIA v6.4.24 generator [25] is used to model the signal and diboson processes. The leptoquark signal samples are generated with masses ranging from 200 to 1000 GeV, and the top squark signal samples are generated with masses ranging from 200 to 800 GeV and the sneutrino mass set to 2000 GeV. The MADGRAPH v5.1.3.30 generator [26] is used to model the tt, W+jets, and Z+jets processes. This generation includes contributions from heavy-flavor and extra jets. The single top-quark production is modeled with the POWHEG 1.0 r138 [27-29] generator. Both the MADGRAPH and POWHEG generators are interfaced with PYTHIA for hadronization and showering. The TAUOLA program [30] is used for tau lepton decays in the leptoquark, tt, W+jets, Z+jets, diboson, and single top-quark samples. Each sample is passed through a full simulation of the CMS detector based on GEANT4 [31] and the complete set of reconstruction algorithms is used to analyze collision data. Cross sections for the leptoquark signal and diboson processes are calculated to next-to-leading order (NLO) [32, 33]. The cross sections for the top squark signal are calculated at NLO in the strong coupling constant, including the resummation of soft gluon emission at next-to-leading-logarithmic accuracy (NLO+NLL) [3438]. The next-to-next-to-leading-order or approximate next-to-next-to-leading-order [39, 40] cross sections are used for the rest of the background processes.
The efficiencies of the trigger and final selection criteria for signal processes are estimated from the simulation. The efficiencies for light leptons and b jets are calculated from data and used where necessary to correct the event selection efficiency estimations from the simulation. No correction is required for hadronically decaying tau leptons.
The tt irreducible background is estimated from an e m sample that is 87% pure in tt events according to the simulation. The contributions from other processes are simulated and subtracted from the observed data. This sample comprises events with one electron and one muon that satisfy the remaining final selection criteria, except that a t h is not required. The potential signal contamination of this sample has been found to be negligible for any signal mass hypothesis. The final yield of the e m sample is scaled by the relative difference in the selection efficiencies between the glyph[lscript] t h and e m samples. The selection efficiencies are measured in the simulation and are corrected to match those from collision data. The estimation of the final yield based on the observed data agrees with both the direct prediction from the simulation and the yield obtained after applying the same method to the Monte Carlo (MC) samples. The S T distribution for the tt irreducible background is obtained from a simulated tt sample that consists exclusively of fully leptonic decays of top quarks.
The major reducible background from tt, W+jets, and Z+jets events in which a jet is misidentified as a hadronically decaying tau lepton is estimated from observed data. The probability of misidentification is measured using events recorded with a Z boson produced in association with jets and decaying to a pair of muons (Z → mm ). The invariant mass of the muon pair is required to be greater than 50 GeV and events are required to contain at least one jet that is incorrectly identified as a t h and may or may not pass the isolation requirement. The misidentification probability f ( p T ( t )) is calculated as the fraction of these t h candidates that pass the isolation requirement and depends on the p T of the candidates. The background yield is estimated from a sample of events satisfying the final selection criteria, except that all t h candidates in the events must fail the isolation requirement. Equation (2) relates the yield of these 'antiisolated' events to the yield of events passing the final selection, using the misidentification probability:
$$N _ { \text{misID} \, \tau } = \sum _ { \text{events} } ^ { ( \text{anti-iso} ) } \frac { 1 - \Pi _ { \tau } \left [ 1 - f ( p _ { T } ( \tau ) ) \right ] } { \Pi _ { \tau } \left [ 1 - f ( p _ { T } ( \tau ) ) \right ] }.$$
The estimation of the final yield based on the observed data agrees with both the direct prediction from the simulation and the estimation performed using the same approach on simulated samples. The S T distribution for the major reducible background is obtained using simulated samples for the W+jets and Z+jets processes and the tt process with exclusively semi-leptonic decays.
The QCD multijet process contributes only in the e t h channel in the leptoquark search and corresponds to 16% of the reducible background. The contribution from multijet events is estimated from a sample of observed events satisfying the final selection criteria for the e t h channel except that the electron and t h must have the same electric charge. The QCD component is included in the distribution of the rest of the major reducible background, described above.
## 5 Systematic uncertainties
There are a number of systematic uncertainties associated with both the background estimation and the signal efficiency. The uncertainty in the total integrated luminosity is 2.6% [41]. The uncertainty in the trigger and lepton efficiencies is 2%, while the uncertainty assigned to the t h identification efficiency is 6%. The uncertainties in the b-tagging efficiency and mistagging probability depend on the h and p T of the jet and are on average 4% and 10%, respectively [42].
Systematic uncertainties, totaling 19-22% depending on the channel and the search, are assigned to the normalization of the tt irreducible background based on statistical uncertainty in the control samples and the propagation of the uncertainties in the acceptances, efficiencies, and subtraction of the contributions from other processes in the e m sample. Systematic uncertainties in the major reducible background are driven by statistical uncertainty in the measured misidentification probability and variation in the misidentification probability based on the event topology. These uncertainties amount to 16-24%, depending on the channel and the search.
Because of the limited number of events in the simulation, uncertainties in the small backgrounds range between 20-50%. Uncertainty due to the effect of pileup modeling in the MC is estimated to be 3%. A 4% uncertainty, due to modeling of initial- and final-state radiation in the simulation, is assigned to the signal acceptance. The uncertainty in the initial- and final-state radiation was found to have a negligible effect on the simulated backgrounds. A 7-32% uncertainty from knowledge of parton distribution functions and a 14-80% uncertainty from QCD renormalization and factorization scales are assigned to the theoretical signal cross-section.
Finally, jet energy scale uncertainties (2-4% depending on h and p T) and energy resolution uncertainties (5-10% depending on h ), as well as energy scale (3%) and resolution (10%) uncertainties for t h, affect both the S T distributions and the expected yields from the signal and background processes.
## 6 Results
The numbers of observed events and expected signal and background events after the final selection for the leptoquark and top squark searches are listed in Tables 1 and 2, respectively, and the selection efficiencies for the two signals are listed in Tables 3 and 4. The S T distributions of the selected events from the observed data and from the background predictions, combining e t h and mt h channels, are shown in Fig. 1 for the leptoquark search and Fig. 2 for the top squark search. The distribution from the 500 GeV (300 GeV) signal hypothesis is added to the background in Fig. 1 (Fig. 2) to illustrate how a hypothetical signal would appear above the background prediction. The data agree well with the SM background prediction.
Anupper bound at 95% confidence level (CL) is set on s B 2 , where s is the cross section for pair production of third-generation LQs (top squarks) and B is the branching fraction for the LQ decay to a tau lepton and a bottom quark (the top squark decay to a ˜ c ± and a bottom quark, with a subsequent decay of the chargino via ˜ c ± → ˜ nt ± → qq t ± ). The symbol M LQ is used for the leptoquark mass and the symbol M ˜ t is used for the top squark mass. The modifiedfrequentist construction CLs [43-45] is used for the limit calculation. A maximum likelihood fit is performed to the S T spectrum simultaneously for the e t h and mt h channels, taking into account correlations between the systematic uncertainties. Expected and observed upper limits on s B 2 as a function of the signal mass are shown in Fig. 3 for the leptoquark search and Fig. 4 for the top squark search.
Weextend the current limits and exclude scalar leptoquarks and top squarks decaying through the coupling l ′ 333 with masses below 740 GeV, in agreement with a limit at 750 GeV, expected in the absence of a signal. We exclude top squarks undergoing a chargino-mediated decay involving the coupling l ′ 3 jk with masses in the range 200-580 GeV, in agreement with the expected exclusion limit in the range 200-590 GeV. These upper limits assume B = 100%. Similar results are obtained when calculating upper bounds using a Bayesian method with a uniform positive prior for the cross section.
The upper bounds for the leptoquark search as a function of the leptoquark branching fraction and mass are shown in Fig. 5. Small B values are not constrained by this search. Results from the CMS experiment on a search for top squarks decaying to a top quark and a neutralino [46] are used to further constrain B . If the neutralino is massless, the final state kinematic distributions for such a signal are the same as those for the pair production of leptoquarks decaying to a tau neutrino and a top quark. Limits can therefore be placed on this signal, which must have a branching fraction of 1 -B if the leptoquark only decays to third-generation fermions. This reinterpretation is included in Fig. 5. The unexcluded region at M LQ = 200-230 GeV corresponds to a portion of phase space where it is topologically very difficult to distinguish between the top squark signal and the tt process, owing to small missing transverse momentum. A top squark excess in this region would imply an excess in the measured tt cross section of ∼ 10%.
Table 1: The estimated backgrounds, observed event yields, and expected number of signal events for the leptoquark search. For the simulation-based entries, the statistical and systematic uncertainties are shown separately, in that order.
| | e t h | mt h |
|----------------------------------------------|--------------------|--------------------|
| tt irreducible | 105.6 ± 18.1 | 66.7 ± 12.6 |
| Major reducible | 147.8 ± 33.0 | 117.3 ± 18.9 |
| Z( glyph[lscript] glyph[lscript] / tt )+jets | 21.4 ± 7.4 ± 4.9 | 7.5 ± 4.6 ± 0.2 |
| Single t | 16.0 ± 2.8 ± 4.4 | 17.3 ± 2.8 ± 4.7 |
| VV | 4.1 ± 0.6 ± 1.3 | 2.6 ± 0.5 ± 0.8 |
| Total exp. bkg. | 294.9 ± 7.9 ± 39.1 | 211.4 ± 5.4 ± 23.4 |
| Observed | 289 | 216 |
| M LQ = 500 GeV | 57.7 ± 1.4 ± 5.9 | 51.6 ± 1.3 ± 5.3 |
| M LQ = 600 GeV | 20.1 ± 0.5 ± 1.9 | 17.7 ± 0.4 ± 1.6 |
| M LQ = 700 GeV | 7.1 ± 0.2 ± 6.3 | 6.2 ± 0.1 ± 5.5 |
| M LQ = 800 GeV | 2.7 ± 0.1 ± 0.2 | 2.3 ± 0.1 ± 0.2 |
Table 2: The estimated backgrounds, observed event yields, and expected number of signal events for the top squark search. For the simulation-based entries, the statistical and systematic uncertainties are shown separately, in that order.
| | e t h | mt h |
|----------------------------------------------|--------------------|--------------------|
| tt irreducible | 88.3 ± 13.7 | 55.0 ± 9.5 |
| Major reducible | 65.7 ± 16.4 | 59.8 ± 13.8 |
| Z( glyph[lscript] glyph[lscript] / tt )+jets | 4.9 ± 2.5 ± 1.1 | 11.6 ± 5.5 ± 2.7 |
| Single t | 3.9 ± 1.5 ± 1.1 | 3.5 ± 1.3 ± 0.9 |
| VV | 0.6 ± 0.2 ± 0.2 | 0.4 ± 0.2 ± 0.1 |
| Total exp. bkg. | 163.4 ± 2.9 ± 21.5 | 130.3 ± 5.6 ± 17.1 |
| Observed | 156 | 123 |
| M t = 300 GeV | 94.3 ± 8.5 ± 13.2 | 82.8 ± 8.0 ± 11.7 |
| ˜ M t = 400 GeV | 43.9 ± 2.6 ± 4.3 | 38.3 ± 2.3 ± 3.8 |
| ˜ M t = 500 GeV | 19.4 ± 0.8 ± 1.8 | 15.4 ± 0.7 ± 1.5 |
| ˜ M t = 600 GeV | 6.9 ± 0.9 ± 0.7 | 5.7 ± 0.3 ± 0.5 |
˜
Table 3: Selection efficiencies in % for the signal in the leptoquark search, estimated from the simulation.
| M LQ (GeV) | e t h | mt h |
|--------------|---------|--------|
| 200 | 0.1 | 0.1 |
| 250 | 0.3 | 0.2 |
| 300 | 1 | 0.8 |
| 350 | 1.9 | 1.5 |
| 400 | 2.4 | 2.3 |
| 450 | 3 | 2.9 |
| 500 | 3.6 | 3.2 |
| 550 | 4 | 3.3 |
| 600 | 4.4 | 3.8 |
| 650 | 4.5 | 4 |
| 700 | 4.7 | 4.1 |
| 750 | 4.9 | 4.2 |
| 800 | 5.1 | 4.3 |
| 850 | 5.4 | 4.4 |
| 900 | 5.1 | 4.4 |
| 950 | 5.4 | 4.3 |
| 1000 | 5.5 | 4.4 |
Table 4: Selection efficiencies in % for the signal in the top squark search, estimated from the simulation.
| M ˜ t (GeV) | e t h | mt h |
|---------------|---------|--------|
| 200 | 0.02 | 0.02 |
| 300 | 0.3 | 0.2 |
| 400 | 0.7 | 0.6 |
| 500 | 1.2 | 1 |
| 600 | 1.5 | 1.2 |
| 700 | 1.8 | 1.4 |
| 800 | 1.8 | 1.3 |
| 900 | 1.5 | 1.1 |
T
Figure 1: The final S T distribution for the leptoquark search with the e t h and mt h channels combined. A signal sample for leptoquarks with the mass of 500 GeV is added on top of the background prediction. The last bin contains the overflow events. The horizontal bar on each observed data point indicates the width of the bin in S T.
Figure 2: The final S T distribution for the top squark search with the e t h and mt h channels combined. A signal sample for top squarks with the mass of 300 GeV is added on top of the background prediction. The last bin contains the overflow events. The horizontal bar on each observed data point indicates the width of the bin in S T.
Figure 3: The expected and observed combined upper limits on the third-generation LQ pair production cross section s times the square of the branching fraction, B 2 , at 95% CL, as a function of the LQ mass. These limits also apply to top squarks decaying directly via the coupling l ′ 333 . The green (darker) and yellow (lighter) uncertainty bands represent 68% and 95% CL intervals on the expected limit. The dark blue curve and the hatched light blue band represent the theoretical LQ pair production cross section, assuming B = 100%, and the uncertainties due to the choice of PDF and renormalization/factorization scales.
t
Figure 4: The expected and observed combined upper limits on the top squark pair production cross section s times the square of the branching fraction, B 2 , at 95% CL, as a function of the top squark mass. These limits apply to top squarks with a chargino-mediated decay through the coupling l ′ 3 kj . The green (darker) and yellow (lighter) uncertainty bands represent 68% and 95% CL intervals on the expected limit. The dark blue curve and the hatched light blue band represent the theoretical top squark pair production cross section, assuming B = 100%, and the uncertainties due to the choice of PDF and renormalization/factorization scales.
Figure 5: The expected (dashed black) and observed (green solid) 95% CL upper limits on the branching fraction for the leptoquark decay to a tau lepton and a bottom quark, as a function of the leptoquark mass. A search for top squark pair production [46] has the same kinematic signature as the leptoquark decay to a tau neutrino and a top quark. This search is reinterpreted to provide the expected (blue hatched) and observed (blue open) 95% CL upper limits for low values of B , assuming the leptoquark only decays to third-generation fermions.
## 7 Summary
A search for pair production of third-generation scalar leptoquarks and top squarks has been presented. The search for leptoquarks and top squarks decaying through the R-parity violating coupling l ′ 333 is performed in final states that include an electron or a muon, a hadronically decaying tau lepton, and at least two jets, at least one of which is b-tagged. The search for top squarks undergoing a chargino-mediated decay involving the R-parity violating coupling l ′ 3 jk is performed in events containing an electron or a muon, a hadronically decaying tau lepton, and at least five jets, at least one of which is b-tagged. No excesses above the standard model background prediction are observed in the S T distributions. Assuming a 100% branching fraction for the decay to a tau lepton and a bottom quark, scalar leptoquarks and top squarks decaying through l ′ 333 with masses below 740 GeV are excluded at 95% confidence level. Top squarks decaying through l ′ 3 jk with masses below 580 GeV are excluded at 95% confidence level, assuming a 100% branching fraction for the decay to a tau lepton, a bottom quark, and two light quarks. The constraint on the third-generation leptoquark mass is the most stringent to date, and this is the first search for top squarks decaying through l ′ 3 jk .
## Acknowledgements
We congratulate our colleagues in the CERN accelerator departments for the excellent performance of the LHC and thank the technical and administrative staffs at CERN and at other CMS institutes for their contributions to the success of the CMS effort. In addition, we gratefully acknowledge the computing centres and personnel of the Worldwide LHC Computing Grid for delivering so effectively the computing infrastructure essential to our analyses. Finally, we acknowledge the enduring support for the construction and operation of the LHC and the CMS detector provided by the following funding agencies: BMWFW and FWF (Austria); FNRS and FWO (Belgium); CNPq, CAPES, FAPERJ, and FAPESP (Brazil); MES (Bulgaria); CERN; CAS, MoST, and NSFC (China); COLCIENCIAS (Colombia); MSES and CSF (Croatia); RPF (Cyprus); MoER, ERC IUT and ERDF (Estonia); Academy of Finland, MEC, and HIP (Finland); CEA and CNRS/IN2P3 (France); BMBF, DFG, and HGF (Germany); GSRT (Greece); OTKA and NIH (Hungary); DAE and DST (India); IPM (Iran); SFI (Ireland); INFN (Italy); NRF and WCU (Republic of Korea); LAS (Lithuania); MOE and UM (Malaysia); CINVESTAV, CONACYT, SEP, and UASLP-FAI (Mexico); MBIE (New Zealand); PAEC (Pakistan); MSHE and NSC (Poland); FCT (Portugal); JINR (Dubna); MON, RosAtom, RAS and RFBR (Russia); MESTD (Serbia); SEIDI and CPAN (Spain); Swiss Funding Agencies (Switzerland); MST (Taipei); ThEPCenter, IPST, STAR and NSTDA (Thailand); TUBITAK and TAEK (Turkey); NASU and SFFR (Ukraine); STFC (United Kingdom); DOE and NSF (USA).
Individuals have received support from the Marie-Curie programme and the European Research Council and EPLANET (European Union); the Leventis Foundation; the A. P. Sloan Foundation; the Alexander von Humboldt Foundation; the Belgian Federal Science Policy Office; the Fonds pour la Formation ` a la Recherche dans l'Industrie et dans l'Agriculture (FRIABelgium); the Agentschap voor Innovatie door Wetenschap en Technologie (IWT-Belgium); the Ministry of Education, Youth and Sports (MEYS) of the Czech Republic; the Council of Science and Industrial Research, India; the HOMING PLUS programme of Foundation for Polish Science, cofinanced from European Union, Regional Development Fund; the Compagnia di San Paolo (Torino); the Consorzio per la Fisica (Trieste); MIUR project 20108T4XTM (Italy); the Thalis and Aristeia programmes cofinanced by EU-ESF and the Greek NSRF; and the National Priorities Research Program by Qatar National Research Fund.
## References
- [1] H. Georgi and S. L. Glashow, 'Unity of All Elementary-Particle Forces', Phys. Rev. Lett. 32 (1974) 438, doi:10.1103/PhysRevLett.32.438 .
- [2] J. C. Pati and A. Salam, 'Lepton number as the fourth 'color'', Phys. Rev. D 10 (1974) 275, doi:10.1103/PhysRevD.10.275 .
- [3] B. Gripaios, 'Composite leptoquarks at the LHC', JHEP 02 (2010) 045, doi:10.1007/JHEP02(2010)045 , arXiv:0910.1789 .
- [4] J. L. Hewett and T. G. Rizzo, 'Low-energy phenomenology of superstring inspired E6 models', Phys. Rept. 183 (1989) 193, doi:10.1016/0370-1573(89)90071-9 .
- [5] E. Eichten and K. Lane, 'Dynamical breaking of weak interaction symmetries', Phys. Lett. B 90 (1980) 125, doi:10.1016/0370-2693(80)90065-9 .
- [6] O. Shanker, ' p glyph[lscript] 2, K glyph[lscript] 3, and K 0 -K 0 constraints on leptoquarks and supersymmetric particles', Nucl. Phys. B 204 (1982) 375, doi:10.1016/0550-3213(82)90196-1 .
- [7] H. P. Nilles, 'Supersymmetry, supergravity and particle physics', Phys. Rept. 110 (1984) 1, doi:10.1016/0370-1573(84)90008-5 .
- [8] H. E. Haber and G. L. Kane, 'The search for supersymmetry: Probing physics beyond the standard model', Phys. Rept. 117 (1985) 75, doi:10.1016/0370-1573(85)90051-1 .
- [9] M. Papucci, J. T. Ruderman, and A. Weiler, 'Natural SUSY endures', JHEP 09 (2012) 035, doi:10.1007/JHEP09(2012)035 , arXiv:1110.6926 .
- [10] R. Kitano and Y. Nomura, 'Supersymmetry, naturalness, and signatures at the CERN LHC', Phys. Rev. D 73 (2006) 095004, doi:10.1103/PhysRevD.73.095004 , arXiv:hep-ph/0602096 .
- [11] R. Barbier et al., ' R -Parity-violating supersymmetry', Phys. Rept. 420 (2005) 1, doi:10.1016/j.physrep.2005.08.006 , arXiv:hep-ph/0406039 .
- [12] J. Evans and Y. Kats, 'LHC searches examined via the RPV MSSM', in 2013 Eur. Phys. Soc. Conf. on High Energy Physics . SISSA, 2013. arXiv:1311.0890 .
- [13] CMS Collaboration, 'Search for Pair Production of Third-Generation Leptoquarks and Top Squarks in pp Collisions at √ s = 7 TeV', Phys. Rev. Lett. 110 (2013) 081801, doi:10.1103/PhysRevLett.110.081801 , arXiv:1210.5629 .
- [14] ATLAS Collaboration, 'Search for third generation scalar leptoquarks in pp collisions at √ s = 7 TeV with the ATLAS detector', JHEP 06 (2013) 033, doi:10.1007/JHEP06(2013)033 , arXiv:1303.0526 .
- [15] CMS Collaboration, 'The CMS high level trigger', Eur. Phys. J. C 46 (2006) 605, doi:10.1140/epjc/s2006-02495-8 , arXiv:hep-ex/0512077 .
- [16] CMS Collaboration, 'The CMS experiment at the CERN LHC', JINST 3 (2008) S08004, doi:10.1088/1748-0221/3/08/S08004 .
- [17] CMS Collaboration, 'Particle-Flow Event Reconstruction in CMS and Performance for Jets, Taus, and MET', CMS Physics Analysis Summary CMS-PAS-PFT-09-001, CERN, 2009.
| [18] | CMS Collaboration, 'Commissioning of the Particle-Flow Reconstruction in Minimum-Bias and Jet Events from pp Collisions at 7 TeV', CMS Physics Analysis Summary CMS-PAS-PFT-10-002, CERN, 2010. |
|--------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [19] | CMS Collaboration, 'Commissioning of the particle-flow event reconstruction with leptons from J/ Y and Wdecays at 7 TeV', CMS Physics Analysis Summary CMS-PAS-PFT-10-003, CERN, 2010. |
| [20] | CMS Collaboration, 'Performance of t -lepton reconstruction and identification in CMS', JINST 7 (2012) 01001, doi:10.1088/1748-0221/7/01/P01001 , arXiv:1109.6034 . |
| [21] | G. P. Salam, 'Towards jetography', Eur. Phys. J. C 67 (2010) 637, doi:10.1140/epjc/s10052-010-1314-6 , arXiv:0906.1833 . |
| [22] | M. Cacciari, G. P. Salam, and G. Soyez, 'The anti- k t jet clustering algorithm', JHEP 04 (2008) 063, doi:10.1088/1126-6708/2008/04/063 , arXiv:0802.1189 . |
| [23] | CMS Collaboration, 'Determination of jet energy calibration and transverse momentum resolution in CMS', JINST 6 (2011) 11002, doi:10.1088/1748-0221/6/11/P11002 , arXiv:1107.4277 . |
| [24] | CMS Collaboration, 'Identification of b-quark jets with the CMS experiment', JINST 8 (2013) P04013, doi:10.1088/1748-0221/8/04/P04013 , arXiv:1211.4462 . |
| [25] | T. Sj¨strand, o S. Mrenna, and P. Skands, 'PYTHIA 6.4 physics and manual', JHEP 05 (2006) 026, doi:10.1088/1126-6708/2006/05/026 , arXiv:hep-ph/0603175 . |
| [26] | J. Alwall et al., 'The automated computation of tree-level and next-to-leading order differential cross sections, and their matching to parton shower simulations', (2014). arXiv:1405.0301 . |
| [27] | S. Frixione, P. Nason, and C. Oleari, 'Matching NLO QCD computations with parton shower simulations: the POWHEGmethod', JHEP 11 (2007) 070, doi:10.1088/1126-6708/2007/11/070 , arXiv:0709.2092 . |
| [28] | S. Alioli, P. Nason, C. Oleari, and E. Re, 'NLO single-top production matched with shower in POWHEG: s - and t -channel contributions', JHEP 09 (2009) 111, doi:10.1088/1126-6708/2009/09/111 , arXiv:0907.4076 . [Erratum: doi:10.1007/JHEP02(2010)011 ]. |
| [29] | E. Re, 'Single-top Wt -channel production matched with parton showers using the POWHEGmethod', Eur. Phys. J. C 71 (2011) 1547, doi:10.1140/epjc/s10052-011-1547-z , arXiv:1009.2450 . |
| [30] | Z. Was, ¸ 'TAUOLA the library for t lepton decay, and KKMC/KORALB/KORALZ...status report', Nucl. Phys. B, Proc. Suppl. 98 (2001) 96, doi:10.1016/S0920-5632(01)01200-2 . |
| [31] | S. Agostinelli et al., 'Geant4-a simulation toolkit', Nucl. Instrum. Meth. A 506 (2003) 250, doi:10.1016/S0168-9002(03)01368-8 . |
| [32] | M. Kr¨mer, a T. Plehn, M. Spira, and P. M. Zerwas, 'Pair production of scalar leptoquarks at the CERN LHC', Phys. Rev. D 71 (2005) 057503, doi:10.1103/PhysRevD.71.057503 , arXiv:hep-ph/0411038 . |
| [33] | J. M. Campbell, R. K. Ellis, and C. Williams, 'Vector boson pair production at the LHC', JHEP 07 (2011) 018, doi:10.1007/JHEP07(2011)018 , arXiv:1105.0020 . |
|--------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [34] | W. Beenakker, R. H¨pker, o M. Spira, and P. M. Zerwas, 'Squark and gluino production at hadron colliders', Nucl. Phys. B 492 (1997) 51, doi:10.1016/S0550-3213(97)80027-2 , arXiv:hep-ph/9610490 . |
| [35] | A. Kulesza and L. Motyka, 'Threshold resummation for squark-antisquark and gluino-pair production at the LHC', Phys. Rev. Lett. 102 (2009) 111802, doi:10.1103/PhysRevLett.102.111802 , arXiv:0807.2405 . |
| [36] | A. Kulesza and L. Motyka, 'Soft gluon resummation for the production of gluino-gluino and squark-antisquark pairs at the LHC', Phys. Rev. D 80 (2009) 095004, doi:10.1103/PhysRevD.80.095004 , arXiv:0905.4749 . |
| [37] | W. Beenakker et al., 'Soft-gluon resummation for squark and gluino hadroproduction', JHEP 12 (2009) 041, doi:10.1088/1126-6708/2009/12/041 , arXiv:0909.4418 . |
| [38] | W. Beenakker et al., 'Squark and gluino hadroproduction', Int. J. Mod. Phys. A 26 (2011) 2637, doi:10.1142/S0217751X11053560 , arXiv:1105.1110 . |
| [39] | K. Melnikov and F. Petriello, 'Electroweak gauge boson production at hadron colliders through O ( a 2 s ) ', Phys. Rev. D 74 (2006) 114017, doi:10.1103/PhysRevD.74.114017 arXiv:hep-ph/0609070 . |
| [40] | N. Kidonakis, 'Differential and total cross sections for top pair and single top production', in Proc. of XX Int. Workshop on Deep-Inelastic Scattering and Related Subjects , p. 831. 2012. arXiv:1205.3453 . doi:10.3204/DESY-PROC-2012-02/251 . |
| [41] | CMS Collaboration, 'CMS Luminosity Based on Pixel Cluster Counting - Summer 2013 Update', CMS Physics Analysis Summary CMS-PAS-LUM-13-001, CERN, Geneva, 2013. |
| [42] | CMS Collaboration, 'Performance of b tagging at √ s = 8 TeV in multijet, tt and boosted topology events', CMS Physics Analysis Summary CMS-PAS-BTV-13-001, CERN, Geneva, 2013. |
| [43] | A. L. Read, 'Presentation of search results: the CL s technique', J. Phys. G 28 (2002) 2693, doi:10.1088/0954-3899/28/10/313 . |
| [44] | T. Junk, 'Confidence level computation for combining searches with small statistics', Nucl. Instrum. Meth. A 434 (1999) 435, doi:10.1016/S0168-9002(99)00498-2 , arXiv:hep-ex/9902006 . |
| [45] | ATLAS and CMS Collaborations, LHC Higgs Combination Group, 'Procedure for the LHC Higgs boson search combination in Summer 2011', Technical Report ATL-PHYS-PUB-2011-11, CMS-NOTE-2011-005, CERN, 2011. |
| [46] | CMS Collaboration, 'Search for top-squark pair production in the single-lepton final state in pp collisions at √ s = 8 TeV', Eur. Phys. J. C 73 (2013) 2677, doi:10.1140/epjc/s10052-013-2677-2 , arXiv:1308.1586 . |
## A The CMS Collaboration
## Yerevan Physics Institute, Yerevan, Armenia
V. Khachatryan, A.M. Sirunyan, A. Tumasyan
## Institut f ¨ r Hochenergiephysik der OeAW, Wien, Austria u
W. Adam, T. Bergauer, M. Dragicevic, J. Er¨ o, C. Fabjan , M. Friedl, R. Fr¨hwirth , V.M. Ghete, 1 u 1 C. Hartl, N. H¨ ormann, J. Hrubec, M. Jeitler , 1 W. Kiesenhofer, V. Kn¨ unz, M. Krammer 1 , I. Kr¨ atschmer, D. Liko, I. Mikulec, D. Rabady 2 , B. Rahbaran, H. Rohringer, R. Sch¨ ofbeck, J. Strauss, A. Taurok, W. Treberer-Treberspurg, W. Waltenberger, C.-E. Wulz 1
## National Centre for Particle and High Energy Physics, Minsk, Belarus
V. Mossolov, N. Shumeiko, J. Suarez Gonzalez
## Universiteit Antwerpen, Antwerpen, Belgium
S. Alderweireldt, M. Bansal, S. Bansal, T. Cornelis, E.A. De Wolf, X. Janssen, A. Knutsson, S. Luyckx, S. Ochesanu, B. Roland, R. Rougny, M. Van De Klundert, H. Van Haevermaet, P. Van Mechelen, N. Van Remortel, A. Van Spilbeeck
## Vrije Universiteit Brussel, Brussel, Belgium
F. Blekman, S. Blyweert, J. D'Hondt, N. Daci, N. Heracleous, J. Keaveney, S. Lowette, M. Maes, A. Olbrechts, Q. Python, D. Strom, S. Tavernier, W. Van Doninck, P. Van Mulders, G.P. Van Onsem, I. Villella
## Universit´ e Libre de Bruxelles, Bruxelles, Belgium
C. Caillol, B. Clerbaux, G. De Lentdecker, D. Dobur, L. Favart, A.P.R. Gay, A. Grebenyuk, A. L´ eonard, A. Mohammadi, L. Perni` e , T. Reis, T. Seva, L. Thomas, C. Vander Velde, P. Vanlaer, 2 J. Wang
## Ghent University, Ghent, Belgium
V. Adler, K. Beernaert, L. Benucci, A. Cimmino, S. Costantini, S. Crucy, S. Dildick, A. Fagot, G. Garcia, J. Mccartin, A.A. Ocampo Rios, D. Ryckbosch, S. Salva Diblen, M. Sigamani, N. Strobbe, F. Thyssen, M. Tytgat, E. Yazgan, N. Zaganidis
## Universit´ e Catholique de Louvain, Louvain-la-Neuve, Belgium
S. Basegmez, C. Beluffi , 3 G. Bruno, R. Castello, A. Caudron, L. Ceard, G.G. Da Silveira, C. Delaere, T. du Pree, D. Favart, L. Forthomme, A. Giammanco 4 , J. Hollar, P. Jez, M. Komm, V. Lemaitre, C. Nuttens, D. Pagano, L. Perrini, A. Pin, K. Piotrzkowski, A. Popov 5 , L. Quertenmont, M. Selvaggi, M. Vidal Marono, J.M. Vizan Garcia
## Universit´ e de Mons, Mons, Belgium
N. Beliy, T. Caebergs, E. Daubie, G.H. Hammad
## Centro Brasileiro de Pesquisas Fisicas, Rio de Janeiro, Brazil
W.L. Ald´ a J ´ nior, G.A. Alves, L. Brito, M. Correa Martins Junior, T. Dos Reis Martins, C. Mora u Herrera, M.E. Pol
## Universidade do Estado do Rio de Janeiro, Rio de Janeiro, Brazil
W. Carvalho, J. Chinellato 6 , A. Cust´ odio, E.M. Da Costa, D. De Jesus Damiao, C. De Oliveira Martins, S. Fonseca De Souza, H. Malbouisson, D. Matos Figueiredo, L. Mundim, H. Nogima, W.L. Prado Da Silva, J. Santaolalla, A. Santoro, A. Sznajder, E.J. Tonelli Manganote 6 , A. Vilela Pereira
## Universidade Estadual Paulista a , Universidade Federal do ABC b , S˜ ao Paulo, Brazil
C.A. Bernardes , S. Dogra , T.R. Fernandez Perez Tomei , E.M. Gregores , P.G. Mercadante , b a a b b S.F. Novaes , Sandra S. Padula a a
## Institute for Nuclear Research and Nuclear Energy, Sofia, Bulgaria
A. Aleksandrov, V. Genchev 2 , P. Iaydjiev, A. Marinov, S. Piperov, M. Rodozov, S. Stoykova, G. Sultanov, V. Tcholakov, M. Vutova
## University of Sofia, Sofia, Bulgaria
A. Dimitrov, I. Glushkov, R. Hadjiiska, V. Kozhuharov, L. Litov, B. Pavlov, P. Petkov
## Institute of High Energy Physics, Beijing, China
J.G. Bian, G.M. Chen, H.S. Chen, M. Chen, R. Du, C.H. Jiang, S. Liang, R. Plestina 7 , J. Tao, X. Wang, Z. Wang
State Key Laboratory of Nuclear Physics and Technology, Peking University, Beijing, China C. Asawatangtrakuldee, Y. Ban, Y. Guo, Q. Li, W. Li, S. Liu, Y. Mao, S.J. Qian, H. Teng, D. Wang, W. Zou
## Universidad de Los Andes, Bogota, Colombia
C. Avila, L.F. Chaparro Sierra, C. Florez, J.P. Gomez, B. Gomez Moreno, J.C. Sanabria
## University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Split, Croatia
N. Godinovic, D. Lelas, D. Polic, I. Puljak
## University of Split, Faculty of Science, Split, Croatia
Z. Antunovic, M. Kovac
## Institute Rudjer Boskovic, Zagreb, Croatia
V. Brigljevic, K. Kadija, J. Luetic, D. Mekterovic, L. Sudic
## University of Cyprus, Nicosia, Cyprus
A. Attikis, G. Mavromanolakis, J. Mousa, C. Nicolaou, F. Ptochos, P.A. Razis
## Charles University, Prague, Czech Republic
M. Bodlak, M. Finger, M. Finger Jr. 8
## Academy of Scientific Research and Technology of the Arab Republic of Egypt, Egyptian Network of High Energy Physics, Cairo, Egypt
Y. Assran 9 , S. Elgammal 10 , M.A. Mahmoud 11 , A. Radi 10,12
## National Institute of Chemical Physics and Biophysics, Tallinn, Estonia
M. Kadastik, M. Murumaa, M. Raidal, A. Tiko
## Department of Physics, University of Helsinki, Helsinki, Finland
P. Eerola, G. Fedi, M. Voutilainen
## Helsinki Institute of Physics, Helsinki, Finland
J. H¨rk¨ onen, V. Karim¨ aki, R. Kinnunen, M.J. Kortelainen, T. Lamp´ en, K. Lassila-Perini, S. Lehti, a T. Lind´ en, P. Luukka, T. M¨ aenp¨ a¨, a T. Peltola, E. Tuominen, J. Tuominiemi, E. Tuovinen, L. Wendland
## Lappeenranta University of Technology, Lappeenranta, Finland
J. Talvitie, T. Tuuva
## DSM/IRFU, CEA/Saclay, Gif-sur-Yvette, France
M. Besancon, F. Couderc, M. Dejardin, D. Denegri, B. Fabbro, J.L. Faure, C. Favaro, F. Ferri, S. Ganjour, A. Givernaud, P. Gras, G. Hamel de Monchenault, P. Jarry, E. Locci, J. Malcles, J. Rander, A. Rosowsky, M. Titov
## Laboratoire Leprince-Ringuet, Ecole Polytechnique, IN2P3-CNRS, Palaiseau, France
S. Baffioni, F. Beaudette, P. Busson, C. Charlot, T. Dahms, M. Dalchenko, L. Dobrzynski, N. Filipovic, A. Florent, R. Granier de Cassagnac, L. Mastrolorenzo, P. Min´ e, C. Mironov, I.N. Naranjo, M. Nguyen, C. Ochando, P. Paganini, S. Regnard, R. Salerno, J.B. Sauvan, Y. Sirois, C. Veelken, Y. Yilmaz, A. Zabi
## Institut Pluridisciplinaire Hubert Curien, Universit´ e de Strasbourg, Universit´ e de Haute Alsace Mulhouse, CNRS/IN2P3, Strasbourg, France
J.-L. Agram 13 , J. Andrea, A. Aubin, D. Bloch, J.-M. Brom, E.C. Chabert, C. Collard, E. Conte 13 , J.-C. Fontaine 13 , D. Gel´ e, U. Goerlach, C. Goetzmann, A.-C. Le Bihan, P. Van Hove
## Centre de Calcul de l'Institut National de Physique Nucleaire et de Physique des Particules, CNRS/IN2P3, Villeurbanne, France
S. Gadrat
## Universit´ e de Lyon, Universit´ e Claude Bernard Lyon 1, CNRS-IN2P3, Institut de Physique Nucl´ eaire de Lyon, Villeurbanne, France
S. Beauceron, N. Beaupere, G. Boudoul 2 , E. Bouvier, S. Brochet, C.A. Carrillo Montoya, J. Chasserat, R. Chierici, D. Contardo 2 , P. Depasse, H. El Mamouni, J. Fan, J. Fay, S. Gascon, M. Gouzevitch, B. Ille, T. Kurca, M. Lethuillier, L. Mirabito, S. Perries, J.D. Ruiz Alvarez, D. Sabes, L. Sgandurra, V. Sordini, M. Vander Donckt, P. Verdier, S. Viret, H. Xiao
## Institute of High Energy Physics and Informatization, Tbilisi State University, Tbilisi, Georgia
Z. Tsamalaidze 8
## RWTHAachen University, I. Physikalisches Institut, Aachen, Germany
C. Autermann, S. Beranek, M. Bontenackels, M. Edelhoff, L. Feld, O. Hindrichs, K. Klein, A. Ostapchuk, A. Perieanu, F. Raupach, J. Sammet, S. Schael, H. Weber, B. Wittmer, V. Zhukov 5
## RWTHAachen University, III. Physikalisches Institut A, Aachen, Germany
M. Ata, M. Brodski, E. Dietz-Laursonn, D. Duchardt, M. Erdmann, R. Fischer, A. G¨ uth, T. Hebbeker, C. Heidemann, K. Hoepfner, D. Klingebiel, S. Knutzen, P. Kreuzer, M. Merschmeyer, A. Meyer, P. Millet, M. Olschewski, K. Padeken, P. Papacz, H. Reithler, S.A. Schmitz, L. Sonnenschein, D. Teyssier, S. Th¨er, M. Weber u
## RWTHAachen University, III. Physikalisches Institut B, Aachen, Germany
V. Cherepanov, Y. Erdogan, G. Fl¨ ugge, H. Geenen, M. Geisler, W. Haj Ahmad, A. Heister, F. Hoehle, B. Kargoll, T. Kress, Y. Kuessel, J. Lingemann , A. Nowack, I.M. Nugent, L. Perchalla, 2 O. Pooth, A. Stahl
## Deutsches Elektronen-Synchrotron, Hamburg, Germany
I. Asin, N. Bartosik, J. Behr, W. Behrenhoff, U. Behrens, A.J. Bell, M. Bergholz 14 , A. Bethani, K. Borras, A. Burgmeier, A. Cakir, L. Calligaris, A. Campbell, S. Choudhury, F. Costanza, C. Diez Pardos, S. Dooling, T. Dorland, G. Eckerlin, D. Eckstein, T. Eichhorn, G. Flucke, J. Garay Garcia, A. Geiser, P. Gunnellini, J. Hauk, G. Hellwig, M. Hempel, D. Horton, H. Jung, A. Kalogeropoulos, M. Kasemann, P. Katsas, J. Kieseler, C. Kleinwort, D. Kr¨cker, u W. Lange, J. Leonard, K. Lipka, A. Lobanov, W. Lohmann 14 , B. Lutz, R. Mankel, I. Marfin, I.-A. Melzer-Pellmann, A.B. Meyer, J. Mnich, A. Mussgiller, S. Naumann-Emme, A. Nayak,
O. Novgorodova, F. Nowak, E. Ntomari, H. Perrey, D. Pitzl, R. Placakyte, A. Raspereza, P.M. Ribeiro Cipriano, E. Ron, M. ¨ . O Sahin, J. Salfeld-Nebgen, P. Saxena, R. Schmidt 14 , T. Schoerner-Sadenius, M. Schr¨ oder, C. Seitz, S. Spannagel, A.D.R. Vargas Trevino, R. Walsh, C. Wissing
## University of Hamburg, Hamburg, Germany
M. Aldaya Martin, V. Blobel, M. Centis Vignali, A.r. Draeger, J. Erfle, E. Garutti, K. Goebel, M. G¨ orner, J. Haller, M. Hoffmann, R.S. H¨ oing, H. Kirschenmann, R. Klanner, R. Kogler, J. Lange, T. Lapsien, T. Lenz, I. Marchesini, J. Ott, T. Peiffer, N. Pietsch, J. Poehlsen, T. Poehlsen, D. Rathjens, C. Sander, H. Schettler, P. Schleper, E. Schlieckau, A. Schmidt, M. Seidel, V. Sola, H. Stadie, G. Steinbr¨ck, D. Troendle, E. Usai, L. Vanelderen, A. Vanhoefer u
## Institut f ¨ r Experimentelle Kernphysik, Karlsruhe, Germany u
C. Barth, C. Baus, J. Berger, C. B¨ser, E. Butz, T. Chwalek, W. De Boer, A. Descroix, A. Dierlamm, o M. Feindt, F. Frensch, M. Giffels, F. Hartmann 2 , T. Hauth 2 , U. Husemann, I. Katkov 5 , A. Kornmayer 2 , E. Kuznetsova, P. Lobelle Pardo, M.U. Mozer, Th. M¨ uller, A. N¨ urnberg, G. Quast, K. Rabbertz, F. Ratnikov, S. R¨ ocker, H.J. Simonis, F.M. Stober, R. Ulrich, J. WagnerKuhr, S. Wayand, T. Weiler, R. Wolf
## Institute of Nuclear and Particle Physics (INPP), NCSR Demokritos, Aghia Paraskevi, Greece
G. Anagnostou, G. Daskalakis, T. Geralis, V.A. Giakoumopoulou, A. Kyriakis, D. Loukas, A. Markou, C. Markou, A. Psallidas, I. Topsis-Giotis
## University of Athens, Athens, Greece
S. Kesisoglou, A. Panagiotou, N. Saoulidou, E. Stiliaris
## University of Io´ annina, Io´ annina, Greece
X. Aslanoglou, I. Evangelou, G. Flouris, C. Foudas, P. Kokkas, N. Manthos, I. Papadopoulos,
E. Paradas
## Wigner Research Centre for Physics, Budapest, Hungary
G. Bencze, C. Hajdu, P. Hidas, D. Horvath 15 , F. Sikler, V. Veszpremi, G. Vesztergombi 16 , A.J. Zsigmond
## Institute of Nuclear Research ATOMKI, Debrecen, Hungary
N. Beni, S. Czellar, J. Karancsi 17 , J. Molnar, J. Palinkas, Z. Szillasi
## University of Debrecen, Debrecen, Hungary
P. Raics, Z.L. Trocsanyi, B. Ujvari
## National Institute of Science Education and Research, Bhubaneswar, India
S.K. Swain
## Panjab University, Chandigarh, India
S.B. Beri, V. Bhatnagar, R. Gupta, U.Bhawandeep, A.K. Kalsi, M. Kaur, M. Mittal, N. Nishu, J.B. Singh
## University of Delhi, Delhi, India
Ashok Kumar, Arun Kumar, S. Ahuja, A. Bhardwaj, B.C. Choudhary, A. Kumar, S. Malhotra, M. Naimuddin, K. Ranjan, V. Sharma
## Saha Institute of Nuclear Physics, Kolkata, India
S. Banerjee, S. Bhattacharya, K. Chatterjee, S. Dutta, B. Gomber, Sa. Jain, Sh. Jain, R. Khurana, A. Modak, S. Mukherjee, D. Roy, S. Sarkar, M. Sharan
## Bhabha Atomic Research Centre, Mumbai, India
A. Abdulsalam, D. Dutta, S. Kailas, V. Kumar, A.K. Mohanty 2 , L.M. Pant, P. Shukla, A. Topkar
## Tata Institute of Fundamental Research, Mumbai, India
T. Aziz, S. Banerjee, S. Bhowmik 18 , R.M. Chatterjee, R.K. Dewanjee, S. Dugad, S. Ganguly, S. Ghosh, M. Guchait, A. Gurtu 19 , G. Kole, S. Kumar, M. Maity 18 , G. Majumder, K. Mazumdar, G.B. Mohanty, B. Parida, K. Sudhakar, N. Wickramage 20
## Institute for Research in Fundamental Sciences (IPM), Tehran, Iran
H. Bakhshiansohi, H. Behnamian, S.M. Etesami 21 , A. Fahim 22 , R. Goldouzian, A. Jafari, M. Khakzad, M. Mohammadi Najafabadi, M. Naseri, S. Paktinat Mehdiabadi, F. Rezaei Hosseinabadi, B. Safarzadeh 23 , M. Zeinali
## University College Dublin, Dublin, Ireland
M. Felcini, M. Grunewald
## INFN Sezione di Bari a , Universit` a di Bari b , Politecnico di Bari c , Bari, Italy
M. Abbrescia a b , , L. Barbone a b , , C. Calabria a b , , S.S. Chhibra a b , , A. Colaleo , D. Creanza a a c , , N. De Filippis a c , , M. De Palma a b , , L. Fiore a , G. Iaselli a c , , G. Maggi a c , , M. Maggi , S. My a a c , , S. Nuzzo a b , , A. Pompili a b , , G. Pugliese a c , , R. Radogna a b , ,2 , G. Selvaggi a b , , L. Silvestris a ,2 , G. Singh a b , , a b , a a
R. Venditti , P. Verwilligen , G. Zito
## INFN Sezione di Bologna a , Universit` a di Bologna b , Bologna, Italy
G. Abbiendi , a A.C. Benvenuti , a D. Bonacorsi a b , , S. Braibant-Giacomelli a b , , L. Brigliadori a b , , R. Campanini a b , , P. Capiluppi a b , , A. Castro a b , , F.R. Cavallo , G. Codispoti a a b , , M. Cuffiani a b , , G.M. Dallavalle , a F. Fabbri , a A. Fanfani a b , , D. Fasanella a b , , P. Giacomelli , a C. Grandi , a L. Guiducci a b , , S. Marcellini , a G. Masetti a ,2 , A. Montanari , a F.L. Navarria a b , , A. Perrotta , a a b , a b , a b , a b , a b , a b ,
- F. Primavera , A.M. Rossi , T. Rovelli , G.P. Siroli , N. Tosi , R. Travaglini
## INFN Sezione di Catania a , Universit` a di Catania b , CSFNSM , Catania, Italy c
S. Albergo a b , , G. Cappello , a M. Chiorboli a b , , S. Costa a b , , F. Giordano a c , ,2 , R. Potenza a b , , a b , a b ,
A. Tricomi , C. Tuve
## INFN Sezione di Firenze a , Universit` a di Firenze b , Firenze, Italy
G. Barbagli , V. Ciulli a a b , , C. Civinini , R. D'Alessandro a a b , , E. Focardi a b , , E. Gallo , S. Gonzi a a b , , V. Gori a b , ,2 , P. Lenzi a b , , M. Meschini , S. Paoletti , G. Sguazzoni , A. Tropiano a a a a b ,
## INFN Laboratori Nazionali di Frascati, Frascati, Italy
L. Benussi, S. Bianco, F. Fabbri, D. Piccolo
## INFN Sezione di Genova a , Universit` a di Genova b , Genova, Italy
F. Ferro , M. Lo Vetere a a b , , E. Robutti , S. Tosi a a b ,
## INFN Sezione di Milano-Bicocca a , Universit` a di Milano-Bicocca b , Milano, Italy
M.E. Dinardo a b , , S. Fiorendi a b , ,2 , S. Gennai a ,2 , R. Gerosa 2 , A. Ghezzi a b , , P. Govoni a b , , M.T. Lucchini a b , ,2 , S. Malvezzi , R.A. Manzoni a a b , , A. Martelli a b , , B. Marzocchi, D. Menasce , a L. Moroni , M. Paganoni a a b , , D. Pedrini , S. Ragazzi a a b , , N. Redaelli , T. Tabarelli de Fatis a a b ,
## INFN Sezione di Napoli a , Universit` a di Napoli 'Federico II' b , Universit` a della Basilicata (Potenza) c , Universit` a G. Marconi (Roma) d , Napoli, Italy
S. Buontempo , a N. Cavallo a c , , S. Di Guida a d , ,2 , F. Fabozzi a c , , A.O.M. Iorio a b , , L. Lista , a a d , ,2 a a ,2
S. Meola , M. Merola , P. Paolucci
## INFN Sezione di Padova a , Universit` a di Padova b , Universit` a di Trento (Trento) c , Padova, Italy
- P. Azzi , N. Bacchetta , D. Bisello a a a b , , A. Branca a b , , R. Carlin a b , , P. Checchia , M. Dall'Osso a a b , ,
- T. Dorigo , a U. Dosselli , a M. Galanti a b , , F. Gasparini a b , , U. Gasparini a b , , P. Giubilato a b , ,
- A. Gozzelino , K. Kanishchev a a c , , S. Lacaprara , M. Margoni a a b , , A.T. Meneguzzo a b , , J. Pazzini a b , ,
- N. Pozzobon a b , , P. Ronchese a b , , F. Simonetto a b , , E. Torassa , a M. Tosi a b , , P. Zotto a b , , A. Zucchetta a b , , G. Zumerle a b ,
## INFN Sezione di Pavia a , Universit` a di Pavia b , Pavia, Italy
M. Gabusi a b , , S.P. Ratti a b , , C. Riccardi a b , , P. Salvini , P. Vitulo a a b ,
## INFN Sezione di Perugia a , Universit` a di Perugia b , Perugia, Italy
M. Biasini a b , , G.M. Bilei a , D. Ciangottini a b , , L. Fan` o a b , , P. Lariccia a b , , G. Mantovani a b , , M. Menichelli , F. Romeo a a b , , A. Saha , A. Santocchia a a b , , A. Spiezia a b , ,2
## INFN Sezione di Pisa a , Universit` a di Pisa b , Scuola Normale Superiore di Pisa c , Pisa, Italy
K. Androsov a ,24 , P . Azzurri a , G. Bagliesi a , J. Bernardini a , T. Boccali a , G. Broccolo a c , , R. Castaldi , a M.A. Ciocci a ,24 , R. Dell'Orso , S. Donato a a c , , F. Fiori a c , , L. Fo` a a c , , A. Giassi , M.T. Grippo a a ,24 , F. Ligabue a c , , T. Lomtadze , L. Martini a a b , , A. Messineo a b , , C.S. Moon a ,25 , F. Palla a ,2 , A. Rizzi a b , , A. Savoy-Navarro a ,26 , A.T. Serban , P. Spagnolo , P. Squillacioti a a a ,24 , R. Tenchini , G. Tonelli a a b , , a a a c , ,2
- A. Venturi , P.G. Verdini , C. Vernieri
## INFN Sezione di Roma a , Universit` a di Roma b , Roma, Italy
- a b , a a b , a b , a a b , a
- L. Barone , F. Cavallari , G. D'imperio , D. Del Re , M. Diemoz , M. Grassi , C. Jorda , E. Longo a b , , F. Margaroli a b , , P. Meridiani , F. Micheli a a b , ,2 , S. Nourbakhsh a b , , G. Organtini a b , , R. Paramatti , S. Rahatlou a a b , , C. Rovelli , F. Santanastasio a a b , , L. Soffi a b , ,2 , P. Traczyk a b ,
## INFN Sezione di Torino a , Universit` a di Torino b , Universit` a del Piemonte Orientale (Novara) c , Torino, Italy
N. Amapane , a b , R. Arcidiacono a c , , S. Argiro a b , ,2 , M. Arneodo a c , , R. Bellan a b , , C. Biino , a N. Cartiglia , S. Casasso a a b , ,2 , M. Costa a b , , A. Degano a b , , N. Demaria , L. Finco a a b , , C. Mariotti , a S. Maselli , a E. Migliore a b , , V. Monaco , a b , M. Musich , a M.M. Obertino a c , ,2 , G. Ortona a b , , L. Pacher a b , , N. Pastrone , M. Pelliccioni , G.L. Pinna Angioni a a a b , , A. Potenza a b , , A. Romero a b , , a c , a b , a b , a a
M. Ruspa , R. Sacchi , A. Solano , A. Staiano , U. Tamponi
## INFN Sezione di Trieste a , Universit` a di Trieste b , Trieste, Italy
S. Belforte , a V. Candelise a b , , M. Casarsa , a F. Cossutti , a G. Della Ricca a b , , B. Gobbo , C. La a Licata a b , , M. Marone a b , , D. Montanino a b , , A. Schizzi a b , ,2 , T. Umer a b , , A. Zanetti a
## Kangwon National University, Chunchon, Korea
S. Chang, A. Kropivnitskaya, S.K. Nam
## Kyungpook National University, Daegu, Korea
D.H. Kim, G.N. Kim, M.S. Kim, D.J. Kong, S. Lee, Y.D. Oh, H. Park, A. Sakharov, D.C. Son
## Chonbuk National University, Jeonju, Korea
T.J. Kim
## Chonnam National University, Institute for Universe and Elementary Particles, Kwangju, Korea
J.Y. Kim, S. Song
## Korea University, Seoul, Korea
S. Choi, D. Gyun, B. Hong, M. Jo, H. Kim, Y. Kim, B. Lee, K.S. Lee, S.K. Park, Y. Roh
## University of Seoul, Seoul, Korea
M. Choi, J.H. Kim, I.C. Park, S. Park, G. Ryu, M.S. Ryu
## Sungkyunkwan University, Suwon, Korea
Y. Choi, Y.K. Choi, J. Goh, D. Kim, E. Kwon, J. Lee, H. Seo, I. Yu
## Vilnius University, Vilnius, Lithuania
A. Juodagalvis
## National Centre for Particle Physics, Universiti Malaya, Kuala Lumpur, Malaysia
J.R. Komaragiri, M.A.B. Md Ali
## Centro de Investigacion y de Estudios Avanzados del IPN, Mexico City, Mexico
H. Castilla-Valdez, E. De La Cruz-Burelo, I. Heredia-de La Cruz 27 , R. Lopez-Fernandez, A. Sanchez-Hernandez
## Universidad Iberoamericana, Mexico City, Mexico
S. Carrillo Moreno, F. Vazquez Valencia
## Benemerita Universidad Autonoma de Puebla, Puebla, Mexico
I. Pedraza, H.A. Salazar Ibarguen
## Universidad Aut´ onoma de San Luis Potos´ ı, San Luis Potos´ ı, Mexico
E. Casimiro Linares, A. Morelos Pineda
## University of Auckland, Auckland, New Zealand
D. Krofcheck
## University of Canterbury, Christchurch, New Zealand
P.H. Butler, S. Reucroft
## National Centre for Physics, Quaid-I-Azam University, Islamabad, Pakistan
A. Ahmad, M. Ahmad, Q. Hassan, H.R. Hoorani, S. Khalid, W.A. Khan, T. Khurshid, M.A. Shah, M. Shoaib
## National Centre for Nuclear Research, Swierk, Poland
H. Bialkowska, M. Bluj, B. Boimska, T. Frueboes, M. G´ orski, M. Kazana, K. Nawrocki, K. Romanowska-Rybinska, M. Szleper, P. Zalewski
## Institute of Experimental Physics, Faculty of Physics, University of Warsaw, Warsaw, Poland
G. Brona, K. Bunkowski, M. Cwiok, W. Dominik, K. Doroba, A. Kalinowski, M. Konecki, J. Krolikowski, M. Misiura, M. Olszewski, W. Wolszczak
## Laborat´ orio de Instrumenta¸ c˜ ao e F´ ısica Experimental de Part´ ıculas, Lisboa, Portugal
- P. Bargassa, C. Beir˜ ao Da Cruz E Silva, P. Faccioli, P.G. Ferreira Parracho, M. Gallinaro, F. Nguyen, J. Rodrigues Antunes, J. Seixas, J. Varela, P. Vischia
## Joint Institute for Nuclear Research, Dubna, Russia
S. Afanasiev, P. Bunin, M. Gavrilenko, I. Golutvin, I. Gorbunov, A. Kamenev, V. Karjavin, V. Konoplyanikov, A. Lanev, A. Malakhov, V. Matveev 28 , P. Moisenz, V. Palichik, V. Perelygin, S. Shmatov, N. Skatchkov, V. Smirnov, A. Zarubin
## Petersburg Nuclear Physics Institute, Gatchina (St. Petersburg), Russia
V. Golovtsov, Y. Ivanov, V. Kim 29 , P . Levchenko, V. Murzin, V. Oreshkin, I. Smirnov, V. Sulimov, L. Uvarov, S. Vavilov, A. Vorobyev, An. Vorobyev
## Institute for Nuclear Research, Moscow, Russia
Yu. Andreev, A. Dermenev, S. Gninenko, N. Golubev, M. Kirsanov, N. Krasnikov, A. Pashenkov, D. Tlisov, A. Toropin
## Institute for Theoretical and Experimental Physics, Moscow, Russia
V. Epshteyn, V. Gavrilov, N. Lychkovskaya, V. Popov, G. Safronov, S. Semenov, A. Spiridonov, V. Stolin, E. Vlasov, A. Zhokin
## P.N. Lebedev Physical Institute, Moscow, Russia
V. Andreev, M. Azarkin, I. Dremin, M. Kirakosyan, A. Leonidov, G. Mesyats, S.V. Rusakov, A. Vinogradov
## Skobeltsyn Institute of Nuclear Physics, Lomonosov Moscow State University, Moscow, Russia
A. Belyaev, E. Boos, V. Bunichev, M. Dubinin 30 , L. Dudko, A. Ershov, V. Klyukhin, O. Kodolova, I. Lokhtin, S. Obraztsov, S. Petrushanko, V. Savrin, A. Snigirev
## State Research Center of Russian Federation, Institute for High Energy Physics, Protvino, Russia
I. Azhgirey, I. Bayshev, S. Bitioukov, V. Kachanov, A. Kalinin, D. Konstantinov, V. Krychkine, V. Petrov, R. Ryutin, A. Sobol, L. Tourtchanovitch, S. Troshin, N. Tyurin, A. Uzunian, A. Volkov
## University of Belgrade, Faculty of Physics and Vinca Institute of Nuclear Sciences, Belgrade, Serbia
P. Adzic 31 , M. Ekmedzic, J. Milosevic, V. Rekovic
## Centro de Investigaciones Energ´ eticas Medioambientales y Tecnol´ ogicas (CIEMAT), Madrid, Spain
J. Alcaraz Maestre, C. Battilana, E. Calvo, M. Cerrada, M. Chamizo Llatas, N. Colino, B. De La Cruz, A. Delgado Peris, D. Dom´ ınguez V´ azquez, A. Escalante Del Valle, C. Fernandez Bedoya, J.P. Fern´ andez Ramos, J. Flix, M.C. Fouz, P. Garcia-Abia, O. Gonzalez Lopez, S. Goy Lopez, J.M. Hernandez, M.I. Josa, G. Merino, E. Navarro De Martino, A. P´ erez-Calero Yzquierdo, J. Puerta Pelayo, A. Quintario Olmeda, I. Redondo, L. Romero, M.S. Soares
## Universidad Aut´ onoma de Madrid, Madrid, Spain
C. Albajar, J.F. de Troc´niz, M. Missiroli, D. Moran o
## Universidad de Oviedo, Oviedo, Spain
H. Brun, J. Cuevas, J. Fernandez Menendez, S. Folgueras, I. Gonzalez Caballero, L. Lloret Iglesias
## Instituto de F´ ısica de Cantabria (IFCA), CSIC-Universidad de Cantabria, Santander, Spain
J.A. Brochero Cifuentes, I.J. Cabrillo, A. Calderon, J. Duarte Campderros, M. Fernandez, G. Gomez, A. Graziano, A. Lopez Virto, J. Marco, R. Marco, C. Martinez Rivero, F. Matorras, F.J. Munoz Sanchez, J. Piedra Gomez, T. Rodrigo, A.Y. Rodr´ ıguez-Marrero, A. Ruiz-Jimeno, L. Scodellaro, I. Vila, R. Vilar Cortabitarte
## CERN, European Organization for Nuclear Research, Geneva, Switzerland
D. Abbaneo, E. Auffray, G. Auzinger, M. Bachtis, P. Baillon, A.H. Ball, D. Barney, A. Benaglia, J. Bendavid, L. Benhabib, J.F. Benitez, C. Bernet 7 , G. Bianchi, P. Bloch, A. Bocci, A. Bonato, O. Bondu, C. Botta, H. Breuker, T. Camporesi, G. Cerminara, S. Colafranceschi 32 , M. D'Alfonso, D. d'Enterria, A. Dabrowski, A. David, F. De Guio, A. De Roeck, S. De Visscher, M. Dobson, M. Dordevic, N. Dupont-Sagorin, A. Elliott-Peisert, J. Eugster, G. Franzoni, W. Funk, D. Gigi, K. Gill, D. Giordano, M. Girone, F. Glege, R. Guida, S. Gundacker, M. Guthoff, J. Hammer,
- M. Hansen, P. Harris, J. Hegeman, V. Innocente, P. Janot, K. Kousouris, K. Krajczar, P. Lecoq, C. Lourenc ¸o, N. Magini, L. Malgeri, M. Mannelli, J. Marrouche, L. Masetti, F. Meijers, S. Mersi, E. Meschi, F. Moortgat, S. Morovic, M. Mulders, P. Musella, L. Orsini, L. Pape, E. Perez, L. Perrozzi, A. Petrilli, G. Petrucciani, A. Pfeiffer, M. Pierini, M. Pimi¨ a, D. Piparo, M. Plagge, A. Racz, G. Rolandi 33 , M. Rovere, H. Sakulin, C. Sch¨ afer, C. Schwick, A. Sharma, P. Siegrist, P. Silva, M. Simon, P. Sphicas 34 , D. Spiga, J. Steggemann, B. Stieger, M. Stoye, Y. Takahashi, D. Treille, A. Tsirou, G.I. Veres 16 , J.R. Vlimant, N. Wardle, H.K. W¨ ohri, H. Wollny, W.D. Zeuner
## Paul Scherrer Institut, Villigen, Switzerland
W. Bertl, K. Deiters, W. Erdmann, R. Horisberger, Q. Ingram, H.C. Kaestli, D. Kotlinski, U. Langenegger, D. Renker, T. Rohe
## Institute for Particle Physics, ETH Zurich, Zurich, Switzerland
F. Bachmair, L. B¨ ani, L. Bianchini, M.A. Buchmann, B. Casal, N. Chanon, A. Deisher, G. Dissertori, M. Dittmar, M. Doneg` a, M. D¨nser, P. Eller, C. Grab, D. Hits, W. Lustermann, u B. Mangano, A.C. Marini, P. Martinez Ruiz del Arbol, D. Meister, N. Mohr, C. N¨ ageli 35 , F. NessiTedaldi, F. Pandolfi, F. Pauss, M. Peruzzi, M. Quittnat, L. Rebane, M. Rossini, A. Starodumov 36 , M. Takahashi, K. Theofilatos, R. Wallny, H.A. Weber
## Universit¨ at Z ¨ rich, Zurich, Switzerland u
C. Amsler 37 , M.F. Canelli, V. Chiochia, A. De Cosa, A. Hinzmann, T. Hreus, B. Kilminster, C. Lange, B. Millan Mejias, J. Ngadiuba, P. Robmann, F.J. Ronga, S. Taroni, M. Verzetti, Y. Yang
## National Central University, Chung-Li, Taiwan
M. Cardaci, K.H. Chen, C. Ferro, C.M. Kuo, W. Lin, Y.J. Lu, R. Volpe, S.S. Yu
## National Taiwan University (NTU), Taipei, Taiwan
P. Chang, Y.H. Chang, Y.W. Chang, Y. Chao, K.F. Chen, P.H. Chen, C. Dietz, U. Grundler, W.-
- S. Hou, K.Y. Kao, Y.J. Lei, Y.F. Liu, R.-S. Lu, D. Majumder, E. Petrakou, Y.M. Tzeng, R. Wilken
## Chulalongkorn University, Faculty of Science, Department of Physics, Bangkok, Thailand
## B. Asavapibhop, N. Srimanobhas, N. Suwonjandee
## Cukurova University, Adana, Turkey
A. Adiguzel, M.N. Bakirci 38 , S. Cerci 39 , C. Dozen, I. Dumanoglu, E. Eskut, S. Girgis, G. Gokbulut, E. Gurpinar, I. Hos, E.E. Kangal, A. Kayis Topaksu, G. Onengut 40 , K. Ozdemir, S. Ozturk 38 , A. Polatoz, D. Sunar Cerci 39 , B. Tali 39 , H. Topakli 38 , M. Vergili
## Middle East Technical University, Physics Department, Ankara, Turkey
I.V. Akin, B. Bilin, S. Bilmis, H. Gamsizkan, G. Karapinar 41 , K. Ocalan, S. Sekmen, U.E. Surat, M. Yalvac, M. Zeyrek
## Bogazici University, Istanbul, Turkey
E. G¨ ulmez, B. Isildak 42 , M. Kaya 43 , O. Kaya 44
## Istanbul Technical University, Istanbul, Turkey
K. Cankocak, F.I. Vardarlı
National Scientific Center, Kharkov Institute of Physics and Technology, Kharkov, Ukraine L. Levchuk, P. Sorokin
## University of Bristol, Bristol, United Kingdom
J.J. Brooke, E. Clement, D. Cussans, H. Flacher, R. Frazier, J. Goldstein, M. Grimes, G.P. Heath, H.F. Heath, J. Jacob, L. Kreczko, C. Lucas, Z. Meng, D.M. Newbold 45 , S. Paramesvaran, A. Poll, S. Senkin, V.J. Smith, T. Williams
## Rutherford Appleton Laboratory, Didcot, United Kingdom
K.W. Bell, A. Belyaev 46 , C. Brew, R.M. Brown, D.J.A. Cockerill, J.A. Coughlan, K. Harder, S. Harper, E. Olaiya, D. Petyt, C.H. Shepherd-Themistocleous, A. Thea, I.R. Tomalin, W.J. Womersley, S.D. Worm
## Imperial College, London, United Kingdom
M. Baber, R. Bainbridge, O. Buchmuller, D. Burton, D. Colling, N. Cripps, M. Cutajar, P. Dauncey, G. Davies, M. Della Negra, P. Dunne, W. Ferguson, J. Fulcher, D. Futyan, A. Gilbert, G. Hall, G. Iles, M. Jarvis, G. Karapostoli, M. Kenzie, R. Lane, R. Lucas 45 , L. Lyons, A.M. Magnan, S. Malik, B. Mathias, J. Nash, A. Nikitenko 36 , J. Pela, M. Pesaresi, K. Petridis, D.M. Raymond, S. Rogerson, A. Rose, C. Seez, P. Sharp , † A. Tapper, M. Vazquez Acosta, T. Virdee, S.C. Zenz
## Brunel University, Uxbridge, United Kingdom
J.E. Cole, P.R. Hobson, A. Khan, P. Kyberd, D. Leggat, D. Leslie, W. Martin, I.D. Reid, P. Symonds, L. Teodorescu, M. Turner
## Baylor University, Waco, USA
J. Dittmann, K. Hatakeyama, A. Kasmi, H. Liu, T. Scarborough
## The University of Alabama, Tuscaloosa, USA
O. Charaf, S.I. Cooper, C. Henderson, P. Rumerio
## Boston University, Boston, USA
A. Avetisyan, T. Bose, C. Fantasia, P. Lawson, C. Richardson, J. Rohlf, D. Sperka, J. St. John, L. Sulak
## Brown University, Providence, USA
J. Alimena, E. Berry, S. Bhattacharya, G. Christopher, D. Cutts, Z. Demiragli, N. Dhingra, A. Ferapontov, A. Garabedian, U. Heintz, G. Kukartsev, E. Laird, G. Landsberg, M. Luk, M. Narain, M. Segala, T. Sinthuprasith, T. Speer, J. Swanson
## University of California, Davis, Davis, USA
R. Breedon, G. Breto, M. Calderon De La Barca Sanchez, S. Chauhan, M. Chertok, J. Conway, R. Conway, P.T. Cox, R. Erbacher, M. Gardner, W. Ko, R. Lander, T. Miceli, M. Mulhearn, D. Pellett, J. Pilot, F. Ricci-Tam, M. Searle, S. Shalhout, J. Smith, M. Squires, D. Stolp, M. Tripathi, S. Wilbur, R. Yohay
## University of California, Los Angeles, USA
R. Cousins, P. Everaerts, C. Farrell, J. Hauser, M. Ignatenko, G. Rakness, E. Takasugi, V. Valuev, M. Weber
## University of California, Riverside, Riverside, USA
K. Burt, R. Clare, J. Ellison, J.W. Gary, G. Hanson, J. Heilman, M. Ivova Rikova, P. Jandir, E. Kennedy, F. Lacroix, O.R. Long, A. Luthra, M. Malberti, H. Nguyen, M. Olmedo Negrete, A. Shrinivas, S. Sumowidagdo, S. Wimpenny
## University of California, San Diego, La Jolla, USA
W. Andrews, J.G. Branson, G.B. Cerati, S. Cittolin, R.T. D'Agnolo, D. Evans, A. Holzner, R. Kelley, D. Klein, M. Lebourgeois, J. Letts, I. Macneill, D. Olivito, S. Padhi, C. Palmer, M. Pieri, M. Sani, V. Sharma, S. Simon, E. Sudano, M. Tadel, Y. Tu, A. Vartak, C. Welke, F. W¨rthwein, u
A. Yagil, J. Yoo
## University of California, Santa Barbara, Santa Barbara, USA
D. Barge, J. Bradmiller-Feld, C. Campagnari, T. Danielson, A. Dishaw, K. Flowers, M. Franco
Sevilla, P. Geffert, C. George, F. Golf, L. Gouskos, J. Incandela, C. Justus, N. Mccoll, J. Richman, D. Stuart, W. To, C. West
## California Institute of Technology, Pasadena, USA
A. Apresyan, A. Bornheim, J. Bunn, Y. Chen, E. Di Marco, J. Duarte, A. Mott, H.B. Newman, C. Pena, C. Rogan, M. Spiropulu, V. Timciuc, R. Wilkinson, S. Xie, R.Y. Zhu
## Carnegie Mellon University, Pittsburgh, USA
V. Azzolini, A. Calamba, B. Carlson, T. Ferguson, Y. Iiyama, M. Paulini, J. Russ, H. Vogel, I. Vorobiev
## University of Colorado at Boulder, Boulder, USA
J.P. Cumalat, W.T. Ford, A. Gaz, E. Luiggi Lopez, U. Nauenberg, J.G. Smith, K. Stenson, K.A. Ulmer, S.R. Wagner
## Cornell University, Ithaca, USA
J. Alexander, A. Chatterjee, J. Chu, S. Dittmer, N. Eggert, N. Mirman, G. Nicolas Kaufman, J.R. Patterson, A. Ryd, E. Salvati, L. Skinnari, W. Sun, W.D. Teo, J. Thom, J. Thompson, J. Tucker, Y. Weng, L. Winstrom, P. Wittich
## Fairfield University, Fairfield, USA
D. Winn
## Fermi National Accelerator Laboratory, Batavia, USA
S. Abdullin, M. Albrow, J. Anderson, G. Apollinari, L.A.T. Bauerdick, A. Beretvas, J. Berryhill, P.C. Bhat, K. Burkett, J.N. Butler, H.W.K. Cheung, F. Chlebana, S. Cihangir, V.D. Elvira, I. Fisk, J. Freeman, Y. Gao, E. Gottschalk, L. Gray, D. Green, S. Gr¨nendahl, O. Gutsche, J. Hanlon, u D. Hare, R.M. Harris, J. Hirschauer, B. Hooberman, S. Jindariani, M. Johnson, U. Joshi, K. Kaadze, B. Klima, B. Kreis, S. Kwan, J. Linacre, D. Lincoln, R. Lipton, T. Liu, J. Lykken, K. Maeshima, J.M. Marraffino, V.I. Martinez Outschoorn, S. Maruyama, D. Mason, P. McBride, K. Mishra, S. Mrenna, Y. Musienko 28 , S. Nahn, C. Newman-Holmes, V. O'Dell, O. Prokofyev, E. Sexton-Kennedy, S. Sharma, A. Soha, W.J. Spalding, L. Spiegel, L. Taylor, S. Tkaczyk, N.V. Tran, L. Uplegger, E.W. Vaandering, R. Vidal, A. Whitbeck, J. Whitmore, F. Yang
## University of Florida, Gainesville, USA
D. Acosta, P. Avery, P. Bortignon, D. Bourilkov, M. Carver, T. Cheng, D. Curry, S. Das, M. De Gruttola, G.P. Di Giovanni, R.D. Field, M. Fisher, I.K. Furic, J. Hugon, J. Konigsberg, A. Korytov, T. Kypreos, J.F. Low, K. Matchev, P. Milenovic 47 , G. Mitselmakher, L. Muniz, A. Rinkevicius,
L. Shchutska, M. Snowball, J. Yelton, M. Zakaria
## Florida International University, Miami, USA
S. Hewamanage, S. Linn, P. Markowitz, G. Martinez, J.L. Rodriguez
## Florida State University, Tallahassee, USA
T. Adams, A. Askew, J. Bochenek, B. Diamond, J. Haas, S. Hagopian, V. Hagopian, K.F. Johnson, H. Prosper, V. Veeraraghavan, M. Weinberg
## Florida Institute of Technology, Melbourne, USA
M.M. Baarmand, M. Hohlmann, H. Kalakhety, F. Yumiceva
## University of Illinois at Chicago (UIC), Chicago, USA
M.R. Adams, L. Apanasevich, V.E. Bazterra, D. Berry, R.R. Betts, I. Bucinskaite, R. Cavanaugh, O. Evdokimov, L. Gauthier, C.E. Gerber, D.J. Hofman, S. Khalatyan, P. Kurt, D.H. Moon, C. O'Brien, C. Silkworth, P. Turner, N. Varelas
## The University of Iowa, Iowa City, USA
E.A. Albayrak 48 , B. Bilki 49 , W. Clarida, K. Dilsiz, F. Duru, M. Haytmyradov, J.-P . Merlo, H. Mermerkaya 50 , A. Mestvirishvili, A. Moeller, J. Nachtman, H. Ogul, Y. Onel, F. Ozok 48 , A. Penzo, R. Rahmat, S. Sen, P. Tan, E. Tiras, J. Wetzel, T. Yetkin 51 , K. Yi
## Johns Hopkins University, Baltimore, USA
B.A. Barnett, B. Blumenfeld, S. Bolognesi, D. Fehling, A.V. Gritsan, P. Maksimovic, C. Martin, M. Swartz
## The University of Kansas, Lawrence, USA
P. Baringer, A. Bean, G. Benelli, C. Bruner, R.P. Kenny III, M. Malek, M. Murray, D. Noonan, S. Sanders, J. Sekaric, R. Stringer, Q. Wang, J.S. Wood
## Kansas State University, Manhattan, USA
A.F. Barfuss, I. Chakaberia, A. Ivanov, S. Khalil, M. Makouski, Y. Maravin, L.K. Saini, S. Shrestha, N. Skhirtladze, I. Svintradze
## Lawrence Livermore National Laboratory, Livermore, USA
J. Gronberg, D. Lange, F. Rebassoo, D. Wright
## University of Maryland, College Park, USA
A. Baden, A. Belloni, B. Calvert, S.C. Eno, J.A. Gomez, N.J. Hadley, R.G. Kellogg, T. Kolberg, Y. Lu, M. Marionneau, A.C. Mignerey, K. Pedro, A. Skuja, M.B. Tonjes, S.C. Tonwar
## Massachusetts Institute of Technology, Cambridge, USA
A. Apyan, R. Barbieri, G. Bauer, W. Busza, I.A. Cali, M. Chan, L. Di Matteo, V. Dutta, G. Gomez Ceballos, M. Goncharov, D. Gulhan, M. Klute, Y.S. Lai, Y.-J. Lee, A. Levin, P.D. Luckey, T. Ma, C. Paus, D. Ralph, C. Roland, G. Roland, G.S.F. Stephans, F. St¨ckli, K. Sumorok, D. Velicanu, o J. Veverka, B. Wyslouch, M. Yang, M. Zanetti, V. Zhukova
## University of Minnesota, Minneapolis, USA
B. Dahmes, A. Gude, S.C. Kao, K. Klapoetke, Y. Kubota, J. Mans, N. Pastika, R. Rusack, A. Singovsky, N. Tambe, J. Turkewitz
## University of Mississippi, Oxford, USA
J.G. Acosta, S. Oliveros
## University of Nebraska-Lincoln, Lincoln, USA
E. Avdeeva, K. Bloom, S. Bose, D.R. Claes, A. Dominguez, R. Gonzalez Suarez, J. Keller, D. Knowlton, I. Kravchenko, J. Lazo-Flores, S. Malik, F. Meier, G.R. Snow
## State University of New York at Buffalo, Buffalo, USA
J. Dolen, A. Godshalk, I. Iashvili, A. Kharchilava, A. Kumar, S. Rappoccio
## Northeastern University, Boston, USA
G. Alverson, E. Barberis, D. Baumgartel, M. Chasco, J. Haley, A. Massironi, D.M. Morse, D. Nash, T. Orimoto, D. Trocino, R.-J. Wang, D. Wood, J. Zhang
## Northwestern University, Evanston, USA
K.A. Hahn, A. Kubik, N. Mucia, N. Odell, B. Pollack, A. Pozdnyakov, M. Schmitt, S. Stoynev,
K. Sung, M. Velasco, S. Won
## University of Notre Dame, Notre Dame, USA
A. Brinkerhoff, K.M. Chan, A. Drozdetskiy, M. Hildreth, C. Jessop, D.J. Karmgard, N. Kellams, K. Lannon, W. Luo, S. Lynch, N. Marinelli, T. Pearson, M. Planer, R. Ruchti, N. Valls, M. Wayne, M. Wolf, A. Woodard
## The Ohio State University, Columbus, USA
L. Antonelli, J. Brinson, B. Bylsma, L.S. Durkin, S. Flowers, C. Hill, R. Hughes, K. Kotov, T.Y. Ling, D. Puigh, M. Rodenburg, G. Smith, B.L. Winer, H. Wolfe, H.W. Wulsin
## Princeton University, Princeton, USA
O. Driga, P. Elmer, P. Hebda, A. Hunt, S.A. Koay, P. Lujan, D. Marlow, T. Medvedeva, M. Mooney, J. Olsen, P. Pirou´ e, X. Quan, H. Saka, D. Stickland 2 , C. Tully, J.S. Werner, A. Zuranski
## University of Puerto Rico, Mayaguez, USA
E. Brownson, H. Mendez, J.E. Ramirez Vargas
## Purdue University, West Lafayette, USA
V.E. Barnes, D. Benedetti, G. Bolla, D. Bortoletto, M. De Mattia, Z. Hu, M.K. Jha, M. Jones, K. Jung, M. Kress, N. Leonardo, D. Lopes Pegna, V. Maroussov, P. Merkel, D.H. Miller, N. Neumeister, B.C. Radburn-Smith, X. Shi, I. Shipsey, D. Silvers, A. Svyatkovskiy, F. Wang, W. Xie, L. Xu, H.D. Yoo, J. Zablocki, Y. Zheng
## Purdue University Calumet, Hammond, USA
N. Parashar, J. Stupak
## Rice University, Houston, USA
A. Adair, B. Akgun, K.M. Ecklund, F.J.M. Geurts, W. Li, B. Michlin, B.P. Padley, R. Redjimi,
J. Roberts, J. Zabel
## University of Rochester, Rochester, USA
B. Betchart, A. Bodek, R. Covarelli, P. de Barbaro, R. Demina, Y. Eshaq, T. Ferbel, A. GarciaBellido, P. Goldenzweig, J. Han, A. Harel, A. Khukhunaishvili, G. Petrillo, D. Vishnevskiy
## The Rockefeller University, New York, USA
R. Ciesielski, L. Demortier, K. Goulianos, G. Lungu, C. Mesropian
## Rutgers, The State University of New Jersey, Piscataway, USA
S. Arora, A. Barker, J.P. Chou, C. Contreras-Campana, E. Contreras-Campana, D. Duggan, D. Ferencek, Y. Gershtein, R. Gray, E. Halkiadakis, D. Hidas, S. Kaplan, A. Lath, S. Panwalkar, M. Park, R. Patel, S. Salur, S. Schnetzer, S. Somalwar, R. Stone, S. Thomas, P. Thomassen,
M. Walker
## University of Tennessee, Knoxville, USA
K. Rose, S. Spanier, A. York
## Texas A&M University, College Station, USA
O. Bouhali 52 , A. Castaneda Hernandez, R. Eusebi, W. Flanagan, J. Gilmore, T. Kamon 53 , V. Khotilovich, V. Krutelyov, R. Montalvo, I. Osipenkov, Y. Pakhotin, A. Perloff, J. Roe, A. Rose, A. Safonov, T. Sakuma, I. Suarez, A. Tatarinov
## Texas Tech University, Lubbock, USA
N. Akchurin, C. Cowden, J. Damgov, C. Dragoiu, P.R. Dudero, J. Faulkner, K. Kovitanggoon, S. Kunori, S.W. Lee, T. Libeiro, I. Volobouev
## Vanderbilt University, Nashville, USA
E. Appelt, A.G. Delannoy, S. Greene, A. Gurrola, W. Johns, C. Maguire, Y. Mao, A. Melo, M. Sharma, P. Sheldon, B. Snook, S. Tuo, J. Velkovska
## University of Virginia, Charlottesville, USA
M.W. Arenton, S. Boutle, B. Cox, B. Francis, J. Goodell, R. Hirosky, A. Ledovskoy, H. Li, C. Lin, C. Neu, J. Wood
## Wayne State University, Detroit, USA
C. Clarke, R. Harr, P.E. Karchin, C. Kottachchi Kankanamge Don, P. Lamichhane, J. Sturdy
## University of Wisconsin, Madison, USA
D.A. Belknap, D. Carlsmith, M. Cepeda, S. Dasu, L. Dodd, S. Duric, E. Friis, R. HallWilton, M. Herndon, A. Herv´ e, P. Klabbers, A. Lanaro, C. Lazaridis, A. Levine, R. Loveless, A. Mohapatra, I. Ojalvo, T. Perry, G.A. Pierro, G. Polese, I. Ross, T. Sarangi, A. Savin, W.H. Smith, D. Taylor, C. Vuosalo, N. Woods
- † : Deceased
- 1: Also at Vienna University of Technology, Vienna, Austria
- 2: Also at CERN, European Organization for Nuclear Research, Geneva, Switzerland
- 3: Also at Institut Pluridisciplinaire Hubert Curien, Universit´ e de Strasbourg, Universit´ e de Haute Alsace Mulhouse, CNRS/IN2P3, Strasbourg, France
- 4: Also at National Institute of Chemical Physics and Biophysics, Tallinn, Estonia
- 5: Also at Skobeltsyn Institute of Nuclear Physics, Lomonosov Moscow State University, Moscow, Russia
- 6: Also at Universidade Estadual de Campinas, Campinas, Brazil
- 7: Also at Laboratoire Leprince-Ringuet, Ecole Polytechnique, IN2P3-CNRS, Palaiseau, France
- 8: Also at Joint Institute for Nuclear Research, Dubna, Russia
- 9: Also at Suez University, Suez, Egypt
- 10: Also at British University in Egypt, Cairo, Egypt
- 11: Also at Fayoum University, El-Fayoum, Egypt
- 12: Now at Ain Shams University, Cairo, Egypt
- 13: Also at Universit´ e de Haute Alsace, Mulhouse, France
- 14: Also at Brandenburg University of Technology, Cottbus, Germany
- 15: Also at Institute of Nuclear Research ATOMKI, Debrecen, Hungary
- 16: Also at E¨tv¨s Lor´ and University, Budapest, Hungary o o
- 17: Also at University of Debrecen, Debrecen, Hungary
- 18: Also at University of Visva-Bharati, Santiniketan, India
- 19: Now at King Abdulaziz University, Jeddah, Saudi Arabia
- 20: Also at University of Ruhuna, Matara, Sri Lanka
- 21: Also at Isfahan University of Technology, Isfahan, Iran
- 22: Also at Sharif University of Technology, Tehran, Iran
- 23: Also at Plasma Physics Research Center, Science and Research Branch, Islamic Azad University, Tehran, Iran
- 24: Also at Universit` a degli Studi di Siena, Siena, Italy
- 25: Also at Centre National de la Recherche Scientifique (CNRS) - IN2P3, Paris, France
- 26: Also at Purdue University, West Lafayette, USA
- 27: Also at Universidad Michoacana de San Nicolas de Hidalgo, Morelia, Mexico
- 28: Also at Institute for Nuclear Research, Moscow, Russia
- 29: Also at St. Petersburg State Polytechnical University, St. Petersburg, Russia
- 30: Also at California Institute of Technology, Pasadena, USA
- 31: Also at Faculty of Physics, University of Belgrade, Belgrade, Serbia
- 32: Also at Facolt` a Ingegneria, Universit` di Roma, Roma, Italy a
- 33: Also at Scuola Normale e Sezione dell'INFN, Pisa, Italy
- 34: Also at University of Athens, Athens, Greece
- 35: Also at Paul Scherrer Institut, Villigen, Switzerland
- 36: Also at Institute for Theoretical and Experimental Physics, Moscow, Russia
- 37: Also at Albert Einstein Center for Fundamental Physics, Bern, Switzerland
- 38: Also at Gaziosmanpasa University, Tokat, Turkey
- 39: Also at Adiyaman University, Adiyaman, Turkey
- 40: Also at Cag University, Mersin, Turkey
- 41: Also at Izmir Institute of Technology, Izmir, Turkey
- 42: Also at Ozyegin University, Istanbul, Turkey
- 43: Also at Marmara University, Istanbul, Turkey
- 44: Also at Kafkas University, Kars, Turkey
- 45: Also at Rutherford Appleton Laboratory, Didcot, United Kingdom
- 46: Also at School of Physics and Astronomy, University of Southampton, Southampton, United Kingdom
- 47: Also at University of Belgrade, Faculty of Physics and Vinca Institute of Nuclear Sciences, Belgrade, Serbia
- 48: Also at Mimar Sinan University, Istanbul, Istanbul, Turkey
- 49: Also at Argonne National Laboratory, Argonne, USA
- 50: Also at Erzincan University, Erzincan, Turkey
- 51: Also at Yildiz Technical University, Istanbul, Turkey
- 52: Also at Texas A&M University at Qatar, Doha, Qatar
- 53: Also at Kyungpook National University, Daegu, Korea
|
10.1016/j.physletb.2014.10.063
|
[
"CMS Collaboration"
] |
2014-08-04T20:12:28+00:00
|
2014-11-10T19:48:48+00:00
|
[
"hep-ex"
] |
Search for pair production of third-generation scalar leptoquarks and top squarks in proton-proton collisions at sqrt(s) = 8 TeV
|
A search for pair production of third-generation scalar leptoquarks and
supersymmetric top quark partners, top squarks, in final states involving tau
leptons and bottom quarks is presented. The search uses events from a data
sample of proton-proton collisions corresponding to an integrated luminosity of
19.7 inverse femtobarns, collected with the CMS detector at the LHC with
sqrt(s) = 8 TeV. The number of observed events is found to be in agreement with
the expected standard model background. Third-generation scalar leptoquarks
with masses below 740 GeV are excluded at 95% confidence level, assuming a 100%
branching fraction for the leptoquark decay to a tau lepton and a bottom quark.
In addition, this mass limit applies directly to top squarks decaying via an
R-parity violating coupling lambda'[333]. The search also considers a similar
signature from top squarks undergoing a chargino-mediated decay involving the
R-parity violating coupling lambda'[3jk]. Each top squark decays to a tau
lepton, a bottom quark, and two light quarks. Top squarks in this model with
masses below 580 GeV are excluded at 95% confidence level. The constraint on
the leptoquark mass is the most stringent to date, and this is the first search
for top squarks decaying via lambda'[3jk].
|
1408.0807v5
|
## Polynomial size linear programs for problems in P
David Avis a, ∗ , David Bremner , Hans Raj Tiwary , Osamu Watanabe b c d a GERAD and School of Computer Science, McGill University and Graduate School of Informatics, Kyoto University b Faculty of Computer Science, University of New Brunswick c
KAM/ITI, Charles University d Department of Mathematical and Computing Sciences, Tokyo Institute of Technology
## Abstract
A perfect matching in an undirected graph G = ( V, E ) is a set of vertex disjoint edges from E that include all vertices in V . The perfect matching problem is to decide if G has such a matching. Recently Rothvoß proved the striking result that the Edmonds' matching polytope has exponential extension complexity. In this paper for each n = | V | we describe a polytope for the perfect matching problem that is different from Edmonds' polytope and define a weaker notion of extended formulation. We show that the new polytope has a weak extended formulation (WEF) Q of polynomial size. For each graph G with n vertices we can readily construct an objective function so that solving the resulting linear program over Q decides whether or not G has a perfect matching. With this construction, a straightforward O n ( 4 ) implementation of Edmonds' matching algorithm using O n ( 2 ) bits of space would yield a WEF Q with O n ( 6 log n ) inequalities and variables. The construction is uniform in the sense that, for each n , a single polytope is defined for the class of all graphs with n nodes. The method extends to solve polynomial time optimization problems, such as the weighted matching problem. In this case a logarithmic (in the weight of the optimum solution) number of optimizations are made over the constructed WEF.
The method described in the paper involves the construction of a compiler that converts an algorithm given in a prescribed pseudocode into a polytope. It can therefore be used to construct a polytope for any decision problem in P which can be solved by a well defined algorithm. Compared with earlier results of Dobkin-Lipton-Reiss and Valiant our method allows the construction of explicit linear programs directly from algorithms written for a standard register model, without intermediate transformations. We apply our results to obtain polynomial upper bounds on the non-negative rank of certain slack matrices related to membership testing of languages in P/poly .
Keywords: Polytopes, extended formulation, extension complexity, perfect matching, linear programming, non-negative rank
## 1. Introduction
A perfect matching in an undirected graph G = ( V, E ) is a set of vertex disjoint edges from E that include all vertices in V . We let n denote the number of vertices and assume n is even throughout the paper. The perfect matching problem is to determine if G contains a perfect matching and this can be decided in polynomial time by running Edmonds' algorithm [9]. As well as this combinatorial algorithm, Edmonds also introduced a related polytope [10] which we will call the Edmonds' polytope EP : n
$$\text{EP} _ { n } = \text{CH} \{ x \in \{$$
where CH is the convex hull operator.
For any S ⊆ V and edge ij ∈ E , we write that ij ∈ δ S ( ) whenever exactly one of the vertices i and j is
∗ Corresponding author
Email addresses: [email protected] (David Avis), [email protected] (David Bremner), [email protected] (Hans Raj Tiwary), [email protected] (Osamu Watanabe)
in S . Edmonds [10] proved that EP n has the following halfspace representation:
$$\begin{array} {$$
This description is exponential in size. Nevertheless, the perfect matching problem can be solved in polynomial time by solving a linear program (LP) over this polytope. Indeed, define an objective function c T x = ∑ 1 glyph[lessorequalslant] i<j glyph[lessorequalslant] n c ij x ij , where c ij = 1 if ij ∈ E and c ij = 0 otherwise. The LP is:
$$z ^ { * } = \max _ { \begin{array} { c c } c ^ {$$
It is easy to verify that if G has a perfect matching then z ∗ = n/ 2 otherwise z ∗ < n/ 2. Since the inequalities defining EP n can be separated in polynomial time, the LP can be solved in polynomial time [14].
Since the perfect matching problem is in P , it seemed possible that EP n could be written as the projection of a polytope with a polynomial size description. This is the topic of extension complexity (see, e.g., Fiorini et al. [12]). We recall the basic definitions here, referring the reader to [12] for further details.
An extended formulation (EF) of a polytope Q ⊆ R q is a linear system
$$E x + F y = g, \ y \geqslant 0$$
in variables ( x, y ) ∈ R q + r , where E,F are real matrices with q, r columns respectively, and g is a column vector, such that x ∈ Q if and only if there exists y such that (3) holds. The size of an EF is defined as the number of inequalities in the system.
An extension of the polytope Q is another polytope Q ′ ⊆ R e such that Q is the image of Q ′ under a linear map. We define the size of an extension Q ′ as the number of facets of Q ′ . Furthermore, we define the extension complexity of Q , denoted by xc ( Q , ) as the minimum size of any extension of Q.
Rothvoß [18] recently proved the surprising result that xc (EP n ) is exponential. Since extension complexity seemed a promising candidate to obtain computational models that separate problems in P from those that are NP -hard, this was a setback. A way of strengthening extension complexity to handle this problem was recently proposed by Avis and Tiwary [3].
Dobkin et al. [8] and Valiant [19] showed that linear programming is P -complete from which it follows that every problem in P has an LP-formulation of polynomial size. We will review this result in Section 3 giving Valiant's construction. This construction applies to P/poly , which is the class of all decision problems L solvable by a family C ,n n ≥ 1 of polynomial size Boolean circuits such that C n solves the restriction of L to inputs of length n . From these circuits it is straightforward to construct a family of LPs.
The main contribution of this paper is to give a direct method to produce polynomial size LPs from polynomial time algorithms, not circuits. Specifically we will construct LPs directly from a polynomial time algorithm expressed in pseudocode that solves a decision problem. Of course a trivial LP formulation can be obtained by first solving the decision problem for a given input and setting c = 1 if the answer is yes and c = 0 otherwise. Then solving the one dimensional LP: max cx, 0 glyph[lessorequalslant] x glyph[lessorequalslant] 1 solves the original problem. To avoid such trivial LPs we limit how much work can be done in constructing the objective function. One such limitation might be, for example, to insist that the objective function can be computed in linear time in terms of the input size. The objective functions we consider in this paper satisfy this condition.
For concreteness, we focus on an explicit construction of a polynomial size LP that can be used to solve the perfect matching problem. Firstly we describe another 'natural' polytope, PM n , for the perfect matching problem. Then we will introduce the notion of a weak extended formulation (WEF). Instead of requiring projection onto PM n we will simply require that LPs solved over the WEF solve the original problem. The objective function used is basically just a ± 1 encoding of the input graph. The approach used is quite general and can be applied to any problem in P for which an explicit algorithm is known. It extends to polynomial time solvable optimization problems also. However in this case a logarithmic (in the weight of the optimum solution) number of optimizations are made over the constructed WEF. Note that when an EF exists both the optimization and decision problems can be solved in a single LP optimization. Hence a WEF is weaker since a single LP solves only the decision problem. We discuss this further in Section 6.
The paper is organized as follows. In the next section we introduce a new polytope for the perfect matching problem and give some basic results about its facet structure. We define the notion of weak extended formulation and state the main theorem. In Section 3 we first give a simple example to illustrate the technique we use to build extended formulations from boolean circuits. Then we prove the main theorem of the paper. In Section 4 we generalize our method to algorithms given in pseudocode rather than as a circuit. We show how programs written in a simple pseudocode can be converted to WEFs. Our method is modeled on Sahni's proof of Cook's theorem given in [15]. Since our pseudocode is clearly strong enough to implement Edmonds' algorithm in polynomial time, our method gives a polynomial size WEF for the perfect matching problem. In Section 5 we use our main theorem to show that the non-negative rank of certain matrices is polynomially bounded above. Finally in Section 6 we give some concluding remarks including a discussion of applying this technique to polynomial time optimization problems such as the maximum weighted matching problem.
## 2. Polytopes for decision problems
## 2.1. Another perfect matching polytope
We use the notation 1 t to denote the t -dimensional vector of all ones, dropping the subscript when it is clear from the context. Let n be an even integer and let x be a binary vector of length ( n 2 ) . We let G x ( ) = ( V, E ) denote the graph with edge incidence vector given by x , let n be the number of its vertices and m = 1 T x the number of its edges. Furthermore, let w x = 1 if G x ( ) has a perfect matching and zero otherwise. We define the polytope PM n as:
$$\text{PM} _ { n } = \text{CH} \{ ( x, w _$$
PM n may be visualized by starting with a hypercube in dimension ( n 2 ) and embedding it in one higher dimension with extra coordinate w . For vertices of the cube corresponding to graphs with perfect matchings w = 1 else w = 0. It is easy to see that PM n has precisely 2 ( n 2 ) vertices. Edmonds' polytope EP n is closely related to PM n , in fact it defines one of its faces. Let
$$F _ { n } = \text{PM} _ { n } \cap \{ ( x, w ) \, \colon \, \mathbb { 1 } ^ { T } x + ( 1 - w ) n ^ { 2 } = \frac { n } { 2 } \, \}$$
Proposition 1. F n defines a face of PM n and in fact
$$F _ { n } = \{ ( x, 1 ) \, \colon x \in \mathrm E P _ { n } \}.$$
Proof. To show that F n is a face we show that the inequality
$$\mathbf 1 ^ { T } x + ( 1 - w ) n ^ { 2 } \geqslant \frac { n } { 2 }$$
is valid for PM n . It suffices to verify it for the extreme points ( x, w x ) given in (4). If w x = 0, (7) holds since 1 T x + n 2 > n 2 . Since the inequality is strict, none of these extreme points lie on F n . Otherwise w x = 1, x is the incidence vector of graph containing a perfect matching, so 1 T x glyph[greaterorequalslant] n/ 2. This shows that F n defines a face of PM . n
The vectors x with w x = 1 and 1 T x = n/ 2 are the incidence vectors of perfect matchings of K n and are precisely those used to define EP n in (1). Hence EP n lifted by adding the coordinate w = 1 is precisely F n .
For a given input graph G x (¯) = ( V, E ) we define the vector c = ( c ij ) as follows:
$$c _ { i j } = \begin{cases} 1 & \text{if $ij\in E$} \\ - 1 & \text{otherwise} \end{cases} \quad 1 \leq i < j \leq n$$
and let d be a constant such that 0 < d glyph[lessorequalslant] 1 / 2. We construct the LP:
$$z ^ { * } = \max _ { \begin{array} { c c } c ^ { T } x + d w \\ s. t. \end{array} } & & ( x, w ) \in \text{PM} _ { n }$$
For any positive integer n , by an n -cube we mean the n -dimensional hypercube whose vertices are the 2 n binary vectors of length n .
Proposition 2. For any edge incidence vector ¯ x ∈ { 0 1 , } ( n 2 ) let m = 1 T ¯ x . The optimum solution to (9) is unique, z ∗ = m + d if G x (¯) has a perfect matching, and z ∗ = m otherwise.
glyph[negationslash]
Proof. Let c be the vector defined by (8) and set m = 1 T ¯. x Note that c T ¯ = x m and that c T x glyph[lessorequalslant] m -1 for any other vertex x of the ( n 2 ) -cube. If G x (¯) has a perfect matching then ( x, w ) = (¯ x, 1) is a feasible solution to (9) with z = m + . Since d x = ¯, x c T x + dw glyph[lessorequalslant] m -1 + d and so (¯ 1) is the unique optimum solution. x,
If G x (¯) does not have a perfect matching then ( x, w ) = (¯ 0) is a feasible solution to (9) with x, z = m . Consider any other vertex x of the ( n 2 ) -cube. Then z = c T x + dw glyph[lessorequalslant] m -1 + 1 / 2 = m -1 / 2. It follows that z ∗ = m and is obtained by the unique solution (¯ x, 0).
## 2.2. Polytopes for decision problems and weak extended formulations
The basic ideas above can be extended to arbitrary polynomial time decision problems. Let X denote a polynomial time decision problem defined on binary input vectors x = ( x , ..., x 1 q ), and an additional bit w x , where w x = 1 if x results in a 'yes' answer and w x = 0 if x results in a 'no' answer. We define the polytope P as:
$$P = \text{CH} \{ ( x, w _ { x } ) \, \colon x \in \{ 0, 1 \} ^ { q } \}$$
For a given binary input vector ¯ we define the vector x c = ( c j ) by:
$$c _ { j } = \begin{cases} 1 & \text{if } \bar { x } _ { j } = 1 \\ - 1 & \text{if } \bar { x } _ { j } = 0 \end{cases} \quad 1 \leq i < j \leq n$$
and let d be a constant such that 0 < d glyph[lessorequalslant] 1 / 2. As before we construct an LP:
$$z ^ { * } = \max \quad & c ^ { T } x + d w \\ & s. t. \quad & ( x, w ) \in P$$
The following proposition can be proved in an identical way to Proposition 2.
Proposition 3. For any ¯ x ∈ { 0 1 , } q let m = 1 T ¯ x . The optimum solution to (12) is unique, z ∗ = m + d if ¯ x has a 'yes' answer and z ∗ = m otherwise.
Definition 1. Let Q be a polytope which is a subset of the ( q + )-cube with variables labeled t x , ..., x 1 q , y 1 , ..., y t . We say that Q has the x-0/1 property if each of the 2 q ways of assigning 0/1 to the variables x , ..., x 1 q uniquely extends to a vertex ( x, y ) of Q and, furthermore, y is 0/1 valued. Q may have additional fractional vertices.
In polyhedral terms, for every binary vector b ∈ R q , the intersection of Q with the hyperplanes x j = b j is a 0/1 vertex. We will show that we can solve a polynomial time decision problem X by replacing P in (9) by a polytope Q of polynomial size, while maintaining the same objective functions. We call Q a weak extended formulation as it does not necessarily project onto P .
## Definition 2. A polytope
$$Q = \{ ( x, w, s ) \, \colon x \in [ 0, 1 ] ^ { q }, w \in [ 0, 1 ], s \in [ 0, 1 ] ^ { r }, A x + b w + C s \leqslant h \}$$
is a weak extended formulation (WEF) of P if the following hold:
- · Q has the x -0/1 property.
- · For any vector ¯ x ∈ { 0 1 , } q let m = 1 T ¯, let x c be defined by (11) and let 0 < d glyph[lessorequalslant] 1 2. / If ¯ has a 'yes' x answer the optimum solution of the LP
$$z ^ { * } = \max \ \{ c ^ { T } x + d w \colon ( x, w, s ) \in Q \}$$
is unique and takes the value z ∗ = m + . Otherwise d z ∗ < m + . d
The first condition states that any vertex of Q that has 0/1 values for the x variables has 0/1 values for the other variables as well. The second condition connects Q to P . For a 0/1 valued vertex ( x, w, s ) of Q we have w = 1 if x encodes a 'yes' answer since z ∗ = m + . If d x encodes a 'no' answer then z ∗ < m + d and we must have w = 0. The purpose of the coefficient d is so that we can distinguish the two answers by simply observing the value of z ∗ .
In the sequel we will be concerned with small positive constants which we will assume are rational and in reduced form d = p/q , for positive integers p and q . The size of d is glyph[ceilingleft] log 2 p glyph[ceilingright] + log glyph[ceilingleft] 2 q glyph[ceilingright] .
In general Q will have fractional vertices and that is why the condition for 'no' answers differs from that given in Proposition 3 (an example is given below). However, for small enough d we can ensure that the LP optimum solution is unique in both cases and corresponds to that given in Proposition 3.
Proposition 4. Let Q be a WEF of P . There is a positive constant d 0 , whose size is polynomial in the size of Q , such that for all d , 0 < d < d 0 , the optimal solution of the LP defined in (13) is unique, z ∗ = m + d if ¯ x has a 'yes' answer and z ∗ = m if ¯ x has a 'no' answer.
glyph[negationslash]
glyph[negationslash]
Proof. The part of the proposition relating to the 'yes' answers is already covered by Definition 2. So consider any ¯ corresponding to a 'no' answer and its associated vector x c given by (11). Since Q has the 0/1 property ¯ x uniquely extends to a 0/1 valued vertex (¯ x, w, s ¯ ¯) of Q with ¯ = 0 as we saw above. So for any w d , 0 < d ≤ 1 / 2, we have c T ¯ + x dw ¯ = c T ¯ = x m . Consider any other vertex ( x, w, s ) of Q . Since x = ¯, we have x c T x < m . Let d x (¯) = min { m -c T x } taken over all vertices ( x, w, s ) with x = ¯. x This minimum exists, since the number of vertices of Q is finite, and is positive. Furthermore, c T x + (¯) d x w ≤ c T x + (¯) d x ≤ m . We define
$$d _ { 0 } = \min \{ d ( y ) \ \colon \ y \in \{ 0, 1 \} ^ { q } \text{ has a ``no` answer} \}.$$
glyph[negationslash]
Now choose any d , 0 < d < d 0 , and any vertex ( x, w, s ) of Q with x = ¯. If x w > 0 then c T x + dw < c T x + d w 0 ≤ c T x + d 0 ≤ m . Otherwise w = 0 and we have c T x + dw = c T x < m as noted above. Since the optimal solution for any LP can always be obtained at a vertex, it follows that the LP defined in (13) has a unique optimum solution z ∗ = m whenever ¯ corresponds to a 'no' answer. x
If follows from standard linear programming theory that each vertex ( x, w, s ) ∈ Q can be represented by a number of bits polynomial in the size of Q . Since c is a ± 1 vector and m ≤ n it follows that m -c T x can also be so represented. Now d 0 is the minimum of a finite number of such quantities so has size polynomial in the size of Q . The proposition follows.
This proof gives, at least in principle, a method of computing the constant d 0 from a list of the vertices of Q . An easier to compute bound can be obtained from the analysis of Khachian's ellipsoid method, see, for example, the text [14]. Following this text, let 〈 Q 〉 be the number of bits required to represent Q . Then the proof of their Lemma 3.1.33 implies that the differences m -c T x in the above proof are all at least 2 -〈 Q 〉 and so we could choose this value for d . This value will generally be much too small to be of practical use. However, if we are able to observe the value of w in the optimum solution of (13) then we may in fact set d = 0. In this case z ∗ = m for both answers and it follows from the 0/1 property that the optimum solution is unique and 0/1 valued. Since Q is a WEF of P the value of w in the optimum solution gives the correct answer. This is the preferred method in practice as it reduces the problem of floating point round off errors which may be caused by small values of d .
We saw in the proof of Proposition 4, the reason that d has to be small is to ensure that fractional vertices of Q do not lead to optimum solutions for ± 1 valued object vectors c . As an example, consider the perfect matching problem and the polytope PM n defined earlier. Suppose Q n is a WEF for PM . n It follows from Propositions 2 and 4 that we can determine whether an input graph G has a perfect matching by solving an LP over either PM n or Q n using the same objective function which is derived directly from the edge adjacency vector of G . As a very simple case, consider n = 2 giving PM 2 = CH (0 0) (1 1) . { , , , } A possible WEF with r = 0 is given by:
## Q 2 = CH (0 0 0) (1 1 1) (1 { , , , , , , / 4 1 1 , , / 2) }
Initially let d = 1 2. / When G x (¯) is an edge, m = 1, c 12 = 1 and z = c T x + dw obtains the same optimum solution of z ∗ = 3 2 = / m + d over both PM 2 and Q 2 . When G x (¯) is a non-edge, m = 0, c 12 = -1 and z = c T x + dw obtains the optimum solution of z ∗ = 0 = m over PM 2 and z ∗ = 1 4 / < 1 / 2 = m + d over Q 2 , at the fractional vertex (1/4,1,1/2). However, if 0 < d < 1 / 4 then z = c T x + dw is negative for the fractional vertex and obtains the unique optimum solution of z ∗ = 0 = m over Q 2 and also, of course, over PM 2 . We see that Q 2 projects onto a triangle in the ( x, w )-space, whereas PM 2 is a line segment.
In the next section we prove the following result:
Theorem 1. Every decision problem X in P/poly admits a weak extended formulation Q of polynomial size.
## 3. From Circuits to Polytopes
In order to show that Linear Programming is P -complete, Valiant [19] gave a construction to transform boolean circuits into a linear sized set of linear inequalities with the x -0/1 property (where x i are the variables corresponding to the inputs of the circuit); a similar construction was used by Yannakakis [20] in the context of the Hamiltonian Circuit problem. In this section we show that Valiant's construction implies Theorem 1. Valiant's point of view is slightly different from ours in that he explicitly fixes the values of the input variables before solving an LP-feasibility problem (as opposed to using different objective functions with a fixed set of inequalities). Showing that the result of this fixing is a 0 / 1-vertex is precisely our x -0/1 property.
We begin with a standard definition, see for example the text by Arora and Barak [1]:
Definition 3. A (boolean) circuit with q input bits x = ( x , x 1 2 , ..., x q ) is a directed acyclic graph in which each of its t nodes, called gates , is either an AND ( ∧ ) gate, an OR ( ∨ ) or a NOT ( ¬ ) gate. We label each gate by its output bit. One of these gates is designated as the output gate and gives output bit w . The size of a circuit is the number of gates it contains and its depth is the maximal length of a path from an input gate to the output gate.
For example, the circuit shown in Figure 1 can be used to compute whether or not a graph on 4 nodes has a perfect matching. The input is the binary edge-vector of the graph and the output is w = 1 if the graph has a matching (e.g. G 1 ) or w = 0 if it does not (e.g. G 2 ). If the graph has a perfect matching, exactly one of y 12 , y 13 or y 14 is one, defining the matching. For each gate we have labeled the output bit by a new variable. We will construct a polytope from the circuit by constructing a system of inequalities on the same variables.
x 1,2
x 3,4
x 1,3
x 2,4
x 1,4
x 2,3
w
Figure 1: A circuit to compute whether a 4 node graph has a perfect matching
From an AND gate, say y 12 = x 12 ∧ x 34 , we generate the inequalities:
$$x _ { 1 2 } & + x _ { 3 4 } - y _ { 1 2 } \ \leqslant \ \ 1 \\ - x _ { 1 2 } & \quad + y _ { 1 2 } \ \leqslant \ \ 0 \\ - x _ { 3 4 } & + y _ { 1 2 } \ \leqslant \ \ 0 \\ & \quad y _ { 1 2 } \ \geqslant \ \ 0$$
The system (14) defines a polytope in three variables whose 4 vertices represent the truth table for the AND gate:
| x 12 | x 34 | y 12 |
|--------|--------|--------|
| 0 | 0 | 0 |
| 0 | 1 | 0 |
| 1 | 0 | 0 |
| 1 | 1 | 1 |
Note that the variables x 12 , x 34 define a 2-cube and so the polytope is an extension of the 2-cube. In the terminology of the last section, it has the { x 12 , x 34 } -0/1 property.
From an OR gate, say s 3 = y 12 ∨ y 13 , we generate the inequalities:
$$\begin{matrix} - y _ { 1 2 } - y _ { 1 3 } + s _ { 3 } & \leq & 0 \\ y _ { 1 2 } & - s _ { 3 } & \leq & 0 \\ y _ { 1 3 } - s _ { 3 } & \leq & 0 \\ & s _ { 3 } & \leq & 1 \end{matrix}$$
The system (15) defines a polytope in three variables whose 4 vertices represent the truth table for the OR gate, as can easily be checked. Indeed, this polytope has the { y 12 , y 13 } -0/1 property.
From a NOT gate, say ¯ y 12 = ¬ y 12 , we could generate the equation
$$\bar { y } _ { 1 2 } & = 1 - y _ { 1 2 }$$
However it is equivalent to just replace all instances of ¯ y 12 by 1 -y 12 in the inequality system, and this is what we will do in the sequel.
The circuit in Figure 1 contains 5 AND gates and 2 OR gates. By suitably replacing variables in (14) and (15) we obtain a system of 28 inequalities in 13 variables. As just mentioned, the NOT gates are handled by variable substitution rather than explicit equations. Let Q 4 denote the corresponding polytope. It will follow by the general argument below that Q 4 is a weak extended formulation (WEF) of PM . 4
We now show that the above construction can be applied to any boolean circuit C to obtain a polytope Q which has the 0/1 property with respect to the inputs of C .
Lemma 1 ([19]) . Let C be a boolean circuit with q input bits x = ( x , x 1 2 , ..., x q ) , t gates labeled by their output bits y = ( y , y 1 2 , ..., y t ) and with circuit output bit w = y t . Construct the polytope Q with 4 t inequalities and q + t variables using the systems (14) and (15) respectively. Q has the the x -0/1 property and for every input x the value of w computed by C corresponds to the value of y t in the unique extension ( x, y ) ∈ Q of x .
Proof. Since C is an acyclic directed graph it contains a topological ordering of its nodes (gates) and we can assume that the labeling y , y 1 2 , ..., y t is such an ordering. Note we can assume w = y t comes last since it cannot be an input to any other gate. For any given input x the output of the circuit can be obtained by evaluating each gate in the order y , ..., y 1 t . Since it is a topological ordering, each input for a gate has been determined before the gate is evaluated.
We proceed by induction. Let Q k be the polytope defined by the 4 k inequalities corresponding to gates y , ..., y 1 k . The inductive hypothesis is that for k = 1 2 , , ..., t
- · Q k has the x -0/1 property, and
- · for each x the value of y k calculated by C corresponds to the value of y k in the unique extension of x in Q k .
This is clearly true for k = 1 as the analysis following (14) and (15) shows. We assume the hypothesis is true for k = 1 2 , , ..., j , where 1 glyph[lessorequalslant] j < t , and prove it for j +1. Indeed, since Q j has the x -0/1 property for each x the values of y , ..., y 1 j are uniquely defined and have 0/1 values. By induction they correspond to the values computed by C . Therefore the analysis following (14) and (15) shows that y j +1 will also be uniquely defined, 0/1 valued, and will correspond to the value computed by C . This verifies the inductive hypothesis for j +1 and since Q = Q t the proof is complete.
Lemma 2. Let C be a circuit that solves a decision problem X with q input bits x = ( x , x 1 2 , ..., x q ) and has associated polytope P as defined in (10). The polytope Q constructed in Lemma 1 is a WEF for P .
Proof. In order to make the correspondence with Definition 2 we relabel the variables in Q , constructed in Lemma 1, so that s = ( y , y 1 2 , ..., y t -1 ) and w = y t . By Lemma 1 we know Q has the x -0/1 property so it remains to prove the second condition in Definition 2.
Let ¯ be any vector in x { 0 1 , } q and set m = 1 T ¯. x Since Q has the x -0/1 property ¯ extends to a unique x binary vertex (¯ x, w, s ¯ ¯) of Q . Define c as in (11). Fix some d , 0 < d glyph[lessorequalslant] 1 / 2 and consider the optimum solution
$$z ^ { * } = \max \ \{ c ^ { T } x + d w \colon ( x, w, s ) \in Q \}.$$
Since Q has the x -0/1 property the maximum of c T x over Q is obtained at c T ¯ = x m at the unique vertex (¯ x, w, s ¯ ¯) of Q . For any other ( x, w, s ) ∈ Q , since x is in the q -cube and not equal to ¯, we have x c T x < m and, since w glyph[lessorequalslant] 1, z = c T x + dw < m + . Therefore, if ¯ has a 'yes' answer then d x w = 1, z ∗ = m + d and (¯ x, w, s ¯ ¯) is the unique optimum solution. Otherwise z ∗ < m + . The lemma follows. d
Theorem 1 follows from Lemmas 1 and 2. Since there is no limitation of uniformity on the circuits used, the theorem holds for all decision problems in P/poly . Since each gate in the circuit gives rise to 4 inequalities and one new variable, we have the following corollary.
Corollary 1. Let X be a decision problem with corresponding polytope P defined by (10). A set of circuits for X with size p n ( ) generates a WEF Q for P with at most 4 p n ( ) inequalities and variables.
In this section we showed how to construct a polynomial size LP from a polynomial size circuit so that the optimum solution of the LP gives the output of the circuit. However it is not immediately clear how to use this to obtain a polytope for the perfect matching problem. It would be required to convert Edmonds' algorithm to a family of circuits. In the next section we bypass this step by showing how to convert a simple pseudocode directly into a polytope without first computing a circuit (See Theorem 2). This can be used to convert polynomial time algorithms into polynomial size LPs directly.
We would like to remark that our construction in Theorem 2 of a WEF from a pseudocode may not be optimal. For example, it would be possible to get roughly O T n ( ( ) log T n ( ))-size circuits simulating a given T n ( )-time bounded Turing machine (see, e.g., Chapter 1 of [1]) from which we can construct a WEF with O T n ( ( ) log T n ( )) inequalities. But since Turing Machines are not commonly used for designing algorithms, we leave the interested reader to check whether a similar idea can be used to define a WEF with smaller size.
## 4. Constructing an LP from pseudocode
In this section we introduce a rudimentary pseudocode that can be used for decision problems. This pseudocode follows the usual practices of specifying algorithms and the tradition of so-called register machines (see e.g. [7]). We show how the code can be translated into a linear program, in a way similar to that shown for circuits in the previous section. This translation works for any pseudocode, but since the focus of this paper is on the class P , we will assume there is a polynomial function p n ( ) so that the the code terminates within p n ( ) steps for any input of size n . In this case we will show that the corresponding LP will also have polynomial size in n .
The pseudocode we use and its translation into an LP is adapted from a proof of Cook's theorem given in [15] which is attributed to Sartaj Sahni. In Sahni's construction the underlying algorithm may be nondeterministic, but we will consider only deterministic algorithms. Furthermore, Sahni describes how to convert his pseudocode into a satisfiability expression. Although it would be possible to convert this expression into an LP, considerable simplifications are obtained by doing a direct conversion from pseudocode to an LP. In this section, for simplicity, we describe only those features of the pseudocode that are necessary for implementing Edmonds' algorithm for the perfect matching problem. Additional features would be needed to handle more
sophisticated problems, such as the weighted matching problem. For full details, the reader is referred to Section 11.2 of [15].
Our pseudocode A has the following form. We assume W is a fixed integer which will represent the word size for integer variables.
- · Variables are binary valued except for indices, which are W -bit integers . Arrays of binary values are allowed and may be one or two dimensional. Dimension information is specified at the beginning of A . We let q n ( ) denote the maximum number of bits required to represent all variables for an input size of n . Sahni argues that q n ( ) = O p n ( ( )) however in our case q n ( ) is significantly smaller. Statements in A are numbered sequentially from 1 to l .
- · An expression contains at most one boolean operator or is the incrementation of an index. Array variables are not used in expressions but may be assigned to simple variables and vice versa.
- · Certain variables are designated as parameters and used to provide input to the program. All other variables are initially zero.
- · A may contain control statements go to k and if c = 1 then go to k endif . Here k is an instruction number and c is a simple binary variable.
- · A terminates by setting a binary variable w to one if the input results in a yes outcome and to zero otherwise. The program then halts.
In our implementation we also allow higher level commands such while and for loops which are first precompiled into the basic statements listed above. As a simple example, here is a pseudocode that produces essentially the same result as the circuit in Figure 1.
```
ally the same result as the circuit
y_12 = x_12 \wedge x_34
y_13 = 0
y_14 = 0
if y_12 then go to 50 endif
y_13 = x_13 \wedge x_24
if y_13 then go to 50 endif
y_14 = x_14 \wedge x_23
if y_14 then go to 50 endif
w = 0
return
50: w = 1
return
hat the lines of the pseudocode wl
```
Note that the lines of the pseudocode which are executed depend on the input values x . This is different from the circuit where all gates are executed for every input. We return to this point below.
The variables in the LP are denoted as follows. They correspond to variables in A as it is being executed on a specific input I.
- · Binary variables B i, t , ( ) 1 glyph[lessorequalslant] i glyph[lessorequalslant] q n , ( ) 0 glyph[lessorequalslant] t < p n ( ).
- B i, t ( ) represents the value of binary variable i in A after t steps of computation. For convenience we may group W consecutive bits together as an integer variable i . I ( i, j, t ) represents the value of the j -th bit of integer variable i in A after t steps of computation. The bits are numbered from right to left, the rightmost bit being numbered 1.
- · Binary arrays Abinary array R m , m [ ] = 0 1 , , ..., u is stored in consecutive binary variables B α ( + m,t , ) 0 glyph[lessorequalslant] m glyph[lessorequalslant] u, 0 glyph[lessorequalslant] t glyph[lessorequalslant] p n ( ) from some base location α . The array index m is stored as a W -bit integer I ( ∗ ∗ , , t ) and so we must have u glyph[lessorequalslant] 2 W -1.
- · 2-dimensional binary arrays A two dimensional binary array R m c [ ][ ], m = 0 1 , , ..., u , c = 0 1 , , ..., v is stored in row major order in consecutive binary variables B α [ + j, t -1], 0 glyph[lessorequalslant] j glyph[lessorequalslant] uv + u + v, 0 glyph[lessorequalslant] t glyph[lessorequalslant] p n ( ) from some base location α . The array indices m and c are stored as W -bit integers I ( ∗ ∗ , , t ) and so we must have u, v glyph[lessorequalslant] 2 W -1.
- · Step counter S i, t ( ) , 1 glyph[lessorequalslant] i glyph[lessorequalslant] l, 1 glyph[lessorequalslant] t glyph[lessorequalslant] p n ( ).
Variable S i, t ( ) represents the instruction to be executed at time t . It takes value 1 if line i of A is being executed at time t and 0 otherwise.
All of the above variables are specifically bounded to be between zero and one in our LP. The last set of variables, the step counter, indicates an essential difference between the circuit model and the pseudocode model. In the former model, all gates are executed for each possible input. The gates can be executed in any topological order consistent with the circuit. For the pseudocode model, however, the step to be executed at any time t will usually depend on the actual input. For each time step t and line i of pseudocode we will develop a system of inequalities which have the x -0/1 property, for some subset of variables x , if line i is executed at time t . I.e., the inequalities should uniquely determine a 0/1 value of all variables given any 0/1 setting of the x variables. However, if step i is not executed at time t then the variables should be free to hold any 0/1 values and these values will be determined by the step that is executed at time t . So in each set of inequalities a control variable (in our case the variable S i, t ( )) will appear for this purpose. More formally, we make the following definition which generalizes Definition 1:
Definition 4. Let Cx + Dy glyph[lessorequalslant] e be a system of inequalities that satisfy the x -0/1 property, i.e. each 0/1 setting of the x variables uniquely defines a 0/1 setting of the y variables. Suppose that Cx + Dy glyph[lessorequalslant] e + 1 is feasible for all 0/1 settings of x and y variables, and let z be a binary variable. The system 1 z + Cx + Dy glyph[lessorequalslant] e + 1 has the ( z ) controlled x-0/1 property .
Note that if z = 0 the new system is always feasible for any 0/1 setting of x and y . If z = 1 then the new system reduces to the old system that has the x -0/1 property.
We now define the 5 different types of linear inequalities needed to simulate the pseudocode which, following Sahni, we label C,D,E,F and G. 1 Recall that the S i, t ( )-variables ensure that at each time t a unique line i is executed, taking the value S i, t ( ) = 1 if it is and 0 otherwise. The inequalities listed below all have the S i, t ( ) controlled x -0/1 property, and so have the form S i, t ( ) + Cx + Dy glyph[lessorequalslant] e +1 for suitably chosen C, D, e, x, y .
- C: (Variable initialization) The variables B i, ( 0) , I ( i, j, 0) 1 , glyph[lessorequalslant] i glyph[lessorequalslant] q n , ( ) 1 glyph[lessorequalslant] j glyph[lessorequalslant] W are set equal to their initial value, if any, else set to zero.
- D: (Step counter initialization) Instruction 1 is executed at time t = 1.
$$S ( 1, 1 ) = 1$$
- E: (Unique step execution) A unique instruction is executed at each time t .
$$\sum _ { i = 1 } ^ { l } S ( i, t ) = 1, \quad 1 \leqslant t \leqslant p ( n )$$
- F: (Flow control) The instruction to execute at time t + 1 is determined, assuming we are at line i of A at time t , i.e. S i, t ( ) = 1. If not, i.e. S i, t ( ) = 0, then all inequalities below are trivially satisfied. This follows since the other variables are constrained to be between zero and one. There are 4 subcases depending on the instruction at line i . Inequalities are generated for each t , 1 glyph[lessorequalslant] t glyph[lessorequalslant] p n ( ).
- (i) (assignment statement) Go to the next instruction.
$$S ( i, t ) - S ( i + 1, t + 1 ) \leqslant 0$$
- (ii) ( go to k )
$$S ( i, t ) - S ( k, t + 1 ) \leqslant 0$$
- (iii) ( return ) Loop on this line until time runs out.
$$S ( i, t ) - S ( i, t + 1 ) \leqslant 0$$
1 Sahni also has a constraint set H which relates to the certificate checking function of his algorithm, and is not needed here.
- (iv) ( if c = 1 then go to k endif ) We assume that bit c is represented by variable B j, t ( -1).
$$S ( i, t ) + B ( j, t - 1 ) - S ( k, t + 1 ) \ \ & \leqslant \ \ 1 \\ S ( i, t ) - B ( j, t - 1 ) - S ( i + 1, t + 1 ) \ \ & \leqslant \ \ 0$$
When S i, t ( ) = 1 cases (i)-(iii) fix the next line to be executed and trivially have the controlled x -0/1 property, where x is empty. For (iv), note we have also the equations E above. When S i, t ( ) = 1, if B j, t ( -1) = 1 then the first inequality fixes S k, t ( + 1) = 1 otherwise the second inequality fixes S i ( +1 , t +1) = 1. The inequalities (iv) have the controlled B j, t ( -1)-0/1 property.
- G: (Control of variables) If we are at line i of A at time t , i.e. S i, t ( ) = 1, all variables are updated to their correct values at time t + 1 following the execution of line i . If not, i.e. S i, t ( ) = 0, then all inequalities below are trivially satisfied. Again there are several cases depending on the instruction at line i . Inequalities are generated for each t , 1 glyph[lessorequalslant] t glyph[lessorequalslant] p n ( ).
- (i) (Reassignment of unchanged variables) All variables left unchanged at a given step t need to be reassigned their previous values. Let k index some bit unchanged at step t .
$$S ( i, t ) + B ( k, t - 1 ) - B ( k, t ) \ \ & \leqslant \ \ 1 \\ S ( i, t ) - B ( k, t - 1 ) + B ( k, t ) \ \ & \leqslant \ \ 1$$
Note that when S i, t ( ) = 1 these inequalities imply that B k,t ( -1) = B k,t ( ). They have the controlled B k,t ( -1)-0/1 property. Similar inequalities are generated for each integer variable I ( k, j, t ) , 1 glyph[lessorequalslant] j glyph[lessorequalslant] W .
In what follows, the above inequalities need to be generated for all variables B k,t ( ) and I ( k, j, t ) not being assigned values at time t in the particular instruction i being considered.
- (ii) ( assignment: s = x and s = ¬ x ) Assume that x, s are stored in B q, t ( -1) , B ( s, t ) respectively. For s = x we generate the two inequalities:
$$S ( i, t ) + B ( q, t - 1 ) - B ( s, t ) \ \ & \leqslant \ 1 \\ S ( i, t ) - B ( q, t - 1 ) + B ( s, t ) \ \ & \leqslant \ 1$$
When S i, t ( ) = 1 the inequalities imply B s, t ( ) = B q, t ( -1) as desired. They have the controlled B q, t ( -1)-property. For s = ¬ x we generate the two inequalities:
$$S ( i, t ) + B ( q, t - 1 ) + B ( s, t ) \ \ & \leqslant \ 2 \\ S ( i, t ) - B ( q, t - 1 ) - B ( s, t ) \ \ & \leqslant \ 0$$
The analysis is similar to that for s = x .
- (iii) ( assignment: s = x ⊕ y ) Assume that x, y, s are stored in B q, t ( -1) , B ( r, t -1) , B ( s, t ) respectively.
$$& S ( i, t ) + B ( q, t - 1 ) - B ( r, t - 1 ) - B ( s, t ) \quad \leqslant \quad 1 \\ & S ( i, t ) - B ( q, t - 1 ) - B ( r, t - 1 ) + B ( s, t ) \quad \leqslant \quad 1 \\ & S ( i, t ) - B ( q, t - 1 ) + B ( r, t - 1 ) - B ( s, t ) \quad \leqslant \quad 1 \\ & \quad \sim \quad \sim \quad \sim \quad \sim \quad \sim \quad \sim \quad \sim \quad \,.$$
$$S ( i, t ) + B ( q, t - 1 ) + B ( r, t - 1 ) + B ( s, t ) \ \leqslant \ 3$$
If S i, t ( ) = 1 then all constants on the right hand side are reduced by one and S i, t ( ) can be deleted. It is easy to check the inequalities have the controlled { B q, t ( -1) , B ( r, t -1) } -0/1 property, and that for each such 0/1 assignment B s, t ( ) is correctly set.
- (iv) ( assignment: s = x ∧ y ) Assume that x, y, s are stored in B q, t ( -1) , B ( r, t -1) , B ( s, t ) respectively.
$$S ( i, t ) - B ( q, t - 1 ) & + B ( s, t ) & \leqslant & \ 1 \\ S ( i, t ) & - B ( r, t - 1 ) + B ( s, t ) & \leqslant & \ 1 \\ S ( i, t ) + B ( q, t - 1 ) + B ( r, t - 1 ) - B ( s, t ) & \leqslant & \ 2$$
If S i, t ( ) = 1 then all constants on the right hand side are reduced by one and S i, t ( ) can be deleted. It is easy to check the inequalities have the controlled { B q, t ( -1) , B ( r, t -1) } -0/1 property, and that for each such 0/1 assignment B s, t ( ) is correctly set.
- (v) ( assignment: s = x ∨ y , s = x 1 ∨ x 2 ∨ ... ∨ x k ) Assume that x, y, s are stored in B q, t ( -1) , B ( r, t -1) , B ( s, t ) respectively.
$$S ( i, t ) + B ( q, t - 1 ) & - B ( s, t ) & \leqslant & 1 \\ S ( i, t ) & + B ( r, t - 1 ) - B ( s, t ) & \leqslant & 1 \\ S ( i, t ) - B ( q, t - 1 ) - B ( r, t - 1 ) + B ( s, t ) & \leqslant & 1$$
The analysis is similar to G(iv) and is omitted. The inequalities have the controlled { B(q,t-1),B(r,t1) } -0/1 property.
The k -way or is an easy generalization which will be needed in the sequel, where we assume that x j is stored in B q , t ( j -1) , j = 1 2 , , ..., k . It is defined by the following inequalities:
$$S ( i, t ) + B ( q _ { j }, t - 1 ) - B ( s, t ) & \ \leqslant \ \ 1 \quad 1 \leqslant j \leqslant k \\ S ( i, t ) - \sum _ { j = 1 } ^ { k } B ( q _ { j }, t - 1 ) + B ( s, t ) & \ \leqslant \ \ 1.$$
- (vi) ( increment integer variable ) Assume that the integer variable is stored in I ( q, j, t -1) , 1 glyph[lessorequalslant] j glyph[lessorequalslant] W and is to be incremented by 1. We require another integer I ( r, j, t ) , 1 glyph[lessorequalslant] j glyph[lessorequalslant] W to hold the binary carries. On overflow, I ( r, W, t ) = 1 and I ( q, j, t ) = 0 1 , glyph[lessorequalslant] j glyph[lessorequalslant] W . The incrementer makes use of two previous operations, G(iii) and G(iv):
$$I ( q, 1, t ) \ & = \ \ I ( q, 1, t - 1 ) \oplus 1 \\ I ( r, 1, t ) \ & = \ \ I ( q, 1, t - 1 ) \wedge 1 \\ I ( q, j, t ) \ & = \ \ I ( q, j, t - 1 ) \oplus I ( r, j - 1, t ) \quad 2 \leqslant j \leqslant W \\ I ( r, j, t ) \ & = \ \ I ( q, j, t - 1 ) \wedge I ( r, j - 1, t ) \quad 2 \leqslant j \leqslant W$$
By appropriate formal substitution of variables, each of the above assignments is transformed into inequalities of the form G(iii) and G(iv), which are controlled by the step counter S i, t ( ). It can be verified that the full system satisfies the controlled { I ( q, j, t ) , 1 glyph[lessorequalslant] j glyph[lessorequalslant] W } -0/1 property because for each 0/1 setting of these variables all other variables are fixed by the above system of equations.
- (vii) ( equality test for integer variables ) Assume that the integer variables are stored in I ( q, j, t -1) and I ( r, j, t -1), 1 glyph[lessorequalslant] j glyph[lessorequalslant] W . We require W +1 temporary variables w.l.o.g. B j, t , ( ) 1 glyph[lessorequalslant] j glyph[lessorequalslant] W +1. If the two integer variables are equal then B W ( +1 , t ) is set to one else it is set to zero.
$$B ( j, t ) \ & = \ \begin{array} { c c } I ( q, j, t - 1 ) \oplus I ( r, j, t - 1 ) \quad 1 \leqslant j \leqslant W \\ B ( W + 1, t ) \ & = \ \end{array} \ \begin{array} { c c } W \\ j = 1 \end{array} \ \begin{array} { c c } 1 \leqslant j \leqslant W \\ B ( j, t ) \end{array}$$
The first equations makes repeated use of G(iii) after appropriate substitution. By combining G(ii) and the k -way or from G v ( ) we may implement the second equation by the inequalities.
$$S ( i, t ) + B ( j, t ) + B ( W + 1, t ) \ \leqslant \ \ 2 \quad 1 \leqslant j \leqslant W \\ S ( i, t ) - \sum _ { j = 1 } ^ { k } B ( j, t ) - B ( W + 1, t ) \ \leqslant \ \ 0.$$
The inequalities have the controlled { I ( q, j, t -1) , I ( r, j, t -1), 1 glyph[lessorequalslant] j glyph[lessorequalslant] W } -0/1 property.
- (viii) ( array assignment ) R m [ ] = x (and x = R m [ ]) We assume that R has dimension u , is stored in B α [ + j, t -1], 0 glyph[lessorequalslant] j glyph[lessorequalslant] u and that x is stored in B x,t ( -1). We further assume that m is stored in an integer variable I ( m,k,t -1) , 1 glyph[lessorequalslant] k glyph[lessorequalslant] W . We need additional binary variables M j,t , ( ) 0 glyph[lessorequalslant] j glyph[lessorequalslant] u to hold intermediate results. Initially we write down some equations and then we use previous results to convert these to inequalities. Firstly we need to discover the memory location for R m [ ]. For any 0 glyph[lessorequalslant] j glyph[lessorequalslant] u let j W W j -1 ...j 1 be the binary representation of j . Then we formally define for k = 1 2 , , ..., W
$$T _ { j } ( m, k, t - 1 ) = \begin{cases} \ I ( m, k, t - 1 ) & \ j _ { k } = 0 \\ 1 - I ( m, k, t - 1 ) & \ j _ { k } = 1 \end{cases}$$
Note this definition is purely formal and has nothing to do with the execution of A . We will assign M j,t ( ) a value via the W -way or given in G(v). For 0 glyph[lessorequalslant] j glyph[lessorequalslant] u :
$$S ( i, t ) + T _ { j } ( m, k, t - 1 ) - M ( j, t ) \ \leqslant \ 1 \quad \ 1 \leqslant k \leqslant W \quad \quad ( 1 7 ) \\ S ( i, t ) - \sum _ { k = 1 } ^ { W } T _ { j } ( m, k, t - 1 ) + M ( j, t ) \ \leqslant \ 1$$
When S i, t ( ) = 1, it can be verified that M j,t ( ) = 0 whenever j = m and is one otherwise. Now we may update all array elements of R at time t and make the assignment R m [ ] = x by the system of inequalities, for all 0 glyph[lessorequalslant] j glyph[lessorequalslant] u :
$$S ( i, t ) + B ( x, t - 1 ) - B ( \alpha + j$$
glyph[negationslash]
To understand these inequalities, first note that they are trivially satisfied unless S i, t ( ) = 1. When j = k we have M j,t ( ) = 0 and the first two inequalities are tight. We have B α ( + j, t ) = B x,t ( -1) updating the array element to x . The second two inequalities are trivially satisfied. Otherwise j = k , M j,t ( ) = 1, the first two inequalities are trivially satisfied and the second two are tight. We have B α ( + j, t ) = B α ( + j, t -1) copying the array element over to time t from time t -1. We remark that there are O uW ( ) inequalities generated above.
Finally note that we can implement x = R m [ ] by using the inequalities
$$S ( i, t ) + B ( x, t ) - B ( \alpha + j, t$$
and letting the array R m [ ] be copied at time t using G(i). Both of these two inequality systems have the controlled { B x,t ( -1) , B ( α + j, t -1) , j = 0 , ..., u } -0/1 property.
- (ix) ( 2-dimensional array assignment ) R m c [ ][ ] = x (or x = R m c [ ][ ]). This is a natural generalization of G(viii). We assume that R has dimensions u and v , is stored in row major order in B α [ + j, t -1], 0 glyph[lessorequalslant] j glyph[lessorequalslant] uv + u + v and that x is stored in B x,t ( -1). We further assume that m and c are stored in an integer variables I ( m,k,t -1) , I ( c, k, t -1) , 1 glyph[lessorequalslant] k glyph[lessorequalslant] W respectively. We need additional binary variables M j,t , ( ) 0 glyph[lessorequalslant] j glyph[lessorequalslant] u and N j, t , ( ) 0 glyph[lessorequalslant] j glyph[lessorequalslant] v to hold intermediate results. Firstly we need to discover the memory location for R m c [ ][ ]. We again use the equations (17) for the row index.
For the column index, as in G(viii), for any 0 glyph[lessorequalslant] j glyph[lessorequalslant] v let j W W j -1 ...j 1 be the binary representation of j . We formally define for k = 1 2 , , ..., W
$$T _ { j } ( c, k, t - 1 ) = \begin{cases} \ I$$
We will assign N j, t ( ) a value via the W -way or given in G(v). For 0 glyph[lessorequalslant] j glyph[lessorequalslant] v :
$$S ( i, t ) + T _ { j } ( c, k, t - 1 ) -$$
When S i, t ( ) = 1, it can be verified that N j, t ( ) = 0 whenever j = c and is one otherwise. Now we may update all array elements of R at time t and make the assignment R m c [ ][ ] = x by the following
system of inequalities. For all 0 glyph[lessorequalslant] j 1 glyph[lessorequalslant] u , 0 glyph[lessorequalslant] j 2 glyph[lessorequalslant] v , r = j 1 ( u +1) + j 2 :
```
\begin{array} {
```
The analysis is similar to G(viii). The above inequalities are all trivial unless S i, t ( ) = 1. Note that for each j 1 and j 2 , index r gives the relative location in the array. If j 1 = m j , 2 = c then M j ,t ( 1 ) = N j , t ( 2 ) = 0, the first two inequalities are tight and the last four loose. The first two inequalities give B α ( + r, t ) = B x,t ( -1). Otherwise either M j ,t ( 1 ) = 1 or N j , t ( 2 ) = 1 or both, and the first two inequalities are trivially satisfied. In the former case the two middle inequalities are tight and we have the equation B α ( + r, t ) = B α ( + r, t -1). In the latter case this equation is formed from the last two inequalities. We remark that there are O uv ( + uW + vW ) inequalities generated above.
For the assignment x = R m c [ ][ ] we need the inequalities
$$S ( i, t ) + B ( x, t ) - B ( \alpha + r, t - 1 ) - M ( j _ { 1 }, t ) - N ( j _ { 2 }, t ) \ \ & \leqslant \ 1 \\ S ( i, t ) - B ( x, t ) + B ( \alpha + r, t - 1 ) - M ( j _ { 1 }, t ) - N ( j _ { 2 }, t ) \ \ & \leqslant \ 1$$
for r = j 1 ( u +1) + j 2 . All array elements of R must also be copied from time t -1 to time t as in G(i).
Both of these two inequality systems have the controlled { B x,t ( -1) , B ( α + j, t ) , j = 0 , ..., uv + + u v } -0/1 property.
Remark: In applications using graphs, a symmetric 2-dimensional array is often used to hold the adjacency matrix. Such symmetric matrices may be implemented in pseudocode by replacing a statement such as R m c [ ][ ] = x by the two statements R m c [ ][ ] = x and R c [ ][ m ] = x . Assignment statements such as x = R m c [ ][ ] may be left as is.
To show the correctness of the above procedure we give two lemmas that are analogous to Lemmas 1 and 2 of the last section. First we show that the above construction can be applied to any pseudocode A , written in the language described, to produce a polytope Q which has the 0/1 property with respect to the inputs of A .
Lemma 3. Let A be a pseudocode, written in the above language, which takes n input bits x = ( x , x 1 2 , ..., x n ) , and terminates by setting a bit w . Construct the polytope Q as described above relabeling B (0 , p ( n )) as w and the additional variables as s = ( s , s 1 2 , ..., s N ) for some integer N . Q has the x -0/1 property and for every input x the value of w computed by A corresponds to the value of w in the unique extension ( x, w, s ) ∈ Q of x .
Proof. (Sketch) As with Lemma 1 the proof is by induction, but this time we use the step counter. By assumption A terminates after p n ( ) steps. Let k = 1 2 , , ..., p ( n ) represent the step counter. Define Q k to be the polytope consisting of precisely those inequalities in Q that use variables: B i, t , ( ) 1 glyph[lessorequalslant] i glyph[lessorequalslant] q n , ( ) 1 glyph[lessorequalslant] t glyph[lessorequalslant] k , I ( i, j, t ) , 1 glyph[lessorequalslant] i glyph[lessorequalslant] q n , ( ) 1 glyph[lessorequalslant] j glyph[lessorequalslant] W, 1 glyph[lessorequalslant] t glyph[lessorequalslant] k and S i, t ( ) , 1 glyph[lessorequalslant] i glyph[lessorequalslant] l, 1 glyph[lessorequalslant] t glyph[lessorequalslant] k .
- · Q k has the x -0/1 property, and
- · for each x , at step k , A with input x is executing line i corresponding to the unique index i where S i, k ( ) = 1 and all variables at that step have the values corresponding to the values of B i, k ( ) , 1 glyph[lessorequalslant] i glyph[lessorequalslant] q n ( ) and I ( i, j, k ) , 1 glyph[lessorequalslant] i glyph[lessorequalslant] q n , ( ) 1 glyph[lessorequalslant] j glyph[lessorequalslant] W .
The inequalities of Q 1 consist of those in groups C, D, and part of E above and the induction hypothesis is readily verified. We assume the hypothesis is true for k = 1 2 , , ..., T , where 1 glyph[lessorequalslant] T < p n ( ), and prove it for T +1. Indeed, since Q k has the x -0/1 property for each x the values of all variables with index t glyph[lessorequalslant] T have been correctly set. It follows that for precisely one index i we have S i, T ( ) = 1, meaning that line i of the pseudocode is executed at time T for this particular input. The inequalities defined in group G all have the
controlled x -0/1 property for the control variable S i, T ( ). The variables B and I with index t = T + 1 are correctly set by the analysis in group G above. The analysis in group F implies that the values of S i, T ( +1) will also be uniquely determined and 0/1, correctly indicating the next line of A to be executed at t = T +1. This verifies the inductive hypothesis for T +1 and since Q = Q p n ( ) this concludes the proof.
The next lemma is simply a restatement of Lemma 2 in the context of our pseudocode rather than circuits. The proof of Lemma 2 makes no reference to how Q was computed, so the same proof holds.
Lemma 4. Let A be a algorithm, written in the above pseudocode, which solves a decision problem X with n input bits x = ( x , x 1 2 , ..., x n ) and has associated polytope P as defined in (10). The polytope Q described in Lemma 3 is a WEF for P .
Lemmas 3 and 4 justify the correctness of the method outlined in this section.
We now analyze the size of the WEF Q created. Recall that q n ( ) is the number of bits of storage required by the algorithm A, which of course consists of a constant number of lines of pseudocode. The variables of Q are the variables B j, t ( ), S i, t ( ) and additional temporary variables created in some of the groups C-G. It can be verified that their number is O p n q n ( ( ) ( )). For fixed t , each of the sets of inequalities described in groups C-G have size at most O q n ( ( )) except possibly the array assignment inequalities described in G(viii) and G(ix). As remarked there, an array of dimension u generates O uW ( ) inequalities. A 2-dimensional array of dimension r by c generates O rc ( + rW + cW ) inequalities. We may assume that W ∈ O (log n ). Then O q n ( ( ) log n ) is an upper bound the number of inequalities generated in either G(viii) or G(ix). Since t is bounded by p n ( ) we see that the WEF has at most O p n ( ( )( q n ( ) log n )) inequalities also. We have:
Theorem 2. Let X be a decision problem with corresponding polytope P defined by (10). An algorithm for X written in the pseudocode described above requiring q n ( ) space and terminating after p n ( ) steps generates a WEF Q for P with O p n q n ( ( ) ( ) log n ) inequalities and variables.
Since Edmonds' algorithm can be implemented in polynomial time in the pseudocode presented our method gives a polynomial size WEF for PM n . So for example, a straightforward O n ( 4 ) implementation of Edmonds' algorithm using O n ( 2 ) space would yield a WEF with O n ( 6 log n ) inequalities and variables. If the O n ( 2 5 . ) time algorithm of Even and Kariv [11] can be implemented in our pseudocode it would yield a considerably smaller polytope.
Sparktope is a compiler built along the lines described above to automatically generate a WEF corresponding to any given pseudocode [2]. The elements described in C-G above have been implemented and tested as well as some complete examples of pseudocode. The polytopes generated are rather large even for short pseudocodes. For example, the pseudocode at the beginning of this section generated a polytope with about 3200 inequalities! This should be compared with 28 inequalities for the circuit in Figure 1 and 4 odd set inequalities for Edmonds' polytope EP . Nevertheless the WEF generated by this method should be significantly smaller than EP 4 n even for relatively small n . The details of the implementation of the compiler and its application to Edmonds' algorithm will be described in a subsequent paper.
## 5. Connections to non-negative rank
In this section we reformulate the results in previous sections in terms of non-negative ranks of certain matrices. Non-negative rank is a very useful tool that captures the extension complexity of polytopes [20]. We will prove a necessary condition for membership testing in a language to be in P/poly based on the existence of a certain matrix with polynomially bounded non-negative rank. This characterization potentially opens the door to proving that a given problem does not belong to P/poly by demonstrating high non-negative rank of the associated matrix. Using a slightly stronger definition of non-negative rank we are also able to give a sufficient condition for membership testing in a language to be in P/poly .
A matrix S is called non-negative if all its entries are non-negative. The non-negative rank of a non-negative matrix S , denoted by rank + ( S ), is the smallest number r such that there exist non-negative matrices T and U such that T has r columns, U has r rows and S = TU . If we require the left factor T in the above definition to only contain numbers that can be encoded using a number of bits only polynomial in the number of bits required to encode any entry of S then the smallest such r is called the succinct non-negative rank of S . To see the usefulness of this apparently asymmetric restriction on the factors, note that when S is the slack matrix of a polytope Ax glyph[greaterorequalslant] b then such a factorization allows one to describe an extended formulation for the polytope using only the entries of A,T and b . So if T is required to be polynomial in the bit complexity of the entries
of S then one can represent the polytope as the shadow of another polytope that can be encoded using a polynomial number of bits.
## 5.1. Polytopal sandwiches
Let P in and P out be two polytopes in R k such that P in ⊆ P out . We say that such a pair defines a polytopal sandwich . With every polytopal sandwich we can associate a non-negative matrix which encodes the slack of the inequalities defining P out with respect to the vertices of P in . That is, if P in = conv( { v , . . . , v 1 n } ) and P out = { x ∈ R k : a T i x glyph[lessorequalslant] b , i 1 glyph[lessorequalslant] i glyph[lessorequalslant] m } then the slack matrix associated with the polytopal sandwich thus defined is S P ( out , P in ) = S with S ij = b i -a T i v . j When P in and P out define the same polytope P we denote the corresponding slack matrix simply as S P . ( ) The next lemma shows the relation between the non-negative rank of the slack of a polytopal sandwich and the smallest size polytope whose shadow fits in the sandwich. We will assume that the polytopes defining our sandwiches are full-dimensional. The same lemma appears in [6] and has roots in [13, 16]. However the proof is simple enough to attribute it to folklore.
Lemma 5. Let P in = conv( { v , . . . , v 1 n } ) and P out = { x ∈ R k : a T i x glyph[lessorequalslant] b , i 1 glyph[lessorequalslant] i glyph[lessorequalslant] m . } Let P min be a polytope with smallest extension complexity such that P in ⊆ P min ⊆ P out . Then, xc( P min ) = rank + ( S P ( out , P in )) .
Proof. Let P be any polytope in the sandwich, i.e. P in ⊆ P ⊆ P out . We can describe P as the convex hull of the vertices of P together with the vertices of P in . Similarly we can describe P as the intersection of all its facetdefining inequalities and the facet-defining inequalities of P out . Now the matrix S P ( out , P in ) is a submatrix of the slack matrix S P ( ) of this particular representation of P. Therefore, rank + ( S P ( )) glyph[greaterorequalslant] rank + ( S P ( out , P in )) . It is easy to see (see, for example, [12]) that the non-negative rank of the slack matrix of a polytope is not changed by adding redundant inequalities and points in its representation. Also, since the non-negative rank of the slack matrix of a polytope is equal to its extension complexity ([12]), we have that xc( P ) glyph[greaterorequalslant] rank + ( S P ( out , P in )) .
Now, suppose that rank + ( S P ( out , P in )) = r. That is there exist non-negative matrices T and U with r columns and rows respectively, such that S P ( out , P in ) = TU. Denote by T i the i -th row of T and U j the j -th column of U . Consider the polytope
$$Q = \{ ( x, y ) \in \mathbb { R } ^ { k + r } \ \colon \, a _ { i } ^ { T } x + T _ { i } y = b _ { i }, 1 \leqslant i \leqslant m, y \geqslant 0 \}$$
and let
$$R = \{ x \in \mathbb { R } ^ { k } \ \colon \ \exists y \in \mathbb { R } ^ { r }, ( x, y ) \in Q \}.$$
Since, by definition, R is a projection of Q and Q has at most r inequalities, we have that xc( R ) glyph[lessorequalslant] rank + ( S P ( out , P in )) .
Next we show that P in ⊆ R ⊆ P out . Suppose x ∈ R. Then ∃ y, ( x, y ) ∈ Q. That is y glyph[greaterorequalslant] 0 and a T i x + T y i = b i for all i. Since T is non-negative, T y i glyph[greaterorequalslant] 0 and therefore a T i x glyph[lessorequalslant] b i for all i. That is, x ∈ P out . Therefore, R ⊆ P out . Suppose x ∈ P in . Then x = ∑ n j =1 λ v , j j ∑ n j =1 λ j = 1 , λ j glyph[greaterorequalslant] 0 , for some λ. Consider y = ∑ n j =1 λ U . j j Then, for each i = 1 2 , , .., m we have that a T i x + T y i = ∑ n j =1 λ j ( a T i v j + T U i j ) = ∑ n j =1 λ j ( b i ) = b . i Clearly y glyph[greaterorequalslant] 0 since U is non-negative. So ( x, y ) ∈ Q and thus x ∈ R. Therefore P in ⊆ R ⊆ P out .
Since R lies in the polytopal sandwich we have xc( R ) glyph[greaterorequalslant] rank + ( S P ( out , P in )) as proved in the first paragraph of the proof. Therefore the inequality is in fact an equation and we may set P min = R completing the proof.
Note that given a polytopal sandwich P in , P out , any lower bound on the non-negative rank of its slack matrix S P ( in , P out ) is also the lower bound on the succinct non-negative rank of S P ( in , P out ). Conversely, any upper bound on the succinct non-negative rank of S P ( in , P out ) is also an upper bound on the non-negative rank of S P ( in , P out ). In the next subsection we will define canonical polytopal sandwiches for binary languages and discuss the relation of the non-negative rank and succinct non-negative rank of the associated slack matrix with whether or not membership testing in the language belongs to P/poly .
## 5.2. Languages and their sandwiches
In the sequel we consider languages over the 0 / 1 alphabet. Let L ⊆ { 0 1 , } ∗ be a such a language. For any positive integer n define the set L n ( ) as
$$L ( n ) = \{ x \in \{ 0, 1 \} ^ { n } \ \colon \ x \in L \} \,.$$
To make the connection with Section 2.2, L plays the role of the decision problem X and L n ( ) is the set of instances of size n that have a 'yes' answer.
Corresponding to any language L n ( ) let us define a polytopal sandwich given by a pair of polytopes. The inner polytope is described by its vertices and is contained in the outer polytope, which in turn is described by a set of inequalities. Both the vertices of the inner polytope and the inequalities for the outer polytope depend only on the language L n . ( ) We call such a sandwich the characteristic sandwich of L n ( ).
For every language L ⊆ { 0 1 , } ∗ we define characteristic functions ψ : { 0 1 , } ∗ → { 0 1 , } and φ : { 0 1 , } ∗ → {-1 1 , } ∗ with
$$\psi ( x ) = \begin{cases} 1 & \quad \text{if $x\in L$} \\ 0 & \quad \text{if $x\notin L$} \end{cases},$$
$$\phi ( x ) _ { i } = \begin{cases} 1 & \quad \text{if $x_{i}=1$} \\ - 1 & \quad \text{if $x_{i}=0$} \end{cases}$$
In terms of Section 2.2, ψ x ( ) will play the role of w x and φ x ( ) will play the role of the objective function vector c . The inner polytope is defined by
$$V ( n ) = C H \{ ( x, \psi ( x ) ) \ \colon \ x \in \{ 0, 1 \} ^ { n } \}.$$
In terms of Section 2.2, V ( n ) plays the role of P and for the perfect matching problem it is PM n . For positive integer n and 0 < d ≤ 1 / 2, define
$$H ( n, d ) = \{ ( x, w ) \in [ 0, 1 ] ^ { n + 1 } \ \colon \ \phi ( a ) ^ { T } x + d w \leqslant 1 ^ { T } a + d \psi ( a ) \ \ \forall a \in \{ 0, 1 \} ^ { n } \}.$$
Note that the normal vectors of the inequalities defining H n,d ( ) are just the optimization directions ( c, d ) that were used in Section 2.2.
By direct substitution we see that each vertex ( x, ψ ( x )) of V ( n ) satisfies the constraints of H n,d ( ) and so V ( n ) ⊆ H n,d ( ). We will show that the following matrix is its slack matrix. For every n consider the 2 n × 2 n non-negative matrix M n,d ( ) defined as follows. Rows and columns of M n,d ( ) are indexed by 0 / 1 vectors a, b of length n and
$$M ( n, d ) [ a, b ] = \mathbb { 1 } ^ { T } a - 2 a ^ { T } b + \mathbb { 1 } ^ { T } b + \alpha ( a, b )$$
where
$$\alpha ( a, b ) = \begin{cases} d & \text{if $a\in L(n),b\notin L(n)$} \\ - d & \text{if $a\notin L(n),b\in L(n)$} \\ 0 & \text{otherwise} \end{cases}.$$
Lemma 6. The slack of H n,d ( ) with respect to V ( n ) is the matrix M n,d . ( )
Proof. Consider two vectors a, b ∈ { 0 1 , } n . The slack of the inequality corresponding to φ a ( ) with respect to ( b, ψ ( b )) is
$$\begin{cases} 1 ^ { T } a + d - d \psi ( b ) - \phi ( a ) ^ { T } b & \quad \text{ if } a \in L \\ 1 ^ { T } a - d \psi ( b ) - \phi ( a ) ^ { T } b & \quad \text{ if } a \not \in L \end{cases}$$
Observing that φ a ( ) = a -( 1 -a ) = 2 a -1 we can see that 1 T a -φ a ( ) T b = 1 T a + 1 T b -2 a T b. Therefore the slack of the inequality corresponding to φ a ( ) with respect to ( b, ψ ( b )) is
$$\begin{cases} 1 ^ { T } a + 1 ^ { T } b - 2 a ^ { T } b + d & \quad \text{ if } a \in L, b \notin L \\ 1 ^ { T } a + 1 ^ { T } b - 2 a ^ { T } b - d & \quad \text{ if } a \notin L, b \in L \\ 1 ^ { T } a + 1 ^ { T } b - 2 a ^ { T } b & \quad \text{ otherwise } \end{cases}$$
The following lemma is analogous to Proposition 3.
Lemma 7. Let P be a polytope such that V ( n ) ⊆ P ⊆ H n,d . ( ) Then, deciding whether a vector a ∈ { 0 1 , } n is in L or not can be achieved by optimizing over P along the direction ( φ a , d ( ) ) .
Proof. Let a be a given vector in { 0 1 , } n . Consider the maxima z v , z p , z h of φ a ( ) T x + dw when ( x, w ) ∈ V ( n ), ( x, w ) ∈ P, and ( x, w ) ∈ H n,d ( ) respectively. Since V ( n ) ⊆ P ⊆ H n,d ( ) we have that z v glyph[lessorequalslant] z p glyph[lessorequalslant] z h . Since ( a, ψ ( a )) ∈ V ( n ), z v ≥ φ a ( ) T a + dψ a ( ) = 1 T a + dψ a ( ). Furthermore, a gives rise to the inequality
$$\phi ( a ) ^ { T } x + d w \leqslant \mathbf 1 ^ { T } a + d \psi ( a )$$
in H n,d ( ) and so z h ≤ 1 T a + dψ a ( ). Since z v ≤ z p ≤ z h , z v = z p = z h = 1 T a + dψ a ( Therefore whether z p = 1 T a + d or not tells us whether a ∈ L or not.
).
Theorem 3. Let L be a language over the 0/1 alphabet and let n be any positive integer. (a) If L belongs to the class P/poly then there is a constant 0 < d < 1 / 2 and slack matrix M n,d ( ) such that both the the size of d and the non-negative rank of M n,d ( ) are polynomial in n . (b) If there is a constant 0 < d < 1 / 2 and slack matrix M n,d ( ) such that both the size of d and the succinct non-negative rank of M n,d ( ) are polynomial in n then L belongs to the class P/poly .
Proof. (a) Suppose L belongs to the class P/poly then by Lemma 1 there is a polytope Q that is a WEF of V ( n ) with size bounded by a polynomial in n . Let P = { ( x, w ) : ( x, w, s ) ∈ Q } be the projection of Q onto its first n + 1 coordinates. Since Q is a WEF of V ( n ) we have V ( n ) ⊆ P . Applying Proposition 4 to Q we obtain a value d 0 , whose size is polynomially bounded in n . Set d = d / 0 2 . For any vector a ∈ { 0 1 , } n this proposition implies that if we optimize φ a x ( ) + dw over Q the maximum value obtained is 1 T a + dψ a ( ). Therefore the inequality corresponding to this a in the definition of H n,d ( ) is satisfied for all ( x, w ) ∈ P . It follows that P ⊆ H n,d ( ) and so P lies in the polytopal sandwich defined by V ( n ) and H n,d ( ). Since M n,d ( ) is the slack of H n,d ( ) with respect to V ( n ), by Lemma 5 the non-negative rank of M n,d ( ) is upper bounded by the extension complexity of any polytope P sandwiched between the two polytopes. Since the size of Q is bounded above by a polynomial in n and Q is the extension of some polytope that can be sandwiched between H n,d ( ) and V ( n ) we have that the non-negative rank of M n,d ( ) is bounded by a polynomial in n.
(b) Suppose for some constant 0 < d < 1 / 2 the slack matrix M n,d ( ) has succinct non-negative rank r and that both d and r are polynomially bounded in n . Since M n,d ( ) is the slack of H n,d ( ) with respect to V ( n , ) by Lemma 5 there exists a polytope P such that the extension complexity of P is r and V ( n ) ⊆ P ⊆ H n,d . ( ) In other words, there exist polytopes P and Q such that Q has r facets, projects to P and V ( n ) ⊆ P ⊆ H n,d ( ). Furthermore, from the proof of Lemma 5 we know there exists such a Q whose description contains only numbers polynomially bounded in n because the rank r non-negative factorization is succinct by our assumption. Using the given value of d , by Lemma 7 we can optimize over P to decide whether x ∈ L n ( ) or not for a given x. Furthermore, optimizing over P can be done by optimizing over Q instead, since Q projects to P . Since Q has polynomial bit complexity, we can use interior point methods to do the optimization and so determine membership in L in polynomial time.
Theorem 3(a) in principle paves a way to show that membership testing in a certain language is not in P/poly by showing that the non negative rank of associated slack matrices cannot be polynomially bounded. Various techniques exist to show lower bounds for the non-negative rank of matrices and have been used to prove super-polynomial lower bounds for the extension complexity of important polytopes like the CUT polytope, the TSP polytope, and the Perfect Matching polytope of Edmonds, among others [12, 18, 4, 17]. Whether one can apply such techniques to show super-polynomial lower bounds on the non-negative rank of the slack of the characteristic sandwich of some language is left as an open problem.
## 6. Concluding remarks
In this paper we have given a method for constructing polynomial size linear programs for solving decision problems in P from their underlying algorithms. As a concrete example we can construct an LP of size O n ( 6 log n ) that decides whether an input graph on n vertices has a perfect matching or not. This LP is based on an O n ( 4 ) implementation of Edmonds' algorithm and smaller polytopes can be obtained by using more efficient implementations. This raises the question of what is the smallest size LP for the perfect matching problem and whether or not there is an LP that has an explicit formulation. Along these lines we mention a recent result in [5] for 2SAT. The standard formulation for this problem has superpolynomial time extension complexity. Applying the method of Section 4 to a standard 2SAT algorithm yields an LP of size O n ( 4 log n ). However in [5] a simple compact WEF for 2SAT of size O n ( 3 ) is given.
The discussion in this paper was centred around decision problems and one may wonder if it can be applied to optimization problems also. Before addressing this let us recall some discussion on the topic in Yannakakis' paper [20]:
Linear programming is complete for decision problems in P ; the P = NP ? question is equivalent to a weaker requirement of the LP (than that expressing the TSP polytope), in some sense reflecting the difference between decision and optimization problems. (P. 445, emphasis ours)
The term 'expressing' in the citation refers to an extended formulation of polynomial size. The method we have described can indeed be used to construct polynomial sized LPs for optimization problems which have polynomial time algorithms. Consider, for example, the problem of finding a maximum weight matching for the complete graph K n , where the edge weights are integers of length W bits. For simplicity we assume the weights are non-negative, but a small extension would handle the general case. We construct a polytope P as defined by (10) as follows. The binary vectors x have length Wn n ( -1) / 2 + W + log glyph[ceilingleft] 2 n glyph[ceilingright] , where the first Wn n ( -1) / 2 bits encode the edge weights and the remaining bits encode an integer k, 0 ≤ k < 2 W n . The bit w x is defined by setting w x = 1 whenever the edge weights specified in x admit a matching of weight k or greater and w x = 0 otherwise. Note that the unweighted maximum matching problem for graphs on n vertices fits into this framework by setting W = 1.
Applying the method of Section 4 to the weighted version of Edmonds' algorithm we obtain a polynomial sized WEF Q for P . We can decide by solving a single LP over Q if a given weighted K n has a matching of weight at least k , for any fixed k : the last W + log glyph[ceilingleft] 2 n glyph[ceilingright] coefficients of the objective function (11) vary depending on k . Therefore, by binary search on k we can solve the maximum weight matching problem for a given input by optimizing O W ( +log n ) times over Q with objective functions depending on k . We do not, however, know how to solve the weighted matching problem by means of a single polynomially sized linear program.
## Acknowledgments
We would like to thank the referees of this paper for their detailed comments leading to several important improvements. This research is supported by a Grant-in-Aid for Scientific Research on Innovative Areas - Exploring the Limits of Computation(ELC), MEXT, Japan. Research of the second author is partially supported by NSERC Canada. Research of the third author is supported by project GA15-11559S of GA CR. ˇ
## References
- [1] S. Arora and B. Barak. Computational Complexity - A Modern Approach . Cambridge University Press, 2009.
- [2] D. Avis and D. Bremner. Sparktope. https://gitlab.com/sparktope/sparktope and http://www.cs. unb.ca/ ~ bremner/research/sparks\_lp/ , 2017.
- [3] D. Avis and H. R. Tiwary. A generalization of extension complexity that captures P. Inf. Process. Lett. , 115(6-8):588-593, 2015.
- [4] D. Avis and H. R. Tiwary. On the extension complexity of combinatorial polytopes. Math. Program. , 153 (1):95-115, 2015.
- [5] D. Avis and H. R. Tiwary. Compact linear programs for 2SAT. CoRR , abs/1702.06723, 2017(European J. Combinatorics, to appear).
- [6] G. Braun, S. Fiorini, S. Pokutta, and D. Steurer. Approximation limits of linear programs (beyond hierarchies). In 53rd Annual IEEE Symposium on Foundations of Computer Science, FOCS 2012, New Brunswick, NJ, USA, October 20-23, 2012 , pages 480-489, 2012.
- [7] S. A. Cook and R. A. Reckhow. Time bounded random access machines. J. Comput. System Sci. , 7: 354-375, 1973. Fourth Annual ACM Symposium on the Theory of Computing (Denver, Colo., 1972).
- [8] D. P. Dobkin, R. J. Lipton, and S. P. Reiss. Linear programming is log-space hard for P. Inf. Process. Lett. , 8:96-97, 1979.
- [9] J. Edmonds. Paths, trees, and flowers. Canad. J. Math. , 17:449-467, 1965.
- [10] J. Edmonds. Maximum matching and a polyhedron with 0 1 vertices. , J. of Res. the Nat. Bureau of Standards , 69 B:125-130, 1965.
- [11] S. Even and O. Kariv. O n ( 2 5 . ) algorithm for maximum matching in general graphs. In FOCS , pages 100-112, 1975.
- [12] S. Fiorini, S. Massar, S. Pokutta, H. R. Tiwary, and R. de Wolf. Exponential lower bounds for polytopes in combinatorial optimization. J. ACM , 62(2):17:1-17:23, 2015.
- [13] N. Gillis and F. Glineur. On the geometric interpretation of the nonnegative rank. Linear Algebra and its Applications , 437(11):2685-2712, dec 2012.
- [14] M. Gr¨ otschel, L. Lov´sz, and A. Schrijver. a Geometric algorithms and combinatorial optimization , volume 2 of Algorithms and Combinatorics . Springer-Verlag, Berlin, second edition, 1993.
- [15] E. Horowitz and S. Sahni. Fundamentals of Computer Algorithms. Computer Science Press, 1978.
- [16] K. Pashkovich. Extended formulations for combinatorial polytopes . PhD thesis, Otto-von-GuerickeUniversit ¨ , Magdeburg, 2012. t
- [17] S. Pokutta and M. V. Vyve. A note on the extension complexity of the knapsack polytope. To appear in Operations Research Letters , 2013.
- [18] T. Rothvoss. The matching polytope has exponential extension complexity. J. ACM , 64(6):41:1-41:19, 2017.
- [19] L. G. Valiant. Reducibility by algebraic projections. Enseign. Math. (2) , 28(3-4):253-268, 1982.
- [20] M. Yannakakis. Expressing combinatorial optimization problems by linear programs. Journal of Computer and System Sciences , 43(3):441-466, 1991.
## Appendices
## A. Valid inequalities and facets of PM n
We give here two classes of valid inequalities for PM n . Firstly, let M ⊆ E define a perfect matching in G . We have:
$$w \geqslant \sum _ { i j \in M } x _ { i j } - \frac { n } { 2 } + 1$$
To see the validity of this inequality, note that if M is a perfect matching the sum becomes n/ 2 and the inequality states that w glyph[greaterorequalslant] 1, i.e. G x ( ) has a perfect matching. We show below that the each inequality of type (20) define a facet of PM . n Since the number of perfect matchings in K n is the double factorial ( n -1)!! = ( n -1) · ( n -3) ... 3 · 1 the number of facet defining inequalities of PM n is therefore super-polynomial.
For a second set of valid inequalities, first let E n be the set of edges of K n . A proper subset S ⊂ E n is hypo-matchable if it has no matching of size n/ 2 but the addition of any other edge from E n \ S to S yields such a matching. Then we have:
$$w \leqslant \sum _ { i j \in E _ { n } \searrow S } x _ { i j }$$
To see the validity of this inequality note that if the sum is zero then no edges from E n \ S are in G x ( ). So G x ( ) has no perfect matching and so w must be zero.
We now prove that the inequalities (20) are facet defining for PM 2 n , where we have replaced n by 2 n to avoid fractions. For any integer s we use the notations I s × s , 1 s × s and O s × s to represent, respectively, the s × s identity matrix, matrix of all ones, and matrix of all zeroes. With only one subscript, the latter two notations represent the corresponding vectors. For an integer n we let t = 2 n n ( -1). Without loss of generality, consider a perfect matching M in K 2 n consisting of the n edges 12 34 56 , , , ..., (2 n -1)2 n and let E t be the t edges of
K 2 n that are not in M . We construct a set of t + n = n (2 n -1) graphs G x ( ) for which inequality (20) is tight and for which the x vectors are affinely independent. The corresponding ( t + n +1) × ( t + n +1)-matrix A of edge vectors x is:
$$A = \left [ \frac { I _ { t \times t } } { \begin{matrix} 1 _ { t \times t } & 1 _ { t \times n } & 1 _ { t } \\ 0 _ { n \times t } & 1 _ { n \times n } & 0 _ { n } \\ 0 _ { t } & 1 _ { n } & 1 \end{matrix} \right ] }$$
We label the columns of A as follows. The first t columns correspond to the edges in E t listed in lexicographical order by ij . The next n columns are indexed by the edges 12 34 , , ..., (2 n -1)2 n of M and the final column by w . The first t rows of A consist of the edge vectors of graphs which contain M and precisely one other edge ij not in M , arranged in lexicographic order by ij . This means that the top left hand block in A is the identity matrix. Since all these graphs contain M , which is a perfect matching, all these remaining entries in the first t rows of A are ones.
The next n rows of A correspond to graphs with edge vectors M \ { ij } , where ij ranges over the perfect matching 12 24 , , ..., (2 n -1)2 n . Clearly the first block of these rows are all zeroes and the second block is 1 n × n -I n × n . The last column is all zero since none of these graphs has a perfect matching. The final row of A corresponds to the graph M .
It is straight forward to perform row operations on A to transform it into an upper triangular matrix with ± 1 on the main diagonal. This can be performed by subtracting the last row from the preceding n rows. The middle block of A is now -I n × n . Finally these rows can then be added to the last row, which is then divided by n -1. It is then all zero except for the last column, which is -1. This completes the proof.
| null |
[
"David Avis",
"David Bremner",
"Hans Raj Tiwary",
"Osamu Watanabe"
] |
2014-08-04T20:12:55+00:00
|
2018-04-25T11:16:33+00:00
|
[
"cs.DM"
] |
Polynomial size linear programs for problems in P
|
A perfect matching in an undirected graph $G=(V,E)$ is a set of vertex
disjoint edges from $E$ that include all vertices in $V$. The perfect matching
problem is to decide if $G$ has such a matching. Recently Rothvo{\ss} proved
the striking result that the Edmonds' matching polytope has exponential
extension complexity. Here for each $n=|V|$ we describe a perfect matching
polytope that is different from Edmonds' polytope and define a weaker notion of
extended formulation. We show that the new polytope has a weak extended
formulation (WEF) $Q$ of polynomial size. For each graph $G$ with $n$ vertices
we can readily construct an objective function so that solving the resulting
linear program over $Q$ decides whether or not $G$ has a perfect matching. The
construction is uniform in the sense that, for each $n$, a single polytope is
defined for the class of all graphs with $n$ nodes. The method extends to solve
poly time optimization problems, such as the weighted matching problem. In this
case a logarithmic (in the weight of the optimum solution) number of
optimizations are made over the constructed WEF.
The method described in the paper involves construction of a compiler that
converts an algorithm given in a prescribed pseudocode into a polytope. It can
therefore be used to construct a polytope for any decision problem in {\bf P}
which can be solved by a given algorithm. Compared with earlier results of
Dobkin-Lipton-Reiss and Valiant our method allows the construction of explicit
linear programs directly from algorithms written for a standard register model,
without intermediate transformations. We apply our results to obtain polynomial
upper bounds on the non-negative rank of certain slack matrices related to
membership testing of languages in {\bf P/Poly}.
|
1408.0808v1
|
## THE IDENTIFICATION OF RR LYRAE AND δ Scuti STARS FROM VARIABLE GALEX ULTRAVIOLET SOURCES.
## T.D. Kinman 1
National Optical Astronomy Observatory, 950 N. Cherry Avenue, Tucson, AZ 85719, USA
Warren R. Brown
Smithsonian Astrophysical Observatory, 60 Garden St., Cambridge, MA 02138, USA
AJ accepted
## ABSTRACT
We identify the RR Lyrae and δ Scuti stars in three catalogs of GALEX variable sources. The NUV amplitude of RR Lyrae stars is about twice that in V , so we find a larger percentage of low amplitude variables than catalogs such as Abbas et al. (2014). Interestingly, the ( NUV -V ) 0 color is sensitive to metallicity and can be used to distinguish between variables of the same period but differing [Fe/H]. This color is also more sensitive to T eff than optical colors and can be used to identify the red edge of the instability gap. We find 8 δ Scuti stars, 17 RRc stars, 1 RRd star and 84 RRab stars in the GALEX variable catalogs of Welsh et al. (2005) and Wheatley et al. (2008). We also classify 6 δ Scuti stars, 5 RRc stars and 18 RRab stars among the 55 variable GALEX sources identified as 'stars' or RR Lyrae stars in the catalog of Gezari et al. (2013). We provide ephemerides and light curves for the 26 variables that were not previously known.
Subject headings: stars: RR Lyrae. stars: δ Scuti, stars: variable. stars: red horizontal branch.
## 1. INTRODUCTION
The Galaxy Evolution Explorer ( GALEX ) (Martin et al. 2005) has imaged the whole sky in the FUV ( λλ 1350 - 1750 ˚ ) and A NUV ( λλ 1750 - 2750 ˚ ) waveA bands, opening the door to studying variable stars in the ultraviolet. One of the most important classes of variable stars are RR Lyraes, evolved horizontal branch stars that have intrinsic luminosities that correlate with pulsation amplitude and period. δ Scuti stars (DSCT ) are a dif2 ferent class of pulsators with colors and pulsation periods comparable to RR Lyraes; they are discussed in the appendix (Sec. A). Importantly, the ultraviolet ( UV ) amplitudes of RR Lyrae stars are much larger in GALEX wavebands than in optical wavebands (Wheatley et al. 2003). What remains unexplored are complete samples of UV variables.
alog exclusively uses NUV magnitudes. Few of these UV variables have optical identifications. Only 25% of the 84 sources in Catalog 1 and less than a third of the 410 sources in Catalog 2 are matched to optical objects of known type. Most of the UV variables with optical identifications are extragalactic objects. Less than 10% of the UV variables in the catalog of Gezari et al. (2013) have identifications with known types of stars.
Interestingly, detailed observations of several bright RR Lyrae stars show that their UV -optical colors correlate with metallicity. Wheatley et al. (2012) fit model light-curves to the FUV and NUV photomtery of six well-observed bright RR Lyrae stars. They find that the FUV and NUV light-curves, which primarily reflect temperature changes during the pulsation cycle, also depend on metallicity. We also find that ( NUV -V ) depends on metallicity.
To date, candidate variable GALEX sources have been cataloged by Welsh (2005) (Catalog 1) and by Wheatley et al. (2008) (Catalog 2). During the course of this work, a third catalog of over a thousand variable GALEX sources was published by Gezari et al. (2013); this cat-
- 1 NOAO is operated by AURA, Inc. under cooperative agreement with the National Science Foundation.
- 2 These pulsating variables (spectral types A0 to F5) have V amplitudes in the range 0.03 to 0.90 mag.; those with V amplitudes greater than 0.20 are often called High Amplitude δ Scuti stars.
The goal of this paper is to identify all of the Galactic stars in the GALEX variable catalogs in a complete and systematic way, with a particular focus on RR Lyrae variables. We investigate the ( NUV -V ) 0 color index as a possible diagnostic for identifying variable types: in Sec. 4.5 and Appendix B, we address the problem of distinguishing between type c RR Lyrae stars and contact binaries (Kinman & Brown 2008). We base our analysis on NUV magnitudes, because NUV magnitudes are more accurate than FUV magnitudes and are available for a larger number of objects. The use of the ( FUV -NUV ) color could provide additional constraint (see for example, Smith et al. 2014) but would limit the sample.
Our approach is to use the VizieR search tool and the JAAVS variable star catalog (VSX, Watson 2006) to identify known variables, such as those identified by Drake et al. (2012) in the Catalina Surveys, by Palaversa et al. (2013) in the LINEAR Survey, and by Abbas et al. (2014) in their search of the combined SDSS, PanSTARRS1 and Catalina Surveys . We look at every ob3 ject in Catalog 1 and in Catalog 2, and the subset of objects identified as 'stars' in the larger Gezari catalog. If we are unable to classify an object with the variable star catalogs, we then use data from the Catalina Survey Data Release 2 (Drake et al. 2009) to measure the ob-
3 The Abbas et al. Catalog was published only shortly before this paper was completed and so it has only been possible to make limited use of it.
Fig. 1.V -magnitude light curves for 4 newly identified RRc and 4 newly identified RRab variables presented in Table 2.
Fig. 2.V -magnitude light curves for 8 new identified RRab variables presented in Table 2.
ject's mean V magnitude and, if possible, its variability. Period-finding with the Catalina Survey data (e.g. using the NASA Exoplanet Archive Periodogram Service) becomes difficult for objects with V > 17 and for objects with V amplitudes /lessorsimilar 0.2 magnitudes; consequently, the types we assign to such variables are less certain. We also review the 38 objects previously identified as RR Lyrae stars in the Gezari catalog.
This paper presents the RR Lyrae and δ Scuti stars that we identify, and describes how GALEX magnitudes can assist in their identification. The ( NUV -V ) 0 color is particularly sensitive for detecting the temperature changes that come from pulsations, and for detecting the onset of this instability as a function of parameters such
Fig. 3.V -magnitude light curves for 7 newly identified δ Scuti stars and one EW variable presented in Table 2. The periods of the lowest amplitude δ Scuti stars are quite uncertain.
Fig. 4.V -magnitude light curves for 2 newly identified Algol (EA) variables presented in Table 2. The period of J221657.3 -000633.0 is uncertain.
as T eff . The identification of other types of stars and variables from the GALEX variability catalogs will be discussed elsewhere.
## 2. THE IDENTIFICATION OF RR LYRAE AND δ Scuti STARS IN THE FIRST TWO GALEX VARIABILITY CATALOGS.
Table 1 presents 110 RR Lyrae and δ Scuti stars that we identify from the first two catalogs of GALEX variables; these stars constitute 22% of all the sources in these catalogs. Our list comprises 8 δ Scuti stars, 17 RRc stars, 1 RRd star and 84 RRab stars. References to the sources of the GALEX magnitudes and the optical data are given in columns 4 and 10 respectively. The ( NUV -V ) 0 colors given in col. 8 refer to mean magnitudes 4 and their largest source of uncertainty lies in the estimation of the mean NUV from the available data; this uncertainty increases with the NUV amplitude and is therefore largest for the RRab variables. We assumed the galactic extinction A( NUV ) = 8.90E( B -V ) (Rey et al. 2007). The E( B -V ) were taken from the MAST archive of GALEX objects. Only 6 stars have E( B -V ) > 0.100 so the uncertainty in the correction for extinc-
4 The mean NUV magnitude is the mean of the maximum and minimum values. The mean V magnitude is the arithmetic mean of all the observations given by the Catalina DataRelease 2. For brighter variables not in this survey, the mean was estimated from the maximum and minimum magnitudes using the formula given in the appendix of Preston et al. (1991).
Fig. 5.Distribution of known RR Lyrae stars from Dambis et al. (2013) in plots of the de-reddened ( NUV -V ) 0 color versus period (upper panel) and versus [Fe/H] (lower panel). The filled circles, crosses, and open circles denote halo RRab, disk RRab, and RRc variables, respectively. There are significant differences in ( NUV -V ) 0 between the disk RRab stars ([Fe/H] ≥ -0 40), . the halo RRab stars ([Fe/H] ≤ -0 80), and the RRc stars. .
tion (Peek & Schiminovich, 2013) is probably significant only for these stars. The periods given in column 9 are those taken from the literature in the case of previously known variables; otherwise, the periods were determined by us.
We discover 2 new δ Scuti stars, 3 RRc stars and 10 RRab stars among the 110 variable stars. We present the ephemerides of the new variables in Table 2, and plot their light curves in Figures 1, 2, 3 & 4. One of these RRc and one of these RRab were later found in the catalog of Drake et al, (2014) but with differing periods; these differences are discussed in Sec. 4.6.
We investigate trends between ( NUV -V ) 0 color and stellar properties by studying a well-defined sample of known RR Lyrae stars. We begin by calculating ( NUV -V ) 0 for 150 nearby RR Lyrae stars using the recent compilation by Dambis et al. (2013), a sample for which we could easily obtain the necessary data. For this sample, we plot ( NUV -V ) 0 against period for RRc and for halo and disk RRab (Fig 5, upper panel), and against [Fe/H] (Fig 5, lower panel).
There is a significant difference in ( NUV -V ) 0 between the RRc and RRab that reflects the average temperature difference between the two types. There is also a significant difference in ( NUV -V ) 0 between the disk and halo RRab that reflects the metallicity difference between the two types as expected from Wheatley et al. (2012).
We also investigate trends between the ( NUV -V ) 0 color and stellar properties for a sample of known high-
Fig. 6.De-reddened ( NUV -V ) 0 color versus pulsation period for our program stars from Table 1. The stars are plotted as: high-amplitude δ Scuti stars (green solid circles), RRc (blue open circles), RRab from Catalog 1 (black open triangles) and RRab from Catalogue 2 (black filled triangles). The blue (solid), black (dotted) and red (dashed) horizontal lines show mean values of ( NUV -V ) 0 for known RRc, halo RRab, and disk RRab variables, respectively. Two samples of known high-amplitude δ Scuti stars variables are plotted with black x's (McNamara 1997) and encircled black x's (Suveges et al. 2012). The heavy blue vertical line shows the range of ( NUV -V ) 0 for blue horizontal branc stars at the blue edge of the instability gap (Kinman et al. 2007).
amplitude δ Scuti stars (HADS) taken from McNamara (1997) and δ Scuti of lower amplitude taken from Suveges et al. (2012). The results are shown in the plot of ( NUV -V ) 0 vs. Period given in Fig. 6. The blue open circles, black open triangles and black filled triangles are respectively the RRc (both Catalogs), RRab from Catalog 1 and RRab from Catalog 2 in Table 1. The horizontal solid blue, dotted black and dashed red lines show the mean values of ( NUV -V ) 0 for the RRc, halo RRab, and disk RRab respectively; their associated vertical lines show the r.m.s. standard errors of these means. There is good agreement between these mean values and the ( NUV -V ) 0 of our program stars. The observed increase in this color with increasing period is nicely matched to the ( NUV -V ) 0 location of the blue edge of the instability gap (Kinman et al. 2007), shown by the heavy blue vertical line at a period of 0.2 days in Fig. 6.
We discover three new RRab of particular interest. J100358.9-270001.4has the largest ( NUV -V ) 0 color and is the only object to lie in the color domain of the disk RR Lyrae stars. Yet at a heliocentric distance of roughly 5 kpc, it must be about 2 kpc above the galactic plane. Its proper motion of 8 ± 2 mas y -1 (Girard et al., 2011) corresponds to a velocity of 200 ± 50 km s -1 . This velocity and its relatively long pulsation period (0.6311 days) suggest that J100358.9-270001.4 is not a disk star, but a halo star. Spectroscopic radial velocity and metallicity measurements are needed to establish its origin. The two RRab stars with the smallest ( NUV -V ) 0 colors are J165153.4-223952.2 and J214038.1-233642.7. J165153.4223952.2 has the largest E( B -V ) = 0 419 and therefore . the most uncertain extinction correction of the sample,
TABLE 1 δ
Identifications of Scuti (DCST) and RR Lyrae Stars in Catalogs 1 and 2. [Available in full in the Online Journal]
| Star a | NUV max b | NUV amp c | Ref d | E( B - V ) | V max e | V amp f | ( NUV - V ) 0 g | Period h | Refs i | Notes |
|---------------------------------------|-------------|-------------|---------|--------------|-------------|-----------|-------------------|---------------------|---------------|---------|
| DCST: | | | | | | | | | | |
| J004347.9+421654.9 | 13.02 | 0.63 | 1,2 | 0.084 | 09.19 | 0.27 | 3.692 | 0.1249078 | 13 | 1 |
| J004347.7-445130.7 | 16.78 | 0.84 | 2 | 0.009 | 14.45 | 0.50 | 2.457 | 0.0574790 | 20 | 2 |
| J033715.2-283238.1 | 15.51 | 0.94 | 2 | 0.011 | 13.20 | 0.65 | 2.397 | 0.0561380 | 11 | |
| J104647.6+591445.5 | 19.86 | 0.93 | 2 | 0.008 | 17.40 | 0.80 | 2.480 | 0.1231520 | 20 | 3 |
| J133757.1+401610.8 | 20.00 | 0.91 | 1 | 0.006 | 18.00 | 0.50 | 2.177 | 0.0641760 | 16 | |
| J160608.8+530519.0 | 18.76 | 1.53 | 2 | 0.014 | 13.53 | 0.55 | 5.647 | 0.1027590 | 20 | 2 |
| J164106.8+404225.2 | 15.70 | 0.87 | 2 | 0.008 | 13.40 | 0.55 | 2.419 | 0.0512960 | 13 | 4 |
| J171250.9+582748.6 | 18.51 | 0.63 | 2 | 0.044 | 15.70 | 0.30 | 2.749 | 0.0803930 | 16,7 | 5 |
| RRc: | | | | | | | | | | |
| J033546.0-290209.8 | 16.09 | 0.90 | 2 | 0.008 | 13.45 | 0.50 | 2.797 | 0.2773379 | 18 | |
| J051855.0-492606.1 | 17.59 | 0.84 | 2 | 0.026 | 14.85 | 0.45 | 2.806 | 0.3822732 | 20 | 2 |
| J103720.9+591653.8 | 18.31 | 0.62 | 2 | 0.008 | 15.70 | 0.40 | 2.679 | 0.3132750 | 10,21,22 | |
| J103759.4+581213.5 | 19.67 | 1.03 | 2 | 0.008 | 16.90 | 0.40 | 3.038 | 0.4023800 | 21 | 2 |
| J105044.4+045115.9 | 19.51 | 0.97 | 2 | 0.027 | 16.80 | 0.45 | 2.813 | 0.2827820 | 16,19.22 | |
| J143823.9+635329.9 | 19.23 | 0.76 | 2 | 0.016 | 16.34 | 0.36 | 2.998 | 0.3559790 | 10,21 | |
| J144033.3-781914.6 | 17.09 | 0.79 | 2 | 0.108 | 13.80 | 0.70 | 2.789 | 0.3281820 | 13 | 6 |
| J154550.8-124936.8 | 19.51 | 0.62 | 2 | 0.146 | 15.96 | 0.28 | 2.970 | 0.2935940 | 20,22 | 2, 45 |
| J155800.6+535233.6 | 18.06 | 0.87 | 2 | 0.012 | 15.12 | 0.36 | 3.124 | 0.3284030 | 16,10,21,22 | |
| J160759.9+535209.9 | 19.20 | 0.90 | 2 | 0.012 | 16.52 | 0.44 | 2.841 | 0.3806390 | 12,22 | |
| J171008.9+585112.9 | 20.06 | 0.92 | 2 | 0.031 | 17.25 | 0.50 | 2.842 | 0.2821010 | 16,21,22 | |
| J203922.4-010346.0 | 16.88 | 0.81 | 2 | 0.084 | 13.57 | 0.40 | 3.089 | 0.3703490 | 14,7,22 | 7 |
| J204024.3-010950.0 | 19.06 | 0.72 | 2 | 0.079 | 15.68 | 0.34 | 3.170 | 0.3626603 | 14,7,21 | 8 |
| J204042.8-003717.3 | 20.28 | 0.72 | 2 | 0.159 | 17.12 | 0.40 | 3.023 | 0.2865740 | 14,21 | |
| J221444.8-182639.3 | 20.00 | 0.96 | 2 | 0.026 | 17.40 | 0.55 | 2.654 | 0.2862060 | 20 | 2 |
| J223601.0+134711.5 | 19.16 | 0.91 | 2 | 0.068 | 15.80 | 0.40 | 3.271 | 0.3760710 | 16,7,22 | 9 |
| J224029.0+120056.3 | 17.56 | 0.85 | 2 | 0.050 | 14.60 | 0.50 | 2.877 | 0.2786230 | 16,7,21,22 | 10 |
| RRd: | | | | | | | | | | |
| J130204.5+463533.7 | 18.01 | 1.45 | 1 | 0.014 | 15.45 | 0.60 | 2.959 | 0.3509191 | 12 | 11 |
| RRab: J001611.6-392721.5 | 16.16 17.25 | 2.02 1.48 | 2 | 0.015 0.007 | 13.50 14.00 | 1.00 0.90 | 3.420 3.747 | 0.5316009 0.6973440 | 11 | |
| J003043.0-422747.7 J004047.4-434948.9 | 18.20 | 1.95 | 2 2 | 0.016 | 15.30 | 0.95 | 3.633 | 0.4941930 | 9 | |
| | | | | | | | | | 20 | 2 |
| J004548.2-435509.1 | 16.00 | 1.38 | 1 | 0.001 | 13.00 | 0.70 | 3.565 | 0.5898310 | 11 | |
| J011010.9-452431.3 | 18.84 | 1.85 | 2 | 0.009 0.055 | 15.90 15.60 | 0.85 | 3.697 | 0.5897626 | 13 | 12 |
| J011805.2-751126.1 J013642.0-062743.1 | 18.61 19.70 | 1.41 0.63 | 2 2 | 0.028 | 16.35 | 1.20 0.25 | 3.032 3.483 | 0.480 0.7775660 | 13 16,7,22 | 13 14 |
| J022521.8-050017.4 | 19.42 | 1.57 | 2 | 0.027 | 16.90 | 0.80 | 3.012 | 0.6212419 | 3,21 | |
| | 16.98 | 1.63 | 2 | 0.009 | | 0.80 | 3.415 | 0.6975764 | | |
| J033500.4-272854.8 | | | | 0.377 | 14.20 | | 3.288 | 0.6229220 | 18 | 2,15 |
| J042832.3+165821.6 J051728.7-481712.8 | 19.30 18.60 | 1.64 1.42 | 2 2 | 0.027 | 14.42 15.70 | 1.00 0.80 | 3.291 | 0.6614350 | 20,7 20 | 2 |
| J052307.7-490336.1 | 18.17 | 2.13 | 2 | 0.030 | 15.47 | 1.10 | 3.397 | 0.5139700 | 20 | 2 |
| J081226.4+033320.5 | 18.95 | 1.95 | 2 | 0.023 | 16.08 | 0.72 | 3.677 | 0.5555151 | 6,7 | 16 |
| J093026.0+071221.6 | 15.83 | 0.97 | 1 | 0.058 | 12.08 | 0.79 | 3.666 | 0.6028494 | 13 | 17 |
| J100133.2+014328.5 | 17.21 | 1.98 | 2 | 0.019 | 14.40 | 1.00 | 3.521 | 0.5423741 | 3,10,19,21 | |
| J100358.9-270001.4 | 18.26 | 1.43 | 1 | 0.079 | 13.78 | 0.84 | | | 20 | |
| J102002.7+611538.9 | | | 1 | 0.007 | | | 4.556 | 0.6311440 | | 2 |
| | 16.71 | 1.62 | | | 14.34 | 0.81 | 3.005 | 0.6428627 | 5,3,10,21 | |
| J102641.8+572858.7 | 18.78 | 1.84 | 2 | 0.011 | 15.80 | 0.85 | 3.719 | 0.6636605 | 10,21 | |
| J103538.2+581549.1 | 15.53 | 1.92 | 2 | 0.008 | 12.87 | 1.02 | 3.384 | 0.5242549 | 13,7 | 18 |
| J104707.6+053349.2 | 17.78 | 2.25 | 2 | 0.024 | 15.00 | 1.00 | 3.642 | 0.5369295 | 10,19,21 | |
| J104803.1+554209.9 | 18.32 | 2.14 2.49 | 2 2 | 0.008 0.011 | 15.70 16.10 | 1.05 | 3.476 | 0.5065280 0.4748198 | 10,21 3,21 | |
| J104844.2+581538.6 | 18.47 | | | | | 1.30 | 3.317 | | | |
| J105130.9+043801.7 | 17.74 | 2.03 | 2 | 0.033 | 15.00 | 1.00 | 3.403 | 0.5243170 | 16,10,19,21 | |
| J105513.8+564747.7 | 17.09 | 2.24 | 2 2 | 0.007 0.009 | 14.65 | 0.75 | 3.519 | 0.5417451 | 16,10 3,10,21 | |
| J105622.3+570522.1 | 17.09 | 2.27 | 1 | 0.042 | 14.70 | 0.90 | 3.402 | 0.5016290 0.4499016 | 13,3,42 | 19 |
| J105926.1-005927.9 | 19.53 18.12 | 2.09 1.50 | 1 | 0.014 | 16.72 | 1.43 0.85 | 3.246 3.514 | 0.6449104 | 3 | |
| J111147.3+510549.4 J112334.9+474014.6 | 19.45 | | | 0.014 | 15.10 | | 3.353 | 0.5224715 | | |
| J113340.3+502328.0 | | 1.07 | 1 | | 16.18 | 1.10 | | | 3,10,21 | |
| | | | | | | | | 0.4976910 | | |
| Truncated for display | 16.71 | 2.93 | 1 | 0.014 | 14.70 | 1.10 | 3.333 | | 13,21 | 20 |
References . - (1) Welsh+(2005)(The GALEX Ultraviolet Variability Catalog); (2) Wheatley+(2008)(The Second GALEX Ultraviolet Varaiability Catalog); (3) Drake+(2013); (4) Keller+(2008); (5) Kinemuchi+(2006); (6) Kraus+(2007); (7) Li+(2011); (8) Miceli+(2008); (9) Norton+(2007); (10) Palaversa+(2013); (11) Pojmanski(2002)(ASAS Catalogue); (12) Poleski(2013); (13) Samus+(2012)(Gen. Cat. Variable Stars); (14) Sesar+(2010); (15) Vivas+(2004)(QUEST Survey); (16) Watson+(2006)(AAVSO VSX Catalog); (17) Wils+(2006); (18) Wils(2010)(Data in VSX Catalog); (19) Zinn+(2014)(Extension of QUEST Survey); (20) This Paper (New Identification). (21) Abbas+ (2014); (22) Drake+ (2014).
Note . -(1) CC And. (2) New identification. Ephemeris in Table 2. (3) New Identification. Type and period uncertain. (4) v1209 Her (SX Phe type). (5) Sp Type A (7) (6) v344 Aps. (7) Sp Type A (7) (8) Sp Type F (7) (9) Sp Type A (7) (10) Sp Typ A (7) (11) Period is first overtone. (12) UU Phe. (13) BG Tuc, B magnitude. (14) Sp Typ A (7) (15) Sp Typ F (7) (16) Sp Typ F (7) (17) WW Leo. (18) v341 Uma;Sp Type A (7). (19) IX Leo. (20) CZ Uma.
- a GALEX identification of source.
- c Amplitude of NUV magnitude.
- b NUV magnitude at maximum.
- d Reference for GALEX source.
- f Amplitude of V magnitude.
- e V magnitude at maximum.
- g Difference beteween mean NUV and V magnitudes corrected for galactic extinction.
- i References for optical data.
- h Period in days.
TABLE 2 Positions, Types and Ephemerides of New † Variables.
| Star a | R.A. b (2000) | Dec. c (2000) | Type d | HJD(max) e 2450000.0+ | Period f (days) | V max g (mag.) | V min h (mag.) |
|--------------------------------------------|-----------------|-----------------|----------|-------------------------|-------------------|------------------|------------------|
| CATALOGS 1 and 2: | | | | | | | |
| J004347.7 - 445130.7 | 010.9488 | - 44.8586 | DCST | 3556.1599 | 0.057479 | 14.45 | 14.95 |
| J160608.8+530519.0 | 241.5371 | +53.0890 | DCST | 3656.5589 | 0.102759 | 13.53 | 13.64 |
| J051855.0 - 492606.1 | 079.7293 | - 49.4349 | RRc | 3578.5038 | 0.382273 | 14.85 | 15.30 |
| J154550.8 - 124936.8 ‡ | 236.4618 | - 12.8269 | RRc | 3523.4753 | 0.293594 | 15.96 | 16.24 |
| J221444.8 - 182639.3 | 333.6870 | - 18.4444 | RRc | 3583.6795 | 0.286206 | 17.40 | 17.95 |
| J004047.4 - 434948.9 | 010.1981 | - 43.8301 | RRab | 3555.8138 | 0.494193 | 15.30 | 16.25 |
| J042832.3+165821.6 | 067.1349 | +16.9726 | RRab | 3469.5764 | 0.622922 | 14.45 | 15.45 |
| J051728.7 - 481712.8 | 079.3700 | - 48.2869 | RRab | 3576.4696 | 0.661435 | 15.70 | 16.50 |
| J052307.7 - 490336.1 | 080.7823 | - 49.0599 | RRab | 3576.4712 | 0.513970 | 15.47 | 16.57 |
| J100358.9 - 270001.4 | 150,9955 | - 27.0002 | RRab | 3694.8098 | 0.631144 | 13.78 | 14.62 |
| J122836.9 - 064230.0 ‡ | 187.1540 | - 06.7085 | RRab | 3496.5066 | 0.501648 | 17.50 | 18.60 |
| J132546.5 - 425141.3 | 201.4437 | - 42.8615 | RRab | 3556.7107 | 0.451759 | 15.25 | 16.25 |
| J165153.4 - 223952.2 | 252.9728 | - 22.6644 | RRab | 3599.5306 | 0.613432 | 13.80 | 14.75 |
| J213856.4 - 240234.6 | 324.7351 | - 24.0430 | RRab | 3597.8294 | 0.594016 | 15.80 | 16.55 |
| J214038.1 - 233642.7 | 325.1592 | - 23.6118 | RRab | 3638.7051 | 0.564344 | 17.25 | 17.90 |
| Gezari et al. (2013): J022709.0 - 032420.6 | 036.7872 | - 03.4058 | DCST | 3648.7343 | 0.1685980 | 16.60 | 16.69 |
| J221812.8+004839.0 | 334.5535 | +00.8106 | DCST | 3507.9368 | 0.0413220 | 17.35 | 17.50 |
| J160552.5+540825.0 | 241.4690 | +54.1404 | DCST | 3880.8828 | 0.0587400 | 17.70 | 18.10 |
| J160905.7+534712.2 ‡ | 242.2740 | +53.7866 | DCST | 3505.5268 | 0.1697210 | 16.50 | 16.85 |
| J021838.1 - 041951.0 | 034.6589 | - 04.3304 | DCST | 3627.8700 | 0.0759360 | 17.70 | 17.80 |
| J022121.9 - 033040.0 ‡ | 035.3413 | - 03.5109 | RRc | 3627.5444 | 0.3438110 | 17.90 | 18.30 |
| J095632.9+011706.0 | 149.1372 | +01.2850 | RRab | 3464.1026 | 0.7711330 | 16.60 | 16.69 |
| J160348.2+545841.0 | 240.9508 | +54.9780 | RRab | 3880.6148 | 0.5418940 | 17.25 | 18.05 |
| J095945.7+005912.1 | 149.9400 | +00.9866 | EW | 3464.4354 | 0.2874170 | 14.62 | 14.75 |
| J221657.3 - 000633.0 | 334.2392 | - 00.1091 | EA | 3502.9800 | 18.32325 | 18.30 | 20.: |
| J022411.7 - 052501.6 ‡ | 036.0485 | - 05.4172 | EA | 3627.8052 | 0.8254050 | 13.95 | 14.3: |
- † All stars in this table had not been identified as variables prior to 2014 May 29.
- b Right Ascension in decimal degrees.
- a GALEX identification of source.
- c Declination in decimal degrees.
- e Heliocentric Julian Date of Maximum.
- d Variable Type.
- f Period in days.
- h V magnitude at minimum.
- g V magnitude at maximum.
- ‡ These stars were identified as variables by Drake et al. (2014).
TABLE 3 Data for the Variable Stars identified by Gezari et al. (2013) .
| (2000) | Dec. b (2000) | NUV amp c (mag.) | LC d | ( NUV - V ) 0 e (mag.) | Period f (days) | V g (mag.) | ( B - V ) h (mag.) | Type i | Type j | Type k | Notes |
|-------------------|-----------------|--------------------|--------|--------------------------|---------------------|--------------|----------------------|-------------|----------|-----------|-----------|
| 212.2576 | 53.1997 | 0.24 | V | 3.298 | · · · | 16.39 | 0.52 | RR | RR | · · · | |
| 36.7721 | -3.7063 | 0.36 | V | 1.896 | · · · | 16.54 | 0.26 | RR | RR | · · · | |
| 36.7872 | -3.4058 | 0.37 | V | 2.103 | 0.1685980 | 15.51 | 0.32 | RR | RR | δ Scuti | (1)(18) |
| 36.4632 | -4.8917 | 0.43 | V | 1.912 | · · · | 16.73 | 0.19 | RR | RR | · · · | |
| 334.5535 | 0.8106 | 0.45 | V | 1.728 | 0.0413220 | 17.3 | 0.21 | RR | RR | δ Scuti | (1) |
| 150.0513 | 1.0615 | 0.46 | V | 3,361 | · · · | 17.19 | 0.49 | RR | RR | · · · | |
| 150.0909 | 1.2796 | 0.49 | V | 4.477 | 0.3162890 | 14.82 | 0.61 | RR | RR | EW | (2)(22) |
| 34.6594 | -4.3309 | 0.51 | V | 2.280 | 0.0759360 | 17.72 | 0.26 | RR | RR | δ Scuti | (1) |
| 332.9349 | -0.423 | 0.63 | V | 1.782 | · · · | 17.42 | 0.5 | RR | RR | · · · | |
| 149.1372 | 1.285 | 0.73 | V | 3.359 | 0.7711330 | 16.69 | 0.37 | RR | RR | RRab | (1) |
| 149.6863 | 2.5218 | 0.79 | V | 3.553 | 0.6468498 | 17.61 | 0.47 | RR | RR | RRab | (3) |
| 241.4690 | 54.1404 | 0.80 | | 2.376 | 0.0587400 | 17.96 | 0.32 | RR | RR | δ Scuti | (1) |
| 242.2740 | 53.7866 | 0.89 | | 3.882 | 0.1697210 | 16.61 | 0.54 | RR | RR | δ Scuti | (1)(23) |
| 241.9999 | 53.8693 | 0.95 | V | 2.619 | 0.3806406 | 16.71 | 0.2 | RR | RR | RRab | (4,5) |
| 35.3413 | -3.5109 | 0.95 | V | 2.820 | 0.3438094 | 18.04 | 0.29 | RR | RR | RRc | (1,7,22) |
| 333.3199 | -0.5376 | 0.95 | | 2.114 | 0.3365870 | 16.66 | -0.45 | RR | RR | RRc | (6,21,22) |
| 333.5874 | 1.3271 | 0.96 | | 4.347 | · · · | 17.22 | 0.55 | RR | RR | · · · | |
| 332.4833 | -0.6473 | 0.99 | V | 2.290 | 0.3016423 | 16.05 | 0.22 | RR | RR | RRc | (8,9,21,2 |
| 215.9569 | 52.6509 | 1.00 | V | 0.961 | · · · | 20.3 | 0.53 | RR | RR | · · · | |
| 150.2291 | 1.3947 | 1.00 | V | 2.651 | 0.3210413 | 15.75 | 0.36 | RR | RR | RRc | (2,21,22) |
| 216.3439 | 52.7295 | > 1.04 | F | > 3.263 | · · · | 19.42 | 0.46 | RR | RR | · · · | |
| 214.0191 | 52.4641 | 1.11 | V | 3.532 | 0.5825432 | 16.41 | 0.51 | RR | RR | RRab | (2,3) |
| 333.8419 | 0.5817 | 1.18 | V | 0.960 | · · · | 20.07 | 0.26 | RR | RR | δ Scuti ? | (10) |
| 212.8810 | 53.5101 | 1.29 | | 3.375 | 0.5807728 | 16.54 | 0.41 | RR | RR | RRab | (11,21,22 |
| 150.5272 | 1.3757 | 1.41 | V | 3.271 | 0.7289104 | 16.28 | 0.5 | RR | RR | RRab | (2,3,21) |
| 332.8058 | -0.3874 | 1.55 | V | 2.612 | 0.5807727 | 16.98 | 0.33 | RR | RR | RRab | (2,21) |
| 242.8745 | 53.7296 | 1.57 | V | 3.052 | 0.5838609 | 16.07 | 0.52 | RR | RR | RRab | (2,4,12,2 |
| 334.4712 | -0.0933 | 1.71 | V | 2.876 | 0.6556452 | 13.93 | 0.37 | RR | RR | RRab | (13) |
| 241.8077 | 55.8589 | 1.85 | | 3.286 | 0.5464062 | 15.92 | 0.18 | RR | RR | RRab | (2,3,4,21 |
| 241.2133 | 54.141 | 1.89 | | 2.997 | 0.5161044 | 16.55 | 0.13 | RR | BHB | RRab | (2,3) |
| 334.2392 | -0.1091 | > 1.91 | F | > 2.754 | 18.32325 | 18.2 | 0.24 | RR | RR | EA (1,14) | |
| 150.3887 | 1.7245 | 1.93 | V | 3.047 | 0.5423790 | 14.91 | 0.26 | RR | RR | RRab | (2,3,4,21 |
| 36.3486 | -5.5392 | 2.07 | V | 2.832 | 0.5363961 | 16.79 | 0.46 | RR | RR | RRab | (3,20,21) |
| 240.9508 | 54.978 | 2.11 | V | 3.465 | 0.5418940 | 17.71 | 0.24 | RR | RR | RRab | (1) |
| 36.3409 | -5.005 | 2.30 | V | 2.888 | 0.6212305 | 17.27 | 0.46 | RR | RR | RRab | (3,21) |
| 150.1707 | 1.3682 | > 2.70 | V | > 3.115 | 0.4888790 | 18.55 | 0.16 | RR | BHB | RRab | (3) |
| 332.9690 | 0.633 | 2.91 | V | 2.633 | 0.5053500 | 16.73 | 0.31 | RR | RR | RRab | (3,21) |
| 333.0054 | -0.0571 | 2.92 | V | 2.680 | 0.5030960 | 17.42 | 0.32 | QSO | RRab | RRab | (6) |
| 36.0485 | -5.4172 | 1.03 | | 5.38 | 0.8254050 | 13.94 | 0.89 | · · · | Star | EA | (1,15,24) |
| | 0.0828 | 0.96 | F F | | | 16.08 | 0.17 | · · · | Star | RRc | (6,22) |
| 150.8023 333.0985 | 0.9722 | 1.00 | V V | 3.74 2.18 | 0.6061280 0.3505063 | 15.85 | 0.55 | · · · | Star | RRab | (3)(19) |
| 149.9732 | 3.2086 | 0.61 | | 5.24 | 0.2738640 | 15.12 | 0.69 | · · · · · · | Star | EA EW | (16,22) |
| 215.2242 | 54.4815 | 0.72 | | 4.91 | 0.3586440 | 13.92 | 0.58 | | Star | | (2,22) |
| 35.5465 | -3.1702 | 0.57 | V | 4.63 | 0.4051308 | 13.14 | 0.6 | · · · | Star | EW | (17.15,22 |
| 149.9400 | 0.9866 | 0.27 | V | 4.44 | 0.2874170 | 14.63 | 0.66 | · · · | Star | EW | (1) |
Note . - f variables. (1) New Identification (this paper). (2) Palaversa et al. (2013). (3) Drake et al. (2012). (4) Wheatley et al. (2008) Second GALEX catalogue o (5) Drake et al. (2014) give type and period. Poleski (2014) gives same period. (6) Sesar et al. (2010). (7) Radial velocity -154 ± 2 km s -1 (SDSS DR 9). (8) Sesar et al. (2012). (9) Radial velocity -213 ± 1 km s -1 (SDSS DR 9). (10) New Identification. Possible HADS but period uncertain. (11) CL Boo. (12) Radial velocity -302 ± 2 km s -1 (SDSS DR 9). (13) GG Aqr. (14) Faint object: Period and depth of minima uncertain. Drake et al. (2014) give Period of 2.23825 days. Radial velocity +29 ± 5 km s -1 (SDSS DR 9). (15) X-ray source (Flesch, 2010) (16) KM Boo. (17) Pojmanski (2002). (18) ( B -V ) = 0.28; [Fe/H] =-1.56; Radial Velocity = -14 ± 14 (Brown et al., 2008). (19) Radial velocites +76 ± 6 & +159 ± 3 (SDSS DR 9). (20) [C/Fe] = +3.4 (Christlieb et al., 2008). (21) Abbas et al. (2014). (22) Drake et al. (2014). (23) Drake et al. (2014) give type as EW and period of 0.339 days. See Sec. 4.6 for discussion. (24) Drake et al. (2014) give same type and period of 0.583 days. See Sec. 4.6 for discussion.
- a Right Ascension in decimal degrees.
- c Amplitude of NUV
- b Declination in decimal degrees.
.
- e Difference between mean NUV and V magnitudes corrected for galactic extinction.
- d Light curve type: V = Stochastic; F = Flaring (Gezari et al., 2013).
- f Period in days.
- h Mean ( B -V ) corrected for extinction.
- g Mean V corrected for extinction.
- i Variable type from color (Gezari et al., 2013).
- j Variable type (Gezari et al., 2013); RR = RR Lyrae, BHB = blue HB star.
- k Variable type (This paper). EA are Algol-type eclipsing systems. EW are W Ursa Majoris-type eclipsing systems.
Fig. 7.The relation between the NUV amplitude and the V amplitude for 84 type RRab variables. The line corresponds to the case where the NUV amplitude is twice the V amplitude. The variables from Catalog 1 (open circles) lie preferentially below the line while those from Catalog 2 (filled circles) are more equally above and below the line.
while J214038.1-233642.7 has an unusually low V amplitude for its period and is possibly multiperiodic.
The high amplitude δ Scuti stars in Table 1 are shown by green filled circles in Fig. 6 and lie within the domain of the known δ Scuti stars which show a wide range in ( NUV -V ) 0 . We note that the star SS Psc, which has a conflicting RRc classification in the GCVS and a δ Scuti star classification by McNamara (1997), is clearly in the δ Scuti star domain.
## 3. THE IDENTIFICATION OF RR LYRAE AND δ Scuti STARS IN THE GEZARI ET AL. (2013) CATALOG.
In Table 3 we present updated classifications and period measurements for 28 of the 38 RR Lyraes stars identified by Gezari et al. (2013). We find that 16 are RRab, 4 are RRc, 6 are δ Scuti stars and 2 are eclipsing systems. The main reason that we could not obtain acceptable light-curves for the remaining objects was because their visual amplitudes were comparable to the observational errors of their magnitudes. A fuller discussion is given in Sec. 4.6.
Interestingly, Gezari et al. (2013) provide a 'light curve type' (LC) which is either Stochastic (V) or Flaring (F) (col. 4, Table 3). We would expect that RR Lyrae and δ Scuti stars to have 'Stochastic' light curves, but two of the variables that Gezari et al. type as RR Lyrae have 'Flaring' light curves in NUV .
Another 17 variables in the Gezari catalog are identified simply as 'Stars.' We are able to make detailed classifications for 7 of these objects, listed at the end of Table 3. We identify 1 RRab, 1 RRc and 5 eclipsing systems. Five of the 'Stars' from the Gezari catalog have 'Flaring' light curves: 1 is an RRab, 2 are eclipsing systems and 2 are of unknown type.
We provide ephemerides for the eleven newly identified variables from the Gezari catalog in Table 2.
## 4. DISCUSSION
We investigate the stellar properties of our sample of UV -selected variables. We compare NUV pulsation amplitudes with optical V amplitudes, with the distribution of pulsation periods, and with the distribution of nonvariable stars. We explore the completeness of the UV -selected sample, the problem of distinguishing between contact binaries and pulsating stars and finally with a comparison of periods obtained in this paper with those independently obtained in a contemporous compilation by Drake et al. (2014).
## 4.1. The relation between the NUV and V amplitudes.
We begin by exploring the distribution of NUV and V pulsation amplitudes of the GALEX variables. For context, consider the five RRab variables with well-observed NUV light curves from Wheatley et al. (2012). The mean of the ratio of the NUV amplitudes to the corresponding V amplitudes (taken from the literature) for these five RRab stars is 2.15 ± 0.18. The [Fe/H] of these stars range from -1 5 . to -1 7 . (mean -1 64), . which means they are typical halo RR Lyrae stars.
In Figure 7 we plot the NUV amplitude vs. the V amplitude for the 84 RRab variables in Table 1. The variables from Catalog 1 are plotted as open circles and those from Catalog 2 by filled circles. The line represents the case where the NUV amplitude is twice that of the V amplitude. In this figure, 11 of the RR Lyrae from Catalog 1 lie above the line and 24 below; 26 of the variables from Catalog 2 lie above the line and 23 below. The Catalog 1 variables lie preferentially below the line (5% level of significance) while those from Catalog 2 are more equally spaced. Thus the NUV amplitudes for the RRab derived from Catalog 1 are on average lower than those from Catalog 2; the latter are generally derived from more observations and approximate more closely to the true amplitudes. The scatter in Fig. 7 is large so that the correlation between the NUV and V amplitudes is not strong. It is clear, however, that on average the RRab NUV amplitudes are twice as large as their V amplitudes.
## 4.2. The log period-amplitude relation for NUV and V amplitudes.
Next we explore the period-amplitude relation of the variables. In Figure 8 we plot the distribution of log(period) versus NUV amplitudes (left panel) and V amplitudes (right panel). For reference, the curve shows the relation for Oosterhoff Type I RRab variables observed in the globular cluster M3 (Cacciari et al. 2008). The curve refers to monoperiodic variables; curves with secondary periods such as RRd or those with Blazhko effect will have smaller amplitudes and lie below the line in this plot. In the right hand plot, the variables with periods shorter than 0.56 days (log P ≤ -0.25) lie mainly below the curve; those with longer periods lie increasingly above the curve with increasing period. We suspect that our V amplitudes (derived by visual inspection from light-curves largely derived from Catalina Survey data (Drake et al., 2009) or in a few cases LINEAR Survey data (Sesar et al., 2013)) are systematically too small and that the admixture of higher amplitude variables with periods greater than 0.56 days comes from a
Fig. 8.Amplitude vs. log period relation for the NUV amplitude (left panel) and V amplitude (right panel) for the 84 RRab variables from Catalog 1 (open circles) and Catalog 2 (solid circles). The curve is the relation found for the Oosteroff type 1 variables in the globular cluster M3 given by Cacciari et al. (2008); the amplitude of the curve has been doubled for the NUV amplitudes in the left panel.
population of Oosterhoff type II variables. The errors in the amplitudes are too great for a clean separation of the two types. A similar interpretation holds for the NUV amplitudes in the left plot in Fig. 8 although the scatter is larger.
Two stars of interest are J013642.0-062743.1 and J171251.8+185452.4, which have periods 0.7775660 and 0.8111936 days, respectively, and amplitudes less than 0.4 mag. Schmidt (2002) has studied stars with periods between 0.60 and 1.0 days and explored the various types of variables in this range. He finds 6 RRab with periods greater than 0.75 days and only 4 with periods greater than 0.80 days. None of these stars have amplitudes less than 0.50 mag. A new survey by Abbas et al. (2014) finds ∼ 20 RRab with periods greater than 0.80 days. Most of these stars have V amplitudes between 0.4 and 0.6 magnitudes, and constitute about 0.03% of their total sample. Here, we find two RRab with smaller amplitudes in this period range, the aforementioned J013642.0-062743.1 and J171251.8+185452.4. These two RRab constitute ∼ 2% of our sample, suggesting that the use of GALEX magnitudes (where observed amplitudes are larger) facilitates the discovery of such low ampltiude stars.
## 4.3. The Red Horizontal Branch (RHB) stars.
In the plot of ( NUV -V ) 0 against period, the hottest RRLyrae stars (type RRc) are coterminous with the blue horizontal branch stars at the blue edge of the instability gap. The stars near this blue edge are nearly all metalpoor; thick disk blue horizontal branch stars are rare (Kinman et al. 2009).
We now consider the non-variable horizontal branch stars on the cool side of the instability gap: the red horizontal branch (RHB) stars. Fig. 9 plots ( NUV -V ) 0 against effective temperature (T eff ) for these stars. The sources for the data in this plot are given in the Appendix (Sec. C). We see that the hottest metal-poor RHB stars have ( NUV -V ) 0 colors similar to the halo RR Lyrae stars. Indeed, we find that some of the hottest metalpoor RHB candidates are actually RR Lyrae stars.
For T eff less than about 5200 K, most of these RHB stars are metal-rich ([Fe/H] > -0.8) and have larger ( NUV -V ) 0 than the metal-poor RHB, similar to the
Fig. 9.-( NUV -V ) 0 color versus spectroscopic effective temperature T eff for a selection of red horizontal branch (RHB) stars. Colors indicate metallicity: [Fe/H] > -0 80 are shown by red filled . circles, -0 8 . > [Fe/H] > -1 70 are blue open circles, and [Fe/H] . < -1 70 are green . filled circles. The two measurements for HD 119516 are joined by a dotted line. Four stars are optical variables and are indicated by encircled symbols (bottom left). The red and green horizontal lines show the mean values of( NUV -V ) 0 for disk and halo RRab variables (as shown in Fig. 6). The data and notes on individual stars are presented in Table A3 in the Appendix.
trend seen in the RRab stars. We do not find any metalrich RHB stars hotter than this limit. As we discuss in Appendix C, this is likely an observational selection effect; there are few constraints on hot metal-rich RHB stars at fainter magnitudes ( V > 10.5). We cannot rule out, however, that hot metal-rich RHB stars may be scarce in the field even though they are found in the metal-rich globular clusters like 47 Tuc.
The change in ( NUV -V ) 0 color with effective temperature is much larger compared with the change in optical color indices. Stars like BD 18 ◦ 2757 and BPS CS 22875-029 have similar metallicity but differ in T eff by ∼ 1250 K. Their difference in ( B -V ) 0 and ( J -K ) 0 are 0.32 and 0.15 mag, respectively, while their difference in ( NUV -V ) 0 is over 2 mag. This makes the ( NUV -V ) 0 color particularly sensitive for detecting the temperature changes that come from pulsations and, in particular, for detecting the onset of this instability as a function of other parameters (such as T eff ).
## 4.4. The Completeness of the Samples.
We explore the magnitude limits of our samples in Figure 10. Fig. 10(a) plots the distribution of the minimum V magnitudes for the 84 RRab variables in Table 1. There is a steep fall-off in the distribution at V ∼ 17.5. Fig. 10(b) and Fig. 10(c) plot the distribution of the faintest observed NUV of the RRab in Catalogs 2 and 1, respectively. There is a steep fall-off at about ( NUV ) = 21.0. These two magnitude limits are consistent with the 3 4 . ≤ ( NUV -V ) ≤ 3 8 mag RRab colors . observed by Wheatley et al. (2012); most of the RRab in Table 1 are also in this range.
A V =17.5 magnitude corresponds to a distance limit of about 25 kpc for our sample of RRab. Most of the stars in our sample are at a sufficiently high galactic latitude
Fig. 10.The distribution of the RRab in Table 3 versus their V magnitude at minimum light (panel a). The distribution versus ( NUV ) magnitude at minimum light for Catalog 2 (panel b) and Catalog 1 (panel c).
that the correction for interstellar extinction is small, E( B -V ) ≤ 0.04 for which E( NUV -V ) ≤ 0.23.
The mean ( NUV -V ) for RRc is somewhat smaller than that of the RRab, and they have lower amplitudes. Consequently, RRc stars will have a fainter V magnitude for a given NUV , and their identification will necessarily be less complete than for the RRab. The NUV light curve minima in the Gezari catalog are all fainter than 20.0 and so, neglecting extinction, the RRc will have V > 17 and will generally be less easy to detect with existing optical data without the use of new techniques.
The 17 RRc and 48 RRab corresponds to ∼ 0.40 vari- ables per square degree for Catalog 2. This is comparable to the 0.455 RR Lyrae per square degree found by Abbas et al. (2014) in a survey of somewhat greater depth (28 kpc) and which they find has a completeness of ∼ 50%. This suggests that GALEX has detected only half of the RRab to V =17.5, or else Catalog 2 is itself only about half complete for these variables.
Catalog 2 contains the variables whose NUV amplitudes are greater or equal to 0.6 mag. The survey covered 169 fields; each field (radius 0. ◦ 55) was visited from between 10 and 187 times. Table 4 shows that that variables were only discovered in about 20% of the fields when the number of visits was 20 or less but this percentage rose to about 40% when there were over 30 visits. The field centered on AE Aqr is also in the well studied SDSS Stripe 82; three RRc and 2 RRab were found in this field but three RRc of comparable brightness that lie within this field were missed. There were 31 visits to this field but the number of actual detections of the RRc variables ranged from 19 for a 13th mag star to 5 for one at 17th mag. Thus the completeness will vary from field to field depending on the number of visits. In the 19 fields that were visited more than 50 times, eleven RR Lyrae variables were found and only two were missed. It is therefore only in these fields (that comprise only 11% of the total) that the Catalog 2 survey is reasonably complete.
## 4.5. Contact Binaries.
Contact binaries can be confused with RRc variables that have half their periods; see discussions by Kinman & Brown (2010) and Drake et al. (2014). Table 2 in the appendix (Sec. B) gives the UV colors for a sample of bright contact binaries and these are plotted against period in Fig. 11.
Drake et al. (2014) use a variety of parameters to distinguish between RRc and contact binaries and conclude that misidentification is most likely to occur among the longer-period bluer contact binaries which are the brightest among such variables; misidentification is therefore not a serious problem for faint surveys. Misidentification apart, the cause of the spread in ( NUV -V ) 0 among contact binaries needs further investigation. It is unlikely to be caused by a spread in metallicity because the range of metallicity in these objects is small (Rucinski et al. 2013). The presence of companions is a more likely cause since it is known that multiplicity is very common among contact binaries (Pribulla & Rucinski 2008). Smith et al. (2014) consider the effect of companion stars on GALEX colors in their discussion of stars in the Kepler field. ( NUV -V ) 0 may therefore be a useful diagnostic in searching for such companion stars.
## 4.6. Uncertainties in determining Variable Type and Period.
Five of our new variables (Table 2) are also found in a new catalogue of periodic variables from the Catalina Surveys (Drake et al. 2014, hereafter D14). The first star is J022121.9-033040.0, for which D14 find the same variable type and period as we do. Thus we do not discuss this star further. For the other four stars, we obtain repeat periodograms using the Lomb-Scargle (LS) and Plavchan (PL) routines in the NASA Exoplanet Archive
TABLE 4
Discovery rate of RR Lyrae stars as a function of number of visits per field for Catalog 2.
| Visits per Field (1) | No. of Fields with this range of visits (2) | Number of Fields in which variables were found (3) |
|------------------------|-----------------------------------------------|------------------------------------------------------|
| 10 to 20 | 86 | 18 |
| 21 to 30 | 39 | 14 |
| 31 to 50 | 25 | 11 |
| > 51 | 19 | 8 |
Fig. 11.-( FUV -NUV ) 0 (upper panel) and ( NUV -V ) 0 (lower panel) plotted against the half-periods of a sample of bright contact binaries (black filled circles). Two of these objects (SW Ret and v1643 Sgr) that are shown encircled, are actually RRc variables. The blue horizontal lines in both figures show the location of RRc variables. Our program RRc variables are plotted as blue filled circles using the full periods in the lower figure.
Periodogram Service. Fig. 12 presents the light curves for these four variables. For each variable, the left-hand panel plots the data folded on our period, and the righthand panel plots the data folded on the D14 period.
With the exception of J154550.8-124936.8, identifying the correct period and type by visual inspection is difficult because of period alias uncertainties. The discrepant results described below illustrate the difficulty in identifying the nature of low-amplitude variables.
J022411.7-052501.6. We identify this as an Algol variable with a period of 0.8254050 days. D14 give the same type with a period of 0.583802 days. The LS routine gives neither of these periods. The PL routine gives both periods; the D14 period has more power in the PL routine and is likely correct.
J122837.9-064230.0. We identify this as an RRab with a period of 0.501648 days. D14 give the same type with a period of 0.333755 days, but call this period 'inexact.'
Fig. 12.Comparison of the light-curves of J022411.7-052501.6, J1222837.9-064230.0, J154550.8-124936.8 and 160905.7+534712.2 obtained in this paper (on the left) and those given by Drake et al. (2014) on the right.
The LS and PL routines give both periods, however our period has more power and is likely correct.
J154550.8-124936.8. We identify this as an RRc with a period of 0.293594 days. D14 give the same type with a period of 0.416068 days. Both the LS and PL routines give more power to our period than to the one at 0.416 days, and our period gives a better looking light curve (Fig. 12).
J160905.7+534712.2. We identify this as a δ Scutiwith a period of 0.1697210 days. D14 identify it with an EW with a period of 0.339438 days. The LP routine gives our period. The PL routine gives both periods with marginally more power to our period. However, the D14 identification is likely correct because the ( g -r ) = 0 32 . color of the variable is probably too red for a δ Scuti.
## 5. SUMMARY
We make as complete an identification as possible of unclassified variable stars in existing GALEX variable source catalogs. We present newly identified δ Scuti, RRc and RRab stars found in what we refer to as Catalog
1 (Welsh et al. 2005) and Catalog 2 (Wheatley et al., 2008). We also examine the sources identified either as 'stars' or RR Lyrae stars in the larger catalog of variable GALEX sources of Gezari et al. (2013).
We identify 8 δ Scuti stars, 17 RRc stars, 1 RRd star and 84 RRab stars in Catalogs 1 and 2. These 110 pulsating variables account for 22% of the sources in these Catalogs. Sixteen of these variables were not previously known; their ephemerides and light curves are given in Table 2 and Figures 1, 2, 3 and 4 respectively.
In Table 3 we present new classifications and period measurements for 28 of the 38 variable GALEX sources that Gezari et al (2013) identify as RR Lyrae stars: 16 are RRab stars, 4 are RRc stars, 6 are δ Scuti stars and 2 are eclipsing systems. We also classify 7 of the 17 GALEX sources that Gezari et al. identify as 'stars': 1 is a RRab star, 1 is a RRc star and 5 are eclipsing systems. The classification of these objects depends on our ability to recognize their variability type from optical data in the Catalina Surveys Data Release 2. This becomes difficult for low-amplitude variables with V > 17 5. . Thus classifications are less complete for the Gezari catalog because the catalog contains lower amplitude variables ( NUV amplitude ≥ 0.2 mag compared with ≥ 0.6 mag for Catalog 2). The light-curves of the variables in Table 2 from the Gezari Catalog are given in Figures 1, 2, 3 and 4.
( NUV -V ) 0 is more sensitive than optical colors to changes in T eff (such as those that occur in stellar pulsation). Four stars that have been described in the literature as RHB stars are actually RR Lyrae stars (e.g. BPS CS 22940-070 which is called non-variable by For et al., (2010)). A plot of ( NUV -V ) 0 against T eff shows that the metal-poor RHB stars merge into the RR Lyraes at T eff ∼ 6000 K which agrees with other estimates of the red edge of the instability gap. Existing surveys such as Afsar et al. (2012) do not show any metal-rich (disk) RHB stars hotter than T eff ∼ 5200 K whereas such stars are known in the metal-rich globular cluster 47 Tuc. We suggest that hotter metal-rich RHB stars could be found by deeper surveys.
We find a large range in both ( NUV -V ) 0 and ( FUV -NUV ) 0 colors in a sample of bright contact binaries that have twice the periods of RRc stars and which may therefore be confused with them. These UV colors do not allow a clear distinction between the two types, but may be of value in detecting companions to the contact binaries.
We find that the ( NUV -V ) 0 color of low metallicity halo RRab stars is smaller than that of the highermetallicity disk RRab stars. The NUV pulsation amplitude of RR Lyrae stars is roughly twice that of their V amplitude (as found earlier by Wheatley et al. (2012)). This allows us to detect low-amplitude variables such as J013642.0-062743.1 and J171251.8+185452.4 which have V amplitudes of 0.25 and 0.38 mag, respectively. A larger number might be found if the Catalogs were more complete. At a rough estimate, the overall completeness of Catalog 2 is similar to that of the catalog of Abbas et al. (2014) which they estimate as 50%. The completeness of Catalog 2 varies from field to field depending on the number of observations made per field.
We acknowledge use of the International Variable Star Index (VSX) database operated at AAVSO, Cambridge, MA, USA. Also the LINEAR survey (available on the SkyDOT website) and funded by NASA at MIT Lincoln Laboratory under Air Force Contract FA8721-05-0002. Also the Catalina Surveys (Data Release 2) that is supported by NASA under grant NNG05GF22G and NSF grants AST-0909182 and AST-1313422, We used the VizieR catalogue access tool, CDC, Strasbourg, France. We acknowledge using data from the SDSS funded by the Alfred P. Sloan Foundation, the Participating Institutions, the NSF and the US D.O.E.. We also used the MAST(Multimission Archive at the STSci which is operated for NASA by AURA), the SIMBAD database (operated at the CDS, Strasbourg, France), ADS (the NASA Astrophysics Data System), and the arXiv e-print server. We also used the NASA Exoplanet Archive which is operated by the California Institute of Technology under contract with NASA.
## REFERENCES
Abbas, M., Grebel, E., Martin, N., Burgett. W. et al., 2014,
Flesch, E. 2010, PASA, 27, 283
- MNRAS, 441, 1230 Afsar, M., Sneden, B., For, B.-Q. 2012, AJ, 144, 20 Avvakumova, E., Malkov, O., Kniazev, A. 2013, AN, 334, 860 Alfonso-Garzon,J., Domingo, A., Mas-Hesse, J. 2012, in 'Interacting Binaries to Exoplanets', I.A.U. Symposium No. 282, (eds. Mercedes T. Richards, & Ivan Hubeny), Cambridge University Press, p.484 Behr, B. 2003, ApJS, 149, 101 Brown, W., Beers, T., Wilhelm, R., Prieto, C. et al. 2008, AJ, 135, 564 Cacciari, C., Corwin, T., Carney, B., 2005, AJ, 129, 267 Christlieb, N., Scharck, T., Friebel, A., Beers, T. et al. 2008, A&A, 484, 721 Corben, P., Carter, B., Banfield, R., Harvey, G. 1972, MNASSA, 31, 7 Dambis. A., Berdnikov, L., Kniazev, A., Kravtsov, A. et al. 2013, MNRAS, 435, 3206 Drake, A., Catelan, S., Djorgovski, S., Torrealba, G. et al. 2013, ApJ, 763, 32 Drake, A., Graham, M., Djorgovski, S., Catelan, S. 2014, ApJS, 213, 9 Dvorak, S. 2004, IBVS, No. 5549
Ferro, A., Pena, J., Jaimes, R. 2013. R. Mx. A. A., 49. 43
- For, B-Q., Sneden, C., 2010, AJ, 140, 1694 Gezari, S., Matin, D., Forster, K., Neill, J. et al. 2013, ApJ, 766, 60 ( GALEX Time Domain Survey) Girard, T., van Altena, W., Zacharias, N., Casetti-Dinescu, D. et al. 2011. yCat. 1320. (SPM 4.0 Catalog) Hansen, C., Nordstrom, B., Bonifacio, P., Spite, M., et al. 2011, A&A, 527, A65 Ishigaki, M., Aoki, W., Chiba, M. 2013, ApJ, 771, 67 Keller, S., Murphy, S., Prior, S., Da Costa, G. et al. 2008, ApJ, 678, 851 Kinman, T., Salim, S., & Clewley, L. 2007, ApJL, 662. 111 Kinman, T., Morrison, H., Brown, W. 2009, AJ, 137, 3198 Kinman, T., Brown, W. 2010, AJ, 139, 2014 Kinemuchi, K., Smith, H., Wo´ zniak, P. & McKay, T. 2006, AJ, 132, 1202 Kraus, A., Craine, E., Giampapa, M., Scharlach, W. et al., 2007, AJ, 134, 1488 Li, Y., Wei, J-Y., He, X-T. 2011, Res. Astron. Astrophys., 11, 833 Malkov, O., Oblak, E., Snegereva, E., Torra, J. 2006, A&A, 446, 785
- Martin, D.C., Fanson, J., Schiminovich, D., Morrisey, P. et al. 2005, ApJL, 619, 1
Miceli. A., Rest, A., Stubbs, C., Hawley, S. et al. 2008, ApJ, 678, 865 McNamara, D. 1997, PASP, 109, 1221 Nichols, J., Henden, A., Huenemoerder, D., Lauer, J. 2010, ApJS, 188, 473 Norton, A., Wheatley, P., West, R., Haswell, C. et al. 2007, A&A, 467, 785 Palaversa, L., Ivezic, Eyer, L., Ruzdjak, D. et al. 2013, AJ, 146, 101 Peek. & Schminovich 2013, ApJ, 771, 68 Petersen, J., Hog, E. 1998, A&A, 331, 989 Pojmanski, G. 2002, Acta Astron., 52, 397 Poleski, R. 2014, PASP, 126, 509 Preston, G., Schectman, S., Beers, T. 1991, ApJ, 121, 375 Pribulla, T., Rucinski, S. 2008, 'Multiple Stars Across the H-R Diagram', ESO Astrophysics Symposia, Springer-Verlag, Berlin Heidelberg, p. 163 Rey, S-C., Rich, R., Sangmo, T., Suk-Yin, Y. et al. 2007, ApJS, 173, 643 Rucinski, S., Pribulla, T., Budaj, J. 2013, AJ, 146, 70 Preston, G., Sneden, C., Thompson, I., Shectman, S. et al. 2006, AJ, 132, 85 Samus, N., Durlevich, O. et al., 2012 (General Catalogue of Variable Stars)
- Schmidt, E. 2002, AJ, 123, 965
- Sesar, B., Ivezic, Z., Grammer, S., Morgan, et al., 2010, ApJ, 708. 717
Sesar, B., 2012, AJ, 144. 114 Sesar, B., Iveezic, Z., Sturat, J., Morgan, D., et al. 2013, AJ, 146. 21 Smith, M., Bianchi, L., Shiao, B. 2014, AJ, 147, 159 Suveges, M., Sesar, B., Veradi, M., Mowlavi, N. et al. 2012, MNRAS, 424, 2528 Szczgiel, D., Fabrycky, D. 2007, MNRAS, 377, 1263 van Leeuwen, F., 2007, A&A, 474, 653 Vivas, A., Zinn, R., Abad, C., et al., 2004, AJ, 127, 1158 Watson, C. 2006, JAVSO, 35, 318 (VSX Catalogue) Welsh, B., Wheatley, J., Heafield, K., Seibert, M. et al. 2005, AJ, 130, 825 (The GALEX Ultraviolet Variability Catalog) Wheatley. J., Welsh, B., Sallman, S., Siegmund, O. 2003, BAAS, 35, 1338 Wheatley. J., Welsh, B., Browne, S. et al. 2008, AJ, 136. 259 (The Second GALEX Ultraviolet Variability Catalog). Wheatley. J., Welsh, B.,2012, PASP, 124, 552 Wils, P., Lloyd, C., Bernhard, K. 2006, MNRAS, 368, 1757 Wils, P. 2010, AAVSO Variable Star Index (VSX Catalogue) Zinn, R., Horowitz, B., Vivas, A., Baltay, C. et al. 2014, ApJ, 781, 22
## APPENDIX
## A. HIGH-AMPLITUDE δ SCUTI STARS (HADS).
The High-Amplitude δ Scuti stars are radially pulsating stars with periods in the range 0.04 to 0.30 days that are largely monoperiodic. They show some overlap in period with the RRc variables but can usually be distinguished from them by having more asymmetric light curves. Table A1 gives the ( NUV -V ) 0 for a sample of these stars drawn from the compilations of McNamara (1997) and Suveges et al. (2012). The HADS are well separated from the RRc variables in the ( NUV -V ) 0 vs. Period plot (Fig. 6); this is clearly the case for SS Psc (which is marked in this figure). This is of interest because Ferro et al. (2013) were unable to classify this star from their Stromgren photometry. The revised Hipparcos parallax of SS Psc is 5.26 ± 1.99 (van Leeuwen, 2007) gives an M v of between +3.5 and +5.2 which is in the expected range for HADS (Peterson & Hog, 1998).
TABLE 1 ( NUV -V ) 0 for δ Scuti Stars.
| R.A. a (deg.) | Dec. b (deg) | Ref c | ( NUV - V ) 0 d (mag.) | Period e (days) | GCVS ID f |
|-----------------|-----------------|---------|--------------------------|---------------------|-------------|
| (1) | (2) | (3) | (4) | (5) | (6) |
| 144.2216 | +44.0668 | 1 | 3.330 | 0.0860100 | AE UMa |
| 231.0294 | +36.8670 | 1 | 3.118 | 0.1040900 | YZ Boo |
| 238.2908 | +06.0907 | 1 | 3.193 | 0.1891500 | CW Ser |
| 013.8256 | +23.1637 | 1 | 2.461 | 0.0786800 | GP And |
| 337.5121 | -08.1075 | 1 | 3.912 | 0.1683800 | GV Aqr |
| 247.8249 | +11.9979 | 1 | 3.392 | 0.1486300 | DY Her |
| 247.5682 | +16.9185 | 1 | 2.550 | 0.0946800 | v1116 Her |
| 357.1914 | -08.1457 | 1 | 3.767 | 0.1978230 | BS Aqr |
| 146.4452 | -12.9940 | 1 | 3.226 | 0.2233891 | VX Hya |
| 020.2182 | +21.7287 | 1 | 3.898 | 0.2877928 | SS Psc |
| 270.6644 | +62.7186 | 2 | 3.393 | 0.1969000 | v395 Dra |
| 278.0269 | +40.5991 | 2 | 2.790 | 0.1021400 | v593 Lyr |
| 240.7758 | +26.2396 | 2 | 4.357 | 0.1924460 | BY CrB |
| 237.4137 | -76.4219 | 2 | 2.182 | 0.0638630 | v360 Aps |
| 343.0172 | -0.8690 | 3 | 3.214 | 0.0495655 | |
| 320.2203 | +0.2755 | 3 | 2.067 | 0.0500164 | |
| 319.3871 | +0.8267 | 3 | 2.813 | 0.0493681 | |
| 318.9853 | -0.8594 | 3 | 2.472 | 0.0512525 | |
| 044.3292 | -0.6536 | 3 | 2.323 | 0.0482597 | |
| 037.4992 | +1.2482 | 3 | 2.380 | 0.0509957 | |
| 341.6031 | +1.1892 | 3 | 2.754 | 0.0502367 | |
| 338.5538 | +0.7835 | 3 | 2.794 | 0.0537832 | |
| 322.4356 | +1.0480 | 3 | 2.421 | 0.0596027 | |
| 320.6782 | +0.7816 | 3 | 2.386 | 0.0538881 | |
| 313.7465 | -0.1246 | 3 | 2.024 | 0.0515025 | |
| 312.4206 | -0.3539 | 3 | 2.749 | 0.0931453 | |
| 322.4695 | -1.1719 | 3 | 2.989 | 0.0805862 | |
| 321.1715 | +0.0444 | 3 | 2.451 | 0.0452487 | |
| 043.1488 | +0.2955 | 3 | 2.480 | 0.0550160 | |
| 355.3285 | +0.7152 | 3 | 3.247 | 0.1198865 | |
| 323.7998 | +0.6225 | 3 | 2.782 | 0.0872579 | |
| 313.1301 | -0.5333 | 3 | 2.554 | 0.0530346 | |
| 305.7112 | -0.1257 | 3 | 2.468 | 0.0521552 | |
| 320.1692 | +0.9076 | 3 | 1.698 | 0.0742401 | |
| 316.8586 | +1.1715 | 3 | 3.410 | 0.0678054 | |
| 310.7717 | +0.0367 | 3 | 2.386 | 0.0457611 | |
| 310.3194 | +0.5634 | 3 | 2.739 | 0.0663556 | |
| 357.0749 | +0.9069 | 3 | 2.562 | 0.0607696 | |
| 314.2421 | +1.0685 | 3 | 2.722 | 0.0702984 | |
| 304.6618 | +1.0123 | 3 | 3.566 | 0.0457270 | |
| 346.1327 | +0.4127 | 3 | 2.390 | 0.0565413 | |
| 002.3078 | -0.1846 | 3 | 2.426 | 0.0519909 | |
| 338.6537 | +0.3735 | 3 | 2.625 | 0.0530301 | |
| 320.8264 | +0.8867 -0.6974 | 3 3 | 2.186 | 0.0706932 0.0581344 | |
| 045.7345 | | | 2.804 | | |
References . - (1) McNamara (1997); (2) Samus et al. (2012), (General Catalogue of Variable Stars); (3) Suveges et al. (2012).
- a Right Ascension in decimal degrees.
- b Declination in decimal degrees.
- d Dereddened difference beteween mean NUV and V magnitudes.
- c Reference for identification as δ Scuti star.
- e Period in days.
- f Identification in General Catalogue of Variable Stars.
TABLE 2
## UV colors for bright Contact Binaries.
| R.A. a (deg.) (1) | Dec. b (deg) (2) | b c (deg.) (3) | ( NUV - V ) 0 d (mag.) (4) | ( FUV - NUV ) 0 e (mag) (5) | X-ray f (6) | P 0 . 5 g (days) (7) | GCVS ID h (8) |
|---------------------|--------------------|------------------|------------------------------|-------------------------------|---------------|------------------------|-----------------|
| 130.007 | 18.9998 | 32.17 | 4.72 | 4.83 | X | 0.1914 | TX Cnc |
| 150.7 | 1.0945 | 41.94 | 3.83 | 4.17 | ... | 0.2099 | Y Sex |
| 258.491 | 16.3502 | 28.73 | 4.48 | 3.73 | X | 0.2108 | AK Her |
| 77.8104 | -8.5569 | -26.13 | 4.98 | ... | X | 0.2117 | ER Ori |
| 141.671 | -13.7518 | 25.68 | 4.71 | ... | ... | 0.2248 | EZ Hya |
| 234.108 | 15.5317 | 50.1 | 2.95 | 3.67 | ... | 0.258 | CC Ser |
| 299.2 | 1.1 | -14 | 3.1 | 3.77 | ... | 0.2588 | v0724 Aql |
| 52.0838 | -64.9772 | -44.8 | 2.68 | 3.82 | ... | 0.2614 | SW Ret † |
| 126.135 | -16.4032 | 12.04 | 3.92 | 5.32 | ... | 0.2782 | AV Pup |
| 79.9754 | -35.9017 | -33.2 | 2.5 | 3.73 | ... | 0.2826 | RZ Col |
| 354.817 | -9.1534 | -65.2 | 2.58 | 2.67 | ... | 0.3065 | EK Aqr |
| 160.138 | 13.5669 | 56.59 | 3.23 | 3.75 | ... | 0.309 | UZ Leo |
| 180.265 | 13.0083 | 71.61 | 4.6 | 1.24 | ... | 0.3213 | AG Vir |
| 143.077 | -28.6278 | 16.61 | 4.51 | ... | ... | 0.3242 | S Ant |
| 277.42 | 31.0003 | 17.9 | 3.26 | ... | ... | 0.3266 | IY Lyr |
| 296.442 | -40.9383 | -27.3 | 2.94 | 3.07 | ... | 0.3398 | v1643 Sgr † |
| 212.364 | -15.5816 | 43.3 | 4.01 | 4.23 | ... | 0.373 | CX Vir |
| 119.546 | 72.7643 | 30.7 | 3.39 | 4.17 | ... | 0.3821 | CD Cam |
| 311.616 | -71.9495 | -34.5 | 3.98 | ... | ... | 0.3974 | MW Pav |
| 274.223 | 27.6628 | 19.3 | 3.07 | 4.06 | ... | 0.434 | MS Her |
| 252.38 | 47.108 | 39.9 | 3.92 | 2.59 | ... | 0.4386 | v921 Her |
Note . -† Both SW Ret and v1643 Sgr are RR Lyrae stars.
- a
- Right Ascension in decimal degrees.
- b Declination in decimal degrees.
- d Dereddened difference beteween mean NUV and V magnitudes.
- c Galactic Latitude.
- e Dereddened difference beteween mean FUV and NUV magnitudes.
- f Given as X-ray source by Flesch (2010).
- g Half of period in days.
- h Identification in General Catalogue of Variable Stars.
## B. CONTACT BINARIES
Table A2 gives the ( NUV -V ) 0 and ( FUV -NUV ) 0 for a sample of bright contact binaries with half-periods between 0.19 and 0.44 days (primarily taken from Malkov et al., 2006). We excluded variables at low galactic latitudes ( b | | < 14 ) in order ◦ to minimize errors in the corrections for extinction. The extinction correction used to derive ( FUV -NUV ) 0 is less than that for ( NUV -V ) 0 , but FUV is available for fewer stars. The data from Table A2 is plotted in Fig. 11.
The ( NUV -V ) 0 of the 17 RRc that we found in Catalogs 1 and 2 range from 2.65 to 3.17 ( σ = 0.14 mag) while in a sample of 50 RRc from the Abbas et al. catalog, the range is from 2.36 to 3.52 ( σ = 0.26 mag). The mean NUV in the former case were derived from multiple values of NUV whereas in the latter case only a single value was usually available; this may well explain the greater spread of ( NUV -V ) 0 found for the RRc in the Abbas et al. catalog.
Half the contact binaries in Table A2 have ( NUV -V ) 0 within the range that we find for the RRc. Two of these stars with low ( NUV -V ) 0 are actually RR Lyrae stars: SW Ret is an RRd (Szczgiel & Fabrycky 2007) and v1643 Sgr is an RRc (Dvorak 2004). The more recent catalog of Avvakumova et al. (2013) continues to identify v1643 Sgr as a contact binary. The mean ( FUV -NUV ) 0 of 16 RRc variables is +2.59 ± 0.14. In Table A2, EK Aqr, AG Vir, v941 Her and the RRc v1643 Sgr are within the range of ( FUV -NUV ) 0 occupied by RR Lyraes; AG Vir is known to have an infra-red excess (Nichols et al. 2010).
## C. RED HORIZONTAL BRANCH (RHB) STARS.
Fig. 9 (Sec. 4.3) shows the trend of ( NUV -V ) 0 with effective temperature (T eff ) for RHB stars in three metallicity ([Fe/H]) ranges. The data used in this plot are given in Table A3. We take all stars with the type RHB in Table 2 of Behr (2003), and we also take NSV 4256 (HD 82590), which Behr lists as RR?, because its T eff (6094 K) is near the red edge of the instability gap. The classification of NSV 4256 is uncertain but it is likely to be one of the lowest amplitude RR Lyrae stars known; both Corben et al. (1972) and Alfonzo-Garzon et al. (2012) give NSV 4256 a V amplitude of 0.06 mag. Further data were taken from: Preston et al. (2006), For & Sneden (2010) and Afsar et al. (2012). We expect that errors in T eff are about ± 150 K. To illustrate this uncertainty, we provide data for HD 119516 from two different sources in Table A3, and plot the points separately (joined by a dotted line) in Fig. 9.
Behr (2003) place the red edge of the instability strip at T eff =6000 K (as plotted in Fig. 9). Preston et al. (2006) and For & Sneden (2010) give 6310 ± 145 K and 5900 K, respectively, for this quantity. These temperatures are appropriate for the lowest metallicity stars (plotted in green in Fig 5.). Bearing in mind the uncertainties in T eff , these estimates for the red edge of the instability strip are in reasonable agreement with our data. We find that ( NUV -V ) 0 increases with increasing metallicity at a given T eff .
There is, however, a large gap in T eff between the hottest metal-rich RHB stars that we have found in the literature and the T eff of the metal-rich disk RR Lyrae stars. We estimate the latter to have T eff ≥ 6500 K based on the ( V -K ) of a small sample of these stars. The blue tip of red horizonal branch of the metal-rich globular cluster 47 Tuc (Salaris et al. 2007) has J -K ∼ 0.40 to 0.45, whereas the hottest of our metal-rich RHB stars has J -K = 0.53. We therefore suspect that the lack of hotter stars among our sample of metal-rich RHB stars is an observational selection effect. The faintest 15 of the 76 stars studied by Afsar et al. (2012) have V between 9.0 and 10.5, which for M = +0.8 corresponds to distances between 436 and 870 pc. v Thus, as they point out, their sample strongly favors thin disk stars. A search for thick disk RHB stars must be made at significantly fainter magnitudes.
TABLE 3 ( NUV -V ) 0 , [Fe/H] and T eff for a sample of RHB stars.
| Identification | ( NUV - V ) 0 a mag. | [Fe/H] | T eff deg K | Ref. | Notes |
|--------------------|------------------------|----------|---------------|--------|---------|
| (1) | (2) | (3) | (4) | (5) | (6) |
| HIP 62325 | 8.117 | -0.06 | 4710 | (6) | |
| BD 25 ◦ 2459 | 7.101 | -0.28 | 4743 | (1) | |
| HIP 115839 | 6.285 | -0.40 | 5000 | (6) | |
| BD 34 ◦ 2371 | 7.262 | -0.41 | 5005 | (1) | |
| HIP 3031 | 8.058 | -0.53 | 5020 | (6) | |
| BD 29 ◦ 2294 | 6.411 | -0.55 | 5132 | (1) | |
| BD 29 ◦ 2231 | 7.076 | -0.65 | 4756 | (1) | |
| HIP 45412 | 6.870 | -0.68 | 5080 | (6) | |
| BD 25 ◦ 2436 | 7.041 | -0.76 | 4847 | (1) | |
| BD 36 ◦ 2303 | 7.860 | -0.77 | 4705 | (1) | |
| HIP 5104 | 6.335 | -0.86 | 5100 | (6) | |
| HD 110930 | 6.854 | -0.94 | 4934 | (1) | |
| HIP 4960 | 5.516 | -1.02 | 5260 | (6) | |
| HD 112030 | 7.111 | -1.12 | 4699 | (1) | |
| HD 202573 | 6.706 | -1.21 | 4853 | (1) | |
| BD 27 ◦ 2057 | 7.246 | -1.25 | 4695 | (1) | |
| HD 6229 | 5.516 | -1.35 | 5200 | (1) | |
| HD 97 | 5.678 | -1.42 | 5272 | (1) | |
| BPS CS 22940-070 | 3.024 | -1.47 | 6100 | (2) | (1) |
| HD 82590 | 3.893 | -1.50 | 6094 | (1) | (2) |
| HD 105546 | 5.437 | -1.67 | 5299 | (1) | |
| HD 63791 | 6.735 | -1.72 | 4954 | (1) | |
| BD 18 ◦ 2890 | 5.334 | -1.78 | 5347 | (1) | |
| HD 119516 | 4.224 | -1.92 | 5689 | (1)(4) | |
| BD 10 ◦ 2495 | 5.040 | -2.07 | 5275 | (1) | |
| BD 17 ◦ 3248 | 4.480 | -2.08 | 5398 | (1) | |
| BPS CS 22886-043 | 3.631 | -2.15 | 6000 | (2) | |
| HD 119516 | 4.233 | -2.16 | 5400 | (3) | |
| BPS CS 22888-047 | 3.487 | -2.35 | 6000 | (2) | (3) |
| BD 18 ◦ 2757 | 5.527 | -2.43 | 4741 | (1) | |
| BPS CS 22882-001 | 3.483 | -2.54 | 5950 | (3) | |
| BPS CS 22875-029 | 3.391 | -2.66 | 6000 | (3) | |
| BPS CS 22191-029 | 3.417 | -2.72 | 6000 | (3) | |
| BPS CS 22881.039 | 3.178 | -2.73 | 6100 | (3) | (4) |
| BPS CS 22186-005 | 3.318 | -2.77 | 6200 | (3) | (5) |
| BPS CS 30317-056 ; | 3.441 | -2.85 | 6000 | (5) | (6) |
References . - (1) Behr (2003); (2) Preston et al. (2006); (3) For & Sneden (2010). (4) Ishigaki et al. (2013); (5) Hansen et al. (2011). (6) Afsar et al. (2012).
Note . - o-Garzon et al. , 2012); type uncertain. (1) RRab variable; Period = 0.5103 days; V amp ∼ 0.9 mag. (this paper) (2) NSV 4526; V amp ∼ 0.06 mag. (Corben et al. 1972; Alfonz (3) [C/Fe] = 1.4 (Christlieb et al., 2008). (4) RRab variable, Period = 0.66876 days; V amp ∼ 0.8 mag. (Hansen et al., 2011); [C/Fe] = 2.0 (Christlieb et al., 2008) (5) [C/Fe] = +1.0 (Christlieb et al., 2008) 6 RRab variable, Period = 0.74851 days; V amp ∼ 0.35 mag. (Hansen et al., 2011)
a Dereddened difference beteween mean NUV and V magnitudes.
|
10.1088/0004-6256/148/6/121
|
[
"T. D. Kinman",
"Warren R. Brown"
] |
2014-08-04T20:25:35+00:00
|
2014-08-04T20:25:35+00:00
|
[
"astro-ph.SR"
] |
The Identification of RR Lyrae and Delta Scuti Stars from Variable GALEX Ultraviolet Sources
|
We identify the RR Lyrae and delta Scuti (DSCT) stars in three catalogs of
GALEX variable sources. The NUV amplitude of RR Lyrae stars is about twice that
in V-band, so we find a larger percentage of low amplitude variables than
catalogs such as Abbas et al (2014). Interestingly, the (NUV-V)_0 color is
sensitive to metallicity and can be used to distinguish between variables of
the same period but differing [Fe/H]. This color is also more sensitive to
T_eff than optical colors and can be used to identify the red edge of the
instability gap. We find 8 DSCT stars, 17 RRc stars, 1 RRd star and 84 RRab
stars in the GALEX variable catalogs of Welsh et al (2005) and Wheatley et al
(2008). We also classify 6 DSCT stars, 5 RRc stars and 18 RRab stars among the
55 variable GALEX sources identified as "stars" or RR Lyraes in the catalog of
Gezari et al (2013). We provide ephemerides and light curves for the 26
variables that were not previously known.
|
1408.0809v1
|
## EF+EX Forest Algebras
Andreas Krebs, University of T¨bingen u Howard Straubing, Boston College
August 6, 2014
## Abstract
We examine languages of unranked forests definable using the temporal operators EF and EX. We characterize the languages definable in this logic, and various fragments thereof, using the syntactic forest algebras introduced by Bojanczyk and Walukiewicz. Our algebraic characterizations yield efficient algorithms for deciding when a given language of forests is definable in this logic. The proofs are based on understanding the wreath product closures of a few small algebras, for which we introduce a general ideal theory for forest algebras. This combines ideas from the work of Bojanczyk and Walukiewicz for the analogous logics on binary trees and from early work of Stiffler on wreath product of finite semigroups.
## 1 Overview
Understanding the expressive power of temporal and first-order logic on trees is important in several areas of computer science, for example in formal verification. Using algebraic methods, in particular, finite monoids, to understand the power of subclasses of the regular languages of finite words has proven to be extremely successful, especially in the characterization of regular languages definable in various fragments of first-order and temporal logics ([CPP93, TW96, Str94]). Here we are interested in sets of of finite trees (or, more precisely, sets of finite forests), where the analogous algebraic structures are forest algebras.
Bojanczyk et. al. [BW08, BSW12] introduced forest algebras, and underscored the importance of the wreath product decomposition theory of these algebras in the study of the expressive power of temporal and first-order logics on finite unranked trees. For languages inside of CTL the associated forest algebras can be built completely via the wreath product of copies of the forest algebra
$$\mathcal { U } _ { 2 } = ( \{ 0, \infty \}, \{ 1, 0, c _ { 0 } \} ),$$
where the vertical element 0 is the constant map to ∞ , and the vertical element c 0 is the constant map to 0 ([BSW12]). The problem of effectively characterizing the wreath product closure of U 2 is thus an important open problem, equivalent
to characterization of CTL . Note that if one strips away the additive structure of U 2 , the wreath product closure is the family of all finite aperiodic semigroups (the Krohn-Rhodes Theorem). Forest algebras have been successfully applied to the obtain characterization of other logics on trees; see, for example [BSS12, BS09].
Here we study in detail the wreath product closures of proper subalgebras of U 2 . In one sense, this generalizes early work of Stiffler [Sti73], who carried out an analogous program for wreath products of semigroups. Along the way, we develop the outlines of a general ideal theory for forest algebras, which we believe will be useful in subsequent work. After developing the algebraic theory, we give an application to logic. We obtain a characterization of the languages of unranked forests definable using the temporal operators EF and EX . This extends the work of Bojanczyk and Walukiewicz [BW06], who obtain a similar characterization for the analogous logics on binary trees. Our proof, however, which proceeds entirely from the algebraic analysis, is completely different. Similar results, again for the case of trees of bounded rank, appear in Esik [ ´ si05]. ´ E
The paper is structured in the following way. First we introduce forest algebras and introduce some general theory (see Section 2). After giving the connections between sublogics of CTL in Section 3, we examine in detail algebraic operations corresponding to closure under the EF quantifier (Section 4), the EX quantifier (Section 5) and then both quantifiers (Section 6). We conclude with our characterization and decidability results in Section 7).
## 2 Forest Algebras
## 2.1 Preliminaries
We refer the reader to [BW08, BSW12] for the definitions of abstract forest algebra, free forest algebra, and syntactic forest algebra. We denote the free forest algebra over a finite alphabet A by A ∆ = ( H ,V A A ) , where H A denotes the monoid of forests over A, with concatenation as the operation, and V A denotes the monoid of contexts over A, with composition as the operation. A subset L of H A is called a forest language over A. We denote its syntactic forest algebra by ( H ,V L L ) , and its syntactic morphism by
$$\mu _ { L } \colon A ^ { \Delta } \rightarrow ( H _ { L }, V _ { L } ).$$
For the most part, our principal objects of study are not the forest algebras themselves, but homomorphisms
$$\alpha \colon A ^ { \Delta } \rightarrow ( H, V ).$$
It is important to bear in mind that each such homomorphism is actually a pair of monoid homomorphisms, one mapping H A to H and the other mapping V A to V. It should usually be clear from the context which of the two component homomorphisms we mean, and thus we denote them both by α. The 'freeness'
of A ∆ is the fact that a homomorphism α into ( H,V ) is completely determined by giving its value, in V, at each a ∈ A.
A homomorphism α as above recognizes a language L ⊆ H A if there exists X ⊆ H such that α -1 ( X ) = L.
If α : A ∆ → ( H,V ) and β : A ∆ → ( H , V ′ ′ ) , are homomorphisms, we say that β factors through α if for all s, s ′ ∈ H , α s A ( ) = α s ( ′ ) implies β s ( ) = β s ( ′ ) . This is equivalent to the existence of a homomorphism ρ from the image of α into ( H , V ′ ′ ) such that β = ρα. A homomorphism α recognizes L ⊆ H A if and only if µ L factors through α. ( [BW08]).
In the course of the paper we will see several congruences defined on free forest algebras. Such a congruence is determined by an equivalence relation ∼ on H A such that for any p ∈ V A , s ∼ s ′ implies ps ∼ ps . ′ This gives a welldefined action of V A on the set of ∼ -classes of H . A We define an equivalence relation (also denoted ∼ ) on V A by setting p ∼ p ′ if for all s ∈ H , ps A ∼ p s. ′ The result is a quotient forest algebra ( H / A ∼ , V A / ∼ ) . In order to prove that an equivalence relation ∼ on H A is a congruence, it is sufficient to verify that s ∼ s ′ implies s + t ∼ s + t ′ and as ∼ as ′ for all s, s , t ′ ∈ H A and a ∈ A.
## 2.2 Horizontally idempotent and commutative algebras
We now introduce an important restriction. Throughout the rest of the paper, we will assume that all of our finite forest algebras ( H,V ) have H idempotent and commutative; that is h + h = h ′ + h and h + h = h for all h, h ′ ∈ H. This is a natural restriction when talking about classes of forest algebras arising in temporal logics, which is the principal application motivating this study.
When H is horizontally idempotent and commutative, the sum of all its elements is an absorbing element for the monoid. While an absorbing element in a monoid is ordinarily written 0, since we use additive notation for H, its identity is denoted 0, and accordingly we denote the absorbing element, which is necessarily unique, by ∞ .
We say that two forests s , s 1 2 ∈ H A are idempotent-and-commutative equivalent if s can be transformed into t by a sequence of operations of the following three types: (i) interchange the order of two adjacent subtrees (that is, if s = p t ( 1 + t 2 ) for some context p and trees t 1 , t 2 , then we transform s to p t ( 2 + t 1 )); (ii) replace a subtree t by two adjacent copies (that is, transform pt to p t ( + )); t (iii) replace two identical adjacent subtrees by a single copy (transform p t ( + ) to t pt ). Since operations (ii) and (iii) are inverses of one another, and operation (i) is its own inverse, this is indeed an equivalence relation.
We have the following obvious lemma:
Lemma 1. Let α : A ∆ → ( H,V ) be a homomorphism, where H is horizontally idempotent and commutative. If s, t ∈ H A are idempotent-and-commutative equivalent, then α s ( ) = α t . ( )
Proof. By idempotence and commutativity of H, each of the three operation types used to transform s into t preserves the value under α.
There is a smallest nontrivial idempotent and commutative forest algebra, U 1 = ( { 0 , ∞} { , 1 0 , } ) . The horizontal and vertical monoids of U 1 are isomorphic, but we use different names for the elements because of the additive notation for the operation in one of these monoids, and multiplicative notation in the other. We have not completely specified how the vertical monoid acts on the horizontal monoid-this is done by setting 0 · x = ∞ for x ∈ { 0 , ∞} .
## 2.3 1-definiteness
In Section 5 we will discuss in detail the notion of definiteness in forest algebras; for this preliminary section, we will only need to consider a special case. A forest algebra homomorphism α : A ∆ → ( H,V ) is said to be 1-definite if for s ∈ H , A the value of α s ( ) depends only on the set of labels of the root nodes of s. We define an equivalence relation ∼ 1 on H A by setting s ∼ 1 s ′ if and only if the sets of labels of root nodes of s and s ′ are equal. This defines a congruence on A . ∆ We denote the homomorphism from A ∆ onto the quotient under ∼ 1 by α A, 1 . It is easy to show that a homomorphism α : A ∆ → ( H,V ) is 1-definite if and only if it factors through α A, 1 .
## 2.4 Wreath Products
We summarize the discussion of wreath products given in [BSW12]. The wreath product of two forest algebras ( H ,V 1 1 ) , ( H ,V 2 2 ) is
$$( H _ { 1 }, V _ { 1 } ) \circ ( H _ { 2 }, V _ { 2 } ) = ( H _ { 1 } \times H _ { 2 }, V _ { 1 } \times V _ { 2 } ^ { H _ { 1 } } ),$$
where the monoid structure of H 1 × H 2 is the ordinary direct product, and the action is given by
$$( v _ { 1 }, f ) ( h _ { 1 }, h _ { 2 } ) = ( v _ { 1 } h _ { 1 }, f ( h _ { 1 } ) h _ { 2 } ),$$
for all h 1 ∈ H , h 1 2 ∈ H , v 2 1 ∈ V , 1 and f : H 1 → V . 2 It is straightforward to verify that the resulting structure satisfies the axioms for a forest algebra. Note that if one forgets about the monoid structure on H 1 and H , 2 this is just the ordinary wreath product of left transformation monoids. Because we use left actions rather than the right actions that are traditional in the study of monoid decompositions, we reverse the usual order of the factors. The projection maps
↦
$$\pi \, \colon ( h _ { 1 }, h _ { 2 } ) \mapsto h _ { 1 }, ( v, f ) \mapsto v,$$
↦
define a homomorphism from the wreath product onto the left-hand factor.
We will view wreath products through the lens of homomorphisms from the free forest algebra. Given such a homomorphism
$$\gamma \colon A ^ { \Delta } \rightarrow ( H _ { 1 }, V _ { 1 } ) \circ ( H _ { 2 }, V _ { 2 } )$$
we can write, for each a ∈ A,
$$\gamma ( a ) = ( v _ { a }, f _ { a } ).$$
This gives rise to a pair of homomorphisms
$$\alpha \colon A ^ { \Delta } \rightarrow ( H _ { 1 }, V _ { 1 } ), \beta \colon ( A \times H _ { 1 } ) ^ { \Delta } \rightarrow ( H _ { 2 }, V _ { 2 } ),$$
where α a ( ) = v , a and β a, h ( ) = f a ( h . ) We write γ = α ⊗ β. Note that α = πγ, where π is the projection onto the left-hand factor. Conversely, any pair of homomorphisms α and β as above gives rise to a homomorphism α ⊗ β into the wreath product. We then have, for any s ∈ H , A
$$\alpha \otimes \beta ( s ) = ( \alpha ( s ), \beta ( s ^ { \alpha } ) ),$$
where s α ∈ H A × H 1 is obtained from s through a relabeling process: if a node x of s is originally labeled a ∈ A, and the tree rooted at x is at, where t ∈ H , A then the label of the same node in s α is ( a, α t ( )) .
The wreath product is an associative operation on forest algebras. Given forest algebras ( H , V i i ) , i = 1 , . . . , r, and homomorphisms
$$\alpha _ { 1 } \colon A ^ { \Delta } \rightarrow ( H _ { 1 }, V _ { 1 } ),$$
$$\alpha _ { i } \colon ( A \times H _ { 1 } \times \cdots H _ { i - 1 } ) ^ { \Delta } \to ( H _ { i }, V _ { i } ),$$
for i = 2 , . . . , r, we can form the homomorphism
$$\alpha _ { 1 } \otimes \cdots \otimes \alpha _ { r } \, \colon A ^ { \Delta } \to ( H _ { 1 }, V _ { 1 } ) \circ \cdots \circ ( H _ { r }, V _ { r } ).$$
A homomorphism
$$\alpha \, \colon A ^ { \Delta } \rightarrow ( H _ {, } V _ { 1 } ) \times ( H _ { 2 }, V _ { 2 } )$$
into a direct product factors through the wreath product in a trivial way: Let α , α 1 2 be the two component homomorphisms, and set β a, h ( ) = α 2 ( a ) for all a ∈ A,h ∈ H . 1 Then α factors through α 1 ⊗ β.
## 2.5 Reachability
Let ( H,V ) be a finite forest algebra. For h, h ′ ∈ H we write h ≤ h ′ if h = vh ′ for some v ∈ V, and say that h is reachable form h . ′ This gives a preorder on H. We set h ∼ = h ′ if both h ≤ h ′ and h ′ ≤ h. An equivalence class of = is ∼ called a reachability class . The preorder consequently results in a partial order on the set of reachability classes of H. We always have h + h ′ ≤ h, because h + h ′ = (1 + h ′ ) h. If h ∈ H and Γ is a reachability class of H then we write, for example, h ≥ Γ to mean that Γ ≤ Γ ′ , where Γ ′ is the class of h.
A reachability ideal in ( H,V ) is a subset I of H such that h ∈ I and h ′ ≤ h implies h ′ ∈ I. If we have a homomorphism
$$\alpha \colon A ^ { \Delta } \rightarrow ( H, V )$$
and a reachability ideal I ⊆ H, we define an equivalence relation ∼ I on H A by setting s ∼ I s ′ if α s ( ) = α s ( ′ ) / I, ∈ or if α s , α s ( ) ( ′ ) ∈ I. Easily s ∼ I s ′ implies
ps ∼ I ps ′ for any p ∈ V A . We thus obtain a homomorphism onto the quotient algebra
$$\alpha _ { I } \colon A ^ { \Delta } \rightarrow ( H / \sim _ { I }, V / \sim _ { I } )$$
which factors through α. Note that I is, in particular, a two-sided ideal in the monoid H, and H/ ∼ I is identical to the usual quotient monoid H/I = ( H -I ) ∪ {∞} . We will thus use the notation ( H/I,V/I ) for the quotient algebra, instead of ( H/ ∼ I , V/ ∼ I ) . If Γ ⊆ H is a reachability class, then both
/negationslash
$$I _ { \Gamma } = \{ h \in H \, \colon h \gtrsim \Gamma \} \text{ and } I _ { \geq \Gamma } = \{ h \in H \, \colon h \gtrsim \Gamma \}$$
/negationslash are reachability ideals. We denote the associated quotients and projection homomorphisms by ( H ,V Γ Γ ) , α , Γ ( H ≥ Γ , V ≥ Γ ), α ≥ Γ .
Given the restriction that H is idempotent and commutative, the absorbing element ∞ is reachable from every element. The reachability class of ∞ is accordingly the unique minimal class, which we denote Γ min . A reachability class Γ is subminimal if Γ min < Γ , but there is no class Λ with Γ min < Λ < Γ . The following lemma will be used several times.
Lemma 2. Let α : A ∆ → ( H,V , ) and let Γ 1 , . . . , Γ r be the subminimal reachability classes of ( H,V . ) Then
$$\alpha _ { \Gamma _ { \min } } \colon A ^ { \Delta } \rightarrow ( H _ { \Gamma _ { \min } }, V _ { \Gamma _ { \min } } )$$
factors through the direct product
$$\left ( \prod _ { j = 1 } ^ { r } \alpha _ { \geq \Gamma _ { j } } \right ) \colon A ^ { \Delta } \to \prod _ { j = 1 } ^ { r } ( H _ { \geq \Gamma _ { j } }, V _ { \geq \Gamma _ { j } } ).$$
Further each of the algebras ( H ≥ Γ j , V ≥ Γ j ) has a unique subminimal reachability class.
/negationslash
/negationslash
Proof. Choose s, s ′ ∈ H A such that α ≥ Γ j ( s ) = α ≥ Γ j ( s ′ ) for all j = 1 , . . . , r. If α s ( ) > Γ min , then α s ( ) ≥ Γ j for some j, and thus α s ( ) = α s ( ′ ) , so in particular α Γ min ( s ) = α Γ min ( s ′ ) . If α s ( ) > Γ min , then α s ( ) ∈ Γ min , by minimality. Thus every α ≥ Γ j ( s ) is the absorbing element ∞ of the quotient algebra, so the same is true for α ≥ Γ j ( s ′ ) . Thus α s ( ′ ) ≥ Γ j for all j, so α s ( ′ ) ∈ Γ min , and α Γ min ( s ) = α Γ min ( s ′ ) . This proves the claim about factorization of the homomorphisms.
Next, for the claim about the subminimal classes of ( H ≥ Γ j , V ≥ Γ j ) , observe that the reachability classes of this algebra are just the reachability classes of ( H,V ) that are greater than or equal to Γ j , along with the minimal class {∞} .
We will also need the following lemma, which concerns the behavior of reachability classes under homomorphisms.
Lemma 3. Let β : ( H ,V 1 1 ) → ( H ,V 2 2 ) be a homomorphism of finite forest algebras. Let Λ ⊆ H 1 be a reachability class. There is a reachability class Γ of ( H ,V 2 2 ) such that β (Λ) ⊆ Γ . If Λ is a minimal class of ( H ,V 1 1 ) satisfying β (Λ) ⊆ Γ , and β is onto, then β (Λ) = Γ . If, further, H 2 is idempotent and commutative, then there is only one such minimal class Λ .
Proof. Let h , h 1 ′ 1 ∈ Λ and let , h 2 = β h ( 1 ) , h ′ 2 = β h ( ′ 1 ) . To prove the first claim, we must show h 2 ∼ = h . ′ 2 There exist v, v ′ ∈ V 1 such that h 1 = v h , h ′ ′ 1 ′ 1 = vh . 1 We then have h 2 = β v ( ′ ) h ′ 2 and h ′ 2 = β v ( ′ ) h , 2 which gives the result.
Now suppose Γ is the class of ( H ,V 2 2 ) containing β (Λ) , and that Λ is a minimal class in the preimage of Γ . Let h ∈ Λ and let , h ′ ∈ Γ We need to show . h ′ ∈ β (Λ) . We have v, v ′ ∈ V 2 such that β h ( ) = v h , ′ ′ h ′ = vβ h . ( ) Since β is onto, there are elements u, u ′ ∈ V 1 with β u ( ) = v, β u ( ′ ) = v . ′ We then have β h ( ) = v vβ h ′ ( ) = β u uh , ( ′ ) but this means the class of u uh ′ maps into Γ . By minimality u uh ′ ∈ Λ , and since , u uh ′ ≤ uh ≤ h, uh ∈ Λ Thus . h ′ = β uh ( ) ∈ β (Λ) .
For the last claim, suppose to the contrary that h, h ′ ∈ V 1 are both in minimal classes mapping into Λ , but are not in the same class. By idempotence, β h ( + h ′ ) = β h ( ) ∈ Γ . We thus have h + h ′ ≤ h, so by minimality h + h ′ ∼ = h. Likewise, h + h ′ ∼ = h , ′ so h ∼ = h , ′ a contradiction.
## 3 Connections to Logic
## 3.1 Temporal logics for forests
We give a description of the temporal operators EF and EX . Our approach closely follows the one given in [BSW12].
We describe the syntax and semantics of our formulas, given a fixed finite alphabet A. One complication, which does not seem to be avoidable, is that we need to treat trees and forests somewhat differently. Thus we define both tree formulas, and a proper subset of these called forest formulas, and give different semantics depending whether we are interpreting a formula in a tree or in a forest.
- · T is a forest formula
- · a is a tree formula, for each a ∈ A.
- · every forest formula is a tree formula
- · both the class of tree formulas and the class of forest formulas are closed under boolean operations
- · if φ is a tree formula, then EF φ, EX φ are forest formulas
The semantics are similarly defined by mutual recursion. There are two satisfaction relations, one for trees satisfying tree formulas, the other for forests satisfying forest formulas.
- · s | = f T for every forest s ∈ H . A
- · as | = t a for every forest s ∈ H . A
- · if φ is a forest formula, then as | = t φ if and only if s | = f φ.
- · boolean operations have their usual interpretation
- · s | = f EF φ iff s x | = t φ, where s x denotes the tree rooted at some node x of s.
- · s | = f EX φ iff s x | = t φ, where x is a root node of s.
We'll call this logic EF + EX , and denote by EF , EX the fragments in which only one of the two operators is used.
Intuitively, when we interpret formulas in trees, EF φ means 'at some time in the future φ ' and EX φ means 'at some next time φ '. When we interpret such formulas in forests, we are in a sense treating the forest as though it were a tree with a phantom root node. Observe that if a ∈ A, we do not interpret the formula a in forests at all. Thus a formula can have different interpretations depending on whether we view it as a tree or a forest formula. For example, as a forest formula EX a means 'there is a root node labeled a ' while as a tree formula it means 'some child of the root is labeled a '.
We are primarily concerned with the forest satisfaction relation, and so we will usually drop the subscript on = | , and assume that = | f is intended. If φ is a forest formula, then we denote by L φ the set of all s ∈ H A such that s | = φ. L φ is the language defined by φ.
Example 4. Consider the following property of forests over { a, b } : There is a tree component containing only a 's, and another tree component that contains at least one b. Now consider the set L of forests s that either have this property, or in which for some node x, the forest of strict descendants of x has the property. The property itself is defined by the forest formula
$$\psi \, \colon \, E X ( a \wedge \neg E F b ) \wedge E X ( b \vee E F b )$$
and L is defined by
$$\psi \vee \text{EF} \psi.$$
In Example 8, we discuss the syntactic forest algebra of L.
## 3.2 Correspondence of operators with wreath products
The principal result of this paper, Theorem 20, is the algebraic characterization of the forest languages using the operators EF and EX , either separately or in combination. It will require some algebraic preparation, in Sections 4, 5 and 6 before we can give the precise statement of this theorem. The bridge between the logic and the algebra is provided by the next two propositions.
Let φ be a tree formula. Then φ can be written as a disjunction
$$\bigvee _ { a \in A } ( a \wedge \psi _ { a } ),$$
where each ψ a is a forest formula. Let Ψ = { ψ a : a ∈ A . } We'll call Ψ the set of forest formulas of φ. We say that a homomorphism
$$\beta \colon A ^ { \Delta } \rightarrow ( H, V )$$
recognizes Ψ if the value of β s ( ) determines exactly which formulas of Ψ are satisfied by s. To construct such a homomorphism, we can take the direct product of the syntactic algebras of L ψ for ψ ∈ Ψ and set , β to be the product of the syntactic morphisms.
The following theorem, adapted from [BSW12], gives the connection between the EF operator and wreath products with U 1 :
Proposition 5. (a) Suppose that φ is a tree formula, Ψ is the set of forest formulas of φ, and that Ψ is recognized by
$$\alpha \colon A ^ { \Delta } \rightarrow ( H, V ).$$
Then
$$E F \phi$$
is recognized by a homomorphism
$$\beta \colon A ^ { \Delta } \rightarrow ( H, V ) \circ \mathcal { U } _ { 1 },$$
where πβ = α.
(b) Suppose that L ⊆ H A is recognized by a homomorphism
$$\beta \colon A ^ { \Delta } \rightarrow ( H, V ) \circ \mathcal { U } _ { 1 }.$$
Then L is a boolean combination of languages of the form EF ( a ∧ φ , ) where L φ is recognized by πβ.
Here we prove an analogous result for the temporal operator EX .
Proposition 6. (a) Suppose that φ is a tree formula, Ψ is the set of forest formulas of φ, and that Ψ is recognized by
$$\alpha \colon A ^ { \Delta } \rightarrow ( H, V ).$$
Then
$$E X \phi$$
is recognized by a homomorphism
$$\alpha \otimes \beta \colon A ^ { \Delta } \rightarrow ( H, V ) \circ ( H ^ { \prime }, V ^ { \prime } ),$$
where β : ( A × H ) ∆ → ( H , V ′ ′ ) is 1-definite.
(b) Suppose that L ⊆ H A is recognized by a homomorphism
$$\alpha \otimes \beta \colon A ^ { \Delta } \rightarrow ( H, V ) \circ ( H ^ { \prime }, V ^ { \prime } ),$$
Suppose further that every language recognized by α is defined by a formula in some set Ψ of formulas. If
$$\beta \colon ( A \times H ) ^ { \Delta } \rightarrow ( H ^ { \prime }, V ^ { \prime } )$$
is 1-definite, then L is a boolean combination of languages of the form L ψ and EX ( a ∧ ψ , ) where ψ ∈ Ψ .
Proof. (a) Let φ be a tree formula. Again, we can write φ as a disjunction of formulas of the form a ∧ ψ, where ψ is a forest formula of φ. Since EX commutes with disjunction, we can write
$$\mathrm E X \phi = \bigvee _ { j = 1 } ^ { m } \mathrm E X ( a _ { j } \wedge \psi _ { j } ),$$
where each ψ j is a forest formula of φ. We now set
$$\beta = \alpha _ { A \times H, 1 } \colon ( A \times H ) ^ { \Delta } / \sim _ { 1 },$$
It follows that if s = a s ′ 1 1 + · · · a ′ n s n , then
$$\alpha \otimes \beta ( s ) = ( \alpha ( s ), \{ ( a ^ { \prime } _ { i }, \alpha ( s _ { i } ) ) \, \colon 1 \leq i \leq n \} ).$$
Thus s | = EX φ if and only if the second component of α ⊗ β s ( ) contains a pair ( a , h ′ ) , where for some 1 ≤ j ≤ m, a ′ = a , j and forests mapping to h under α satisfy ψ . j So α ⊗ β recognizes EX φ .
(b) For the converse, suppose L is recognized by a homomorphism α ⊗ β as described. Let s ∈ H , A and write s = a s 1 1 + · · · + a n s n . Then α ⊗ β s ( ) = ( α s , h ( ) ′ ) . By 1-definiteness of β, the value of h ′ is completely determined by the set of pairs { ( a , α s i ( i )) : 1 ≤ i ≤ n . } The first component of α ⊗ β s ( ) is h ∈ H, if and only if s ∈ α -1 ( h ) which is defined by a formula ψ h ∈ Ψ The . second component of contains the element ( a, h ) if and only if s | = EX ( a ∧ ψ h ) . Thus L is a boolean combination of sets of the required form.
## 4 EF -algebras
Following [BSW12], we define:
Definition 1. A finite forest algebra ( H,V ) is an EF -algebra if it satisfies the identities
$$h + h ^ { \prime } = h ^ { \prime } + h, v h + h = v h$$
for all h, h ′ ∈ H and v ∈ V. The second identity with v = 1 gives h + h = h. Thus every EF -algebra is horizontally idempotent and commutative.
The following result is proved in [BSW12], and is the key element in the characterization of languages definable in one of the temporal logics we consider in Section 3. We will give a new proof, as it provides a good first illustration of how we use the reachability ideal theory introduced above in decomposition arguments.
## Theorem 7. Let
$$\alpha \colon A ^ { \Delta } \rightarrow ( H, V )$$
be a homomorphism onto a forest algebra. ( H,V ) is an EF -algebra if and only if α factors through a homomorphism
$$\beta \colon A ^ { \Delta } \to \mathcal { U } _ { 1 } \circ \dots \circ \mathcal { U } _ { 1 }.$$
Proof. For the 'if' direction, suppose α : A ∆ → ( H,V ) factors through such a homomorphism β. Then ( H,V ) divides ( i.e., is a quotient of a subalgebra of) the iterated wreath product. Since it is obvious that the identities defining EF -algebras are preserved under subalgebras and quotients, and are satisfied by U 1 , we only need to prove that the wreath product of two EF -algebras is an EF -algebra.
Clearly idempotence and commutativity of the horizontal monoid are preserved by the wreath product, since we are just forming the direct product of the component horizontal monoids. It remains to show that the identity vh + h = vh is preserved by the wreath product. We have
$$( v, f ) ( h _ { 1 }, h _ { 2 } ) + ( h _ { 1 }, h _ { 2 } ) \ & = \ \ ( v h _ { 1 }, f ( h _ { 1 } ) h _ { 2 } ) + ( h _ { 1 }, h _ { 2 } ) \\ & = \ \ ( v h _ { 1 } + h _ { 1 }, f ( h _ { 1 } ) h _ { 2 } + h _ { 2 } ) \\ & = \ \ ( v h _ { 1 }, f ( h _ { 1 } ) h _ { 2 } ) \\ & = \ \ ( v, f ) ( h _ { 1 }, h _ { 2 } ).$$
We now prove the converse. We suppose that ( H,V ) is an EF -algebra, and show by induction on | H | that any homomorphism into ( H,V ) factors through such an iterated wreath product. If | H | = 2 then the EF identities force H = { 0 , ∞} , and either | V | = 1 , or ( H,V ) = U 1 . So we may suppose | H > | 2 . H has trivial reachability classes, because h ≤ h ′ implies h = vh ′ = vh ′ + h ′ = h + h , ′ and similarly h ′ ≤ h implies h ′ = h + h , ′ so elements in the same reachability class are all equal. In particular Γ min has a single element, so ( H Γ min , V Γ min ) = ( H,V . ) Thus by Lemma 2, any homomorphism onto ( H,V ) factors through a direct product, and hence a wreath product, of EF -algebras with a unique subminimal reachability class. So it suffices to prove the theorem in the case where ( H,V ) has a single trivial subminimal class Γ = { h ∗ } . We claim that in this case α : A ∆ → ( H,V ) factors through
$$\gamma = \alpha _ { \Gamma } \otimes \beta \, \colon A ^ { \Delta } \rightarrow ( H _ { \Gamma }, V _ { \Gamma } ) \circ \mathcal { U } _ { 1 }$$
for some homomorphism β : ( A × H Γ ) ∆ →U 1 . Since | H Γ | = | H | -1 , the desired result follows from the inductive hypothesis.
We define the homomorphism β by giving the value of β a, h ( ) ∈ { 1 0 , } for every ( a, h ) ∈ A × H . Γ If h > Γ we set β a, h ( ) = 0 if α a ( ) · h = ∞ in ( H,V ) and β a, h ( ) = 1 otherwise. If h is the minimal element of H , Γ we set β a, h ( ) = 0 if a · h ∗ = ∞ , and β a, h ( ) = 1 otherwise.
/negationslash
We first establish the following fact: α s ( ) = ∞∈ H if and only if there is a node x in s such that the tree rooted at x is at, with β a, α ( Γ ( )) = 0 t . It follows trivially from the definition that if such a node exists then α at ( ) = ∞ ∈ H, and thus α s ( ) = ∞ , since ∞ is the unique minimal element of H. Conversely, suppose that α s ( ) = ∞ . There must be some node x such that the tree at rooted at x has α at ( ) = ∞ : the alternative would be that s = t 1 + · · · + t r , where each t i is a tree with α t ( i ) = ∞ but α s ( ) = ∞ . This cannot occur, because then by the uniqueness of the subminimal element, each α t ( i ) ≥ h , ∗ and the identities for EF -algebras would then give
$$h ^ { * } = h ^ { * } + \alpha ( t _ { 1 } ) + \cdots + \alpha ( t _ { r } ) = h ^ { * } + \alpha ( s ) + \infty = \infty,$$
/negationslash a contradiction. We thus choose a node of maximal depth such that the tree at rooted at this node has α at ( ) = ∞ . By the maximal depth condition, no tree component of t is mapped by α to ∞ , and by the argument we just gave α t ( ) = ∞ . So α t ( ) ≥ h , ∗ and thus β a, α ( Γ ( )) = 0 t .
We now have
$$\gamma ( s ) = ( \alpha _ { \Gamma } ( s ), \beta ( s ^ { \alpha _ { \Gamma } } ) ).$$
The left-hand component of γ s ( ) determines α s ( ) except for distinguishing between α s ( ) = h ∗ and α s ( ) = ∞ . The fact that we just proved shows that the right-hand component of γ s ( ) is ∞ if and only if α s ( ) = ∞ . Thus γ s ( ) completely determines α s . ( )
/negationslash
A classic result of Stiffler [Sti73] shows that a right transformation monoid ( Q,M ) divides an iterated wreath product of copies of the transformation monoid U 1 = ( 0 1 { , } { , 0 1 , } ) if and only if M is R -trivial. In terms of transformation monoids this means there is no pair of distinct states q = q ′ ∈ Q such that qm = q , q m ′ ′ ′ = q for some m,m ′ ∈ M. Since forest algebras are left transformation monoids, the analogous result would suggest that a forest algebra ( H,V ) divides an iterated wreath product of copies of U 1 if and only if V is L -trivial-that is, if and only if ( H,V ) has trivial reachability classes. We have already seen that this condition is necessary.
However, the following example shows that it is not sufficient.
/negationslash
Example 8. Figure 1 below defines the syntactic forest algebra of the language L of Example 4. The nodes in the diagram represent the elements of the horizontal monoid, and the arrows give the action of a generating set of letters A = { a, b } on the horizontal monoid. The letter transitions, together with the conventions about idempotence and commutativity, and the meaning of 0 and ∞ , completely determine the addition and the action. Since ∞ = a + b = a + ba = ba = b, this is not an EF -algebra, but the reachability classes are singletons.
Figure 1: An algebra with trivial reachability classes that is not an EF -algebra
## 5 Definiteness
## 5.1 Definite homomorphisms
Let k > 0 . A finite semigroup S is said to be reverse k -definite if it satisfies the identity
$$x _ { 1 } x _ { 2 } \cdots x _ { k } y = x _ { 1 } \cdots x _ { k }.$$
The reason for the word 'reverse' is that definiteness of semigroups was originally formulated in terms of right transformation monoids, so the natural analogue of definiteness in the setting of forest algebras corresponds to reverse definiteness in semigroups. Observe that the notions of definiteness and reverse definiteness in semigroups do not really make sense for monoids, since only the trivial monoid can satisfy the underlying identities. For much the same reason, we define definiteness for forest algebras not as a property of the algebras themselves, but of homomorphisms
$$\alpha \colon A ^ { \Delta } \rightarrow ( H, V ).$$
The depth of a context p ∈ V A is defined to be the depth of its hole; so for instance a context with its hole at a root node has depth 0. We say that the homomorphism α is k -definite, where k > 0 , if for every p ∈ V A of depth at least k, and for all s, s ′ ∈ H , α ps A ( ) = α ps ( ′ ) . Easily, if α , α 1 2 are k -definite homomorphisms, then so are α 1 × α 2 and ψα , 1 where ψ : ( H,V ) → ( H , V ′ ′ ) is a homomorphism of forest algebras.
A context is guarded if it has depth at least 1, that is, if the hole is not at the root. We denote by V gu A the subsemigroup of V A consisting of the guarded contexts.
Lemma 9. Let k > 0 . A homomorphism α : A ∆ → ( H,V ) is k -definite if and only if α V ( gu A ) is a reverse k -definite semigroup.
Proof. Let α be k -definite, and let p , . . . , p 1 k ∈ V gu A . Then the context p = p 1 · · · p k has depth at least k. Thus if q ∈ V gu A , and s ∈ H , A we have
$$\alpha ( p _ { 1 } \cdots p _ { k } q s ) = \alpha ( p q s ) = \alpha ( p s ) = \alpha ( p _ { 1 } \cdots p _ { k } s ).$$
As this holds for arbitrary s, faithfulness of the action implies
$$\alpha ( p _ { 1 } \cdots p _ { k } q ) = \alpha ( p _ { 1 } \cdots p _ { k } ),$$
and thus α V ( gu A ) is a reverse k -definite semigroup.
/negationslash
Conversely, suppose that α V ( gu A ) is a reverse k -definite semigroup. Let p be a context of depth at least k. By following the path from the hole of p to a root, we obtain a factorization p = q 1 · · · q , r where r ≥ k, and each q i has the form t + a /square + t , ′ where a ∈ A and t, t ′ ∈ H . A In particular, we can write p = p 1 · · · p , k where each p i ∈ V gu A . Let s ∈ H , A with s = 0 . Then we can write s = q · 0 , where q ∈ V gu A . Thus
$$\alpha ( p s ) = \alpha ( p _ { 1 } \cdots p _ { k } q \cdot 0 ) = \alpha ( p _ { 1 } \cdots p _ { k } \cdot 0 ) = \alpha ( p \cdot 0 ).$$
As this holds for arbitrary s, α is k -definite.
Definition 2. An EX -homomorphism is a homomorphism that is k -definite for some k ∈ N .
## 5.2 Free k -definite algebra
We construct what we will call free k -definite algebra over an alphabet A. This is a slight abuse of terminology, since as we noted above, it is the homomorphism into this algebra, and not the algebra itself, that is k -definite. We do this by recursively defining a sequence of congruences ∼ k on A . ∆ If k = 0 , then ∼ 0 is just the trivial congruence that identifies all forests. If k ≥ 0 and ∼ k ha been defined then we associate to each forest s = a s 1 1 + · · · a s , r r where each a i ∈ A, s i ∈ H , A the set
$$T _ { s } ^ { k + 1 } = \{ ( a _ { i }, [ s _ { i } ] _ { \sim _ { k } } ) \colon 1 \leq i \leq r \},$$
where [] ∼ k denotes the ∼ k -class of a forest. We then define s ∼ k +1 s ′ if and only if T k +1 s = T k +1 s ′ .
Proposition 10. Let k ≥ 0 . Then ∼ k +1 refines ∼ k . ∼ k is a congruence of finite index on A , ∆ with a horizontally idempotent and commutative quotient.
Proof. Obviously ∼ k is an equivalence relation of finite index. We prove by induction on k that it is also a congruence with a horizontally idempotent and commutative quotient. The case k = 0 is obvious. Assume now that ∼ k is a congruence with an idempotent and commutative quotient. If s ∼ k +1 s , ′ then idempotence and commutativity of ∼ k gives
$$s \sim _ { k } \sum _ { ( a, [ t ] ) \in T _ { s } ^ { k + 1 } } a t \sim _ { k } s ^ { \prime },$$
which proves the first claim. We further have
$$T _ { a s } ^ { k + 1 } = \{ ( a, [ s ] _ { \sim _ { k } } ) \} = \{ ( a, [ s ^ { \prime } ] _ { \sim _ { k } } ) \} = T _ { a s ^ { \prime } } ^ { k + 1 },$$
so as ∼ k +1 as . ′ Moreover, if s i ∼ k +1 s ′ i for i = 1 2 , , then s 1 + s 2 ∼ k +1 s ′ 1 + s , ′ 2 since addition of forests corresponds to union of the associated sets. Since the equivalence is preserved under application of a letter and under addition, it is a congruence. The observation about addition and union implies that the quotient is idempotent and commutative.
Intuitively, s ∼ k s ′ means that the forests s and s ′ are identical at the k levels closest to the root, up to idempotent and commutative equivalence. In fact, this intuition provides an equivalent characterization of ∼ k , which we give below. We omit the simple proof.
Lemma 11. Let s, s ′ ∈ H A and k > 0 . Let ¯ s, ¯ s , ′ denote, respectively, the forests obtained from s and s ′ by removing all the nodes at depth k or more. Then s ∼ k s ′ if and only if ¯ s and ¯ s ′ are idempotent-and-commutative equivalent.
Let us denote by α A,k the homomorphism from A ∆ onto its quotient by ∼ k . In the case where k = 1 , we will identify H / A ∼ 1 with the monoid ( P ( A , ) ∪ ) , and the horizontal component of α A, 1 with the map that sends each forest to the set of its root nodes.
The following theorem gives both the precise sense in which this is the 'free k -definite forest algebra', as well as the wreath product decomposition of k -definite homomorphisms into 1-definite homomorphisms into a forest algebra with horizontal monoid { 0 , ∞} .
Theorem 12. Let α : A ∆ → ( H,V ) be a homomorphism onto a finite forest algebra. Let k > 0 . The following are equivalent.
- (a) α is k -definite.
- (b) α factors through α A,k .
- (c) α factors through
$$\beta _ { 1 } \otimes \dots \otimes \beta _ { k } \colon A ^ { \Delta } \to ( H _ { 1 }, V _ { 1 } ) \circ \dots \circ ( H _ { k }, V _ { k } ),$$
where each
$$\beta _ { i } \colon ( A \times H _ { 1 } \times \cdots \times H _ { i - 1 } ) ^ { \Delta } \to ( H _ { i }, V _ { i } )$$
is 1-definite.
- (d) α factors through an iterated wreath product of 1-definite homomorphisms into U 2 .
Proof. (( a ) ⇒ ( b )). Let s ∼ k s ′ and let ¯ s, ¯ s ′ denote the forests obtained by removing all nodes at depth k or more from s and s . ′ Let α : A ∆ → ( H,V ) be k -definite. By Lemma 11, ¯ and ¯ are idempotent-and-commutative equivalent. s s ′ Thus by Lemma 1, α s (¯) = α s ( ¯ ) ′ . If t 2 ∈ H A is obtained from t 1 by removing a node of depth k together with all its descendants, then there is a context p with a hole at depth k and a forest u such that t 1 = pu, t 2 = p 0 Thus by k -definiteness, α t ( 2 ) = α t ( 1 ) . Consequently for any forest s, α s (¯) = α s . ( ) Thus with s ∼ k s ′ as above, we have
$$\alpha ( s ) = \alpha ( \bar { s } ) = \alpha ( \bar { s ^ { \prime } } ) = \alpha ( s ^ { \prime } ).$$
So α factors through α A,k , as required.
(( b ) ⇒ ( c )). It suffices to show that for k ≥ 1 , α A,k +1 factors through a homomorphism γ = α A,k ⊗ β, where β is 1-definite. In fact, we can choose β = α A × H / A ∼ k , 1 . We then have, for s ∈ H , A
$$\gamma ( s ) \ & = \ \left ( [ s ] _ { \sim _ { k } }, \beta ( s ^ { \alpha _ { A, k } } ) \right ) \\ & = \ \left ( [ s ] _ { \sim _ { k } }, T _ { s } ^ { k + 1 } \right )$$
so that in fact s ∼ k +1 s ′ if and only if γ s ( ) = γ s ( ′ ) . (That is, the two homomorphisms factor through one another, and so are essentially identical.)
(( c ) ⇒ ( d )) It will suffice to show that for any finite alphabet A, the homomorphism α A, 1 factors through a wreath product of 1-definite homomorphisms into U 2 . In fact, we will show that α A, 1 is isomorphic to a direct product of such homomorphisms. For each a ∈ A define
$$\beta _ { a } \colon A ^ { \Delta } \rightarrow \mathcal { U } _ { 2 }$$
by setting β a ( a ) = 0 , β a ( b ) = c 0 for b ∈ A -{ } a . If s ∈ H , A then β a ( s ) = ∞ if some root node of s is a, and β a ( s ) = 0 otherwise. In particular, β a ( s ) depends only on the set of labels of the root nodes of s and so is 1-definite. Consider the direct product
$$\beta = \left ( \prod _ { a \in A } \beta _ { a } \right ) \colon A ^ { \Delta } \to \prod _ { a \in A } \mathcal { U } _ { 2 }.$$
Then β s ( ) is an A -tuple from { 0 , ∞} in which the components with value ∞ are exactly those corresponding to the labels of the root nodes of s. Since α A, 1 ( s ) is the set of labels of root nodes of s, the two homomorphisms are equivalent.
(( d ) ⇒ ( a )). It suffices to show that for homomorphisms
$$\alpha \colon A ^ { \Delta } \rightarrow ( H _ { 1 }, V _ { 1 } ), \beta \colon ( A \times H _ { 1 } ) ^ { \Delta } \rightarrow ( H _ { 2 }, V _ { 2 } ),$$
where α is k -definite and β is 1-definite, that γ = α ⊗ β is ( k +1)-definite. That is, we will show that if p is a context with a hole at depth k +1 and s is a forest, then γ ps ( ) is independent of s. We can write ps = u + aqs + v, where u, v ∈ H , A a ∈ A, and q is a context with a hole at depth k. It is thus sufficient to show that γ aqs ( ) is independent of s. But we have
$$\gamma ( a q s ) \ & = \ \gamma ( a ) \gamma ( q s ) \\ & = \ \gamma ( a ) ( \alpha ( q s ), \beta ( ( q s ) ^ { \alpha } ) ) \\ & = \ \left ( \alpha ( a ) \cdot \alpha ( q s ), \beta ( a, \alpha ( q s ) ) \cdot \beta ( ( q s ) ^ { \alpha } ) \right ).$$
Since α is k -definite, α qs ( ) depends only on q. Since β is 1-definite, the righthand coordinate depends only on β a, α qs ( ( )) , which depends only on a and q. Thus the value is independent of s, as required.
## 6 ( EF EX , ) -algebras
## 6.1 The principal result
Definition 3. An ( EF EX , ) -homomorphism α : A ∆ → ( H,V ) is one that factors through an iterated wreath product
$$\beta _ { 1 } \otimes \dots \otimes \beta _ { k },$$
where each β i either maps into U 1 or is 1-definite. By Theorem 12 we can suppose that each 1-definite β i maps into U 2 .
The principal result of this paper is an effective necessary and sufficient condition for a homomorphism to be a ( EF EX , )-homomorphism.
Definition 4. Suppose α : A ∆ → ( H,V ) . Let s , s 1 2 ∈ H , k > A 0 , and Γ ⊆ H a reachability class for ( H,V . ) We say that s , s 1 2 are ( α, k, Γ) -confused, and write s 1 ≡ α,k, Γ s , 2 if
$$( s _ { 1 } ) ^ { \alpha _ { \Gamma } } \sim _ { k } ( s _ { 2 } ) ^ { \alpha _ { \Gamma } }, \ \alpha ( s _ { 1 } ), \alpha ( s _ { 2 } ) \in \Gamma.$$
Observe that the equivalence relation ∼ k in the first item is over the extended alphabet A × H . Γ It is worth emphasizing what ( s ) α Γ is when α s ( ) ∈ Γ: We are tagging each node of x of s with the value α t ( ) ∈ H if the tree rooted at x is at and α t ( ) > Γ , but we are tagging the node by ∞ -effectively leaving the node untagged-if α t ( ) ∈ Γ Since . α s ( ) ∈ Γ every node is of one of these two types. ,
Definition 5. A homomorphism α is nonconfusing if and only if there exists k > 0 such that ≡ α,k, Γ is equality for reachability classes Γ .
It follows from Proposition 10 that ≡ α,k +1 Γ , refines ≡ α,k, Γ , so that if α is nonconfusing with associated parameter k, then it is nonconfusing for all m>k. Our main result is:
Theorem 13. Let α : A ∆ → ( H,V ) be a homomorphism into a finite forest algebra. α is a ( EF EX , ) homomorphism if and only if it is nonconfusing.
The proof of Theorem 13 will be given in the next two subsections.
Example 14. Consider once again the algebra of Examples 4 and 8 and the associated homomorphism α from { a, b } ∆ . Since the algebra has trivial reachability classes, α is nonconfusing for all k , so Theorem 13 implies that α is an ( EF EX , ) -homomorphism. We will see in the course of the proof of the main theorem how the wreath product decomposition is obtained.
Example 15. Consider again the forest algebra U 2 = ( { 0 , ∞} { , 1 , c ∞ , c 0 } ) , and the homomorphism α from { a, b, c } ∆ onto U 2 that maps a to 1, b to c 0 and c to c ∞ . There is a unique reachability class Γ , so for any forest s, s α Γ is identical to s. Now observe that a b k ∼ k a c, k but that these are mapped to different elements under α. So by our main theorem, α is not an ( EF EX , ) -homomorphism.
## 6.2 Sufficiency of the condition
We will use the ideal theory developed earlier to prove that every nonconfusing homomorphism factors through a wreath product decomposition of the required kind. The structure of our proof resembles the one given for Theorem 7 Once again, we proceed by induction on | H . | The base of the induction is the trivial case | H | = 1 . Let us suppose that
$$\alpha \colon A ^ { \Delta } \rightarrow ( H, V )$$
is nonconfusing with parameter k, that | H > | 1 , and that every nonconfusing homomorphism into a forest algebra with a smaller horizontal monoid factors through a wreath product of the required kind.
Let Γ = Γ min . Suppose first that | Γ | > 1 . We claim that α factors through
$$\beta = \alpha _ { \Gamma } \otimes \alpha _ { B, k } \colon A ^ { \Delta } \rightarrow ( H _ { \Gamma }, V _ { \Gamma } ) \circ B ^ { \Delta } / \sim _ { k }$$
where B = A × H . Γ Since | H Γ | < H | | and α Γ is also nonconfusing, the induction hypothesis gives the desired decomposition of α. To establish the claim, let s ∈ H . A Then
$$\beta ( s ) = ( \alpha _ { \Gamma } ( s ), [ s ^ { \alpha _ { \Gamma } } ] _ { \sim _ { k } } ).$$
If s / ∈ Γ , then the value of the left-hand coordinate determines α s . ( ) If s ∈ Γ , then by the nonconfusion condition, the value of the right-hand coordinate determines α s . ( ) Thus α factors through β as required.
So let | Γ = 1 | . Then Γ = {∞} and ( H ,V Γ Γ ) = ( H,V . ) Lemma 2 implies that we can suppose ( H,V ) has a single subminimal reachability class, because each of the component homomorphisms in the direct product is nonconfusing, and the direct product factors through the wreath product.
Thus we have a unique minimal element ∞ , and a unique subminimal ideal Γ ′ . We claim that α factors through
$$\beta = \alpha _ { 1 } \otimes \alpha _ { 2 } \otimes \alpha _ { 3 } \, \colon A ^ { \Delta } \to ( H _ { \Gamma ^ { \prime } }, V _ { \Gamma ^ { \prime } } ) \circ B ^ { \Delta } / \sim _ { k } \circ \, \mathcal { U } _ { 1 },$$
where
$$\alpha _ { 1 } = \alpha _ { \Gamma ^ { \prime } }.$$
$$\alpha _ { 2 } = \alpha _ { B, k }, \, \text{where } B = A \times H _ { \Gamma ^ { \prime } }.$$
$$\alpha _ { 3 } \colon ( B \times 2 ^ { B } ) ^ { \Delta } \to \mathcal { U } _ { 1 } \text{ will be defined below}.$$
To see how α 3 should be defined, let us consider what this homomorphism needs to tell us. If α s ( ) > Γ ′ , then the first coordinate of β s ( ) determines α s . ( ) If α s ( ) ∈ Γ ′ , then the first two components of β s ( ) determine α s , ( ) by nonconfusion. So we will use the third component to distinguish between α s ( ) ∈ Γ ′ and α s ( ) = ∞ . The value of the first component already determines whether or not α s ( ) ∈ Γ ′ ∪ {∞} , so we really just need to be able to tell when α s ( ) = ∞ . There are several cases to consider, depending on whether or not s contains a
tree t such that α t ( ) = ∞ . If not, then s = t 1 + · · · + t r , where α t ( i ) ≥ Γ ′ for all i. Observe that if this is the case, then the set of values { α t ( 1 ) , · · · , α ( t r ) } is determined by the second component { [ t α 1 1 ] ∼ k , . . . , [ t α 1 r ] ∼ k } of β s . ( ) If s contains a tree t such that α t ( ) = ∞ , pick such a tree at maximal depth. Then t = a t ( 1 + · · · + t r ) , where once again α t ( i ) ≥ Γ ′ for all i, and the set of values { α t ( 1 ) , · · · , α ( t r ) } is determined by the second component of β s . ( ) We now specify the value of α 3 ( a, h, Q . ) As remarked above, Q determines a set of values all in Γ ′ or strictly higher. Let h Q ∈ H A be the sum of these values. If either h Q = ∞ , or ah Q = ∞ , set α 3 ( a, h, Q ) = 0 . Otherwise, α 3 ( a, h, q ) = 1 .
The third component of β s ( ) will be ∞ if and only if there is some subtree a t ( 1 + · · · + t r ) such that
$$\alpha _ { 3 } ( a, \alpha _ { 1 } ( t _ { 1 } + \dots + t _ { r } ), \{ [ t _ { 1 } ^ { \alpha _ { 1 } } ] _ { \sim _ { k } }, \dots, [ t _ { r } ^ { \alpha _ { 1 } } ] _ { \sim _ { k } } \} ) = 0.$$
If we pick the subtree of maximal depth at which this occurs, then as argued above, α s ( ) = ∞ . The only other way we can have α s ( ) = ∞ is if there is no such subtree, but s = t 1 + · · · + t r where each α t ( i ) ≥ Γ ′ and the sum of these values is ∞ . In this case, the fact that no such subtree exists is determined by the third coordinate of β s ( ) being 1, and the set of α t ( i ) ≥ Γ is determined by ′ the second coordinate of β s . ( ) So in all cases β s ( ) determines α s . ( )
## 6.3 Necessity of the condition.
To prove the converse, we have to show preservation of nonconfusion under quotients and wreath products with the allowable factors. This is carried out in the following three lemmas. Preservation under quotients (Lemma 16) is the most difficult of the three to show.
## Lemma 16. Let
$$\alpha \colon A ^ { \Delta } \rightarrow ( H _ { 1 }, V _ { 1 } ), \beta \colon A ^ { \Delta } \rightarrow ( H _ { 2 }, V _ { 2 } ),$$
be homomorphisms onto finite forest algebras such that β factors through α . If α is nonconfusing then so is β.
Proof. Suppose that α is nonconfusing with parameter k. We will show that β is nonconfusing with the same parameter. To this end, let Γ ⊂ H 2 be a reachability class, and let s , s 1 2 ∈ H A be forests with
$$s _ { 1 } ^ { \beta _ { \Gamma } } \sim _ { k } s _ { 2 } ^ { \beta _ { \Gamma } } \text{ and } \beta ( s _ { 1 } ), \beta ( s _ { 2 } ) \in \Gamma.$$
We must show β s ( 1 ) = β s ( 2 ) .
Since β factors through α, there is an onto forest algebra homomorphism η : ( H ,V 1 1 ) → ( H ,V 2 2 ) such that β = ηα. Choose h 2 ∈ Γ , and let h 1 ∈ H 1 be such that η h ( 1 ) = h , 2 and h 1 is ≤ -minimal for this property. Let Λ be the reachability class of h . 1 By Lemma 3, η (Λ) = Γ .
We now perform a little surgery on the forests s 1 and s 2 : For each h ∈ H , 2 we choose s h ∈ H A such that β s ( h ) = h, and such that if h ∈ Γ then , α s ( h ) ∈ Λ .
(This is where we use the fact that η (Λ) = Γ ) Look at a nodes of . s 1 or s 2 at depth k -1 . If the tree rooted at such a node is at, where t ∈ H , A we replace the forest t by s β t ( ) . We do this for every node at depth k -1 of the two forests and obtain new forests ¯ s 1 and ¯ s . 2 Obviously we have not changed the values of these forests under β, so we have β s (¯ ) = i β s ( i ) for i = 1 2 , . Thus it is sufficient to show β s (¯ ) = 1 β s (¯ ) 2 .
We claim that
$$( \bar { s } _ { 1 } ) ^ { \alpha _ { \Lambda } } \sim _ { k } ( \bar { s } _ { 2 } ) ^ { \alpha _ { \Lambda } } \text{ and } \alpha _ { \Lambda } ( \bar { s } _ { 1 } ) = \alpha _ { \Lambda } ( \bar { s } _ { 2 } ).$$
This claim gives our desired result. To see this, note that the second equality in the claim implies α s (¯ ) 1 > Λ if and only if α s (¯ ) 2 > Λ If . α s (¯ ) 1 , α (¯ ) s 2 > Λ we , have
$$\alpha ( \bar { s } _ { 1 } ) = \alpha _ { \Lambda } ( \bar { s } _ { 1 } ) = \alpha _ { \Lambda } ( \bar { s } _ { 2 } ) = \alpha ( \bar { s } _ { 2 } ).$$
/negationslash
On the other hand, if α s (¯ ) 1 , α (¯ ) s 2 > Λ we must have , α s (¯ ) 1 , α (¯ ) s 2 ∈ Λ because , ηα s (¯ ) = i β s (¯ ) i ∈ Γ . Then the nonconfusing property of α gives α s (¯ ) = 1 α s (¯ ) 2 . So in all cases we have α s (¯ ) 1 = α s (¯ ) 2 . Applying η gives β s (¯ ) 1 = β s (¯ ) 2 , as required.
We prove the claim by induction on k. More precisely, for each k ≥ 1 , we will show that if s β Γ ∼ k t β Γ , then (¯) s α Λ ∼ k ( ¯ ) t α Λ , and α Λ (¯) s = α Λ ( ¯ ) t . First suppose k = 1 , so s β Γ ∼ 1 t β Γ . A root node of (¯) s α Λ has the form ( a, α Λ ( s h )) , for some h ∈ H . 2 This means that s has a component tree of the form au, where β u ( ) = h. Thus s β Γ has a root node labeled ( a, h ) or ( a, ∞ ) , depending on whether h ∈ Γ Since . s β Γ ∼ 1 t β Γ , the same root node occurs in t β Γ . If h / ∈ Γ , then t contains a component av with β v ( ) = h, and thus ¯ has a component t as h , so that ( ¯ ) t α Λ has a root node ( a, α Λ ( s h )) . If h ∈ Γ then , t contains a component av with β v ( ) = h ′ ∈ Γ so that ( ¯ ) , t α Λ contains a root node
$$( a, \alpha _ { \Lambda } ( s _ { h ^ { \prime } } ) ) = ( a, \infty ) = ( a, \alpha _ { \Lambda } ( s _ { h } ) ),$$
so that every root node of (¯) s α Λ is also a root node of ( ¯ ) t α Λ . We get the converse inclusion by symmetry. So (¯) s α Λ ∼ 1 ( ¯ ) t α Λ . Moreover, we also have α Λ (¯) s = α Λ ( ¯ ) t . This is because if no root node of (¯) s α Λ has the form ( a, ∞ ) , then α Λ (¯) s is determined by the sum of the α as ( h ) , and α Λ ( ¯ ) t is determined by the sum of the same set of terms. If some root node of (¯) s α Λ is ( a, ∞ ) , then the same is true for some root node of ( ¯ ) t α Λ , and we have α Λ (¯) = s ∞ = α Λ ( ¯ ) t .
Our induction hypothesis is now that k ≥ 1 , and that whenever s β Γ ∼ k t β Γ , we have both (¯) s α Λ ∼ k ( ¯ ) t α Λ , and α Λ (¯) = s α Λ ( ¯ ) t . We show that these properties are preserved at level k +1 If we write .
$$s = a _ { 1 } s _ { 1 } + \dots + a _ { k } s _ { k }, t = b _ { 1 } t _ { 1 } + \dots + b _ { p } t _ { p },$$
where the s , t i j belong to H , A and the a , b i j to A, then we have
$$\bar { s } = a _ { 1 } \bar { s } _ { 1 } + \dots + a _ { k } \bar { s } _ { k }, t = b _ { 1 } \bar { t } _ { 1 } + \dots + b _ { p } \bar { t } _ { p }.$$
↦
It is important to understand precisely what the operator u → ¯ means in these u equations: On the left-hand sides we are performing the substitution at nodes
of ¯ s, t ¯ at level k ; on the right-hand sides we carry out the operation at nodes of level k -1 . The ∼ k +1 -class of (¯) s α Λ is determined by the set T k +1 ¯ s α Λ of pairs of the form (( a , α i Λ (¯ )) s i , [(¯ ) s i α Λ ] ∼ k ) . Let 1 ≤ i ≤ r. The corresponding set T k +1 s β Γ for the ∼ k +1 -class of s β Γ contains the pair (( a , β i Γ ( s i )) , [ s β Γ i ] ∼ k ) . Thus there is some j such that a i = b j and s β Γ i ∼ k t β Γ j . By the induction hypothesis, we have both ¯ s α Λ i ∼ k ¯ t α Λ j and α Λ (¯ ) = s i α Λ ( ¯ t j ) . Thus the pair (( a , α i Λ (¯ )) s i , [(¯ ) s i α Λ ] ∼ k ) , also occurs in T k +1 ¯ t α Λ This shows T k +1 ¯ s α Λ ⊆ T k +1 ¯ t α Λ . We get the converse inclusion by symmetry. So (¯) s α Λ ∼ k +1 ( ¯ ) t α Λ . We obtain α Λ (¯) = s α Λ ( ¯ ) t just as we did in the case k = 1 .
## Lemma 17. Suppose that
$$\alpha \colon A ^ { \Delta } \rightarrow ( H, V ) \circ \mathcal { U } _ { 1 }$$
is a homomorphism, and that β = πα, where π is the projection morphism onto ( H,V , ) is nonconfusing. Then α is nonconfusing.
Proof. Let k be the nonconfusion parameter for β. Let ∆ ⊆ H × { 0 , ∞} be a reachability class in the image of α. Suppose s, t ∈ H A with ( s ) α ∆ ∼ k ( ) t α ∆ and α s , α t ( ) ( ) ∈ ∆ We must show . α s ( ) = α t . ( )
We can never reach an element of the form ( h, 0) from one of the form ( h , ′ ∞ ) , since
$$( v, f ) ( h ^ { \prime }, \infty ) = ( v h, f ( h ) \cdot \infty ) = ( v h, \infty ).$$
So, since α s ( ) = ∼ α t , ( ) they must agree in the right-hand coordinate. It remains to show that the left coordinates β s , β t ( ) ( ) are equal.
/negationslash
As we argued in the proof of Lemma 16, π (∆) is contained in a reachability class Γ of H. Let us look at the corresponding nodes of ( s ) α ∆ , ( ) t α ∆ and of ( s ) β Γ , ( ) t β Γ . If a node of ( s ) α ∆ has a label of the form ( a, h , ) where h > ∆ then the , corresponding node of ( s ) β Γ will either be labeled ( a, π ( h )) or ( a, ∞ ) , and this is entirely determined by the value of h. If, on the other hand, a node of ( s ) α ∆ has the label ( a, ∞ ) , then h > ∆ But since . α s ( ) ∈ ∆ this implies , h ∈ ∆ so that , π h ( ) ∈ Γ and thus the node also has the label ( , a, ∞ ) in ( s ) β Γ . Thus the labels of nodes of ( s ) β Γ and ( ) t β Γ , are determined by applying a mapping H ∆ → H Γ to the right coordinates of the node labels of ( s ) α ∆ , ( ) t α ∆ As a consequence, ( s ) β Γ ∼ k ( ) t β Γ . Also, for i = 1 2 , , β s ( ) = πα s ( ) ∈ Γ , and likewise β t ( ) ∈ Γ , so nonconfusion gives β s ( ) = β t , ( ) as required.
## Lemma 18. Suppose that
$$\alpha = \beta \otimes \gamma \, \colon A ^ { \Delta } \rightarrow ( H, V ) \circ ( H ^ { \prime }, V ^ { \prime } )$$
is a homomorphism, that β is nonconfusing, and that γ : ( A × H ) ∆ → ( H , V ′ ′ ) is 1-definite. Then α is nonconfusing.
Proof. Let us suppose in particular that β is nonconfusing with parameter k. We claim that α is nonconfusing with parameter k + 1 . Let Γ ⊆ H × H ′ be a reachability class in the image of α, and let s, t ∈ H A with ( s ) α Γ ∼ k +1 ( ) t α Γ and α s , α t ( ) ( ) ∈ Γ We will show . α s ( ) = α t . ( ) We begin by proving that β s ( ) = β t , ( ) using the nonconfusing property of β, and then use 1-definiteness to show that the right-hand coordinates are also equal.
We write both s and t as sums of the component trees:
$$s = a _ { 1 } s _ { 1 } + \cdots + a _ { q } s _ { q }, \ t = b _ { 1 } t _ { 1 } + \cdots + b _ { r } t _ { r }.$$
Since ( s ) α Γ ∼ k ( ) t α Γ , the two sets of pairs
$$\{ ( a _ { j }, [ ( s _ { j } ) ^ { \alpha _ { \Gamma } } ] _ { \sim _ { k } } ) \, \colon 1 \leq j \leq q \}, \{ ( b _ { j }, [ ( t _ { j } ) ^ { \alpha _ { \Gamma } } ] _ { \sim _ { k } } ) \, \colon 1 \leq j \leq r \}$$
are equal. As we argued in the previous lemma, if
$$( s _ { j } ) ^ { \alpha _ { \Gamma } } \sim _ { k } ( t _ { j ^ { \prime } } ) ^ { \alpha _ { \Gamma } },$$
then nonconfusion for β makes their values under β equal. Let us denote this common value by h. We then have, for a ∈ A,
$$\alpha ( a s _ { j } ) = \alpha ( a ) ( h, h ^ { \prime } ) = ( \beta ( a ) h, \gamma ( a, h ) h ^ { \prime } ).$$
We similarly have
$$\alpha ( a t _ { j ^ { \prime } } ) = \alpha ( a ) ( h, h ^ { \prime \prime } ) = ( \beta ( a ) h, \gamma ( a, h ) h ^ { \prime \prime } ),$$
and the two values are the same by 1-definiteness of γ. Thus
$$\{ \alpha ( a _ { j } s _ { j } ) \, \colon 1 \leq j \leq q \} = \{ \alpha ( b _ { j } t _ { j } ) \, \colon 1 \leq j \leq r \}.$$
We get α s ( 1 ) = α s ( 2 ) by summing over these two sets and using idempotence and commutativity of addition.
## 6.4 Decidability
In this section we show that we can effectively determine if a given forest algebra homomorphism α : A ∆ → ( H,V ) is nonconfusing. Our method is a variant of the one given by Bojanczyk and Walukiewicz for the binary tree case. [BW06]
We suppose that α is onto, and that Γ ⊆ H is a reachability class. We first describe an algorithm that constructs a finite sequence of subsets B , B , B , . . . 0 1 2 of Γ × Γ .
- · Set
/negationslash
$$B _ { 0 } = \{ ( h, h ^ { \prime } ) \in \Gamma \times \Gamma \, \colon h \neq h ^ { \prime } \}.$$
- · For j = 0 1 , , . . . ,
$$\begin{array} {$$
- -If there exist ( h, h ′ ) ∈ B j +1 and g ∈ H such that ( h + g, h ′ + ) g ∈ B , 0 add ( h + g, h ′ + ) to g B j +1 .
- -If there exist ( h, h ′ ) , ( g, g ′ ) ∈ B j +1 such that ( h + g, h ′ + g ′ ) ∈ B , 0 add ( h + g, h ′ + g ′ ) to B j +1 .
- -Repeat the preceding two steps until no new elements can be added to B j +1 .
Since there are at most 2 | H | 2 different possibilities for the B , i the algorithm will eventually cycle, so we terminate the execution as soon as we find some B i = B j for i < j. In fact, we will see below that the algorithm requires considerably less time and storage than this crude analysis suggests. We prove the following crucial property of the algorithm:
/negationslash
Theorem 19. Let h, h ′ ∈ Γ with h = h , ′ and let k ≥ 0 . Then ( h, h ′ ) ∈ B k if and only if there exist s, t ∈ H A with ( s ) α Γ ∼ k ( ) t α Γ , α s ( ) = h, and α t ( ) = h . ′ We also have B k +1 ⊆ B k for all k ≥ 0 .
Proof. Let C k denote the set of pairs ( h, h ′ ) satisfying the condition in the statement of the theorem. We will prove by induction on k that B k = C k for all k ≥ 0 . The case k = 0 is trivial, so assume that B k = C k for some k ≥ 0 . We show B k +1 = C k +1 .
We first prove B k +1 ⊆ C k +1 by induction on the construction of B k +1 . A pair added to B k +1 at the initial step of the construction has the form ( α a h, α a h ( ) ( ) ′ ) ∈ B . 0 where a ∈ A, ( h, h ′ ) ∈ B . k By the induction hypothesis, ( h, h ′ ) ∈ C , k so there is a pair of forests s, t with ( ) s α Γ ∼ k ( ) t α Γ and α s ( ) = h, α t ( ) = h . ′ Since α s , α t ( ) ( ) ∈ Γ we have ,
$$( a s ) ^ { \alpha _ { \Gamma } } = ( a, \infty$$
so that ( as ) α Γ ∼ k +1 ( at ) α Γ . Thus ( α a h, α a h ( ) ( ) ′ ) ∈ C k +1 .
Now suppose that a pair ( h, h ′ ) is added to B k +1 after this initial step. Then we have either h = h 1 + g, h ′ = h 2 + g, where ( h , h 1 2 ) ∈ B k +1 was added at an earlier step; or h = h 1 + g , 1 h ′ = h 2 + g , 2 where ( h , h 1 2 ) , ( g , g 2 2 ) ∈ B k +1 were added at earlier steps. In the first case, the hypothesis of induction by construction gives ( h , h 1 2 ) ∈ C k +1 , so there exists a pair s, t of forests with ( s ) α Γ ∼ k +1 ( ) t α Γ , and α s ( ) = h , α t 1 ( ) = h . 2 Let u be any forest such that α u ( ) = g. Then
$$( s + u ) ^ { \alpha _ { \Gamma } } = ( s ) ^ {$$
and α s ( + u ) = h, α t ( + u ) = h , ′ so ( h, h ′ ) ∈ C k +1 . We argue the second case similarly, replacing u by a second pair of forests s , t ′ ′ with ( s ′ ) α Γ ∼ k +1 ( t ′ ) α Γ mapping to g , g 1 2 . This shows B k +1 ⊆ C k +1 .
We now prove the opposite inclusion C k +1 ⊆ B k +1 . It is sufficient to show that whenever ( s ) α Γ ∼ k +1 ( ) t α Γ , and α s , α t ( ) ( ) ∈ Γ then either , α s ( ) = α t , ( ) or ( α s , α t ( ) ( )) ∈ B k +1 . We write
$$s = a _ { 1 } s _ { 1 } + \cdots + a _ { m }$$
For each i = 1 , . . . , m there is j = 1 , . . . , n such that ( a s i i ) α Γ ∼ k +1 ( a t ′ j j ) α Γ (which implies in particular that a i = a ′ j ) and vice-versa. Thus by duplicating and reordering terms, we can assume that
$$s & = a _ { 1 } s _ { 1 } + \dots + a _ { m$$
where ( a s j j ) α Γ ∼ k +1 ( a t j j ) α Γ for each j. We show by induction on j that for each pair of partial sums
$$s ^ { ( j ) } = a _ { 1 } s _ { 1 } + \dots + a _ { j } s _ { j },$$
$$t ^ { ( j ) } = a _ { 1 } t _ { 1 } + \dots + a _ { j } t _ { j },$$
/negationslash we either have α s ( ( j ) ) = α t ( ( j ) ) , or ( α s ( ( j ) ) , α ( t ( j ) )) ∈ B k +1 . This is true for j = 0 , since we can take s (0) = t (0) = 0 , the empty forest. Suppose it holds for some j ≥ 0 and consider the partial sums s ( j +1) , t ( j +1) . There are several cases to consider, depending on whether or not α s ( ( j ) ) = α t ( ( j ) ) , α s ( ( j +1) ) = α t ( ( j +1) ) , and α a ( j +1 s j +1 ) = α a ( j +1 t j +1 ) . We will treat in detail the case where we have inequality for all three of these pairs; the other cases are proved similarly, and are easier. Let h 1 = α s ( ( j ) ) , h 2 = α t ( ( j ) ) . Since we assume h 1 = h , 2 the induction hypothesis on j gives ( h , h 1 2 ) ∈ B k +1 . Since α a ( j +1 s j +1 ) = α a ( j +1 t j +1 ) , we must have ( a j +1 s j +1 ) α Γ = ( a j +1 , ∞ )( s j +1 ) α Γ , ( a j +1 t j +1 ) α Γ = ( a j +1 , ∞ )( t j +1 ) α Γ . This is because ∼ k +1 -equivalence implies the root nodes must be equal, and if the common value was ( a j +1 , h ) for some h > Γ then we would , get α a ( j +1 s j +1 ) = α a ( j +1 ) h = α a ( j +1 t j +1 ) , contrary to assumption. Thus α s ( j +1 ) , α ( t j +1 ) ∈ Γ . By the induction hypothesis (on k ), ( α s ( j +1 ) , α ( t j +1 )) ∈ B , k so we get ( α a ( j +1 s j +1 ) , α ( a j +1 t j +1 ) ∈ B k +1 . Finally
/negationslash
$$\alpha ( s ^ { ( j + 1 ) } ) = h _ { 1 } + \alpha ( a _ { j + 1 } s _ { j + 1 } ), \alpha ( t ^ { ( j + 1 ) } ) = h _ { 2 } + \alpha ( a _ { j + 1 } t _ { j + 1 } ),$$
so ( α s ( ( j +1) ) , α ( t ( j +1) )) ∈ B k +1 . This completes the proof that B k = C k for all k.
Since ∼ k +1 refines ∼ k , we obtain C k +1 ⊆ C , k and thus B k +1 ⊆ B k for all k ≥ 0 .
/negationslash
Because B k +1 ⊆ B , k the algorithm will terminate when for some k, either B k = ∅ or B k +1 = B k = ∅ . Theorem 19 then tells us that if B k = ∅ for all reachability classes Γ , then α is nonconfusing. Otherwise, there is some Γ for which C k is nonempty for every k, and thus α cannot be nonconfusing.
Since | B k | strictly decreases at each execution of the outer loop of the algorithm, this loop will not be executed more than | Γ | 2 < H | | 2 times. Computing each set B j in the inner loop also takes time that is polynomial in | H | and | A , | assuming that we have access to the table of operations in ( H,V ) and the values of α a ( ) for a ∈ A. Finally, given the table of addition in H and the action of letters of A on H, we can construct the graph of the reachability order and
compute its strongly connected components in time polynomial in | H | and | A , | so the running time of the entire algorithm is polynomial in | H | and | A | .
We summarize these observations as follows:
## Theorem 20.
- (a) A homomorphism α : A ∆ → ( H,V ) is an ( EF EX , ) -homomorphism if and only if it is nonconfusing. Furthermore, α is nonconfusing, then it is nonconfusing with parameter | H . | 2
- (b) We can determine in time polynomial in ( | A | + | H | ) whether a given α is an ( EF + EX ) -homomorphism.
## 7 Results
Using the wreath product characterizations of EF -algebras, EX -homomorphisms, and ( EF EX , )-homomorphisms of the previous three sections, we get:
Theorem 21. Let A be a finite alphabet, and let L ⊆ H . A
- (a) L is defined by an EF -formula if and only if ( H ,V L L ) is an EF -algebra.
- (b) L is defined by an EX -formula if and only if µ L is an EX -homomorphism.
- (c) L is defined by an EF + EX -formula if and only if µ L is an ( EF EX , ) -homomorphism.
- (d) There are effective procedures for determining, given a finite tree automaton recognizing L , whether L is definable by an EF -, EX -, or EF + EX -formula, and for producing a defining formula in case one exists.
Proof. The first three assertions are proved similarly; we give the proof for the third one, as it is the most general. First, suppose L is defined by an EF + EX -formula. We prove by induction on the depth of nesting of the operators in the formula that µ L is an ( EF EX , )-homomorphism. The base case is when the depth of nesting is 0. The only forest formulas with nesting depth 0 are T and F , in which case L is either H A or ∅ , and the syntactic forest algebra is trivial. We now suppose that L is defined by a forest formula φ, with nesting depth k > 0 , and that the syntactic morphism of every language defined by a formula of smaller depth is an ( EF EX , )-homomorphism. We can write φ as a boolean combination of formulas of the form EF τ and EX τ, where τ is a tree formula. It suffices to show that the syntactic morphisms of the languages defined by EF τ and EX τ are ( EF EX , )-homomorphisms: this is because the syntactic morphism of the union or intersection of two languages factors through the direct product of the syntactic morphisms of the two languages, which in turn factors through the wreath product. By the inductive hypothesis, the languages defined by the
forest formulas of τ are recognized by ( EF EX , )-homomorphisms, so by Propositions 5 and 6, so are the languages defined by EF τ and EX τ. Thus the syntactic morphisms of these languages are ( EF EX , )-homomorphisms.
Conversely, suppose the syntactic morphism of L ⊆ H A is an ( EF EX , )-homomorphism. Then L is recognized by a wreath product α 1 ⊗···⊗ α , r where each component homomorphism is either 1-definite or maps into U 1 . It follows from Propositions 5 and 6 and induction on r that L is defined by an EF + EX -formula.
We turn to the results about effectively determining definability and producing formulas. We can construct both the syntactic forest algebra and syntactic morphisms for a language L from any automaton recognizing L. For the case of definability by EF formulas we only need to verify the identities for EF -algebras. For EX formulas, we need to test whether µ L ( V gu ) is a reverse-definite semigroup. A semigroup S is reverse-definite if and only if es = s for all e, s ∈ S with e idempotent, so this too can be determined effectively. Using the characterization of Theorem 13 our results in Section 6.4 show that we can effectively determine definability by EF + EX -formulas. While we have not provided a streamlined algorithm for producing the defining formulas themselves, our proofs of wreath product decompositions are entirely constructive, and the formulas themselves can be derived from the construction of these decompositions along with Propositions 5 and 6.
## References
| [BS09] | Michael Benedikt and Luc Segoufin. Regular tree languages definable in FO and in FO mod . ACM Trans. Comput. Log. , 11(1), 2009. |
|----------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [BSS12] | Mikolaj Bojanczyk, Luc Segoufin, and Howard Straubing. Piecewise testable tree languages. Logical Methods in Computer Science , 8(3), 2012. |
| [BSW12] | Mikolaj Bojanczyk, Howard Straubing, and Igor Walukiewicz. Wreath products of forest algebras, with applications to tree logics. Logical Methods in Computer Science , 8(3), 2012. |
| [BW06] | Mikolaj Bojanczyk and Igor Walukiewicz. Characterizing EF and EX tree logics. Theor. Comput. Sci. , 358(2-3):255-272, 2006. |
| [BW08] | Mikolaj Bojanczyk and Igor Walukiewicz. Forest algebras. In J¨rg o Flum, Erich Gr¨del, a and Thomas Wilke, editors, Logic and Automata , volume 2 of Texts in Logic and Games , pages 107-132. Amsterdam University Press, 2008. |
| [CPP93] | Jo¨lle e Cohen, Dominique Perrin, and Jean-Eric Pin. On the expressive power of temporal logic. J. Comput. Syst. Sci. , 46(3):271-294, 1993. |
| [Esi05] ´ | Zoltan ´ Esik. An algebraic characterization of the expressive power of temporal logics on finite trees. 1st Int. Conf. Algebraic Informatics, Aristotle Univ. of Thessaloniki , pages 53-110, 2005. |
|-------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [Sti73] | P. E. Stiffler. Extension of the fundamental theorem of finite semi- groups. Advances in Mathematics , 11(2):159 - 209, 1973. |
| [Str94] | Howard Straubing. Finite Automata, Formal Logic, and Circuit Com- plexity . Birkh¨user, a Boston, 1994. |
| [TW96] | Denis Th´rien e and Thomas Wilke. Temporal logic and semidirect products: An effective characterization of the until hierarchy. In FOCS , pages 256-263. IEEE Computer Society, 1996. |
| null |
[
"Andreas Krebs",
"Howard Straubing"
] |
2014-08-04T20:29:08+00:00
|
2014-08-04T20:29:08+00:00
|
[
"cs.LO"
] |
EF+EX Forest Algebras
|
We examine languages of unranked forests definable using the temporal
operators EF and EX. We characterize the languages definable in this logic, and
various fragments thereof, using the syntactic forest algebras introduced by
Bojanczyk and Walukiewicz. Our algebraic characterizations yield efficient
algorithms for deciding when a given language of forests is definable in this
logic. The proofs are based on understanding the wreath product closures of a
few small algebras, for which we introduce a general ideal theory for forest
algebras. This combines ideas from the work of Bojanczyk and Walukiewicz for
the analogous logics on binary trees and from early work of Stiffler on wreath
product of finite semigroups.
|
1408.0810v2
|
## Galilean Conformal Electrodynamics
## Arjun Bagchi a,b , Rudranil Basu a,c , Aditya Mehra a
- a Indian Institute of Science Education and Research Dr Homi Bhabha Road, Pashan. Pune 411008. INDIA
- b Institute of Theoretical Physics, Vienna University of Technology Wiedner Hauptstr. 8-10/136, A-1040 Vienna, Austria
- c The Institute of Mathematical Sciences Tharamani, Chennai 600113 INDIA
- [email protected], [email protected]
[email protected]
Abstract: Maxwell's Electrodynamics admits two distinct Galilean limits called the Electric and Magnetic limits. We show that the equations of motion in both these limits are invariant under the Galilean Conformal Algebra in D = 4, thereby exhibiting non-relativistic conformal symmetries. Remarkably, the symmetries are infinite dimensional and thus Galilean Electrodynamics give us the first example of an infinitely extended Galilean Conformal Field Theory in D > 2. We examine details of the theory by looking at purely non-relativistic conformal methods and also use input from the limit of the relativistic theory.
## Contents
| 1. | Introduction | Introduction | 2 |
|------|----------------------------------------------------------|----------------------------------------------------------|-----|
| 2. | Conformal Invariance and Electrodynamics | Conformal Invariance and Electrodynamics | 4 |
| 3. | Non-relativistic Conformal Symmetries | Non-relativistic Conformal Symmetries | 7 |
| | 3.1 | Contraction and Infinite Extension | 7 |
| | 3.2 | The case of D = 2 | 9 |
| | 3.3 | Representations of GCA: Scale-Spin Primaries | 10 |
| 4. | Galilean Electrodynamics | Galilean Electrodynamics | 13 |
| | 4.1 | The Tale of Two Limits | 14 |
| | 4.2 | Galilean Conformal Transformations in GED: Contractions | 15 |
| 5. | Galilean Conformal Symmetries in GED: Intrinsic Analysis | Galilean Conformal Symmetries in GED: Intrinsic Analysis | 17 |
| | 5.1 | Invariance under GCA: Electric limit | 17 |
| | 5.2 | Invariance under GCA: Magnetic limit | 18 |
| | 5.3 | Correlation functions in GED | 20 |
| 6. | GED in D = 3 | GED in D = 3 | 24 |
| | 6.1 | Dual scalar formulation | 24 |
| | 6.2 | Invariance under GCA | 25 |
| 7. | Conclusions and Future Directions | Conclusions and Future Directions | 25 |
| | 7.1 | Summary of results | 25 |
| | 7.2 | Connections with Flatspace Holography | 27 |
| | 7.3 | Other future directions | 28 |
## 1. Introduction
Throughout history, symmetries have been central to our attempts at understanding Nature and symmetry principles continue to play very important roles in the construction of modern theoretical physics. In recent times, conformal symmetry has been a principal actor in many diverse sub-fields. Conformal symmetry and field theories endowed with conformal symmetry, viz. conformal field theories (CFT) have been the key to understanding of apparently unrelated subjects like critical exponents in statistical mechanics, correlations in the cosmic microwave background and many different aspects of string theory and quantum gravity.
Conformal symmetry is also known to appear as accidental symmetries of Maxwell's equations of electrodynamics in four spacetime dimensions. Although these symmetries become anomalous in the quantum regime, there are important lessons to be learnt here. When we look to generalise to Yang-Mills theories and then supersymmetrise to get to Supersymmetric Yang-Mills, conformal symmetries also survive in the quantum regime.
One of the cornerstones of research in the past decade and half has been the AdS/CFT correspondence which has provided a concrete exemplification of the Holographic Principle in the context of string theory. The Holographic Principle [1, 2] is the bold assertion that a theory of quantum gravity in a certain spacetime is completely equivalent to a theory without gravity living on the boundary of the same spacetime. The AdS/CFT correspondence realises this by relating Type IIB string theory on AdS 5 (times S 5 ) to N = 4 Super Yang-Mills (SYM) living on the boundary of AdS [3]. The quantum conformal invariance of N = 4 SYM [4] has been central to the enormous progress in the gauge-gravity duality.
The true power of conformal symmetry manifests itself in two spacetime dimensions, where the symmetry algebra is enhanced to two copies of the infinite dimensional Virasoro algebra [5]. Armed with infinite dimensional symmetries, one can now compute various quantities of interest in conformally invariant systems without even resorting to a Lagrangian description. There has been interest in trying to implement general symmetry arguments to understand CFTs in D=3 and D=4. These include trying to understand classification of primary operators and conformally invariant operator product expansions, conformal blocks and the conformal bootstrap procedure. While some of this goes back a long way [6, 7, 8], it has only recently been understood how to put these together to understand dynamical features about CFTs [9, 10, 11]. Of late there has been a resurgence of activities in this field with a special emphasis on trying to solve the 3D Ising model by general CFT methods like the conformal bootstrap [12].
In an ideal world, it would have been wonderful to have infinite symmetries in all spacetime dimensions which one could use to constrain various quantities of interest. This apparently utopian wish is actually fulfilled if one looks at Galilean Conformal Field Theories (GCFTs) which are the nonrelativistic limits of conformal field theories [13]. The relativistic conformal algebra undergoes an In¨n¨-Wigner contraction to the finite Galilean Conformal Algebra (f-GCA) with the same number o u of generators. It is then observed that the finite GCA can be given an infinite lift to an algebra which we will call the GCA (without the 'f'). In two dimensions, the idea of an infinitely extended non-relativistic algebra is not unexpected and it can indeed be shown that the GCA is just a con-
traction of (linear combinations of) two copies of the Virasoro algebra [14]. But the infinite algebra in dimensions D > 2 is indeed remarkable.
To justify the claim that there is indeed an enhancement to infinite dimensional symmetries in non-relativistic conformal systems for D > 2 and this is not just some mathematical jugglery, one needs physical examples. One partial example was provided in [13] where it was mentioned that the non-relativistic Navier-Stokes equations of hydrodynamics realises a part of the symmetries of the GCA. But so far, there has been no concrete example of a GCFT in dimensions D > 2. Our goal in this paper is to construct the first such example. For this we would be looking at the non-relativistic limit of Maxwell's electrodynamics in D = 4.
Galilean Electromagnetism was first discussed by Le Ballac and Levy-Leblond in 1973 [15]. They found that there were two distinct limits that they could perform on the theory, the Electric limit ( E >> B i i ) and the Magnetic limit ( B >> E i i ). Here, and elsewhere in this paper, we will set the speed of light c = 1. In our discussions below, we shall reformulate this into a scaling of the spacetime co-ordinates and two different scalings of the components of the gauge field A µ . We shall show that both sets of equations of Galilean Electrodynamics (GED) in the Electric and Magnetic limits exhibit invariance under the f-GCA in D = 4 and that there is a further infinite enhancement of symmetries to the full infinite dimensional GCA. We would be following recent work [16] which makes explicit the conformal structures of Maxwell's theory and would adapt our calculations to the non-relativistic case. Our paper would follow two parallel themes. We shall look at the limit of the appropriate analyses in the relativistic theory and also independently perform intrinsically non-relativistic calculations based on just the symmetries of the GCA. The analysis just in terms of the symmetries of the non-relativistic theory is obviously more general, as we do not expect all GCFTs to descend as limits of CFTs, even in the 2D case [14]. We stress that the intrinsic approach is essential for D > 2 in the studies of infinite extended GCA, which do not have any relativistic analogue. But, as we shall see, we need to resort to the limit to understand specifics of the particular example that we are dealing with. Certain constants which define the non-relativistic theory are determined through the limit from the relativistic theory.
Before we go into the details of the computations, let us provide some further justification of why we are interested in this study over and beyond the obviously important goal of trying to find the first example of a non-relativistic conformal field theory in D > 2 with infinite symmetries. As mentioned earlier, the quantum conformal invariance of N = 4 SYM is one of the main reasons why the AdS/CFT duality is so powerful. We expect that in the non-relativistic limit, Yang-Mills theory and subsequently Super Yang-Mills will continue to show invariance under the GCA. In particular, the hope is that in the supersymmetric setting, the non-relativistic limit of N = 4 SYM would exhibit GCA invariance also in the quantum domain. The existence of a non-relativistic sector with infinite dimensional symmetries raises very interesting possibilities. If this is a closed sub-sector in N = 4 SYM with infinite symmetries, it could very well turn out to be an integrable sub-sector and something that could be solved exactly, like the planar limit of N = 4 SYM.
Integrability is usually a feature of two dimensions and the reason why the planar limit of N = 4 SYM is integrable is because this is linked through the AdS/CFT duality to the integrability of the string theory world-sheet which is a 2D non-linear sigma model. The non-relativistic subsector
of N = 4 SYM could well be similarly linked to some limiting case of the string world sheet and hence exhibit similar features. It is interesting to point out that a tensionless limit of the closed string worldsheet actually exhibits symmetries of the 2D GCA [19]. It would be of great interest to explore these connections in more detail. But we will postpone discussions along these lines for future work.
Let us now briefly outline the contents of this paper. In Sec 2, we revisit the analysis of [16, 17] about the conformal invariance in relativistic electrodynamics which will help us set the stage for our non-relativistic ventures. In Sec 3, we recapitulate the construction of the GCA from the relativistic conformal symmetries. We then move on to constructing representations of the GCA. These are scale-spin representations as opposed to scale-boost ones which were considered earlier in [18] and play a central role in our analysis. In Sec 4, we first review the two distinct non-relativistic limits of [15] before reformulating this in the language of contractions of spacetime co-ordinates and gauge field components. The action of the conformal generators on the fields in the two limits is then studied. In Sec 5, we demonstrate the invariance of the two limiting theories under first the finite part of the GCA and then exhibit the enhancement to the full infinite dimensional symmetry. We go on to construct correlation functions for the gauge fields in both sectors. In Sec 6, we turn our attention to the formulation of electrodynamics in 3D with a scalar field and show that its non-relativistic limit again exhibits GCA invariance which is again infinite dimensional. This provides us with another example of a GCFT, now in D = 3. We close with a summary of results, connections to other related work and more future directions in Sec 7.
## 2. Conformal Invariance and Electrodynamics
Conformal invariance, as stressed in the introduction, is of great importance in many different branches of theoretical physics. In general D space-time dimensions, the conformal group consists of the Poincare group along with scaling and special conformal transformations (SCT). There has been a debate in the literature on whether the Poincare group along with scaling necessarily leads to conformal invariance. While this hold true in dimensions D = 2 for well-behaved quantum field theories (unitary, with a well defined energy-momentum tensor and a discrete spectrum) [20, 21], there are examples of scale invariant, non-conformal systems in higher dimensions, with classical electrodynamics being one of the simplest examples [16, 17].
Before we go into the details of conformal invariance in Maxwellian Electrodynamics, in order to set some notation, let us remind the reader of some details of the conformal algebra in D = 4. The Poincare sub-algebra consists of spacetime translations ( ˜ P i = ∂ , H i ˜ = -∂ t ), rotations ( ˜ J ij = x ∂ i j -x ∂ i j ) and boosts ( ˜ B i = x ∂ i t + t∂ i ). The other generators are the Dilatation operator ( ˜ D = -x · ∂ ) and the Special conformal generators ( ˜ K µ = -2 x µ ( x · ∂ ) + ( x · x ∂ ) µ ). The conformal algebra in D dimensions is isomorphic to so ( D, 2). Let us indicate a few important commutation relations below which will help us highlight the differences between the relativistic case considered here and the non-relativistic case that would be the central theme of this paper:
$$[ \tilde { P } _ { i }, \tilde { B } _ { j } ] = - \delta _ { i j } \tilde { H }, \ [ \tilde { B } _ { i }, \tilde { B } _ { j } ] = \tilde { J } _ { i j }, \ [ \tilde { K } _ { i }, \tilde { B } _ { j } ] = \delta _ { i j } \tilde { K }, \ [ \tilde { K } _ { i }, \tilde { P } _ { j } ] = 2 \tilde { J } _ { i j } + 2 \delta _ { i j } \tilde { D }. \quad \ ( 2. 1 )$$
We now go on to describe, following [16], the transformation laws of fields under conformal transformations. Let us consider transformations of a multi-component field Φ. For Poincare transformations, which leave the action for Φ invariant, we have
$$\tilde { \delta } _ { \mu } ^ { \text{trans} } \, \Phi ( x ) = \partial _ { \mu } \Phi ( x ), \quad \tilde { \delta } _ { \mu \nu } ^ { \text{Lorentas} } \, \Phi ( x ) = ( x _ { \mu } \partial _ { \nu } - x _ { \nu } \partial _ { \mu } + \Sigma _ { \mu \nu } ) \Phi ( x )$$
Transformations under scaling takes the following form:
$$\tilde { \delta } ^ { \text{scale} } \, \Phi ( x ) = ( x \cdot \partial + \tilde { \Delta } ) \Phi$$
where ∆ is the scaling dimension of the field Φ and to make the kinetic term of the corresponding ˜ action scale invariant one chooses
$$\tilde { \Delta } & = \frac { D - 2 } { 2 }.$$
The transformation under special conformal transformation can be obtained by looking the commutation relations of the conformal algebra. Specialising to the case of primary fields, we find that
$$\tilde { \delta } _ { \mu } ^ { \text{SCT} } \, \Phi ( x ) = \{ 2 x _ { \mu } ( x \cdot \partial ) - x ^ { 2 } \partial _ { \mu } + 2 \tilde { \Delta } x _ { \mu } - 2 x ^ { \nu } \Sigma _ { \nu \mu } \} \, \Phi ( x ).$$
We shall also be considering non-primary or descendent fields, like derivatives of primary fields, in our analysis. The transformation laws of these fields can be obtained from the transformations of their corresponding primary fields.
Let us now consider free Maxwell theory in D-dimensional spacetime. The theory is best expressed in terms of a field strength F µν = ∂ µ A ν -∂ A ν µ where A µ is the fundamental dynamic variable, the gauge field. F µν satisfies
$$\text{Equations of Motion} \colon \quad \partial ^ { \mu } F _ { \mu \nu } = 0$$
$$\text{Bianchi identity} \colon \quad \partial _ { \rho } F _ { \mu \nu } + \partial _ { \nu } F _ { \rho \mu } + \partial _ { \mu } F _ { \nu \rho } = 0.$$
The equations of motion can be derived from the well-known Lagrangian
$$\mathcal { L } = - \frac { 1 } { 4 } F _ { \mu \nu } F ^ { \mu \nu }$$
While the above Lagrangian is manifestly invariant under Poincare transformations:
$$\tilde { \delta } _ { \mu } ^ { \text{trans} } A _ { \nu } ( x ) = \partial _ { \mu } A _ { \nu } ( x )$$
$$\tilde { \delta } _ { \mu \nu } ^ { \text{Lorents} } A _ { \rho } ( x ) = ( x _ { \mu } \partial _ { \nu } - x _ { \nu } \partial _ { \mu } ) A _ { \rho } ( x ) + g _ { \rho \mu } A _ { \nu } ( x ) - g _ { \rho \nu } A _ { \mu } ( x ),$$
scale and special conformal transformations act non-trivially on it. In terms of the field variable A µ , the transformations are the following
$$\tilde { \delta } ^ { \text{scale} } A _ { \mu } ( x ) = ( x \cdot \partial + \tilde { \Delta } ) A _ { \mu } ( x )$$
$$\tilde { \delta } _ { \mu } ^ { \text{SCT} } \, A _ { \nu } ( x ) \, = \, \{ 2 x _ { \mu } ( x \cdot \partial ) - x ^ { 2 } \partial _ { \mu } + 2 \tilde { \Delta } x _ { \mu } \} A _ { \nu } ( x ) - 2 x _ { \nu } A _ { \mu } ( x ) + g _ { \mu \nu } x ^ { \rho } A _ { \rho } ( x ), \quad ( 2. 1 1 )$$
where the conformal dimension of the field is ∆ = ˜ D -2 2 . The last line is the conformal transformation where the gauge field has been specialised to a primary field. While to prove conformal invariance, it is important to show that the Lagrangian (2.8) is invariant up to a total divergence under conformal
transformations, we shall concentrate on understanding the conformal invariance of the equations of motion (2.6). This is because in the non-relativistic setting in which we are interested, there is no straightforward action or Lagrangian that one can construct. We would thus focus on the symmetries of the equations of motion. For this we require the transformations of the field strength. These can be derived directly from the transformations of the gauge fields that we have outlined above in (2.10). Under scaling and special conformal transformations, the field strength transforms as:
$$\tilde { \delta } ^ { \text{scale} } F _ { \mu \nu } ( x ) & = ( x \cdot \partial + \frac { D } { 2 } ) F _ { \mu \nu } ( x ) \\ \tilde { \delta } ^ { \text{scale} } \dots$$
$$\tilde { \delta } _ { \sigma } ^ { \text{SCT} } \, F _ { \mu \nu } ( x ) & = \Xi _ { \sigma } F _ { \mu \nu } + ( D - 4 ) ( g _ { \sigma \alpha } A _ { \beta } - g _ { \sigma \beta } A _ { \alpha } )$$
$$\text{where} \ \Xi _ { \sigma } F _ { \mu \nu } = ( 2 x _ { \sigma } x _ { \tau } - g _ { \sigma \tau } x ^ { 2 } ) \partial ^ { \tau } F _ { \mu \nu } + D x _ { \sigma } F _ { \mu \nu } + 2 ( g _ { \sigma \mu } x ^ { \tau } F _ { \tau \nu } + g _ { \sigma \nu } x ^ { \tau } F _ { \mu \tau } - x _ { \mu } F _ { \sigma \nu } - x _ { \nu } F _ { \mu \sigma } ).$$
Following [16], one can show that the Lagrangian (2.8) is invariant (up to total derivative terms) under dilatations. In the case of special conformal transformations, this is true only in dimension D = 4. The equations of motion show a similar behaviour. If they are to be invariant under special conformal transformations, ∂ µ ˜ δ SCT σ F µν should be zero. It can be shown by considering the above transformations that glyph[negationslash]
$$\partial ^ { \mu } \, \tilde { \delta } _ { \sigma } ^ { \text{SCT} } \, F _ { \mu \nu } = ( D - 4 ) ( \partial _ { \nu } A _ { \sigma } - g _ { \nu \sigma } \partial ^ { \tau } A _ { \tau } ) \neq 0 \quad \text{for} \quad D \neq 4$$
glyph[negationslash]
This implies that Maxwell's equations are only conformally invariant in D = 4. In dimensions D > 4, Maxwellian electrodynamics forms an example of a scale invariant but non-conformal relativistic theory. In D = 3, a special formulation of electrodynamics is possible where one can formulate everything in terms of a scalar potential and recover conformal symmetry [16]. We shall return to this discussion near the end of this paper.
Conformal invariance of Maxwell's equations in D = 4 or the lack of it in D > 4 can be expressed in terms of correlation functions of the theory starting from the Lagrangian (2.8) [17]. Although, we shall not be using this approach in our non-relativistic approach, it is good to keep this in mind. We eventually hope to have a Lagrangian description which would possibly depend on Newton-Cartan structures as outlined briefly near the end of the paper and this approach should prove valuable there. In position space in D = 4, the two-point function of gauge fields is given by
$$\left \langle A _ { \mu } ( x ) \, A _ { \nu } ( 0 ) \right \rangle = \frac { g _ { \mu \nu } } { x ^ { 2 } } + \text{gauge dependent terms}$$
Since this is gauge dependent, the correlator is obviously not a physical one, but it helps to define the two-point function of the gauge independent field strengths which in turn defines the theory.
$$\langle F _ { \mu \nu } F _ { \rho \sigma } \rangle = \frac { I _ { \mu \rho } I _ { \nu \sigma } - I _ { \mu \sigma } I _ { \nu \rho } } { x ^ { 4 } }, \quad \text{where } I _ { \mu \nu } = g _ { \mu \nu } - \frac { 2 x _ { \mu } x _ { \nu } } { x ^ { 2 } }$$
The stress energy tensor can be constructed out of the field strength:
$$T _ { \mu \nu } = F _ { \mu \rho } F _ { \nu } ^ { \rho } - \frac { 1 } { 4 } g _ { \mu \nu } F _ { \rho \sigma } F ^ { \rho \sigma }.$$
The common way of arriving at the above expression is by varying the Lagrangian (2.8) with respect to the metric. But [17] argues that it is enough to look at the expression (2.17) and construct its correlation functions and check if they are conserved and satisfy Ward identities. In order to now
glyph[negationslash]
check whether the Maxwellian theory is conformally invariant, one can compute the trace of the energy momentum tensor which in D = 4 turns out to be zero. This however is not as straightforward in dimensions other that four. Although one can straightaway check that the trace of the EM tensor defined through (2.17) is non-zero when D = 4:
$$T _ { \mu } ^ { \mu } & = \frac { D - 4 } { 4 } F _ { \mu \nu } F ^ { \mu \nu }$$
one can add 'improvement' terms to the EM tensor which may potentially make the trace zero. It can be argued that this is not possible for Maxwell's theory [17]. Another way of testing the presence of conformal symmetry is to look at the correlators of the theory. One can look at the operator Φ = : ( F µν ) 2 : and look for example at its three point function which in a conformal field theory is fixed by symmetries. The three point function 〈 Φ( x )Φ( y )Φ( z ) 〉 actually vanishes in D = 4 and this is consistent with the assertion that this is a primary in the CFT. The structure of (2.16) is also consistent with the two point function of a tensor valued anti-symmetric primary.
We shall not be adopting this approach as our non-relativistic construction does not have a straight forward Lagrangian description and hence the construction of correlation functions is a problem. In the non-relativistic conformal case in D = 4, we would be able to construct correlation functions from the symmetries. But presently, we do not have a way of cross-checking the forms of the correlators with explicit answers from an action principle.
## 3. Non-relativistic Conformal Symmetries
## 3.1 Contraction and Infinite Extension
We wish to consider the non-relativistic equivalent of the conformal symmetry. For this we shall consider the Inonu-Wigner contraction of the conformal algebra. We shall be performing this limit at the level of the spacetime and the contraction will be performed in units where the speed of light c = 1. In the non-relativistic limit, velocities would be small compared to the speed of light and hence in these units, it should correspond to v i → 0. We shall achieve this by the parametric contraction
$$x _ { i } \to \epsilon x _ { i }, \ t \to t, \ \epsilon \to 0$$
so that v i ∼ x i t → 0. In a contraction, the number of generators does not change from the parent to the daughter algebra, but the expressions for some of the generators change. Let us demonstrate this for one example below.
$$\tilde { B } _ { i } = t \partial _ { i } + x _ { i } \partial _ { t } \rightarrow \frac { 1 } { \epsilon } t \partial _ { i } + \epsilon x _ { i } \partial _ { t } \Rightarrow B _ { i } = \lim _ { \epsilon \rightarrow 0 } \epsilon \tilde { B } _ { i } = t \partial _ { i }$$
The algebra we will get after the contraction will be called the finite Galilean Conformal Algebra (f-GCA). The generators of the f-GCA are
$$H = - \partial _ { t }, \ \ D = - ( t \partial _ { t } + x _ { i } \partial _ { i } ), \ \ K = - ( t ^ { 2 } \partial _ { t } + 2 x ^ { i } t \partial _ { i } )$$
$$P _ { i } = \partial _ { i }, \ \ B _ { i } = t \partial _ { i }, \ \ K _ { i } = t ^ { 2 } \partial _ { i }$$
$$J _ { i j } & = x _ { [ i } \partial _ { j ] }$$
where [] is the anti-symmetrizer and is used without any combinatorial factor.
Here the Galilean subalgebra is generated by { H,P ,B ,J i i ij } . When we compute the commutators, we see that the difference with the relativistic conformal algebra arises in the analogues of (2.1). The right-hand sides of all these commutators vanish in the non-relativistic limit. We see that the generators can be written in the suggestive form for n = 0 , ± 1:
$$L ^ { ( n ) } = - t ^ { n + 1 } \partial _ { t } - ( n + 1 ) t ^ { n } x _ { i } \partial _ { i } \ \ ( L ^ { ( - 1, 0, + 1 ) } = H, D, K )$$
$$M _ { i } ^ { ( n ) } = t ^ { n + 1 } \partial _ { i } \ \ ( M _ { i } ^ { ( - 1, 0, + 1 ) } = P _ { i }, B _ { i }, K _ { i } )$$
The identifications with the contracted generators (3.3) are written in parenthesis. The f-GCA can be written as
$$& [ L ^ { ( n ) }, L ^ { ( m ) } ] = ( n - m ) L ^ { ( n + m ) }, \quad [ L ^ { ( n ) }, M _ { i } ^ { ( m ) } ] = ( n - m ) M _ { i } ^ { ( n + m ) } \quad [ M _ { i } ^ { ( n ) }, M _ { j } ^ { ( m ) } ] = 0 \quad ( 3. 8 ) \\ & [ L ^ { ( n ) }, J _ { i j } ] = 0, \quad [ M _ { i } ^ { ( n ) }, J _ { j k } ] = M _ { [ k } ^ { ( n ) } \delta _ { j ] i }$$
We now observe that the above algebra continues to hold even if we consider generators (3.6) for any integer value of n . 1 The generators { L n } form a sub-algebra which with suitable central extensions becomes a Virasoro algebra. The symmetry algebra above (3.8) is an infinite dimensional algebra for any space-time dimension. We have hence the tantalising prospect of infinite symmetry enhancements in the non-relativistic limit of a conformal field theory in any dimensions.
The obvious question is whether this infinite dimensional symmetry is realised in any physical example. As we will go on to describe next, in D = 2, there are surprising appearances of this algebra in different apparently unconnected fields. In D > 2, we have the example of hydrodynamics. The non-relativistic Navier-Stokes equation is invariant under arbitrary time-dependent accelerations, the M ( n ) i 's of the above algebra, thus realising an infinite dimensional sub-set. The viscous term in the Navier-Stokes introduces a scale in the system and hence the dilatation operator is no longer a symmetry. So the Navier-Stokes equation only realises a part of the infinte dimensional GCA. If we turn off viscosity to move to the Euler equation, the f-GCA becomes a symmetry. But the higher L n modes do not leave the equation invariant.
So, before the present paper, there have been no examples in D > 2 of systems realising the full infinite dimensional GCA as their symmetry algebra.
1 In earlier work, it has been noted that even the rotation generators could be given an infinite lift following
$$J _ { a } ^ { ( n ) } \equiv J _ { i j } ^ { ( n ) } = t ^ { n } ( x _ { i } \partial _ { j } - x _ { j } \partial _ { i } ) & & ( 3. 1 0 )$$
These generators constitute the following infinite dimensional algebra
$$& [ L ^ { ( n ) }, L ^ { ( m ) } ] = ( n - m ) \, L ^ { ( n + m ) }, \quad [ L ^ { ( n ) }, M ^ { ( m ) } _ { i } ] = ( n - m ) \, M ^ { ( n + m ) } _ { i }, \quad [ M ^ { ( n ) } _ { i }, M ^ { ( m ) } _ { j } ] = 0 \\ & [ J ^ { ( n ) } _ { a }, J ^ { ( m ) } _ { b } ] = f _ { a b } ^ { c } \, J ^ { ( n + m ) } _ { c }, \quad [ L ^ { ( n ) }, J ^ { ( m ) } _ { a } ] = m \, J ^ { ( n + m ) } _ { a } \quad [ M ^ { ( n ) } _ { i }, J ^ { ( m ) } _ { j k } ] = M ^ { ( n + m ) } _ { [ k } \delta _ { j ] i }.$$
In this paper, we would not be considering the infinite dimensional lift for the rotation generators. This is primarily because of some complications in the analysis to follow which we would like to avoid for the present considerations. It is important to note that in the supersymmetric case [44], it has been noted that the rotations cannot be given this infinite lift. Since our ultimate goal is to look at Supersymmetric Yang-Mills theories, the present restriction would not be an issue when one generalises to the supersymmetric analysis.
## 3.2 The case of D = 2
In D = 2, the appearance of an infinite non-relativistic conformal algebra is much less surprising given the fact that the 2d relativistic conformal algebra itself is enhanced to two copies of the infinite dimensional Virasoro algebra. The 2d GCA can actually be obtained by contraction of linear combinations of these two Virasoro algebras. If we denote L n , ¯ L n as the two copies of relativistic Virasoro:
$$[ \mathcal { L } _ { n }, \mathcal { L } _ { m } ] = ( n - m ) \mathcal { L } _ { n + m } + \frac { c } { 1 2 } \delta _ { n + m, 0 } ( n ^ { 3 } - n ), \ \ [ \mathcal { L } _ { n }, \bar { \mathcal { L } } _ { m } ] = 0$$
$$[ \bar { \mathcal { L } } _ { n }, \bar { \mathcal { L } } _ { m } ] = ( n - m ) \bar { \mathcal { L } } _ { n + m } + \frac { \bar { \bar { c } } } { 1 2 } \delta _ { n + m, 0 } ( n ^ { 3 } - n )$$
then the 2d GCA is obtained by looking at the limit
$$L _ { n } = \mathcal { L } _ { n } + \bar { \mathcal { L } } _ { n }, \ \ M _ { n } = \epsilon ( \mathcal { L } _ { n } - \bar { \mathcal { L } } _ { n } ).$$
This leads to the 2d GCA:
$$& [ L _ { n }, L _ { m } ] = ( n - m ) L _ { n + m } + \frac { c _ { L } } { 1 2 } \, \delta _ { n + m, 0 } ( n ^ { 3 } - n ), & ( 3. 1 5 ) \\ & [ L _ { n }, M _ { m } ] = ( n - m ) M _ { n + m } + \frac { c _ { M } } { 1 2 } \, \delta _ { n + m, 0 } ( n ^ { 3 } - n ), \quad [ M _ { n }, M _ { m } ] = 0$$
It is interesting to note that in this case, the GCA in 2d admits a central extension between L n and M m . This in higher dimensions would be a vector and hence not a central charge. The two central extensions are related to their relativistic versions by the relations:
$$c _ { L } = c + \bar { c }, \ \ c _ { M } = \epsilon ( c - \bar { c } ).$$
The fact that (3.14) is a non-relativistic limit can be understood by looking at the familiar representations of the Witt algebra in terms of holomorphic and anti-holomorphic co-ordinates, rewriting them in terms of x, t and performing the contraction (3.1)
$$\mathcal { L } _ { n } = z ^ { n + 2 } \partial _ { z }, \ \bar { \mathcal { L } } _ { n } = \bar { z } ^ { n + 2 } \partial _ { \bar { z } } ; \ \ z, \bar { z } = t \pm \epsilon x$$
The 2d GCA can also be obtained by another contraction which combines the generators as
$$L _ { n } = \mathcal { L } _ { n } - \bar { \mathcal { L } } _ { - n }, \ \ M _ { n } = \epsilon ( \mathcal { L } _ { n } + \bar { \mathcal { L } } _ { - n } ).$$
It can be easily checked that again the commutators (3.15) are reproduced. The identifications of the central terms also change to c L = c -¯ and c c M = glyph[epsilon1] ( c + ¯). c From the point of view of spacetime, (3.17) actually can be motivated from an ultra-relativistic contraction as opposed to the non-relativistic one (3.14). For this, it is important to look at the equivalent of the expressions (3.16) on the cylinder and look at the contraction which sends the speed of light to zero instead of infinity. This contraction is important for considerations in flat holography. It turns out that the asymptotic symmetry algebra of flat space in three dimensions [24, 25, 26] is isomorphic to the 2d GCA and hence a dual field theory to 3d flat spacetimes is actually a Galilean Conformal Field Theory. This fact has been exploited to build aspects of flat holography in 3d [27]. A incomplete set of references include [27] - [42]. It is interesting to point out here that there has been the construction of an
explicit field theory example where the flat limit of Liouville theory was performed and the Poisson algebra of the conserved charges realise the infinite dimensional algebra (3.15) [33].
Apart from the uses in non-relativistic physics and also in the development of flat space holography in 3D, it has also been noticed that the 2D GCFTs would have a central role to play in the tensionless limit of bosonic string theory. The symmetries (3.15) arise as residual symmetries on the world sheet in the 'conformal' gauge of the tensionless closed string, and this is analogous to the appearance of the 2d conformal symmetry in the case of the usual tensile bosonic closed string.
The unexpected appearance of the GCA in various apparently unrelated fields of physics seem to point to a universality in the behaviour of limits of conformally invariant systems. It is increasingly becoming of importance to explore these symmetries and our current paper is another very important step in this direction as it deals with the appearance of a GCFT in its full infinite dimensional form, now in a higher dimensional example.
## 3.3 Representations of GCA: Scale-Spin Primaries
The representations of the GCA can be constructed along lines similar to that of the relativistic conformal algebra. We will label the states with their weights under the dilatation operator L (0) .
$$L ^ { ( 0 ) } | \Phi \rangle = \Delta | \Phi \rangle$$
We notice that the weights are lowered by the operators L ( n ) , M ( n ) i for positive n and raised by ones with negative n .
$$L ^ { ( 0 ) } L ^ { ( n ) } | \Phi \rangle = ( \Delta - n ) L ^ { ( n ) } | \Phi \rangle, \ \ L ^ { ( 0 ) } M _ { i } ^ { ( n ) } | \Phi \rangle = ( \Delta - n ) M _ { i } ^ { ( n ) } | \Phi \rangle.$$
This helps us build primary states in the theory in analogy with CFTs by demanding that the spectrum be bounded from below. We define primary states as
$$L ^ { ( n ) } | \Phi \rangle _ { p } = M _ { i } ^ { ( n ) } | \Phi \rangle _ { p } = 0 \ \forall n > 0$$
We now have two different options in labelling these states further. Since glyph[negationslash]
$$[ L ^ { ( 0 ) }, M _ { i } ^ { ( 0 ) } ] = 0 \ \ [ L ^ { ( 0 ) }, J _ { i j } ] = 0 \ \ [ M _ { i } ^ { ( 0 ) }, J _ { i j } ] \neq 0,$$
we can choose to label the states further by the weights under M (0) i or J ij , but not with both of them. In [18], we chose to label the primaries with M (0) i and this was also useful in 2D [13] as this was identified with L -L 0 ¯ 0 in the non-relativistic limit. But in this paper, we shall be interested in understanding operators with spin 2 and hence we will look at primaries which are labelled under J ij .
$$J _ { i j } ^ { ( 0 ) } | \Phi \rangle _ { p } = \Sigma _ { i j } \rhd | \Phi \rangle _ { p }$$
where Σ ij glyph[triangleright] denotes action of J ij in that particular representation of SO D ( -1), to which | Φ 〉 p belongs. The representations of the GCFT are built by acting with the raising operators ( L ( n ) , M ( n ) i for negative n ) on these primary states.
2 At this stage, by 'spin' we would mean the weight vector or the set of all quantum numbers specifying a particular irreducible representation of so ( D -1) generated by J ij for D ≥ 4
For sake of explicitly studying the action of GCA on the operators of our theory, we would use the state-operator correspondence and would denote by Φ( t, x ) primary operator(s) relating the primary state and the vacuum:
$$| \Phi \rangle _ { p } = \Phi ( 0, 0 ) | 0 \rangle.$$
We would assume as usual, the action of the finite part of GCA on these primaries consistently with (3.19), (3.18) as:
$$\text{as.} \\ \quad \left [ L ^ { ( - 1 ) }, \Phi ( t, x ) \right ] = \partial _ { t } \Phi ( t, x ) \quad \right ( 3. 2 1 a ) \\ \left [ M _ { i } ^ { ( - 1 ) }, \Phi ( t, x ) \right ] = - \partial _ { i } \Phi ( t, x ) \quad \right ( 3. 2 1 b ) \\ \left [ J _ { i j }, \Phi ( 0, 0 ) \right ] = \Sigma _ { i j } \rhd \Phi ( 0, 0 ) \quad \right ( 3. 2 1 c ) \\ \left [ L ^ { ( 0 ) }, \Phi ( 0, 0 ) \right ] = \Delta \Phi ( 0, 0 ) \quad \right ( 3. 2 1 d ) \\ \left [ L ^ { ( + 1 ) }, \Phi ( 0, 0 ) \right ] = 0 = \left [ M _ { i } ^ { ( + 1 ) }, \Phi ( 0, 0 ) \right ] \quad \right ) \quad \left ( 3. 2 1 e \right ) \\ \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon$$
Let's now briefly recapitulate how the actions of J ij , L (0 1) , and M (+1) i can be worked out on an operator at general space-time points ( t, x ). According to our conventions,
$$\Phi ( t, x ) = U \Phi ( 0, 0 ) U ^ { - 1 }$$
where U = e ( tL ( -1) -x M i ( -1) i ) . For a general GCA element glyph[star] , we therefore use:
$$\left [ \tilde { \kappa }, \Phi ( t, x ) \right ] = U \left [ U ^ { - 1 } \tilde { \kappa } U, \Phi ( 0, 0 ) \right ] U ^ { - 1 }$$
and then exploit the Baker-Campbell-Hausdorff formula (BCH) and GCA (3.8) to evaluate U -1 glyph[star] U . This is straightforward for J ij and L (0) :
$$& [ J _ { i j }, \Phi ( t, x ) ] = x _ { [ i } \partial _ { j ] } \Phi ( t, x ) + \Sigma _ { i j } \rhd \Phi ( t, x ) \\ & \left [ L ^ { ( 0 ) }, \Phi ( t, x ) \right ] = ( t \partial _ { t } + x ^ { i } \partial _ { i } + \Delta ) \Phi ( t, x )$$
However we encounter a problem as soon as we try to find the action of L (+1) and M (+1) i , i.e. the generators of Galilean version of special conformal transformation. This is because we need the role of boost operator for these evaluations. We have avoided it up to this point, since we do not want boost to label our primaries. This is the opportune moment to digress on the action of B i (which is M (0) i according to our notation).
In this direction, let us look at the following equation found from the relevant Jacobi identity:
$$\left [ J _ { i j }, \left [ B _ { k }, \Phi ( 0, 0 ) \right ] \right ] = \left [ B _ { k }, \Sigma _ { i j } \rhd \Phi ( 0, 0 ) \right ] + \delta _ { k [ i } \left [ B _ { j ] }, \Phi ( 0 ) \right ].$$
Moreover, a similar Jacobi identity guarantees that B i preserves scaling dimension of Φ. With Galilean ED as a motivation, we study following two cases explicitly.
## · Φ is so ( D -1) scalar
Let φ denote Φ(0 0) here. Hence from (3.24) we infer: ,
$$[ J _ { i j }, [ B _ { k }, \phi ] ] = \delta _ { k [ i } \left [ B _ { j }, \phi \right ]$$
In its most general form:
$$[ B _ { i }, \phi ] = a _ { i } \, \phi + b \, \phi _ { i } + \text{ higher spin operators}$$
where a i is a c-number valued vector and φ i is a vector operator (spin-1 for D = 4) at spacetime origin. Note that presence of a i as a c-number valued vector breaks the SO D ( -1). Hence, we must have a i = 0 and
$$[ B _ { i }, \phi ] = b \, \phi _ { i } + \text{ higher spin operators}$$
Structure of the higher spin (tensor) operators in general could only have a term of the form: ε i i 1 2 i ...i D -2 φ i 1 2 i ...i D -2 , since ε is the only so ( D -1) invariant tensor other than δ ij . However, the right hand side of (3.24) transforms in a vector representation, negating the possibility of any higher spin operator as a consequence. On the other hand, if one prefers not to have rotation as an invariance, a i essentially measures the 'rapidity' of φ , in case b = 0, as in [18].
One must keep note of the fact that both the scalar φ and the vector it transforms into φ i , share the same scaling dimension.
## · Φ is so ( D -1) vector
Let ˜ φ i denote Φ(0 0) here. Hence from (3.24) we have: ,
$$\left [ J _ { i j }, \left [ B _ { k }, \tilde { \phi } _ { l } \right ] \right ] = \left [ B _ { k }, \delta _ { l [ i } \tilde { \phi } _ { j ] } \right ] + \delta _ { k [ i } \left [ B _ { j ] }, \tilde { \phi } _ { l } \right ]$$
We use the rotational symmetry arguments presented in the earlier case again to solve this most generically by
$$\left \lceil B _ { i }, \tilde { \phi } _ { j } \right \rceil = a \delta _ { i j } \tilde { \phi },$$
where a is a c-number and ˜ is a spin-less operator. φ In general these ˜ and φ ˜ φ i need not have to do anything with the φ and φ i appearing in the previous case. However, if we demand (which we should, for the case of concrete example of Galilean Electrodynamics) φ and φ i to be of same scaling dimension, then φ ∼ ˜ and φ φ i ∼ ˜ φ i upto some real multiplier, which we choose to be 1 in what follows.
We see that we are yet to determine the constants a, b appearing in (3.27), (3.29) using all of kinematical symmetry. Another way of saying this is that our representations are labelled by two constants, in addition to weight and spin. It will become apparent in the following analysis that a, b get fixed for particular physical systems. Determination of these constants demand inputs from dynamics.
With the above information at hand, it is easy to generalize the action of boost on operators at finite space-time points:
$$[ B _ { i }, \phi ( t, x ) ] \, = \, - t \partial _ { i } \phi ( t, x ) + b \phi _ { i } ( t, x )$$
$$[ B _ { i }, \phi _ { j } ( t, x ) ] \, = \, - t \partial _ { i } \phi _ { j } ( t, x ) + a \delta _ { i j } \phi ( t, x )$$
Now we are ready to study the action of L (+1) and M (+1) i on our scalar and vector primaries of weight ∆ at arbitrary space-time points. We use the scheme of using BCH and (3.8) outlined above
in deriving (3.23),(3.30):
$$\left [ L ^ { ( + 1 ) }, \phi ( t, x ) \right ] \, = \, ( t ^ { 2 } \partial _ { t } + 2 t x ^ { i } \partial _ { i } + 2 t \Delta ) \phi ( t, x ) - 2 b \, x ^ { i } \phi _ { i } ( t, x ) \, \quad \ \ ( 3. 3 1 \mathbf a )$$
$$\left [ L ^ { ( + 1 ) }, \phi _ { i } ( t, x ) \right ] \, = \, ( t ^ { 2 } \partial _ { t } + 2 t x ^ { i } \partial _ { i } + 2 t \Delta ) \phi _ { i } ( t, x ) - 2 a \, x ^ { i } \phi ( t, x ) \, \quad \ \ ( 3. 3 1 b )$$
$$\left [ M _ { i } ^ { ( + 1 ) }, \phi ( t, x ) \right ] \, = \, - t ^ { 2 } \partial _ { i } \phi ( t, x ) + 2 b t \phi _ { i } ( t, x ) \, & & ( 3. 3 1 c )$$
$$\left [ M _ { i } ^ { ( + 1 ) }, \phi _ { j } ( t, x ) \right ] \, = \, - t ^ { 2 } \partial _ { i } \phi _ { j } ( t, x ) + 2 a \, t \delta _ { i j } \phi ( t, x ) & & ( 3. 3 1 d )$$
Let's pause for a moment to take stock of the progress made so far in accomplishing our goal of studying action of GCA on scale-spin primaries. From the equations (3.21a), (3.21b), (3.23), (3.30) and (3.31) we conclude that for the global or finite part of GCA, we have successfully produced what we desired to, additionally labelled by two numbers a, b . Conceptually, it should not be hard to do the same for infinite extension of GCA (3.8). The essentially non-trivial computational step lies in simplification of adjoint action of Galilean Hamiltonian and momentum operators on arbitrary modes of GCA:
$$U ^ { - 1 } \tilde { \kappa } ^ { ( n ) } U.$$
This can be performed again appealing to BCH and keeping in mind that U = e ( tL ( -1) -x M i ( -1) i ) and repeatedly using (3.8) [18]. To be more concrete, for non-negative modes:
$$U ^ { - 1 } L ^ { ( n ) } U = \sum _ { \substack { k = 0 \\ \cdot \cdot \cdot } } ^ { n + 1 } \binom { n } { k } \left ( t ^ { k } L ^ { ( n - k ) } - k \, t ^ { k - 1 } x ^ { i } M _ { i } ^ { ( n - k ) } \right ) \quad \quad ( 3. 3 2 a )$$
$$U ^ { - 1 } M _ { i } ^ { ( n ) } U & = \sum _ { k = 0 } ^ { n + 1 } \binom { n } { k } t ^ { k } \, M _ { i } ^ { ( n - k ) }$$
Using these, we arrive at (here we have assumed, as usual, that the positive modes of GCA generators annihilate the primaries.):
$$\cdot \, \begin{array} { c } \cdot & \, ^ { \prime } \\ \left [ L ^ { ( n ) }, \phi ( t, x ) \right ] & = t ^ { n } \left ( t \partial _ { t } + ( n + 1 ) x ^ { i } \, \partial _ { i } + ( n + 1 ) \Delta \right ) \phi ( t, x ) \\ & - \, b \, n \, ( n + 1 ) t ^ { n - 1 } \, x ^ { i } \phi _ { i } ( t, x ) \\ \left [ L ^ { ( n ) }, \phi _ { i } ( t, x ) \right ] & = t ^ { n } \left ( t \partial _ { t } + ( n + 1 ) x ^ { j } \, \partial _ { j } + ( n + 1 ) \Delta \right ) \phi _ { i } ( t, x ) \\ & \, \cdot \, \end{array} \cdot$$
$$\left [ L ^ { n }, \varphi ( t, x ) \right ] & = t ^ { n } \left ( t \partial _ { t } + ( n + 1 ) x ^ { 2 } \, \partial _ { i } + ( n + 1 ) \Delta \right ) \varphi ( t, x ) \\ & - b \, n \, ( n + 1 ) t ^ { n - 1 } \, x ^ { i } \phi _ { i } ( t, x ) & ( 3. 3 3 a ) \\ \left [ L ^ { ( n ) }, \phi _ { i } ( t, x ) \right ] & = t ^ { n } \left ( t \partial _ { t } + ( n + 1 ) x ^ { j } \, \partial _ { j } + ( n + 1 ) \Delta \right ) \phi _ { i } ( t, x ) \\ & - \, a \, n \, ( n + 1 ) t ^ { n - 1 } \, x _ { i } \phi ( t, x ) & ( 3. 3 3 b ) \\ \left [ M _ { i } ^ { ( n ) }, \phi ( t, x ) \right ] & = - t ^ { n + 1 } \partial _ { i } \phi ( t, x ) + b \, ( n + 1 ) \, t ^ { n } \phi _ { i } ( t, x ) & ( 3. 3 3 c ) \\ \left [ M _ { i } ^ { ( n ) }, \phi _ { j } ( t, x ) \right ] & = - t ^ { n + 1 } \partial _ { i } \phi _ { j } ( t, x ) + a \, ( n + 1 ) \, t ^ { n } \delta _ { i j } \phi ( t, x ) & ( 3. 3 3 d ) \\ \intertext { these at hand we have successfully constructed modules defined by the quadrunelet $\Lambda$\ \, $N$} \, a b$$
With these at hand, we have successfully constructed modules defined by the quadruplet ∆ Σ , , a, b serving as representations of GCA.
## 4. Galilean Electrodynamics
In this section, following the work by Le Bellac and Levy-Leblond [15], we first briefly review the Galilean version of Maxwell's electrodynamics. Then we reformulate the same in terms of scalings of the spacetime and the components of the gauge field. We then go on to describe the action of the Galilean Conformal generators on GED by implementing this limiting procedure from relativistic electrodynamics.
## 4.1 The Tale of Two Limits
The fundamental observation of [15] was that Maxwell's theory admits two different non-relativistic limits . This can be understood from the usual Lorentz transformation of a four-vector u µ = ( u , u 0 i ) which admits two different Galilean limits:
$$u _ { 0 } ^ { \prime } = u _ { 0 }, \ \ u _ { i } ^ { \prime } = u _ { i } - v _ { i } u _ { 0 }$$
$$u _ { 0 } ^ { \prime } = u _ { 0 } - v ^ { i } u _ { i }, \ \ u _ { i } ^ { \prime } = u _ { i }$$
Here we have set the speed of light c = 1 and v i is the velocity of the transformation. (4.1) applies when v << 1 and u 0 >> √ u 2 i and (4.2) applies when v << 1 and u 0 << √ u 2 i . In the case of Electromagnetism, (4.1) would correspond to the case where | ρ | >> glyph[vector] | j | and is called the Electric limit. In this limit, true to its name, the electric effects dominate over the magnetic ones. (4.2) applies when | ρ | << glyph[vector] | j | and is called the Magnetic limit. Here magnetic effects dominate over the electric ones. It is straight-forward to write down the analogues of the Maxwell equations in these limits. These are displayed below.
$$\underline { \text{Electric limit} } _ { i } E _ { i } ^ { e } = \frac { \rho _ { e } } { \varepsilon _ { 0 } }, \ \ \epsilon _ { i j k } \partial _ { j } E _ { k } ^ { e } = 0, \ \ \partial _ { i } B _ { i } ^ { e } = 0, \ \ \epsilon _ { i j k } \partial _ { j } B _ { k } ^ { e } = \varepsilon _ { 0 } \mu _ { 0 } \partial _ { t } E _ { i } ^ { e } + \mu _ { 0 } j _ { i } ^ { e }. \quad \ \ ( 4. 3 )$$
$$\frac { \text{Magnetic limit} \colon } { \partial _ { i } E _ { i } ^ { m } = \frac { \rho _ { m } } { \varepsilon _ { 0 } }, \ \ \epsilon _ { i j k } \partial _ { j } E _ { k } ^ { m } = - \partial _ { t } B _ { i } ^ { m }, \ \ \partial _ { i } B _ { i } ^ { m } = 0, \ \ \epsilon _ { i j k } \partial _ { j } B _ { k } ^ { m } = \mu _ { 0 } j _ { i } ^ { m }. \quad \ \ ( 4. 4 ) }$$
To connect to our formalism of previous sections, we shall re-derive the above equations (4.3), (4.4) in terms of our scaling limits. We shall work in terms of the gauge fields. We continue to use the non-relativistic scaling of the co-ordinates (3.1). Now we have the additional freedom of choosing the scaling of the gauge field components. We shall scale these in the following way:
$$\text{Electric limit} \colon \ A _ { 0 } \to A _ { 0 }, \, A _ { i } \to \epsilon A _ { i } & & ( 4. 5 a )$$
$$\text{Magnetic limit} \colon \ A _ { 0 } \to \epsilon A _ { 0 }, \, A _ { i } \to A _ { i } & & ( 4. 5 b )$$
For the moment, let us consider having switched off all sources. It is then easy to check that the source-free Galilean Electrodynamic equations in the two different limits are obtained from the relativistic equations of motion
$$\partial _ { \mu } F ^ { \mu \nu } = 0$$
by looking at the above contractions of the gauge fields. Suitably rescaled constants ε , µ 0 0 give rise to the equations (4.3) - (4.4). The 'equations of motion' in the limit defined by (4.5a) for source-free case are:
$$\partial ^ { i } \partial _ { i } A _ { 0 } & = 0 & ( 4. 6 a )$$
$$\partial ^ { j } \partial _ { j } A _ { i } - \partial _ { i } \partial _ { j } A ^ { j } - \partial _ { t } \partial _ { i } A _ { 0 } = 0$$
whereas for the magnetic limit (4.5b) these are:
$$( \partial ^ { j } \partial _ { j } ) A _ { i } - \partial _ { i } \partial _ { j } A ^ { j } = 0$$
$$\partial _ { i } \partial _ { t } A ^ { i } + ( \partial ^ { j } \partial _ { j } ) A _ { 0 } = 0.$$
We have used the trivial metric δ ij on R 3 for all tensor contractions. Although we need not distinguish between contra or covariant indices, we use them purely for aesthetic reasons.
## 4.2 Galilean Conformal Transformations in GED: Contractions
In Sec 2, we looked at the conformal invariance of Maxwell's electromagnetism by referring to the transformations of the fields under the conformal group. The fields A µ transformed non-trivially under dilatation and special conformal transformation (2.10). Here we wish to find the non-relativistic versions of these transformations. In Sec 3.3 we worked out, purely from the algebra of Galilean conformal symmetry, its action on SO D ( -1) scalars and vectors with definite scaling dimension. However there is another, more straightforward way of looking at this analysis. As pointed out in Sec 3.1, the finite part of GCA can be extracted as a contraction of the relativistic conformal algebra. In Galilean Electrodynamics, we have just discussed two distinct physically meaningful limits the contraction naturally induces on the field space (4.5). We would exploit these two contractions (space-time ones, reflecting through generators and ones acting on field space) directly on well-known fully relativistic conformal transformation laws to see the parallels of (3.31). More specifically we find the Galilean versions of (2.9),(2.10) apart from usual Poincare transformations on the relativistic 1-form gauge field in D space-time (Minkowski) dimension. As a prescription, Galileanization of transformation rules (2.9),(2.10) would mean performing the following steps:
- 1. First we perform a canonical space-time split of the transformation formula.
- 2. Then replace all relevant objects (ie, space and time coordinates, derivative operators with respect to them, field variables) with their transformed version using (3.1) and (4.5).
- 3. This would reduce the right hand side of the space-time split version of equations (2.9), (2.10) to the form:
$$\sum _ { n = p _ { 1 } } ^ { p _ { 2 } } \epsilon ^ { n } f _ { \mathcal { A } } ^ { ( n ) } ( x, A, \partial A )$$
p 1 ≤ p 2 and both of them are integers, positive negative or zero, in general. A stands for arbitrary index structure.
- 4. In view of the above, now define Galilean version of (2.9)(2.10) as:
$$\tilde { \delta } _ { \mathcal { A } } A = \lim _ { \epsilon \rightarrow 0 } \epsilon ^ { - p _ { 1 } } \sum _ { n = p _ { 1 } } ^ { p _ { 2 } } \epsilon ^ { n } f _ { \mathcal { A } } ^ { ( n ) } ( x, A, \partial A )$$
This last step essentially extracts out the finite part of the relativistic transformation in the limit.
The two distinct contractions on fields (4.5) are expected to give different transformation rules under the plan of contraction outlined above. Let's tabulate all of them consecutively for the electric limit (4.5a) first.
Now let us focus our attention on the two parts of special conformal transformation in its Galileanized version, for the scalar and the vector parts. Comparing with (3.31), it is easy to read out that the arbitrary constants that would label our representations of GCA get indeed fixed as a e = -1 , b e = 0
Table 1: Galilean transformations in the Electric limit
| Relativistic Transformation | Galileanized Transformation: Electric version | p 1 |
|-------------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|----------|
| Translation | ˜ δ trans 0 A 0 = ∂ t A 0 =: [ H,A 0 ( t, x )] ˜ δ trans 0 A i = ∂ t A i =: [ H,A i ( t, x )] ˜ δ trans i A 0 = - ∂ i A 0 =: [ P i , A 0 ( t, x )] ˜ δ trans i A j = - ∂ i A j =: [ P i , A j ( t, x )] | 0 1 -1 0 |
| Lorentz = Boost+ SO ( D - 1) | ˜ δ Lorentz 0 i A 0 = - t ∂ i A 0 =: [ B i , A 0 ( t, x )] ˜ δ Lorentz 0 i A j = - t∂ i A j - δ ij A 0 =: [ B i , A j ( t, x )] ˜ δ Lorentz ij A 0 = x [ i ∂ j ] A 0 =: [ J ij , A 0 ( t, x )] ˜ δ Lorentz ij A k = x [ i ∂ j ] A k + δ k [ i A j ] =: [ J ij , A k ( t, x )] | -1 0 0 1 |
| Dilatation | ˜ δ scale A 0 = ( t∂ t + x j ∂ j +∆) A 0 =: [ D,A 0 ( t, x )] ˜ δ scale A i = ( t∂ t + x j ∂ j +∆) A i =: [ D,A i ( t, x )] | 0 1 |
| Special Conformal | ˜ δ SCT 0 A 0 = [ t 2 ∂ t +2 t x i ∂ i +2∆ t ] A 0 =: [ K,A 0 ( t, x )] ˜ δ SCT 0 A i = [ t 2 ∂ t +2 t x j ∂ j +2∆ t ] A i +2 x i A 0 =: [ K,A i ( t, x )] ˜ δ SCT i A 0 = - t 2 ∂ i A 0 =: [ K i , A 0 ( t, x )] ˜ δ SCT i A j = - t 2 ∂ i A j - 2 δ ij tA 0 =: [ K i , A j ( t, x )] | 0 1 -1 0 |
in this particular sector of GED. (Here the subscript on a, b indicates the type of limit in GED.) We also infer that the scaling dimension of the fields { A , A 0 i } are given by
$$\Delta = \tilde { \Delta } = \frac { D - 2 } { 2 },$$
i.e. their values are unchanged from the relativistic ones. The set { ∆ , a, b } provides the necessary dynamical input required to construct our specific example of the Electric sector of GED. Apart from providing a cross-check of the general analysis we considered earlier in Sec 3.3, this contraction scheme is vitally important to generate the above mentioned dynamical information required to focus attention on the specifics of a particular example of a GCFT.
The very same analysis is repeated with the magnetic limit (4.5b) to extract similar information. The corresponding transformation rules differ from the electric case in some places. This is explicitly displayed in the Magnetic table 2 below, which summarises the results of the contraction procedure.
When compared with (3.31), we gather that a m = 0 , b m = -1 are values relevant in the magnetic case. The values of a and b are curiously reversed from the electric case. We will have more to say about this in the next section.
Table 2: Galilean transformations in the Magnetic limit
| Relativistic Transformation | Galileanized Transformation: Magnetic version | Same as Electric | p 1 |
|-------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-----------------------------------------------------|------------|
| Translation | ˜ δ trans 0 A 0 = ∂ t A 0 =: [ H,A 0 ( t, x )] ˜ δ trans 0 A i = ∂ t A i =: [ H,A i ( t, x )] ˜ δ trans i A 0 = - ∂ i A 0 =: [ P i , A 0 ( t, x )] ˜ δ trans i A j = - ∂ i A j =: [ P i , A j ( t, x )] | glyph[check] glyph[check] glyph[check] glyph[check] | -1 0 -2 -1 |
| Lorentz= Boost+ SO ( D - 1) | ˜ δ Lorentz 0 i A 0 = - t ∂ i A 0 - A i =: [ B i , A 0 ( t, x )] ˜ δ Lorentz 0 i A j = - t∂ i A j =: [ B i , A j ( t, x )] ˜ δ Lorentz ij A 0 = x [ i ∂ j ] A 0 =: [ J ij , A 0 ( t, x )] ˜ δ Lorentz ij A k = x [ i ∂ j ] A k + δ k [ i A j ] =: [ J ij , A k ( t, x )] | X X glyph[check] glyph[check] | 0 -1 1 0 |
| Dilatation | ˜ δ scale A 0 = ( t∂ t + x j ∂ j +∆) A 0 =: [ D,A 0 ( t, x )] ˜ δ scale A i = ( t∂ t + x j ∂ j +∆) A i =: [ D,A i ( t, x )] | glyph[check] glyph[check] | 1 0 |
| Special Conformal | ˜ δ SCT 0 A 0 = [ t 2 ∂ t +2 t x i ∂ i +2∆ t ] A 0 +2 x i A i =: [ K,A 0 ( t, x )] ˜ δ SCT 0 A i = [ t 2 ∂ t +2 t x j ∂ j +2∆ t ] A i =: [ K,A i ( t, x )] ˜ δ SCT i A 0 = - t 2 ∂ i A 0 - 2 tA i =: [ K i , A 0 ( t, x )] ˜ δ SCT i A j = - t 2 ∂ i A j =: [ K i , A j ( t, x )] | X X X X | 1 0 0 -1 |
## 5. Galilean Conformal Symmetries in GED: Intrinsic Analysis
We have already built up the representation of GCA in its full infinitely extended form in Sec 3, considering specifically spinless (scalar) and spin-1 (actually vector in a fundamental representation on so ( D -1)) primaries. We observed that these representations are labelled by two additional c-numbers ( a, b ), in addition to scaling dimension and spin of the primaries. The goal of the present section is to study these symmetries in the context of Galilean Electrodynamics discussed above. More specifically, we have two distinct dynamical systems here, namely the electric and the magnetic limits of Galilean Electrodynamics. Before delving into the full infinite GCA we would like to check invariance of the equations of motion describing these distinct systems under the actions of scalar ( K ) and vector parts ( K i ) of special conformal transformation. This would lead us to two different classes of information. Firstly, as expected, Galilean conformal invariance is observed for space-time dimension D = 4, as for the relativistic case. Secondly, the two constants, that label our representations also get fixed to real values consistent with those arrived at in Sec 4.2 from contraction procedure. After that, invariance under infinite GCA will be presented. Note that by invariance, we would always mean on-shell invariance. We would be dealing with Electric and the Magnetic limit separately in the following analysis for the ease of perusal.
## 5.1 Invariance under GCA: Electric limit
For checking invariance of (4.6) we would take the following strategy. This concerns checking whether the equations still continue to hold with transformed field variables Φ( t, x ). In other words, if an equation of motion has the schematic form: glyph[square] Φ( t, x ) = J then if the following also holds, we would have shown that the equation has the proposed symmetry:
$$\Box \delta _ { \vec { \mathbf x } } \Phi ( t, x ) = \Box \left [ \vec { \mathbf x }, \Phi ( t, x ) \right ] = 0.$$
glyph[star] here denotes any transformation generator that is relevant here. In what follows, we would be using (3.31) for the expression of [ glyph[star] , Φ( t, x )], keeping in mind the definition: K = L (+1) , K i = M (+1) i . We note that any source J in the right hand side should anyway be annihilated by transformation operators, since they are non-dynamical.
Checking for (4.6a): It is trivial to see that under scalar and the vector SCT respectively:
$$( \partial \cdot \partial ) \left [ K, A _ { 0 } ( t, x ) \right ] = 2 b _ { e } \left ( 2 \partial _ { i } A ^ { i } + x ^ { i } ( \partial \cdot \partial ) A _ { i } \right ) & & ( 5. 2 a )$$
$$( \partial \cdot \partial ) \left [ K ^ { i }, A _ { 0 } ( t, x ) \right ] = - 2 b _ { e } \, t ( \partial \cdot \partial ) A ^ { i }$$
Hence for this equation to be invariant under these transformation, we demand b e = 0 as the only possible solution.
Checking for (4.6b): Checking invariance of this equation under scalar SCT is a somewhat long, but straight-forward exercise. However the vector one is shorter and we display it here:
$$( \partial \cdot \partial ) \left [ K, A _ { i } \right ] - \partial _ { i } \partial _ { j } \left [ K, A ^ { j } \right ] - \partial _ { t } \partial _ { i } \left [ K, A _ { 0 } \right ] & = - 2 ( a _ { e } \, D + \Delta - 2 a _ { e } + 1 ) \partial _ { i } A _ { 0 } \\ & \quad - 2 ( a _ { e } + 1 ) x ^ { k } \partial _ { k } A _ { i }$$
$$( \partial \cdot \partial ) \left [ K ^ { i }, A _ { j } \right ] - \partial _ { j } \partial _ { k } \left [ K ^ { i }, A ^ { k } \right ] - \partial _ { t } \partial _ { j } \left [ K ^ { i }, A _ { 0 } \right ] = 2 ( a _ { e } + 1 ) \partial ^ { i } \partial _ { j } A _ { 0 } \\ - 2 b _ { e } \partial ^ { i } \partial _ { t } ( A _ { j } )$$
$$( 5. 3 b )$$
Using the solution b e = 0 from the invariance condition (5.2) of (4.6a) we infer that a e = -1 should hold for (5.3b) to satisfy invariance. Note that these are the exact values of a e and b e that we inferred from the results presented in table 1. Substituting this in (5.3a) and using (4.9), we get:
$$\text{Right Hand Side} = ( D - 4 ) \, \partial _ { i } A _ { 0 }.$$
Therefore the equations (4.6) are Galilean Conformal invariant only at space-time dimension 4.
Having proved Galilean conformal invariance for the finite transformations, it is now straightforward to extend the same analysis for the infinite number of other modes of the GCA.
Checking for (4.6a): It is trivial to see that under L ( n ) and M ( n ) i respectively:
$$( \partial \cdot \partial ) \left [ L ^ { ( n ) }, A _ { 0 } ( t, x ) \right ] = 0, \quad ( \partial \cdot \partial ) \left [ M _ { i } ^ { ( n ) }, A _ { 0 } ( t, x ) \right ] = 0$$
Checking for (4.6b): A very similar analysis as the previous case now gives
$$( \partial \cdot \partial ) \left [ L ^ { ( n ) }, A _ { i } \right ] - \partial _ { i } \partial _ { j } \left [ L ^ { ( n ) }, A ^ { j } \right ] - \partial _ { i } \partial _ { i } \left [ L ^ { ( n ) }, A _ { 0 } \right ] = - \frac { 1 } { 2 } \, n ( n + 1 ) \, t ^ { n - 1 } ( D - 4 ) \partial _ { i } A _ { 0 } \quad ( 5. 6 a ) \\ ( \partial \cdot \partial ) \left [ M _ { i } ^ { ( n ) }, A _ { j } \right ] - \partial _ { j } \partial _ { k } \left [ M _ { i } ^ { ( n ) }, A ^ { k } \right ] - \partial _ { i } \partial _ { j } \left [ M _ { i } ^ { ( n ) }, A _ { 0 } \right ] = 0$$
Again we conclude that these equations (4.6) are invariant only at space-time dimension 4 under the modes of the infinitely extended GCA.
## 5.2 Invariance under GCA: Magnetic limit
For GED in magnetic limit, we would proceed in the very same manner as done in the last subsection. It's now obvious that we would be checking how these equations of motion (4.7) behave under the scalar and vector special conformal transformation. The results are as follows:
Checking for (4.7a): Under scalar and the vector SCT, we get, respectively:
$$( \partial \cdot \partial ) \left [ K, A _ { i } ( t, x ) \right ] - \left ( \partial _ { j } \partial _ { i } \right ) \left [ K, A ^ { j } \right ] = - 2 a _ { m } \left [ ( D - 2 ) \partial _ { i } + ( x \cdot \partial ) \partial _ { i } - x _ { i } ( \partial \cdot \partial ) \right ] A _ { 0 } \quad ( 5. 7 a )$$
$$( \partial. \partial ) [ K ^ { i }, A _ { j } ] - ( \partial _ { k } \partial _ { j } ) [ K ^ { i }, A ^ { k } ] = 2 a _ { m } t \left [ \delta ^ { i } _ { j } ( \partial \cdot \partial ) - ( \partial ^ { i } \partial _ { j } ) \right ] A _ { 0 }$$
Therefore in order to have keep equation invariant under these transformation, we demand a m = 0 as the only possible solution.
Checking for (4.7b): Under scalar and vector SCT, the second (4.7b) magnetic equation yields:
$$\partial _ { t } \partial _ { j } [ K ^ { i }, A ^ { j } ] + ( \partial. \partial ) [ K ^ { i }, A _ { 0 } ] = 2 a _ { m } ( \partial ^ { i } - t \partial _ { t } \partial ^ { i } ) A _ { 0 } - 2 ( b _ { m } + 1 ) t ( \partial \cdot \partial ) A ^ { i } \, \quad \quad ( 5. 8 a )$$
$$\partial _ { t } \partial _ { i } [ K, A ^ { i } ] + ( \partial \cdot \partial ) [ K, A _ { 0 } ] = ( D + 4 b _ { m } ) ( \partial \cdot A ) + 2 a _ { m } \left [ ( D - 1 ) + ( x \cdot \partial ) \right ] \partial _ { t } A _ { 0 }$$
$$+ 2 ( b _ { m } + 1 ) x ^ { j } ( \partial \cdot \partial ) A _ { j }$$
Using the solution a m = 0 from the invariance condition (5.7) of (4.7a) we infer that b m = -1 should hold for (5.8a) to satisfy invariance. These values are in conformity with those concluded from the analysis of table 2. Substituting this in (5.8b), we get
$$\text{Right Hand Side} = ( D - 4 ) \, \partial \cdot A.$$
Therefore the magnetic limit equations (4.7) too are GC invariant only at space-time dimension D = 4.
We see an interesting feature emerging in the above. This is the duality of the parameters labelling GCA representations. The coefficients a and b interchange their values in the electric and magnetic limits. We will have more to say about this shortly. It is now reasonably straightforward to see that the magnetic limit of GED is also invariant under the infinite non-negative modes of GCA:
Checking for (4.7a): The scalar and vector modes of GCA keep this equation invariant only in space-time dimension 4:
$$( \partial _ { j } \partial _ { t } [ M _ { i } ^ { ( n ) }, A _ { j } ] + ( \partial. \partial ) [ M _ { i } ^ { ( n ) }, A _ { 0 } ] ) = 0$$
$$( \partial _ { i } \partial _ { i } [ L ^ { ( n ) }, A _ { i } ] + ( \partial. \partial ) [ L ^ { ( n ) }, A _ { 0 } ] ) = - \frac { 1 } { 2 } n ( n + 1 ) ( D - 4 ) t ^ { n - 1 } ( \partial. A ) \quad \ \ ( 5. 1 0 b )$$
Checking for (4.7b): This equation is however invariant under GCA in all dimensions
$$( \partial. \partial ) [ L ^ { ( n ) }, A _ { i } ] - \partial _ { i } \partial _ { j } [ L ^ { ( n ) }, A _ { j } ] = 0 & & ( 5. 1 1 \mathbf a )$$
$$( \partial. \partial ) [ M _ { i } ^ { ( n ) }, A _ { k } ] - \partial _ { k } \partial _ { j } [ M _ { i } ^ { ( n ) }, A _ { j } ] = 0$$
The observation we make here is that the magnetic limit, like the electric one is invariant under GCA in 3+1 dimensions only.
The above analysis constitutes the main result of the paper: in both the electric and magnetic limits, the equations of motion of Galilean Electrodynamics have the underlying infinite dimensional symmetry of the Galilean Conformal Algebra (3.8) in D = 4.
Before moving on to an example of the use of this symmetry, a couple of comments are in order. First, let us stress again that this is the first example of a Galilean Conformal Field Theory in D > 2.
We have also observed an interesting 'duality' between the electric and magnetic limits. This is the remnant of the parent electro-magnetic duality in the source-free Maxwell's equations. It has been pointed out that Poincare symmetry is a direct consequence of an electro-magnetic duality [22], and hence our non-relativistic limits individually cannot hope to exhibit this. The exchange of E e i ↔ B m i and B e i ↔ E m i in the non-relativistic equations reflects the re-emergence of this duality, albeit in a different avatar. This has been reported earlier in [23]. In our analysis, we find further evidence of this duality in the exchange of a e ↔ b m , b e ↔ a m . We shall look at correlation functions below and there as well one would see the exchange of roles of A 0 ↔ A i in the two limits indicating the emergence of this duality.
## 5.3 Correlation functions in GED
Given that the two limits of Galilean Electrodynamics are invariant under the infinite dimensional GCA, we can now call upon the strength of the infinite symmetry to learn more about these systems. Here to give a flavour of what can be calculated, we present the calculation of correlation functions of the field variables { A , A 0 i } . We shall restrict ourselves to the finite part of the GCA and look at the two-point functions. It is of interest to look at the full power of the infinite symmetries to write down Ward identities following the case of 2d GCFTs [14]. We defer this to future work. In this section we wish to present the correlation functions of our dynamical variables, namely A , A 0 i . We start with the assumption that our vacuum state is invariant under the 'global' part of GCA, i.e.: { H,D,P ,B ,K,K ,J i i i ij } . Moreover, the assumptions that A 0 and A i are scale ∆ objects continue here also. We would be presenting the analysis in Electric and Magnetic limit respectively in what follows.
As has been done repeatedly earlier in this article we again take this opportunity to summarize our convention and strategy for calculating the correlators.
- 1 Let Φ( t, x ) and Φ( ˜ t, x ) be two definite spin primaries (in our case spin-0 and 1 objects suffice) with definite scaling dimensions. We would always assume Galilean time-ordering prescription of keeping earlier time operators at the left of any product of operators.
- 2 Any global generator (assumed to be self-adjoint) glyph[star] being a symmetry of the vacuum, would mean glyph[star] | 0 〉 = 0 and 〈 0 | glyph[star] = 0.
- 3 When used in the context of correlators, this gives:
$$\langle 0 | \left [ \tilde { \mathbf k }, \Phi ( t _ { 1 }, x _ { 1 } ) \right ] \tilde { \Phi } ( t _ { 2 }, x _ { 2 } ) | 0 \rangle + \langle 0 | \Phi ( t _ { 1 }, x _ { 1 } ) \, \left [ \tilde { \mathbf k }, \tilde { \Phi } ( t _ { 2 }, x _ { 2 } ) \right ] | 0 \rangle = 0$$
When the transformation rules relevant to our case, as exhibited in Tables 1 and 2, these boil down to differential equations of the correlators. Note that, our transformation rules can mix primary operators in addition to differential action 3 . This results in a set of simultaneous differential equations of correlators. Solutions to these would give us the required correlators.
3 For example, in (3.31a) φ i appears in the transformation of φ apart from derivatives of φ
## Correlations in the Electric Limit
## · 〈 A 0 ( t 1 , x 1 ) A 0 ( t 2 , x 2 ) 〉 :
We want to impose first the invariance under temporal ( H ), spatial ( P i ) and rotational ( J ij ) transformation of the GCFT vacuum. As expected, it follows from the differential equations thus generated, that:
$$G _ { 0 0 } ( x, t ) = \langle A _ { 0 } ( t _ { 1 }, x _ { 1 } ) A _ { 0 } ( t _ { 2 }, x _ { 2 } ) \rangle = \sum _ { m, n \in \mathbb { Z } } c _ { m, n } t ^ { m } r ^ { n }$$
where x i = x i 1 -x , t i 2 = t 1 -t 2 and r = x x i i . Now imposing scale invariance, we can restrict it further to (keeping in mind that A 0 are scale ∆ operators):
$$\langle A _ { 0 } ( t _ { 1 }, x _ { 1 } ) A _ { 0 } ( t _ { 2 }, x _ { 2 } ) \rangle = \sum _ { k \in \mathbb { Z } } c _ { k } t ^ { - 2 \Delta + k } r ^ { - k }.$$
The next step is to impose the invariance under either of B , K, K i i . For example, the invariance under K i results in:
$$( t _ { 1 } ^ { 2 } - t _ { 2 } ^ { 2 } ) \sum _ { k } k \, c _ { k } \, t ^ { - 2 \Delta + k } r ^ { - k } = 0,$$
glyph[negationslash]
which in turn imply that c k = 0 , ∀ k = 0. Note that we have used the concerned transformation rules as presented in Table 1, for action of the generators on the dynamical operators. Hence:
$$\langle A _ { 0 } ( t _ { 1 }, x _ { 1 } ) A _ { 0 } ( t _ { 2 }, x _ { 2 } ) \rangle = c \, t ^ { - 2 \Delta }.$$
Invariance under rest of the generators ( B , K i ) give nothing new and simply respect this form of the correlator, as expected.
## · 〈 A 0 ( t 1 , x 1 ) A t ,x i ( 2 2 ) 〉 :
While using invariance of the vacuum under H,P i and J ij , we keep note of the fact that this correlator must transform as a vector under the action of J ij . Invariance under H,P ,D i and J ij of the vacuum now dictates, as above
$$G _ { 0 i } ( x, t ) = \langle A _ { 0 } ( t _ { 1 }, x _ { 1 } ) A _ { i } ( t _ { 2 }, x _ { 2 } ) \rangle = \sum _ { k } b _ { k } \, x ^ { i } \, t ^ { - 2 \Delta + k - 1 } r ^ { - k }$$
Let's now implement invariance under K (we could have done with B i or K i at this step with same result). We derive:
$$& \langle 0 | \left [ K, A _ { 0 } ( t _ { 1 }, x _ { 1 } ) A _ { i } ( t _ { 2 }, x _ { 2 } ) \right ] | 0 \rangle = \sum _ { k } b _ { k } t ^ { - 2 \Delta - k - 1 } [ - t ( x _ { 1 } ^ { i } + x _ { 2 } ^ { i } ) r ^ { k } \\ + & k r ^ { k - 2 } x ^ { i } \left ( ( t _ { 1 } + t _ { 2 } ) \, r ^ { 2 } \right ) - 2 x _ { j } ( t _ { 1 } x _ { 1 } ^ { j } - t _ { 2 } x _ { 2 } ^ { j } ) ] + 2 c \, x _ { 2 } ^ { i } \, t ^ { - 2 \Delta }$$
glyph[negationslash]
Here, the coefficient c is same as the one appearing in (5.15). For the vanishing of the right hand side of (5.17) one may firstly impose b k = 0 , ∀ k = 0 for simplification. Then the expression becomes:
$$\left [ 2 c \, x _ { 2 } ^ { i } - b _ { 0 } ( x _ { 1 } ^ { i } + x _ { 2 } ^ { i } ) \right ] t ^ { - 2 \Delta }.$$
As the next step one should demand c = b 0 = 0 as the only solution for this to be zero 4 . This forces vanishing of both G 00 and G 0 i . One could have taken any other route, ie, first imposing invariance under B i or K i . But all of these give the same result.
## · 〈 A t ,x i ( 1 1 ) A j ( t 2 , x 2 ) 〉 :
We would see that this is the sole non-zero correlator among the three we are discussing. The most general expression for this (after implementation of H,P ,D i and J ij as symmetry):
$$\langle A _ { i } ( t _ { 1 }, x _ { 1 } ) A _ { j } ( t _ { 2 }, x _ { 2 } ) \rangle = \sum _ { n } t ^ { - 2 \Delta - n } r ^ { n } \left [ a _ { n } \delta _ { i j } + b _ { n } \epsilon _ { i j k } \frac { x ^ { k } } { r } + c _ { n } \frac { x ^ { i } x ^ { j } } { r ^ { 2 } } \right ] \quad \ \ ( 5. 1 9 )$$
Note that the term glyph[epsilon1] ijk x k is special, occurring only in 3 dimensional case (here our spatial dimension). Had we been dealing with a 3 dimensional relativistic space-time, this term would have to be dropped in favour of principal of micro-causality of Bosonic operators A i .
glyph[negationslash]
Now acting this with K k and using G 0 i = 0, we derive that b n = 0 = c n , ∀ n and a n = 0 , ∀ n = 0. This finally yields:
$$\langle A _ { i } ( t _ { 1 }, x _ { 1 } ) A _ { j } ( t _ { 2 }, x _ { 2 } ) \rangle = a \, \delta _ { i j } \, t ^ { - 2 \Delta }.$$
## Correlations in the Magnetic Limit
As expected correlation functions in the Magnetic limit play a role dual to their Electric counterparts. This would be clear as we go along calculating them. As is apparent from transformation properties (Table 2), the vector (spin-1) potential does not involve the scalar one under boost or the scalar and vector conformal transformation. It would therefore be wise to consider the spin-1 spin-1 correlator first.
## · 〈 A t ,x i ( 1 1 ) A j ( t 2 , x 2 ) 〉 :
The general form of this correlator, after imposing invariance under H,P ,J i ij and D is same as (5.19). It gets more restricted to the form:
$$\langle A _ { i } ( t _ { 1 }, x _ { 1 } ) A _ { j } ( t _ { 2 }, x _ { 2 } ) \rangle = a \, t ^ { - 2 \Delta } \delta _ { i j }$$
when one imposes invariance under boost. No more information can be extracted even after checking with K and K i .
## · 〈 A 0 ( t 1 , x 1 ) A t ,x i ( 2 2 ) 〉 :
Arguments very similar to those made while deriving (5.16) can be made here to arrive at (after imposing invariance of H,P ,J i ij , D ):
$$\langle A _ { 0 } ( t _ { 1 }, x _ { 1 } ) A _ { i } ( t _ { 2 }, x _ { 2 } ) \rangle = \sum _ { m \in \mathbb { Z } } a _ { m } t ^ { - 2 \Delta - m } \, r ^ { m - 1 } \, x _ { i }.$$
glyph[negationslash]
4 Note that, for c = b 0 = 0, we have the expression only functions of t and x i :
$$\langle [ K, A _ { 0 } ( t _ { 1 }, x _ { 1 } ) A _ { i } ( t _ { 2 }, x _ { 2 } ) ] \rangle = c x ^ { i } t ^ { - 2 \Delta } = x ^ { i } \, G _ { 0 0 } ( x, t ).$$
Invariance under B j would now mean:
$$\sum _ { m \in \mathbb { Z } } a _ { m } \left [ ( m - 1 ) ( t _ { 1 } x _ { 1 } ^ { j } - t _ { 2 } x _ { 2 } ^ { j } ) r ^ { m - 3 } x _ { i } + t r ^ { m - 1 } \delta _ { i j } \right ] - a t ^ { - 2 \Delta } \delta _ { i j } = 0. \quad \ ( 5. 2 3 )$$
a appearing above is the same one of the right hand side of (5.21). This is because B i , in the magnetic limit while acting on A 0 , brings along A i . This is why the spin-1 - spin-1 correlator becomes relevant in the present analysis.
glyph[negationslash]
Now the above equation is satisfied if a m = 0 , ∀ m = 1 and a 1 = a . However, as one wishes to check whether the present form still continues to hold under the invariance condition given by K i , one finds:
$$a \, t ^ { - 2 \Delta + 1 } \delta _ { i j } = 0.$$
Hence we find a = 0, forcing both the above spin-1 - spin-1 and scalar - spin-1 two point functions to vanish.
## · 〈 A 0 ( t 1 , x 1 ) A 0 ( t 2 , x 2 ) 〉 :
Formally, the scalar scalar two point function would involve the above ones in magnetic limit. But as we have just observed, their vanishing nullifies this possibility. Deriving the present correlator would therefore be a repetitive task, already performed for the Electric limit. It is of the form:
$$\langle A _ { 0 } ( t _ { 1 }, x _ { 1 } ) A _ { 0 } ( t _ { 2 }, x _ { 2 } ) \rangle = c \, t ^ { - 2 \Delta }.$$
This time, there are no restrictions c , though.
The above correlators, both in the electric and magnetic case, can be obtained as a limit of the relativistic correlation function (2.15). In the electric case, for example, it turns out that some of the correlation functions can have interesting space dependent structures as well. But when we look at the scaling of the gauge field as in (4.5a), all the correlation functions except 〈 A A i j 〉 become zero. There is a similar mechanism which matches the correlation functions in the magnetic limit. While we admit that ultra-local correlation functions dependent only on the temporal coordinate is somewhat disappointing, we point out that these are answers dependent on the specific choice of gauge. Gauge transformations remain an intriguing question in the context of non-relativistic electromagnetism. One choice of gauge in the electric limit may be entirely incompatible with the magnetic limit. For example, the Lorentz gauge
$$\partial _ { \mu } A ^ { \mu } = 0$$
remains a good choice in the magnetic limit as x i → glyph[epsilon1]x i , t → t and A 0 → glyph[epsilon1]A 0 , A i → A i keeps this unchanged while this becomes the Coloumb gauge ∂ A i i = 0 in the electric limit. We need to have a better understanding gauge invariance in the context of non-relativistic electrodynamics in order to grasp physics behind the form of the correlation functions displayed above. It is possible that we may have imposed stricter conditions than required and missed spatial dependent pieces from our present correlators. We hope to return to the details of this in later work.
## 6. GED in D = 3
## 6.1 Dual scalar formulation
It is a curious, but less studied fact that a particular description of Electrodynamics in 3 space-time (Minkowski) dimension also exhibits conformal invariance [16]. This is done by taking the hodge dual of the 3 dimensional Electro-magnetic field strength:
$$\partial _ { \mu } \Psi = \frac { 1 } { 2 } \varepsilon _ { \mu \nu \rho } F ^ { \nu \rho }.$$
The scalar field Ψ defined this way, captures the dynamics of the system. The duality that is inherent to this approach manifests through the reversal of roles of the Bianchi identity and the equation of motion:
| Equation of motion | Identity |
|---------------------------------|---------------------|
| ∂ µ F µν = 0 | ε µνρ ∂ µ ∂ ν Ψ = 0 |
| glyph[square] Ψ ≡ ∂ µ ∂ µ Ψ = 0 | ε µνρ ∂ µ F νρ = 0 |
The Lagrangian density in terms of the scalar reads:
$$\mathcal { L } = - \frac { 1 } { 4 } F ^ { \mu \nu } F _ { \mu \nu } = - \frac { 1 } { 2 } \partial _ { \mu } \Psi \partial ^ { \mu } \Psi.$$
This is consistent with the equation of motion displayed in the above table. As it represents a theory of a massless scalar, the theory is manifestly scale invariant. Moreover, this fixes the scaling dimension of Ψ to 1 / 2 (by demanding scale invariance of the Lagrangian). It is also easy to see that the theory is conformally invariant under the special conformal transformation considering Ψ as a dimension 1 / 2 primary scalar:
$$\tilde { \delta } _ { \mu } ^ { \text{SCT} } \Psi = - x ^ { 2 } \partial _ { \mu } + 2 x _ { \mu } x ^ { \nu } \partial _ { \nu } + 2 \Delta x _ { \mu } \Psi$$
where ∆ = 1 2 is the scaling dimension.
The goal of this section is to see whether this invariance continues to hold for Galilean conformal transformations. We would see that the answer is affirmative and it is also invariant under its infinite extension.
In this direction, let us first see the non-relativistic or Galilean version of the equation of motion ( glyph[square] Ψ = 0 ). We would be using the space-time contraction (3.1) and will not scale the field Ψ. It is easy to see, that any scaling on Ψ would be redundant as far as the equation of motion is concerned. The equation of motion in Galilean setting reads:
$$\partial _ { i } \partial ^ { i } \Psi \equiv \nabla ^ { 2 } \Psi = 0.$$
## 6.2 Invariance under GCA
Time and space translation generators act on Ψ in usual manner (3.21a), (3.21b). There is a single component of spatial rotation in this case: J := 1 2 ε ij J ij . The action of this and the dilatation operator is also easily realized:
$$[ J, \Psi ( t, x ) ] = \varepsilon ^ { i j } x _ { i } \partial _ { j } \Psi ( t, x ) & & ( 6. 5 a )$$
$$\left [ L ^ { ( 0 ) }, \Psi ( t, x ) \right ] = [ D, \Psi ( t, x ) ] \, = \, ( t \partial _ { t } + x ^ { i } \partial _ { i } + \frac { 1 } { 2 } ) \Psi ( t, x ) \, \quad \quad \quad ( 6. 5 b )$$
For the action of the boost operator, we take note of the fact that unlike in the D = 4 case, we do not have any analogue of vector potential in our set of primary fields. More precisely, we can construct our representation of GCA for this particular system, with the single primary field Ψ. This prevents transforming the scalar into a vector, as in (3.27). Moreover, invariance under J also prohibits possibility of rapidity labelling the representation. This results in [ M (0) i , Ψ(0) ] = 0.
Apart from this, [ L (+1) , Ψ(0) ] = 0 = [ M (+1) i , Ψ(0) ] are also assumed. Introducing space-time dependence is also fairly straightforward, given that we have already performed the same exercise in Sec 3.3. The results are:
$$\left [ M _ { i } ^ { ( 0 ) }, \Psi ( t, x ) \right ] = - t \, \partial _ { i } \Psi ( t, x )$$
$$\left [ L ^ { ( + 1 ) }, \Psi ( t, x ) \right ] & = ( t ^ { 2 } \partial _ { t } + 2 t x ^ { i } \partial _ { i } + 2 t \Delta ) \Psi ( t, x ) & ( 6. 6 b )$$
$$\left [ M _ { i } ^ { ( + 1 ) }, \Psi ( t, x ) \right ] \, = \, - t ^ { 2 } \, \partial _ { i } \, \Psi ( t, x )$$
From this, we infer that (6.6) is the special case of (3.31) now only with a scalar primary and with a = 0 = b . We can consider even the infinite extension of these modes, as done earlier in (3.33) for arbitrary spatial dimension. We should keep in mind, while using (3.33), that the constants a, b appearing there are both zero in the present context. For checking the invariance of the equation of motion (6.4), we employ the same tactics as in the Sec 5.1 and 5.2. A number steps involving pure algebraic manipulation with partial derivatives now gives us that:
$$\nabla ^ { 2 } [ L ^ { ( n ) }, \Psi ] & = 0$$
$$\nabla ^ { 2 } [ M _ { i } ^ { ( n ) }, \Psi ] = 0$$
on shell, as before. We therefore conclude that the 'hidden' conformal symmetry of 3D Electrodynamics not only holds in its Galilean version, but also it enjoys an infinite dimensional symmetry similar to its counterpart in D = 4.
## 7. Conclusions and Future Directions
## 7.1 Summary of results
In this paper, we have initiated a study of symmetries in the non-relativistic limits of Maxwellian electrodynamics and found that infinite dimensional Galilean conformal symmetries emerge in both the Electric and Magnetic limits of Galilean Electrodynamics.
A central point of our analysis was the construction of a new type of representation theory of the GCA where we considered scale-spin primaries instead of scale-boost primaries considered earlier. We saw how these representations were determined by not only the scaling dimension and the spin of the primaries but also a couple of constants which were fixed by the dynamical information of the particular system under consideration.
For the specific example of Galilean Electromagnetism, we fixed these constants by implementing a procedure of contraction from the parent relativistic system. With this information, we could examine the symmetries of the equations of motion and found that the Electric and Magnetic limits both exhibited invariance under the infinite dimensional GCA in dimensions D = 4. Curiously, the values of these constants are the only ones which actually allow the extended symmetries.
While the invariance under the finite GCA could have been expected from the relativistic Maxwell equations in D = 4, the invariance under the extended infinite dimensional symmetries is indeed remarkable and we thus found the first example of a (infinitely extended) Galilean Conformal Field Theory in D > 2.
We went on to look at correlation functions of the gauge fields to demonstrate in a particular example of how one can exploit the symmetries of Galilean Electrodynamics. Although here we did not use the power of infinite symmetries and restricted our attention to just the finite GCA, this is a problem we wish to return to in the near future.
We then looked at GED in D = 3 and its formulation in terms of a scalar field and found that even here Galilean conformal structures emerge as symmetries of the equations of motion. This constitutes another example of a GCFT, now in D = 3.
Before we move on, we would like to address a couple of concerns that the reader may have . 5 A potential question could be whether the GCA is the remnant of some infinite dimensional higher spin algebra some free relativistic theories enjoy [17]. Let us argue why this is not the case. In two dimensions, as explained in detail in Sec 3.2, the infinite GCA emerges as a contraction of two copies of the Virasoro algebra. The relativistic higher spin algebras in 2d conformal field theories are given by W -algebras. The non-relativistic versions of these algebras are given by In¨n¨-Wigner o u contractions of these W -algebras. The structures of these non-relativistic W -algebras can e.g. be found in [35]. As is evident from this, the GCA forms only the spin-2 version of this algebra and there are infinite families of other generators corresponding to the higher spins. The structure of the contracted higher spin algebras are indeed more rich than just the GCA. In higher dimensions, we expect this feature to hold, i.e. if we were to perform a contraction on a higher spin algebra, the GCA can at most emerge as a sub-algebra of the entire higher spin algebra and there would be additional (infinite number of) generators corresponding to the higher spins.
This can also be argued from the point of view of non-relativistic holography. The GCA was first discussed in the holographic context in [13], where the dual AdS d +2 gravitational theory was shown to be a Newton-Cartan like AdS 2 × R d in the non-relativistic limit. The Virasoro sub-algebra of the GCA emerged as the asymptotic symmetries of AdS . 2 If we suppose that a higher-spin field theory would have a bulk description in terms of a higher spin theory of gravity on AdS in one
5 We thank the referee for raising these important questions.
higher dimension, and we were to look at the non-relativistic limit of this higher spin version of the AdS/CFT correspondence, we would again have a similar AdS 2 × R d split of the underlying spacetime. The AdS 2 part would now give rise to a part in the higher spin version of the field theory which would now contain a chiral W -algebra [47]. The symmetries of the higher spin nonrelativistic field theory would contain the GCA, but along with it an infinite number of other generators corresponding to higher spins.
We have also exclusively looked at the free Galilean Electromagnetic theory in this paper. The reader may worry whether these extended infinite symmetries are only a feature of the free theory. We are currently studying some interacting theories where GED is coupled with massless matter fields and hope to report on these issues elsewhere. But even in the case where these symmetries belong to just the free theory, we remind the reader of the primary goal of our programme. We wish to uncover a new solvable sector in the AdS 5 /CFT 4 correspondence which would be different from the planar limit of N = 4 Super Yang-Mills. If the GCA exists in our generalisations to Yang-Mills and then Super Yang-Mills and is responsible for the integrability of this sector, even if this is restricted to the free theory, this would be a potentially very important result.
## 7.2 Connections with Flatspace Holography
In Sec 3.2, we had commented on the connection of the GCA to the asymptotic symmetries of flat space. Let us now elaborate on this connection and connected it to our discussions of Galilean Electrodynamics presented in this paper. The asymptotic symmetries of flat space at null infinity in three and four dimensions enlarges from the Poincare algebra to what is known as the BondiMetzner-Sachs (BMS) algebra. The bms 3 and bms 4 algebras are infinite dimensional algebras and the existence of such infinite symmetry structures raises tantalising prospects of applications to holography, which had not been utilised until recently. The bms 3 is actually isomorphic to the 2d GCA (3.12) with c L = 0 , c M = 3 /G [27] and as mentioned in Sec 3.2, the mapping between AdS and flatspace is achieved by looking at the Ultra-relativistic contraction (3.17). In two field theory dimensions, the algebra turns out to be isomorphic to the one obtained by non-relativistic contraction (3.14) as one can exchange space and time directions without affecting the symmetry structure [28]. This is similar to the isomorphism of the asymptotic symmetry algebras of dS and AdS (of the same dimensions). In bulk dimensions higher than three, this isomorphism does holds in a different sense 6 and one needs to look at Carrolian Conformal Algebras (CCA) as opposed to the GCA.
In order to link our present explorations of Galilean Electrodynamics to possible dual theories of flat space, it is thus instructive to look at an ultra-relativistic limit of the Maxwellian equations as opposed to the non-relativistic one. In the four dimensional context, we expect that the Carrollian limit will exhibit finite symmetries as the CCA does not have infinite extensions in dimensions higher than three. The presence of similar Electric and Magnetic domains in this ultra-relativistic limit have already been noticed in [23]. We expect symmetry enhancements in both these limits.
6 There is an isomorphism between BMS algebras and what are called 'Semi'- Galilean Conformal Algebras (sGCA) in higher dimensions [27]. These sGCAs are related to the ultra-relativistic ones (CCAs) again by an exchange of spatial and temporal directions.
The goal of this particular project is to construct a particular example of flat holography in five bulk dimensions and thereby link it to the AdS /CFT 5 4 duality by looking at generalisations to N = 4 Super Yang Mills and to ultra-relativistic limits in that theory. It is also of interest to explore connections of the results we have established in this work to similar ones in the ultrarelativistic case and see if there exists any potential 'dual' formulation between the non-relativistic and ultra-relativistic sectors similar to the one proposed in [19] in a two-dimensional context for the tensionless limit of string theory.
The D = 3 case holds other interesting prospects. Here we can hope for infinite dimensional symmetry enhancements of the scalar formulation of Ultra-Relativistic Electrodynamics to what is the bms 4 algebra. It would be of interest to explore the extensions to Yang-Mills and Supersymmetric versions if these infinite dimensional actually do exist in the dual scalar formulation in this ultra-relativistic limit.
Some of the above is work in progress and we hope to report on this soon.
## 7.3 Other future directions
We have already outlined that the next steps in our programme are to first generalise the analysis in this paper to the Yang-Mills theories and then look at supersymmetric versions so as to understand the non-relativistic limit of Super Yang-Mills and study its potential relation to the non-relativistic limit of the AdS 5 /CFT 4 correspondence. We expect the Super-GCA [44, 45, 46] to play a vital role here. Studies of the SGCA indicate that there are similar infinite dimensional structures that arise in the supersymmetric version. We expect the Galilean versions of Super Yang Mills to exhibit invariance under these extended symmetries.
Let us now describe some of the other numerous avenues that we wish to explore in relation to this work.
We have touched upon the debate of scale invariant versus conformally invariant systems in Sec 2 in the relativistic setting and pointed to electromagnetism in D > 4 as an example of a scale invariant, non-conformal theory. The non-relativistic analogue of this question is whether Galilean invariance coupled with scale invariance in the non-relativistic theory naturally extends to Galilean Conformal invariant systems. We have seen that this does not seem to be the case and Galilean Electrodynamics for D > 4, like the relativistic case, provides a simple counter-example. It also pointed out earlier in the paper that in two spacetime dimensions, the Zamalodchikov-Polchinski theorem implies that under certain assumptions scale invariance together with Lorentz invariance actually implies conformal invariance. We would like to understand the non-relativistic analogue of this theorem. Among the assumptions of the Zamalodchikov-Polchinski theorem, unitarity plays a vital role. This is a potential stumbling block for us. In D = 2, using highest weight representations labelled by L 0 and M 0 , we are naturally lead to a non-unitary theory. But there are non-trivial unitary sectors in this theory. It would be good to understand if we are directly lead to such sub-sectors by the non-relativistic Zamalodchikov-Polchinski theorem.
The Galilean Conformal Algebra is not the only candidate for a non-relativistic conformal algebra. There has been considerable interest in understanding the Schrodinger algebra, the symmetry of
the free Schrodinger equations, as a non-relativistic algebra [48, 49, 50] and this has also been extensively used in the context of non-relativistic holography [51, 52]. It had been observed in the past that the Schrodinger algebra, like the GCA, can be extended to an infinite dimensional algebra called the Schrodinger-Virasoro algebra [53]. Both the GCA and Schrodinger algebras can be cast into the more general structure of l -conformal Galilie Algebras [54] where l = 1 2 gives / the Schrodinger algebra and l = 1 leads to the GCA. It is natural to ponder if there could be a more general formulation of Galilean Electrodynamics than what we have presented here. Let us point out here that the Schrodinger algebra cannot be obtained as a contraction of the relativistic conformal algebra and it is much less clear if there can be a formulation of electrodynamics in the non-relativistic limit that is invariant under these symmetries. One possible hurdle is the fact that the Schrodinger algebra contains a mass parameter (a central term in the [ P , B i j ] commutator) and this seems to go against the very nature of electromagnetism. But since we are talking about Galilean electromagnetism, there could be more intricate ways of coming down from the relativistic theory and we defer investigations along this line of thought to future work.
Non-relativistic physics seems to come hand in hand with a non-metric Newton-Cartan like formulation of spacetime. In the context of the non-relativistic limit of AdS/CFT, it was pointed out in [13] that these exotic structures can naturally describe both the bulk and the boundary. In the bulk, the GCA were shown to arise as asymptotic symmetries of this Newton-Cartan structure. This analysis was further elaborated and generalised to include the Schrodinger-Virasoro algebra in [43]. Of late, there has been a number of works pointing out the importance of Newton-Cartan structures to non-relativistic field theories. See e.g. [55]. There have also been efforts to understand the importance of these structures to the formulation of non-relativistic supergravity theories [56]. It would be instructive to understand the significance of these structures in our formulation of Galilean Electrodynamics and understand how the extended symmetries emerge in this context. The present lack of a Lagrangian description of the two limits could also perhaps be addressed when one looks at these exotic structures and attempts to formulate the problem in a language more intrinsically non-relativistic.
We have been interested in field theories in D = 4 in this piece of work. It would also be of interest to look at examples in lower dimensions, e.g. studies of non-relativistic conformal quantum mechanics by studying analogues of the dAFF [57] model in the Galilean context. See [58] for related work. There have been studies of the holographic duals of the dAFF models in [59] as an effort to understand the AdS /CFT 2 1 connection. It would be very interesting to understand similar holographic structures in the case of the non-relativistic theories as well. What is particularly intriguing about this is that the non-relativistic limit of any AdS d +1 /CFT d seems to pick out a particular AdS /CFT 2 1 connection. So if we were to now concentrate on the limit where d = 1, then it is far from obvious what one would end up with.
Electromagnetism and Yang-Mills theory, with massless charged particles in D = 4, has also been in focus recently due to the analysis of asymptotic symmetries [60] which turn out to be infinite dimensional at the null boundary of flat space. It is of interest to understand the question in the non-relativistic limit and see whether this has any link to the infinite structures that we have discussed in this paper. The lack of a Lagrangian description could be a possible stumbling block
in attempting the canonical analysis in this case. So the importance of Newton-Cartan structures discussed above again comes to the fore.
Lastly, we would like to point out a possible relation of our work with the tensionless limit of string theory. In usual tensile string theory, when one fixes conformal gauge in the bosonic closed string theory, there is still some residual gauge freedom which takes the form of conformal invariance on the world sheet. In the tensionless limit of closed bosonic string theory, this residual symmetry turns out to be isomorphic to the 2d GCA and hence it has been suggested that the 2d GCFT would play a central role, very similar to that played by 2d CFTs for tensile strings, in the theory of the tensionless string [19]. The quantum worldsheet theory of strings leads to the spectrum of physical particles in the spacetime. In particular, the quantising of the open string leads to gauge fields in the spacetime. It is very plausible that the spectrum of the open string in the tensionless limit may lead to the Galilean version of the photon which would be the quantum of Galilean Electromagnetism. In order to venture into this programme, it is obviously important first to construct unambiguously the theory of tensionless open strings in the light of the analysis of the closed string in [19].
## Acknowledgements
It is a pleasure to thank Daniel Grumiller and Joan Simon for discussions and comments on the manuscript. We would also like to thank Pulastya Parekh for initial collaboration. AB is partially supported by an INSPIRE award by the Department of Science and Technology, India and a Lise Meitner Award by the FWF Austria. AB would like to thank the string theory groups at University of Santiago de Compostela and ULB Brussels for warm hospitality during the completion of this work. RB thanks IISER Pune for hospitality during a number of visits which resulted in this collaboration.
## References
- [1] G. 't Hooft, 'Dimensional reduction in quantum gravity,' gr-qc/9310026.
- [2] L. Susskind, 'The World as a hologram,' J. Math. Phys. 36 , 6377 (1995) [hep-th/9409089].
- [3] J. M. Maldacena, 'The large N limit of superconformal field theories and supergravity,' Adv. Theor. Math. Phys. 2 , 231 (1998) [Int. J. Theor. Phys. 38 , 1113 (1999)] [arXiv:hep-th/9711200].
- [4] N. Beisert, C. Ahn, L. F. Alday, Z. Bajnok, J. M. Drummond, L. Freyhult, N. Gromov and R. A. Janik et al. , 'Review of AdS/CFT Integrability: An Overview,' Lett. Math. Phys. 99 , 3 (2012) [arXiv:1012.3982 [hep-th]].
- [5] A. A. Belavin, A. M. Polyakov and A. B. Zamolodchikov, 'Infinite Conformal Symmetry in Two-Dimensional Quantum Field Theory,' Nucl. Phys. B 241 , 333 (1984).
- [6] G. Mack and A. Salam, 'Finite component field representations of the conformal group,' Annals Phys. 53 , 174 (1969).
- [7] S. Ferrara, A. F. Grillo, G. Parisi and R. Gatto, 'Covariant expansion of the conformal four-point function,' Nucl. Phys. B 49 , 77 (1972).
- [8] S. Ferrara, A. F. Grillo and R. Gatto, 'Tensor representations of conformal algebra and conformally covariant operator product expansion,' Annals Phys. 76 , 161 (1973).
- [9] R. Rattazzi, V. S. Rychkov, E. Tonni and A. Vichi, 'Bounding scalar operator dimensions in 4D CFT,' JHEP 0812 , 031 (2008) [arXiv:0807.0004 [hep-th]].
- [10] R. Rattazzi, S. Rychkov and A. Vichi, 'Bounds in 4D Conformal Field Theories with Global Symmetry,' J. Phys. A 44 , 035402 (2011) [arXiv:1009.5985 [hep-th]].
- [11] D. Poland, D. Simmons-Duffin and A. Vichi, 'Carving Out the Space of 4D CFTs,' JHEP 1205 , 110 (2012) [arXiv:1109.5176 [hep-th]].
- [12] S. El-Showk, M. F. Paulos, D. Poland, S. Rychkov, D. Simmons-Duffin and A. Vichi, 'Solving the 3D Ising Model with the Conformal Bootstrap,' Phys. Rev. D 86 , 025022 (2012) [arXiv:1203.6064 [hep-th]].
- [13] A. Bagchi and R. Gopakumar, 'Galilean Conformal Algebras and AdS/CFT,' JHEP 0907 , 037 (2009) [arXiv:0902.1385 [hep-th]].
- [14] A. Bagchi, R. Gopakumar, I. Mandal and A. Miwa, 'GCA in 2d,' JHEP 1008 , 004 (2010) [arXiv:0912.1090 [hep-th]].
- [15] M. Le Bellac and J.-M. L´vy-Leblond, 'Galilean Electromagnetism,' Nuovo Cimento e 14B (1973), 217
- [16] R. Jackiw and S. -Y. Pi, 'Tutorial on Scale and Conformal Symmetries in Diverse Dimensions,' J. Phys. A 44 , 223001 (2011) [arXiv:1101.4886 [math-ph]].
- [17] S. El-Showk, Y. Nakayama and S. Rychkov, 'What Maxwell Theory in D <> 4 teaches us about scale and conformal invariance,' Nucl. Phys. B 848 , 578 (2011) [arXiv:1101.5385 [hep-th]].
- [18] A. Bagchi and I. Mandal, 'On Representations and Correlation Functions of Galilean Conformal Algebras,' Phys. Lett. B 675 , 393 (2009) [arXiv:0903.4524 [hep-th]].
- [19] A. Bagchi, 'Tensionless Strings and Galilean Conformal Algebra,' JHEP 1305 , 141 (2013) [arXiv:1303.0291 [hep-th]].
- [20] A. B. Zamolodchikov, 'Irreversibility of the Flux of the Renormalization Group in a 2D Field Theory,' JETP Lett. 43 , 730 (1986) [Pisma Zh. Eksp. Teor. Fiz. 43 , 565 (1986)].
- [21] J. Polchinski, 'Scale and Conformal Invariance in Quantum Field Theory,' Nucl. Phys. B 303 , 226 (1988).
- [22] C. Bunster and M. Henneaux, 'Duality invariance implies Poincare invariance,' Phys. Rev. Lett. 110 (2013) 011603 [arXiv:1208.6302 [hep-th]].
| [23] | C. Duval, G. W. Gibbons, P. A. Horvathy and P. M. Zhang, 'Carroll versus Newton and Galilei: two dual non-Einsteinian concepts of time,' Class. Quant. Grav. 31 , 085016 (2014) [arXiv:1402.0657 [gr-qc]]. |
|--------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [24] | H. Bondi, M. G. J. van der Burg and A. W. K. Metzner, 'Gravitational waves in general relativity. 7. Waves from axisymmetric isolated systems,' Proc. Roy. Soc. Lond. A 269 , 21 (1962). |
| [25] | R. Sachs, 'Asymptotic symmetries in gravitational theory,' Phys. Rev. 128 , 2851 (1962). |
| [26] | G. Barnich and G. Compere, 'Classical central extension for asymptotic symmetries at null infinity in three spacetime dimensions,' Class. Quant. Grav. 24 , F15 (2007) [gr-qc/0610130]. |
| [27] | A. Bagchi, 'The BMS/GCA correspondence,' Phys. Rev. Lett. 105 , 171601 (2010) [arXiv:1006.3354 [hep-th]]. |
| [28] | A. Bagchi and R. Fareghbal, 'BMS/GCA Redux: Towards Flatspace Holography from Non-Relativistic Symmetries,' JHEP 1210 , 092 (2012) [arXiv:1203.5795 [hep-th]]. |
| [29] | G. Barnich, A. Gomberoff and H. A. Gonzalez, 'The Flat limit of three dimensional asymptotically anti-de Sitter spacetimes,' Phys. Rev. D 86 , 024020 (2012) [arXiv:1204.3288 [gr-qc]]. |
| [30] | A. Bagchi, S. Detournay and D. Grumiller, 'Flat-Space Chiral Gravity,' Phys. Rev. Lett. 109 , 151301 (2012) [arXiv:1208.1658 [hep-th]]. |
| [31] | A. Bagchi, S. Detournay, R. Fareghbal and J. Simon, 'Holography of 3d Flat Cosmological Horizons,' Phys. Rev. Lett. 110 , 141302 (2013) [arXiv:1208.4372 [hep-th]]. |
| [32] | G. Barnich, 'Entropy of three-dimensional asymptotically flat cosmological solutions,' JHEP 1210 095 (2012) [arXiv:1208.4371 [hep-th]]. |
| [33] | G. Barnich, A. Gomberoff and H. A. Gonzalez, 'BMS 3 invariant two dimensional field theories as flat limit of Liouville,' arXiv:1210.0731 [hep-th]. |
| [34] | A. Bagchi, S. Detournay, D. Grumiller and J. Simon, 'Cosmic evolution from phase transition of 3-dimensional flat space,' Phys. Rev. Lett. 111 , 181301 (2013) [arXiv:1305.2919 [hep-th]]. |
| [35] | H. Afshar, A. Bagchi, R. Fareghbal, D. Grumiller and J. Rosseel, 'Higher spin theory in 3-dimensional flat space,' Phys. Rev. Lett. 111 , 121603 (2013) [arXiv:1307.4768 [hep-th]]. |
| [36] | H. A. Gonzalez, J. Matulich, M. Pino and R. Troncoso, 'Asymptotically flat spacetimes in three-dimensional higher spin gravity,' JHEP 1309 , 016 (2013) [arXiv:1307.5651 [hep-th]]. |
| [37] | A. Bagchi and R. Basu, '3D Flat Holography: Entropy and Logarithmic Corrections,' JHEP 1403 020 (2014) [arXiv:1312.5748 [hep-th]]. |
| [38] | R. Fareghbal and A. Naseh, 'Flat-Space Energy-Momentum Tensor from BMS/GCA Correspondence,' JHEP 1403 , 005 (2014) [arXiv:1312.2109 [hep-th]]. |
| [39] | C. Krishnan, A. Raju and S. Roy, 'A Grassmann path from AdS 3 to flat space,' JHEP 1403 , 036 (2014) [arXiv:1312.2941 [hep-th]]. |
| [40] | S. Detournay, D. Grumiller, F. Scholler and J. Simon, 'Variational principle and 1-point functions in 3-dimensional flat space Einstein gravity,' Phys. Rev. D 89 , 084061 (2014) [arXiv:1402.3687 [hep-th]]. |
| [41] | C. Duval, G. W. Gibbons and P. A. Horvathy, 'Conformal Carroll groups and BMS symmetry,' Class. Quant. Grav. 31 , 092001 (2014) [arXiv:1402.5894 [gr-qc]]. |
| [42] | C. Duval, G. W. Gibbons and P. A. Horvathy, 'Conformal Carroll groups,' arXiv:1403.4213 [hep-th]. |
| [43] | C. Duval and P. A. Horvathy, 'Non-relativistic conformal symmetries and Newton-Cartan structures,' J. Phys. A 42 , 465206 (2009) [arXiv:0904.0531 [math-ph]]. |
| [44] | A. Bagchi and I. Mandal, 'Supersymmetric Extension of Galilean Conformal Algebras,' Phys. Rev. D 80 , 086011 (2009) [arXiv:0905.0580 [hep-th]]. |
| [45] | M. Sakaguchi, 'Super Galilean conformal algebra in AdS/CFT,' J. Math. Phys. 51 , 042301 (2010) [arXiv:0905.0188 [hep-th]]. |
- [46] J. A. de Azcarraga and J. Lukierski, 'Galilean Superconformal Symmetries,' Phys. Lett. B 678 , 411 (2009) [arXiv:0905.0141 [math-ph]].
- [47] D. Grumiller, M. Leston and D. Vassilevich, 'Anti-de Sitter holography for gravity and higher spin theories in two dimensions,' Phys. Rev. D 89 , no. 4, 044001 (2014) [arXiv:1311.7413 [hep-th]].
- [48] C. R. Hagen, 'Scale and conformal transformations in galilean-covariant field theory,' Phys. Rev. D 5 , 377 (1972).
- [49] U. Niederer, 'The maximal kinematical invariance group of the free Schrodinger equation.,' Helv. Phys. Acta 45 , 802 (1972).
- [50] Y. Nishida and D. T. Son, 'Nonrelativistic conformal field theories,' Phys. Rev. D 76 , 086004 (2007) [arXiv:0706.3746 [hep-th]].
- [51] D. T. Son, 'Toward an AdS/cold atoms correspondence: A Geometric realization of the Schrodinger symmetry,' Phys. Rev. D 78 , 046003 (2008) [arXiv:0804.3972 [hep-th]].
- [52] K. Balasubramanian and J. McGreevy, 'Gravity duals for non-relativistic CFTs,' Phys. Rev. Lett. 101 , 061601 (2008) [arXiv:0804.4053 [hep-th]].
- [53] M. Henkel, 'Schrodinger invariance in strongly anisotropic critical systems,' J. Statist. Phys. 75 , 1023 (1994) [hep-th/9310081].
- [54] D. Martelli and Y. Tachikawa, 'Comments on Galilean conformal field theories and their geometric realization,' JHEP 1005 , 091 (2010) [arXiv:0903.5184 [hep-th]].
- [55] D. T. Son, 'Newton-Cartan Geometry and the Quantum Hall Effect,' arXiv:1306.0638 [cond-mat.mes-hall].
- [56] R. Andringa, E. A. Bergshoeff, J. Rosseel and E. Sezgin, '3D Newton-Cartan supergravity,' Class. Quant. Grav. 30 , 205005 (2013).
- [57] V. de Alfaro, S. Fubini and G. Furlan, 'Conformal Invariance in Quantum Mechanics,' Nuovo Cim. A 34 , 569 (1976).
- [58] S. Fedoruk, E. Ivanov and J. Lukierski, 'Galilean Conformal Mechanics from Nonlinear Realizations,' Phys. Rev. D 83 , 085013 (2011) [arXiv:1101.1658 [hep-th]].
- [59] C. Chamon, R. Jackiw, S. Y. Pi and L. Santos, 'Conformal quantum mechanics as the CFT 1 dual to AdS ,' Phys. Lett. B 2 701 , 503 (2011) [arXiv:1106.0726 [hep-th]].
- [60] A. Strominger, 'Asymptotic Symmetries of Yang-Mills Theory,' arXiv:1308.0589 [hep-th].
|
10.1007/JHEP11(2014)061
|
[
"Arjun Bagchi",
"Rudranil Basu",
"Aditya Mehra"
] |
2014-08-04T20:29:58+00:00
|
2014-11-24T04:33:45+00:00
|
[
"hep-th"
] |
Galilean Conformal Electrodynamics
|
Maxwell's Electrodynamics admits two distinct Galilean limits called the
Electric and Magnetic limits. We show that the equations of motion in both
these limits are invariant under the Galilean Conformal Algebra in D=4, thereby
exhibiting non-relativistic conformal symmetries. Remarkably, the symmetries
are infinite dimensional and thus Galilean Electrodynamics give us the first
example of an infinitely extended Galilean Conformal Field Theory in D>2. We
examine details of the theory by looking at purely non-relativistic conformal
methods and also use input from the limit of the relativistic theory.
|
1408.0811v2
|
## Photonuclear production of vector mesons in ultra-peripheral Pb-Pb collisions at the LHC
Joakim Nystrand (for the ALICE Collaboration)
Department of Physics and Technology University of Bergen, Bergen, Norway [email protected]
## Abstract
Vector mesons are copiously produced in ultra-peripheral nucleus-nucleus collisions. In these collisions, the nuclei are separated by impact parameters larger than the sum of the nuclear radii, and the interaction is mediated by the electromagnetic field. The interaction effectively corresponds to a photonuclear interaction between a photon, generated from the electromagnetic field of one of the nuclei, and the target nucleus. The ALICE Collaboration has previously published results on exclusive J/ψ photoproduction at mid and forward rapidities in Pb-Pb collisions. The cross section for this process is a particularly good measure of the nuclear gluon distribution. In this talk, the latest results on exclusive production of light and heavy vector mesons from ALICE in Pb-Pb collisions will be presented.
Keywords: photoproduction, vector mesons, ultra-peripheal collisions
## 1 Introduction
The strong electromagnetic fields accompanying heavy-ions accelerated at the LHC may lead to particle production in ultra-peripheral collisions where there is no overlap between the colliding nuclei. This has been exploited by the ALICE Collaboration to constrain the nuclear and proton gluon distributions through exclusive photoproduction of J/ ψ mesons [1, 2, 3]. This presentation extends these studies to exclusive photoproduction of ρ 0 and ψ (2S). It also contains results on two-photon production of e + e -pairs in a novel invariant mass range.
During the heavy-ion runs in 2010 and 2011, ALICE used dedicated triggers for collecting vector mesons produced in ultra-peripheral collisions. The ALICE detector and the triggers used during the 2011 run are described in [1, 2]. The midrapidity triggers were based on input from the ALICE Time-of-Flight (TOF), Silicon Pixel (SPD), and VZERO detectors. At least two hits were required in TOF and SPD, while the VZERO detectors, covering 2 8 . < η < 5 2 and . -3 7 . < η < -1 7, . were used as veto detectors and were required to be empty. In 2011, a cut on azimuthal angle was applied which required the hits in TOF to be back-to-back, thereby restricting the final states to have invariant masses ≥ 2 GeV/c . The ultra-peripheral trigger used in 2010 was similar to the one in 2011 but without the angular cut. During 2 the early part of the 2010 run a trigger using input only from the TOF was also used. The integrated luminosities for the analyses presented below were about 260 mb -1 ( ρ 0 and e + e -pairs, 2010 data) and 22 µ b -1 ( ψ (2S), 2011 data).
Figure 1: Left: Invariant mass distribution of exclusive π + π -pairs with p T < 0.15 GeV/c. The curves show the fit to a Breit-Wigner distribution, as discussed in the text. Right: The measured d σ/ d y for coherent photoproduction of ρ 0 compared with model predictions.
## 2 Coherent ρ 0 production
The events used for the analysis were required to satisfy the ultra-peripheral trigger, to have exactly two reconstructed tracks identified as pions from the dE/dx in the Time Projection Chamber (TPC), and to have a reconstructed primary vertex within ± 10 cm from the center of the interaction region along the beam axis. To select coherent production the pairp T was required to be < . 0 15 GeV/c. The rapidity of the pair was required to be within | y | < . 0 5 to avoid edge effects. The data were corrected for acceptance and efficiency bin-by-bin in invariant mass using simulated events processed through the detector response simulation and reconstructed back using the same selection as for real data. The simulated events consisted of two pions of opposite charge with a flat distribution in invariant mass (2 m π ≤ M ππ ≤ 1.5 GeV/c ), rapidity 2 ( | y | < 0 5), . and transverse momentum ( p T < 0.15 GeV/c). The decay angle distribution was not measured. In the simulation, it was, as expected for the decay from a spin-1 state, assumed to be d n/ dcos( θ ) ∝ sin 2 ( θ ) in the ππ center of mass. The systematic error is estimated and includes the errors in signal extraction, track selection, luminosity, trigger efficiency, incoherent contribution, and particle identification.
The corrected invariant mass distribution is shown in Fig. 1 (left). The distribution has been fitted to a relativistic Breit-Wigner function plus a continuum term:
$$\frac { d \sigma } { d M _ { \pi \pi } } = \left | A \frac { \sqrt { M _ { \pi \pi } M _ { \rho } } \Gamma ( M _ { \pi \pi } ) } { M _ { \pi \pi } ^ { 2 } - M _ { \rho } ^ { 2 } + i M _ { \rho } \Gamma ( M _ { \pi \pi } ) } + B \right | ^ { 2 } \quad \text{with} \ \ \Gamma ( M _ { \pi \pi } ) = \Gamma _ { \rho } \frac { M _ { \rho } } { M _ { \pi \pi } } \left [ \frac { M _ { \pi \pi } ^ { 2 } - 4 m _ { \pi } ^ { 2 } } { M _ { \rho } ^ { 2 } - 4 m _ { \pi } ^ { 2 } } \right ] ^ { \frac { 3 } { 2 } } \, \\ \text{where $A$ and $B$ are the amplitudes for the resonant and continuum contributions, respectively. $M_{\rho}$ and $\Gamma_{\rho}$ are the mass}$$
∣ ∣ where A and B are the amplitudes for the resonant and continuum contributions, respectively. M ρ and Γ ρ are the mass and width of the ρ 0 meson, and m π is the mass of the pion. Although other parameterizations of the ρ 0 shape are possible, the above is chosen to be consistent with the most recent ρ 0 photoproduction measurements by ZEUS [4] and STAR [5]. The fit gives M ρ = 761 6 . ± 2 3( . stat. ) +6 1 . -3 0 . ( sys. ) MeV/c 2 and Γ ρ = 150 2 . ± 5 5( . stat. ) +12 0 . -5 6 . ( sys. ) MeV/c , in good agreement 2 with the PDG values. The ratio of the non-resonant to resonant amplitudes is | B/A | = 0 50 . ± 0 04( . stat. ) +0 10 . -0 04 . ( sys. ) (GeV/c ) 2 -1 / 2 .
The cross section for coherent ρ 0 production is obtained by integrating the resonant part of the d σ/ d M ππ (dashed curve in Fig. 1 (left)) from 2 m π to 1.5 GeV/c . The contribution from incoherent photoproduction below 2 p T < 0.15 GeV/c is estimated to be 5% and is subtracted. The result is d σ/ d y = 420 ± 10( stat. ) +39 -55 ( sys. ) mb, which is shown and compared with model predictions in Fig. 1 (right).
The measured cross section is in agreement with STARLIGHT [6] and the calculation by Goncalves and Machado (GM) [7] while the GDL (Glauber-Donnachie-Landshoff) prediction [8] is about a factor of 2 higher than data. The calculation by GM is based on the Color Dipole model, while STARLIGHT and GDL use the photon-proton cross section σ γ ( + p → ρ 0 + p ) constrained from data as input. All calculations use the Glauber model to scale the cross section from γ -nucleon to γ -nucleus. The agreement with STARLIGHT is a bit surprising since the calculation does not include the elastic part of the total cross section, which is included in the GDL model, but a similar trend was observed by STAR at RHIC [5].
Figure 2: Measured d σ/ d y for coherent ψ (2S) photoproduction compared with model calculations (STARLIGHT [6], AN [9], GDGM [10], LM [11]).
## 3 Coherent ψ (2S) production
Photoproduced ψ (2S) mesons can be studied in the dilepton decay channel using the ALICE triggers for ultra-peripheral collisions [2]. The trigger at midrapidity is, however, also sensitive to the decay channel ψ (2S) → J/ψ + π + π -, followed by J/ψ → e + e -/µ + µ -, which has a more advantageous branching ratio of 34%. These events are characterized by two hard tracks with p T > 1 GeV/c from the decay of the J/ψ and two soft tracks with p T < 0.4 GeV/c from the two pions.
The triggered events were required to have exactly 2 or 4 reconstructed tracks and a reconstructed primary vertex. The dileptons can be separated into e + e -or µ + µ -using the TPC dE/dx. To select coherently produced ψ (2S) a requirement p T < 0.15 (0.30) GeV/c was applied for the decay channels containing muons (electrons).
The yield in the direct dilepton channel ( e + e -and µ + µ -combined) was obtained by fitting the invariant mass distribution to the sum of a Crystal Ball function for the signal and an exponential for the γγ → l + l -background. The 4 track samples have a very high purity and the yield is obtained by counting the number of events within selected intervals in phase space determined from Monte Carlo simulations. The invariant mass of the dileptons are required to be within 3 0 . ≤ M inv ≤ 3 2 (2 6 . . ≤ M inv ≤ 3 2) GeV/c . 2 for µ + µ -( e + e -) to select pairs coming from a J/ψ decay. The µ + µ -π + π -( e + e -π + π -) final state is required to have an invariant mass 3 6 . ≤ M inv ≤ 3 8 (3 1 . . ≤ M inv ≤ 3 8) GeV/c . 2 to count as a ψ (2S). This gives a total of 17 (11) events for the µ + µ -π + π -( e + e -π + π -) decay channel, with a like-sign background of 1 (0) event.
The extracted yield is corrected for acceptance and efficiency using STARLIGHT events processed through the detector response simulation and reconstructed back using the same selection as for real data. The cross section is obtained from a weighted mean of the dilepton and 4-track decay channels, giving d σ/ d y = 0 83 . ± 0 19( . stat. + sys. ) mb. This is compared to model predictions [6, 9, 10, 11] in Fig. 2.
The ψ (2S) is a hard probe, the scale being set by its mass, and is therefore expected to be sensitive to the nuclear gluon distribution in the same way as the J/ψ . The experimental error is, however, larger for the ψ (2S) than for the J/ψ measurement, mainly because of the limited statistics. It is also important to note that the underlying γ + p → V + p cross section has a considerably larger uncertainty for ψ (2S) than for J/ψ . This is illustrated by the large discrepancy between the AN [9] and STARLIGHT predictions without nuclear effects in the figure. One can nevertheless conclude that the measured cross section disfavors models with no nuclear effects and models with strong gluon shadowing.
## 4 Two-photon production of e + e -pairs
Two-photon production of e + e -pairs has a topology similar to that of exclusive vector mesons production followed by decay into a pair of dileptons or pions. This process is of interest since the coupling between the photon and the emitting nucleus is enhanced by a factor Z (the nuclear charge). One can therefore expect higher order terms to be important. The results from ALICE on γγ → e + e -from the 2011 Pb-Pb run were, however, found to be in good agreement with STARLIGHT, which includes only the leading order QED terms [2]. Because of the requirement on azimuthal angle in
T
e
e
Figure 3: Left: Transverse momentum distribution of exclusive e + e -pairs with 0 6 . ≤ M ee ≤ 2 0 GeV/c . 2 compared with STARLIGHT. Right: Differential cross section, d σ/ d M ee , for Pb + Pb → Pb + Pb + e + e -compared with STARLIGHT.
the trigger, the results from 2011 were restricted to high invariant masses, M ee > 2.2 GeV/c . 2 The 2010 data allow to go down to M ee = 0.6 GeV/c . 2
The event selection was similar to that for coherent ρ 0 production, with the selection on the TPC dE/dx modified to accept electrons rather than pions. The raw data were corrected for acceptance and efficiency using events generated by STARLIGHT. Figure 3 (left) shows the p T distribution of the selected e + e -pairs. The distribution is well described by STARLIGHT, indicating that there is no background not accounted for in the sample. The measured cross section for the selection 0 6 . ≤ M ee ≤ 2 0 . GeV/c 2 and | η 1 2 , | < 0.9 ( η 1 2 , are the pseudorapidities of the two tracks) is 9 8 . ± 0 6( . stat. ) +0 9 . -1 2 . ( sys. ) mb. The STARLIGHT prediction for the same selection is σ = 9.7 mb. The differential cross section, d σ/ M d ee , is shown in Fig. 3 (right) together with the previous ALICE measurement for M ee > 2.2 GeV/c 2 and the cross section from STARLIGHT. The variation of d σ/ d M ee with M ee is well described by STARLIGHT.
## 5 Conclusions
The ALICE Collaboration has made the first measurements of coherent ρ 0 and ψ (2S) photoproduction in Pb-Pb collisions at the LHC. The measured cross section for the ψ (2S) disfavors models with no nuclear effects and models with strong gluon shadowing. This is thus consistent with the conclusions from the J/ψ measurements [1, 2] but less strong because of the larger uncertainties for ψ (2S) discussed above. The results on ρ 0 show that a straightforward scaling of the γ -p cross section using the Glauber model [8] overpredicts the cross section. This confirms what was observed in AuAu collisions at RHIC energies. The cross section for two-photon production of e + e -pairs at midrapidity in Pb-Pb collisions is in good agreement with leading order QED as implemented in STARLIGHT in the invariant mass range 0 6 . ≤ M ee ≤ 10 0 GeV/c . . 2
## References
- [1] B. Abelev et al. [ALICE Collaboration], Phys. Lett. B 718 (2013) 1273.
- [2] E. Abbas et al. [ALICE Collaboration], Eur. Phys. J. C 73 (2013) 2617.
- [3] B. Abelev et al. [ALICE Collaboration], arXiv:1406.7819 [nucl-ex].
- [4] J. Breitweg et al. [ZEUS Collaboration], Eur. Phys. J. C 2 (1998) 247.
- [5] B. I. Abelev et al. [STAR Collaboration], Phys. Rev. C 77 (2008) 034910.
- [6] S. R. Klein, J. Nystrand, Phys. Rev. C 60 (1999) 014903; A. J. Baltz, Y. Gorbunov, S. R. Klein, J. Nystrand, Phys. Rev. C 80 (2009) 044902; http://starlight.hepforge.org/.
- [7] V. P. Goncalves, M. V. T. Machado, Phys. Rev. C 84 (2011) 011902.
- [8] L. Frankfurt, M. Strikman, M. Zhalov, Phys. Lett. B 537 (2002) 51; V. Rebyakova, M. Strikman, M. Zhalov, Phys. Lett. B 710 (2012) 647.
- [9] A. Adeluyi, T. Nguyen, Phys. Rev. C 87 (2013) 027901.
- [10] M. B. G. Ducati, M. T. Griep, M. V. T. Machado, Phys. Rev. C 88 (2013) 014910.
- [11] T. Lappi, H. Mantysaari, Phys. Rev. C 87 (2013) 032201; T. Lappi, H. Mantysaari, private communication.
|
10.1016/j.nuclphysa.2014.09.018
|
[
"Joakim Nystrand"
] |
2014-08-04T20:36:30+00:00
|
2014-10-31T16:10:32+00:00
|
[
"nucl-ex",
"hep-ex"
] |
Photonuclear production of vector mesons in ultra-peripheral Pb-Pb collisions at the LHC
|
Vector mesons are copiously produced in ultra-peripheral nucleus-nucleus
collisions. In these collisions, the nuclei are separated by impact parameters
larger than the sum of the nuclear radii, and the interaction is mediated by
the electromagnetic field. The interaction effectively corresponds to a
photonuclear interaction between a photon, generated from the electromagnetic
field of one of the nuclei, and the target nucleus. The ALICE Collaboration has
previously published results on exclusive J/psi photoproduction at mid and
forward rapidities in Pb-Pb collisions. The cross section for this process is a
particularly good measure of the nuclear gluon distribution. In this talk, the
latest results on exclusive production of light and heavy vector mesons from
ALICE in Pb-Pb collisions will be presented.
|
1408.0812v1
|
## Lower Bounds for Structuring Unreliable Radio Networks ∗
Calvin Newport Georgetown University [email protected]
## Abstract
In this paper, we study lower bounds for randomized solutions to the maximal independent set (MIS) and connected dominating set (CDS) problems in the dual graph model of radio networks-a generalization of the standard graph-based model that now includes unreliable links controlled by an adversary. We begin by proving that a natural geographic constraint on the network topology is required to solve these problems efficiently (i.e., in time polylogarthmic in the network size). In more detail, we prove that in the absence of this constraint, for a network of size n : every MIS algorithm now requires Ω( n 1 -/epsilon1 ) rounds to solve the problem, for any constant /epsilon1, 0 < /epsilon1 ≤ 1, and every CDS algorithm that provides a reasonable approximation of a minimum CDS now requires Ω( √ n/ log n ) rounds. We then prove the importance of the assumption that nodes are provided advance knowledge of their reliable neighbors (i.e, neighbors connected by reliable links). In more detail, we prove that in the absence of this assumption, for any CDS algorithm that guarentees a g n ( )-approximation of a minimum CDS in f ( n ) rounds, it follows that g n ( )+ f ( n ) = Ω( n ). This holds even if we assume the geographic constraint and the weakest possible adversary controlling the unreliable links. Finally, we show that although you can efficiently build an MIS without advance neighborhood knowledge, this omission increases the problem's dependence on the geographic constraint. When both constraints are missing, every MIS algorithm now requires Ω( n ) rounds, even if we assume the weakest possible adversary. Combined, these results answer an open question by proving that the efficient MIS and CDS algorithms from [3] are optimal with respect to their dual graph model assumptions. They also provide insight into what properties of an unreliable network enable efficient local computation.
∗ This work is supported in part by NSF grant number CCF 1320279 and the Ford Motor Company University Research Program. An extended abstract also appears in the 2014 proceedings of the International Symposium on Distributed Computing (DISC).
## 1 Introduction
This paper proves four new lower bounds on the maximal independent set (MIS) and connected dominating set (CDS) problems in radio networks with unreliable links. These bounds establish the necessary model assumptions for building structures efficiently in this dynamic setting. In doing so, they also prove that the MIS and CDS algorithms of [3] are optimal with respect to their assumptions. As emphasized in previous studies (e.g., [13, 16, 3]), these two problems are important in the radio setting as they provide clusterings and routing backbones, respectively, both of which are useful to higher-level applications.
The Dual Graph Model. Theoreticians have studied distributed algorithms in radio network models since the 1980s [1]. Most of this existing work assumes static models in which receive behavior depends on a fixed set of deterministic rules. This property is true, for example, of both the popular graph-based [4, 1] and signal-strength [17] models. We argue that it is important, however, to also study radio network models that are more dynamic and less predictable. This type of model uncertainty can abstract the complex behavior observed in real wireless networks [18, 19], and therefore improve the likelihood that properties proved in the theory setting will remain satisfied in a practical deployment. Our call for dynamic radio network models, in other words, is an attempt to help close the gap between theory and practice.
In a recent series of papers motivated by this argument, we study distributed computation in a dynamic radio network environment that we call the dual graph model [12, 11, 3, 8, 9]. This model generalizes the well-studied graph-based models [4, 1] to now include two topology graphs. The first graph captures reliable links that are present in every round of the computation 1 and the second captures unreliable links that come and go as determined by a (bounded) adversary. The collision rules in each round are the same as in the standard graph-based models.
Results. In previous work [3], we studied the MIS and CDS problems in the dual graph model with an adaptive adversary and the following two strong assumptions: (1) a natural geographic constraint holds with respect to the dual graphs (see Section 2); and (2) the nodes are provided the ids of their reliable neighbors (i.e., neighbors in the reliable link graph) at the beginning of the execution. Under these assumptions, we described randomized MIS and CDS algorithms that are efficient, which we define in the following to mean time polylogarthmic in the network size. Furthermore, the CDS algorithm guarantees a structure that is a constant-approximation of a minimum CDS in the network. We note that in the standard graph-based model without unreliable links, the best known solutions to these problems are also polylogarthmic [16], indicating that the above assumptions enable algorithms to minimize the impact of unreliability.
In this paper, we explore the necessity of these two assumptions. We begin by proving that the geographic constraint is required to efficiently build an MIS or CDS in the dual graph model. In more detail, in Section 3.1 we prove that without this assumption, every randomized MIS algorithm now requires Ω( n 1 -/epsilon1 / log n ) rounds to solve the problem in a network of size n , for any constant /epsilon1, 0 < /epsilon1 ≤ 1. We then prove, in Section 3.2, that any randomized CDS algorithm that guarantees to provide at least a o ( √ n )-approximation of the minimum CDS now requires Ω( √ n/ log n ) rounds to solve the problem. In both cases, these results hold even when we weaken the adversary from the offline adaptive adversary assumed in [3] (which knows the nodes' random bits)
1 Notice, a more general approach to modeling unreliability would be to assume a single graph that changes from round to round. The dual graph model assumes the same reliable sub-graph is present in each each round because it enables more natural and simple definitions of standard problems; e.g., to define broadcast , we can simply say the message gets to all nodes connected to the source in the reliable sub-graph, and to define structuring algorithms, we can require that the structures to be correct with respect to this sub-graph-definitions that are complicated without this stability.
to the weaker online adaptive adversary (which does not know these bits). Note that these lower bounds are exponentially worse than what is possible with the geographic constraint-underscoring its importance.
To prove our MIS lower bound, we show that any algorithm that works efficiently in this setting must work in a ring with a (non-geographic) unreliable link topology that allows a clever adversary to prevent many segments of the ring from receiving any messages. The nodes in these isolated segments must then make an MIS decision based only on their id and the ids of their neighbors (which, by assumption, they are provided). By repurposing a key combinatorial result due to Linial [15], we are able to show that for a particular method of assigning ids to the ring, it is likely that some isolated segments will make mistakes. To prove the CDS result, we use simulations of the algorithm in question to carefully build a challenging (non-geographic) dual graph network and id assignment in which it is likely that either the CDS is too large (leading to a bad approximation) or is not connected (violating correctness).
We proceed by exploring the necessity of the second assumption which provides nodes advance knowledge of their reliable neighbors. We emphasize that for structuring problems, nodes need some way to distinguish reliable links from unreliable links, as the problem definitions require that the structures be correct with respect to the reliable link graph (see Section 2). They do not, however, necessarily require advance knowledge of their reliable neighbors. With this in mind, we study what happens when we replace this advance knowledge assumption with a passive alternative that simply labels messages received from a reliable neighbor as reliable-leaving it up to the algorithm to discover these nodes.
We prove in Section 4.1 that the advance knowledge assumption is necessary to efficiently solve the CDS problem. In more detail, we prove that with a geographic constraint, the weakest possible adversary (a static adversary that never changes the unreliable links it includes), but only passive neighborhood knowledge, for any randomized CDS algorithm that guarantees to construct a g n ( )-approximation of the minimum CDS in f ( n ) rounds, it follows that g n ( ) + f ( n ) = Ω( n ). We then turn our attention to the MIS problem. In Section 4.2, we first show that the MIS solution from [3] still works with passive knowledge-identifying a gap with respect to the CDS problem. We then prove, however, that the switch to passive knowledge increases the dependence of any MIS solution on the assumption of a geographic constraint. In particular, we prove that with the passive neighborhood knowledge, the static adversary, and no geographic constraint, every randomized MIS algorithm now requires Ω( n ) rounds to solve the problem.
In both bounds, we rely on a static adversary that adds unreliable links between all nodes in all rounds. In such a network, only one node can successfully send a message in a given round (if any two nodes send, there will be a collision everywhere), but any successful message will be received by all nodes in the network. The key to the arguments is the insight that a received message is only useful if it comes from a reliable neighbor, and therefore, in each round, at most a small fraction of the network receives useful information. If we run the algorithm for a sufficiently small number of rounds, a significant fraction of nodes will end up making an MIS or CDS decision without any knowledge of their reliable neighborhood (as they did not receive any useful messages and we assume no advance knowledge of reliable neighbors). Our bounds use reductions from hard guessing games to careful network constructions to prove that many nodes are subsequently likely to guess wrong.
Implications. In addition to proving the algorithms from [3] optimal, our lower bounds provide interesting general insight into what enables efficient local computation in an unreliable environment. They show us, for example, that geographic network topology constraints are crucial-without such constraints, the MIS and CDS problems cannot be solved efficiently in the dual graph model, even with strong assumptions about neighborhood knowledge. Though existing structuring results in
other radio network models all tend to use similar constraints (e.g., [13, 16]), to the best of our knowledge this is the first time they are shown to be necessary in a radio setting. Our lower bounds also identify an interesting split between the MIS and CDS problems, which are typically understood to be similar (building an MIS is often a key subroutine in CDS algorithms). In particular, the MIS problem can still be solved efficiently with passive neighborhood knowledge, but the CDS problem cannot. Our intuition for this divide, as highlighted by the details of our proof argument (see Section 4.1), is that a CDS's requirement for reliable connectivity necessitates, in the absence of advance neighborhood knowledge, a sometimes laborious search through a thicket of unreliable links to find the small number of reliable connections needed for correctness.
Related Work. The dual graph model was introduced independently by Clementi et al. [5] and Kuhn et al. [12], and has since been well-studied [11, 3, 8, 9]. In [3], we presented an MIS and CDS algorithm that both require O (log 3 n ) rounds, for a network of size n and the strong assumptions described above. It was also shown in [3], that an efficient CDS solution is impossible if provided imperfect advance neighborhood knowledge (i.e., a list of reliable neighbors that can contain a small number of mistakes). In the classical graph-based radio network model [4, 2], which does not include unreliable edges, the best known MIS algorithm requires O (log 2 n ) rounds [16] (which is tight [10, 7, 6]), and assumes a similar geographic constraints as in [3] (and which we prove necessary in the dual graph model in this paper). Strategies for efficiently building a CDS once you have an MIS in the classical radio network model (with geographic constraints) are well-known in folklore. It is sufficient, for example, to simply connect all MIS nodes within 3 hops, which can be accomplished in this setting in O (log 2 n ) rounds with a bounded randomized flood (see [3] for more discussion). Finally, we note that the dual graph model combined with the geographic constraint defined below is similar to the quasi-unit disk graph model [14]. The key difference, however, is that the dual graph model allows the set of unreliable links selected to change from round to round.
## 2 Model & Problems
The dual graph model describes a synchronous multihop radio network with both reliable and unreliable links. In more detail, the model describes the network topology with two graphs on the same vertex set: G = ( V, E ) and G ′ = ( V, E ′ ), where E ⊆ E ′ . The n = | V | nodes in V correspond to the wireless devices and the edges describe links. An algorithm in this model consists of n randomized processes . An execution of an algorithm in a given network ( G,G ′ ) begins with an adversary assigning each process to a node in the graph. This assignment is unknown to the processes. To simplify notation, we use the terminology node u , with respect to an execution and vertex u , to refer to the process assigned to node u in the graph in the execution. Executions then proceed in synchronous rounds. In each round r , each node decides whether to transmit a message or receive based on its randomized process definition. The communication topology in this round is described by the edges in E (which we call the reliable links) plus some subset (potentially empty) of the edges in E ′ \ E (which we call the unreliable links). This subset, which can change from round to round, is determined by a bounded adversary (see below for the adversary bounds we consider).
Once a topology is fixed for a given round, we use the standard communication rules for graphbased radio network models. That is, we say a node u receives a message m from node v in round r , if and only if: (1) node u is receiving; (2) node v is transmitting m ; and (3) v is the only node transmitting among the neighbors of u in the communication topology fixed by the adversary for r . Notice, the dual graph model is a strict generalization of the classical graph-based radio network model (they are equivalent when G = G ′ ).
/negationslash
Network Assumptions. To achieve the strongest possible lower bounds, we assume nodes are assigned unique ids from [ n ] (where we define [ ], for any integer i i > 0, to be the sequence { 1 2 , , ..., i } ). Structuring algorithms often require constraints on the network topology. In this paper, we say a dual graph ( G,G ′ ) satisfies the geographic constraint , if there exists some constant γ ≥ 1, such that we can embed the nodes in our graph in a Euclidean plane with distance function d , and ∀ u, v ∈ V , u = v : if d u, v ( ) ≤ 1 then ( u, v ) is in G , and if d u, v ( ) > γ , ( u, v ) is not in G ′ . This constraint says that close nodes can communicate, far away nodes cannot, and for nodes in the grey zone in between, the behavior is controlled by the adversary.
We consider two assumptions about nodes' knowledge regarding the dual graph. The advance neighborhood knowledge assumption provides every node u , at the beginning of the execution, the ids of its neighbors in G (which we also call u 's reliable neighbors). This assumption is motivated by the real world practice of providing wireless algorithms a low-level neighbor discovery service. The passive neighborhood knowledge assumption, by contrast, labels received messages at a node u with a 'reliable' tag if and only if the message was sent by a reliable neighbor. This assumption is motivated wireless cards' ability to measure the signal quality of received packets.
Adversary Assumptions There are different assumptions that can be used to bound the adversary that decides in the dual graph model which edges from E ′ \ E to include in the communication topology in each round. Following the classical definitions of adversaries in randomized analysis, in this paper we consider the following three types: (1) the offline adaptive adversary, which when making a decision on which links to include in a given round r , can use knowledge of the network topology, the algorithm being executed, the execution history through round r -1, and the nodes' random choices for round r ; (2) the online adaptive adversary, which is a weaker version of the offline adaptive variant that no longer learns the nodes' random choices in r before it makes its link decisions for r ; and (3) a static adversary, which includes the same set of unreliable links in every round. In this paper, when we refer to the ' 〈 adversary type 〉 dual graph model ', we mean the dual graph model combined with adversaries that satisfy the 〈 adversary type 〉 constraints.
/negationslash
The MIS and CDS Problems. Fix some undirected graph H = ( V, E ). We say S ⊆ V is a maximal independent set (MIS) of H if it satisfies the following two properties: (1) ∀ u, v ∈ S u , = v : { u, v } ∈ / E (no two nodes in S are neighbors in H ); and (2) ∀ u ∈ V \ S , ∃ v ∈ S : { u, v } ∈ E (every node in H is either in S or neighbors a node in S ). We say C ⊆ V is a connected dominating set (CDS) of H if it satisfies property 2 from the MIS definition (defined now with respect to C ), and C is a connected subgraph of H .
In this paper, we assume the structuring algorithms used to construct an MIS or CDS run for a fixed number of rounds then have each node output a 1 to indicate it joins the set and a 0 to indicate it does not (that is, we consider Monte Carlo algorithms). It simplifies some of the lower bounds that follow to exactly specify how a node makes its decision to output 1 and 0. With this in mind, in this paper, we assume that at the end of a fixed-length execution, the algorithm provides the nodes a function that each node will use to map the following information to a probability p ∈ [0 , 1] of joining the relevant set: (1) the node's id; (2) the ids of the node's neighbors (in the advance neighborhood knowledge setting); and (3) the node's message history (which messages it received and in what rounds they were received). The node then outputs 1 with probability p and 0 with probability 1 -p . For a given algorithm A , we sometimes use the notation A .out to reference this function.
We say a structuring algorithm A solves the MIS problem in f ( n ) rounds if it has each node output after f ( n ) rounds, for network size n , and this output is a correct MIS with respect to G (the reliable link graph) with at least constant probability. We say an algorithm A solves the CDS problem in f ( n ) rounds and provides a g n ( ) -approximation if it has each node output after f ( n ) rounds, it guarantees that this output is a correct CDS with respect to G , and it guarantees the
size of the CDS is within a factor of g n ( ) of the size of the minimum-sized CDS for G , for network size n , also with at least constant probability.
## 3 The Necessity of Geographic Constraints
We begin by proving that the geographic constant is necessary to efficiently solve the MIS and CDS problem in the dual graph model. In both bounds we assume an online adaptive adversary, which is weaker than the offline adaptive adversary assumed in [3]-strengthening our results.
## 3.1 MIS Lower Bound
We prove that without the geographic constraint every MIS solution requires a time complexity that is arbitrarily close to Ω( n/ log n ) rounds. This is exponentially worse than the O (log 3 n )-round solution possible with this constraint. Our proof argument begins by introducing and bounding an abstract game that we call selective ring coloring . This game is designed to capture a core difficulty of constructing in MIS in this unreliable setting. We then connect this game to the MIS problem using a reduction argument.
The Selective Ring Coloring Game. The ( g, n )-selective ring coloring game is defined a function g : N → N and some integer n > 0. The game is played between a player and a referee (both formalized as randomized algorithms) as follows. Let t ( n ) be the set containing all n ! ( n -3)! ordered triples of values from [ n ]. In the first round, the player generates a mapping C : t ( n ) →{ 1 2 3 , , } that assigned a color from { 1 2 3 , , } to each triple in t ( n ). Also during the first round, the referee assigns unique ids from [ n ] to a ring of size n . In particular, we define the ring as the graph R = ( V, E ), where V = { u , u 1 2 , ..., u n } , and E = {{ u , u i i +1 } | 1 ≤ i < n } ∪ { u , u n 1 } . Let /lscript : V → [ n ] be the bijection describing the referee's assignment of ids to this ring. The player and referee have no interaction during this round-their decisions regarding C and /lscript are made independently.
At the beginning of the second round, the player sends the referee C and the referee sends the player /lscript . Consider the coloring that results when we color each u i in the ring with color C /lscript ( ( u CC i ) , /lscript ( u i ) , /lscript ( u C i )), where u CC i and u C i are u i 's counterclockwise and clockwise neighbors, respectively. Notice, it is possible that this graph suffers from some coloring violations. This brings us to the third round. In the third round, the player generates a set S containing up to g n ( ) ids from [ n ]. It sends this set of exceptions to the referee. The referee considers the coloring of the nodes left in R once the exceptions and their incident edges are removed from the graph. If any coloring violations still remain, the referee declares that the player loses . Otherwise, it declares that the player wins .
A Lower Bound for Selective Ring Coloring. We now prove a fundamental limit on solutions to the selective ring coloring game. In particular, we prove that to win the game with constant probability requires a value of g n ( ) that is close to n . To prove this lower bound we will make use of a useful combinatorial result established in Linial's seminal proof of the necessity of Ω(log ∗ n ) rounds to constant-color a ring in the message passing model. This result concerns the following graph definition which captures relationships between possible t -neighborhoods of a ring with ids from [ m ]:
Definition 3.1. Fix two integers t and m , where t > 0 and m> t 2 +1 . We define the undirected graph B t,m = ( V t,m , E t.m ) as follows:
- · V t.m = ( { x , x 1 2 , ..., x 2 +1 t ) | ∀ i, j ∈ [2 t +1] : x i ∈ [ m,i ] = j ⇒ x i = x j } .
/negationslash
/negationslash
- · E t,m = {{ v , v 1 2 } | v 1 = ( x , ..., x 1 2 +1 t ) , v 2 = ( y, x 1 , ..., x 2 t ) , y = x 2 +1 t } .
/negationslash
Notice that in the context of the message passing model, each node ( x , x 1 2 , ..., x 2 +1 t ) in B t,m represents a potential view of a target node x t +1 in an execution, where view describes what ids a given node in the ring learns in a t round execution in this model; i.e., its id, and the id of nodes within t hops in both directions. The following result (adapted from Theorem 2 1 of [15]) bounds . the chromatic number of B t,m .
Lemma 3.2 (From [15]) . Fix two integers t and m , where t > 0 and m> t 2 +1 , and consider the graph B t,m . It follows: χ B ( t,m ) = Ω(log (2 ) t m ) , where log (2 ) t m is the 2 t times iterated logarithm of m .
We use this lemma in a key step in our following multi-step proof of the need for close to n exceptions to win selective ring coloring.
Lemma 3.3. Let P be a player that guarantees to solve the ( g, n ) -selective ring coloring game with constant probability, for all n . It follows that for every constant /epsilon1, 0 < /epsilon1 ≤ 1 : g n ( ) = Ω( n 1 -/epsilon1 ) .
Proof. Assume for contradiction that for some constant /epsilon1 that satisfies the constraints of the lemma statement, some g n ( ) = o n ( 1 -/epsilon1 ), and some player P that guarantees for all n to win the ( g, n )-selective ring coloring game with constant probability.
We start by describing a referee that will give P trouble. In more detail, to define /lscript , the referee first assigns i to node u i , for all i ∈ [ n ]. It then partitions the ring with this preliminary assignment into consecutive sequences of nodes each of length f ( n ) = n /epsilon1/ 5 . Finally, for each partition, it takes the ids assigned to nodes in the partition and permutes them with uniform randomness. We emphasize that the permutation in each partition is independent. We reference the n/f n ( ) partitions 2 of R as P , P 1 2 , ..., P n/f n ( ) . We use I 1 , I 2 , ..., I n/f n ( ) to describe the corresponding ids in each partition.
To understand the effectiveness of this strategy we start by exploring the difficulty of correctly coloring these partitions. Intuitively, we note that a 1-round coloring algorithm in the message passing model needs more than a constant number of colors to guarantee to correctly color all permutations of a non-constant-sized partition. This intuition will provide the core of our upcoming argument that many of our referee's random id assignments generate coloring violations for any given C provided by P . To formalize this intuition we leverage the result from Linial we established above. To do so, fix some P i . Let b be a bijection from [ | P i | ] to I i . Next consider B b 1 ,f ( n ) , which we define the same as B 1 ,f ( n ) , except we now relabel each vertex ( x, y, z ) as ( b x , b ( ) ( y , b ) ( z )). By Lemma 3.2, we know that χ B ( 1 ,f ( n ) ) = Ω(log (2) f ( n )). Clearly, this same result still holds for χ B ( b 1 ,f ( n ) ) (as we simply transformed the labels). Notice, because f ( n ) = ω (1), it follows that for sufficiently large n χ B , ( b 1 ,f ( n ) ) is strictly larger than 3. Fix this value of n for the remainder of this proof argument.
We now consider a specific instance of our game with P , our referee, and our fixed value of n . Focus as above on partition P i . Let C be the coloring function produced by the player and /lscript the assignment produced by our referee. Because we just established that the chromatic number of B b 1 ,f ( n ) is larger than 3, if we color this graph with C (which uses only three colors), there are (at least) two neighbors v 1 = ( x , x 1 2 , x 3 ) and v 2 = ( y, x 1 , x 2 ) in the graph that are colored the same.
It follows that if /lscript happens to assign the sequence of ids y, x 1 , x 2 , x 3 to four consecutive nodes in P i , C will color x 1 and x 2 the same, creating a coloring violation. 3 We can now ask what is
2 For simplicity of notation, we assume f ( n ) and n/f n ( ) are whole numbers. We can handle the other case through the (notationally cluttered) use of ceilings and floors.
3 A subtlety in this step is that we need | P i | = f ( n ) ≥ 4. If this is not true for the value of n fixed above we can just keep increasing this value until it becomes true.
the probability that this bad sequence of ids is chosen by /lscript ? This probability is crudely lower bounded by | I i | -4 = f -4 ( n ). We can now expand our attention to the total number of partitions with coloring violations. To do so, we define the following indicator variables to capture which partitions have coloring violations:
$$\forall j \in [ n / f ( n ) ], X _ { j } = \begin{cases} 1 & \text{if $P_{j}$ has a coloring violation w.r.t. $C$ and $\ell$,} \\ 0 & \text{else.} \end{cases}$$
We know from above that for any particular j , Pr[ X j = 1] > f -4 ( n ). It follows directly from our process for defining /lscript that this probability is independent for each X j . If Y = X 1 + X 2 + ... + X n/f n ( ) is the total number of coloring violations, therefore, by linearity of expectation, and the fact that f ( n ) = n /epsilon1/ 5 , the following holds:
$$\mathbb { E } [ Y ] = \mathbb { E } [ X _ { 1 } ] + \mathbb { E } [ X _ { 2 } ] + \dots + \mathbb { E } [ X _ { n / f ( n ) } ] > \frac { n } { f ^ { 5 } ( n ) } = n ^ { 1 - \epsilon }.$$
A straightforward application of Chernoff tells us that Y is within a constant factor of this expectation with high probability in n . We are now ready to pull together the pieces to reach our contradiction. We have shown that with high probability our referee strategy, combined with player P , generates at least n 1 -/epsilon1 coloring violations, with high probability. We assumed, however, that g n ( ) = o n ( 1 -/epsilon1 ). It follows (for sufficiently large n ) that with high probability, the player will not have enough exceptions to cover all the coloring violations. His success probability, therefore, is subconstant. This contradicts our assumption that the player wins with at least constant probability for this definition of g .
Connecting Selective Ring Coloring to the MIS Problem. Our next step is to connect the process of building an MIS in our particular wireless model to achieving efficient solutions to the ring coloring game we just bounded. At a high-level, this argument begins by noting that if you can build an MIS fast then you can three color a ring fast. It then notes this if you can three-color a ring fast in our online adaptive model, then you can do so with an adversary that ends up forcing many partitions in the ring to decide without receiving a message (and therefore, base their decision only on the ids of their reliable neighbors). To conclude the proof, we show that the coloring generated by this function can be used to win the selective ring coloring game. The faster the original MIS algorithm works, the smaller the g for which it can solve selective ring coloring.
Lemma 3.4. Let A be an algorithm that solves the MIS problem in g n ( ) rounds, for some polynomial g , in the online adaptive dual graph model with a network size of n , advance neighborhood knowledge, but no geographic constraint. It follows that there exists a player P A that solves the ( g , n ′ ) -selective coloring game with some constant probability p ′ , where g ′ ( n ) = O g n ( ( ) · log n ) .
Proof. Fix some A , g , and p matching the assumptions of the lemma statement. We first transform A into an algorithm A ′ that guarantees to provide a correct 3-coloring (also with probability p ) when executed in a ring. This new algorithm A ′ works as follows: (1) it runs A to generate an MIS in the ring; (2) all MIS nodes take color 1, then send a 'left' message to their left neighbor and a 'right' message to their right neighbor; (3) any non-MIS node the receives a 'left' message adopts color 2, and all other non-MIS nodes adopt color 3. (This transformation depends on our assumption that the ring is oriented, but since we are proving a lower bound, adding this assumption strengthens our result.)
We now define a selective ring coloring player P A that uses A ′ to generate its plays. In particular, in round 1 of the selective coloring game, the player must define C i, j, k ( ) for each ordered unique
triple ( i, j, k ) ∈ t ( n ), where n is the size fixed for this instance of the game. To do so, it starts by generating with uniform randomness a bit string κ i,j,k for each ( i, j, k ) ∈ t ( n ), where κ i,j,k is sufficiently long to generate a decision for any probability returned by A ′ .out (see Section 2). Next, for each such ( i, j, k ), it defines C i, j, k ( ) by using κ i,j,k to choose a color using the probability specified by A ′ .out when this function is applied to id j , (left and right) neighbors i and k , and an empty message history.
Let /lscript be the id assignment passed by the referee after this round. At this point, P A must use /lscript and C to generate its list of exceptions. To do so, it simulates A ′ in a dual graph network where G = R (i.e, the ring), G ′ is the complete graph over all nodes, and id assignments are from /lscript . It runs this simulation right up to the point where nodes need to make their coloring decisions. To then make the final coloring decisions, for each node j with left and right neighbors i and k , respectively, the player uses κ i,j,k to select the color using the probability specified by A ′ .out for j 's message history in this simulation. (Recall, it already generated all these κ strings during the first round. We are using the same κ definitions here.)
To implement this process, however, we must specify how the player handles message receive behavior in its simulation. To do so, the player implements an online adaptive adversary to control the unreliable links in the network. In particular, at the beginning of each round r , we note that by definition an online adaptive adversary knows the probability that each node u will broadcast in r . The player can therefore simulate the adversary calculating E [ B r ] at the beginning of round r , where B r is the number of nodes that end up broadcasting in r after probabilistic choices for this round are resolved. An adversary of this type can calculate E [ B r ] at the beginning of r , but not B r (as it does not know the bits in advance that will be used to resolve this round's probabilistic choices). The player simulates the adversary's decision in r based on the value of this expectation. In particular:
- · If E [ B r ] ≥ c log n , for some positive constant c we will fix later, the player has the adversary include all edges from G ′ \ G in the graph for the round.
- · If E [ B r ] < c log n , the player has the adversary include no edges from G ′ \ G .
Call this g n ( )-round simulation α . Let X be the set containing the id of every node that received a message in α . We have our player return X as its exception set.
We now analyze how our player fares with this playing strategy. To do so, we first note that the probability that the κ values chosen by the player generate a correct coloring in α is at least p , where p is the constant probability with which our algorithm guarantees to work. We next note that for an appropriately chosen constant c used in our adversary definition, with high probability, no more than O (log n ) nodes receive a message in any given round of α . This follows because in the case where E [ B r ] ≥ c log n (which has the adversary include all unreliable links in the topology), there is a high probability that at least two nodes broadcast, causing a collision everywhere and preventing any receptions, and when E [ B r ] < c log n (which has the adversary remove all unreliable links) with high probability no more than O (log n ) broadcast, and only their constant number of neighbors in the ring can receive these messages.
We now ask what is the probability that both events occur. That is, that α produces a correct coloring and has no more than O (log n ) nodes receive a message in any given round. Notice that these two events are not necessarily independent. If the adversary fails to bound communication, for example, perhaps the probability of a correct coloring increases or decreases. We can, however, easily deal with the dependence in this case by applying a union bound to the failure probability for each event. The probability that the coloring fails is upper bounded by some constant 1 -p , and the probability that the communication bound property fails is upper bounded by some sub-constant
value that is no more than 1 /n . As we grow n , this sums to some constant that approaches (1 -p ) from above. The probability that both events do occur, therefore, can be seen as some constant that approaches 1 -(1 -p ) = p as n increases. We can therefore fix some constant value n 0 for n and constant p ′ , such that for any n > n 0 , these two events holds with probability at least p ′ . For the constant number of network sizes that are of size ≤ n 0 , we can hardcode a trivial solution where all nodes broadcast their ids and their neighbor ids in round robin in order, then locally and consistently calculate a solution that guarantees both properties hold. For all larger values, we default to the standard algorithm. With this slightly modified algorithm it follows that both properties hold with constant probability.
With constant probability p ′ , therefore, α generates a correct coloring and an exception set X of size O g n ( ( ) · log n ). Because the ring is correctly colored, if we remove the nodes in X from the graph, then the remaining nodes are still correctly colored. Notice, however, that these nonX nodes received no messages in α (by definition), and therefore each such node choose the same color as returned by C for /lscript . Pulling together the pieces, we have shown that with some constant probability at least p ′ , the player P A wins the ( g , n ′ )-selective coloring game for some g ′ ( n ) = O g n ( ( ) · log n ).
Our final theorem follows directly from Lemmata 3.3 and 3.4:
Theorem 3.5. Let A be an algorithm that solves the MIS problem in f ( n ) rounds in the online adaptive dual graph model with a network size of n , advance neighborhood knowledge, and no geographic constraint. It follows that for every constant /epsilon1 , 0 < /epsilon1 ≤ 0 , f ( n ) = Ω( n 1 -/epsilon1 / log n ) .
As an immediate corollary to the above, we note that the family of functions described by Ω( n 1 -/epsilon1 / log n ) is equivalent to the family described by Ω( n 1 -/epsilon1 ), allowing for the omission of the log n divisor if desired in describing the bound.
## 3.2 CDS Lower Bound
We now prove the necessity of the geographic constraint for the CDS problem. In particular, we prove that in the absence of this constraint, any CDS algorithm that guarantees a reasonable approximation now requires Ω( √ n/ log n ) rounds. This is worse than the O (log 3 n ) solution that provides a O (1)-approximation that is possible with this constraint. Unlike our lower bound in the previous section, we do not use a reduction argument below. We instead deploy the more traditional strategy of using the definition of a fixed algorithm to construct a network in which the algorithm performs poorly.
Theorem 3.6. Let A be an algorithm that solves the CDS problem in f ( n ) rounds and provides a o ( √ n ) -approximation in the online adaptive dual graph model with a network size of n , advance neighborhood knowledge, and no geographic constraint. It follows that f ( n ) = Ω( √ n/ log n ) .
Proof. Assume for contradiction that there exists some algorithm A that achieves an o ( √ n )-approximation in f ( n ) ≤ √ n/ (2 log n ) rounds with (at least) constant probability p . Our proof proceeds in two steps. During the first step, we use the definition of A to construct a challenging dual graph network ( G ,G A ′ A ) and assignment of ids to nodes in that network. The second step describes and analyzse an online adaptive adversary that causes A , with sufficiently high probability, to either violate correctness or produce (at best) an Ω( √ n )-approximation of the minimum CDS when run in this network with these id assignments. This yields the needed contradiction.
Beginning with the first step, we fix k = √ n (assume for simplicity that √ n is a whole number, the proof easily extends to fractional values, but at the expense of increased notational cluttering).
To construct our dual graph ( G ,G A ′ A ), we first fix G ′ A to be the complete graph over all n nodes. (It is here we potentially violate the geographic constraint.) To define G A , we partition the set I = [ n ] of unique ids from 1 to n into sets C , C , ..., C 1 2 k of size k . We will now create a subgraph of size k in G A for each C h and assign ids from C h to these nodes. In particular, for each id partition C h , let i 0 ∈ C h be the smallest id in C h . We add a node to G A and assign it id i 0 . We call this the core node for C h . Moving forward in our process, let C ′ h = C h \ { i 0 } . We call C ′ h a point set .
We must now add nodes corresponding to the ids in point set C ′ h . To do so, for each i ∈ C ′ h , we define p h i to be the probability that i joins the CDS as defined by the function A .out applied to id i , neighbor set C ′ h \ { } i , and an empty message history (see Section 2). We call each such p h value a join probability . How we add nodes to the graph associated with point set C ′ h depends on the join probability values. In more detail, we consider two cases:
Case 1 : ∀ i ∈ C ′ h : p h i ≥ 1 2 / . In this case, we add a clique of size k -1 to the graph. We then assign the ids in C ′ h to nodes in this clique arbitrarily. Finally, we choose one i ∈ C ′ h to act as a connector , and connect the node with this id to the core node for C h that we previously identified. Notice, the neighbor set for i is different now than it was when we calculated p h i , but for all other nodes with ids in C ′ h , the neighbor sets are the same.
Case 2 : ∃ i ∈ C ′ h : p h i < 1 2 / . In this case, we add a clique of size k -2 to the graph, then add an edge from a single connector node in the clique to a new node, that we call the extender , then connect the extender to our previously identified core node for this set. Let i be the id from C ′ h for which the property that defines this case holds. We assign this id to the connector. We then assign the ids from C ′ h \ { } i to the clique and extender nodes arbitrarily. Notice, in this case, the node with the id i is the only id in C ′ h for which its neighbor set is the same here as it was when its join probability was calculated.
We repeat this behavior for every set C h , h ∈ [ k ]. Finally, to ensure our graph is connected, we add edges between all /lscript core nodes to form a clique.
Having now used the definition of A to define a specific reliable link graph G A , and an id assignment to this graph, consider the behavior of A when executed in ( G ,G A ′ A ), with this specfied id assignment, and an online adaptive adversary that behaves as follows in each round r . By definition, the adversary knows the probability that each node in the network will broadcast in this round, so it can therefore calculate E [ B r ], the expected value of B r , the actual number of broadcasters in round r . If E [ B r ] ≥ b log n , for a constant b > 0 we will fix later, then the adversary includes all edges in the network for r , and otherwise it includes no extra edges from G ′ \ G A .
At this point, we remind ourselves of our assumption that f ( n ) ≤ √ n/ (2 log n ). If our communication bound from above holds, it would then hold that in f ( n ) rounds, at least half of the √ n point sets received no messages. Moving forward, assume this property holds. Let us consider what will happen when the nodes in these silent point sets decide whether or not to join the CDS by using the probabilities specified by A .out applied to their neighborhood ids and an empty message history.
A Notice, this is the same online adaptive adversary strategy we used in the proof of Lemma 3.4, and as in that proof, a standard Chernoff analysis tells us that for any constant c ≥ 1, there exists a constant b that guarantees that with probability at least 1 -n -c , in any round in which more than log n nodes broadcast, all edges from G ′ A are included in the network by the adversary. If we combine this property with the observation that no node in G A neighbors more than one set C ′ h (by 'neighbors C h ' we mean neighbors at least one node in C h ), it follows that with this same high probability no more than log n point sets include a node that receives a message in any given round.
There are two possibilities. The first possibility is that half or more of these silent points sets fell under Case 1 from our above procedure. For each such point set, there are √ n -2 nodes that
will now join the CDS with probability at least 1 / 2 (i.e., the nodes in C ′ h with the exception of the connector). The expected number of nodes that join from this point set is therefore at least √ n -2 2 . (Key in this result is the fact that these nodes are in silent sets, which means they have received no messages, and therefore their behavior is based on an independent coin flip weighted according to the probability returned by A .out .) Given that we have at least √ n/ 2 such silent point sets, the total expected number of nodes that join is in Ω( n ) (by linearity of expectation). A Chernoff bound concentrates this expectation around the mean and provides that with high probability in n , the total number of nodes that join is within a constant factor of this linear expectation.
The second possibility is that half or more of these silent point sets fall under Case 2. For each such silent point set, the connector node does not join with probability at least 1 2. / Notice, if the connector does not join, then its point set is disconnected from the rest of the network, and therefore, the overall CDS is not correct. Because there are at least √ n/ 2 silent point sets in this case that are violating correctness with probability at least 1 / 2, the probability that this CDS is correct is exponentially small in n .
We are left to combine the probabilities of the relevant events. We have shown that with high probability, a f ( n )-round execution of A in ( G ,G A ′ A ), with our above adversary strategy, concludes with at least half of the point sets having received no messages. If this event occurs, then there are two possibilities analyzed above concerning whether these silent point sets mainly fall under Case 1 or 2 from our graph construction procedure. We proved that the first possibility leads to a Ω( √ n )-approximation with high probability (as with this probability, Ω( n ) nodes join in a network where O ( √ n ) nodes is sufficient to form a CDS), and the second possibility leads to a lack of connectivity with (very) high probability. A union bound on either of these two events (many silent sets and bad performance given many silent sets) failing yields a sub-constant probability. This probability, however, upper bounds the probability of the algorithm satisfying the theorem. This provides our contradiction.
## 4 The Necessity of Advance Neighborhood Knowledge
In this section we explore the importance of advance neighborhood knowledge by proving new lower bounds for the MIS and CDS problems when provided the weaker assumption of passive neighborhood knowledge.
## 4.1 CDS Lower Bound
In this section, we prove that any CDS solution requires both the geographic constraint and advance neighborhood knowledge. In more detail, we prove below that if we assume the geographic constraint but only passive neighborhood knowledge, any CDS algorithm now requires either a slow time complexity or a bad approximation factor (formally, these two values must add to something linear in the network size). Our bound reduces k -isolation, a hard guessing game, to the CDS problem.
The k -Isolation Game. The game is defined for an integer k > 0 and is played by a player P modeled as a synchronous randomized algorithm. At the beginning of the game, a referee chooses a target value t ∈ [ k ] with uniform randomness. The player P now proceeds in rounds. In each round, the player can guess a single value i ∈ [ k ] by sending it to referee. If i = , the player wins. t Otherwise, it is told it did not win and continues to the next round. Once again, the straightforward probabilistic structure of the game yields a straightforward bound:
Lemma 4.1. Fix some k > 1 and r ∈ [ k ] . No player can win the k -isolation game in r rounds with probability better than r/k .
Connecting Isolation to the CDS Problem. We now reduce this isolation game to CDS construction. To do so, we show how to use a CDS algorithm to construct an isolation game player that simulates the algorithm in a barbell network (two cliques connected by a single edge) with the bridge nodes indicating the target.
Theorem 4.2. Let A be an algorithm that solves the CDS problem in f ( n ) rounds and provides a g n ( ) -approximation in the static dual graph model with a network of size n , passive neighborhood knowledge, and the geographic constraint. It follows that g n ( ) + f ( n ) = Ω( n ) .
Proof. Assume for contradiction that there exists some A that solves the problem in this setting for some f ( n ) + g n ( ) = o n ( ) with some constant probability p . We will use A to construct a player P A for the k -isolation game that contradicts the lower bound of Lemma 4.1. In more detail, given a target t for the isolation game, we define a graph G t of size k that has two cliques of size k/ 2 connected by a single edge. 4 The first (resp. second) clique is assigned ids from 1 to k/ 2 (resp. k/ 2 + 1 to k ). If t < k/ 2 (resp. t ≥ k/ 2) we add the bridge from the node with id t to the node with id k/ 2 + t (resp. t to t -k/ 2).
Our player P A simulates A running in G t with a complete G ′ and the static adversary that includes all G ′ edges in every round. (Notice that this dual graph definition satisfies our geographic constraint.) The player, of course, does not know the full definition of G t in advance as it does not know t in advance, but we will show that its guesses allow it to keep the simulation correct until it wins. In more detail, in each round r of the simulation, the player simulates the round up until the broadcast decisions are made. Let B r be the set of the ids of the nodes that broadcast in r . If | B r | = 0 or | B r | > 1 then no node will receive any message (regardless of the identity of the bridge), so the player can simply simulate silence for this round. The interesting case is when B r = { } i . Here, the player will make two guesses (i.e., play two rounds of the game). If i ≤ k/ 2, it will guess i then k/ 2 + i , and if i > k/ 2, it will guess i and i -k/ 2. If i is a bridge node, the player wins the game. If i is not a bridge node, then the player learns this (by not winning) and can simulate all nodes receiving the message but only nodes in the same clique as the sender receiving a 'reliable' tag. If after f ( k ) rounds of simulation, the player has still not won the game, it then guesses the id of every node that joins the CDS at the end of this round.
To conclude the proof, we note that a correct CDS contains the bridge nodes (to satisfy connectivity). We also note that in an f ( k ) round simulation the player makes at most O f k ( ( ) + g k ( )) guesses (two guesses per round, and one guess per node in a CDS that is within a factor of g k ( ) of the constant size minimum CDS for this network). Therefore, if A satisfies our above assumptions, P A wins the k -isolation game with constant probability with O f k ( ( ) + g k ( )) = o k ( ) guessescontradicting Lemma 4.1.
## 4.2 MIS Lower Bound
It is straightforward to show that the MIS algorithm from [3] still works if we replace the advance neighborhood knowledge assumption with its passive alternative (the algorithm uses this knowledge only to discard messages it receives from unreliable neighbors). We prove below, however, that the passive assumption increases the fragility of any MIS solution. In particular, we show that when we
4 We assume for now that k is even, the proof is easily extended at the expense of extra notation to handle the odd case.
switch from advance to passive, the bound from Section 3.1 now increases to Ω( n ) and still holds even with a static adversary. As before, we use a reduction argument from a hard guessing game. The k -Bit Revealing Game The game is defined for an integer k > 0 and is played by a player P modeled as a synchronous randomized algorithm. At the beginning of the game, a referee generates a sequence κ of k bits, where each bit is determined with uniform and independent randomness. In the following, we use the notation κ i [ ], for i ∈ [ k ], to refer to the i th bit in this sequence. The player P now proceeds in rounds. In each round, it can request a value i ∈ [ k ], and the adversary will respond by returning κ i [ ]. At the end of any round (i.e., after the bit is revealed), the player can decide to guess κ by sending the referee a sequence ˆ of κ k bits. If ˆ = κ κ , the player wins; otherwise it loses. We say a player P solves the k -bit revealing game in f ( k ) rounds with probability p , if with probability p it wins the game by the end of round f ( k ). Given the well-behaved probabilistic structure of this game, the following bound is straightforward to establish.
Lemma 4.3. Fix some k > 1 and t ∈ [ k ] . No player can solve the k -bit revealing game in t rounds with probability p > 2 -( k -t ) .
Connecting Bit Revealing to the MIS Problem. We now reduce our bit revealing game to the more complex task of building an MIS in a non-geographic static dual graph network. In particular, we will show how to use an MIS algorithm to solve the bit revealing game by having a player simulate the algorithm in a carefully constructed dual graph network. In this network, we partition nodes into sets, such that we can match these sets to bits, and use the MIS decisions of nodes in a given set to guess the corresponding bit in the revealing game.
Theorem 4.4. Let A be an algorithm that solves the MIS problem in f ( n ) rounds in the static dual graph model with a network of size n , passive neighborhood knowledge, and no geographic constraint. It follows that f ( n ) = Ω( n ) .
Proof. Assume for contradiction that there exists some A that solves the problem in f ( n ) = o n ( ) rounds with some constant probability p . We will use A to construct a player P A for the k -bit revealing game that contradicts the lower bound of Lemma 4.3.
In more detail, for a given bit string κ of length k , we define a graph G κ of size 6 k as follows. Start with a line of k nodes. Call each node an anchor and label them (for the sake of this construction process), 1 2 3 , , , ..., k . Partition the remaining 5 k nodes into sets. For anchor i , if κ i [ ] = 0, then arrange the nodes in set i into a line and connect anchor i to an endpoint of this line. On the other hand, if κ i [ ] = 1, then arrange the nodes in set i into a clique and connect anchor i to a single node in this clique.
In a given instance of the k -bit revealing game with target bit string κ , our player P A works by simulating A for n = 6 k in the dual graph consisting of G = G κ and G ′ as the complete graph over all n nodes. (Notice, for sufficiently large k , this network does not satisfy our geographic constraints for dual graph networks.) In this simulation, the player assumes that every edge in G ′ is included in the network in every round. Our player will simulate one round of A for each round of the game, using the behavior of the nodes in the simulation to help determine its guess, and using the response to the guess to help keep its simulation correct. When the simulation terminates after f ( n ) = f (6 k ) rounds, the player will generate a guess ˆ based on the MIS generated in the κ simulation. We will prove that if A outputs a correct MIS in this simulation then P A will correctly guess κ . Of course, P A does not know κ in advance and therefore does not know the definition of G κ in advance. We will show, however, that this does not matter as we will keep our simulation valid for the unknown G κ in every step.
In more detail, each round of r , 1 ≤ r ≤ f (6 k ) of the player's simulation work as follows. Let B r be the set of nodes that broadcast in r . If | B r | = 0 or | B r | > 1 then in our simulated network it is
clear to see that no node receives a message (recall that G ′ is the complete graph and all edges are included). The player, therefore, can simulate the nodes in B r broadcasting and no node receiving any message. It can then make any arbitrary request in this round of the bit revealing game and ignore the answer.
The interesting case is where B r = { u } . Let i correspond to u 's position in G κ (i.e., u is either anchor i or in set i ). To correctly simulate receive behavior we must know if set i is arranged in a clique or a line. To determine this, we have the player request κ i [ ] from the referee. The value of κ i [ ] allows the player to determine the structure of set i in G κ and therefore correctly simulate receive behavior. (Notice, in this case, the question of who receives the message is straightforward, as given our definition of G ′ and the static adversary, all nodes will receive it. The unknown solved by learning κ i [ ] is which nodes will receive the message with u labelled as a G -neighbor .)
Finally, after f (6 k ) rounds, A has each node output 1 to indicate it is in the MIS and 0 to indicate it is not. We have our player uses these MIS decisions to generate its guess ˆ for the bit κ revealing game. In particular, for each bit position i , we determine ˆ[ ] based on the output of κ i nodes in set i in our simulation as follows: (1) if ≤ 1 node in set i outputs 1 then set ˆ[ ] κ i ← 1; (2) if ≥ 2 nodes in set i outputs 1 then set ˆ[ ] κ i ← 0. We claim that if A outputs a correct MIS in its network then ˆ = κ κ . To see why, notice that a correct MIS cannot have more than 1 node in a given set i if set i is a clique (as this would generate an independence violation), and it cannot have less than 2 nodes in a given set i if set i is a line (as a line of length 5 requires at least 2 MIS nodes to satisfy maximality, even if anchor i is in the MIS as well).
Pulling together the pieces, we see that if A solves the MIS problem in f (6 k ) = o k ( ) rounds with constant probability p , then it solves the k -bit revealing game in o k ( ) rounds also with probability at least p . By Lemma 4.3, however, Ω( ) rounds are k needed for this success probability. A contradiction.
## References
| [1] | R. Bar-Yehuda, O. Goldreich, and A. Itai. On the Time Complexity of Broadcast in Radio Networks: an Exponential Gap Between Determinism and Randomization. In Proceedings of the ACM Conference on Distributed Computing , 1987. |
|-------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [2] | Reuven Bar-Yehuda, Oded Goldreigch, and Alon Itai. On the Time-Complexity of Broadcast in Multi-Hop Radio Networks: An Exponential Gap between Determinism and Randomization. Journal of Computer and System Sciences , 45(1):104-126, 1992. |
| [3] | K. Censor-Hillel, S. Gilbert, F. Kuhn, N. Lynch, and C. Newport. Structuring Unreliable Radio Networks. In Proceedings of the ACM Conference on Distributed Computing , 2011. |
| [4] | I. Chlamtac and S. Kutten. On Broadcasting in Radio Networks-Problem Analysis and Pro- tocol Design. IEEE Transactions on Communications , 33(12):1240-1246, 1985. |
| [5] | A. Clementi, A. Monti, and R. Silvestri. Round Robin is Optimal for Fault-Tolerant Broad- casting on Wireless Networks. Journal of Parallel and Distributed Computing , 64(1):89-96, 2004. |
| [6] | S. Daum, F. Kuhn, and C. Newport. Efficient Symmetry Breaking in Multi-Channel Radio Networks. In Proceedings of the International Conference on Distributed Computing , 2012. |
| [7] | Martin Farach-Colton, Rohan J. Fernandes, and Miguel A. Mosteiro. Lower Bounds for Clear Transmissions in Radio Networks. In Proceedings of the Latin American Symposium on The- oretical Informatics , 2006. |
| [8] | M. Ghaffari, B. Haeupler, N. Lynch, and C. Newport. Bounds on Contention Management in Radio Networks. In Proceedings of the International Conference on Distributed Computing , 2012. |
| [9] | Mohsen Ghaffari, Nancy Lynch, and Calvin Newport. The Cost of Radio Network Broadcast for Different Models of Unreliable Links. In Proceedings of the ACM Conference on Distributed Computing , 2013. |
| [10] | T. Jurdzinski and G. Stachowiak. Probabilistic Algorithms for the Wakeup Problem in Single- Hop Radio Networks. In Proceedings of the International Symposium on Algorithms and Com- putation , 2002. |
| [11] | F. Kuhn, N. Lynch, C. Newport, R. Oshman, and A. Richa. Broadcasting in Unreliable Radio Networks. In Proceedings of the ACM Conference on Distributed Computing , 2010. |
| [12] | Fabian Kuhn, Nancy Lynch, and Calvin Newport. Brief Announcement: Hardness of Broad- casting in Wireless Networks with Unreliable Communication. In Proceedings of the ACM Conference on Distributed Computing , 2009. |
| [13] | Fabian Kuhn, Thomas Moscibroda, and Rogert Wattenhofer. Initializing Newly Deployed Ad Hoc and Sensor Networks. In Proceedings of the International Conference on Mobile Computing and Networking , 2004. |
| [14] | Fabian Kuhn, Rogert Wattenhofer, and Aaron Zollinger. Ad-Hoc Networks Beyond Unit Disk Graphs. In Proceedings of the International Workshop on the Foundations of Mobile Computing , 2003. |
| [15] | Nathan Linial. Locality in Distributed Graph Algorithms. SIAM Journal on Computing , 21(1):193-201, 1992. |
|--------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [16] | Thomas Moscibroda and Roger Wattenhofer. Maximal Independent Sets in Radio Networks. In Proceedings of the ACM Conference on Distributed Computing , 2005. |
| [17] | Thomas Moscibroda and Roger Wattenhofer. The Complexity of Connectivity in Wireless Networks. In Proceedings of the IEEE International Conference on Computer Communications , 2006. |
| [18] | Calvin Newport, David Kotz, Yougu Yuan, Robert S Gray, Jason Liu, and Chip Elliott. Experimental Evaluation of Wireless Simulation Assumptions. Simulation , 83(9):643-661, 2007. |
| [19] | K. Srinivasan, M.A. Kazandjieva, S. Agarwal, and P. Levis. The β -Factor: Measuring Wireless Link Burstiness. In Proceedings of the Conference on Embedded Networked Sensor System , 2008. |
| null |
[
"Calvin Newport"
] |
2014-08-04T20:39:55+00:00
|
2014-08-04T20:39:55+00:00
|
[
"cs.DC"
] |
Lower Bounds for Structuring Unreliable Radio Networks
|
In this paper, we study lower bounds for randomized solutions to the maximal
independent set (MIS) and connected dominating set (CDS) problems in the dual
graph model of radio networks---a generalization of the standard graph-based
model that now includes unreliable links controlled by an adversary. We begin
by proving that a natural geographic constraint on the network topology is
required to solve these problems efficiently (i.e., in time polylogarthmic in
the network size). We then prove the importance of the assumption that nodes
are provided advance knowledge of their reliable neighbors (i.e, neighbors
connected by reliable links). Combined, these results answer an open question
by proving that the efficient MIS and CDS algorithms from [Censor-Hillel, PODC
2011] are optimal with respect to their dual graph model assumptions. They also
provide insight into what properties of an unreliable network enable efficient
local computation.
|
1408.0813v1
|
## DISCOVERY OF A COMPANION CANDIDATE IN THE HD169142 TRANSITION DISK AND THE POSSIBILITY OF MULTIPLE PLANET FORMATION
Maddalena Reggiani 1 , Sascha P. Quanz 1 , Michael R. Meyer 1 , Laurent Pueyo 2 , Olivier Absil 3 , Adam Amara 1 , Guillem Anglada 4 , Henning Avenhaus 1 , Julien H. Girard 5 , Carlos Carrasco Gonzalez 6 , Graham James 7 , Dimitri Mawet 5 , Farzana Meru 1 , Julien Milli 5 , Mayra Osorio 4 , Schuyler Wolff 2 , and Jose-Maria Torrelles 8
## ABSTRACT
We present L ′ and J -band high-contrast observations of HD169142, obtained with the VLT/NACO AGPM vector vortex coronagraph and the Gemini Planet Imager, respectively. A source located at 0'.156 ± 0'.032 north of the host star ( PA =7.4 ◦ ± 11.3 ◦ ) appears in the final reduced L ′ image. At the distance of the star ( ∼ 145 pc), this angular separation corresponds to a physical separation of 22.7 ± 4.7 AU, locating the source within the recently resolved inner cavity of the transition disk. The source has a brightness of L ′ =12.2 ± 0.5 mag, whereas it is not detected in the J band ( J > 13.8 mag). If its L ′ brightness arose solely from the photosphere of a companion and given the J -L ′ color constraints, it would correspond to a 28-32 M Jupiter object at the age of the star, according to the COND models. Ongoing accretion activity of the star suggests, however, that gas is left in the inner disk cavity from which the companion could also be accreting. In this case the object could be lower in mass and its luminosity enhanced by the accretion process and by a circumplanetary disk. A lower mass object is more consistent with the observed cavity width. Finally, the observations enable us to place an upper limit on the L ′ -band flux of a second companion candidate orbiting in the disk annular gap at ∼ 50 AU, as suggested by millimeter observations. If the second companion is also confirmed, HD169142 might be forming a planetary system, with at least two companions opening gaps and possibly interacting with each other.
Subject headings: planet-disk interactions - planets and satellites: formation - protoplanetary disks stars: low-mass, brown dwarfs - stars:individual (HD169142)
## 1. INTRODUCTION
2013).
To understand how planet formation proceeds and in which chemical and physical conditions it occurs, young, gas rich disks have been studied by numerous observing programs. The so-called 'transition disks' show inner holes, bright rims and annular gaps, which could be tracing on-going planet formation (Espaillat et al. 2014).
Afew planet candidates have been found in these disks. Some of them were detected with sparse aperture masking (SAM) observations in the disk gaps around their host stars (e.g. LkCa15 b, TCha b; Kraus & Ireland 2012; Hu´ elamo et al. 2011), but disk features or scattered light from inner disk rims have been suggested to explain some of these observations (e.g. Cieza et al. 2013; Thalmann et al. 2014). Coronagraphy supported angular differential imaging (ADI, Marois et al. 2006) and spectroastrometry of CO rovibrational lines provided direct empirical evidence for two companions orbiting the young star HD100546 (Quanz et al. 2013a; Brittain et al.
- 1 Institute for Astronomy, ETH Zurich, CH-8093 Zurich, Switzerland; [email protected]
- 3 D´ epartement d'Astrophysique, G´ophysique et Oc´anographie, e e Universit´ de Li`ge, 17 All´e du Six Aoˆt, 4000 Li`ge, Belgium e e e u e
- 2 Space Telescope Science Institute, 3700 San Martin Dr, Baltimore, MD 21218, United States
- 4 Instituto de Astrof´ ısica de Andaluc´ ıa, CSIC, Glorieta de la Astronom´ ıa s/n, 18008 Granada, Spain
- 5 European Southern Observatory, Alonso de Cordova 3107, Casilla 19001, Vitacura, Santiago 19, Chile
- 6 Centro de Radioastronom´ ıa y Astrof´ ısica (UNAM), Apartado Postal 3-72 (Xangari), 58089 Morelia, Mexico
- 7 University of California, 644 Campbell Hall, Berkeley
- 8 Instituto de Ciencias del Espacio (CSIC)-UB/IEEC, Universitat de Barcelona, Mart´ ı i Franqu`s 1, 08028 Barcelona, Spain e
In this letter we present L ′ - and J -band high-contrast imaging observations of HD169142 , 9 which reveal the presence of a companion candidate in the inner cavity of its transition disk.
HD169142 is a young Herbig Ae/Be star (see Table 1 for stellar properties) with a complex circumstellar disk structure, that has been extensively studied (e.g., Dent et al. 2006; Grady et al. 2007; Meeus et al. 2010; Honda et al. 2012; Quanz et al. 2013b). It has a small central disk ( < 0.7 AU) with either a hot-dust halo (Honda et al. 2012) or a hot inner wall (Osorio et al. 2014), an inner cavity, a bright rim ( ∼ 25 AU), and an annular gap from 40 to 70 AU. The latter features have recently been resolved with polarimetric differential imaging (PDI) in the H band (Quanz et al. 2013b). Furthermore, EVLA 7-mm observations of the disk detected the thermal dust emission of the bright rim, seen in polarized light, and revealed a compact emission source inside the annular gap (at ∼ 50 AU; Osorio et al. 2014). These recent observations as well as the morphology of the disk suggest that HD169142 could be hosting young planetary companions.
## 2. OBSERVATIONS AND DATA REDUCTION
## 2.1. NACO L' band
HD169142 was observed on June 28, 2013 with the VLT/NACO annular groove phase mask (AGPM) vector vortex coronagraph (Mawet et al. 2013) in pupil sta-
- 9 Based on observations made with ESO Telescopes at the Paranal Observatory under program 291.C-5020(A) and with Gemini during GPI Early Science runs under program GS-2014A-SV407
TABLE 1 Stellar Parameters.
| Parameter | Value | Reference |
|-----------------------------------|----------------------------|---------------|
| R.A. (J2000) | 18h 24m 29.785s | [1] |
| Dec. (J2000) | -29 ◦ 46' 49.829' | [1] |
| Distance (pc) | 145-151 | [6]; [3] |
| J (mag) | 7.31 ± 0.02 | [1] |
| H (mag) | 6.91 ± 0.04 | [1] |
| K s (mag) | 6.41 ± 0.02 | [1] |
| L' (mag) | 5.66 ± 0.03 | [7] |
| Sp. Type | A9III/IVe/A7V | [2]; [3] |
| v sin i (km s - 1 ) | 55 ± 5 | [2] |
| Age (Myr) | 1-5/12/3-12 | [2]; [3]; [4] |
| T eff | 7500 ± 200/6500/7650 ± 150 | [2]; [3] |
| Mass ( M /circledot ) | ∼ 1.65 | [3] |
| L ∗ ( L /circledot ) | ∼ 8.6-13 | [3]; [8] |
| R ∗ ( R /circledot ) | ∼ 1.6 | [3]; [5] |
| ˙ M (10 - 9 M /circledot yr - 1 ) | ∼ 3.1/ ≤ 1 . 25 ± 0 . 55 | [3]; [4] |
| log g | 3.7 ± 0.1/4.0-4.1 | [2]; [5] |
References : [1] From 2MASS point source catalog (Cutri et al. 2003) and corrected for proper motions to the epoch of our VLT observations; [2] Guimar˜es et al. (2006); [3] Blondel & Djie a (2006); [4] Grady et al. (2007); [5] Meeus et al. (2010); [6] Sylvester et al. (1996); [7] van der Veen et al. (1989); [8] Mari˜as et al. (2011) n bilized mode. All images were taken with the L27 camera (plate scale ∼ 27.15 mas pixel -1 ) using the L ′ filter ( λ c = 3 8 . µm , ∆ λ = 0 62 . µm ). The detector reads were recorded in 'cube' mode and the integration time per read was set to 0.25 s. We obtained 111 minutes of onsource integration time and ∼ 159 ◦ of field rotation. The sky was observed every ∼ 20 minutes and unsaturated images (0.05 s) of the star were acquired to calibrate the photometry. Table 2 summarizes the observations.
We subtracted the background from each frame adopting for each cube the mean of the closest sky measures in time. We applied a bad pixel/cosmic ray correction, adopting a 5-sigma threshold and replacing every anomalous pixel with the mean value of the 8 surrounding pixels. Since the AGPM already requires a mandatory centering accuracy of ∼ 0.3 pixel, which has been checked every 10-30 minutes during the observations, we did not apply any further centering to our images. As a test, additional centering was performed in a second data reduction pipeline (see Section 3), yielding consistent results. Finally from each image we created a ∼ 2' × 2' sub-image (75 × 75 pixels) centered on the star, resulting in a stack of ∼ 25000 sub-images.
To subtract the stellar PSF from all the sub-images we used the principal component analysis (PCA) based package PYNPOINT 10 (Amara & Quanz 2012). PYNPOINT creates a set of orthogonal basis functions to reproduce the stellar PSF and fits it to the individual frames with a chosen number of PCA coefficients. It subtracts the PSF from each frame, de-rotates the frames to the same field rotation, averages them and convolves them with a Gaussian kernel (0.5 × FWHM PSF ) to get the final image of the stack. We adopted the 20 PCA PYNPOINT image as final reference, as it shows lower residual noise compared to images obtained with higher or lower numbers of PCA coefficients.
10 Public version of the code can be found at http://pynpoint.ethz.ch and additional material in Amara, Quanz, and Akeret, Astronomy and Computing (submitted).
## 2.2. GPI J band
Spectral data were obtained in J band (1.12-1.35 µm , R =35-39) on April 26, 2014 with the Gemini Planet Imager (GPI, Macintosh et al. 2014) using an apodized Lyot coronagraph in pupil stabilized mode (see Table 2). The total on source integration time was 52 minutes with a total field rotation of ∼ 6 . ◦
We adopted the GPI pipeline (Maire, Perrin, Doyon, et al. 2010, 2012) for bad-pixel removal, de-striping, flat-fielding, wavelength calibration, and to convert the data into spectral data cubes. In total, the data are made of 52 cubes, consisting of 37 spectral channels each. The cubes were then aligned and registered according to the procedure described in Crepp et al. (2011) and Pueyo et al. (2012) in order for the data to be placed in a frame where the speckles scale is fixed as a function of wavelength while putative point sources move radially. The KLIP algorithm was then carried out as presented in Soummer et al. (2012), using the variations for IFS data described in Pueyo et al 2014 (in review). Finally, we reduced each slice of each cube separately and then mean combined them both in wavelength and time.
## 3. RESULTS
## 3.1. Detection of an Emission Source in the L' band
In the final L ′ -band image an emission source is revealed north of HD169142 (see Figure 1a). To assess its reliability, we performed a series of tests.
- 1. We varied the number of PCA coefficients used in PYNPOINT between 5 and 120.
- 2. We divided the dataset into different subsets containing either half or a third of the frames, but spanning the full field rotation.
- 3. We did two 'blind' data reductions to confirm the result using both a separate PCA-based pipeline (Absil et al. 2013; Mawet et al. 2013) and the LOCI algorithm (Lafreni` ere et al. 2007).
In each case, we always found a bright emission source at the same location. Since the residual speckle noise does not follow a Gaussian distribution, the calculation of Gaussian confidence levels may not be appropriate (see e.g. Kasper et al. 2007). To estimate the statistical confidence of the detection we used the final image and selected 28 pixels in 2 concentric rings around the star as noise reference. 10 pixels had the same separation from the star as the peak flux of the companion, the ring of the other 18 pixels had a radius of 0'.23 and included a bright residual feature east of the central star. These pixels were chosen to be statistically independent in the convolved image and were used to compute the mean, variance and skewness of the distribution and built a probability density function (PDF) assuming a log-normal underlying distribution. From this PDF we estimated the likelihood of finding a pixel value equaling the companion's peak flux or higher to be p < 0 2%. .
The results of all these tests give us confidence that the detection is real. None of the other features in the final image is a reliable detection, based on these tests.
TABLE 2 Summary of Observations.
| Instrument | Filter | No. of detector reads × exp. time | No. of data cubes | Parallactic angle start/end | Airmass range |
|--------------|----------|-------------------------------------|---------------------|-------------------------------|-----------------|
| VLT/NACO | L' | 60 × 0.25 s | 444 | -84.29/74.70 | 1.097 -1.038 |
| Gemini/GPI | J-coro | 1 × 60 s | 52 | -96.65/-102.96 | 1.048 -1.001 |
Fig. 1.a) NACO/AGPM L ′ image of HD169142, using PYNPOINT with 20 PCA coefficients. A bright source is detected north of the central star. The image is scaled with respect to the maximum flux. b) H -band PDI image of the circumstellar disk of HD169142 (Quanz et al. 2013b). The inner cavity ( < 25 AU), the bright rim, and the annular gap (40-70 AU) are clearly visible. Overplotted in red contours is the detected L ′ source. The green diamond indicates the location of the compact 7-mm emission detected by Osorio et al. (2014).
To derive the astrometry and photometry of the source, we inserted negative artificial planets in the individual exposures varying at the same time their brightness (with steps of 0.25 mag) and location (with steps of 0.25 pixel) and then re-ran PYNPOINT . To generate them we used a MonteCarlo photon generator with customizable FWHM. We adopted the FWHM measured from the unsaturated images of the photometric calibration dataset and we scaled the flux of the objects relative to the star, accounting for the difference in exposure time. We performed some simulations to evaluate the astrometric distortion induced by the vortex. At 1 λ/D , the offset is 0.075 FWHM, well below the speckle noise induced errors, and thus negligible. Each time, we used the final PYNPOINT image to calculate the deviation of the remaining flux at the object's location compared to the background noise in an annulus of 1 FWHM around the detection. We chose as brightness and astrometry of the source the combination of flux and position that yields the lowest deviation; i.e. the best subtraction.
To conclude, the source is located at 0'.156 ± 0'.032 from the central star at a position angle of PA =7.4 ◦ ± 11.3 . ◦ Our best estimate of the contrast for the object is ∆ L ′ =6.5 ± 0.5 mag. The errors on these measurements are the 1σ deviation quantities. These estimates are consistent with the expected per- formance of the AGPM at ∼ 0'.16 (see e.g. Mawet et al. 2013). We also inserted an artificial positive planet of the same brightness at the same separation but at a different position angle and we were able to recover it (see Figure 2). The artificial planet appears elongated, showing that the final shape of point-like sources at these small separations is affected by image processing. The observed magnitude for HD169142 is L ′ =5.66 ± 0.03 mag (van der Veen et al. 1989). Thus, we derived an apparent magnitude of L ′ =12.2 ± 0.5 mag for the newly detected source. The uncertainty is the square-root of the sum of squares of the errors on the stellar and the object's magnitudes.
An independent confirmation of this detection with the same instrument and at the same wavelength is provided by Biller et al., submitted. The astrometry and photometry of the detections are consistent within the errors.
## 3.2. Non-Detection in the J band
No source was detected in the J band images. We derived the 5σ detection limits by match filtering the reduced data using empirical PSF templates based on the astrometric spots. We then set as a conservative upper limit the maximum of the matched filtered PSF in various annuli around the star and corrected for self subtraction. Given the small separation of the L ′ band
## NACOIAGPM HD169142
Fig. 2.NACO/AGPM L ′ image of HD169142 with an artificial planet of the same brightness and angular separation as the detection. This image shows the final outcome of PYNPOINT with 20 PCA coefficients. Besides the bright source detected north of the central star, we could recover the artificial planet at PA /similarequal 270 ◦ . candidate, a trade between self-subtraction and speckle suppression was found, based on synthetic companions injected in the dataset at 6 azimuthal zones of radius 20 pixels and a radial exclusion criteria of ∼ 0.7 FWHM. According to our limits, the L ′ -band detection has J > 13.8 mag 11 .
## 3.3. Non-Detection of the L ′ Counterpart of a Millimeter Emission Source in the Annular Gap
H -band PDI images of HD169142 provided evidence for a low surface brightness annular gap in polarized light which extends from ∼ 40 to 70 AU (Quanz et al. 2013b). EVLA 7-mm observations revealed an unresolved source (0.10 mJy) in this gap (Osorio et al. 2014), which could be tracing circumplanetary dust material associated with a forming planet. The angular resolution of the 7-mm map is 0'.23 × 0'.16 and the signal-to-noise ratio of the compact emission is ∼ 5 σ above the background. Since no L ′ counterpart of the mm-emission appears in the L ′ -band images (see Figure 1), we can place an upper limit on its luminosity.
To estimate our sensitivity at the expected location of this emission, we inserted a positive artificial planet with increasing flux until we were able to detect it with PYNPOINT and the signal deviated more than 3 σ from the mean background value. According to this calculation, the compact source has L > ′ 14.0 ± 0.5 mag.
## 4. DISCUSSION
## 4.1. The Emission Source in the Inner Cavity
Given the object's angular separation, the distance to HD169142, and assuming the disk is seen face-on
11 We acknowledge that this magnitude limit is not yet representative of the expected GPI performances.
(Quanz et al. 2013b), the physical separation from the central star is 22.7 ± 4.7 AU. This suggests that it is located within the inner cavity (see Figure 1b), right inside the inner edge of the bright rim ( ∼ 25 AU, Quanz et al. 2013b). Different scenarios can explain the L ′ -band observations.
Although unlikely, the detected source could have an instrumental origin. Such a feature could be an AO tip/tilt residual, but it should rotate and be subtracted with the PSF of the star. In case of bad centering behind the AGPM, point-like features can also be generated. However, given the extensive set of tests we did and the data from Biller et al., submitted, we can be confident about the physical origin of the detection.
The bright source could be a background star. However, according to the Besancon galactic model (Robin et al. 2003), in a 1 deg 2 portion of the sky at the location of HD169142 we do not expect to have a single object with apparent magnitude L ′ ≤ 12.2 mag and J -L ′ ≥ 1 6 mag. .
Due to the limited precision in the astrometry for both the object and the nearby disk structures, we cannot fully exclude that the emission is originated in the disk rim. However, on the one hand, the bright rim is mostly axisymmetric in scattered light (see Figure 1b) and there is no maximum in the direction of the PYNPOINT detection. On the other hand, in case of thermal emission, the L ′ -band source must be hotter than ∼ 260 K and smaller than ∼ 1.9 AU, given that it is unresolved in the L ′ -band images. Such a locally concentrated and warm emission is best explained with a self-luminous compact source.
We can assume that the emission is coming from the photosphere of a companion in quasi-hydrostatic equilibrium undergoing Kelvin-Helmholtz contraction. Under this assumption, the L ′ luminosity suggests a mass of 3580 M Jupiter for an age of 3-12 Myrs (Grady et al. 2007), according to the COND models (Baraffe et al. 2003). In this case we should have detected the companion in J band, given our sensitivity at 0'.156 and the predicted J -band flux ( J =13.6-13.7 mag) for such object from the COND models. However, according to the same models, the J -band upper limit is still consistent within the uncertainties with the L ′ photometry and a 28-32 M Jupiter object at 3 Myrs. From the theoretical point of view, the presence of a 28-32 M Jupiter companion would be difficult to reconcile with the morphology of the innermost 30 AU. According to classical gap opening theories (e.g. Lin & Papaloizou 1993), we can estimate the width ∆ of the gap if we assume that a single body is carving out the cavity. Under several assumptions, such as the disk scale height at the object's location (2.8 AU from a disk model for HD169142 by Meeus et al. 2010), the geometric factor ( f ≈ 0.836, Lin & Papaloizou 1993), and the effective disk viscosity ( α ∼ 0.001), we find a value of ∆ =54-59 AU for object masses in the range 28-32 M Jupiter . A 10 M Jupiter planet would already be enough to explain the observed cavity size (∆ /similarequal 25 AU).
Thus, a final possibility could be the detection of a planet during its formation, as it has been proposed for HD100546b (Quanz et al. 2013a). As the star is still accreting, the object itself might still be gathering gaseous material from within the disk cavity. Such an accretion process would increase the observed luminosity due to the presence of a circumplanetary disk, allowing a much
lower mass for the object. A drastic increase in the planet luminosity during its formation is expected during the runaway gas accretion phase at a few million years (Marley et al. 2007; Mordasini et al. 2012). The J -band non-detection would also support the hypothesis of a cooler and/or more extincted object.
## 4.2. The Non-Detection in the Annular Gap
Concerning the non-detection in the L ′ -band images at the location of the 7-mm emission, our dataset allows us to put an upper limit on the mass of a possible object orbiting in the annular gap. At the distance of the star, our magnitude limit of L > ′ 14.0 ± 0.5 mag would correspond to < 11-18 M Jupiter at 3-12 Myrs, based on the COND models.
There is therefore a range of possible solutions for the planet-circumplanetary disk system hypothesis. If we assume that the 7-mm flux of 0.10 mJy arises from circumplanetary material emitting like a single-temperature black-body, the non-detection in the L ′ -band implies that it must be cooler than ∼ 240 K and that the emitting area is larger than ∼ 1.5 AU in radius to be consistent with the 7-mm flux. Taking this size as the minimum Hill radius at the expected location of the planet ( ∼ 50 AU), we get a lower limit for the planet mass of 0.1 M Jupiter .
In summary, our data are consistent with a planetary mass object with a mass between 0.1-18 M Jupiter surrounded by a circumplanetary disk cooler than ∼ 240 K and with a minimum radius of 1.5 AU. High-spatial resolution (sub-)mm data of the 7-mm source would allow us to test this scenario.
## 4.3. Possible multiple planet interaction and evolution
If the second object is also confirmed, it is interesting to speculate about the possibility of sequential planet formation and how it would affect the evolution of the disk. As suggested by Bryden et al. (2000), the accumulation of solid particles at the outer edge of a gap, that has been carved out by a protoplanet, could lead to the formation of an additional protoplanetary core at a larger orbital radius. Pierens & Nelson (2008) have shown that when two massive gap-opening planets are embedded in a disk, the gas in between the two gaps is cleared as the two gaps join together. The bright rim seen in the scattered light may be the region in between the gaps, before they merge. Such regions consist of gas surface density maxima (and hence pressure maxima) where dust can be trapped.
Together with HD100546, HD169142 might be a great laboratory to study multiple, and possibly sequential, planet formation empirically.
## 5. CONCLUSIONS
In this letter, we present L ′ and J -band observations of HD169142 with the VLT/NACO AGPM coronagraph and GPI, respectively. These images suggest the presence of a low-mass companion in the inner cavity of the transitional disk, at a separation of ∼ 23 AU. Whether this object is a BD or a forming planet still remains to be investigated. In any case, it is likely that this companion affected the disk morphology. Given the proper motion of HD169142, observations as early as mid-2015 will allow us to rule out the hypothesis of a background source. Upcoming instruments, such as VLT/SPHERE, will be crucial for proper motion confirmation and followup. Furthermore, given the expected orbital time of the object ( ∼ 86 years), it should move of ∼ 40 ◦ in 10 years.
Finally, our images do not exclude the possibility of a second object (0.1-18 M Jupiter ) forming in the annular gap (40-70 AU), as suggested by recent millimeter observations (Osorio et al. 2014). If future observations (e.g. mm and sub-mm data) confirm it, HD169142 would be forming a planetary system with at least two planets, and would boost our understanding of multiple and possibly sequential planet formation.
Based on observations obtained at the Gemini Observatory, operated by the Association of Universities for Research in Astronomy, Inc., under a cooperative agreement with the NSF on behalf of the Gemini partnership: NSF (United States), the National Research Council (Canada), CONICYT (Chile), the Australian Research Council (Australia), Minist´rio da Ciˆncia, Tecnologia e e e Inova¸ c˜o (Brazil) and Ministerio de Ciencia, Tecnolog´ ıa a e Innovaci´n Productiva (Argentina). o We thank Bruce Macintosh and the GPI team for support during the observations. GA, CC-G, MO, and JMT acknowledge support from MINECO (Spain) AYA2011-30228-C03 grant. Facilities: VLT:Yepun (NACO).
## REFERENCES
Absil, O., Milli, J., Mawet, D., et al. 2013, A&A, 559, L12 Amara, A., & Quanz, S. P. 2012, MNRAS, 427, 948
- Honda, M., Maaskant, K., Okamoto, Y. K., et al. 2012, ApJ, 752,
- Baraffe, I., Chabrier, G., Barman, T. S., Allard, F., & Hauschildt, P. H. 2003, A&A, 402, 701
- Blondel, P. F. C., & Djie, H. R. E. T. A. 2006, A&A, 456, 1045 Brittain, S. D., Najita, J. R., Carr, J. S., et al. 2013, ApJ, 767, 159 Bryden, G., R´˙ zyczka, M., Lin, D. N. C., & Bodenheimer, P. 2000, o ApJ, 540, 1091
- Cieza, L. A., Lacour, S., Schreiber, M. R., et al. 2013, ApJ, 762, L12
- Crepp, J. R., Pueyo, L., Brenner, D., et al. 2011, ApJ, 729, 132 Cutri, R. M., Skrutskie, M. F., van Dyk, S., et al. 2003, VizieR Online Data Catalog, 2246, 0
- Dent, W. R. F., Torrelles, J. M., Osorio, M., Calvet, N., & Anglada, G. 2006, MNRAS, 365, 1283
- Espaillat, C., Muzerolle, J., Najita, J., et al. 2014, arXiv:1402.7103 Grady, C. A., Schneider, G., Hamaguchi, K., et al. 2007, ApJ, 665, 1391
- Guimar˜ aes, M. M., Alencar, S. H. P., Corradi, W. J. B., & Vieira, S. L. A. 2006, A&A, 457, 581
- 143 Hu´ elamo, N., Lacour, S., Tuthill, P., et al. 2011, A&A, 528, L7 Kasper, M., Apai, D., Janson, M., & Brandner, W. 2007, A&A, 472, 321 Kraus, A. L., & Ireland, M. J. 2012, ApJ, 745, 5 Lafreni`re, D., Marois, C., Doyon, R., Nadeau, D., & Artigau, E. e ´ 2007, ApJ, 660, 770 Lin, D. N. C., & Papaloizou, J. C. B. 1993, Protostars and Planets III, 749 Macintosh, B., Graham, J. R., Ingraham, P., et al. 2014, arXiv:1403.7520 Maire, Perrin, Doyon, et al., Data Reduction Pipeline for the Gemini Planet Imager, Proc SPIE, 2010, 7735 Maire, Perrin, Doyon, et al., Test Results for the Gemini Planet Imager Data Pipeline, Proc. SPIE, 2012, 8451
- Mari˜as, N., Telesco, C. M., Fisher, R. S., & Packham, C. 2011, n ApJ, 737, 57
- Marley, M. S., Fortney, J. J., Hubickyj, O., Bodenheimer, P., & Lissauer, J. J. 2007, ApJ, 655, 541
- Marois, C., Lafreni`re, D., Doyon, R., Macintosh, B., & Nadeau, e
- D. 2006, ApJ, 641, 556 Mawet, D., Absil, O., Delacroix, C., et al. 2013, A&A, 552, L13 Meeus, G., Pinte, C., Woitke, P., et al. 2010, A&A, 518, L124
- Mordasini, C., Alibert, Y., Klahr, H., & Henning, T. 2012, A&A, 547, A111
- Osorio, M., Anglada, G., Carrasco-Gonz´lez, a C., et al. 2014, arXiv:1407.6549v1
- Pierens, A., & Nelson, R. P. 2008, A&A, 482, 333
- Pueyo, L., Crepp, J. R., Vasisht, G., et al. 2012, ApJS, 199, 6
- Quanz, S. P., Amara, A., Meyer, M. R., et al. 2013, ApJ, 766, L1
- Quanz, S. P., Avenhaus, H., Buenzli, E., et al. 2013, ApJ, 766, L2 Robin, A. C., Reyl´, C., Derri`re, S., & Picaud, S. 2003, A&A, 409, e e 523
- Soummer, R., Pueyo, L., & Larkin, J. 2012, ApJ, 755, L28
- Sylvester, R. J., Skinner, C. J., Barlow, M. J., & Mannings, V. 1996, MNRAS, 279, 915
- Thalmann, C., Mulders, G. D., Hodapp, K., et al. 2014, A&A, 566, A51
- van der Veen, W. E. C. J., Habing, H. J., & Geballe, T. R. 1989, A&A, 226, 108
|
10.1088/2041-8205/792/1/L23
|
[
"Maddalena Reggiani",
"Sascha P. Quanz",
"Michael R. Meyer",
"Laurent Pueyo",
"Olivier Absil",
"Adam Amara",
"Guillem Anglada",
"Henning Avenhaus",
"Julien H. Girard",
"Carlos Carrasco Gonzalez",
"Graham James",
"Dimitri Mawet",
"Farzana Meru",
"Julien Milli",
"Mayra Osorio",
"Schuyler Wolff",
"Jose-Maria Torrelles"
] |
2014-08-04T20:41:52+00:00
|
2014-08-04T20:41:52+00:00
|
[
"astro-ph.EP"
] |
Discovery of a Companion Candidate in the HD169142 Transition Disk and the Possibility of Multiple Planet Formation
|
We present L' and J-band high-contrast observations of HD169142, obtained
with the VLT/NACO AGPM vector vortex coronagraph and the Gemini Planet Imager,
respectively. A source located at 0".156+/-0".032 north of the host star
(PA=7.4+/-11.3 degrees) appears in the final reduced L' image. At the distance
of the star (~145 pc), this angular separation corresponds to a physical
separation of 22.7+/-4.7 AU, locating the source within the recently resolved
inner cavity of the transition disk. The source has a brightness of
L'=12.2+/-0.5 mag, whereas it is not detected in the J band (J>13.8 mag). If
its L' brightness arose solely from the photosphere of a companion and given
the J-L' color constraints, it would correspond to a 28-32 MJupiter object at
the age of the star, according to the COND models. Ongoing accretion activity
of the star suggests, however, that gas is left in the inner disk cavity from
which the companion could also be accreting. In this case the object could be
lower in mass and its luminosity enhanced by the accretion process and by a
circumplanetary disk. A lower mass object is more consistent with the observed
cavity width. Finally, the observations enable us to place an upper limit on
the L'-band flux of a second companion candidate orbiting in the disk annular
gap at ~50 AU, as suggested by millimeter observations. If the second companion
is also confirmed, HD169142 might be forming a planetary system, with at least
two companions opening gaps and possibly interacting with each other.
|
1408.0814v2
|
## Object Detection Through Exploration With A Foveated Visual Field
Emre Akbas, Miguel P . Eckstein
Abstract -We present a foveated object detector (FOD) as a biologically-inspired alternative to the sliding window (SW) approach which is the dominant method of search in computer vision object detection. Similar to the human visual system, the FOD has higher resolution at the fovea and lower resolution at the visual periphery. Consequently, more computational resources are allocated at the fovea and relatively fewer at the periphery. The FOD processes the entire scene, uses retino-specific object detection classifiers to guide eye movements, aligns its fovea with regions of interest in the input image and integrates observations across multiple fixations. Our approach combines modern object detectors from computer vision with a recent model of peripheral pooling regions found at the V1 layer of the human visual system. We assessed various eye movement strategies on the PASCAL VOC 2007 dataset and show that the FOD performs on par with the SW detector while bringing significant computational cost savings.
Index Terms -object detection, visual search, eye movements, latent linear discriminant analysis, foveated visual field
F
## 1 INTRODUCTION
There has been substantial progress (e.g. [7], [14], [22], [29], [37], [46], [53] to name a few) in object detection research in recent years. However, humans are still unsurpassed in their ability to search for objects in visual scenes. The human brain relies on a variety of strategies [9] including prior probabilities of object occurrence, global scene statistics [33], [45] and object co-occurrence [10], [28], [35] to successfully detect objects in cluttered scenes. Object detection approaches have increasingly included some of the human strategies [1], [2], [11], [14], [41]. One remaining crucial difference between the human visual system and a modern object detector is that while humans process the visual field with decreasing resolution away [26], [39], [43], [49] from the fixation point and make saccades to collect information, typical object detectors [14] scan all locations at the same resolution and repeats this at multiple scales. The goal of the present work is to investigate the impact on object detector performance of using a foveated visual field and saccade exploration rather than the dominant sliding window paradigm [14], [29], [53]. Such endeavor is of interest for two reasons. . First, from the computer vision perspective, using a visual field with varying resolution might lead to reduction in computational complexity, consequently the approach might lead to more efficient object detection algorithms. Second, from a scientific perspective, if a foveated object detection model can achieve similar performance accuracy as a non-foveated sliding window approach, it might suggest a possible reason for the evolution of foveated systems in organisms: achieving successful object detection while
- · E. Akbas and M. P. Eckstein are with the Vision and Image Understanding Laboratory at the Department of Psychological and Brain Sciences, University of California Santa Barbara.
minimizing computational and metabolic costs.
Contemporary object detection research can be roughly outlined by the following three important components of a modern object detector: the features, the detection model and the search model. The most popular choices for these three components are Histogram of Oriented Gradients (HOG) features [6], mixture of linear templates [14], and the sliding window (SW) method, respectively. Although there are efforts to go beyond these standard choices (e.g. new features [37], [46]; alternative detection models [22], [46], whether object parts should be modeled or not [8], [54]; and alternative search methods [13], [14], [20], [24], [46]), HOG, mixture of linear templates and SW form the crux of modern object detection methods ( [14], [29], [53]). Here, we build upon the 'HOG + mixture of linear templates' framework and propose a biologically inspired alternative search model to the sliding window method, where the detector searches for the object by making saccades instead of processing all locations at fine spatial resolution (See Section 4 for a more detailed discussion on related work).
The human visual system is known to have a varying resolution visual field. The fovea has higher resolution and this resolution decreases towards the periphery [26], [39], [43], [49]. As a consequence, the visual input at and around the fixation location has more detail relative to peripheral locations away from the fixation point. Humans and other mammals make saccades to align their high resolution fovea with the regions of interest in the visual environment. There are many possible methods to implement such a foveated visual field in an object detection system. In this work, we opt to use a recent model [15] which specifies how responses of elementary sensors are pooled at the layers (V1 and V2) of the human visual cortex. The model specifies the shapes and sizes of V1, V2 regions which pool responses from the visual field. We use a simplified version of this
Fig. 1. The foveated visual field of the proposed object detector. Square blue boxes with white borders at the center are foveal pooling regions. Around them are peripheral pooling regions which are radially elongated. The sizes of peripheral regions increase with distance to the fixation point which is at the center of the fovea. The color within the peripheral regions represent pooling weights.
Fig. 2. Two example detections by our foveated object detector (FOD). Yellow dots show fixation points, numbers in yellow fonts indicate the sequence of fixations and the bounding box is the final detection. Note that FOD does not have to fixate on the target object in order to localize it (example on the right).
model as the foveated visual field of our object detector (Figure 1). We call our detector as 'the foveated object detector (FOD)' due to its foveated visual field.
The sizes of pooling regions in the visual field increase as a function of eccentricity from the fixation location. As the pooling regions get larger towards the periphery, more information is lost at these locations, which might seem to be a disadvantage, however, the exploration of the scene with the high resolution fovea through a guided search algorithm might mitigate the apparent loss of peripheral information. On the other hand, fewer computational resources are allocated to process these low resolution areas which, in turn, lower the computational cost. In this paper, we investigate the impact of using a foveated visual field on the detection performance and its computational cost savings.
## 1.1 Overview of our approach
The foveated object detector (FOD) mimics the process by which humans search for objects in scenes utilizing eye movements to point the high resolution fovea to points of interest (Figure 2). The FOD gets assigned an initial fixation point on the input image and collects information by extracting image features through its foveated visual field. The features extracted around the fixation point are at fine spatial scale while features extracted away from the fixation location at coarser scale. This fine-to-coarse transition is dictated by the pooling region sizes of the visual field. Then, based on the information collected, the FOD chooses the next fixation point and makes a saccade to that point. Finally, the FOD integrates information collected through multiple saccades and outputs object detection predictions.
Training such an object detector entails learning templates at all locations in the visual field. Because the visual field has varying resolution, the appearance of a target object varies depending on where it is located within the visual field. We use the HOG [6] as image features and a simplified version of the V1 model [15] to compute pooled features within the visual field. A mixture of linear templates is trained at selected locations in the visual field using a latent-SVM-like [14], [18] framework.
## 1.2 Contribution
We present an object detector that has a foveated visual field based on physiological measurements in primate visual cortex [15] and that models the appearance of target objects not only in the high resolution fovea but also in the periphery. . Importantly, the model is developed in the context of a modern object detection algorithm and a standard data-set (PASCAL VOC) allowing for the first time direct evaluation of the impact of a foveated visual system on an object detector.
We believe that object detection using a foveated visual field offers a novel and promising direction of research in the quest for an efficient alternative to the sliding window method, and also a possible explanation for why foveated visual systems might have evolved in organisms. We show that our method achieves greater computational savings than a state-of-the-art cascaded detection method. Another contribution of our work is the latent-LDA formulation (Section 2.4.2) where linear discriminant analysis is used within a latent-variable learning framework.
In the next section, we describe the FOD in detail and report experimental results in Section 3 which is followed by the related work section, conclusions and discussion.
## THE FOVEATED OBJECT DETECTOR (FOD)
## 2 2.1 Foveated visual field
The Freeman-Simoncelli (FS) model [15] is neuronal population model of V1 and V2 layers of the visual cortex.
The model specifies how responses are pooled (averaged together) hierarchically beginning from the lateral geniculate nucleus to V1 and then the V2 layer. V1 cells encode information about local orientation and spatial frequency whereas the cells in V2 pools V1 responses non-linearly to achieve selectivity for compound features such as corners and junctions. The model is based on findings and physiological measurements of the primate visual cortex and specifies the shapes and sizes of the receptive fields of the cells in V1 and V2. According to the model, the sizes of receptive fields increase linearly as a function of the distance from the fovea and this rate of increase in V2 is larger than that of V1, which means V2 pools larger areas of the visual field in the periphery. The reader is referred to [15] for further details.
We simplify the FS model in two ways. First, the model uses a Gabor filter bank to compute image features and we replace these with the HOG features [6], [14]. Second, we only use the V1 layer and leave the nonlinear pooling at V2 as future work. We use this simplified FS model as the foveated visual field of our object detector which is shown in Figure 1. The fovea subtends a radius of 2 degrees. We also only simulate a visual field with a radius of 10 degrees which is sufficient to cover the test images presented at a typical viewing distance of 40 cm. The square boxes with white borders (Figure 1 represent the pooling regions within the fovea. The surrounding colored regions are the peripheral pooling regions. While the foveal regions have equal sizes, the peripheral regions grow in size as a function - which is specified by the FS model - of their distance to the center of the fovea. The color represents the weights that are used in pooling, i.e. weighted summation of, the underlying responses. A pooling region partly overlaps with its neighboring pooling regions (see the supplementary material of [15] for details). Assuming a viewing distance of 40 cm, the whole visual field covers about a 500 x 500 pixel area (a pixel subtends 0 08 . ◦ ). The foveal radius is 52 pixels subtending a visual angle of 4 degrees.
Given an image and a fixation point, we first compute the gradient at each pixel and then for each pooling region, the gradient magnitudes are pooled per orientation for the pixels that fall under the region. At the fovea, where the pooling regions are 8 x 8 pixels, we use the HOG features at the same spatial scale of the original DPM model [14], and in the periphery, each pooling region takes a weighted sum of HOG features of the 8 x 8 regions that are covered by that pooling region.
## 2.2 The model
The model M consists of a mixture of n components
$$M = \{ ( \mathbf w _ { i }, \ell _ { i } ) \, \colon i = 1, 2, \dots, n \} \quad \quad ( 1 ) \ \text{ be at}$$
where w i is a linear template and glyph[lscript] i is the location of the template with respect to the center of the visual field. The location variable glyph[lscript] i defines a unique bounding box within the visual field for the i th template. Specifically, glyph[lscript] i = ( ω , h , x , y i i i i ) is a 4 -tuple whose variables respectively denote width, height and x y , coordinates of the i th template within the visual field. The template, w i , is a matrix of weights on the features extracted from the pooling regions underlying the bounding box glyph[lscript] i . The dimensionality of w i , i.e. the total number of weights, depends both on the width and height of its bounding box and its location in the visual field. A component within the fovea covers a larger number of pooling regions compared to a peripheral component with the same width and height, hence the dimensionality of a foveal template is larger. Three example components are illustrated in Figure 3 where the foveal component (red) covers 7 x 5 = 35 pooling regions while the (blue and green) peripheral components cover 15 and 2 regions, respectively. Since a fixed number of features 1 is extracted from each pooling region (regardless of its size), foveal components have higher-resolution templates associated with them.
## 2.2.1 Detection model
Suppose that we are given a model M that is already trained for a certain object class. The model is presented with an image I and assigned an initial fixation location f . We are interested in searching for an object instance in I . Because the size of a searched object is not known apriori, the model has to analyze the input image at various scales. We use the same set of image scales given in [14] and use σ to denote a scale from that set. When used as a subscript to an image, e.g. I σ , it denotes the scaled version of that image, i.e. width (and height) of I σ is σ times the width (and height) of I . σ also applies to fixation locations and bounding boxes: if f denotes a fixation location ( f x , f y ) , then f σ = ( σ f x , σ f y ) ; for a bounding box b = ( w,h,x,y ) , b σ = ( σw,σh,σx,σy ) .
To check whether an arbitrary bounding box b within I contains an object instance, while the model is fixating at location f, we compute a detection score as
$$\text{s} ( I, b, f ) = \max _ { \sigma } \max _ { c \in G ( b _ { \sigma }, f _ { \sigma } ) } \text{w} ^ { T } \Psi ( I _ { \sigma }, f _ { \sigma }, c ) \quad ( 2 )$$
where Ψ( I σ , f σ , c ) is a feature extraction function which returns the features of I σ for component c (see Equation (1)) when the model is fixating at f σ . The vector w is the blockwise concatenation of the templates of all components. Ψ( ) · effectively chooses which component to use, that is w T Ψ( I σ , f σ , c ) = w T c Ψ( I σ , f σ , c ) . The fixation location , f σ , together with the component c define a unique location, i.e. a bounding box, on I σ . G ( b σ , f σ ) returns the set of all components whose templates have a predetermined overlap (intersection over union should be at least 0 7 . as in [14]) with b σ when the model is fixating at f σ . During both training and testing, σ and c are latent variables for example ( I, b ) .
1. We use the feature extraction implementation of DPM (rel5) [14], [17], which extracts a 31 -dimensional feature vector.
Fig. 3. Illustration of the visual field of the model. (a) The model is fixating at the red cross mark on the image. (b) Visual field (Figure 1) overlaid on the image, centered at the fixation location. White line delineate the borders of pooling regions. Nearby pooling regions do overlap. The weights (Figure 1) of a pooling region sharply decrease outside of its shown borders. White borders are actually iso-weight contours for neighboring regions. Colored bounding boxes show the templates of three components on the visual field: red, a template within the fovea; blue and green, two peripheral templates at 2.8 and 7 degree periphery, respectively. (c,d,e) Zoomed in versions of the red (foveal), blue (peripheral) and green (peripheral) templates. The weights of a template, w i , are defined on the gray shaded pooling regions.
Ideally, s I, ( b f , ) > 0 should hold for an appropriate f when I contains an object instance within b . For an image that does not contain an object instance, s I, ( b = ∅ , f ) < 0 should hold for any f . For this to work, a subtlety in G ( ) · 's definition is needed: G ( ∅ , f ) returns all components of the model (Equation (1)). During training (Section 2.4), this will enforce the responses of all components for a negative image to be suppressed down.
## 2.2.2 Integrating observations across multiple fixations
So far, we have looked at the situation where the model has made only one fixation. We describe in Section 2.3 how the model chooses the next fixation location. For now, suppose that the model has made m fixations, f 1 , f 2 , . . . , f m , and we want to find out whether an arbitrary bounding box b contains an object instance. This computation involves integrating observations across multiple fixations, which is a considerably more complicated problem than the single fixation case. The Bayesian decision on whether b contains an object instance is based on the comparison of posterior probabilities:
$$\frac { P ( y _ { b } = 1 | f _ { 1 }, f _ { 2 }, \dots, f _ { m }, I ) } { P ( y _ { b } = 0 | f _ { 1 }, f _ { 2 }, \dots, f _ { m }, I ) } \lesssim 1 \quad \quad ( 3 ) \quad 2. 3$$
where y b = 1 denotes the event that there is an object instance at location b . We use the posteriors' ratio as a detection score, the higher it is the more likely b contains an instance. Computing the probabilities in (3) requires training a classifier per combination of fixation locations for each different value of m , which is intractable. We approximate it using a conditional independence assumption (derivation given in Appendix A):
$$\frac { P ( y _ { b } = 1 | f _ { 1 }, f _ { 2 }, \dots, f _ { m }, I ) } { P ( y _ { b } = 0 | f _ { 1 }, f _ { 2 }, \dots, f _ { m }, I ) } \approx \prod _ { i = 1 } ^ { m } \frac { P ( y _ { b } = 1 | f _ { i }, I ) } { P ( y _ { b } = 0 | f _ { i }, I ) }. \ \ ( 4 ) \quad \text{in real} \quad \text{the high}$$
We model the probability P y ( b = 1 | f , I ) using a classifier and use the sigmoid transfer function to convert raw classification scores to probabilities:
$$P ( y _ { b } = 1 | f, I ) = \frac { 1 } { 1 + e ^ { ( - s ( I, b, f ) ) } }. \quad \quad ( 5 )$$
We simplify the computation in (4) by taking the log (derivation given in Appendix B):
$$\log \left ( \prod _ { i = 1 } ^ { m } \frac { P ( y _ { b } = 1 | f _ { i }, I ) } { P ( y _ { b } = 0 | f _ { i }, I ) } \right ) = \sum _ { i = 1 } ^ { m } s ( I, b, f _ { i } ). \quad ( 6 )$$
Taking the logarithm of posterior ratios does not alter the ranking of detection scores for different locations, i.e. b 's, because logarithm is a monotonic function. In short, the detection score computed by the FOD for a certain location b , is the sum of the individual scores for b computed at each fixation.
After evaluating (6) for a set of candidate locations, final bounding box predictions are obtained by nonmaxima suppression [14], i.e. given multiple predictions for a certain location, all predictions except the one with the maximal score are discarded.
## 2.3 Eye movement strategy
We use the maximum-a-posteriori (MAP) model [4] as the basic eye movement strategy of the FOD. The MAP model is shown to be consistent with human eye movements in a variety of visual search tasks [4], [47]. Studies have demonstrated that in some circumstances human saccade statistics better match an ideal searcher [31] that makes eye movements to locations that maximize the accuracy of localizing targets, yet in many circumstances the MAP model approximates the ideal searcher [32], [51] but is computationally more tractable for objects in real scenes. The MAP model select the location with the highest posterior probability of containing the target
Fig. 4. Two bounding boxes (A,B) are shown on the visual field. While box A covers a large portion of the pooling regions that it intersects with, box B's coverage is not as good. Box B is discarded as it does not meet the overlap criteria (see text), therefore a component for B in the model is not created.
object as the next fixation location, that is f i +1 = center of glyph[lscript] ∗ where
$$\ell ^ { * } = \arg \max _ { \ell } P ( y _ { \ell } = 1 | f _ { 1 }, f _ { 2 }, \dots, f _ { i }, I ). \quad ( 7 ) \begin{array} { c } \text{Opt} \\ \text{mixtu} \\ \text{1} \dots \dots \end{array}.$$
Finding the maximum of the posterior above is equivalent to finding the maximum of the posterior ratios,
$$\arg \max _ { \ell } P ( y _ { \ell } = 1 | f _ { 1 }, \dots, f _ { i }, I ) = \arg \max _ { \ell } \frac { P ( y _ { \ell } = 1 | f _ { 1 }, \dots, f _ { i }, I ) } { P ( y _ { \ell } = 0 | f _ { 1 }, \dots, f _ { i }, I ) } \frac { \text{As an all} } { \text{Recently},} \\ \cdot \quad \epsilon \quad \cdot \quad \cdot \quad \cdot \quad \epsilon \quad \cdot \quad \epsilon / \quad \text{used in } \epsilon \\ \dots \quad \epsilon \quad \cdot \quad \cdot \quad \cdot \quad \epsilon / \quad \text{used in } \epsilon$$
since for two arbitrary locations glyph[lscript] 1 , glyph[lscript] 2 ; let p 1 = P y ( glyph[lscript] 1 = 1 |· ) and p 2 = P y ( glyph[lscript] 2 = 1 |· ) , then we have
$$\frac { p _ { 1 } } { 1 - p _ { 1 } } > \frac { p _ { 2 } } { 1 - p _ { 2 } } \implies p _ { 1 } > p _ { 2 }. \quad \ \ ( 9 ) \, \begin{array} { c } \, \stackrel { \mu _ { 1 } \, \infty } { \text{exam} } \\ \text{is the} \end{array} \,.$$
## 2.4 Training the model
## 2.4.1 Initialization
A set of dimensions (width and height) is determined from the bounding box statistics of the examples in the training set as done in the initialization of the DPM model [14]. Then, for each width and height, new components with these dimensions are created to tile the entire visual field. However, the density of components in the visual field is not uniform. Locations, i.e. bounding boxes, that do not overlap well with the underlying pooling regions are discarded. To define goodness of overlap, a bounding box is said to intersect with an underlying pooling region if more than one fifth of that region is covered by the bounding box. Overlap is the average coverage across the intersected regions. If the overlap is more than 75% , then a component for that location is created, otherwise the location is discarded (see Figure 4 for an example). In addition, no components are created for locations that are outside of the visual field. Weights of the component templates ( w i ) are initialized to arbitrary values. Training the model is essentially optimizing these weights on a given dataset.
## 2.4.2 Training
Consider a training set D = { ( I , i b i ) } K i =1 where I i is an image and b i a bounding box and K is the total number of examples. If I i does not contain any positive examples, i.e. object instances, then b i = ∅ . Following the DPM model [14], we train model templates using a latent-SVM formulation:
$$\text{on the} \quad \text{(or the} \quad \text{arg} \min \frac { 1 } { 2 } | | w | | _ { 2 } ^ { 2 } + C \sum _ { i = 1 } ^ { K } \sum _ { f \in F ( I _ { i }, b _ { i } ) } \max ( 0, 1 - y _ { i } s ( I _ { i }, b _ { i }, f ) ). \\ \text{for $B$ in} \quad \text{(who $w$ - 1$ i f $b$ -\theta$ and $w$ - -1$ otherwise $The cost}$$
glyph[negationslash]
glyph[negationslash]
where y i = 1 if b i = ∅ and y i = -1 , otherwise. The set F I , ( i b i ) denotes the set of all feasible fixation locations for example ( I , i b i ) . For b i = ∅ , a fixation location is considered feasible if there exists a model component whose bounding box overlaps with b i . For b i = ∅ , all possible fixation locations on I i are considered feasible.
Optimizing the cost function in (10) is manageable for mixtures with few components, however, the FOD has a large number of components in its visual field (typically, for an object class in the PASCAL VOC 2007 dataset [12], there are around 500 700 -) and optimizing this cost function becomes prohibitive in terms of computational cost. As an alternative, cheaper linear classifiers can be used. Recently, linear discriminant analysis (LDA) has been used in object detection ( [18]) producing surprisingly good results with much faster training time. Training a LDAclassifier amounts to computing Σ -1 ( µ 1 -µ 0 ) where µ 1 is the mean of the feature vectors of the positive examples, µ 0 is the same for the negative examples and Σ is the covariance matrix of these features. Here, the most expensive computation is the estimation of Σ , which is required for each template with different dimensions. However, it is possible to estimate a global Σ from which covariance matrices for templates of different dimensions can be obtained [18]. For the FOD, we estimate the covariance matrices for the foveal templates and estimate the covariance matrices for peripheral templates by applying the feature pooling transformations to the foveal covariance matrices.
We propose to use LDA in a latent-SVM-like framework as an alternative to the method in [18] where positive examples are clustered first and then a LDA classifier is trained per cluster. Consider the t th template, w t . LDA gives us that LDA gives us that w t, LDA = Σ -1 t ( µ pos t -µ neg t ) where Σ t is the covariance matrix for template t , µ pos t and µ neg t are the mean of positive and negative feature vectors, respectively, assigned to template t . We propose to apply an affine transformation to the LDA classifier:
$$\inf _ { \substack { \mathbf v _ { i } \\ \mathbf v _ { i } } ) } \quad w _ { t } = \begin{bmatrix} \alpha _ { t } & 0 \\ \alpha _ { t } & 0 \\ \ddots & \\ 0 & \alpha _ { t } & \beta _ { t } \end{bmatrix} \begin{bmatrix} w _ { t, L D A } \\ 1 \end{bmatrix} = \begin{bmatrix} \alpha _ { t } w _ { t, L D A } \\ \beta _ { t } \end{bmatrix} \quad ( 1 1 )$$
and modify the cost function as glyph[negationslash]
$$\arg \min _ { \alpha, \beta } \left ( \frac { 1 } { 2 } | | \mathbf w | | _ { 2 } ^ { 2 } + C \sum _ { t = 1 } ^ { N } \max ( 0, 1 + \mathbf w _ { t } ^ { T } \mu _ { t } ^ { n e g } ) + \text{conex} } \text{ora} \\ C \sum _ { i \in \{ i | \mathbf b _ { i } \neq \emptyset \} } \sum _ { f \in F ( I _ { i }, \mathbf b _ { i } ) } \max ( 0, 1 - y _ { i } s ( I _ { i }, \mathbf b _ { i }, f ) ) \right ) \text{ \quad \ } \text{doing} \, \varepsilon \\ ( \text{12} ) \text{ \ } \text{train} \, \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \text{$\ } \end{cases}$$
where the first summation pushes the score of the mean of the negative examples to under zero and the second summation, taken over positive examples only, pushes the scores to above 0. α and β are appropriate blockwise concatenation of α t and β t s. C is the regularization constant. Overall, this optimization effectively calibrates the dynamic ranges of different templates' responses in the model so that the scores of positive examples and negative means are pushed away from each other while the norm of w is constraint to prevent overfitting. This formulation does not require the costly mining of hard-negative examples of latent-SVM. We call this formulation (Equation (12)) as latent-LDA.
To optimize (12), we use the classical coordinatedescent procedure. We start by initializing w by training on warped-positive examples as in [14]. Then, we alternate between choosing the best values for the latent variables while keeping w fixed, and optimizing for w while keeping the latent variables of positive examples fixed.
## 3 EXPERIMENTS
We evaluated our method on the PASCAL VOC 2007 detection ( comp3 ) challenge dataset and protocol (see [12] for details). All results are obtained by training on the train+val split and testing on the test split.
## 3.1 Comparison of SW based methods
We first compared our SW implementation, which corresponds to using foveal templates only, to three stateof-the-art methods that are also SW based [14], [18], [29]. Table 1 gives the AP (average precision) results, i.e. area under the precision-recall curve per class, and mean AP (mAP) over all classes. Originally, the deformable parts model (DPM) uses object parts, however, in order to make a fair comparison with our model, we disabled its parts. The first row of Table 1 shows the latest version of the DPM system [17] with the parts-learning code disabled. The second row shows results for another popular SVM-based system, known as the exemplarSVM (E-SVM), which also only models whole objects, not its parts. Finally, the third row shows results from a LDA-based system, 'discriminative decorrelation for classification' (DCC) [18]. All three systems are based on HOG features and mixture of linear templates. The results show that SVM based systems perform better than the LDA based systems, which is not a surprising finding since it is well known that discriminative models outperform generative models in classification tasks. However,
LDA's advantage against this performance loss is that it is ultra fast to train, which is exactly the reason we chose to use LDA instead of SVM. Once the background covariance matrices are estimated (which can be done once and for all [18]), training is as easy as taking the average of the feature vectors of positive examples and doing a matrix multiplication. We estimated the time that training a SVM based system for our FOD to be about 300 hours (approximately 2 weeks) for a single object class, whereas the LDA based system can be trained under an hour on the same machine which has an Intel i7 processor.
Although our SW method achieves the same mean AP (mAP) score as the DCC method [18], the latter has a detection model with higher computational cost. We use 2 templates per class while DCC trains more than 15 templates per class within an exemplar-SVM [29]-like framework. DCC considers the dot product of the feature vector of the detection window with every exemplar within a cluster, which basically means that a detection window is compared to all positive examples in the training set. In our case, the number of dot products considered per detection window is equal to the number of templates, which is 2 in this paper, which clearly demonstrates the advantage of our latent-LDA approach over DCC [18].
## 3.2 Comparison of FOD with SW
Next, we compared the performance of FOD with our SW method. We experimented with two eye movement strategies, MAP (Section 2.3) and random strategy to demonstrate the importance of guidance of eye movements.
Table 2 shows the AP scores for FOD with different eye movement strategies and different number of fixations. We also include in this table the 'Our SW' result from Table 1 for ease of reference. The MAP and random strategies are denoted with MAP and RAND, respectively. Because the model accuracy results will depend on initial point of fixation, we ran the models with different initial points of fixation. The presence of a suffix on a model refers to the location of the initial fixation: '-C' stands for the center of the input image, i.e. (0 5 . , 0 5) . in normalized image coordinates where the topleft corner is taken as (0 0) , and the bottom-right corner is (1 , 1) ; and '-E' for the two locations at the left and right edges of the image, 10% of the image width away from the image border, that is (0 1 . , 0 5) . and (0 9 . , 0 5) . . MAP-E and RAND-E results are the performance average of two different runs, one with initial fixation close to the left edge of the image, the other run close to the right edge of the image. For the random eye movement, we report the 95% confidence interval for AP over 10 different runs. We ran all systems for a total of 5 fixations. Table 2 shows results for after 1 3 , and 5 fixations. A condition with one fixation is a model that makes decisions based only on the initial fixation.
TABLE 1
Average precision (AP) scores of SW based methods on the PASCAL VOC 2007 dataset.
TABLE 2
| DPM [14] | aero bike 23.6 48.6 | bird 9.7 | boat 11.0 | bottle 19.3 | bus car 40.4 45.2 | cat chair 12.4 15.4 | 19.4 | cow | table dog 17.4 | horse 4.0 | mbike 44.7 36.4 | person 31.2 | plant 10.9 | sheep 14.1 | sofa 19.5 | train 32.2 | tv 37.0 | mAP 24.6 |
|------------|-----------------------|------------|-------------|---------------|---------------------|-----------------------|-----------|-------|------------------|-------------|-------------------|---------------|--------------|--------------|-------------|--------------|-----------|------------|
| E-SVM [29] | 20.4 40.7 | 9.3 | 10 | 10.3 | 31.0 40.1 | 9.6 | 10.4 | 14.7 | 2.3 9.7 | 38.4 | 32.0 | 19.2 | 9.6 | 16.7 | 11 | 29.1 | 31.5 | 19.8 |
| DCC [18] | 17.4 35.5 | 9.7 | 10.9 | 15.4 | 17.2 40.3 | 10.6 | 10.3 | 14.3 | 4.1 1.8 | 39.7 | 26.0 | 23.1 | 4.9 | 14.1 | 8.7 | 22.1 | 15.2 | 17.1 |
| Our SW | 17.5 28.6 | 9.7 | 10.4 | 17.3 | 29.8 36.7 | 7.9 | 11.2 21.0 | | 2.3 | 2.7 | 30.9 21.1 | 19.7 | 3 | 9.2 | 13.7 | 23.5 | 25.2 | 17.1 |
AP scores and relative computational costs of SW and FOD on the PASCAL VOC 2007 dataset.
Fig. 5. Ratio of mean AP scores of FOD systems relative to that of the SW system. Graph shows two eye movement algorithms: maximum aposteriori probability (MAP) and random (RAND) and two starting points (C: center; E: edge).
| Our SW | 17.5 | 28.6 | 9.7 | 10.4 | 17.3 | 29.8 36.7 | 7.9 | 11.2 | 21.0 | 2.3 | 2.7 | 30.9 | 21.1 | 19.7 | 3.0 | 9.2 | 13.7 | 23.5 | 25.2 17.1 | 100 |
|----------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|-----------------------------------------|
| | 1 17.0 | 21.1 | 4.9 | 9.8 | 9.3 | 27.4 27.9 | | 8.5 | 3.7 | 12.8 2.0 | 4.3 | 29.7 | 19.7 | 18.2 | 1.2 | 10.7 | 14.0 | 26.2 21.8 | 14.5 | 11.5 |
| MAP-C | 3 17.4 | 27.7 | 10.1 | 10.6 | 10.4 | 30.8 31.6 | 8.4 | | 10.4 | 17.2 2.1 | 3.4 | 33.3 | 21.1 | 18.7 | 3.4 | | 7.6 | 26.4 | 23.5 16.5 | 31.2 |
| MAP-C | 5 17.0 | 28.6 | 10.0 | 10.7 | 11.2 | 31.0 34.0 | 8.3 | 10.6 | 18.2 | 2.1 | 3.4 | 34.2 | 21.8 | 19.7 | 2.8 | 8.1 | 15.4 15.1 | 27.8 | 24.0 16.9 | 49.6 |
| MAP-E | 1 1.6 | 7.1 | 4.1 | 5.6 | 9.1 | 8.7 11.7 | 6.0 | 3.6 | 10.2 | 2.0 | 2.2 | 8.5 | 10.2 | 13.5 | 1.3 | 6.8 | 8.0 | 10.6 | 10.3 7.1 | 8.7 |
| | 3 13.0 | 24.6 | 9.9 | 9.8 | 10.7 | 27.2 29.3 | 7.4 | 10.4 | | 16.4 3.7 | 2.2 | 30.6 | 20.8 | 16.9 | 3.3 | 11.2 | 13.8 | 23.0 | 24.1 15.4 | 28.1 |
| | 5 15.1 | 28.0 | 9.9 | 10.4 | 11.6 | 29.9 33.0 | 8.3 | 10.6 | 18.7 | 2.7 | 4.1 | 33.7 | 22.6 | 18.9 | 3.1 | 7.1 | 14.7 | 25.5 | 25.2 16.7 | 46.9 |
| RAND | 1 8.2 | 9.3 | 5.5 | 9.3 | 7.8 | 12.2 16.2 | 6.1 | 6.8 | 7.5 | 1.6 | 2.5 | 10.6 | 9.1 | 9.9 | 1.9 | 5.0 | 6.7 | 11.2 | 10.0 7.9 ± 1.4 | |
| | 3 9.6 | 13.0 | 3.2 | 9.6 | 9.3 | 16.9 23.5 | 8.8 | 9.4 | 9.9 | 1.8 | 3.2 | 16.5 | 12.3 | 12.2 | 2.7 | 3.9 | 9.3 | 16.9 | 11.7 10.2 ± 0.9 | similar |
| | 5 10.9 | 15.3 | 3.8 | 9.7 | 9.6 | 20.5 26.3 | 9.3 | 9.5 | 10.6 | 1.5 | 3.1 | 20.9 | 13.7 | 13.5 | 2.7 | 3.9 | 12.0 | 18.9 | 12.4 11.4 ± 1.0 | to above |
| | | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, | 1 This row is the same with the 'MAP-C, |
| RAND-C | 3 17.5 | 20.4 | 3.7 | 10.0 | 9.3 | 28.6 27.4 | 11.5 | 6.7 | 11.8 | 1.7 | 3.5 | 31.7 | 1' 18.0 | above. 15.4 | 2.7 | 5.4 | 15.2 | 26.1 | 15.8 14.1 ± 0.5 | ' |
| | 5 17.6 | 21.4 | 5.2 | 9.9 | 9.7 | 28.1 28.6 | 11.4 | 9.6 | 12.1 | 1.6 | 3.5 | 30.0 | 17.9 | 15.3 | 3.7 | 6.7 | 14.4 | 25.4 | 15.9 14.4 ± 0.7 | |
| | 1 This row is the same with the 'MAP-E, | | | | | | | | | | | | | | | | | | | |
| RAND-E | 3 9.1 | 13.1 | 2.8 | 9.7 | 9.4 | 17.8 22.5 | 9.0 | 6.6 | 10.7 | 2.3 | 3.7 | 14.9 | 1' 12.0 | above. 14.9 | 1.3 | 3.9 | 2.4 | 13.6 | 14.1 9.7 ± 0.7 | ' |
| | 5 10.7 | 15.9 | 4.1 | 8.7 | 9.5 | 21.9 26.0 | 8.2 | 9.7 | 11.6 | 1.7 | 4.3 | 17.6 | 13.7 | 14.1 | 1.9 | 5.7 | 4.8 | 15.7 | 15.8 11.1 ± 1.1 | |
The results show that the FOD using the MAP rule with 5 fixations (MAP-C,5 for short) performs nearly as good as the SW (a difference of 0 2 . in mean AP).
Figure 5 shows the ratio of mean AP for the FOD with the various eye movement strategies to that of the SW system (relative performance) as a function of fixation. The relative performance of the MAP-C to SW (AP of MAP-C divided by AP of SW) is 98 8% . for 5 fixations, 96 5% . for 3 fixations and 84 8% . for 1 fixation. The FOD with eye movement guidance towards the target (MAPC,5) achieves or exceeds SW's performance with only
Fig. 6. AP scores achieved by SW and MAP-E per class.
1 fixation in 4 classes, with 3 fixations in 7 classes, with 5 fixations in 2 classes. For the remaining of 7 classes, FOD needs more than 5 fixations to achieve SW's performance.
MAP-C performs quite well ( 84 8% . relative performance) even with 1 fixation. The reason behind this result is the fact that, on average, bounding boxes in the PASCAL dataset cover a large portion of the images (average bounding box area normalized by image area is 0 2 . ) and are located at and around the center [44]. To reduce the effects of these biases about the location of object placement on the results, we assessed the models with an initial fixation close to the edge of the image (MAP-E). When the initial fixation is closer to the edge of the image, performance is initially worse than when the initial fixation is at the center of the image, The difference in performance diminishes achieving similar performance with five fixations ( 0 2 . difference in mean
AP). Figure 6 shows how the distribution of AP scores for different object classes for MAP-E improves from 1 fixation to 5 fixations
## 3.2.1 Importance of the guidance algorithm
To assess the importance of guided saccades towards the target we compared performance of the MAP model against FOD that guides eye movements based on a random eye movement generator.
Figure 5 allows comparisons of the relative performance of the MAP FOD and those with a random eye movement strategy. The performance gap between MAPC, RAND-C pair and MAP-E,RAND-E pair shows that MAP eye movement strategy is effective in improving the performance of the system.
## 3.3 Computational cost
The computational complexity of the SW method is easily expressed in terms of image size. However, this is not the case for our model. The computational complexity of FOD is O mn ( ) where m is the number of fixations and n is the total number of components, hence templates, on the visual field. These numbers do not explicitly depend on the image size; so in this sense, the complexity of FOD is O (1) in terms of image size. Currently, m is given as an input parameter but if it were to be automated, e.g. to achieve a certain detection accuracy, m would implicitly depend on several factors such as the difficulty of the object class, the location and size distribution of positive examples. Targets that are small (relative to the image size) and that are located far away from the initial fixation location would require more fixations to get a certain detection accuracy. The number of components, n , depends on both the visual field parameters (number of angle and eccentricity bins which, in our case, are fixed based on the Freeman-Simoncelli model [15]) and the bounding box statistics of the target object. These dependencies make it difficult to express the theoretical complexity in terms of input image size. For this reason, we compare the computational costs of FOD and SW in a practical framework, expressed in terms of the total number of operations performed in template evaluations.
In both SW based methods and the FOD, linear template evaluations, i.e. taking dot-products, is the main time consuming operation. We define the computational cost of a method based on the total number of template evaluations it executes (as also done in [46]). A model may have several templates with different sizes, so instead of counting each template evaluation as 1 operation, we take into account the dimensionalities of the templates. For example, the cost of evaluating a (6-region)x(8-region) HOG template is counted as 48 operations.
It is straightforward to compute the computational cost (as defined above) of the SW method. For the FOD, we run the model on a subset of the testing set and count
Fig. 7. FOD-DPM's performance (mean AP over 20 classes) as a function of number of fixations. FOD-DPM achieves DPM's performance at 11 fixations and exceeds it with more fixations.
the number of operations actually performed. Note that, in order to compute a detection score, the FOD first performs a feature pooling (based on the location of the component in the visual field) and then a linear template evaluation. Since these are both linear operations, we combine them into a single linear template. The last column of Table 2 gives the computational costs of the SW method and the FOD. For the FOD the computational cost is reported as a function of different number of fixations. For ease of comparison, we normalized the costs so that the SW method performs 100 operations in total. The results show that FOD is computationally more efficient than SW. FOD achieves 98 8% . of SW's performance at 49 6% . of the computational cost of SW. Note that this saving is not directly comparable to that of the cascaded detection method reported in [13] because FOD's computational savings comes about from fewer root filter evaluations, whereas in [13] a richer model (DPM, root filters and part filters) is used and the savings are associated to fewer evaluations in the part filters (i.e., the model applies the root filters at all locations first and sequentially running other filters on the non-rejected locations).
## 3.4 Using richer models to increase performance
To directly compare the computational savings of the FOD model to a cascade-type object detector, we used a richer and more expensive detection model at the fovea. This is analogous to the cascaded detection idea where cheaper detectors are applied first and more expensive detectors are applied later on the locations not rejected by the cheaper detectors. To this end, we run our FOD and after each fixation we evaluate the full DPM detector (root and part filters together) [17] only at foveal locations that score above a threshold which is determined on the training set to achieve high recall rate ( 95% ). We call this approach 'FOD-DPM cascade' or FOD-DPM for short. Table 3 and Figure 7 give the
TABLE 3 AP scores and relative computational costs of FOD-DPM and DPM on the PASCAL VOC 2007 dataset.
| DPM(rel5) | DPM(rel5) | aero 33.2 60.3 | bike bird boat | 10.2 16.1 | bottle 27.3 | bus car 54.3 58.2 | cat 23.0 | chair 20.0 | cow 24.1 | table 26.7 | dog 12.7 | horse 58.1 | mbike 48.2 | person 43.2 | plant 12.0 | sheep 21.1 | sofa train 36.1 46.0 | tv mAP 43.5 33.7 | Comp. Cost 100 |
|-------------|-------------|------------------|------------------|-------------|---------------|---------------------|------------|--------------|------------|--------------|------------|--------------|--------------|---------------|--------------|--------------|------------------------|--------------------|------------------|
| | 1 | 31.0 37.1 | 10 | 14.3 | 12.9 | 47.1 46.7 | 28.0 | 9.3 | 15.5 | 26.2 | 10.7 | 56 | 39.7 | 29.4 | 9.8 | 15.5 | 27.6 43.4 | 21.5 26.6 | 0.46 |
| FOD-DPM | 5 | 32.3 50.0 | 9.8 | 15.2 | 21.8 | 50.0 63.0 25.9 | | 17.1 | 20.5 | 25.4 | 9.7 | 61.4 | 44.6 | 38 | 9.2 | 19.7 | 30.1 43.1 | 32.1 31.0 | 1.84 |
| FOD-DPM | 9 | 33.2 56.6 | 9.9 | 15.6 | 25.3 | 54.6 65.3 | 25.3 | 19.8 | 22 | 24.9 | 9.4 | 60.9 | 50.8 | 41.7 | 10 | 20.4 | 34.9 44.3 37.3 | 33.1 | 3.09 |
| FOD-DPM | 13 33.4 | 59.9 | 10 | 15.7 | 27.2 | 54.8 65.7 | 25.0 | 20.5 | 22 | 24.8 | 9.2 | 62 | 51.9 | 44.5 | 10.2 | 20.9 | 36.8 46.2 | 40.9 34.1 | 4.16 |
Fig. 8. AP scores achieved by FOD-DPM and DPM per class.
performance result of this approach. FOD-DPM achieves a similar average performance to that of DPM ( 98 2% . relative performance, 0 6 . AP gap) using 9 fixations and exceeds DPM's performance starting from 11 fixations. On some classes (e.g. bus, car, horse), FOD-DPM exceeds DPM's performance probably due to lesser number of evaluations and reduced false positives; on other cases (e.g. bike, dog, tv) FOD-DPM underperforms probably due to low recall rate of the FOD detector for these classes. Figure 8 gives per class AP scores of FOD-DPM and DPM to demonstrate the improvement from 1 to 9 fixations.
We compare the computational complexities of FODDPM and DPM by their total number of operations as defined above. For a given object class, DPM model has 3 root filters and 8 6x6 part filters. It is straightforward to calculate the number of operations performed by DPM as it uses the SW method. For FOD-DPM, the total number of operations is calculated by adding: 1) FOD's operations and 2) DPM's operations at each high-scoring foveal detection b , one DPM root filter (with the most similar shape as b ) and 8 parts evaluated at all locations within the boundaries of this root filter. Note that we ignore the time for optimal placing of parts in both DPM and FOD-DPM. Cost of feature extraction is also not included as the two methods use the same feature extraction code. We report the computational costs of FODDPM and DPM in the last column of Table 3. The costs are normalized so that DPM's cost is 100 operations. Results show that FOD-DPM drastically reduces the cost from 100 to 3 09 . for 9 fixations. Assuming both methods are implemented equally efficiently, this would translate to an approximately 32 x speed-up which is better than the 20 x speed-up reported for a cascaded evaluation of DPM [13]. These results demonstrate the effectiveness of our foveated object detector in guiding the visual search.
Finally, in Figure 9 we give sample detections by the FOD system. We ran the trained bicycle, person and car models on an image outside of the PASCAL datasaet. The models were assigned the same initial location and we ran them for 3 fixations. Results show that the each model fixates at different locations, and these locations are attracted towards instances of the target objects being searched.
## 4 RELATED WORK
The sliding window (SW) method is the dominant model of search in object detection. The complexity of identifying object instances in a given image is O ( mn ) where m is the number of locations to be evaluated and n is the number of object classes to be searched for. Efficient alternatives to sliding windows can be categorized in two groups: (i) methods aimed at reducing m , (ii) methods aimed at reducing n . Since typically m >> n , the are a larger number efforts in trying to reduce m , however, reducing the contribution of the number of object classes has recently been receiving increasing interest as search for hundreds of thousands of object classes has started to be tackled [7]. According to this categorization, our proposed FOD method falls into the first group as it is designed to locate object instances by making a set of sequential fixations where in each fixation only a sparse set of locations are evaluated.
## 4.1 Reducing the number of evaluated locations ( m )
In efforts to reduce the number of locations to be evaluated, one line of research is the branch-and-bound methods ( [20], [24]) where an upper bound on the quality function of the detection model is used in a global branch and bound optimization scheme. Although the authors provide efficiently computable upper bounds for popular quality functions (e.g. linear template, bag-ofwords, spatial pyramid), it might not be trivial to derive suitable upper bounds for a custom quality function. Our method, on the other hand, uses binary classification detection model and is agnostic to the quality function used.
Another line of research is the casdaded detection framework ( [13], [23], [48]) where a series of cheap to expensive tests are done to locate the object. Cascaded detection is similar to our method in the sense that
Fig. 9. Fixation locations and bounding box predictions of FOD for different object classes (bicycle, person, and car from left to right) but for the same image and initial point of fixation.
simple, coarse and cheap evaluations are used together with complex, fine and expensive evaluations. However, we differ with it in that it is essentially a sliding window method with a coarse-to-fine heuristic used to reduce the number of total evaluations. Another coarse-to-fine search scheme is presented in [34] where a set of low to high resolution templates are used. The method starts by evaluating the lowest resolution template - which is essentially a sliding window operation - and selecting the high responding locations for further processing with higher resolution templates. Our method, too, uses a set of varying resolution templates; however, these templates are evaluated at every fixation instead of serializing their evaluations with respect to resolution.
In [46], a segmentation based method is proposed to yield a small set of locations that are likely to corresponds to objects, which are subsequently used to guide the search in a selective manner. The locations are identified in an object class-independent way using an unsupervised multiscale segmentation approach. Thus, the method evaluates the same set of locations regardless of which object class is being searched for. In contrast, in our method, selection of locations to be foveated is guided by learned object class templates.
The method in [1], similar to ours, works like a fixational system: at a given time step, the location to be evaluated next is decided based on previous observations. However, there are important differences. In [1], only a single location is evaluated at a time step whereas we evaluate all template locations within the visual field at each fixation. Their method returns only one box as the result whereas our method is able to output many predictions.
There are also vector quantization based methods [19], [21], [40] aiming to reduce the time required to compute linear template evaluations. These methods to reduce the contribution of m in O ( mn ) are orthogonal to our foveated approach. Thus, vector quantization approaches can be integrated with the proposed foveated object detection method.
## 4.2 Reducing the number of evaluations of object classes( n )
Works in this group aim to reduce the time complexity contributed by the number of object classes. The method proposed in [7] accelerates the search by replacing the costly linear convolution by a locality sensitive hashing scheme that works on non-linearly coded features. Although they evaluate all locations in a given image, their approach scale constantly over the number of classes, which enables them to evaluate thousands of object classes in a very short amount of time.
Another method [42] uses a sparse representation of object part templates, and then uses the basis of this representation to reconstruct template responses. When the number of object categories is large, sparse representation serves as a shared dictionary of parts and accelerates the search.
Another line of research (e.g. [3], [16], [36]) accelerate search by constructing classifier hierarchies. These generally work by pruning unlikely classes while descending the classifier hierarchy.
Importantly, the way the methods in this group accelerate search is orthogonal to the savings proposed by using a foveated visual field. Therefore, these methods are complementary and can be integrated with our method to further accelerate search.
In the context of the references listed in this and the previous sections, our method of search through fixations using a non-uniform foveated visual field is novel.
## 4.3 Biologically inspired methods
There have been previous efforts, (e.g. [41]), on biologically inspired object recognition. However, these models do not have a foveated visual field and thus do not execute eye movements. More recent work has implemented biologically inspired search methods. In [11], a fixed, pre-attentive, low-resolution wide-field camera is combined with a shiftable, attentive, highresolution narrow-field camera, where the pre-attentive camera generates saccadic targets for the attentive, highresolution camera. The fundamental difference between this and our method is that while their pre-attentive system has the same coarse resolution everywhere in
the visual field, our method, which is a model of the V1 layer of the visual cortex, has a varying resolution that depends on the radial distance to the center of the fovea. There have been previous efforts to create foveated search models with eye movements [30], [31], [38], [51]. Such models have been applied mostly to detect simple signals in computer generated noise [31], [51] and used as benchmarks to compare against human eye movements and performance.
Other biologically inspired methods include the target acquisition model (TAM) [50], [52], the Infomax model [5] and artificial neural network based models [2], [25]. TAM is a foveated model and it uses scale invariant feature transform (SIFT) features [27] for representation and utilizes a training set of images to learn the appearance of the target object. However, it does not include the variability in object appearance due to scale and viewpoint, and the evaluation is done by placing the objects on a uniform background. The Infomax, on the other hand, can use any previously trained object detector and works on natural images. They report successful results on a face detection task. Both TAM and Infomax uses the same template for all locations in the visual field while our method uses different templates for different locations. [25] was applied to image categorization and [2] to object tracking in videos. Critically, none of these models have been tested on standard object detection datasets nor they have been compared to a SW approach to evaluate the potential performance loss and computational savings of modeling a foveated visual field.
## 5 CONCLUSIONS AND DISCUSSION
We present an implementation of a foveated object detector with a recent neurobiologically plausible model of pooling in the visual periphery and report the first ever evaluation of a foveated object detection model on a standard data set in computer vision (PASCAL VOC 2007). Our results show that the foveated method achieves nearly the same performance as the sliding window method at 49 6% . of sliding window's computational cost. Using a richer model (such as DPM [14]) to evaluate high-scoring locations, FOD is able to outperform the DPM with more computational savings than a state-of-the-art cascaded detection system [13]. These results suggest that using a foveated visual system offers a promising potential for the development of more efficient object detectors.
## APPENDIX A APPROXIMATION OF THE BAYESIAN DECISION
Derivation for Equation (4):
$$\frac { P ( y _ { b } = 1 | f _ { 1 }, \dots, f _ { m }, I ) } { P ( y _ { b } = 0 | f _ { 1 }, \dots, f _ { m }, I ) } = \frac { P ( f _ { 1 }, \dots, f _ { m } | y _ { b } = 1, I ) P ( y _ { b } = 1 | I ) } { P ( f _ { 1 }, \dots, f _ { m } | y _ { b } = 0, I ) P ( y _ { b } = 0 | I ) } \begin{array} { c } \left [ 1 4 \right ] & P. E. \\ \text{Obje} \\ \text{IEEE} \\ 1 6 4 5 \end{array}$$
$$\underset { \text{$\text{ter of}$} } { \text{$--$
\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{ $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text
$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text<$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text#$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text}$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text`$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text'$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text1$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text.$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\texts$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$$
## APPENDIX B
## DETECTION SCORE AFTER MULTIPLE FIXATIONS
Derivation for Equation (6):
$$\stackrel { \infty } { 2 } \text{ and } \stackrel { \infty } { [ } 5 ] {. } \\ \text{ant fea-} \\ \text{ion and } \quad \text{inc} \alpha \, \text{ 5 \, \text{w} \, \text{w} \, \text{or} \, \text{t} }$$
using (5), we get
$$\underset { \substack { \text{causes} \\ \text{due the} \\ \text{d view} } } \sum _ { i = 1 } ^ { m } \log \left ( \frac { P ( y _ { b } = 1 | f _ { i }, I ) } { 1 - P ( y _ { b } = 1 | f _ { i }, I ) } \right ) = \sum _ { i = 1 } ^ { m } \log \left ( \frac { \frac { 1 } { 1 + e ^ { - s ( I, b, f _ { i } ) } } } { 1 - \frac { 1 } { 1 + e ^ { - s ( I, b, f _ { i } ) } } } \right ) \\ \underset { \text{e other} } } \text{causes}$$
$$= \sum _ { i = 1 } ^ { m } \log \left ( \frac { 1 } { e ^ { - s ( I, b, f _ { i } ) } } \right ) = \sum _ { i = 1 } ^ { m } s ( I, b, f _ { i } ). \quad ( 1 7 )$$
## REFERENCES
- [1] B. Alexe, N. Heess, Y. W. Teh, and V. Ferrari. Searching for objects driven by context. In Advances in Neural Information Processing , pages 1-9, 2012.
- [2] L. Bazzani, N. de Freitas, H. Larochelle, V. Murino, and J.-A. Ting. Learning attentional policies for tracking and recognition in video with deep networks. In Int'l Conf. on Machine Learning , 2011.
- [3] S. Bengio, J. Weston, and D. Grangier. Label Embedding Trees for Large Multi-Class Tasks. In J. Lafferty, C. Williams, J. ShaweTaylor, R. Zemel, and A. Culotta, editors, Advances in Neural Information Processing , page 163171. 2010.
- [4] B. R. Beutter, M. P. Eckstein, and L. S. Stone. Saccadic and perceptual performance in visual search tasks. i. contrast detection and discrimination. Journal of Optical Society of America , 20:1341 1355, 2003.
- [5] N. J. Butko and J. R. Movellan. Infomax control of eye movements. IEEE Trans. on Auton. Ment. Dev. , 2(2):91-107, June 2010.
- [6] N. Dalal and B. Triggs. Histograms of oriented gradients for human detection. In Conf. on Computer Vision and Pattern Recognition , pages 886-893, 2005.
- [7] T. Dean, M. A. Ruzon, M. Segal, J. Shlens, S. Vijayanarasimhan, and J. Yagnik. Fast , Accurate Detection of 100,000 Object Classes on a Single Machine. In Conf. on Computer Vision and Pattern Recognition , 2013.
- [8] S. Divvala, A. Efros, and M. Hebert. How important are 'deformable parts' in the deformable parts model? In European Conf. on Computer Vision , Workshop on Parts and Attributes, pages 3140, 2012.
- [9] M. P. Eckstein. Visual search: a retrospective. Journal of vision , 11(5):14-, Jan. 2011.
- [10] M. P. Eckstein, B. A. Drescher, and S. S. Shimozaki. Attentional cues in real scenes, saccadic targeting, and Bayesian priors. Psychological science , 17(11):973-80, Nov. 2006.
- [11] J. Elder, S. Prince, Y. Hou, M. Sizintsev, and E. Olevskiy. PreAttentive and Attentive Detection of Humans in Wide-Field Scenes. Int'l Journal of Computer Vision , 72(1):4766, 2007.
- [12] M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman. The PASCAL Visual Object Classes Challenge 2007 (VOC2007) Results. http://www.pascalnetwork.org/challenges/VOC/voc2007/workshop/index.html.
- [13] P. Felzenszwalb, R. Girshick, and D. McAllester. Cascade object detection with deformable part models. In Conf. on Computer Vision and Pattern Recognition , 2010.
- [14] P. F. Felzenszwalb, R. B. Girshick, D. McAllester, and D. Ramanan. Object detection with discriminatively trained part based models. IEEE Trans. on Pattern Analysis and Machine Intelligence , 32(9):16271645, 2010.
- [15] J. Freeman and E. P. Simoncelli. Metamers of the ventral stream. Nature Neuroscience , 14(9):1195-1201, 2011.
- [16] T. Gao and D. Koller. Discriminative learning of relaxed hierarchy for large-scale visual recognition. In Int'l Conf. on Computer Vision , pages 2072-2079, 2011.
- [17] R. B. Girshick, P. F. Felzenszwalb, and D. McAllester. Discriminatively trained deformable part models, release 5. http://people.cs.uchicago.edu/ rbg/latent-release5/.
- [18] B. Hariharan, J. Malik, and D. Ramanan. Discriminative decorrelation for clustering and classification. In European Conf. on Computer Vision , 2012.
- [19] H. Jegou, M. Douze, and C. Schmid. Product quantization for nearest neighbor search. IEEE Trans. on Pattern Analysis and Machine Intelligence , 33(1):117-128, Jan 2011.
- [20] I. Kokkinos. Rapid deformable object detection using dual-tree branch-and-bound. In Advances in Neural Information Processing , 2011.
- [21] I. Kokkinos. Bounding part scores for rapid detection with deformable part models. In 2nd Parts and Attributes Workshop, in conjunction with ECCV , pages 41-50, 2012.
- [22] P. Kontschieder, S. R. Bul, A. Criminisi, P. Kohli, M. Pelillo, and H. Bischof. Context-sensitive decision forests for object detection. In Advances in Neural Information Processing , 2012.
- [23] C. H. Lampert. An efcient divide-and-conquer cascade for nonlinear object detection. In Conf. on Computer Vision and Pattern Recognition , 2010.
- [24] C. H. Lampert, M. B. Blaschko, and T. Hofmann. Efficient subwindow search: A branch and bound framework for object localization. IEEE Trans. on Pattern Analysis and Machine Intelligence , 31(12):2129-2142, Dec 2009.
- [25] H. Larochelle and G. Hinton. Learning to combine foveal glimpses with a third-order Boltzmann machine. In Advances in Neural Information Processing , pages 1-9, 2010.
- [26] D. M. Levi, S. A. Klein, and A. P. Aitsebaomo. Vernier acuity, crowding and cortical magnification. Vision Research , 25(7):963977, 1985.
- [27] D. G. Lowe. Distinctive Image Features from Scale-Invariant Keypoints. International Journal of Computer Vision , 60(2):91-110, Nov. 2004.
- [28] S. C. Mack and M. P. Eckstein. Object co-occurrence serves as a contextual cue to guide and facilitate visual search in a natural viewing environment. Journal of vision , 11(9):1-16, Jan. 2011.
- [29] T. Malisiewicz, A. Gupta, and A. A. Efros. Ensemble of exemplarsvms for object detection and beyond. In ICCV , 2011.
- [30] C. Morvan and L. T. Maloney. Human visual search does not maximize the post-saccadic probability of identifying targets. PLoS computational biology , 8(2):e1002342, Feb. 2012.
- [31] J. Najemnik and W. S. Geisler. Optimal eye movement strategies in visual search. Nature , 434:387 - 391, 2005.
- [32] J. Najemnik and W. S. Geisler. Simple summation rule for optimal fixation selection in visual search. Vision research , 49(10):1286-94, June 2009.
- [33] M. B. Neider and G. J. Zelinsky. Scene context guides eye movements during visual search. Vision research , 46(5):614-21, Mar. 2006.
- [34] M. Pedersoli, A. Vedaldi, and J. Gonzalez. A coarse-to-fine approach for fast deformable object detection. In Conf. on Computer Vision and Pattern Recognition , pages 1353-1360, 2011.
- [35] T. J. Preston, F. Guo, K. Das, B. Giesbrecht, and M. P. Eckstein. Neural representations of contextual guidance in visual search of real-world scenes. The Journal of neuroscience : the official journal of the Society for Neuroscience , 33(18):7846-55, May 2013.
- [36] N. Razavi, J. Gall, and L. Van Gool. Scalable multi-class object detection. In Conf. on Computer Vision and Pattern Recognition , pages 1505-1512, 2011.
- [37] X. Ren and D. Ramanan. Histograms of sparse codes for object detection. In Conf. on Computer Vision and Pattern Recognition , 2013.
- [38] L. W. Renninger, J. M. Coughlan, P. Verghese, and J. Malik. An information maximization model of eye movements. In Advances in Neural Information Processing , pages 1121-1128, 2004.
- [39] J. Rovamo, L. Leinonen, P. Laurinen, and V. Virsu. Temporal integration and contrast sensitivity in foveal and peripheral vision. Perception , 13(6):665-74, Jan. 1984.
- [40] M. Sadeghi and D. Forsyth. Fast Template Evaluation with Vector Quantization. In Advances in Neural Information Processing , 2013.
- [41] T. Serre, L. Wolf, and T. Poggio. Object recognition with features inspired by visual cortex. In Conf. on Computer Vision and Pattern Recognition , 2005.
- [42] H. O. Song, S. Zickler, T. Althoff, R. Girshick, M. Fritz, C. Geyer, P. Felzenszwalb, and T. Darrell. Sparselet Models for Efficient
- Multiclass Object Detection. In European Conf. on Computer Vision , 2012.
- [43] H. Strasburger, I. Rentschler, and M. J¨ uttner. Peripheral vision and pattern recognition: a review. Journal of vision , 11(5):13, Jan. 2011.
- [44] B. W. Tatler. The central fixation bias in scene viewing: selecting an optimal viewing position independently of motor biases and image feature distributions. Journal of vision , 7(14):4.1-17, Jan. 2007.
- [45] A. Torralba, A. Oliva, M. S. Castelhano, and J. M. Henderson. Contextual guidance of eye movements and attention in realworld scenes: The role of global features in object search. Psychological Review , 113(4):766-786, 2006.
- [46] K. E. A. van de Sande, J. R. R. Uijlings, T. Gevers, and A. W. M. Smeulders. Segmentation as selective search for object recognition. In Int'l Conf. on Computer Vision , 2011.
- [47] P. Verghese. Active search for multiple targets is inefficient. Vision Research , 74:61-71, 2012.
- [48] P. Viola and M. J. Jones. Robust real-time face detection. Int'l Journal of Computer Vision , 57(2):137-154, May 2004.
- [49] T. Wertheim. ¨ ber U die indirekte sehsch¨ arfe. Zeitschrift f ¨ r u Psychologie und Physiologie der Sinnesorgane , 7:172-187, 1894.
- [50] G. J. Zelinsky. A theory of eye movements during target acquisition. Psychological Review , 115:787-835, 2008.
- [51] S. Zhang and M. P. Eckstein. Evolution and optimality of similar neural mechanisms for perception and action during search. PLoS Computational Biology , 6(9):e1000930, 2010.
- [52] W. Zhang, H. Yang, D. Samaras, and G. J. Zelinsky. A computational model of eye movements during object class detection. In Advances in Neural Information Processing , 2006.
- [53] L. Zhu, Y. Chen, A. Yuille, and W. Freeman. Latent hierarchical structural learning for object detection. In Conf. on Computer Vision and Pattern Recognition , 2010.
- [54] X. Zhu, C. Vondrick, D. Ramanan, and C. Fowlkes. Do we need more training data or better models for object detection? In British Machine Vision Conf. , pages 1-11, 2012.
|
10.1371/journal.pcbi.1005743
|
[
"Emre Akbas",
"Miguel P. Eckstein"
] |
2014-08-04T20:49:26+00:00
|
2017-11-06T07:17:44+00:00
|
[
"cs.CV"
] |
Object Detection Through Exploration With A Foveated Visual Field
|
We present a foveated object detector (FOD) as a biologically-inspired
alternative to the sliding window (SW) approach which is the dominant method of
search in computer vision object detection. Similar to the human visual system,
the FOD has higher resolution at the fovea and lower resolution at the visual
periphery. Consequently, more computational resources are allocated at the
fovea and relatively fewer at the periphery. The FOD processes the entire
scene, uses retino-specific object detection classifiers to guide eye
movements, aligns its fovea with regions of interest in the input image and
integrates observations across multiple fixations. Our approach combines modern
object detectors from computer vision with a recent model of peripheral pooling
regions found at the V1 layer of the human visual system. We assessed various
eye movement strategies on the PASCAL VOC 2007 dataset and show that the FOD
performs on par with the SW detector while bringing significant computational
cost savings.
|
1408.0816v2
|
## Molecular gas content in strongly-lensed z GLYPH<24> 1 5 : GLYPH<0> 3 star-forming galaxies with low IR luminosities ?
M. Dessauges-Zavadsky 1 , M. Zamojski 1 , D. Schaerer 1 2 ; , F. Combes 3 , E. Egami 4 , A. M. Swinbank 5 , J. Richard 6 , P. Sklias 1 , T. D. Rawle 7 , M. Rex 4 , J.-P. Kneib 8 9 ; , F. Boone 2 , and A. Blain 10
- 1 Observatoire de Gen` eve, Universit´ e de Gen`ve, 51 Ch. des Maillettes, 1290 Versoix, Switzerland e
- 2 CNRS, IRAP, 14 Avenue E. Belin, 31400 Toulouse, France
- 3 Observatoire de Paris, LERMA, 61 Avenue de l'Observatoire, 75014 Paris, France
- 4 Steward Observatory, University of Arizona, 933 North Cherry Avenue, Tucson, AZ 85721, USA
- 5 Institute for Computational Cosmology, Durham University, South Road, Durham DH1 3LE, UK
- 6 CRAL, Observatoire de Lyon, Universit´ e Lyon 1, 9 Avenue Ch. Andr´ e, 69561 Saint Genis Laval Cedex, France
- 7 ESAC, ESA, PO Box 78, Villanueva de la Canada, 28691 Madrid, Spain
- 8 Laboratoire d'Astrophysique, Ecole Polytechnique F´ ed´rale e de Lausanne (EPFL), Observatoire de Sauverny, 1290 Versoix, Switzerland
- 9 Aix Marseille Universit´ e, CNRS, LAM, UMR 7326, 13388, Marseille, France
- 10 Department of Physics & Astronomy, University of Leicester, University Road, Leicester LE1 7RH, UK
## ABSTRACT
To extend the molecular gas measurements to more typical star-forming galaxies (SFGs) with SFR < 40 M GLYPH<12> yr GLYPH<0> 1 and M GLYPH<3> < 2 5 : GLYPH<2> 10 10 M GLYPH<12> at z GLYPH<24> 1 5 : GLYPH<0> 3, we have observed CO emission with the IRAM Plateau de Bure Interferometer and 30 m telescope for five strongly-lensed galaxies, selected from the Herschel Lensing Survey. These observations are combined with a compilation of CO measurements from the literature. From this, we infer the CO luminosity correction factors r 2 1 ; = 0 81 : GLYPH<6> 0 20 and : r 3 1 ; = 0 57 : GLYPH<6> 0 15 for : the J = 2 and J = 3 CO transitions, respectively, valid for SFGs at z > 1. The combined sample of CO-detected SFGs at z > 1 shows a large spread in star formation e GLYPH<14> ciency (SFE) with a dispersion of 0.33 dex, such that the SFE extend well beyond the low values of local spirals and overlap the distribution of z > 1 sub-mm galaxies. We find that the spread in SFE (or equivalently in molecular gas depletion timescale) is due to variations of several physical parameters, primarily the specific star formation rate, but also stellar mass and redshift. Correlations of the SFE with the o GLYPH<11> set from the main-sequence and the compactness of the starburst are less clear. The possible increase of the molecular gas depletion timescale with M GLYPH<3> now revealed by low stellar mass SFGs at z > 1 and also observed at z = 0 is in contrast to the constant molecular gas depletion timescale generally admitted and refutes the linearity of the Kennicutt-Schmidt relation. A net rise of the molecular gas fraction ( f gas ) is observed from z GLYPH<24> 0 2 to : z GLYPH<24> 1 2, followed by a very : mild increase toward higher redshifts, as found in earlier studies. At each redshift the molecular gas fraction shows a large dispersion, mainly due to the dependence of f gas on stellar mass, producing a gradient of increasing f gas with decreasing M GLYPH<3> . We provide the first measurement of the molecular gas fraction of z > 1 SFGs at the lowM GLYPH<3> end between 10 9 4 : < M GLYPH<3> = M GLYPH<12> < 10 9 9 : , reaching a mean h f gas i = 0 69 : GLYPH<6> 0 18 which shows a clear : f gas upturn at these lower stellar masses. Finally, we find evidence for a non-universal dust-to-gas ratio among high-redshift SFGs and sub-mm galaxies, local spirals and ultra-luminous infrared galaxies with near-solar metallicities, as inferred from a homogeneous analysis of their rest-frame 850 GLYPH<22> m luminosity per unit gas mass. z > 1 SFGs show a trend for a lower L GLYPH<23> (850 GLYPH<22> m) = M gas mean by 0.33 dex compared to the other galaxy populations.
Key words. cosmology: observations - gravitational lensing: strong - galaxies: high-redshift - ISM: molecules - galaxies: evolution
## 1. Introduction
Gas, stars, dust, and metals are the basic galaxy constituents. They determine the majority of observable properties in galaxies at all wavelengths. Therefore, the characterization of these observable properties from present up to high-redshifts provides stringent tests and anchors for the physical processes that regulate the galaxy evolution.
Several empirical relationships between gas, stars, dust, and metals have been observationally identified. The first relationship is the correlation between the star formation rate (SFR) and stellar mass ( M GLYPH<3> ). It determines the so-called 'main-
Send o GLYPH<11> print requests to : [email protected]
? Based on observations carried out with the IRAM Plateau de Bure Interferometer and the IRAM 30 m telescope. IRAM is supported by CNRS INSU (France), the MPG (Germany) and the IGN (Spain). /
sequence', representing the locus where local and high-redshift star-forming galaxies (SFGs) lie in the SFRM GLYPH<3> plane. The correlation is slightly sublinear and evolves with redshift, such that high-redshift galaxies form more stars per unit time than the low-redshift ones with the same stellar mass (e.g., Noeske et al. 2007; Daddi et al. 2007; Elbaz et al. 2007; Rodighiero et al. 2010; Salmi et al. 2012). This implies an increase of the specific star formation rate, defined as the ratio of the star formation rate over the stellar mass, with redshift. The second relationship is the correlation between the gas-phase metallicity and stellar mass (e.g., Tremonti et al. 2004; Savaglio et al. 2005; Erb et al. 2006; Maiolino et al. 2008). It shows a redshift evolution toward lower metallicities at higher redshifts. More recently, Mannucci et al. (2010, 2011) found the 'fundamental' metallicity relation (FMR) that connects at the same time metallicity, stellar mass, and star formation rate, as revealed by local and z < 2 5 galaxies. : The third important relationship is the Kennicutt-Schmidt rela-
tion, or the star-formation relation, that relates the star formation rate surface density to the total gas (H I + H2) surface density through a power law (Kennicutt 1998a). Mostly constrained by local galaxies so far (Leroy et al. 2008; Bigiel et al. 2008, 2011), the tightest correlation is observed between the star formation rate surface density and the H2 gas surface density well parametrized by a linear relation, meaning that the molecular gas is being consumed at a constant rate within a molecular gas depletion timescale of about 1.5 Gyr. However, both the COLD GASS survey of local massive galaxies (log( M GLYPH<3> = M ) GLYPH<12> > 10) from Saintonge et al. (2011) and the recent ALLSMOG survey of local low-mass galaxies (9 < log( M GLYPH<3> = M ) GLYPH<12> < 10) from Bothwell et al. (2014) show an increase of the molecular gas depletion timescale with stellar mass and thus bring evidence against a linear Kennicutt-Schmidt relation. A linear KennicuttSchmidt relation for the H2 gas surface densities seems to hold for high-redshift SFGs, but with a much shorter molecular gas depletion timescale of about 0.7 Gyr (Tacconi et al. 2013), even though they are more gas-rich (e.g., Daddi et al. 2010a; Tacconi et al. 2010; Genzel et al. 2010). Finally, there is the relationship between the dust-to-gas ratio and metallicity seen in nearby galaxies (e.g., Issa et al. 1990; Dwek 1998; Edmunds 2001; Inoue 2003; Draine et al. 2007; Leroy et al. 2011). Attempts to measure the dust-to-gas ratios of high-redshift galaxies support a similar correlation, as well as a trend toward smaller dust-togas ratios at higher redshifts (Saintonge et al. 2013; Chen et al. 2013).
Most of these empirical correlations are both qualitatively and quantitatively consistent with the so-called 'bathtub' model, which gives a good analytical representation of the gross features of the star-forming galaxy evolution (e.g., Bouch´ e et al. 2010; Lagos et al. 2012; Lilly et al. 2013; Dekel & Mandelker 2014). Solely based on the equation of conservation of gas mass in a galaxy, this model assumes that galaxies lie in a quasisteady state equilibrium, where their ability to form stars is regulated by the availability of gas replenished through the accretion rate dictated by cosmology and the amount of material they return into the intergalactic medium through gas outflows. The predicted redshift and mass dependences are the following (although slightly di GLYPH<11> ering from authors to authors): the average gas accretion rate varies as M 1 1 : halo (1 + z ) 2 2 : , the specific star formation rate as (1 + z ) 2 2 : , the gas depletion timescale as (1 + z ) GLYPH<0> 1 5 : , and the molecular gas fraction steadily increases with redshift and decreases with stellar mass. In agreement with cosmological hydrodynamic simulations from Dav´ e et al. (2011, 2012), the latter, in addition, predict a decrease of the gas depletion timescale with the stellar mass, instead of a constant gas depletion timescale.
Improvements in the sensitivity of the IRAM Plateau de Bure Interferometer have made it possible over the last years to start getting a census of the molecular gas content in star-forming galaxies near the peak of the cosmic star formation activity. However, the samples of CO-detected objects at z = 1 GLYPH<0> 1 5 : and z = 2 GLYPH<0> 3 are still small and confined to the high-SFR and highM GLYPH<3> end of main-sequence SFGs (Daddi et al. 2010a; Genzel et al. 2010; Tacconi et al. 2010, 2013). In this paper, we extend the dynamical range of star formation rates and stellar masses of star-forming galaxies with observationally constrained molecular gas contents below SFR < 40 M GLYPH<12> yr GLYPH<0> 1 and M GLYPH<3> < 2 5 : GLYPH<2> 10 10 M , and thus reach the GLYPH<12> L ? to subL ? domain of z GLYPH<24> 1 5 : GLYPH<0> 3 galaxies (Gruppioni et al. 2013). This domain of physical parameters is accessible to CO emission measurements only with the help of gravitational lensing, a technique that proved to be e GLYPH<14> cient by a few objects already(Baker et al. 2004; Coppin et al. 2007; Saintonge et al. 2013). Therefore, our five galaxies selected for CO follow-up observations come from the Herschel Lensing Survey of massive galaxy cluster fields (HLS; Egami et al. 2010), designed to detect lensed, highredshift background galaxies and probe more typical, intrinsically fainter galaxies than those identified in large-area, blankfield surveys. The gaseous, stellar, and dust properties inferred for these (sub-) L ? main-sequence star-forming galaxies (see also Sklias et al. 2014) are put face-to-face with a large comparison sample of local and high-redshift galaxies with CO measurements reported in the literature, which together provide new tests and anchors for galaxy evolution models.
In Sect. 2 we describe the target selection and their physical properties, and present the comparison sample of CO-detected galaxies from the literature. In Sect. 3 we report on CO observations performed with the IRAM Plateau de Bure Interferometer and 30 m telescope and discuss the CO results. In Sect. 4 we infer the CO luminosity correction factors for the J = 2 and J = 3 CO rotational transitions. The gaseous and stellar properties of our strongly-lensed galaxies are placed in the general context of galaxies with CO measurements in Sect. 5, where the main objective is to understand what drives the large spread in star formation e GLYPH<14> ciency observed in high-redshift SFGs. In Sect. 6 we explore the redshift evolution and the stellar mass dependence of their molecular gas fractions. In Sect. 7 we discuss the universality of the dust-to-gas ratio inferred from a homogeneous analysis of the rest-frame 850 GLYPH<22> mcontinuum. Summary and conclusions are given in Sect. 8. Individual CO properties and inferred kinematics of our selected strongly-lensed galaxies are described in Appendix A.
Throughout the paper, we adopt the initial mass function (IMF) of Chabrier (2003) and scale the values from the literature by the factor of 1.7 when the Salpeter (1955) IMF is used. The designation to 'gas' always refers to the molecular gas (H2) only and not the total gas (H I + H2). All the molecular gas masses are derived from the observed CO emission via the 'Galactic' CO(1-0)-H2 conversion factor X CO = 2 GLYPH<2> 10 20 cm GLYPH<0> 2 = (K km s GLYPH<0> 1 ), or GLYPH<11> = 4 36 M : GLYPH<12> = (K km s GLYPH<0> 1 pc 2 ) which includes the correction factor of 1.36 for helium. We use the cosmology with H0 = 70 km s GLYPH<0> 1 Mpc GLYPH<0> 1 , GLYPH<10> M = 0 29, and : GLYPH<10> GLYPH<3> = 0 71. :
## 2. High-redshift samples of CO-detected galaxies
## 2.1. Target selection
The HLS is providing us with unique targets ideal for CO followup studies. We used this dataset to select galaxies at z GLYPH<24> 1 5 : GLYPH<0> 3 with low intrinsic (delensed) IR luminosities L IR < 4 GLYPH<2> 10 11 L GLYPH<12> , as derived from the Herschel / PACS (100 and 160 GLYPH<22> m) and SPIRE (250, 350, and 500 GLYPH<22> m) SED modelling. These galaxies are of particular interest, because they allow probing the molecular gas content at high redshift in a regime of star formation rates SFR < 40 M GLYPH<12> yr GLYPH<0> 1 still very poorly explored, while being typical of 'normal' star-forming galaxies. In addition, we required that our targets (1) are strongly lensed with well-known magnification factors derived from robust lens modelling (Richard et al. 2007, 2011) in order to make the CO emission of these intrinsically faint L IR objects accessible to current millimeter instruments, (2) have known spectroscopic redshifts from our optical and near-IR observing campaigns (Richard et al. 2007, 2011), (3) have well-sampled global SEDs from optical, near-IR to IR obtained with ground-based telescopes, the Hubble Space Telescope (HST), and the Spitzer satellite in order to have tight estimates of their stellar masses and star formation rates, and
Table 1. Physical properties derived from SED fitting
| Source | z H GLYPH<11> | GLYPH<22> a | L IR b (10 11 L GLYPH<12> ) | M GLYPH<3> c (10 10 M GLYPH<12> ) | SFR c (M GLYPH<12> yr GLYPH<0> 1 ) | M dust d (10 7 M GLYPH<12> ) | T dust d (K) | L GLYPH<23> (850 GLYPH<22> m) d (10 29 erg s GLYPH<0> 1 Hz GLYPH<0> 1 ) |
|--------------|-----------------|---------------|-------------------------------|-------------------------------------|--------------------------------------|--------------------------------|----------------|---------------------------------------------------------------------------|
| A68-C0 | 1.5864 | 30 | 1 : 18 GLYPH<6> 0 : 07 | 2 : 0 GLYPH<6> 0 : 65 | 9 + 3 : 3 GLYPH<0> 3 9 | 4.7 | 34.5 | 5.0 |
| A68-HLS115 | 1.5869 | 15 | 3 : 42 GLYPH<6> 0 : 16 | 0 : 81 + 0 : 17 GLYPH<0> 0 20 | : 25 + 2 : 6 GLYPH<0> 6 | 3.0 | 37.5 | 7.7 |
| MACS0451-arc | 2.013 | 49 | 1 : 36 GLYPH<6> 0 : 13 e | : 0 : 25 GLYPH<6> 0 : 02 | 19 + 0 : 6 GLYPH<0> 0 2 | 0.6 | 47.4 | 3.4 |
| A2218-Mult | 3.104 | 14 | 2 : 26 GLYPH<6> 0 : 15 | | : | | | |
| A68-h7 | 2.15 f | 3 | 18 : 3 + 0 : 9 GLYPH<0> 1 2 | 15 : 4 + 0 : 9 GLYPH<0> 3 8 | 76 GLYPH<6> 3 : 4 | 18.2 | 43.3 | 27.6 |
| MS1512-cB58 | 2.729 | 30 | : 3 : 04 GLYPH<6> 0 : 07 | : 0 : 44 GLYPH<6> 0 : 03 | 37 + 2 : 4 GLYPH<0> 2 9 | 1.2 | 50.1 | 2.0 |
| Cosmic Eye | 3.0733 | 28 | 3 : 41 + 0 : 16 GLYPH<0> 0 23 | 2 : 4 GLYPH<6> 0 : 06 | : 33 GLYPH<6> 0 : 6 | 2.2 | 46.3 | 3.3 |
:
Notes. All values correspond to intrinsic values corrected from magnification factors and come from Sklias et al. (2014) except for A2218-Mult. ( a ) Magnification factors derived from robust lens modeling (Richard et al. 2007, 2011). ( b ) Infrared luminosities integrated over the [8,1000] GLYPH<22> m interval. ( c ) Stellar masses and star formation rates as derived from the best energy conserving SED fits, obtained under the hypothesis of an extinction, A V, fixed at the observed ratio of L IR over L UV following the prescriptions of Schaerer et al. (2013). ( d ) Dust masses, dust temperatures, and rest-frame 850 GLYPH<22> m luminosities derived from a modified black-body fit applied to the Herschel photometry with the GLYPH<12> -slope fixed to 1.5 for T dust and to 1.8 for M dust and L GLYPH<23> (850 GLYPH<22> m). ( e ) IR luminosity corrected from the AGN contribution identified by the excess of the flux in the Herschel / PACS100 GLYPH<22> m band observed in the southern part of the MACS0451-arc, indicating very hot dust. A decomposition of the IR emission of the southern part between the AGN and stellar emission indicates a starburst-component contributing to about 45% of the total L IR (south) (for a detailed analysis, see Zamojski et al. in prep). ( f ) Redshift determined from weak absorption lines in the rest-frame UV instead of the H GLYPH<11> line.
Fig. 1. IR luminosities as a function of redshift of our lowL IR selected SFGs (squares) compared to our compilation of galaxies with CO measurements from the literature (see Sect. 2.2). Symbols in black refer to L IR as derived from the Herschel farIR photometry. Our SFGs populate the regime with the lower L IR reported at z > 1 5: they reach the : L ? and below domain as delimited by the hatched area showing the L ? limit of the Herschel far-IR luminosity function as determined by Magnelli et al. (2013, upper band) and Gruppioni et al. (2013, lower band) and extend the blank field studies to fainter luminosities as it is shown by the solid line which defines the minimal L IR at each redshift that produces a flux & 2 mJy in the Herschel / PACS160 GLYPH<22> m band ( GLYPH<24> 3 GLYPH<27> detection limit in GOODSN, Elbaz et al. 2011).
(4) have high-resolution HST images useful to get information on their morphology.
The four galaxies-A68-C0, A68-HLS115, MACS0451-arc, and A2218-Mult-selected for CO observations with the IRAM interferometer at the Plateau de Bure, France, do satisfy all the
Fig. 2. Location of our lowL IR selected SFGs (squares) in the star-formation-stellar-mass plane compared to our compilation of galaxies with CO measurements from the literature (see Sect. 2.2). Our (sub-) L ? galaxies probe one order of magnitude smaller stellar masses than the bulk of z > 1 SFGs from the literature. The blue dotted, red short-dashed, cyan long-dashed, and black solid lines are the best-fits of SFRM GLYPH<3> MS at z ' 0, 1.2, 2.2, and 3, respectively, parametrized by Eq. (1). The colorcoding of SFGs (our sample plus the z > 1 SFGs from the literature) refers to three redshift intervals: h z 1 2 : i = [1 ; 1 6] (red), : h z 2 2 : i = [2 ; 2 5] (cyan), and : h z 3 0 : i = [2 7 : ; 3 2] (black) in line with : the MS fits.
above specifications. The fifth object-A68-h7- observed with the IRAM 30 m telescope at Pico Veleta, Spain, has a higher intrinsic IR luminosity and hence has properties resembling more the high-redshift galaxies with CO measurements already available. A description of these selected targets at z > 1 5, as well : as their detailed multi-wavelength SED analysis from optical to far-IR sub-mm can be found in Sklias et al. (2014, except for /
A2218-Mult). In Table 1 we summarize their physical properties: spectroscopic redshifts ( z H GLYPH<11> ), magnification factors ( GLYPH<22> ), IR luminosities ( L IR ), stellar masses ( M GLYPH<3> ), star formation rates (SFR), dust masses ( M dust ), dust temperatures ( T dust ), and restframe 850 GLYPH<22> m luminosities ( L GLYPH<23> (850 GLYPH<22> m)). We add to this sample, MS 1512-cB58 (Yee et al. 1996, hereafter 'cB58') and the Cosmic Eye (Smail et al. 2007, hereafter 'Eye'), two well-known strongly-lensed galaxies with similar characteristics, for which we provide in Sklias et al. (2014) revised physical properties obtained from updated SED fitting and Herschel photometry.
In Fig. 1 we show the intrinsic IR luminosities of the lowL IR selected SFGs as a function of redshift and compare them to the compilation of galaxies with CO measurements from the literature (Sect. 2.2). Our sample of SFGs populates the regime with the lower L IR reported at z > 1 5 as expected from the : selection criterion and reach the L ? to subL ? domain according to the Herschel far-IR luminosity function of galaxies at z GLYPH<24> 1 5 : GLYPH<0> 3 with log( L ? = L GLYPH<12> ) varying between 11.4 and 12.3 (see the hatched area in Fig. 1 as delimited from Gruppioni et al. (2013) and Magnelli et al. (2013)). These L IR , derived from Herschel far-IR photometry, also fall below the Herschel detection limit and are only accessible with gravitational lensing. All the data points from the literature below the Herschel detection limit refer either to other lensed galaxies with Herschel L IR measurements or to galaxies with L IR measurements determined from their star formation rates via the Kennicutt (1998b) relation, L IR (L GLYPH<12> ) = 10 10 GLYPH<2> SFR (M GLYPH<12> yr GLYPH<0> 1 ), scaled to the Chabrier (2003) IMF.
Our sample of strongly-lensed galaxies also probes very small stellar masses M GLYPH<3> < 2 5 : GLYPH<2> 10 10 M GLYPH<12> (see Table 1), on average one order of magnitude smaller compared to the stellar masses of the bulk of z > 1 SFGs with CO measurements reported in the literature; making our sample of particular interest for the study of the molecular gas content in a new regime of smaller stellar masses in addition to the lower star formation rates. This is in line with the well-defined empirical SFRM GLYPH<3> relation. Our galaxies indeed nicely follow the main-sequence (MS) of z ' 1 2, 2.2, and 3 galaxies, respectively, as shown in : Fig. 2. We consider here the best-fit parametrization of the SFRM GLYPH<3> MS given by Tacconi et al. (2013) based on the samples of Bouch´ e et al. (2010), Noeske et al. (2007), Daddi et al. (2007), Rodighiero et al. (2010), and Salmi et al. (2012):
$$\text{SFR} _ { \text{MS} } ( z, M _ { * } ) \left ( M _ { \odot } \, \text{yr} ^ { - 1 } \right ) \, = \\ \text{45} ( M _ { * } / 6. 6 \times 1 0 ^ { 1 0 } \, ( M _ { \odot } ) ) ^ { 0. 6 5 } ( ( 1 + z ) / 2. 2 ) ^ { 2. 8 } \,, \quad \quad ( 1 ) \quad 3. \, \text{Obs}$$
which we compute at three redshifts, the redshift medians of our comparison sample of CO-detected z > 1 SFGs from the literature (Sect. 2.2), plus our sample of lowL IR selected galaxies, within three redshift intervals: h z 1 2 : i = [1 ; 1 6], : h z 2 2 : i = [2 ; 2 5], : and h z 3 0 : i = [2 7 : ; 3 2]. : No excess in the specific star formation rate is observed among our objects, they all lie within 0 3 : < sSFR sSFRMS = GLYPH<17> SFR = M GLYPH<3> SFRMS = M GLYPH<3> < 3, the admitted thickness of the main-sequence (e.g., Magdis et al. 2012b). On the other hand one galaxy-the Eye-is located below the MS at its respective redshift with sSFR = sSFRMS = 0 27. :
## 2.2. Comparison sample
In order to have a comparison sample to put face to face with our sample of lowL IR selected galaxies, we built up a large compilation of local and high-redshift galaxies with CO measurements reported in the literature. We refer to the same comparison sample throughout the paper. This compilation is exhaus- tive for star-forming galaxies that have the highest priority in the comparative work to our sample of galaxies, very complete for high-redshift sub-mm galaxies (SMGs), but remains incomplete for all local galaxies including the spiral galaxies, luminous IR galaxies (LIRGs), and ultra-luminous IR galaxies (ULIRGs). All these CO-detected galaxies are classified along the following sub-categories and are designated by the same symbols (but not by the same colors) in all figures.
- -Local spirals with L IR < 10 11 L GLYPH<12> (crosses): Helfer et al. (2003), Gao & Solomon (2004), Leroy et al. (2008, 2009), Papadopoulos et al. (2012), Wilson et al. (2012), and Bothwell et al. (2014).
- -Local LIRGs with 10 11 < L IR < 10 12 L GLYPH<12> (filled diamonds): Solomon et al. (1997), Gao & Solomon (2004), Wilson et al. (2008), Leech et al. (2010), and Papadopoulos et al. (2012).
- -Local ULIRGs with L IR > 10 12 L GLYPH<12> (small open circles): Solomon et al. (1997) and Papadopoulos et al. (2012).
- -High-redshift SMGs ULIRGs / at z > 0 2 with : L IR > 10 12 L GLYPH<12> (big open circles): Greve et al. (2005), Kneib et al. (2005), Iono et al. (2006), Tacconi et al. (2006), Daddi et al. (2009a,b), Knudsen et al. (2009), Weiss et al. (2009), Bothwell et al. (2010, 2013), Carilli et al. (2010), Harris et al. (2010), Ivison et al. (2010, 2011), Riechers et al. (2010, 2011), Swinbank et al. (2010), Yan et al. (2010), Braun et al. (2011), Casey et al. (2011), Combes et al. (2011, 2012, 2013), Danielson et al. (2011), Fu et al. (2012), Sharon et al. (2013), Rawle et al. (2014), and Magdis et al. (2014).
- -SFGs at z < 0 4 (open pentagons) and at : z > 1 (filled triangles) including field and lensed galaxies: Dannerbauer et al. (2009), Geach et al. (2009), Geach et al. (2011), Knudsen et al. (2009), Aravena et al. (2010, 2012, 2014), Genzel et al. (2010), Tacconi et al. (2010, 2013), Daddi et al. (2010a, 2014), Johansson et al. (2012), Magdis et al. (2012a,b), Magnelli et al. (2012), Bauermeister et al. (2013), Saintonge et al. (2013) , Tan et al. (2013), and 1 Magdis et al. (2014).
## 3. Observations, data reduction, and results
## 3.1. Plateau de Bure data
Table 2 summarizes the IRAM Plateau de Bure Interferometer (PdBI) observations of the galaxies-A68-C0, A68-HLS115, MACS0451-arc, and A2218-Mult-selected from the HLS with intrinsic L IR < 4 GLYPH<2> 10 11 L GLYPH<12> . The observations were conducted under typical summer conditions between June and September 2011 using 5 antennas in the compact D-configuration. The compact D-configuration provides the highest sensitivity with the smallest spatial resolution ( GLYPH<24> 5 5 : 00 at our typical tuned frequencies), which is ideal for detection projects. The frequencies were tuned to the expected redshifted frequency of the CO(2-1) or CO(3-2) line chosen according to the redshift of the targets in order to perform all the observations in the 3 mm band. On-source integration times were varying between 5 hours and 30 hours in
1 For the physical properties of the 8 o'clock arc (stellar mass and star formation rate), we use the values derived in Dessauges-Zavadsky et al. (2011).
Table 2. Observation summary
| Source | Coordinates | Coordinates | T on GLYPH<0> source | CO | GLYPH<23> obs a | Synthesized beam b | Synthesized beam b | Bandwidth c | rms d |
|--------------|---------------|---------------|------------------------|------|-------------------|------------------------|----------------------|---------------|---------|
| | RA (J2000) | DEC (J2000) | (hours) | line | (GHz) | Size ( 00 ) | PA ( GLYPH<14> ) | (MHz) | (mJy) |
| A68-C0 | 00:37:07.404 | + 09:09:26.57 | 5.2 | 2-1 | 89.131 | 6 : 41 GLYPH<2> 5 : 53 | 174 | 15 | 0.72 |
| A68-HLS115 | 00:37:09.503 | + 09:09:03.80 | 4.6 | 2-1 | 89.128 | 5 : 94 GLYPH<2> 5 : 79 | 111 | 15 | 0.71 |
| MACS0451-arc | 04:51:57.093 | + 00:06:10.44 | 30.2 | 3-2 | 114.768 | 5 : 45 GLYPH<2> 4 : 12 | 163 | 19 | 0.84 |
| A2218-Mult | 16:35:48.919 | + 66:12:13.81 | 9.2 | 3-2 | 84.258 | 6 : 35 GLYPH<2> 5 : 90 | 90 | 15 | 0.77 |
| A68-h7 e | 00:37:01.41 | + 09:10:22.31 | 5.5 | 3-2 | 109.777 | | | 24 | 0.94 |
Notes. ( a ) Observed line frequency used for tuning. ( b ) Beam resulting after combining all available imaging with natural weighting. These are the beam sizes and position angles displayed in Fig. 3. ( c ) Bandwidth achieved after resampling the PdBI data to a resolution of 50 km s GLYPH<0> 1 and the 30 m telescope data to a resolution of 65 km s GLYPH<0> 1 . Such resampling perfectly suits observed CO lines with full widths at half maximum > 200 km s GLYPH<0> 1 . ( d ) Noise per beam obtained for the specified bandwidth and averaged over the full 3.6 GHz spectral range after excluding channels where CO emission is detected. ( e ) Source observed at the 30 m single dish telescope, while all the other sources were observed with the PdBI.
Table 3. CO emission properties
| Source | z CO | CO line | S = N det a | FWHM CO b (km s GLYPH<0> 1 ) | F CO b (Jy km s GLYPH<0> 1 ) | L 0 CO(1 GLYPH<0> 0) c (10 9 K km s GLYPH<0> 1 pc 2 ) | M gas d (10 10 M GLYPH<12> ) | f gas d |
|--------------|--------|-----------|---------------|--------------------------------|--------------------------------|---------------------------------------------------------|--------------------------------|-----------|
| A68-C0 | 1.5854 | 2-1 | 11 | 334 GLYPH<6> 76 | 1 : 89 GLYPH<6> 0 : 28 | 2 : 6 GLYPH<6> 0 : 4 | 1.2 | 0.38 |
| A68-HLS115 | 1.5859 | 2-1 | 14 | 267 GLYPH<6> 18 | 2 : 00 GLYPH<6> 0 : 30 | 5 : 5 GLYPH<6> 0 : 9 | 2.4 | 0.75 |
| MACS0451-arc | 2.0118 | 3-2 | 4-5 | 261 GLYPH<6> 41 | 1 : 27 GLYPH<6> 0 : 32 | 1 : 0 GLYPH<6> 0 : 3 | 0.4 | 0.62 |
| A2218-Mult | 3.104 | 3-2 | undetected | - | < 0 : 62 e | < 3 : 8 | < 1 : 7 | |
| A68-h7 | 2.1529 | 3-2 | 3-4 | 282 GLYPH<6> 61 | 1 : 12 GLYPH<6> 0 : 22 | 16 : 9 GLYPH<6> 3 : 3 | 7.4 | 0.33 |
| MS1512-cB58 | 2.727 | 1-0 | | 174 | 0 : 052 GLYPH<6> 0 : 013 f | 0 : 6 GLYPH<6> 0 : 15 | 0.3 | 0.41 |
| Cosmic Eye | 3.074 | 1-0 | | 200 | 0 : 077 GLYPH<6> 0 : 013 f | 1 : 2 GLYPH<6> 0 : 2 | 0.5 | 0.18 |
Notes. ( a ) Signal-to-noise ratio of the CO emission detection. ( b ) Full widths half maximum and 'observed' CO(2-1) or CO(3-2) line integrated fluxes in Jy km s GLYPH<0> 1 and their 1 GLYPH<27> errors, as derived from fitting Gaussian function(s) to the observed CO profile. ( c ) Lensing-corrected CO(1-0) integrated line luminosities obtained with the following prescriptions: (1) the Solomon et al. (1997) formula L 0 CO(2 GLYPH<0> 1) CO(3 ; GLYPH<0> 2) (K km s GLYPH<0> 1 pc 2 ) = 3 25 : GLYPH<2> 10 7 F CO(2 GLYPH<0> 1) CO(3 ; GLYPH<0> 2) GLYPH<23> GLYPH<0> 2 obs D 2 L (1 + z ) GLYPH<0> 3 , where GLYPH<23> obs is the observed line frequency in GHz and D L is the luminosity distance in Mpc; (2) the correction factors r 2 1 ; = L 0 CO(2 GLYPH<0> 1) = L 0 CO(1 GLYPH<0> 0) = 0 81 and r3 1 : ; = L 0 CO(3 GLYPH<0> 2) = L 0 CO(1 GLYPH<0> 0) = 0 57 adopted for SFGs to account for the CO(2-1) and CO(3-2) transitions : being slightly sub-thermally excited (see Sect. 4); and (3) the lensing magnification correction, GLYPH<22> , given in Table 1. ( d ) Molecular gas masses, M gas = GLYPH<11> L 0 CO(1 GLYPH<0> 0) , obtained by assuming the 'Galactic' CO-H2 conversion factor X CO = 2 GLYPH<2> 10 20 cm 2 GLYPH<0> = (K km s GLYPH<0> 1 ), or GLYPH<11> = 4 36 M : GLYPH<12> = (K km s GLYPH<0> 1 pc 2 ) including the correction factor of 1.36 for helium. The molecular gas fraction is expressed as f gas = M gas = ( M gas + M GLYPH<3> ). Both quantities correspond to intrinsic values corrected from magnification factors. ( e ) 4 GLYPH<27> upper limit estimated by assuming a full width half maximum FWHMCO = 200 km s GLYPH<0> 1 for the undetected CO(3-2) line. ( f ) 'Observed' CO(1-0) integrated line fluxes and their 1 GLYPH<27> errors from Riechers et al. (2010).
the extreme case of MACS0451-arc. We used the WideX correlator that provides a continuous frequency coverage of 3.6 GHz in dual-polarization with a fixed channel spacing of 1.95 MHz resolution.
Standard data reduction was performed using the IRAM GILDAS software packages CLIC and MAP, where bandpass calibration was done using observations of calibrators that were best adapted to each of our targets individually. All data were mapped with the CLEAN procedure using the 'clark' deconvolution algorithm and combined with 'natural' weighting, resulting in synthesized beams listed in Table 2. The final noise per beam in all our cleaned, weighted images reaches an rms between 0.71 and 0.84 mJy over a resolution of 15 MHz (50 km s GLYPH<0> 1 ). The velocity-integrated maps of the CO emission are obtained by averaging the cleaned, weighted images over the spectral channels where emission is detected, and the corresponding spectra are obtained by spatially integrating the cleaned, weighted images over the 1 GLYPH<27> CO detection contours.
## 3.2. 30 m telescope data
Table 2 also summarizes the IRAM 30 m telescope observations of the HLS source-A68-h7-undertaken on September 35, 2011 under good summer conditions. We used the four single pixel heterodyne EMIR receivers, two centred on the E0 band (3 mm) and two on the E1 band (2 mm) tuned, respectively, to the redshifted frequencies of the CO(3-2) and CO(4-3) lines. The data were recorded using the WILMA autocorrelator providing a spectral resolution of 2 MHz. The observations were conducted in wobbler-switching mode, with a switching frequency of 0.5 Hz and a symmetrical azimuthal wobbler throw of 50 00 to maximize the baseline stability. Series of 12 ON OFF subscans / of 30 seconds each were performed and calibrations were repeated every 6 minutes. A total on-source integration time of 5.5 hours was obtained on this target.
The data reduction was completed with the IRAM GILDAS software package CLASS. The scans obtained with the two receivers tuned on the CO(3-2) line were averaged using the temporal scan length as weight. The resulting 3 mm spectrum was then Hanning smoothed to a resolution of 24 MHz (65 km s GLYPH<0> 1 ) and reaches an rms noise level of 0.94 mJy.
## 3.3. CO results
The CO emission has been successfully detected at the expected frequency for four out of the five lowL IR selected galaxies, it remains undetected in A2218-Mult. CO velocity-integrated maps are shown in the left panels of Fig. 3 for the three positive detections obtained with PdBI. In the right panels of Fig. 3 we show the CO contours overlaid on the HST images of the corresponding galaxies. None of these three CO detections is spatially re-
Fig. 3. From top to bottom, we plot A68-C0, A68-HLS115 and MACS0451-arc. Left panels: Velocity-integrated, cleaned maps of the CO(2-1) or CO(3-2) emission detected in three of our lowL IR selected galaxies observed with the PdBI. The maps are integrated over the cyan-shaded channels shown in Fig. 4. Contour levels start at GLYPH<6> 1 GLYPH<27> and are in steps of 1 GLYPH<27> . The size and orientation of the beam is indicated by the black ellipse in the bottom left corner. None of these three CO detections is spatially resolved. The cross in each pannel corresponds to the coordinates of the optical HST continuum position as listed in Table 2 and is GLYPH<6> 2 00 ( GLYPH<6> 17 kpc at z ' 2) in size. The coding of the color bar is in units of integrated flux Jy km s GLYPH<0> 1 . Right panels: CO(2-1) or CO(3-2) contours overlaid on the HST images in the F110W band for A68-C0, the F702W band for A68-HLS115, and the F140W band for MACS0451-arc. Contour levels start at GLYPH<6> 2 GLYPH<27> and are in steps of 2 GLYPH<27> , except for MACS0451-arc where they start at GLYPH<6> 1 GLYPH<27> and are in steps of 1 GLYPH<27> .
6
4
2
0
-2
-1000
-500
0
500
1000
-1000
-500
0
500
1000
8
6
4
2
0
-2
6
4
2
0
-2
-1000
-500
0
500
1000
Fig. 4. Spectra of the CO(2-1) or CO(3-2) emission line detected in our lowL IR selected galaxies, binned in steps of 50 km s GLYPH<0> 1 except for A68-h7 where bins have a resolution of 65 km s GLYPH<0> 1 . The cyan-shaded regions indicate channels where positive emission is detected. These channels have been used to derive total CO integrated fluxes and velocity-integrated maps shown in Fig. 3. The solid red lines are the best-fitting single or double Gaussian profiles to the observed CO line profiles. The zero velocity marked by the vertical bar corresponds to the tuning frequency of our targets listed in Table 2 and derived from the H GLYPH<11> line or rest-frame UV absorption lines redshifts.
solved, the observed spatial extension of the CO emission in our objects is similar to the PdBI beam size.
The resulting spectra of the CO detections can be found in Fig. 4. There is no evidence for continuum in any of our targets. The properties of the CO emission-velocity centroid ( z CO), the line full width half maximum (FWHMCO), and the observed integrated line flux ( F CO)-are evaluated by applying a single or double Gaussian fitting procedure on the observed line profiles based on the nonlinear GLYPH<31> 2 minimization and the LevenbergMarquardt algorithm. Errors on the values of z CO, FWHMCO, and F CO are estimated using a Monte Carlo approach, whereby the observed spectrum is perturbed with a random realization of the error spectrum and refitted. The process is repeated 1000 times and the error in each quantity is taken to be the standard deviation of the values generated by the 1000 Monte Carlo runs. The derived best-fitting results are shown in Fig. 4. In addition to the measure of CO line integrated fluxes from spectra, we have obtained independent integrated flux measurements for A68-C0, A68-HLS115 and MACS0451-arc by fitting either a circular or elliptical Gaussian model to the two-dimensional CO emission observed in velocity-integrated maps. The two respective methods lead to consistent integrated CO line fluxes within 1 GLYPH<27> errors.
Table 3 summarizes the measured CO emission properties. To convert the measured CO(2-1) and CO(3-2) luminosities to the fundamental CO(1-0) luminosity, which at the end gives a measure of the total H2 mass, we apply the luminosity correction factors r 2 1 ; = L 0 CO(2 GLYPH<0> 1) = L 0 CO(1 GLYPH<0> 0) = 0 81 : GLYPH<6> 0 20 : and r 3 1 ; = L 0 CO(3 GLYPH<0> 2) = L 0 CO(1 GLYPH<0> 0) = 0 57 : GLYPH<6> 0 15 to account for the lower : Rayleigh-Jeans brightness temperature of the 2-1 and 3-2 transitions relative to 1-0, as determined in Sect. 4. To estimate the uncertainties on the luminosities, we simply propagate the uncertainties on the CO line integrated fluxes. The molecular gas masses, M gas , and gas fractions, f gas = M gas = ( M gas + M GLYPH<3> ), correspond to values computed with the 'Galactic' CO-H2 con-
version factor X CO = 2 GLYPH<2> 10 20 cm GLYPH<0> 2 = (K km s GLYPH<0> 1 ), or GLYPH<11> = 4 36 M : GLYPH<12> = (K km s GLYPH<0> 1 pc 2 ) which includes the correction factor of 1.36 for helium. In the case of the CO(3-2) non-detection in A2218-Mult, we provide the 4 GLYPH<27> upper limits computed from the rms noise achieved in the PdBI observations of this galaxy and assume a typical full width half maximum FWHMCO = 200 km s GLYPH<0> 1 . The CO properties and inferred kinematics of the individual lowL IR selected targets are described in Appendix A.
## 4. CO luminosity correction factors
The measure of the molecular gas mass requires the luminosity of the fundamental CO(1-0) line, L 0 CO(1 GLYPH<0> 0) . Nevertheless, as soon as we are interested in objects at z > 0 4, accessing the : fundamental CO(1-0) line at 115.27 GHz becomes impossible with mm sub-mm receivers, and CO(1-0) must be replaced by / rotationally excited J > 1 CO transitions. Consequently, corrections for the ratio of the intrinsic Rayleigh-Jeans brightness temperatures in the 1-0 line to that in the rotationally excited line become necessary and need to be determined to access the fundamental L 0 CO(1 GLYPH<0> 0) . The corresponding CO luminosity correction factors vary with the J -transition of the CO line considered, the optical thickness, the thermal excitation and gas density of the medium, and hence the galactic type and redshift (Papadopoulos et al. 2012; Lagos et al. 2012; Narayanan et al. 2014).
Large e GLYPH<11> orts are done to get measurements of CO(1-0) and highJ CO lines within the same objects to determine the CO spectral line energy distribution (SLED) of local and high-redshift galaxies (e.g., Weiss et al. 2007; Dannerbauer et al. 2009; Papadopoulos et al. 2010; Daddi et al. 2014). The comparison sample of CO-detected galaxies from the literature (Sect. 2.2), particularly exhaustive at high redshift, offers a unique opportunity to obtain a comprehensive view on the CO luminosity correction factors, defined as r J ; 1 = L 0 CO( J GLYPH<0> ( J GLYPH<0> 1)) = L 0 CO(1 GLYPH<0> 0) , for the J = 2 and J = 3 CO rotational transitions. In Fig. 5 we plot the r2 1 ; and r3 1 ; correction factors as a function of the IR luminosity for galaxies from the comparison sample. The histograms show, respectively, r2 1 ; and r3 1 ; separately for z = 0 galaxies (open histograms) and z > 1 galaxies (hatched histograms), and for three L IR intervals: L IR < 10 11 L GLYPH<12> (the 'spiral' regime), 10 11 L GLYPH<12> < L IR < 10 12 L GLYPH<12> (the 'LIRG' regime), and L IR > 10 12 L GLYPH<12> (the 'ULIRG' regime).
The mean h r 2 1 ; i > 0 9 and : h r 3 1 ; i ' 0 55 : GLYPH<0> 0 6 values of : z = 0 galaxies within the 'spiral' and 'LIRG' regimes reveal relatively well-excited CO(2-1) and CO(3-2) lines. The observed luminosity correction factors are in line with the 1 & r 2 1 ; & r 3 1 ; & r 4 1 ; & ::: sequence which is expected if the average state is dominated by one warm optically thick and thermally excited phase (Papadopoulos et al. 2012). On the other hand, z = 0 galaxies within the 'ULIRG' regime show a predominance of a high-excitation phase with h r 3 1 ; i & h r 2 1 ; i and a mean h r 3 1 ; i close to unity. This suggests highly-excited media and optically thin CO SLEDs in the higher L IR z = 0 galaxies. This view is supported by the models of Lagos et al. (2012) that predict flatter CO SLEDs for the brightest IR galaxies than for the fainter IR counterparts at z = 0.
For z > 1 galaxies the available r 2 1 ; and r3 1 ; statistics is still small, except for r 3 1 ; within the 'ULIRG' regime. Both in the 'LIRG' and 'ULIRG' regimes, the mean h r 2 1 ; i ' 0 8 : and h r 3 1 ; i ' 0 6 values point toward a slightly lower-excitation : medium in high-redshift galaxies compared to z = 0 galaxies, where the higherJ CO lines are fainter as the subthermal excitation sets in more rapidly. The r3 1 ; correction factors within
Fig. 5. Bottom panel: Luminosity correction factors, r 2 1 ; = L 0 CO(2 GLYPH<0> 1) = L 0 CO(1 GLYPH<0> 0) (in red) and r3 1 ; = L 0 CO(3 GLYPH<0> 2) = L 0 CO(1 GLYPH<0> 0) (in black), plotted as a function of the IR luminosity for galaxies from our comparison sample (see Sect. 2.2). Upper, middle panels: r 2 1 ; and r3 1 ; distributions, respectively, of z = 0 galaxies (open histograms) and z > 1 galaxies (hatched histograms) plotted for three L IR intervals: from left to right L IR < 10 11 L GLYPH<12> (the 'spiral' regime), 10 11 L GLYPH<12> < L IR < 10 12 L GLYPH<12> (the 'LIRG' regime), and L IR > 10 12 L GLYPH<12> (the 'ULIRG' regime). The numbers in the upper right corner of each panel are the mean values of the r J ; 1 distributions ('thick' numbers refer to z = 0 galaxies and 'thin' numbers to z > 1 galaxies).
the 'ULIRG' regime also show no significant sign of evolution toward higher excitation in the higher L IR z > 1 galaxies, unlike what is observed in z = 0 galaxies. These results again are in a nice agreement with the models of Lagos et al. (2012) that predict (1) shallower CO SLEDs for high-redshift galaxies compared to their z = 0 counterparts at a fixed IR luminosity, and (2) smaller di GLYPH<11> erences in the CO SLEDs of faint- and bright-IR galaxies at z > 1 than for z = 0 galaxies. This is due to the increasing average gas kinetic temperature in molecular clouds with redshift. Given the fact that the observed highJ CO transitions of our sample of z > 1 galaxies are on average not highly excited, we may consider the 'Galactic' CO-H2 conversion factor as a sensible assumption for both SFGs and SMGs at z > 1.
Throughout the paper, we adopt the following luminosity correction factors for SFGs (including our lowL IR selected galaxies) at z > 1: r2 1 ; = L 0 CO(2 GLYPH<0> 1) = L 0 CO(1 GLYPH<0> 0) = 0 81 : GLYPH<6> 0 20 and : r 3 1 ; = L 0 CO(3 GLYPH<0> 2) = L 0 CO(1 GLYPH<0> 0) = 0 57 : GLYPH<6> 0 15. These are the means of : observed luminosity correction factors as measured at high redshift, instead of extrapolations from partial CO SLEDs or CO luminosity correction factors derived for local galaxies. Compared to the canonical value r3 1 ; = 0 5 assumed in the literature for : z > 1 SFGs (Tacconi et al. 2013; Saintonge et al. 2013), our r 3 1 ; value is 14% higher, but well within 1 GLYPH<27> . The di GLYPH<11> erence between our r2 1 ; value and the values found in the literature for
Fig. 6. IR luminosities as a function of CO(1-0) luminosities of our lowL IR selected SFGs (squares) compared to our compilation of galaxies with CO measurements from the literature (see Sect. 2.2). The color-coding of the various symbols refers to the J = 1, 2, and 3 CO transitions used to infer L 0 CO(1 GLYPH<0> 0) by applying the CO luminosity correction factors determined in Sect. 4. The best-fitting bisector linear relation to the entire sample of galaxies, log( L IR ) = (1 17 : GLYPH<6> 0 03) log( : L 0 CO(1 GLYPH<0> 0) ) + (0 28 : GLYPH<6> 0 23), is : plotted as a solid line. The z > 1 SFGs have on average a scatter in L IR as large as 1 dex at a given value of L 0 CO(1 GLYPH<0> 0) and a 1 GLYPH<27> dispersion of 0.38 dex in the y -direction about the best-fitting L IR -L 0 CO(1 GLYPH<0> 0) relation.
z > 1 SFGs is smaller: r2 1 ; = 0 84 in Daddi et al. (2010a) and : r 2 1 ; = 0 75 in Magnelli et al. (2012). The luminosity correction : factors we derived for both high-redshift SFGs and SMGs also well agree with the values compiled by Carilli & Walter (2013), although based on other prescriptions for the z > 1 SFGs.
## 5. LowL IR selected galaxies in the general context of galaxies with CO measurements
We discuss the overall physical properties derived from the new CO measurements achieved in our lowL IR selected galaxies. We combine and compare their CO luminosities, IR luminosities, star formation e GLYPH<14> ciencies, and molecular gas depletion timescales to our compilation of CO-detected galaxies from the literature (Sect. 2.2). The goal is to investigate whether the extended dynamical range toward lower star formation rates SFR < 40 M GLYPH<12> yr GLYPH<0> 1 and smaller stellar masses M GLYPH<3> < 2 5 : GLYPH<2> 10 10 M GLYPH<12> shows evidence for new trends and correlations.
## 5.1. CO luminosity to IR luminosity relation
It is now well established that there is a relation between L IR and L 0 CO(1 GLYPH<0> 0) (see the review by Carilli & Walter 2013) and this despite the uncertainties on L 0 CO(1 GLYPH<0> 0) mostly inferred from higher J > 1 CO rotational transitions and on L IR often derived from various star formation tracers (UV, H GLYPH<11> , mid-IR, submm, or radio) instead of direct far-IR photometry. The first high-
Fig. 7. Star formation e GLYPH<14> ciencies, SFE = L IR = L 0 CO(1 GLYPH<0> 0) (left yaxis), or molecular gas depletion timescales, t depl = M gas = SFR, assuming a 'Galactic' CO-H2 conversion factor for all galaxies (right y-axis), as a function of IR luminosities of our lowL IR selected SFGs (squares) compared to our compilation of galaxies with CO measurements from the literature (see Sect. 2.2). This illustrates in another way the large dispersion about the L IR -L 0 CO(1 GLYPH<0> 0) relation observed in Fig. 6.
redshift datasets showed evidence for a 'bimodal' L IR -L 0 CO(1 GLYPH<0> 0) behaviour between the so-called 'sequence of disks' or 'MS starforming galaxies' which includes local spirals plus high-redshift SFGs, and the 'sequence of starbursts' or 'mergers' which includes local ULIRGs plus high-redshift SMGs ULIRGs (e.g., / Daddi et al. 2010b; Genzel et al. 2010; Sargent et al. 2014).
The results for the most recent compilation of CO-detected galaxies from the literature (Sect. 2.2), including our sample of lowL IR selected SFGs, are shown in Fig. 6. The color-coding allows to track the original J = 1, 2, or 3 CO transition used to compute the CO(1-0) luminosity. Our new sources populate a new regime of low IR luminosities, L IR < 4 GLYPH<2> 10 11 L GLYPH<12> , and low CO(1-0) luminosities. They overlap with the domain of local LIRGs, their L IR counterparts at z = 0, and perfectly extend the observed L IR versus L 0 CO(1 GLYPH<0> 0) distribution of previously studied z > 1 SFGs with higher IR luminosities. They show an increased scatter toward lower L 0 CO(1 GLYPH<0> 0) at given values of L IR . This is a general trend which is highlighted by the new compilation of CO-detected SFGs at z > 1, in contrast to what was previously observed, where all the high-redshift SFGs' CO luminosities were characterized by higher values at given L IR compared to the local ULIRG and high-redshift SMG ULIRG populations / (Daddi et al. 2010b; Genzel et al. 2010; Carilli & Walter 2013).
As a result, a single linear relation now best reproduces the distribution within the L IR -L 0 CO(1 GLYPH<0> 0) log plane of the various COdetected galaxies spanning 5 orders of magnitude in the IR luminosity, having redshifts between z = 0 and 5.3, and sampling diverse galaxy types from main-sequence galaxies to mergers. The best-fitting bisector linear relation has a slope of 1 17 : GLYPH<6> 0 03. : The 'bimodal' behaviour is clearly smeared out: the o GLYPH<11> set between the di GLYPH<11> erent galaxy populations is embedded in the large dispersion of data points about the linear L IR -L 0 CO(1 GLYPH<0> 0) relation in
log space. Indeed, the z > 1 SFGs only 2 show a scatter in L IR as large as 1 dex at a given value of L 0 CO(1 GLYPH<0> 0) and a 1 GLYPH<27> dispersion of 0.3 dex in the y -direction about the best-fit of their L IR -L 0 CO(1 GLYPH<0> 0) relation. The o GLYPH<11> set of 0.46-0.5 dex in the normalization between the 'sequence of disks' and the 'sequence of starbursts' reported by Daddi et al. (2010a) and Sargent et al. (2014) hence is within 1 5 : GLYPH<27> dispersion of the current sample of z > 1 SFGs.
## 5.2. Star formation efficiency and gas depletion timescale: What determines their behaviour at high redshift ?
Another way to represent the CO luminosity-IR luminosity relation, that may help understand the dispersion in this relation, is through the star formation e GLYPH<14> ciency, SFE, defined as SFE = L IR = L 0 CO(1 GLYPH<0> 0) , or equivalently SFE = SFR = M gas by assuming for all galaxies a 'Galactic' CO-H2 conversion factor GLYPH<11> = 4 36 M : GLYPH<12> = (K km s GLYPH<0> 1 pc 2 ) which includes the correction factor of 1.36 for helium. The star formation e GLYPH<14> ciency is intimately linked to the molecular gas depletion timescale defined as the inverse of the SFE, t depl = M gas = SFR. This physical parameter describes how long each galaxy could sustain star formation at the current rate before running out of fuel, assuming that the gas reservoir is not replenished.
In Fig. 7 we plot the star formation e GLYPH<14> ciency and the molecular gas depletion timescale as a function of the IR luminosity for our compilation of galaxies with CO measurements from the literature (Sect. 2.2), including our sample of lowL IR selected SFGs. This shows: (1) High-redshift SFGs and SMGs have very comparable star formation e GLYPH<14> ciency distributions and spreads. Their respective mean SFE (very similar to their respective median SFE) at z > 1 are h log(SFE) i SFGs = 1 89 : GLYPH<6> 0 33 L : GLYPH<12> = (K km s GLYPH<0> 1 pc 2 ) ( h t depl i SFGs = 560 Myr) and h log(SFE) i SMGs ULIRGs = = 2 14 : GLYPH<6> 0 30 : L GLYPH<12> = (K km s GLYPH<0> 1 pc 2 ) ( h t depl i SMGs = 315 Myr); they are the same within 1 GLYPH<27> dispersion. As a result, z > 1 SFGs have not their SFE confined to the low values of local spirals and z > 1 SMGs ULIRGs do not / systematically show an excess in L IR = L 0 CO(1 GLYPH<0> 0) , in contrast to the reported 'bimodality' (Daddi et al. 2010a,b; Genzel et al. 2010; Sargent et al. 2014). (2) The IR luminosity hence appears as a weak tracer of the star formation e GLYPH<14> ciency, although a correlation between L IR (or SFR) and SFE remains, as the slope of the L IR versus L 0 CO(1 GLYPH<0> 0) relation in log space is not exactly equal to unity (it is equal to 1.17, see Fig. 6).
In what follows, we focus on z > 1 SFGs with the aim to try to understand what drives their large spread in SFE and what di GLYPH<11> erentiates galaxies with high SFE from those with low SFE. Various physical parameters may induce di GLYPH<11> erences in the star formation e GLYPH<14> ciencies, or respectively molecular gas depletion timescales. We investigate below the link between the SFE (or t depl ) and the following physical parameters: specific star formation rate, stellar mass, redshift, o GLYPH<11> set from the main-sequence, and compactness of the starburst.
## 5.2.1. Specific star formation rate
Apossible physical parameter which may control the SFE or t depl of galaxies is the specific star formation rate, sSFR = SFR = M GLYPH<3> ,
2 Those include our sample of strongly-lensed lowL IR galaxies, the BzK galaxies from Daddi et al. (2010a), the PHIBSS (EGS, BM BX) / galaxies from Tacconi et al. (2010, 2013) and Genzel et al. (2010), the PEP galaxies from Magnelli et al. (2012), and strongly-lensed SFGs from Saintonge et al. (2013) and others (see Sect. 2.2).
Fig. 8. Molecular gas depletion timescales, t depl = M gas = SFR, as a function of specific star formation rates, sSFR = SFR = M GLYPH<3> , plotted for our lowL IR selected SFGs (squares) and our comparison sample of SFGs and local spirals with CO measurements from the literature (see Sect. 2.2). The molecular gas depletion timescales are computed assuming a 'Galactic' CO-H2 conversion factor for all galaxies. t depl shows a strong dependence on sSFR. The best-fitting bisector linear relation to the z = 0 COLD GASS sample from Saintonge et al. (2011) is plotted as a dashed blue line. z > 1 SFGs are displaced with respect to the local relation, their best-fitting bisector linear relation, log( t depl ) = ( GLYPH<0> 0 83 : GLYPH<6> 0 08) log(sSFR) : + (1 43 : GLYPH<6> 0 70), : is plotted as a solid black line. The color-coding of SFGs (our sample plus z > 1 SFGs from the literature) refers to three redshift intervals: h z 1 2 : i = [1 ; 1 6] (red), : h z 2 2 : i = [2 ; 2 5] (cyan), and : h z 3 0 : i = [2 7 : ; 3 2] (black). :
i.e. the SFR normalized by the stellar mass which gives the timescale of formation of all the stellar mass in a galaxy at the given present SFR. In their study of z = 0 massive star-forming galaxies with 10 0 : < log( M GLYPH<3> = M ) GLYPH<12> < 11 5 from the COLD : GASS survey, Saintonge et al. (2011) found the strongest dependence of t depl to be precisely on the sSFR. Their t depl -sSFR anti-correlation in fact highlights comparable timescales for the gas consumption and the stellar mass formation. They interpret it as the result of the increased dynamical stirring as one proceeds to more strongly star-forming galaxies, since in local starburst galaxies the very high values of sSFR are achieved because dynamical disturbances act to compress the available interstellar medium atomic gas and create giant molecular clouds and stars.
We also find a good anti-correlation between t depl and sSFR for z > 1 SFGs as shown in Fig. 8. We overplot the best-fit relation obtained for the COLD GASS sample and observe that z > 1 SFGs are generally distributed above the z = 0 relation and have longer t depl by about 0.75 dex than local galaxies with the same sSFR (see the best-fitting relation obtained for z > 1 SFGs). The observed t depl -sSFR anti-correlation and its shift with redshift are in agreement with the 'bathtub' model predictions (Bouch´ e et al. 2010) and confirm the findings by Saintonge et al. (2011) and Combes et al. (2013). They attribute this displacement of z = 1 GLYPH<0> 3 star-forming galaxies with re-
spect to the local galaxy population to their significantly larger molecular gas fractions that a GLYPH<11> ord longer molecular gas depletion times at a given value of sSFR. We do confirm in Sect. 6 the increased available molecular gas reservoir as one proceeds toward higher redshifts, as well as, on average, as one proceeds toward more strongly star-forming galaxies (higher sSFR), which certainly triggers the t depl -sSFR anti-correlation observed at z > 1. Nevertheless, while a clear evolution of the t depl -sSFR relation appears to be present between local and z > 1 galaxies, no clear trend seems to emerge between the three redshift bins h z 1 2 : i = [1 ; 1 6], : h z 2 2 : i = [2 ; 2 5], and : h z 3 0 : i = [2 7 : ; 3 2]. This : agrees with the steep increase of the molecular gas fraction between z = 0 and z GLYPH<24> 1 followed by a quasi non-evolution toward higher redshifts (but with a large scatter) as observed in Fig. 13.
In addition to the shift of z > 1 SFGs with respect to local galaxies due to their larger molecular gas fractions, the specific star formation rates of galaxies in the local Universe are sealed on low values because of the accumulation of more and more old stars in their bulge at z = 0. These old stars have an important weight in the total stellar mass budget, but none on the current SFR and thus imply low sSFR values. As a result, this accentuates the displacement between the respective distributions of local and high-redshift galaxies in the t depl -sSFR plane.
## 5.2.2. Stellar mass
Another physical parameter which may trigger di GLYPH<11> erences in SFE or t depl from galaxy to galaxy is the stellar mass. In their analysis of z = 0 massive star-forming galaxies from the COLD GASS survey, Saintonge et al. (2011) could observe an increase in t depl by a factor of 6 over the stellar mass range of 10 10 M GLYPH<12> to 10 11 5 : M , from about 0.7 Gyr to 4 Gyr. They assign this GLYPH<12> t depl -M GLYPH<3> correlation to the more bursty star formation history of low-mass galaxies which leads to enhanced SFE, inversely reduced t depl , due to minor starburst events produced either by weak mergers associated with distant tidal encounters, variations in the intergalactic medium accretion rate, or secular processes within galactic disks. On the other hand, morphological quenching and feedback in high-mass galaxies prevent the molecular gas from forming stars, while at the same time not destroying the gas leads to an increased reservoir of molecular gas. This dependence of t depl on M GLYPH<3> in local SFGs has been recently confirmed over the full stellar mass range down to M GLYPH<3> = 10 9 M GLYPH<12> by the ALLSMOG survey from Bothwell et al. (2014).
The increase of the molecular gas depletion timescale with the stellar mass is also observed in our compilation of z > 1 SFGs with CO measurements from the literature, including our lowL IR selected SFGs, as shown in Fig. 9. It is confirmed with a Kendall's tau probability of 0.0017 per cent and a slope M 0 5 : GLYPH<3> . On average, z > 1 SFGs have shorter t depl than local galaxies with the same M GLYPH<3> . This t depl -M GLYPH<3> correlation in order to be compatible with the t depl -sSFR anti-correlation discussed above (Sect. 5.2.1), implies that a galaxy at z > 1 with a given M GLYPH<3> necessarily has its SFR higher than a galaxy at z = 0 with the same M GLYPH<3> . This is in line with the SFRM GLYPH<3> MS relation (see Fig. 2).
The increase in t depl by a factor of 10 over the stellar mass range of 10 9 4 : M GLYPH<12> to 10 11 5 : M , from about 0.15 Gyr to 1.5 Gyr, GLYPH<12> we observe in our sample of z > 1 SFGs, questions the constant t depl of 0.7 Gyr found by Tacconi et al. (2013) in their sample of z = 1 GLYPH<0> 3 SFGs with stellar masses confined to M GLYPH<3> > 10 10 4 : M . GLYPH<12> In Fig. 9, we may see that the t depl -M GLYPH<3> correlation at high redshift is triggered by galaxies with small stellar masses. This resembles the first z = 0 surveys that were unable to highlight the t depl
Fig. 9. Molecular gas depletion timescales, t depl = M gas = SFR, as a function of stellar masses plotted for our lowL IR selected SFGs (squares) and our comparison sample of SFGs and local spirals with CO measurements from the literature (see Sect. 2.2). The molecular gas depletion timescales are computed assuming a 'Galactic' CO-H2 conversion factor for all galaxies. A t depl -M GLYPH<3> correlation emerges for z > 1 SFGs; their best-fitting bisector linear relation, log( t depl ) = (0 52 : GLYPH<6> 0 16) log( : M GLYPH<3> ) + (3 25 : GLYPH<6> 1 74), : is plotted as a solid black line. The best-fitting bisector linear relation to the z = 0 COLD GASS sample from Saintonge et al. (2011) is shown by the dashed blue line.
correlation with M GLYPH<3> because of their limited dynamical range in M GLYPH<3> . Indeed, in the smaller range of galaxies explored in the THINGS sample, Bigiel et al. (2008) and Leroy et al. (2008) initially found a constant molecular gas depletion timescale.
If true, such a correlation between t depl and M GLYPH<3> observed both in local and z > 1 SFGs has two important implications. First, it is in contradiction with cosmological hydrodynamic simulations by Dav´ e et al. (2011) which predict the opposite, namely an anti-correlation between M GLYPH<3> and t depl , where t depl drops to larger stellar masses roughly as M GLYPH<0> 0 3 : GLYPH<3> 3 . Second, it refutes the linearity of the Kennicutt-Schmidt relation, i.e. the proportionality of SFR and molecular gas mass, such that GLYPH<6> SFR / GLYPH<6> N gas with N , 1. This is fundamental, as it contradicts one of the hypotheses of the 'bathtub' model that assumes a constant molecular gas depletion timescale, and hence a linear Kennicutt-Schmidt relation (e.g., Bouch´ e et al. 2010; Lilly et al. 2013; Dekel & Mandelker 2014). Therefore, we may wonder whether the t depl -M GLYPH<3> correlation really exists, or whether we simply see a constant t depl embedded in a very large spread. Only getting more molecular gas depletion timescale measurements for z > 1 galaxies with small stellar masses will bring a definitive answer.
A way to reconcile the current empirical finding with a constant t depl is to invoke the possible CO-H2 conversion factor dependence on metallicity (e.g., Isreal 1997; Leroy et al. 2011; Feldmann et al. 2012; Genzel et al. 2012). Indeed, it is ex-
3 The dependence of t depl on M GLYPH<0> 0 3 : GLYPH<3> comes from the star formation relation assumed ( GLYPH<6> SFR / GLYPH<6> 1 4 : gas ) and the empirical relation measured in Dav´ e et al. (2011) simulations of GLYPH<6> gas / M 3 4 = GLYPH<3> .
pected that galaxies with lower stellar masses have lower metallicities, according to the mass-metallicity relation (see Sect. 1), and would hence require higher CO-H2 conversion factors to be applied that would, in reality, lead to longer molecular gas depletion timescales in low stellar mass galaxies in comparison to what we get when assuming a 'Galactic' CO-H2 conversion factor. However, whether this is su GLYPH<14> cient to compensate the t depl dependence on M 0 6 : GLYPH<3> needs to be further tested.
## 5.2.3. Redshift
Since the star-forming galaxies span a large interval of the Hubble time, the redshift could also be at the origin of their large SFE or t depl dispersion. Models do predict a cosmic evolution of the molecular gas depletion timescale for main-sequence galaxies (e.g., Hopkins & Beacom 2006; Dav´ e et al. 2011, 2012). The temporal scaling of t depl can be most easily understood when using the formulation of the star formation relation given by SFR / M gas = t dyn , where t dyn is the dynamical time of the star formation region (see also Bouch´ e et al. 2010; Genel et al. 2010). This then gives t depl / t dyn , which in a canonical disk model scales as (1 + z ) GLYPH<0> 1 5 : (Mo et al. 1998).
The decrease in t depl with redshift is supported by observations, as already reported by Combes et al. (2013), Tacconi et al. (2013), Saintonge et al. (2013), and Santini et al. (2014). In Fig. 10 we plot the molecular gas depletion timescale as a function of redshift for our compilation of local spirals and SFGs with CO measurements from the literature, including our lowL IR selected SFGs. Four redshift bins are considered h z 0 2 : i = [0 055 0 4], : ; : h z 1 2 : i = [1 ; 1 6], : h z 2 2 : i = [2 ; 2 5], : and h z 3 0 : i = [2 7 : ; 3 2], their respective mean values are : h log( t h z 0 2 : i depl ) i = 8 94 : GLYPH<6> 0 18 (870 Myr), : h log( t h z 1 2 : i depl ) i = 8 79 : GLYPH<6> 0 25 (620 Myr), : h log( t h z 2 2 : i depl ) i = 8 72 : GLYPH<6> 0 43 (520 Myr), and : h log( t h z 3 0 : i depl ) i = 8 27 : GLYPH<6> 0 33 (190 Myr) for a 'Galactic' CO-H2 : conversion factor. This confirms a shallow t depl decrease from z = 0 to z = 3 2, in agree-: ment with the expected (1 + z ) GLYPH<0> 1 5 : dependence. As a result, high redshift SFGs do form stars with a much higher SFE, and consume the molecular gas over a much shorter timescale, than local SFGs. The t depl spread and dispersion per redshift bin are significant, they result from the t depl -sSFR and t depl -M GLYPH<3> relations discussed above (Sects. 5.2.1 and 5.2.2). Indeed, these relations induce a gradient in t depl per redshift bin, such that galaxies with the higher sSFR and smaller M GLYPH<3> have the shorter t depl .
## 5.2.4. Offset from the main-sequence
Apositive empirical correlation between SFE and the o GLYPH<11> set from the main-sequence, sSFR sSFRMS, was found by Magdis et al. = (2012b), Saintonge et al. (2012), and Sargent et al. (2014), suggesting higher SFE for galaxies with enhanced sSFR and, in particular, for galaxies with sSFR = sSFRMS GLYPH<29> 3, namely beyond the accepted MS thickness, and this independent of galaxies' redshift. In Fig. 11 we show the SFE of our lowL IR selected galaxies and z > 1 SFGs from the literature (Sect. 2.2) as a function of their o GLYPH<11> set from the MS. The main-sequence, SFRMS( z ; M GLYPH<3> ), is defined here by Eq. (1) and is computed at three redshifts covering three redshift intervals h z 1 2 : i = [1 ; 1 6], : h z 2 2 : i = [2 ; 2 5], : and h z 3 0 : i = [2 7 : ; 3 2] of the SFGs. While the enlarged sample of : z > 1 SFGs confirms the general trend of higher SFE for galaxies with larger o GLYPH<11> sets from the MS (Kendall's tau probability of 0.025 per cent), the MS galaxies themselves, with o GLYPH<11> sets from the MS restricted to 0 3 : < sSFR sSFRMS = < 3 (80% of the sam-
Fig. 10. Molecular gas depletion timescales, t depl = M gas = SFR, as a function of redshift plotted for our lowL IR selected SFGs (squares) and our comparison sample of SFGs and local spirals with CO measurements from the literature (see Sect. 2.2). The molecular gas depletion timescales are computed assuming a 'Galactic' CO-H2 conversion factor for all galaxies, but for z < 0 4 and : z > 1 SFGs (pentagons, triangles, and squares) the dotted lines, in addition, show the interval of possible t depl values as determined with two extreme CO-H2 conversion factors GLYPH<11> = 4 36 ('Galactic' value) and 1.1 (local ULIRG value). A : redshift evolution of t depl is observed in agreement with the expected 1 5 : GLYPH<2> (1 + z ) GLYPH<0> 1 5 : dependence of Dav´ e et al. (2012), where the normalization is set to the typical depletion time of 1.5 Gyr observed in local galaxies (solid line). Four redshift bins are considered, h z 0 2 : i = [0 055 0 4], : ; : h z 1 2 : i = [1 ; 1 6], : h z 2 2 : i = [2 ; 2 5], : and h z 3 0 : i = [2 7 : ; 3 2], : for the t depl means shown by the large black crosses with their 1 GLYPH<27> dispersion.
ple), have roughly a constant SFE given the Kendall's tau probability of 0.23 per cent with a large dispersion of 0.32 dex, reaching SFE values even higher than those of galaxies beyond the accepted MS thickness. Consequently, the large spread in SFE observed among MS galaxies over 1.7 orders of magnitude cannot be attributed to the thickness of the SFRM GLYPH<3> relation, i.e. the relative position of galaxies with respect to the main-sequence.
Saying things on the other way, these results tell us that it is not the star formation e GLYPH<14> ciency of MS galaxies that drives the thickness of the SFRM GLYPH<3> relation, although the SFE certainly contributes to it in some way within its large spread. What physical parameter then dominates and triggers this thickness ? In Fig. 12 we show that the strongest dependence of the o GLYPH<11> set from the MS is observed on M gas = M GLYPH<3> , the molecular gas mass over the stellar mass ratio, with a Kendall's tau probability of 0 per cent and this over two orders of magnitude. This confirms the findings by Magdis et al. (2012b) and Sargent et al. (2014), who support that the thickness of the SFRM GLYPH<3> relation at any redshift is driven by the variations of the molecular gas fractions of MSgalaxies rather than by variations of their star formation e GLYPH<14> -ciencies. According to Magdis et al. (2012b), this prevalence of M gas = M GLYPH<3> implies that MS galaxies can have higher SFR (within the thickness of the MS) mainly because they have more raw molecular gas material to produce stars, which hence favours the
Fig. 11. Star formation e GLYPH<14> ciencies, SFE = L IR = L 0 CO(1 GLYPH<0> 0) (left yaxis), or molecular gas depletion timescales, t depl = SFR = M gas , assuming a 'Galactic' CO-H2 conversion factor for all galaxies (right y-axis), as a function of galaxies' o GLYPH<11> sets from the mainsequence, sSFR sSFRMS, plotted for our low-= L IR selected SFGs (squares) and our comparison sample of z > 1 SFGs with CO measurements from the literature (triangles). The shaded area defines the accepted thickness of the MS, i.e. a scatter about the SFRM GLYPH<3> relation between 0 3 : < sSFR sSFRMS = < 3. The enlarged sample of z > 1 SFGs confirms the general trend of higher SFE for galaxies with larger o GLYPH<11> sets from the MS with a Kendall's tau probability of 0.025 per cent. However, considering MS galaxies only within the shaded area, their SFE appear roughly constant, given the Kendall's tau probability of 0.23 per cent with a large dispersion of 0.32 dex.
existence of a global star-formation relation L IR (or SFR)M gas for all MS galaxies (see Fig. 6).
## 5.2.5. Compactness of the starburst
Downes & Solomon (1998) were the first to point out that the compactness of the nuclear starburst regions was linked to the star formation e GLYPH<14> ciency in extreme local ULIRGs. Combes et al. (2013) found, in this sense, a weak trend for SFE to be anticorrelated with the half-light radii of high-redshift galaxies, R 1 2 = , or correlated with their compactness. The five SFGs with the highest SFE from our compilation of z > 1 SFGs, including our lowL IR selected galaxies, have R 1 2 = ranging between 1.5 and 4 kpc with one object having R 1 2 = as large as 8 kpc. These half-light radii are typical of high-redshift star-forming galaxies (Tacconi et al. 2013), and thus do not highlight a particular compactness among SFGs with the highest SFE. Certainly, we have to increase the number of available data, in parallel with the angular resolution of CO measurements to confirm or a GLYPH<14> rm the role of the starburst's compactness in the SFE (or t depl ) spread.
## 5.2.6. Summary
The answer to the question 'what determines the behaviour of the star formation e GLYPH<14> ciency and molecular gas depletion
Fig. 12. Molecular gas mass over stellar mass ratios, M gas = M GLYPH<3> , as a function of galaxies' o GLYPH<11> sets from the main-sequence, sSFR sSFRMS, plotted for our low-= L IR selected SFGs (squares) and our comparison sample of z > 1 SFGs with CO measurements from the literature (triangles). The shaded area defines the accepted thickness of the MS as above. A clear correlation is observed with a Kendall's tau probability of 0 per cent.
timescale in star-forming galaxies at high redshift' is not straightforward from what we have discussed above. The dependence of SFE (or t depl ) on five physical parameters has been investigated. While the dependence appears to be the strongest on the specific star formation rate, there is, in addition, a trend for a correlation with the stellar mass and a clear evolution with redshift. The dependences on the o GLYPH<11> set from the main-sequence and the compactness of the starburst are less clear. Consequently, there is a tight interplay between the SFE (or t depl ) and, respectively, the specific star formation rate, the stellar mass, and the redshift of galaxies, which implies that not only one single physical parameter drives the observed large spread in SFE of z > 1 SFGs, but a combination of all of them.
In addition, an intrinsic spread in SFE and t depl certainly exists due to other, more complex, and hence more di GLYPH<14> cult to test, e GLYPH<11> ects. For example, some galaxies very likely are not in a quasisteady state equilibrium, because the accretion rate is not constant, in contrast to what is assumed in the ideal 'bathtub' model where the accretion rate and the star formation rate timescale are supposed to vary slowly enough. And obviously, environment and mergers, as well as other variability e GLYPH<11> ects, like an episodic star formation, probably imply a not so smooth galaxy evolution.
## 6. Molecular gas fraction of high-redshift galaxies
The molecular gas fraction, defined as f gas = M gas = ( M gas + M GLYPH<3> ), is a very important physical quantity to know, but also very di GLYPH<14> -cult to access, since dependent on the CO-H2 conversion factor. We can also express the molecular gas fraction as:
$$f _ { \text{gas} } = \frac { 1 } { 1 + ( t _ { \text{dep} } s S F R ) ^ { - 1 } } \,,$$
meaning that it is closely linked to the molecular gas depletion timescale and the specific star formation rate. Observations show
Fig. 13. Molecular gas fractions, f gas = M gas = ( M gas + M GLYPH<3> ), as a function of redshift plotted for our lowL IR selected SFGs (squares) and our comparison sample of SFGs and local spirals with CO measurements from the literature (see Sect. 2.2). The molecular gas fractions are computed assuming a 'Galactic' CO-H2 conversion factor for all galaxies, but for z < 0 4 and : z > 1 SFGs (pentagons, triangles, and squares) the dotted lines, in addition, show the interval of possible f gas values as determined with two extreme CO-H2 conversion factors GLYPH<11> = 4 36 : ('Galactic' value) and 1.1 (local ULIRG value). Four redshift bins are considered, h z 0 2 : i = [0 055 0 4], : ; : h z 1 2 : i = [1 ; 1 6], : h z 2 2 : i = [2 ; 2 5], : and h z 3 0 : i = [2 7 : ; 3 2], : for the f gas means shown by the large black crosses with their 1 GLYPH<27> dispersion. A net rise of f gas is observed from h z 0 2 : i to h z 1 2 : i , followed by a very mild increase toward higher redshifts. The color-coding refers to four stellar mass intervals: log( M GLYPH<3> = M ) GLYPH<12> < 10 (black), 10 2 : < log( M GLYPH<3> = M ) GLYPH<12> < 10 7 (red), 10 7 : : < log( M GLYPH<3> = M ) GLYPH<12> < 11 (green), and log( M GLYPH<3> = M ) GLYPH<12> > 11 (blue), as defined in Sect. 6.2.
a clear trend of increasing f gas with sSFR (Tacconi et al. 2013), but with a large dispersion (larger than the systematic uncertainties in sSFR and f gas measurements) induced by the di GLYPH<11> erent molecular gas depletion timescales found among z > 1 SFGs, ranging between 0.15 Gyr and 1.5 Gyr (see Sect. 5.2).
The various physical processes at play in the evolution of galaxies (e.g., accretion, star formation, and feedback) have direct impact on the behaviour of their molecular gas fraction. Consequently, a solid way to test galaxy evolution models is to confront their predictions with the observed f gas behaviour. We consider below two main observables, the redshift and the stellar mass, and investigate the respective redshift evolution and stellar mass dependence of f gas in high-redshift SFGs.
## 6.1. Redshift evolution of the molecular gas fraction
The cosmic evolution of the molecular gas fraction with the Hubble time directly results from the expansion of the Universe itself. New matter falling on galaxies through accretion from filaments comes from further out as the time progresses and, consequently, high angular momentum matter settles in the outer parts of galaxies. This leads to the size growth of galaxies with
Fig. 14. Molecular gas fractions, f gas = M gas = ( M gas + M GLYPH<3> ), as a function of stellar masses plotted for our lowL IR selected SFGs (squares) and our comparison sample of SFGs and local spirals with CO measurements from the literature (see Sect. 2.2). The molecular gas fractions are computed assuming a 'Galactic' CO-H2 conversion factor for all galaxies. Four stellar mass bins are considered 9 4 : < log( M GLYPH<3> = M ) GLYPH<12> < 9 9, 10 2 : : < log( M GLYPH<3> = M ) GLYPH<12> < 10 7, 10 7 : : < log( M GLYPH<3> = M ) GLYPH<12> < 11 0, and 11 0 : : < log( M GLYPH<3> = M ) GLYPH<12> < 11 5. The corresponding : f gas means with their 1 GLYPH<27> dispersion of z > 1 SFGs are shown by the large black crosses and of z < 0 4 : SFGs by the large green crosses. The M GLYPH<3> dependence of f gas sustains an upturn of the molecular gas fraction at the lowM GLYPH<3> end, a mild decrease toward higher stellar masses of SFGs at z > 1, and a shift toward lower f gas values for SFGs at lower redshifts. The color-coding of SFGs (our sample plus z > 1 SFGs from the literature) refers to three redshift intervals: h z 1 2 : i = [1 ; 1 6] (red), : h z 2 2 : i = [2 ; 2 5] (cyan), and : h z 3 0 : i = [2 7 : ; 3 2] (black). :
the cosmic time, such that galaxies were more compact in the early Universe than today. A finding first proposed by dark matter simulations (e.g., Fall & Efstathiou 1980; Mo et al. 1998), more recently by cosmological hydrodynamic simulations including cold gas accretion (Stewart et al. 2013), and now observationally confirmed (e.g., Bouwens et al. 2004; Trujillo et al. 2006; Buitrago et al. 2008). The size growth of galaxies has a direct impact on the H2 / HI ratio evolution. Indeed, the pressure on the midplane of galaxies strongly depends on the gas surface density, in the way that a smaller radius increases the gas surface density. Since the H2 / HI ratio is strongly correlated to this pressure, for the same gas mass the high-redshift galaxies, which are smaller (more compact) and hence have a higher internal pressure, will have a higher H2 / HI ratio and that means a higher H2 content than local galaxies (e.g., Obreschkow & Rawlings 2009; Lagos et al. 2011, 2014). This together with the trend for high-redshift galaxies to be intrinsically more gas-rich, i.e. have a higher total gas mass (H I + H2), leads to even larger di GLYPH<11> erence between high-redshift and local galaxies and hence to a net increase of the molecular gas fraction with redshift.
From Eq. (2) we see that the evolution of the molecular gas fraction depends on the evolution of both t depl and sSFR. We have shown in Sect. 5.2.3 that the molecular gas depletion timescale evolves with (1 + z ) GLYPH<0> 1 5 : . The 'bathtub' model, in par-
allel, predicts a continuing rise of the specific star formation rate, generally driven by the redshift evolution of the average accretion rate (inflow), which scales as (1 + z ) GLYPH<11> with GLYPH<11> varying between 5 3 and 3 according to the prescriptions considered (Bouch´ e et / al. 2010; Dav´ e et al. 2012; Lilly et al. 2013). The steeper rise of the specific star formation rate with respect to the decline of the molecular gas depletion timescale leads to the steady increase of f gas with redshift.
Observationally, the cosmic evolution of the molecular gas fraction with the Hubble time, in the way that higher redshift galaxies are more gas-rich, is now well established (e.g., Daddi et al. 2010a; Geach et al. 2011; Tacconi et al. 2010, 2013; Saintonge et al. 2013; Sargent et al. 2014; Carilli & Walter 2013). In Fig. 13 we show f gas as a function of redshift for our compilation of local spirals and SFGs with CO measurements from the literature, including our lowL IR selected SFGs. We consider four redshift bins h z 0 2 : i = [0 055 0 4], : ; : h z 1 2 : i = [1 ; 1 6], : h z 2 2 : i = [2 ; 2 5], : and h z 3 0 : i = [2 7 : ; 3 2]. : The respective mean values h f h z 0 2 : i gas i = 0 19 : GLYPH<6> 0 07, : h f h z 1 2 : i gas i = 0 46 : GLYPH<6> 0 15, : h f h z 2 2 : i gas i = 0 45 : GLYPH<6> 0 22, and : h f h z 3 0 : i gas i = 0 51 : GLYPH<6> 0 23 for a 'Galactic' CO-H2 : conversion factor show a net rise of the molecular gas fraction from h z 0 2 : i to h z 1 2 : i , followed by a very mild increase toward higher redshifts, as found in earlier studies. If further confirmed, such an f gas redshift evolution does not really correspond to a steady redshift increase of f gas as predicted by the 'bathtub' model. It requires a di GLYPH<11> erent redshift-dependence of either t depl or sSFR (see Eq. (2)). The sSFR redshift evolution parametrised by Lilly et al. (2013) proposes a steep sSFR increase as (1 + z ) 3 out to z = 2 and a slow down to (1 + z ) 5 3 = at z > 2. This model is among the closest to the observations, since it predicts a flattening of the f gas evolution beyond z = 2 (see also Saintonge et al. 2013).
The spread and dispersion of the molecular gas fractions per redshift bin are significant, in particular for z > 1 SFGs. As discussed in Carilli & Walter (2013), a large f gas dispersion at a given redshift is expected: it is mainly due to the strong dependence of f gas on the stellar mass (shown in Sect. 6.2), which should produce an f gas gradient per redshift bin, such that galaxies with the smaller M GLYPH<3> have the larger f gas (see Fig. 14). To highlight the e GLYPH<11> ect of this dependence, the color-coding of the data points in Fig. 13 refers to four stellar mass bins. We may see that the data points within the lowest M GLYPH<3> bin (log( M GLYPH<3> = M ) GLYPH<12> < 10) do show the larger f gas per redshift bin. The expected f gas gradient with M GLYPH<3> per redshift bin for galaxies within the other M GLYPH<3> bins is more scattered, but we still see its e GLYPH<11> ect on the z < 0 4 SFGs : which have the smallest dispersion because they cover the two higher M GLYPH<3> bins only. The stellar mass is not the only physical parameter inducing an f gas dispersion per redshift bin. The star formation rate plays an important role too, since f gas increases with SFR as a consequence of the Kennicutt-Schmidt relation (see the fundamental f gas -M GLYPH<3> -SFR relation proposed by Santini et al. (2014)), as well as environmental e GLYPH<11> ects (outflows, disk hydrostatic pressure, etc.).
## 6.2. Stellar mass dependence of the molecular gas fraction
Bouch´ e et al. (2010), Dav´ et al. (2011), and Lagos et al. (2014), e all show a drop in the molecular gas fraction with increasing stellar mass and an upturn of f gas at the lowM GLYPH<3> end. The location and strength of that upturn depends on the model adopted and, in particular, on whether any outflow is considered. A comparable M GLYPH<3> dependence of the molecular gas fraction is predicted for both local and high-redshift galaxies, but with a shift in f gas toward higher values at higher redshifts, a shift that results from the redshift evolution of f gas (see Fig. 13). While in a closed-box model the molecular gas fraction is expected to decrease with the stellar mass because of the gas conversion into stars, this turns out to be similar in the 'bathtub' model as the modelled galaxies approach a steady state where the SFR, which is proportional to the gas mass, represents the rate at which gas is turned into stars, follows the net gas accretion rate dictated by cosmology plus a gas outflow rate (Dekel & Mandelker 2014).
The M GLYPH<3> dependence of f gas at z > 1 has already been observationally constrained by Tacconi et al. (2013) and Santini et al. (2014) for galaxies at the intermediate high-/ M GLYPH<3> end 10 4 : < log( M GLYPH<3> = M ) GLYPH<12> < 12 (see also Magdis et al. 2012b; Sargent et al. 2014). We provide for the first time insights on f gas of highredshift SFGs at the lowM GLYPH<3> end 9 4 : < log( M GLYPH<3> = M ) GLYPH<12> < 9 9. : Our new sample of lowL IR selected SFGs doubles the number of galaxies at z > 1 (3 out of a total of 6) with achieved f gas measurements at the lowM GLYPH<3> end. In Fig. 14 we show the resulting f gas as a function of stellar mass. We divide the sample of z > 1 SFGs with intermediate high-/ M GLYPH<3> in three stellar mass bins 10 2 : < log( M GLYPH<3> = M ) GLYPH<12> < 10 7, 10 7 : : < log( M GLYPH<3> = M ) GLYPH<12> < 11 0, : and 11 0 : < log( M GLYPH<3> = M ) GLYPH<12> < 11 5, each containing a compara-: ble number of galaxies. The respective mean values h f h M GLYPH<3> 10 5 : i gas i = 0 48 : GLYPH<6> 0 17, : h f h M GLYPH<3> 10 9 : i gas i = 0 43 : GLYPH<6> 0 15, and : h f h M GLYPH<3> 11 1 : i gas i = 0 40 : GLYPH<6> 0 15 : for a 'Galactic' CO-H2 conversion factor show a mild decrease of f gas with stellar mass within a very large dispersion. On the other hand, the mean h f h M GLYPH<3> 9 7 : i gas i = 0 69 : GLYPH<6> 0 18 value of the lowest : stellar mass bin 9 4 : < log( M GLYPH<3> = M ) GLYPH<12> < 9 9 highlights an upturn : of f gas at the lowM GLYPH<3> end of SFGs at z > 1 as predicted. More data in the lowM GLYPH<3> end, tackling the f gas upturn, may help disentangling specific outflow feedback wind recipes. / /
The mean values of f gas within the two higher stellar mass bins defined above were also computed for SFGs at z < 0 4 and : are plotted in Fig. 14. They nicely show the shift of the M GLYPH<3> dependence toward smaller molecular gas fractions that is expected for star-forming galaxies at lower redshifts, because of the redshift evolution of f gas (Sect. 6.1). This evolution is also partly responsible for the large dispersion observed within each M GLYPH<3> bin of z > 1 SFGs, as shown by the color-coding of the data points of z > 1 SFGs highlighting three redshift bins.
The combination of the redshift evolution and M GLYPH<3> dependence of the molecular gas fraction shows not only that the average f gas of star-forming galaxies increases with redshift (Sect. 6.1), but that this increase is even more substantial for low stellar mass galaxies than for the high stellar mass ones given the upturn of f gas at the lowM GLYPH<3> end for z > 1 SFGs. A behaviour judged by Bouch´ e et al. (2010) and Santini et al. (2014) as being a direct result of the 'downsizing' scenario, which claims that more massive galaxies formed earlier and over a shorter period of time (e.g., Cowie et al. 1996; Heavens et al. 2004; Jimenez et al. 2007; Thomas et al. 2010), i.e., consume their molecular gas more quickly. Consequently, massive galaxies have already consumed most of their gas at high redshifts, while less massive galaxies still have a large fraction of molecular gas.
## 7. Dust-to-gas ratio
The Herschel / PACS + SPIRE and longer wavelength data of our lowL IR selected SFGs allow accessing their dust masses (see Table 1 and Sklias et al. 2014). When combined with CO emission data, we may hence investigate another key physical parameter that is the dust-to-gas mass ratio, GLYPH<14> DGR = M dust = M gas . This is even more important, as the dust-to-gas mass ratio and
dust masses start to be used to estimate the CO-H2 conversion factor both in local and high-redshift galaxies (e.g., Leroy et al. 2011; Magdis et al. 2011, 2012b; Magnelli et al. 2012) and to determine molecular gas masses when CO measurements are not available (Santini et al. 2014; Scoville et al. 2014).
It is now accepted both from observations and interstellar medium evolution models that GLYPH<14> DGR scales linearly with the oxygen abundance (e.g., Issa et al. 1990; Dwek 1998; Edmunds 2001; Inoue 2003; Draine et al. 2007; Leroy et al. 2011; Saintonge et al. 2013; Chen et al. 2013). Nevertheless, the measure of the dust-to-gas mass ratio, in particular, in highredshift galaxies remains very uncertain, because of a number of important assumptions that are done:
- 1. The dust mass derivation from the far-IR sub-mm SED is / tributary to several assumptions, such as the dust model, the dust emissivity index GLYPH<12> within the modified black-body (MBB) fits, and the dust mass absorption cross section. The freedom on these parameters can lead to large variations in the M dust estimates. Magnelli et al. (2012) already pointed out a factor of 3 di GLYPH<11> erence between dust masses as inferred from MBB fits and the Draine & Li (2007) model.
- 2. A CO-H2 conversion factor needs to be assumed to determine the molecular gas mass. Since the CO-H2 conversion factor is suspected to vary with metallicity, this could induce a false GLYPH<14> DGR-metallicity dependence.
- 3. At high redshift we consider M H2 GLYPH<29> M HI, or equivalently M gas GLYPH<25> M H2 = M dust =GLYPH<14> DGR. This is supported by the high molecular gas fractions measured in z > 1 SFGs (see Fig. 13), which leave little room for a significant atomic gas component within the total mass budget as inferred from dynamical analysis (Daddi et al. 2010a; Tacconi et al. 2010) and theoretical arguments like those discussed in Sect. 6.1 (e.g., Obreschkow & Rawlings 2009; Lagos et al. 2011, 2014). However, no direct H I measurements exist in highredshift galaxies so far.
To alleviate the uncertainty on the above first assumption, Scoville et al. (2014) proposed to consider the long-wavelength far-IR sub-mm continuum as a dust mass tracer, since the long-/ wavelength Rayleigh-Jeans tail of dust emission is generally optically thin and thus provides a direct probe of the total dust. They chose 850 GLYPH<22> mas the fiducial wavelength and measured, in a homogeneous way, GLYPH<14> DGR GLYPH<25> L GLYPH<23> (850 GLYPH<22> m) = M gas in the Galaxy, local spirals, local ULIRGs, and high-redshift SMGs by assuming for all a spectral slope GLYPH<12> = 1 8 for the far-IR sub-mm dust : / emission flux density as determined from the extensive Planck data throughout the Galaxy (Planck Collaboration 2011a,b) and a 'Galactic' CO-H2 conversion factor 4 .
Following the Scoville et al. (2014) prescriptions, we add to their sample our lowL IR selected SFGs (see their rest-frame 850 GLYPH<22> m luminosities in Table 1) and the compilation of z > 1 SFGs with CO measurements from the literature (Sect. 2.2) for which published far-IR sub-mm / photometry is available (Magdis et al. 2012b; Saintonge et al. 2013). In addition, in order to bypass e GLYPH<11> ects linked to the possible dust-to-gas ratio dependence on metallicity, we consider only the star-forming galaxies with near-solar metallicities Z Z = GLYPH<12> > 0 8. The metallicities : are estimated using the mass-metallicity relation at z GLYPH<24> 2 2 : from Erb et al. (2006) recalibrated by Maiolino et al. (2008)
4 Scoville et al. (2014) provide a solid justification of why using the 'Galactic' CO-H2 conversion factor in local ULIRGs and high-redshift SMGs instead of the much smaller GLYPH<11> = 1 1 factor often proposed for : these galaxies.
Fig. 15. Dust emission at rest-frame 850 GLYPH<22> m per unit mass of gas, L GLYPH<23> (850 GLYPH<22> m) = M gas , plotted for our lowL IR selected SFGs (squares) and our comparison sample of z > 1 SFGs with CO measurements from the literature and published far-IR sub-mm / photometry (triangles). Only SFGs with metallicities Z Z = GLYPH<12> > 0 8 are considered. These results are compared with the compi-: lation from Scoville et al. (2014) for the SMGs at z = 1 4 : GLYPH<0> 3 and local spirals and ULIRGs. We assume a 'Galactic' CO-H2 conversion factor for all objects and M gas ' M H2 for high-redshift galaxies. All galaxies have a similar L GLYPH<23> (850 GLYPH<22> m) = M gas within 1 GLYPH<0> 2 GLYPH<27> , with a trend toward a lower dust-to-gas ratio in z > 1 SFGs by GLYPH<24> 0 33 dex at fixed near-solar metallicity, as shown : by the lines corresponding, respectively, to the L GLYPH<23> (850 GLYPH<22> m) = M gas means of local spirals and ULIRGs (blue dotted line), highredshift SMGs (green dashed line), and z > 1 SFGs (red solid line).
and scaled to the Chabrier (2003) IMF, when direct metallicity measurements from nebular emission lines are not available. 12 out of 73 CO-detected SFGs at z > 1 (including our lowL IR selected SFGs) satisfy the above prescriptions. Similarly to Scoville et al., we adopt the 'Galactic' CO-H2 conversion factor, which in the case of these high-redshift SFGs is reasonable given their near-solar metallicities (and see also Sect. 4). The L GLYPH<23> (850 GLYPH<22> m) = M gas results of these z > 1 SFGs are shown in Fig. 15 as a function of L GLYPH<23> (850 GLYPH<22> m) and are compared to local spirals, local ULIRGs, and high-redshift SMGs from Scoville et al. The respective means and dispersions are:
$$\langle$$
$$\begin{array} { r c l } t { - } r a m e \\ f _ { \, z \, > \, 1 } & \langle L _ { \nu } ( 8 5 0 \mu ) / M _ { \text{gas} } \rangle _ { \text{SFGs} } & = \, 4. 6 6 ^ { + 3. 0 3 } _ { - 1. 8 4 } \times 1 0 ^ { 1 9 } \, \text{erg} \, s ^ { - 1 } \, H z ^ { - 1 } \, M _ { \odot } ^ { - 1 } \\ \text{ct. 2.2} \\ \text{vailable} \\ \text{in order} & \langle L _ { \nu } ( 8 5 0 \mu ) / M _ { \text{gas} } \rangle _ { \text{SMGs} } & = \, 9. 5 4 ^ { + 6. 0 0 } _ { - 3. 6 8 } \times 1 0 ^ { 1 9 } \, \text{erg} \, s ^ { - 1 } \, H z ^ { - 1 } \, M _ { \odot } ^ { - 1 } \\ \text{end} \\ \text{depen-} & \quad \, \bullet \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \delta \, \end{array}$$
Measurements in local galaxies (spirals and ULIRGs), z = 1 4 : GLYPH<0> 3 SMGs, and z > 1 SFGs, at fixed near-solar metallicity ( Z = Z GLYPH<12> > 0 8), show similar : L GLYPH<23> (850 GLYPH<22> m) = M gas means within 1 GLYPH<0> 2 GLYPH<27> dispersion. While the means of local galaxies and highredshift SMGs are nearly the same, z > 1 SFGs sustain a trend for a shift toward a lower L GLYPH<23> (850 GLYPH<22> m) = M gas mean by GLYPH<24> 0 33 dex : with a large dispersion of 0.22 dex. Such a shift in z > 1 SFGs was already reported by Saintonge et al. (2013, by about 0.23 dex) when compared to Local Group galaxies (Leroy et
al. 2011). If the dust-to-gas ratio truly is lower in high-redshift SFGs, this may suggest that a smaller fraction of dust grains is produced for the same metallicity under the specific conditions prevailing in the interstellar medium of these galaxies in comparison with local galaxies and high-redshift SMGs.
The data thus support a non-universal GLYPH<14> DGR. Hence, deriving molecular gas masses from direct CO measurements remains highly recommended, as CO still appears a better tracer of the molecular gas mass, despite the uncertainty linked with the COto-H2 conversion factor, than dust.
## 8. Summary and conclusions
We report new CO observations performed with the IRAM PdBI and 30 m telescope for five strongly-lensed star-forming galaxies at z GLYPH<24> 1 5 : GLYPH<0> 3. These galaxies were selected from the Herschel Lensing Survey for their low intrinsic IR luminosities and high magnification factors. Four of them have IR luminosities as low as L IR < 4 GLYPH<2> 10 11 L GLYPH<12> (SFR < 40 M GLYPH<12> yr GLYPH<0> 1 ) and reach the L ? to subL ? domain. While this regime is typical of SFGs, the molecular gas content of such galaxies has been poorly explored so far, because of mm sub-mm instrumental sensitivity limitations. / Thanks to the gravitational lensing, we achieve three CO emission detections in L ? to subL ? SFGs and a fourth in an SFG with a slightly higher L IR . These SFGs not only are among galaxies with the lower L IR with accessible CO luminosities known, but also with the smaller stellar masses M GLYPH<3> < 2 5 : GLYPH<2> 10 10 M . GLYPH<12> We add to this sample, cB58 and the Eye, two well-known strongly-lensed galaxies characterized by similar low L IR and small M GLYPH<3> properties, as derived by Sklias et al. (2014) from revised Herschel photometry and SED fitting. To put these SFGs in the general context of all CO-detected galaxies, we built up a large comparison sample of local and high-redshift galaxies with COmeasurements reported in the literature that is exhaustive for high-redshift SFGs, complete for high-redshift SMGs, and only indicative for local spirals, LIRGs, and ULIRGs.
This comparison sample allows us to obtain good estimates of the CO luminosity correction factors r2 1 ; and r3 1 ; for, respectively, the J = 2 and J = 3 CO rotational transitions for both z = 0 and z > 1 galaxies distributed over three L IR intervals. Two main trends pop out in agreement with the models of Lagos et al. (2012): (1) shallower CO SLEDs for high-redshift galaxies compared to their z = 0 counterparts at a fixed IR luminosity, and (2) smaller di GLYPH<11> erences in the CO SLEDs of faint- and bright-IR galaxies at z > 1 than for z = 0 galaxies. The inferred means of the observed CO luminosity correction factors valid for SFGs at z > 1 are r2 1 ; = 0 81 : GLYPH<6> 0 20 and r3 1 : ; = 0 57 : GLYPH<6> 0 15. :
The combination of the overall physical properties derived from the new CO measurements achieved in our SFGs covering a new dynamical range of SFR < 40 M GLYPH<12> yr GLYPH<0> 1 and M GLYPH<3> < 2 5 : GLYPH<2> 10 10 M GLYPH<12> with the comparison sample of CO-detected galaxies from the literature leads to the following main results.
- 1. A single linear relation within the L IR -L 0 CO(1 GLYPH<0> 0) log plane is observed, which best-fit gives a slope of 1.17. The current larger sample of z > 1 SFGs shows an increased 1 GLYPH<27> dispersion of 0.3 dex in the y-direction about the best-fit of their L IR -L 0 CO(1 GLYPH<0> 0) relation, which is su GLYPH<14> cient to hide the reported bimodal behaviour between the 'sequence of disks' and the 'sequence of starbursts'.
- 2. Another way to represent the L IR -L 0 CO(1 GLYPH<0> 0) relation is through the star formation e GLYPH<14> ciency, defined as SFE = L IR = L 0 CO(1 GLYPH<0> 0) GLYPH<17> SFR = M gas = 1 = t depl . SFGs and SMGs at z > 1 show very similar SFE distributions and large spreads;
the SFE of z > 1 SFGs are not confined to the low values of local spirals any more. The investigation of the SFE and t depl dependence on several physical parameters leads us to conclude that it is the combination of the specific star formation rate, stellar mass, and redshift that drives the large spread in SFE of z > 1 SFGs. The respective o GLYPH<11> set of SFGs from the main-sequence within the thickness of the SFRM GLYPH<3> relation, as well as the compactness of the starburst seem to play a minor role in the observed SFE spread.
- 3. The strongest dependence of t depl is observed on the specific star formation rate both for z > 1 SFGs and local galaxies. SFGs at z > 1 show longer t depl by about 0.75 dex than local galaxies with the same sSFR. This displacement of the t depl -sSFR relation with redshift is attributed to the larger molecular gas fractions found in z > 1 SFGs that a GLYPH<11> ord longer molecular gas depletion times at a given value of sSFR. In addition, the sSFR of z = 0 galaxies are sealed on low values, because of the accumulation of more and more old stars in their bulge that have an important weight in the total stellar mass budget, but none on the present SFR.
- 4. Acorrelation between the molecular gas depletion timescale and stellar mass is observed thanks to the enlarged dynamical range of stellar masses down to M GLYPH<3> = 10 9 4 : M GLYPH<12> sampled by our lowL IR selected SFGs. Although it needs to be confirmed with additional z > 1 SFGs with small stellar masses, a t depl increase with M GLYPH<3> is seen in z = 0 galaxies over the M GLYPH<3> range from 10 9 M GLYPH<12> to 10 11 5 : M . If true, this GLYPH<12> t depl -M GLYPH<3> correlation observed both in local galaxies and z > 1 SFGs is opposed to the constant molecular gas depletion timescale assumed in the 'bathtub' model and refutes the linearity of the Kennicutt-Schmidt relation.
- 5. We confirm the (1 + z ) GLYPH<0> 1 5 : evolution of the molecular gas depletion timescale for main-sequence galaxies as predicted by various models. The mean t depl drops from 870 Myr at h z 0 2 : i = [0 055 0 4] down to 190 Myr at : ; : h z 3 0 : i = [2 7 : ; 3 2] : with a significant dispersion per redshift bin. This dispersion results from the t depl -sSFR and t depl -M GLYPH<3> relations, which induce a gradient of t depl per redshift bin, such that galaxies with the higher sSFR and smaller M GLYPH<3> have the shorter t depl .
- 6. Observationally, it is now well established that high-redshift galaxies are gas-rich. What remains debated is the steady increase of the molecular gas reservoir with redshift predicted by models. While we do observe a net rise of the mean f gas from 0 19 : GLYPH<6> 0 07 : up to 0 46 : GLYPH<6> 0 15 : between h z 0 2 : i = [0 055 0 4] and : ; : h z 1 2 : i = [1 ; 1 6], it : is followed by a very mild increase toward higher redshifts. A significant f gas dispersion is seen per redshift bin, it is due to the strong dependence of the molecular gas fraction on the stellar mass, which we start testing observationally.
- 7. An f gas drop with increasing M GLYPH<3> because of the gas conversion into stars, as well as an f gas upturn at the lowM GLYPH<3> end whose strength is dependent on the outflow are predicted. z > 1 SFGs show a mild f gas decrease between 10 10 4 : < M GLYPH<3> = M GLYPH<12> < 10 11 5 : . The large dispersion per stellar mass bin comes from the f gas redshift evolution, as nicely illustrated by z < 0 4 SFGs that have their : M GLYPH<3> dependence shifted toward lower f gas . We provide the first insights on f gas of z > 1 SFGs at the lowM GLYPH<3> end 10 9 4 : < M GLYPH<3> = M GLYPH<12> < 10 9 9 : . The corresponding mean h f gas i = 0 69 : GLYPH<6> 0 18 value supports : an f gas upturn. This shows that the average f gas of SFGs not only increases with redshift, but this increase is even more substantial for low M GLYPH<3> galaxies than for the high M GLYPH<3> ones, a behaviour resulting from the 'downsizing' scenario.
- 8. To alleviate the uncertainties linked to the dust mass estimates, we consider the rest-frame 850 GLYPH<22> m continuum as a dust mass tracer and compare the dust-to-gas mass ratios, GLYPH<14> DGR GLYPH<25> L GLYPH<23> (850 GLYPH<22> m) = M gas , inferred in a homogeneous way, of z > 1 SFGs to those of local spirals, local ULIRGs, and high-redshift SMGs. While the L GLYPH<23> (850 GLYPH<22> m) = M gas means of local galaxies and high-redshift SMGs are nearly the same, the one of high-redshift SFGs scatters systematically by a factor of two at fixed near-solar metallicity ( Z = Z GLYPH<12> > 0 8). : z > 1 SFGs sustain a shift toward a lower L GLYPH<23> (850 GLYPH<22> m) = M gas mean by 0.33 dex with a very large dispersion of 0.22 dex. This supports a non-universal GLYPH<14> DGR. Thus, deriving molecular gas masses from direct CO measurements remains highly recommended.
Acknowledgements. We are very grateful to M´ elanie Krips, Philippe Salom´ e, Claudia del P. Lagos, and Nicolas Bouch´ e for helpful discussions, advises, and adapted model predictions. We thank the IRAM sta GLYPH<11> of both the Plateau de Bure Interferometer and the 30 m telescope for the high-quality data acquired and for their support during observations and data reduction. This work was supported by the Swiss National Science Foundation. JPK acknowledges support from the ERC advanced grant LIDA and JR from the ERC starting grant CALENDS.
## References
Agertz, O., Teyssier, R., & Moore, B. 2009, MNRAS, 397, L64
- Aravena, M., Carilli, C., Daddi , E., et al. 2010, ApJ, 718, 177
- Aravena, M., Carilli, C. L., Salvato, M., et al. 2012, MNRAS, 426, 258
- Aravena, M., Hodge, J. A., Wagg, J., et al. 2014, MNRAS, 442, 558
- Baker, A. J., Tacconi, L. J., Genzel, R., Lehnert, M. D., & Lutz, D. 2004, ApJ, 604, 125
Bauermeister, A., Blitz, L., Bolatto, A., et al. 2013, ApJ, 768, 132
- Bigiel, F., Leroy, A., Walter, F., Brinks, E., de Blok, W. J. G., Madore, B., & Thornley, M. D. 2008, AJ, 136, 2846
- Bigiel, F., Leroy, A., Walter, F., et al. 2011, ApJL, 730, L13
- Bournaud, F., Dekel, A., Teyssier, R., et al. 2011, ApJ 730, 4
- Bouwens, R. J., Illingworth, G. D., Blakeslee, J. P., Broadhurst, T. J., & Franx, M. 2004, ApJ, 611, L1
- Bothwell, M. S., Chapman, S. C., Tacconi, L., et al. 2010, MNRAS, 405, 219
Bothwell, M. S., Smail, I., Chapman, S. C., et al. 2013, MNRAS, 429, 3047
- Bothwell, M. S., Wagg, J., Cicone, C., et al. 2014, MNRAS, 44 5, 2599
- Bouch´ e, N., Dekel, A., Genzel, R., et al. 2010, ApJ, 718, 1001
- Braun, R., Popping, A., Brooks, K., & Combes, F. 2011, MNRAS, 416, 2600 Buitrago, F., Trujillo, I., Conselice, C. J., Bouwens, R. J., Dickinson, M., & Yan,
- H. 2008, ApJ, 687, L61
- Carilli, C. L., Daddi, E., Riechers, D., et al. 2010, ApJ, 714, 1407
- Casey, C. M., Chapman, S. C., Neri, R., et al. 2011, MNRAS, 415, 2723 Chabrier, G. 2003, PASP, 115, 763
- Carilli, C. L., & Walter, F. 2013, ARA&A, 51, 105
- Chen, B., Dai, X., Kochanek, C. S., & Chartas, G. 2013, ApJ, submitted [arXiv:1306.0008]
- Combes, F., Garc´ ıa-Burillo, S., Braine, J., Schinnerer, E., Walter, F., & Colina, L. 2011, A&A, 528, 124
- Combes, F., Rex, M., Rawle, T. D., et al. 2012, A&A, 538, L4
- Combes, F., Garc´ ıa-Burillo, S., Braine, J., Schinnerer, E., Walter, F., & Colina, L. 2013, A&A, 550, 41
- Coppin, K. E. K., Swinbank, A. M., Neri, R., et al. 2007, ApJ, 665, 936
- Cowie, L. L., Songaila, A., Hu, E. M., & Cohen, J. G. 1996, AJ, 112, 839
- Daddi, E., Dickinson, M., Morrison, G., et al. 2007, ApJ, 670, 156
- Daddi, E., Dannerbauer, H., Stern, D., et al. 2009a, ApJ, 694, 1517
- Daddi, E., Dannerbauer, H., Krips, M., Walter, F., Dickinson, M., Elbaz, D., & Morrison, G. E. 2009b, ApJ, 695, L176
- Daddi, E., Bournaud, F., Walter, F., et al. 2010a, ApJ, 713, 686
- Daddi, E., Elbaz, D., Walter, F., et al. 2010b, ApJL, 714, L118
- Daddi, E., Dannerbauer, H., Liu, D., et al. 2014, [arXiv:1409.8158]
- Danielson, A. L. R., Swinbank, A. M., Smail, I., et al. 2011, MNRAS, 410, 1687
- Dannerbauer, H., Daddi, E., Riechers, D. A., et al. 2009, ApJ, 698, L178
- Dav´ e, R., Finlator, K., & Oppenheimer, B. 2011, MNRAS, 416, 1354
- Dav´ e, R., Finlator, K., & Oppenheimer, B. 2012, MNRAS, 421, 98
- Dekel, A., & Mandelker, N. 2014, MNRAS, 444, 2071
- Dessauges-Zavadsky, M., D'Odorico, S., Schaerer, D., Modigliani, A., Tapken,
- C., & Vernet, J. 2010, A&A, 510, 26 Dessauges-Zavadsky, M., Christensen, L., D'Odorico, S., Schaerer, D., & Richard, J. 2011, A&A, 533, 15
- Downes, D., & Solomon, P. M. 1998, ApJ, 507, 615
- Draine, B. T., & Li, A. 2007, ApJ, 657, 810
- Draine, B. T., Dale, D. A., Bendo, G., et al. 2007, ApJ, 663, 866
- Dwek, E. 1998, ApJ, 501, 643
- Edmunds, M. G. 2001, MNRAS, 328, 223
- Egami, E., Rex, M., Rawle, T. D., et al. 2010, A&A, 518, L12
- Elbaz, D., Daddi, E., Le Borgne, D., et al. 2007, A&A, 468, 33
- Elbaz, D., Dickinson, M., Hwang, H. S., et al. 2011, A&A, 533, 119
- Erb, D. K., Steidel, C. C., Shapley, A. E., et al. 2006, ApJ, 644, 813
- Fall, S. M., & Efstathiou, G. 1980, MNRAS, 193, 189
Feldmann, R., Gnedin, N. Y., & Kravtsov, A. V. 2012, ApJ, 747, 124
Fu, H., Jullo, E., Cooray, A., et al. 2012, ApJ, 753, 134
- Gao, Y., & Solomon, P. M. 2004, ApJS, 152, 63
- Geach, J. E., Smail, I., Coppin, K., Moran, S. M., Edge, A. C., & Ellis, R. S. 2009, MNRAS, 395, L62
- Geach, J. E., Smail, I., Moran, S. M., MacArthur, L. A., Lagos, C. del P., & Edge, A. C. 2011, ApJL, 730, L19
- Genel, S., Bouch´ e, N., Naab, T., Sternberg, A., & Genzel R. 2010, ApJ, 719, 229 Genzel, R., Tacconi, L. J., Gracia-Carpio, J., et al. 2010, MNRAS, 407, 2091
- Genzel, R., Tacconi, L. J., Combes, F., et al. 2012, ApJ, 746, 69
- Greve, T. R., Bertoldi, F., Smail, I., et al. 2005, MNRAS, 359, 1165
Gruppioni, C., Pozzi, F., Rodighiero, G., et al. 2013, MNRAS, 432, 23
Harris, A. I., Baker, A. J., Zonak, S. G., et al. 2010, ApJ, 723, 1139
Heavens, A., Panter, B., Jimenez, R., & Dunlop, J. 2004, Nature, 428, 625
Helfer, T. T., Thornley, M. D., Regan, M. W., et al. 2003, ApJS, 145, 259
Hopkins, A. M., & Beacom, J. F. 2006, ApJ, 651, 142
- Inoue, A. K. 2003, PASJ, 55, 901
Iono, D., Tamura, Y., Nakanishi, K., et al. 2006, PASJ, 58, 957
- Israel, F. P. 1997, A&A, 328, 471
- Issa, M. R., MacLaren, I., & Wolfendale, A. W. 1990, A&A, 236, 237
- Ivison, R. J., Smail, I., Papadopoulos, P. P., et al. 2010, MNRAS, 404, 198
- Ivison, R. J., Papadopoulos, P. P., Smail, I., Greve, T. R., Thomson, A. P., Xilouris, E. M., & Chapman, S. C. 2011, MNRAS, 412, 1913
- Jimenez, R., Bernardi, M., Haiman, Z., Panter, B., & Heavens, A. F. 2007, ApJ, 669, 947
Johansson, D., Horellou, C., Lopez-Cruz, O., et al. 2012, A&A, 543, 62
Jones, T. A., Swinbank, A. M., Ellis, R. S., et al. 2010, MNRAS, 404, 1247
- Kennicutt, R. C. Jr. 1998a, ApJ, 498, 541
- Kennicutt, R. C. Jr. 1998b, ARA&A, 36, 189
- Kneib, J.-P., Neri, R., Smail, I., Blain, A., Sheth, K., van der Werf, P., & Knudsen, K. K. 2005, A&A, 434, 819
- Knudsen, K. K., Neri, R., Kneib, J.-P., & van der Werf, P. P. 2009, A&A, 496, 45 Lagos, C. Del P., Baugh, C. M., Lacey, C. G., Benson, A. J., Kim, H.-S., & Power, C. 2011, MNRAS, 418, 1649
- Lagos, C. del P., Bayet, E., Baugh, C. M., et al. 2012, MNRAS, 426, 2142
Lagos, C. del P., Baugh, C. M., Zwaan, M. A., et al. 2014, MNRAS, 440, 920
- Leech, J., Isaak, K. G., Papadopoulos, P. P., Gao, Y., & Davis, G. R. 2010, MNRAS, 406, 1364
- Leroy, A. K., Walter, F., Brinks, E., et al. 2008, ApJ, 136, 2782
- Leroy, A. K., Walter, F., Bigiel, F., et al. 2009, ApJ, 137, 4670
- Leroy, A. K., Bolatto, A., Gordon, K., et al. 2011, ApJ, 737, 12
- Lilly, S. J., Carollo, C. M., Pipino, A., Renzini, A., & Peng, Y. 2013, ApJ, 772, 119
- Maiolino, R., Nagao, T., Grazian, A., et al. 2008, A&A, 488, 463 McCracken, H. J., Capak, P., Salvato, M., et al. 2010, ApJ, 708, 202
- Magdis, G. E., Daddi, E., Elbaz, D., et al. 2011, ApJL, 740, L15
- Magdis, G. E., Daddi, E., Sargent, M., et al. 2012a, ApJL, 758, L9
- Magdis, G. E., Daddi, E., B´ e thermin, M., et al. 2012b, ApJ, 760, 6
Magdis, G. E., Rigopoulou, D., Hopwood, R., et al. 2014, ApJ, 796, 63
- Magnelli, B., Saintonge, A., Lutz, D., et al. 2012, A&A, 548, 22
- Magnelli, B., Popesso1, P., Berta, S., et al. 2013, A&A, 553, 132
- Maiolino, R., Nagao, T., Grazian, A., et al. 2008, A&A, 488, 463
- Mancini, C., F¨ster Schreiber, N. M., Renzini, A., et al. 2011, ApJ, 743, 86 o Mannucci, F., Cresci, G., Maiolino, R., Marconi, A., & Gnerucci, A. 2010, MNRAS, 408, 2115
Mannucci, F., Salvaterra, R., & Campisi, M. A. 2011, MNRAS, 414, 1263
Mo, H. J., Mao, S., & White, S. D. M. 1998, MNRAS, 295, 319
- Narayanan, D., & Krumholz, M. R. 2014, MNRAS, 442, 1411
- Noeske, K. G., Weiner, B. J., Faber, S. M., et al. 2007, ApJL, 660, L43
- Obreschkow, D., & Rawlings, S. 2009, ApJ, 696, L129
- Papadopoulos, P. P., van der Werf, P., Isaak, K., & Xilouris, E. M. 2010, ApJ, 715, 775
- Papadopoulos, P. P., van der Werf, P. P., Xilouris, E. M., Isaak, K. G., Gao, Y ., & M¨ uhle, S. 2012, MNRAS, 426, 2601
- Planck Collaboration 2011a, A&A, 536, 21
Planck Collaboration 2011b, A&A, 536, 25
- Quider, A., Pettini, M., Shapley, A. E., & Steidel, C. C. 2009, MNRAS, 398, 1263
- Rawle, T. D., Egami, E., Bussmann, R. S., et al. 2014, ApJ, 783, 59
- Richard, J., Kneib, J.-P., Jullo, E., et al. 2007, ApJ, 662, 781
- Richard, J., Jones, T., Ellis, R., et al. 2011, MNRAS, 413, 643
- Riechers, D. A., Capak, P. L., Carilli, C. L., et al. 2010, ApJL, 720, L131
- Riechers, D. A., Cooray, A., Omont, A., et al. 2011, ApJL, 733, L12
- Rodighiero, G., Cimatti, A., Gruppioni, C., et al. 2010, A&A, 518, 25
- Saintonge, A., Kau GLYPH<11> mann, G., Wang, J., et al. 2011, MNRAS, 415, 61
Saintonge, A., Tacconi, L. J., Fabello, S., et al. 2012, ApJ, 758, 73
- Saintonge, A., Lutz, D., Genzel, R., et al. 2013, ApJ, 778, 2
- Salmi, F., Daddi, E., Elbaz, D., et al. 2012, ApJ, 754, 14
Salpeter, E. E. 1955, ApJ, 121, 161
Santini, P., Maiolino, R., Magnelli, B., et al. 2014, A&A, 562, 30
- Sargent, M. T., Daddi, E., B´ e thermin, M., et al. 2014, ApJ, 793, 19
- Savaglio, S., Glazebrook, K., Le Borgne, D., et al. 2005, ApJ, 635, 260
Schaerer, D., de Barros, S., & Sklias, P. 2013, A&A, 549, 4
Scoville, N., Aussel, H., Sheth, K., et al. 2014, ApJ, 783, 84
- Sharon, C. E., Baker, A. J., Harris, A. I., & Thomson, A. P. 2013, ApJ, 765, 6
- Sklias, P., Zamojski, M., Schaerer, D., et al. 2014, A&A, 561, 149
- Smail, I., Swinbank, A. M., Richard, J., et al. 2007, ApJL, 654, L33
- Solomon, P. M., Downes, D., Radford, S. J. E., et al. 1997, ApJ, 478, 144
- Stewart, K. R., Brooks, A. M., Bullock, J. S., Maller, A. H., Diemand, J., Wadsley, J., & Moustakas, L. A. 2013, ApJ, 769, 74
- Swinbank, A. M.; Smail, I., Longmore, S., et al. 2010, Nature, 464, 733
- Tacconi, L. J., Neri, R., Chapman, S. C., et al. 2006, ApJ, 640, 228
- Tacconi, L. J., Genzel, R., Neri R., et al. 2010, Nature, 463, 781
- Tacconi, L. J., Neri, R., Genzel, R., et al. 2013, ApJ, 768, 74
- Tan, Q., Daddi, E., Sargent, M., et al. 2013, ApJL, 776, L24
- Thomas, D., Maraston, C., Schawinski, K., Sarzi, M., & Silk, J. 2010, MNRAS, 404, 1775
- Tremonti, C. A., Heckman, T. M., Kau GLYPH<11> mann, G., et al. 2007, ApJ, 613, 898
- Trujillo, I., F¨rster Schreiber, N. M., Rudnick, G., et al. 2006, ApJ, 650, 18 o
- Weiss, A., Downes, D., Neri, R., et al. 2007, A&A, 467, 955
- Weiss, A., Ivison, R. J., Downes, D., Walter, F., Cirasuolo, M., & Menten, K. M. 2009, ApJ, 705, L45
- Wilson, C. D., Petitpas, G. R., Iono, D., et al. 2008, ApJS 178, 189
- Wilson, C. D., Warren, B. E., Israel, F. P., et al. 2012, MNRAS, 424, 3050
- Yan, L., Tacconi, L. J., Fiolet, N., et al. 2010, ApJ, 714, 100
- Yee, H. K. C., Ellingson, E., Bechtold, J., Carlberg, R. G., & Cuillandre, J.-C. 1996, AJ, 111, 1783
## Appendix A: CO properties and kinematics of the individual lowL IR selected galaxies
A.1. A68-C0
The CO(2-1) intensity peak is detected at 11 GLYPH<27> in A68-C0, which makes it a very secure CO detection in this z H GLYPH<11> = 1 5864 : galaxy, as shown in Figs. 3 and 4. The CO(2-1) emission is well centred on the HST images to within 0 5 : 00 . The HST images clearly reveal the presence of two counterpart images of the same galaxy with a morphology typical of a spiral galaxy. The spiral structure is strongly stretched and extended by gravitational lensing, but the two counterpart images that are mirrored are easily identified by the two bright bulges observed north and south with the critical line passing in between. The two counterpart images of this galaxy are not resolved in CO, but they both contribute to the detected CO emission given the CO emission intensity precisely peaking on the critical line and the CO elongation in the direction of the stretch. The two counterpart images are amplified by the same magnification factor.
The CO(2-1) spectrum reveals a double-peaked emission line profile with a total full width half maximum FWHMCO = 334 km s GLYPH<0> 1 and an observed integrated flux F CO = 1 89 : GLYPH<6> 0 28 Jy km s : GLYPH<0> 1 . The measured CO redshift agrees very well with the H GLYPH<11> redshift to within GLYPH<1> z = ( z H GLYPH<11> GLYPH<0> z CO) = 0 001. The derived : lensing-corrected molecular gas fraction f gas = 0 38 is much : higher than the molecular gas fractions of local spirals, but is comparable to the f gas distribution of SFGs with CO measurements from the literature in the h z 1 2 : i = [1 ; 1 6] interval which : has a mean h f h z 1 2 : i gas i = 0 46 : GLYPH<6> 0 15 (see Sect. 6.1). :
Fig. A.1. Velocity-integrated map of the CO(2-1) emission in A68-C0 overlaid on the HST image in band F110W. The extended blue contours are the contours integrated over the channels that define the blue-shifted component of the double-peaked CO emission line profile shown in Fig. 4 and the compact red contours are the contours integrated over the channels that define the red-shifted component of the CO profile. Contour levels start at GLYPH<6> 2 GLYPH<27> and are in steps of 2 GLYPH<27> . The critical line (yellow line) is located right in the middle of the two multiple images easily identified by the two bright bulges resolved in the HST image.
1 The observed double-peaked CO emission line profile is a first strong indication of rotation in A68-C0 (Daddi et al. 2010a). Two additional observational facts support rotation in this galaxy. First, its spiral morphology as seen in HST images (see Fig. 3). Second, as shown in Fig. A.1 the CO contours integrated over the channels that define the blue-shifted component (blue contours) of the double-peaked CO emission line profile (Fig. 4) appear to be much more extended in comparison to the contours integrated over the channels that define the redshifted component (red contours) of the CO profile that are more compact. This is a direct consequence of rotation combined with gravitational lensing. Indeed, the critical line is located right in the middle of the two multiple images and is at an angle relative to the major axis of the spiral structure. Because the blue-shifted velocities relative to the observer are the farthest from the critical line and the red-shifted velocities the closest to the critical line, we see the blue-shifted CO component to be much more extended than the red-shifted component when plotting the CO blue-shifted and red-shifted velocity-integrated contours separately in Fig. A.1.
## A.2. A68-HLS115
The CO(2-1) line is also very well detected in A68-HLS115 as it can be seen in Figs. 3 and 4, with its intensity peaking at 14 GLYPH<27> . The CO(2-1) emission is nicely centred on the HST images to within 0 5 : 00 . The HST images show that this z H GLYPH<11> = 1 5869 galaxy : has a very clumpy structure. The clumps are bright in the restframe UV, betraying intense star formation. Unfortunately they
are unresolved in our CO observations, but would be of high interest for a deeper CO study at high spatial resolution.
The CO(2-1) spectrum reveals a strong emission line very well fitted by a single Gaussian function with a full width half maximum FWHMCO = 267 km s GLYPH<0> 1 and an observed integrated flux F CO = 2 00 : GLYPH<6> 0 30 Jy km s : GLYPH<0> 1 . The measured CO redshift perfectly agrees with the H GLYPH<11> redshift to within GLYPH<1> z = ( z H GLYPH<11> GLYPH<0> z CO) = 0 001. The derived lensing-corrected molecular : gas fraction f gas = 0 75 is very high; A68-HLS115 is among the : gas-richer z > 1 SFGs known (see Sect. 6). With a strong singlepeaked CO emission line profile, we can hardly invoke evidence for rotation in this galaxy.
## A.3. MACS0451-arc
The HST image reveals that this z H GLYPH<11> = 2 013 galaxy is strongly : stretched by gravitational lensing as testified by the striking, elongated arc extending over 20 00 and composed of multiple images of the source. We detect the CO(3-2) emission at 4 GLYPH<0> 5 GLYPH<27> . It does not cover the entire arc, but is confined to the southern part of the arc (see Fig. 3). Such a confinement may either indicate that the lensing magnification is di GLYPH<11> erential along the arc, and or that the molecular gas is located in specific regions of the / galaxy. The spatially resolved kinematics of this arc has been studied in detail with H GLYPH<11> observations by Jones et al. (2010). This work allowed, in particular, to obtain a precise lens modelling and the reconstruction of the MACS0451-arc in the source plane. They show: (1) the southern part of the arc is the image of the whole source and it also corresponds to the image in which the central red region of the source, the region where most of the CO emission is likely to originate, is most strongly magnified; and (2) the northern part of the arc contains two images of the source very weakly magnified, in particular the central red region of the source is invisible in the HST images, the emission is dominated by one star-forming clump of the source close to the critical line which is popping up because of the increased magnification. Thus, the lack of CO detection in the northern part of the MACS0451-arc can be explained by the sensitivity limit of our PdBI data: if the total integrated CO flux of the southern part of the arc is detected at only S = Ndet = 4 GLYPH<0> 5 GLYPH<27> , the northern part of the arc is expected to have an integrated CO flux below 2 GLYPH<27> of the noise. We may conclude with confidence that the detected CO emission provides a good estimate of the total molecular gas content of this strongly-lensed galaxy. Finally, the small o GLYPH<11> set of 1 00 observed between the HST images and the CO emission, as well as the elongated shape of the CO emission which does not follow the exact orientation of the arc in HST images (see Fig. 3) are both most likely due to noise e GLYPH<11> ects. Indeed, the low signal-to-noise ratio of the MACS0451-arc's CO detection combined with the low spatial resolution of D-configuration data imply that only the detection itself is reliable, while any spatial information remains uncertain.
There is evidence that the MACS0451-arc contains an AGN component, as determined from the analysis of its IR SED. In particular, the southern part of the arc, which contains the most magnified image of the center of the source (as said above), shows a much hotter IR SED than the northern part of the arc. Indeed, while the SED of the northern part peaks at 250 GLYPH<22> m, that of the southern part rises down to 100 GLYPH<22> m (corresponding to a GLYPH<21> rest < 33 GLYPH<22> m). This excess of flux at short wavelengths indicates the presence of very hot dust, which cannot be reproduced by a stellar component only. By modelling the IR emission of the southern part, Zamojski et al. (in prep) find that roughly 55% of the IR-flux originates from an AGN, while the remaining 45% is coming from star formation. This is taken into account when estimating the total IR luminosity of the MAS0451-arc corrected from the AGN component (Table 1).
The CO(3-2) spectrum indicates the presence of a doublepeaked emission line profile (see Fig. 4). The total full width half maximum is FWHMCO = 261 km s GLYPH<0> 1 and the observed integrated flux F CO = 1 27 : GLYPH<6> 0 32 Jy km s : GLYPH<0> 1 . The measured CO redshift agrees very well with the H GLYPH<11> redshift to within GLYPH<1> z = ( z H GLYPH<11> GLYPH<0> z CO) = 0 0012. The derived lensing-corrected molecular : gas fraction f gas = 0 62 is high, it is the higher among SFGs : with CO measurements from the literature in the h z 2 2 : i = [2 ; 2 5] : interval (see Sect. 6.1).
The observed double-peaked CO emission line profile in the MACS0451-arc may appear as tentative because of the low S Ndet = = 4 GLYPH<0> 5 GLYPH<27> , however the fit with two Gaussian functions is clearly favoured relative to the fit with a single Gaussian function. This prevents us from a GLYPH<14> rming conclusively whether the observed double-peaked CO emission line profile really originates from the presence of rotation (Daddi et al. 2010a), or is simply an artefact of a noisy line profile. Jones et al. (2010) showed that the kinematics of this galaxy as traced by the integrated H GLYPH<11> emission line flux can be explained by rotation. Because, their data cover only part of the galaxy, either scenario, rotation versus merger, remains possible. The recent reconstructed morphology of the MACS0451-arc galaxy in the source plane found in Sklias et al. (2014) shows two tails, which can be most easily explained by a merger. Such a scenario would also naturally explain the presence of the AGN seen in the southern part of the arc.
## A.4. A2218-Mult
This z H GLYPH<12> = 3 104 galaxy is the highest redshift : galaxy from our sample of lowL IR selected galaxies. HST images reveal two counterpart images (Richard et al. 2011). We searched for CO in the western counterpart image. Comparatively we achieved similarly deep PdBI observations with an rms of 0.77 mJy for this object, but the CO(3-2) emission remains undetected despite a high magnification factor GLYPH<22> = 14. We provide 4 GLYPH<27> upper limits on the CO(3-2) integrated line flux and its respective derivatives CO(1-0) luminosity and molecular gas mass in Table 3. This object was not included in the study of Sklias et al. (2014), but we used the same tools and methods to derive its IR luminosity.
## A.5. A68-h7
This z CO = 2 1529 galaxy is less magnified with : GLYPH<22> = 3 and characterized by a higher intrinsic IR luminosity L IR = 18 3 : GLYPH<2> 10 11 L GLYPH<12> in comparison to the other galaxies from our sample. It has thus been observed at the IRAM 30 m telescope. The CO(3-2) spectrum shows a broad line detected at 3 GLYPH<0> 4 GLYPH<27> and well fitted by a single Gaussian function (see Fig. 4). It has a full width half maximum FWHMCO = 282 km s GLYPH<0> 1 and an observed integrated flux F CO = 1 12 : GLYPH<6> 0 22 Jy km s : GLYPH<0> 1 . A redshift o GLYPH<11> set is observed in A68-h7 between the CO redshift and the redshift determined in this case from weak absorption lines in the rest-frame UV, GLYPH<1> z = ( z rest GLYPH<0> UV GLYPH<0> z CO) = GLYPH<0> 0 0029. This o : GLYPH<11> -set is not surprising, because interstellar medium lines are expected to be blueshifted with respect to the systemic redshift of the galaxy as set here by the CO(3-2) line in presence of outflows (Quider et al. 2009; Dessauges-Zavadsky et al. 2010). The derived lensing-corrected molecular gas fraction f gas = 0 33 is : comparable to the f gas distribution of SFGs with CO measure-
ments from the literature in the h z 2 2 : i = [2 ; 2 5] interval which : has a mean h f h z 2 2 : i gas i = 0 45 : GLYPH<6> 0 22 (see Sect. 6.1). :
The observed double-peaked CO emission line profile, which would require higher signal-to-noise ratio data to be confirmed (in particular for the second component sampled by two channels only given its small full width half maximum FWHMCO = 90 km s GLYPH<0> 1 ; see Fig. 4), sustains evidence for rotation in A68-h7 (Daddi et al. 2010a). Its morphology in the HST images reveals several components, which probably all contribute to the CO emission signal and may be the analogues of giant star-forming clumps seen in disk-like simulated systems (Agertz et al. 2009; Genel et al. 2010; Bournaud et al. 2011).
## A.6. cB58 and Eye
We adopt for these two strongly-lensed galaxies at z H GLYPH<11> = 2 729 : and 3.0733, respectively, the CO(1-0) integrated fluxes obtained with the Expanded Very Large Array (EVLA) by Riechers et al. (2010). The Eye is one of the rare z > 3 star-forming galaxies with a CO detection. Detections of the CO(3-2) line with the PdBI and measurements of their respective integrated fluxes were also reported by Baker et al. (2004) in cB58 and Coppin et al. (2007) in the Eye. Those are used in Sect. 4 to study the CO luminosity correction factors of star-forming galaxies at z > 1. Combined with our updated stellar masses from Sklias et al. (2014), we infer lensing-corrected molecular gas fractions f gas = 0 41 in cB58 and : f gas = 0 18 in the Eye. Although they are : very similar to the molecular gas fractions reported by Riechers et al. (2010), f gas = 0 32 in cB58 and : f gas = 0 13 in the Eye, : if we consider the stellar masses assumed by Riechers et al. and our prescriptions for the CO luminosity correction factor and the CO-H2 conversion factor, the corresponding molecular gas fractions get significantly higher by a factor of 2 and 2.5, respectively, compared to ours. The revised molecular gas fraction of the Eye is particularly low, it is the lowest among z > 1 SFGs with CO measurements from the literature (see Sect. 6). This may partly be attributed to the 'post-starburst' nature of the Eye as inferred by Sklias et al. (2014) from a detailed SED analysis.
|
10.1051/0004-6361/201424661
|
[
"M. Dessauges-Zavadsky",
"M. Zamojski",
"D. Schaerer",
"F. Combes",
"E. Egami",
"A. M. Swinbank",
"J. Richard",
"P. Sklias",
"T. D. Rawle",
"M. Rex",
"J. -P. Kneib",
"F. Boone",
"A. Blain"
] |
2014-08-04T20:51:21+00:00
|
2015-02-11T23:48:41+00:00
|
[
"astro-ph.GA"
] |
Molecular gas content in strongly-lensed z~1.5-3 star-forming galaxies with low IR luminosities
|
To extend the molecular gas measurements to typical star-forming galaxies
(SFGs) with SFR < 40 Msun yr^{-1} and M* < 2.5x10^{10} Msun at z~1.5-3, we have
observed CO emission with the IRAM Plateau de Bure Interferometer and 30m
telescope for five strongly-lensed galaxies selected from the Herschel Lensing
Survey. These observations are combined with a compilation of CO measurements
from the literature. We infer the luminosity correction factors r2,1 =
0.81+/-0.20 and r3,1 = 0.57+/-0.15 for the J=2 and J=3 CO transitions,
respectively, valid for SFGs at z>1. The combined sample of CO-detected SFGs at
z>1 shows a large spread in star formation efficiency (SFE), such that SFE
extend beyond the low values of local spirals and overlap the distribution of
z>1 sub-mm galaxies. We find that the spread in SFE (or equivalently in
molecular gas depletion timescale) is due to primarily the specific star
formation rate, but also stellar mass and redshift. Correlations of SFE with
the offset from the main-sequence and the compactness of the starburst are less
clear. The increase of the molecular gas depletion timescale with M* now
revealed by low M* SFGs at z>1 and observed at z=0 is in contrast to the
admitted constant molecular gas depletion timescale and the linear
Kennicutt-Schmidt relation. We confirm an increase of the molecular gas
fraction (fgas) from z~0.2 to z~1.2, followed by a very mild increase toward
higher redshifts. At each redshift fgas shows a large dispersion due to the
dependence of fgas on M*, producing a gradient of increasing fgas with
decreasing M*. We provide the first measure of fgas of z>1 SFGs at the low-M*
end (10^{9.4} < M*/Msun < 10^{9.9}), reaching a mean fgas = 0.69+/-0.18, which
shows a clear fgas upturn. Finally, we find evidence for a non-universal
dust-to-gas ratio among high-redshift SFGs and sub-mm galaxies, local spirals
and ULIRGs with near-solar metallicities.
|
1408.0817v1
|
## The connection between stellar granulation and oscillation as seen by the Kepler mission
T. Kallinger 1 , J. De Ridder 2 , S. Hekker 3 4 , , S. Mathur 5 6 , , B. Mosser 7 , M. Gruberbauer 8 1 , , R. A. Garc´ ıa 9 , C. Karo ff 10 11 , , and J. Ballot 12 13 ,
- 1 Institute for Astronomy (IfA), University of Vienna, T¨rk enschanzstrasse 17, 1180 Vienna, Austria u
- 2 Instituut voor Sterrenkunde, KU Leuven, Celestijnenlaan 200D, 3001 Leuven, Belgium
- 3
- Max-Planck-Institut f¨ ur Sonnensystemforschung, Justus -von-Liebig-Weg 3, 37077 G¨ ottingen, Germany
- 4 Astronomical Institute 'Anton Pannekoek', University of Amsterdam, PO Box 94249, 1090 GE Amsterdam, The Netherlands
- 5 High Altitude Observatory, NCAR, P.O. Box 3000, Boulder, CO 80307, USA
- 6 Space Science Institute, 4750 Walnut street Suite 205, Boulder, CO 80301, USA
- 7 LESIA, CNRS, Universit´ e Pierre et Marie Curie, Universit´ e Denis Diderot, Observatoire de Paris, 92195 Meudon cedex, France
- 8 Department of Astronomy and Physics, Saint Marys University, Halifax, NS B3H 3C3, Canada
- 9 Laboratoire AIM, CEA DSM-CNRS, Universit´ e Paris 7 Diderot, IRFU SAp, Centre de Saclay, 91191, Gif-sur-Yvette, France / /
- 10 Stellar Astrophysics Centre, Department of Physics and Astronomy, Århus University, Ny Munkegade 120, DK-8000 Århus C , Denmark
- 11 Department of Geoscience, Århus University, Høegh-Guldbe rgs Gade 2, 8000 Århus C, Denmark
- 12 CNRS, Institut de Recherche en Astrophysique et Plan´ etologie, 14 avenue Edouard Belin, 31400 Toulouse, France ´
- 13 Universit´ e de Toulouse, UPS-OMP, IRAP, Toulouse, France
Received 30 May 2014 / Accepted 31 July 2014
## ABSTRACT
Context. The long and almost continuous observations by Kepler show clear evidence of a granulation background signal in a large sample of stars, which is interpreted as the surface manifestation of convection. It has been shown that its characteristic timescale and rms intensity fluctuation scale with the peak frequency ( ν max) of the solar-like oscillations. Various attempts have been made to quantify the observed signal, to determine scaling relations for its characteristic parameters, and to compare them to theoretical predictions.
Aims. We aim to study di ff erent approaches to quantifying the signature of stellar granulation and to search for a unified model that reproduces the observed signal best in a wide variety of stars. We then aim to define empirical scaling relations between the granulation properties and ν max and various other stellar parameters.
Methods. Weuse a probabilistic method to compare di ff erent approaches to extracting the granulation signal. We fit the power density spectra of a large set of Kepler targets, determine the granulation and global oscillation parameter, and quantify scaling relations between them.
Results. We establish that a depression in power at about ν max / 2, known from the Sun and a few other main-sequence stars, is also statistically significant in red giants and that a super-Lorentzian function with two components is best suited to reproducing the granulation signal in the broader vicinity of the pulsation power excess. We also establish that the specific choice of the background model can a ff ect the determination of ν max, introducing systematic uncertainties that can significantly exceed the random uncertainties. We find the characteristic frequency and amplitude of both background components to tightly scale with ν max for a wide variety of stars, and quantify a mass dependency of the latter. To enable comparison with theoretical predictions, we computed e ff ective timescales and bolometric intensity fluctuations and found them to approximately scale as τ e ff ∝ g -0 85 . T -0 4 . and A gran ∝ ( g M 2 ) -1 4 / (or more conveniently R M / 3 4 / ), respectively. Similarly, the bolometric pulsation amplitude scales approximately as A puls ∝ ( g M 2 ) -1 3 / (or R 4 3 / / M ), which implicitly verifies a separate mass and luminosity dependence of A puls .
Conclusions. Weprovide a thorough analysis of the granulation background signal in a large sample of stars, from which we establish a unified model that allows us to accurately extract the granulation and global oscillation parameter.
Key words. stars: late-type - stars: oscillations - stars: fundamental parameters - stars: interior
## 1. Introduction
Modern theoretical physics is still lacking a complete description of turbulence, which has proven to be highly resistant to full modelling because of the wide range in scale from macroscopic to microscopic. A promising opportunity for studying this process is provided by stellar astrophysics. Turbulent convection is an important energy transport mechanism that is essential for modelling the structure and evolution of our Sun and stars in general.
Send o ff print requests to : [email protected]
The surface of the Sun shows an irregular cellular pattern, which is known as the surface-visible signature of convection. It is due to hot plasma that rises from the outer convective zone to the photosphere, where it forms bright cells (the so-called granules), cools down, and descends at the darker inter-granular lanes. These granulation structures on the solar surface were first discovered by Herschel (1801), and thanks to numerical hydrodynamical simulations, their properties are nowadays explained
well (e.g., Muller 1999). Other phenomena related to convection are the so-called solar-like oscillations that are due to turbulent motions in the convective envelope producing an acoustic noise that stochastically drives resonant, intrinsically damped acoustic oscillations (e.g., Christensen-Dalsgaard 2002).
In the Hertzsprung-Russel (HR) diagram, the red border of the classical δ Sct instability strip (ISRB) is expected to mark the transition from opacity mechanisms driving stellar oscillations on the blue side of the ISRB to stochastic excitation on the red side. This is also the region where stars develop an appreciable upper convective envelope giving rise to surface granulation. In contrast to this simplistic picture, evidence has been found that even the very shallow convective envelopes of hot stars ( < 1% in radius) are su ffi cient to produce an observable granulation signal (Kallinger & Matthews 2010) and potentially excite solarlike oscillations (Antoci et al. 2011; Belkacem et al. 2009).
However, all stars with an convective envelope show granulation. The granules on their surface evolve with time and produce quasi-periodic brightness fluctuations on a wide range of timescales and amplitudes (see e.g., Harvey 1985; Aigrain et al. 2004, for the Sun). Thanks to the high precision photometric measurements of CoRoT (Baglin et al. 2006) and Kepler (Borucki et al. 2010; Koch et al. 2010) such brightness fluctuations are subject to continuous long-term monitoring for a large variety of stars. First results for a large sample study have been presented by Mathur et al. (2011), who investigated the characteristics of granulation in red giants observed by Kepler . They found that the characteristic timescale ( τ e ff ) and the rms brightness fluctuation ( σ ) of the granulation signal scale to a first approximation with the peak frequency ( ν max) of the solar-like oscillations 1 as τ e ff ∝ ν -0 89 . max and σ ∝ ν -0 45 . max , respectively, which is consistent with basic theoretical predictions (e.g., Kjeldsen & Bedding 2011; Chaplin et al. 2011a). The authors also performed hydrodynamical 3D simulations of granulation in red giants on the basis of the Ludwig (2006) ab initio approach. Even though these simulations match the observed scaling relations well in terms of trends, they found large systematic di ff erences of up to a factor of three. Recently, Samadi et al. (2013a,b) provided a simple theoretical model that supports the observed variations of the granulation properties with ν max, but also suggests that the turbulent Mach number in the photosphere plays an important role in controlling granulation. As in Mathur et al. (2011), they found their theoretical granulation parameters to match the observations well on a global scale but to systematically underestimate the measurements. From these attempts it appears that the general trends of the physical properties of granulation across the HR diagram are reproducible from a modelling point of view, but the source of the rather large systematic deviations remains unknown. The reason for that is not necessarily entirely due to insu ffi cient models. There is also no consensus yet on how to exactly extract the granulation parameters from the observations and large systematic di ff erences have been found between di ff erent methods (see Mathur et al. 2011, for a comparison).
To infer the characteristics of stellar granulation from the power spectrum of a stars' intensity variations is not straightforward and di ff erent analysis methods are found in the literature. These methods use one or more components to model the background signal locally around ν max, and di ff er from each other
1 The peak frequency ( ν max; also known as the frequency of maximum oscillation power) is expected to be proportional to the acoustic cuto ff -frequency ( ν ac ∝ g / √ T e ff ) of the stellar atmosphere (e.g., Brown et al. 1991; Kjeldsen & Bedding 1995; Belkacem et al. 2011).
not only in the number of components used but also in the functional form of the individual components. Usually models of the form P ( ν ) ∝ 1 [1 / + ( πτν ) c ] are implemented, in which the exponent c plays an important role as it controls how fast the power, P , decays with increasing frequency. Originally Harvey (1985) adopted a value of two (i.e., a Lorentzian function, which is in this context also often named the 'classical Harvey model') to model the solar background signal but already mentioned that the actual exponent may well be di ff erent from two. The reason why this is of particular interest is because the exponent defines the general shape of the background component and the use of an 'incorrect' value can easily result in systematic deviations of the other background parameters. This e ff ect has not yet been investigated thoroughly. In fact, it is di ffi cult to measure the actual exponent from the observations (even for the Sun). Therefore values di ff erent from two were established empirically, like a fixed exponent of four (e.g., Michel et al. 2009; Kallinger et al. 2010c) but also more general models with c being a free parameter (resulting in values as high as 6.2 for the Sun, Karo ff 2012).
Another problem arises from the number of components that are superposed to reproduce the stellar background signal. Observational evidence for more than one background component in the vicinity of the pulsation power excess is given by a slight depression in power at about ν max / 2. Such a feature has been first found in the solar irradiance data from the SOHO VIRGO instrument / (e.g., V´ azquez Rami´ et al. o 2005) but is meanwhile also found to be significant in other stars (e.g., Karo ff et al. 2013). The physical origin of this kink in the power spectrum (and therefore an additional signal component) is still a matter of debate and is either attributed to the occurrence of bright points (e.g., Harvey et al. 1993), faculae (Karo ff 2012), changing properties of the granules (Andersen et al. 1998), or a second granulation population (e.g., V´ azquez Rami´ et a l. o 2005).
A more phenomenological approach of granulation was recently demonstrated by Bastien et al. (2013), who filtered the Kepler time series with a 8 hour high-pass filter and found that the rms scatter of the residuals are a good indicator for the surface gravity of the stars. The advantage of this method is that it is much faster and easier to apply than asteroseismic techniques and that it gives a more accurate surface gravity than the usual spectroscopic methods.
The aim of the current analysis is twofold. Firstly, we investigate various approaches to model the granulation background signal in a large sample of stars observed by Kepler . The sample includes stars from the main sequence to high up the giant branch and asymptotic giant branch (compared to the analysis in Mathur et al. 2010, which was restricted to red giants) and therefore basically covers the entire parameter space for which we can expect solar-like oscillations and surface granulation. We use a probabilistic approach to search for a unified model that reproduces the observed signal best for all stars in our sample, where the complexity of the model should be driven by the data quality alone. We find that the superposition of two super-Lorentzian functions (i.e, Harvey-like models with c = 4) is well suited to reproduce the granulation signal in the broader vicinity of the pulsation power excess (about 0.1 to 10 × ν max) and fit this model to all stars in our sample.
Secondly, we investigate the resulting granulation timescales and rms brightness fluctuations and find tight correlations between these parameters and ν max. We then define empirical scaling relations between the granulation properties and the surface gravity (and other fundamental parameters) of a star and compare them to theoretical predictions. Furthermore, we study the
pulsation amplitudes and how they scale with other parameters of the star.
## 2. Observations
The Kepler space telescope was launched in March 2009 with the primary goal of searching for transiting Earth-sized planets in and near the habitable zones of Sun-like stars. Kepler houses a 95-cm aperture telescope that points at a single field in the constellations of Cygnus and Lyra, feeding a photometer with a 115 square-degree wide field of view (FOV) to continuously monitor the brightnesses of over 145 000 stars. The exquisite precision and accuracy of the photometry makes Kepler also an ideal instrument for asteroseismology and the Kepler Asteroseismic Science Consortium (KASC) has been set up to study many of the observed stars (see Gilliland et al. 2010 for an overview and first results).
Kepler observations are subdivided into quarters, starting with the initial commissioning run (10d, Q0), followed by a short first quarter (34 d, Q1) and subsequent full quarters of 90 d length. Photometry of most of the stars is conducted at a long cadence (LC) of 29.42 minutes, but a subset of up to 512 stars can be observed at a short cadence (SC) of 58.82 seconds (for more details see, e.g., Jenkins et al. 2010). Our studies are primarily based on the LC data for 1289 red-giant stars spanning from Q0 to Q13 with a total of about 51 800 measurements per star. Apart from occasional losses of the satellite's fine pointing and scheduled re-orientations of the spacecraft, the about 1142 day-long observations were continuous, with an overall duty cycle of about 93 %. The sample includes only stars that show a clear power excess due to pulsation and which have already been used for other studies, such as the investigation of di ff erent seismic (Huber et al. 2010, 2011; Mosser et al. 2012a) and granulation (Mathur et al. 2011) observables, the seismic determination of fundamental parameters (Kallinger et al. 2010b), a comparison of global oscillation parameters derived from different methods (Hekker et al. 2011, 2012), a detailed analysis of an individual star (di Mauro et al. 2011) and of the radial (Kallinger et al. 2012) and non-radial mixed mode spectra (Bedding et al. 2011; Mosser et al. 2012c), and how the latter are used to constrain rotational properties (Beck et al. 2012) and their evolution (Mosser et al. 2012b).
Naturally, this sample contains only stars that oscillate with timescales longer than about 1 h (twice the sampling rate) and is therefore limited to stars on the giant branch. To extend the sample towards the subgiant branch and main sequence we also include SC data of a subsample of solar-type and subgiant stars presented by Chaplin et al. (2011b). The data were obtained between Q5 an Q8 and cover up to one year of continuous observations with up to 480 000 individual measurements per star. Out of the original sample (500 stars), we selected those for which we have at least one full quarter of observations. Out of the remaining 113 stars we selected the 'best' 75 stars, that show a clear power excess with reasonably high signal-to-noise ratios and more importantly also the high-frequency part (see Fig. 1 above ∼ 50 µ Hz ) of the granulation signal (i.e., signal with timescales shorter than the pulsations) is well above the white noise. The latter is quite often not the case, even for the high-precision observations of Kepler . This does not result in a selection e ff ect, so that we select only stars with a high granulation amplitude and ignore those with low amplitudes. The fact that we see the high-frequency part of the granulation signal is predominantly due to a low white noise level, which basically reflects the apparent magnitude of the stars. While the original sample roughly
Fig. 1. Comparison between the heavily smoothed power density spectra of KIC 5517118 based on 'noise-optimised' (black line) and 'stability-optimised' (red line) data, showing additional frequency-dependent(instrumental) noise in the latter. The insert shows the relative di ff erence (PDSnoise-PDSstab ) PDSstab . / . and a Lorentzian fit (dashed line).
covers stars in the range 6 5 . /lessorsimilar Kp /lessorsimilar 12 mag, our SC sample is limited to Kp /lessorsimilar 10 mag.
## 3. Data analysis
## 3.1. Flux extraction
Generally spoken, the Kepler raw data are prepared so that the point-to-point scatter of the individual measurements is minimised ('noise-optimised'), which is important to detect and characterize planetary transits. This implies that the apertures on the CCD's from which the photometry is extracted are kept generally small to include only a minimum of the sky background signal. However, this leads to a number of problems. So it happens for example that a target star (but also background stars) drifts across the aperture on the CCD leading to longterm trends in the extracted photometry. Additionally, artefacts are introduced by occasional losses of the satellite fine pointing or during the thermal recovery after Kepler is pointed to the Earth. Even though these instrumental e ff ects can be corrected to some extent (see, e.g., Garc´ ıa et al. 2011; Smith et al. 2012, and references therein), they prevent us from accessing intrinsic variability timescales longer than a few weeks. This is not a problem for most of the stars in our sample. It, however, limits the sample to stars well below the tip of the giant branch. To improve the situation Mathur et al. (2014, in preperation) developed a new approach, whose basic principle is to adopt the individual apertures so that trends and jumps are minimised ('stabilityoptimised'). Subsequently, we smooth the data with a triangular filter to suppress residual instrumental long-term trends with timescales longer than about 40 d.
The improved long-term stability comes, however, at the price of additional high-frequency noise. This is illustrated in Fig. 1, where we compare the power density spectra of a particu-
Fig. 2. Power spectrum of KIC 7949599. The black and grey lines indicate the spectrum of the raw data (duty cycle of ∼ 93%) and after filling all gaps shorter than 1 day (duty cycle of ∼ 95%), respectively, showing how low-frequency signal leaks into the high-frequency range via the spectral window (see insert). The blue and red lines correspond to heavily smoothed versions of the raw and gap-filled spectrum, respectively.
lar star based on the original 'noise-optimised' (black line) and the new 'stability-optimised' (red line) data reductions. Apart from the di ff erent apertures the data sets are treated equally. Clearly, the new approach adds significant power at high frequencies 2 . The relative di ff erence of the two spectra reveals that the additional noise component is frequency dependent but follows a simple Lorentzian. In the modelling of the power density spectra (see Sec. 4 below) it should therefore be su ffi cient to add a frequency dependent term to the usually used constant white noise component.
We note that the presented example is a extreme case. In fact, the additional noise is negligible (i.e., its amplitude is much smaller than the white noise) for most stars in our sample. We also did not find any correlations between the amplitudes or timescales of the Lorentzian and any other fitted parameter and assume that this noise feature is largely defined by how exactly the aperture was chosen and how the sky background looks like in this aperture. However, there is a tendency that the amplitude becomes larger for stars with low ν max, which are usually also the brightest stars with a low white noise.
## 3.2. Gap filling
Even though Kepler is supposed to observe continuously, there are several operations a ff ecting the scientific data acquisition producing quasi-regular gaps (see, e.g., Garcıa et al. 2014). If we were to keep these gaps the spectral window would be degraded with, e.g., up to 0.5% high regular peaks (see the insert of Fig. 2). This might seem negligible but we have to keep in mind that the power contrast between the low-frequency part of the spec-
2 Note that the improved long-term stability is not noticable in this plot as the di ff erences are well below 2 µ Hz
trum with its typically high amplitudes and the high-frequency white noise is at least for red giants several orders of magnitudes. Leaving the data gaps untouched would smear out the intrinsic structure in the high-frequency part of the spectrum. To correct for this one can fill the gaps by, e.g., linear interpolation. In Fig. 2 we compare the raw and gap-filled spectrum of a typical Kepler data set of a red giant. Clearly, the aliasing signal from the gaps in the original time series buries the structure of the spectrum where the granulation signal fades into the white noise. On the other hand, filling gaps as long as, e.g., 16 days (resulting from a malfunction of the satellite at the beginning of Q8) would introduce spurious modulation peaks. As a compromise we only fill gaps by linear interpolation if they are shorter than 3 / ν max ( ∼ 1 day for most of the red giants), which improves the duty cycle from about 93 to about 95%. This appears as a small improvement but as can be seen from Fig. 2 it is su ffi cient to correct for most of the objectionable regularities in the spectral window function. A detailed description of the impact of nominal regular gaps in the Kepler time series is given in Garcıa et al. (2014). We have to mention that gap-filling has only a minor e ff ect on the extraction of the parameters relevant for the subsequent analysis but it becomes important if one wants to study the detailed structure of the granulation background signal (see Sec. 4.2).
The power spectra are then computed using a Discrete Fourier Transform algorithm (DFT; Deeming 1975) and converted to power density (see Appendix A).
## 3.3. Sampling effects
The variations we observe in stars are generally continuous but as soon as we measure them we naturally have to discretise them, i.e. the intrinsic signal is integrated for a certain time during each
Fig. 3. Heavily smoothed (5 µ Hz boxcar filter) power density spectrum of the about 540 and 1000 day-long SC (grey line) and LC K epler time series (light-red line), respectively, of KIC5596656. The black lines correspond to the best-fit model for the SC data with (full line) and without (dashed line) the Gaussian component. The red lines indicate the SC model corrected for the amplitude suppression due to the longer sampling. The vertical dotted line marks the Nyquist frequency for the LC sampling.
measurement. This is not a problem if the timescales of the variations are long compared to the integration time. For shorter timescales, however, the discretisation yields a partial cancelation of the signal, so that the resulting time series contains a damped signal. The damping is frequency dependent and becomes larger for increasing signal frequencies approaching the Nyquist frequency (given that the sampling is similar as the integration time, i.e. with little or no dead time). For the amplitude of a harmonic oscillation with a frequency ν this e ff ect can be expressed by a damping factor,
$$\eta = \sin c \left ( \frac { \pi } { 2 } \frac { \nu } { \nu _ { \text{nq} } } \right ). & & \text{compat} \\ & &.. \quad.$$
Event though this is well known (e.g., Chaplin et al. 2011a), the e ff ect is often ignored, which can lead to a quite significant underestimation of the intrinsic amplitudes, especially for the Kepler LC data. So it happens that, e.g., the amplitude of a mode at 200 µ Hz is diminished by roughly 20% ( ∼ 35% in power) in the LC data, which should no longer be ignored.
The situation is similar for the granulation signal. What we observe for granulation is a quasi-stochastic signal covering a wide range of timescales with decreasing amplitudes for increasing frequencies. To measure the granulation amplitudes we are not so much interested in the amplitude at a particular frequency but in the integrated power at all timescales covered by the observations. Consequently, there is no simple correction factor for the signal damping as it is the case for oscillation modes and it was even unclear if the granulation signal is a ff ected by the cancelation e ff ects at all. In Fig. 3 we show that it indeed is. It compares heavily smoothed power density spectra of one of the few cases where a red giant has been observed in LC as well as SC mode. Clearly, the power density of the LC time series (light-red line) drops faster towards higher frequencies than in the SC spectrum (grey line). To test that this is indeed due to the signal damping we computed a global model fit for the SC data (solid black line; see Sec. 4) and multiplied it with η 2 (red line). The resulting 'corrected' model almost perfectly reproduces the LC spectrum. In reverse, we can therefore use Eq. 1 to distort the global model during the fitting process in the same way as the intrinsic signal is distorted by the sampling in order to access the unperturbed granulation parameters. Ignoring the signal damping does barely a ff ect most of the stars in our LC sample as their oscillation and granulation timescales are mostly well below the LC Nyquist frequency. But it introduces a systematically increasing underestimation of all granulation parameters (also the timescales are a ff ected) for stars with higher ν max.
Finally, we note that the white noise component of the power density spectrum is not a ff ected by this sampling e ff ect just like the binning of a light curve does not change the white noise level in the Fourier domain.
## 4. Power density spectra modelling
Modelling the power density spectrum of a solar-type oscillator appears to be straight forward but actually is a quite complex task. Before the Kepler era, the background signal was mostly treated as a parasitic component one needs to get rid of quickly to access the oscillations. But the background signal itself includes interesting information on, e.g., the granulation or activity signal of the star. To extract this information most reliably, one needs an appropriate model whose complexity should be driven by the data quality. However, to our knowledge there is yet no consensus in the community about such a model and various approaches have been used in the past.
Some early work was done by Mathur et al. (2011), who qualitatively compared granulation parameters that result from various background models for a large sample of red giants, observed by Kepler for about 13 months. They found that even though the granulation parameters generally follow the same trends (e.g., as a function of ν max) there are significant di ff erences between the di ff erent models. However, at the time the aim of the study was not to pick a 'best model' nor were the data long enough to make firm conclusions. Now the time series are almost three times as long, enabling us to carry out a detailed comparison of di ff erent background models.
## 4.1. The background model
Originally, Harvey (1985) used a function like P ( ν ) ∝ 1 [1 / + ( πντ ) c ] to model the solar background signal, where an exponent c of 2 (i.e., a Lorentzian function) was adopted. The basic idea of this is that the granules cause a signal on the stellar surface that can be approximated by a sudden pulse with an exponential decay with a typical timescale τ . The Fourier transform of such a pulse is a Lorentzian function. Later on it was shown (using better data) that an exponent of 4 is more appropriate (e.g. Michel et al. 2009 for the Sun and Kallinger et al. 2010b for red giants) but also linear or exponential models have been used to describe the background underneath the oscillation signal. Note that a model with an exponent of 4 represents a simplification of 1 [1 / + ( πντ ) 2 ] 2 , which is the Fourier transform of a symmetrically rising and decaying pulse.
Another question is if more than one background component is required to su ffi ciently fit the observations. Each component is believed to represent a separate class of physical process such as stellar activity, di ff erent scales of granulation, or faculae, which are all strongly connected to the turbulent motions in the convective envelope. With the long time series observations of the Sun (e.g., from SOHO VIRGO; Frohlich et al. 1997) it became / evident that there is a slight depression in power density at about ν max / 2 (see, e.g., Kallinger et al. 2010b), which can not be explained by a single background component in the vicinity of the oscillations. Karo ff (2012) interpreted this as the superposed signal of granulation and faculae and similar was found in other solar-type oscillators (Karo ff et al. 2013).
The reason why the exponents of Harvey-like models and the number of components needed to su ffi ciently reproduce the observed background are poorly determined is because one needs to access the high-frequency part of the spectrum (beyond the pulsation power excess, where the power rapidly drops). Until recently, this was only possible for the Sun and a few red giants with very good SNR but for most of the previously available data of solar-type oscillators, this part of the spectrum is hidden in the white noise. Apart from the Sun, however, the particular choice of the background model was not critical for preKepler data as the blurry background signal could be almost equally well modelled by di ff erent models. But with the increasing length of the Kepler time series it becomes increasingly important to use an appropriate model that accounts for the complex structure of the backgroundand that allows to access the oscillation modes in the best possible way. We have to keep in mind that the choice of the background model potentially influences the global parameters of the pulsation power excess (the e ff ect is small but tilting the underlying background redistributes the excess power and shifts ν max) as well as the mode parameters itself (see Sec. 4.3). We therefore investigated di ff erent background models in more detail.
In all cases, the power density spectra are modelled by the superposition of instrumental noise , the contribution of one to 3 three super-Lorentzian 4 functions, and a power excess hump approximated by a Gaussian,
$$P ( \nu ) = P _ { n } ^ { \prime } + \eta ( \nu ) ^ { 2 } \left [ \sum _ { i } \frac { \xi _ { i } a _ { i } ^ { 2 } / b _ { i } } { 1 + ( \nu / b _ { i } ) ^ { c _ { i } } } + P _ { g } \exp \frac { ( \nu - \nu _ { \max } ) ^ { 2 } } { 2 \sigma ^ { 2 } } \right ], \ ( 2 ) \quad \int _ { 1 0 ^ { 3 } } ^ { 1 0 } \\ \text{where } P _ { n } ^ { \prime } \text{ corresponds to the instrumental noise contribution and } \quad \int _ { 2 } ^ { 1 0 }$$
↦
where Pn ′ corresponds to the instrumental noise contribution and ai and ci are the rms amplitude and exponent of the i -th background component, respectively. The parameter bi corresponds to the frequency at which the power of the component is equal to half its value at zero frequency and is called the characteristic frequency (which is equal to (2 πτ ) -1 , with τ being the characteristic timescale). Pg , ν max , and σ are the height, the central frequency, and the width of the power excess hump, respectively. η ν ( ) is computed according to Eq.1 and distorts the background model in the same way as the observations are distorted by the sampling. The factor ξ is used to normalise ∫ ∞ 0 ( ξ/ b ) / [1 + ( ν/ b ) c ] d ν = 1 so that a 2 corresponds to the area under the super-Lorentzian function in the power density spectrum (which in turn is equal to the variance in the time series that originates, for example, from granulation). Obviously, ξ depends on c and is for c = 2 → ξ = 2 /π , for c = 4 → ξ = 2 √ 2 /π . For c di ff erent from an even integer (i.e., c is a free parameter in the fit), the integral can't be solved analytically so that ξ is undefined. We then fitted Ai (i.e., the power density at zero frequency) instead of ξ a 2 i / bi , and numerically integrated the fit to determine ai .
↦
In the current analysis we chose to use a frequencydependent instrumental noise,
$$P _ { n } ^ { \prime } & = P _ { n } + \frac { 2 \pi \alpha ^ { 2 } / \beta } { 1 + ( \nu / \beta ) ^ { 2 } }, & ( 3 ) & \text{ively } o ^ { \text{compon} }$$
where Pn corresponds to the (constant) white noise and α and β to the amplitude and characteristic frequency of the noise component introduced by the aperture optimisation. We note that neither a constant nor a coloured noise is a ff ected by the sampling e ff ect described in Sec. 3.3, which is why Pn ′ is not corrected for η ν ( ) in Eq. 2.
## 4.2. Model comparison
To fit the models to the power density spectra we have used a Bayesian inference tool ( MultiNest ; Feroz et al. 2009) assuming that the di ff erence between the model and observed spectrum follow a χ 2 statistic with two degrees of freedom (Gabriel 1994) 5 . MultiNest provides the posterior distributions for the parameter estimation as well as a realistic global model evidence. The big advantage of this global evidence compared to other statistical tools is that it is properly normalised and evaluated over the entire parameter space. It therefore allows to reliably rate how good a given model represents a given data set compared to another model with little risk to over-fit the data, which is often not given by a direct comparison of, e.g., the maximum likelihoods. This
3 which is either 'white' or 'coloured'
4 Note that this function is often called Harvey or Harvey-like model. However, as the original Harvey model actually is a Lorentzian function we prefer the term super-Lorentzian
5 To ensure that the gap-filling has no (or only marginal) influence on this basic assumption we tested the residual signal in raw and gap-filled spectra for a number of stars and found no significant deviations.
Fig. 4. Raw (grey) and heavily smoothed (black) power density spectrum of KIC 7949599. Over-plotted in colour are the best fits model with (top panel) and without (bottom panel) the Gaussian component. Note that model A and E and model F and G e ff ectively overlap. The inserts show enlarged sections around ν max.
is because, roughly speaking, a more complex model (i.e., with more free parameters) tends to fit some data better (i.e., has a better likelihood) than a less complex model but in the Bayesian concept a model gets assigned a penalty for its complexity and needs to fit the data considerably better to get a higher model evidence than a less complex model. With this statistical measurement in hand, which tells us how good a given model represents the observations we compared the following models:
- A: the classical Harvey model with a single component and a fixed c = 2,
- B: a single super-Lorentzian function with c fixed to 4,
- C: a single component of the form 1 / [1 + ( ν/ b ) 2 ] 2 ,
- D: a single component with c being a free parameter,
- E-H: same as A-D but with two individual components
To compare the di ff erent models we obviously have to assume that at least one model is true (i.e., the total probability is 1) and can then determine the individual model probabilities to pi = zi / ∑ j z j , with z being the global model evidence as delivered by MultiNest . In Bayesian statistics, the model complexity is automatically covered by the multi-dimensional parameter volume that needs to be integrated over when computing z . This, however, depends also on the priors, as it is often the priors that determine the e ff ective volume. Even tough MultiNest uses per default uniform priors (i.e., no a priori knowledge is encoded) for all parameters, the particular choice of the parameter range that is evaluated influences z . The model evidence is not very
Fig. 5. Histograms (grey bars) and cumulative distributions (black lines) of the probability density for the parameters determined for model F (pointing upwards) and model H (pointing downwards) for KIC 7949599, showing that all common parameters are practically equivalent (within the uncertainties). Horizontal solid and dashed lines indicate the median value and the ± 1 σ limits (of a normal distribution), respectively. Vertical dotted lines mark the centre of a Gaussian fit to the histograms. The parameters units are the same as in Table 1.
sensitive to this (given that the posterior distributions are well covered), but to allow a 'fair' model comparison, a given parameter, for example ν max, needs to be evaluated over the same (or at least similar) range for all models.
We tested these models for 10 di ff erent red giants covering various ν max and exemplarily list the model probabilities and best-fit parameters for KIC 7949599 in Tab. 1, for which the model comparison gives a clear picture. This is also typical for the other stars. The first interesting result is that the original Harvey models (A and E) are completely ruled out as they can not reproduce the observed spectrum at all. This is not surprising as can be seen in Fig. 4, where we show the observed spectrum along with some of the model fits. Obviously, the slope in the high-frequency part of the spectrum, above the power excess, is much too steep to be reproduced by a model with an exponent of 2. Also interesting is that none of the models with a single background component (A to D) gets assigned a model
Fig. 6. Relative di ff erences in ν max between model F and various other models for a sample of 50 red giants. The average relative di ff erence and standard deviation is given in brackets. The top and middle panel show δν max that result from fitting the entire spectrum (i.e., global fit) while in the bottom panel the fits are restricted to the range around the power excess hump (i.e., local fit). In both cases, there are clear systematic di ff erences between 1- and 2-component models.
probability that comes even close to those of the 2 component models. One can therefore safely assume that the background signal in red giants consists of multiple components, which is in agreement with what was found for main-sequence stars (e.g., Karo ff et al. 2013). Among the eight tested models there are only two that result in a high enough probability to be considered a good representation (in our sample of models) of the data. Even though there is a preference for model H the probability contrast of about 5 between model F and H is not enough to provide significant evidence for one particular choice. According to Je ff reys (1998) an odds ratio of up to 3:1 is called weak evidence, only for > 10:1 one can speak of strong evidence. In fact, the best-fit exponents of model H are close to 4 (within ∼ 1.3 σ ) but more important here is that model F and H result in similar parameters (actually identical within the uncertainties). This can be seen from Fig. 5, where we plot histograms of the posterior probability density for all parameters of model F (upwards) and H (downwards) along with their cumulative distributions. Obviously, the posterior distributions for all (common) parameters strongly overlap with a tendency that model H results in broader distributions, i.e. larger uncertainties. The models can therefore be considered practically equivalent. Fig. 5 also illustrates that the posterior distributions are not always Gaussian. It can therefore be misleading to define the best-fit parameter and its uncertainty as the average value and its standard deviation (which is equivalent to fitting a Gaussian to the distribution). A more general definition for the best-fit parameter and its uncer-
Table 1. Summary of the background model comparison for KIC 7949599. The model probabilities are determined according to pi = zi / ∑ j z j , where z corresponds to the global model evidence and z 0 is an abritrary reference value (ln( z 0 ) = -138898). Pg is given in 1000 ppm 2 / µ Hz ai in ppm, respectively. All frequency parameters are in µ Hz. An asterisks indicates that the parameter is kept fixed during the fitting. Last digit uncertainties are given in parenthesis.
| | ln( z / z 0 ) | p | Gaussian | Gaussian | Gaussian | 1 st component | 1 st component | 1 st component | 2 nd component | 2 nd component | 2 nd component |
|----|-----------------|------------|------------|------------|------------|------------------|------------------|------------------|------------------|------------------|------------------|
| | | | P g | ν max | σ | a 1 | b 1 | c 1 | a 2 | b 2 | c 2 |
| A | -1587.7 | < 10 - 200 | 5.4(2) | 30.38(02) | 13.1(2) | 560(12) | 2.3(1) | 2 ∗ | | | |
| B | -255.7 | ∼ 10 - 111 | 4.8(3) | 35.7(3) | 5.1(2) | 624(6) | 23.7(2) | 4 ∗ | | | |
| C | -75.8 | ∼ 10 - 33 | 5.5(3) | 34.5(2) | 6.0(1) | 606(6) | 22.5(2) | 2 / 4 ∗ | | | |
| D | -243.4 | ∼ 10 - 102 | 5.1(3) | 35.2(2) | 5.7(2) | 601(28) | 20.8(4) | 3.7(1) | | | |
| E | -1592.4 | < 10 - 200 | 5.4(2) | 30.42(02) | 13.2(2) | 571(15) | 2.3(2) | 2 ∗ | 31(4) | 34.1(6) | 2 ∗ |
| F | -1.7 | 0.166 | 5.5(2) | 33.8(4) | 6.1(2) | 466(14) | 9.4(5) | 4 ∗ | 399(19) | 31.9(1) | 4 ∗ |
| G | -36.6 | ∼ 10 - 16 | 5.7(2) | 33.9(2) | 6.4(2) | 352(26) | 8.5(9) | 2 / 4 ∗ | 502(18) | 25.7(6) | 2 / 4 ∗ |
| H | -0.1 | 0.833 | 5.6(3) | 33.5(5) | 6.1(3) | 470(35) | 9.7(6) | 3.6(3) | 365(59) | 35.8(3) | 4.2(2) |
tainty limits is the value at which the cumulative distribution is equal to 0.5 (i.e., the median) and 0.5 ± 0.3414 (which is equivalent to ± 1 σ uncertainties in case of a Gaussian ), respectively. Note that, even though the uncertainties can be asymmetric we provide only a single average value for the sake of simplicity.
The result of our model comparison would have been different if we would have used the classical likelihood ratio test. The likelihood computed from the di ff erence between the best-fit models and the observed spectrum, clearly favours model H with an odds ratio of about 99:1, which is, as shown by the Bayesian analysis clearly over-interpreting the data. Or with other words, the data do not (yet) support to explicitly determine the exponent of the background model.
The main conclusion for the other 9 test stars is similar and we do not find a case that di ff ers significantly from that of KIC 7949599. For some stars in this representative sample model H is the most probable, for some model F is but for none of them the probability contrast is large enough to strongly support a specific choice. We do also not find any indication for a correlation between the specific model choice and the stellar properties like the surface gravity. Even though model H might be more realistic in some cases, we found no compelling argument to prefer model H over model F and since model F gives practically the same background parameters, is more robust (as it has less free parameters), and tends to result in smaller uncertainties, we decided to use model F for the subsequent analysis.
## 4.3. Systematic effects on ν max
The global model fit does not only allow to determine the background and global oscillation parameters but also provides an estimate for the background signal underneath the pulsation modes, which is an important ingredient for the subsequent analysis of the oscillations. As can be seen from the insert in Fig. 4 it is di ffi cult to decide which model gives the most realistic background as we can not easily disentangle the pulsation from the granulation signal and all models (apart from the classical Harvey model) seem to estimate the background equally well. However, there are small di ff erences, especially at the low-frequency end of the power excess. These di ff erences redistribute the power in the hump and potentially lead to different estimates for ν max. We therefore selected 50 red giants with a low white noise component (where about half of the stars with ν max below 70 µ Hz were chosen to be RC stars) and compare ν max that results from model F and various other models
Fig. 7. Power density spectra of three typical stars with ν max /similarequal 22, 220, and 2200 µ Hz, respectively, showing that all timescales and amplitudes (granulation as well as pulsation) scale simultaneously. Grey and black lines indicate the raw and heavily smoothed spectrum, respectively. The global fit is shown with (red) and without (blue) the Gaussian component. Green lines indicate the individual background and white noise components of the fit.
in Fig. 6. Whereas 2-component models give roughly the same ν max (see middle panel) with only minor systematic deviations, 1-component models result in ν max estimates that clearly di ff er from those determined with model F (see top panel). Obviously there is a systematic trend as a function of ν max, so that ν max from 1-component models are systematically over- and underestimated for stars with low and high ν max, respectively. The quite significant deviations of up to ∼ 6%disagree to some extend with what was found by, e.g., Hekker et al. (2012) who argued that
ν max from di ff erent fitters (using di ff erent background models) is generally in good agreement. An explanation for that might be that the individual deviations cancel out when looking at the entire sample as a whole. The mean di ff erence of the sample in Fig. 6 is always within the typical uncertainties of the individual measurements. Another reason could be that the methods mentioned in Hekker et al. (2012) that use only one background component do not try to fit the whole spectrum, but only the region around the pulsation power excess. They are therefore less sensitive to the overall structure of the spectrum (and not suitable to deliver granulation parameters) and might give a better estimate of ν max than in our analysis, where we needed to fit the whole spectrum to allow the model comparison. However, to test if the situation changes in a local analysis we redid the fits for the range 0.5 to 5 times ν max and show the results in the bottom panel of Fig. 6. Note that for model A we had to trim the spectrum even more to [0.5 - 2] × ν max, as it is not possible to follow the rapid decay of the granulation signal at high frequencies with the classical Harvey model. Obviously also a local treatment of the background can systematically e ff ect the determination of ν max although the e ff ect is reduced. We therefore advise against 1) the use of only one background component (even in a local analysis), and 2) the use of the classical Harvey model. A 1-component Harvey-like model might be applicable if the instrumental white noise is high enough so that the highfrequency part of the granulation signal is hidden therein. For stars like the one shown in the bottom panel of Fig. 7, where the contrast between the low-frequency granulation signal and the white noise is only about 10:1 (compared to about 10 :1 for the 4 star in the top panel) the functional form and number of background components become less important. In an extreme case with the white noise being at a similar level as the granulation signal even a straight line (i.e, a white component) will do.
The above analysis shows that a di ff erent treatment of the background systematically e ff ects the determination of ν max, but it does not reveal which of the background models gives the correct ν max. This question can only be answered with simulated data set with the actual ν max known, which we try in the following. As an input for the pulsation signal we use the l = 0, 1, and 2 adiabatic frequencies 6 of a 1.25 M /circledot near solarcalibrated YREC (Yale Stellar Evolution Code; Guenther et al. 1992; Demarque et al. 2008) model with a radius and e ff ective temperature chosen so that ν max is close to 50 µ Hz (where we find the largest systematic deviations, see Fig. 6). For details about the constitutive physics we refer to Kallinger et al. (2010a) and references therein. Mode amplitude ratios are modulated between 0 and 1 by a Gaussian centred on 50 µ Hz (and a width of 8 µ Hz), where we assume a mode visibility of 1.0, 1.5, and 0.5 for l = 0, 1, and 2 modes, respectively. The amplitude ratios of l = 1 and 2 modes are additionally modulated according to their inverse mode inertia (i.e., modes with a low inertia have large amplitudes and vice versa; see, e.g., Dupret et al. 2009). The l = 0 and 2 modes are assumed to have a constant mode lifetime of 30 and 40 days, respectively, and the lifetime of l = 1 modes is modulated between 100 and 200 days according to their mode inertia. To compute the artificial 1142 day-long time series (sampled each ∼ 5.9 minutes, see below) for this set of solar-like modes we add up damped and randomly re-excited oscillations, generated by the method described by Chaplin et al. (1997). In order to get realistic absolute amplitudes we rescale the artificial time series so that its rms scatter is equal to 200 ppm
6 computed with Guenther's nonradial nonadiabatic stellar pulsation program; Guenther (1994)
(which corresponds to the total pulsation amplitude, see Sec. 5.5 below) and add random scatter so that the white noise in the resulting power density spectrum is equal to 3 ppm 2 /µ Hz. Since the modes in this simulated power excess are not symmetrically distributed within the power excess, the real ν max does not have to correspond to input value of the centre. To check this, we fit a Gaussian to the simulated power excess and find the power to be symmetrically distributed around ν max ref , = 49.61 ± 0.06 µ Hz. We note that in reality the shape of the pulsation power excess is not necessarily Gaussian but as long as the power is distributed roughly symmetric (and there is no indication that it is not, see, e.g., Kallinger et al. 2010c) any symmetric function will give a good estimate of the intrinsic centre of the power excess.
For the granulation background we add two sets of random symmetric exponential pulses, with decay times that correspond to characteristic frequencies, bi = 14 and 46 µ Hz. The characteristic amplitudes ai are set to 250 ppm by rescaling the time series to the corresponding rms scatter. To also account for the sampling e ff ect described in Sec. 3.3, the original time series is oversampled by a factor of 5 and finally re-sampled to the LC sampling by averaging.
Wethen analyse the simulated spectrum between 10 µ Hz and the Nyqvist frequency. For the 2-component models, model F results in a ν max = 0.995 ± 0.01 ν max ref , and is therefore in excellent agreement with the reference value. Furthermore all other input parameters are reproduced to within ∼ 2 σ . Similar can be said for model G and H. Only for model E, the peak frequency is grossly underestimated to ν max = 0.94 ± 0.01 ν max ref , , even though we truncate the spectrum to [0.5 - 2] × ν max to get a useful fit. As we expected, the 1-component models yield peak frequencies that are significantly di ff erent from the reference value. While model A underestimates ν max even more (0.91 ν max ref , ), model B, C, and D overestimate the peak frequency to ν max = 1.03, 1.02, and 1.02 ν max ref , , respectively, where the uncertainties are always on the order of 1%.
As mentioned earlier, these deviations are presumably due to the fact that an inappropriate background model redistributes the power in the power excess and shifts the centre. If so, the e ff ect should become smaller if the shape of background signal is less distinctive, i.e. large parts of it are hidden in the white noise. This is indeed what we observe. If we increase the white noise in our simulated spectrum by a factor of 200 (which dilutes the contrast between the low-frequency granulation signal and the white noise to about 10:1) the deviations between the various models become less pronounced. While we get almost exactly the same ν max as before for model F, all other models result in a ν max that is within [0.97-1.0] × ν max ref , . Interestingly, increasing the white noise by such a large amount has only small e ff ects on the uncertainties of ν max, which are now about 50% larger than before.
This test with a simulated spectrum seems to supports our conclusions from above, namely that a model with two superLorentzian functions is suited best to reproduce the observed granulation signal and to determine ν max. Neither a 1-component model nor a model with an exponent significantly di ff erent from four should be used (or at least only under certain circumstances; see above). For completeness we also simulated a star with a much higher ν max = 150 µ Hz and find the peak frequencies from the various models to more or less converge to the same value, which is in agreement with what can be seen in Fig. 6.
## 4.4. Fitting the sample
As mentioned above we fit model F to our sample of LC and SC data. In addition to the two components in the vicinity of the power excess, we apply a third component to account for long-period signal intrinsic to the star (e.g., due to activity) with periods that are usually much longer than 10 / ν max, but also residual instrumental signal. From previous work (e.g., Mathur et al. 2011) it is know that basically all (intrinsic) parameters scale with ν max, for which we have already good initial guesses from, e.g., Kallinger et al. (2012). To optimise the computational performance we tried to keep the parameter ranges relatively narrow. During the fit ν max is kept within ± 10% of its initial guess. The characteristic frequencies bi are allowed to vary between 0.01 to 0.3, 0.1 to 0.7, and 0.7 to 2 times the initial guess of ν max. All amplitude parameters ( ai and α ) are kept below the rms scatter of the time series (i.e., the square root of the integrated power in the spectrum). The height and width of the Gaussian are allowed to vary between 0 to half the maximum power in the spectrum and 0.05 to 0.3 times the initial guess of ν max, respectively, and β may vary between 0 and the Nyquist frequency. MultiNest then delivers the posterior probability distributions for the 12 free parameters. We check if they are well sampled within the parameter range and finally marginalise them to determine the most probable parameters and their 1 σ uncertainties. One parameter that is not directly fitted but evaluated in the subsequent analysis is the so-called total pulsation amplitude A puls . It is supposed to represent the entire pulsation energy in the power excess, i.e., the sum of all individual mode amplitudes. During the fitting procedure we compute A puls as ( σ Pg √ 2 π ) 1 2 / , which is the square root of the area under the Gaussian component of our model, for each evaluated combination of Pg and σ to build a posterior distribution from which we determine the most probable value of A puls and its uncertainty.
The long Kepler time series allowed us to determine ν max on average to within about 0.9%, the characteristic granulation frequencies b 1 and b 2 to within about 5.3 and 3.6%, respectively, the granulation amplitudes a 1 and a 2 to better than 3.2 and 5.2%, respectively, and A puls to within about 2.8% on average.
In Fig. 7 we show the power density spectra of three representative stars, whose ν max di ff er by a factor of ∼ 10 from one to the next, along with the best-fit model (F) and its individual components.
## 5. Scaling relations
## 5.1. Granulation timescales (or frequencies)
Using a few basic physical assumptions one can estimate the granulation timescales and amplitudes from the stellar parameters. For example, Kjeldsen & Bedding (2011) and Mathur et al. (2011) argued that convection cells cover a vertical distance that is proportional to the atmospheric pressure scale height, Hp ∝ T e ff / g , at a speed approximately proportional to the speed of sound, cs ∝ √ T e ff , where g is the surface gravity and T e ff is the e ff ective temperature. Therefore the characteristic timescale of granulation or more conveniently the characteristic frequency (since we measure them in a frequency spectrum) can be expressed as ν gran ∝ cs / Hp ∝ g / √ T e ff ∝ ν max, or with other words there should be a tight relation between the characteristic granulation frequency and ν max that should be linear in a first approximation. This is indeed what we observe. The top panel of Fig. 8 shows that there is a tight relation between the characteristic frequencies ( b 1 and b 2 ) and ν max, covering a range in ν max of more than 3 orders of magnitude from the Sun to high up the giant branch. A power law fit (black lines) reveals that both relations are almost linear as both exponents are close to 1. The best fit coe ffi cients and their uncertainties are listed in Tab. 2. Furthermore we find that the dispersion along this relation is with about 10 and 9% only about twice as large as the average uncertainties of the individual measurements. This indicates that a power law is too simplistic but also that other properties (like the evolutionary state, chemical composition, magnetic activity, etc.) play only a minor role.
The uncertainties of the fit coe ffi cients might appear unrealistically small but we note that they are based on uncertainties in bi as well as ν max. For the fit we again used MultiNest , which can only consider uncertainties in the dependent variable (as most other fitting algorithms). To account also for the uncertainties of the independent variable we add normally distributed uncertainties to ν max (with σ being equal to the actual uncertainties of the individual measurements) and redo the fit adding up the probability density distributions. We iterate this procedure until each fitting parameter converges to a value that stays within one fifth of the corresponding uncertainty. Additionally we note that the high correlation coe ffi cients (see Tab. 2) between bi and ν max indicate that one can accurately constrain the power-law parameters.
## 5.2. Granulation amplitudes
Using similar assumptions as above, Kjeldsen & Bedding (2011) argued that the granulation power at ν max scales as P gran ( ν max) ∝ ν -2 max . However, P gran is not really qualified to serve as an 'amplitude' parameter for granulation as it gives the power at a certain frequency and not, e.g., a proxy for the total energy that is contained in the granulation signal. More appropriate is the rms amplitude a , which is defined as the square root of the integrated granulation signal in the power density spectrum and represents the rms intensity fluctuations in the time series that are due to granulation. From Eq. 2 follows that P gran ( ν max) ∝ ( a 2 / b ) / [1 + ( ν max / b ) c ]. With P gran ( ν max) ∝ ν -2 max and b ∝ ν max we then infer that a ∝ ν -0 5 . max . Basically this comes from the fact that the intensity fluctuations scale as the inverse of the square root of the number of granules on the stellar surface (e.g. Ludwig 2006). This is again similar to what we observe. The bottom panel of Fig. 8 illustrates the rms amplitude of both granulation components ( a 1 and a 2 ) as a function of ν max showing that both components have comparable amplitudes that are tightly correlated to ν max. A power law fit (green line) to both amplitudes reveals that a ∝ ν -0 61 . max , which is close to what we expected (based on some very basic assumptions) and to what has been observed by others (e.g., Hekker et al. 2012, who fitted the variance of a sample of Kepler red giants, which is dominated by the granulation signal). In contrast to the granulation frequency scaling, the dispersion along the fit is with about 16% significantly larger (three to five times) than the average uncertainties of the individual measurements. This indicates that the granulation amplitudes are primarily correlated to ν max (which in turn is dominated by the stellar radius) but also other parameters like the stellar mass and or the chemical composition po-/ tentially have a significant influence on the granulation amplitude (see, e.g., Mosser et al. 2012a). To test this we correct the granulation amplitudes for the dependency on ν max and show the residuals as a function of stellar mass (insert in the bottom panel of Fig. 8). Stellar fundamental parameters have been computed with the method described by Kallinger et al. (2010b), where we
Table 2. Parameters of power law fits to the granulation. The surface gravity g is in cgs units, frequency parameters ( ν max, b 1 , and b 2 ) are in µ Hz, τ e ff is in seconds, and amplitude parameters ( a , A gran , and A puls ) are in ppm. The stellar mass M is in solar units and T = T e ff / 5777. σ indicates the rms variations of the residuals in percent. If a coe ffi cient is left blank means that the corresponding parameter is not included in the fit. Last digit uncertainties are given in parenthesis. ρ Yi , Xi is the magnitude of the correlation coe ffi cient (in percent) between Xi = log 10 ( xi ) and Yi = log 10 yi , with yi = y 1 / ∏ i -1 j = 0 xj = k x c i , with x 0 = 1. The index i indicates the independent variable (from left to right) of the corresponding power law fit.
| | b 1 = k ν s max | b 2 = k ν s max | a = k ν s max M t | a = k ν s max M t | τ e ff = k ν s max = k g s T t | τ e ff = k ν s max = k g s T t | A gran = k ν s max |
|-------------|-------------------|-------------------|---------------------|------------------------|----------------------------------------------|----------------------------------------------|----------------------------------------------|
| k | 0.317(2) | 0.948(3) | 3382(9) | 3710(21) | 836(4) | 2.00(4) · 10 6 | 3335(9) |
| s | 0.970(2) | 0.992(2) | -0.609(2) | -0.613(2) | -0.886(2) | -0.853(3) -0.41(5) | -0.564(2) |
| t σ | 10.2 | 8.7 | 16.5 95 | -0.26(3) 14.4 44 | 11.0 99 | 9.3 99,52 | 13.5 |
| ρ Y i , X i | 99 | 99 | | | | | 98 |
| | A gran = k g s | M t T | A gran = k g s | A puls = k g s M t T u | A gran = k A s puls T t = k A s puls g t M u | A gran = k A s puls T t = k A s puls g t M u | A gran = k A s puls T t = k A s puls g t M u |
| k | 9799(43) | 10887(127) | 11054(385) | 9344(348) | 4.57(3) | 4.79(4) | 36(3) |
| s | -0.549(6) | -0.555(6) | -0.556(4) | -0.657(4) | 0.855(3) | 0.837(3) | 0.63(1) |
| t | | -0.24(2) | -0.24(3) | -0.35(3) | | 0.24(3) | -0.15(1) |
| u | | | 0.05(9) | 0.04(8) | | | -0.06(1) |
| σ | 13.1 | 11.0 | 10.7 | 13.6 | 8.0 | 10.3 | 7.3 |
| ρ Y i , X i | 97 | 44 | 80 | 97,41,76 | 99 | 57 | 99,91,25 |
use ν max, ∆ ν , and T e ff as an input for the LC sample and ν max and T e ff for the SC sample. E ff ective temperatures are taken from the Kepler Input Catalog (Brown et al. 2011) and corrected according to Thygesen et al. (2012).
Obviously there is a correlation with mass so that high-mass stars tend to have smaller amplitudes than low-mass stars. To account for this additional mass dependency ( ν max already depends on mass) we add a term and fit a = k ν s max M t instead of a = k ν s max , which results in a considerably better fit (the global evidence of the MulitNest fit is many orders of magnitudes better). The best fit coe ffi cients and their uncertainties (determined with the procedure described in Sec. 5.1) are listed in Tab. 2.
trum) one can define an e ff ective timescale τ e ff as the e-folding time of the autocorrelation function (ACF) of the time series (e.g., Mathur et al. 2011). The characteristic timescale of any description of the granulation signal in a power spectrum measures the temporal correlation of the signal in the time domain, which can also be expressed by the width of the ACF. Recalling that the autocorrelation of a signal corresponds to the Fourier transform of its power spectrum, we can determine τ e ff of our twocomponent granulation model by computing the time where the Fourier transform (i.e., the ACF) of the two superposed granulation components drops below e -1 . The result is shown in the top panel of Fig. 9, where we choose to plot τ e ff as a function of log g instead of ν max, as the surface gravity is a more basic stellar property and easy to determine for the models.
## 5.3. Characteristic timescale, τ e ff
For the Sun, any background signal between a few 100 to a few 1000 µ Hz is usually attributed to granulation (see e.g. Karo ff et al. 2013, for a summary of various interpretations found in the literature). A major problem in this frequency range results from a kink that is visible in the background spectrum just below the oscillation power excess. This feature has been first identified in the solar irradiance data from the SOHO VIRGO / instrument (e.g., Michel et al. 2009, and references therein) but has meanwhile been observed by Kepler in other stars ranging from red giants (e.g., Mathur et al. 2011) to main-sequence stars (e.g., Karo ff et al. 2013). The depression in power at about ν max / 2 (see Fig. 7) arises from the fact that the background signal in the vicinity of the oscillation signal consists not of one but two statistically significant (see Sec. 4.2) individual components. One component certainly can be attributed to granulation but the physical origin of the second component is not yet clear and still subject of debates (see e.g., Samadi et al. 2013a,b, for a summary) and is therefore missing in theoretical models. Thus a direct comparison between our measured granulation parameters and those from theoretical models is di ffi cult as they measure di ff erent things.
To still be able to compare characteristic granulation timescales obtained from models and observations (or also from observations using di ff erent descriptions of the granulation spec-
From a theoretical point of view and with 3-D modelling, Samadi et al. (2013b) suggested that the characteristic granulation timescale can be described as τ e ff ∝ ( ν max M a ) 1 , where M a is the turbulent Mach number in the photosphere. The Mach number is di ffi cult (or even impossible) to determine observationally but the authors also found that it approximately scales as M ∝ a T 2 4 . e ff / g 0 15 . . With this and with ν max ∝ g / √ T e ff the above theoretical description translates to τ e ff ∝ g -0 85 . T -1 9 . e ff . A powerlaw fit to the observation reveals,
$$\underset { \text{one} } { \text{ang} } \quad \tau _ { \text{eff} } \in g ^ { - 0. 8 5 } T ^ { - 0. 4 },$$
where T = T e ff / 5777 K . Interestingly, the scaling with g is exactly as expected. Only the temperature dependency is much weaker than theoretically anticipated. A potential problem in this analysis are the rather large systematic uncertainties of the KIC temperatures. We therefore redid the analysis for the 624 stars in our sample for which well-calibrated e ff ective temperatures from Sloan Digital Sky Survey photometry (Pinsonneault et al. 2012) are available but found no significant di ff erences in the fit. Nonetheless, our sample is dominated by red giants, which naturally cover only a relatively small range in e ff ective temperature (a few hundred K at a given log g , with typical uncertainties of 150K for individual stars). As a result it is di ffi cult to define any temperature dependency in our sample of Kepler stars. However, the dispersion along the fit is with about 9% rather small so that
Fig. 8. Characteristic frequencies (top) and rms amplitudes (bottom) as a function of ν max for our sample of LC (red orange) and / SC (blue magenta) stars. Orange and magenta coloured symbols / correspond to the lower frequency component ( a 1 and b 1 ). Black lines indicate power law fits (see Tab. 2) and red open circles solar values. The insert shows the rms amplitudes of the LC sample divided by the power law fit as a function of stellar mass.
we can expect no other properties of the star to play a significant role. For completeness we also fit τ e ff as a function of ν max and find it to approximately scale as τ e ff ∝ ν -0 89 . max , which is in perfect agreement to what was found by others (e.g., Mathur et al. 2011).
For comparison we show a power law fit to the theoretical values of τ e ff (dashed line) presented by Mathur et al. (2011) and obtained from 3D simulations of convective red-giant atmospheres (Trampedach et al. 2013). They do not provide the e ff ective temperatures of their models, which is why we restricted the fit to the surface gravity. Interestingly, the exponent is exactly what we obtain from the observations, only the absolute values are smaller by about 60%. This shift in absolute value is likely caused by the fact that we fit the observed granulation signal with two individual components, whereas the theoretical spectra are fitted only with one component as the kink around ν max / 2 is (due to presumably missing physics in the models) not present in the theoretical spectra. If we would fit the observations with only one component, as we have done for, e.g., KIC 7949599 in our background model comparison, the resulting τ e ff is with ∼ 1 17 . · 10 4 s about 60% smaller than the ∼ 1 99 . · 10 4 s that we determine for a two-component fit.
Samadi et al. (2013b) computed granulation timescales and intensity fluctuations for a grid of 3D hydrodynamical models based on an approach described by Samadi et al. (2013a). Their grid also includes three K-type and three F-type dwarf models. While the K-type models are not covered by our observations,
Fig. 9. Characteristic timescale τ eft (top) and intensity fluctuations A gran (bottom) as a function of log g for our sample of LC (red) and SC (blue) stars. Solid lines correspond to power law fits. The dashed line in the top panel corresponds to a power law fit to e ff ective timescales obtained from 3D simulations (Mathur et al. 2011). Diamond symbols indicate values obtained from the granulation spectrum of 3D hydrodynamical models (Samadi et al. 2013b) and the dotted line corresponds to a power law fit to these models. The dashed line in the bottom panel shows the relation between the 8hr-flicker amplitude and log g .
the F-type models are known to systematically overestimate the granulation amplitudes (see also Ludwig et al. 2009). We therefore exclude them from the further analysis and plot the remaining models in Fig. 9. From a power-law fit we find the characteristic timescales of the models to be almost perfectly reproduced as τ e ff ∝ g -0 86 . T -1 2 . (the residuals scatter by less than 3%). The correlation with g is fully compatible with the observed one (Eq. 4) as well as with the theoretically expected. The temperature dependency, however, is much stronger as for the observations but weaker than originally anticipated. This might be because the models cover a di ff erent range in mass (1.4 - 4 M /circledot compared to ∼ 0.7 - 2.5 M /circledot and therefore di ff erent temperatures) on the RGB and a larger range in e ff ective temperature ( ∼ 1200K compared to a few hundred K) on the main sequences than the Kepler sample and the resulting scaling might therefore be of di ff erent sensitivity to the e ff ective temperature. To solve this discrepancy is, however, beyond the scope of this paper. As for the Mathur et al. (2011) simulations the absolute values of model τ e ff underestimate the observations (by about 40 %), which is again likely due to the missing kink in the theoretical spectra. Hence, apart from the missing kink and a minor issue with the temperature dependency (for the observed sample, changing the exponent from -0.4 to -1.2 changes τ e ff by a few percent at most), the agreement between the observations and the model predictions are surprisingly good.
## 5.4. Intensity fluctuation, A gran
The rms amplitudes we measured for the two granulation components represent the rms intensity fluctuations in the time series that originate from two physical processes on the stellar surface and that are both likely associated to granulation. It is however not relevant for our purpose to identify the exact physical process that causes each part of the signal. We are more interested in the total background signal (or so to say a proxy for the total energy) locally around ν max and therefore have to merge the two individual amplitudes to a single value that can be compared to other observations or model predictions. This is straightforward and we define the total bolometric intensity fluctuation due to granulation as A 2 gran = C 2 bol ( a 2 1 + a 2 2 ), where C bol is a bolometric correction that scales for the Kepler bandpass as C bol = ( T e ff / T 0 ) α , where T 0 = 5934K and α = 0 8 (Ballot et al. 2011; Michel et al. . 2009). The estimated uncertainties of ± 250K in T e ff typically add about 1-2% uncertainty to A gran , which are then about 7% on average.
In a first step we fit A gran as a function of the peak frequency and find it to approximately scale as A gran ∝ ν -0 56 . max . This roughly corresponds to P gran ∝ ν -2 1 . max and is therefore di ff erent from what was found by, for example, Mathur et al. (2011), who derived an exponent of 1.89.
In the bottom panel of Fig. 9 we show A gran as a function of log g for our total sample of stars. A power law fit indicates that A gran ∝ g -0 55 . , which is in agreement to what one would expect. However, there is some structure in the correlation indicating a more complex scaling relation and there are a number of stars that fall significantly below the fit. We therefore follow the same approach as in Sec. 5.2 and add a mass term to the power law, which again considerably improves the fit (in terms of global evidence of the fit and the rms scatter of the residuals). The best fit coe ffi cients and their uncertainties are listed in Tab. 2.
We could not compare our measurements to the models presented in Mathur et al. (2011) as the authors did not provide estimates for the intensity fluctuations. Samadi et al. (2013b), however, do and we plot their values in the bottom panel of Fig. 9 for comparison. Whereas the models approximately reproduce the observations on the red-giant branch, they are di ff erent for mainsequence stars. For a given log g , the models cover a much larger range in A gran than the observations. This might be because A gran is supposed to be also sensitive to the e ff ective temperature and the main-sequence models cover a wider range in T e ff . To test this we add a temperature term to our power law and find the intensity fluctuations to scale as A gran ∝ g -0 56 . M -0 24 . T 0 05 . . Even though we find a quite weak temperature dependency, the global evidence of the fit is much better than for the fit without the temperature term indicating again an improvement in the scaling relation. For the sake of simplicity we ignore the temperature dependency and the granulation amplitude can then be approximated to scale as
$$\begin{smallmatrix} A _ { \text{gran} } \circ ( g ^ { 2 } M ) ^ { - s }, & & & & & ( 5 ) & \text{dampin} \\ \cdot & \cdot \cdot \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot \end{smallmatrix}$$
where s /similarequal 1 / 4. We note that it might be more convenient to express this scaling as A gran ∝ R M / 3 4 / (since g already depends on M ) but prefer it the way given above as g is the parameter that can be determined best from the observations and the mass dependency is only weak. Eq. 5 is, however, incompatible with the classical scaling relation (e.g., Kjeldsen & Bedding 2011) A gran ∝ ( ν max M ) -1 2 / T 3 4 / , which translates into A gran ∝ T gM ( ) -1 / 2 .
As for the characteristic timescales, we fit our observational scaling relation to the model predictions from Samadi et al.
Fig. 10. Total pulsation amplitude as a function of log g for our sample of LC (circles) and SC (squares) stars, with the colour encoding stellar mass. Black crosses indicate pulsation amplitudes predicted from a power law fit of the form A puls ∝ ( g M 2 ) -s (shifted by a factor of 2 for better visibility).
(2013b) and find them to approximately scale as A gran ∝ g -0 88 . M -0 61 . T 6 0 . . Apart from the di ff erent exponents for g and M , the strong dependancy on T with such a high power explains the large dispersion in Fig. 9 but is not compatible to what we find for the observations. Part of the discrepancy should come from the di ff erences of overall stellar properties between our sample and the synthetic models.
In Fig. 9 we also indicate the relation between the 8-hr flicker amplitude and log g from Bastien et al. (2013), who demonstrated that the rms scatter of the Kepler time series (after applying a 8 hour high-pass filter) can be used to accurately estimate the surface gravity of stars that show a surface granulation signal. Interestingly, their flicker amplitude almost perfectly resembles our A gran scaling for stars with a log g between about 3 to 3.5, indicating that they indeed measure the intensity fluctuations due to granulation. For stars with smaller or larger log g , however, the flicker amplitude represents only a part of the granulation amplitude. We note that this method is not applicable for many of the red giants because the fixed 8 hour filter suppresses the granulation (and pulsation) signal in stars with ν max /lessorsimilar 35 µ Hz (i.e., with periods longer than 8 hours).
## 5.5. Pulsation amplitudes
Oscillation amplitudes are a di ffi cult to measure and model asteroseismic quantity. From a theoretical point of view, the amplitude of a mode is determined by the competing driving and damping mechanism, involving rather complex physics (e.g., Houdek et al. 1999; Samadi et al. 2007). In solar-type oscillating stars, the outer convective layer is believed to drive modes with a resonant frequency v CV / Hp , where v CV is the convection velocity, and a mode energy that is roughly equal to the kinetic energy of a single convection eddy. The underlying physics is still not properly understood but a number of scaling relations aiming to predict mode amplitudes by scaling from the Sun's values have been derived and discussed (see, e.g., Corsaro et al. 2013, and references therein). Testing such scaling relations with observational data is vital for a better understanding of stellar oscillations but not only their physical interpretation is still a mat-
ter of debate, also the actual measurements are far from being straightforward to perform.
Ideally, it would require to fit a sequence of Lorentzian profiles on top of an appropriate background model to determine the individual mode amplitudes. This is already a challenging task for individual stars with good SNR and frequency resolution (e.g., Gruberbauer et al. 2009) and currently practically impossible for a large sample of stars. In practice, the power density spectra are heavily smoothed, corrected for the background signal and converted to amplitude per oscillation mode. This requires not only a good knowledge of the granulation background (which is often not the case, see Sec. 4) but is also based on the assumption that modes of di ff erent degree having a di ff erent spatial response are excited to the same intrinsic amplitude (e.g., Kjeldsen et al. 2008), which is also often not the case (see next. Sec.). Furthermore, sampling e ff ects as described in Sec. 3.3, which potentially introduce significant systematic uncertainties are frequently not accounted for. Alternatively, we use what we call the total pulsation amplitude A puls , which is a direct output of our power spectra modelling and which represents the sum of the amplitudes of all excited modes (even those that are not detectable individually) and is therefore a good measure for the total intensity fluctuations due to oscillations that we observe on the stellar surface.
From theoretical considerations it is assumed that the bolometric mode amplitude scales as ( L M p / ) ( T e ff ) -t , where it is not yet fully clear whether or not it scales with L and M to the same power and what are the specific values of the exponents (e.g., Stello et al. 2011; Huber et al. 2011; Corsaro et al. 2013). For the vast majority of the rather faint stars observed with Kepler , however, the luminosity is simply unknown (or poorly determined via the seismic radius and e ff ective temperature) and it is therefore di ffi cult to test such a scaling relation. Instead we prefer to adopt the e ff ective temperature and surface gravity as independent variables since L M / ∝ T 4 e ff / g (given g ∝ M R / 2 and L ∝ R T 2 4 e ff ). Huber et al. (2011) and Stello et al. (2011) suggested independent exponents for L and M , which means that we have to add a mass term to our scaling relation. This can also be seen from Fig. 10, where we plot A puls as a function of log g , with the stellar mass colour-coded. Clearly, mass a ff ects the pulsation amplitudes so that low-mass stars tend to have larger amplitudes than high-mass stars at a given log g . A power law fit indicates that the total (bolometric) pulsation amplitude roughly scales as A puls ∝ g -0 66 . M -0 35 . T 0 04 . . As for the granulation amplitude, MultiNest found the exponent of the temperature term to be close to zero, which means that the pulsation amplitude of a star can be su ffi ciently well explained by the stars mass and surface gravity (i.e., mass and radius) and T e ff has no (or only marginal) e ff ect. Therefore we can approximate the pulsation amplitude to scale as
$$A _ { p u l s } & \in ( g ^ { 2 } M ) ^ { - s }, & ( 6 ) & \text{ at } \nu _ { \max } 1 \\ & \text{tions fo} \\ & \sim \cdot.$$
where s /similarequal 1 / 3. Substituting g by M R / 2 and R 2 by L T / 4 , this translates to
$$A _ { p u l s } \in R ^ { 4 / 3 } / M \in ( L ^ { 2 } / M ^ { 3 } ) ^ { s } T ^ { - 8 s }, & & ( 7 ) \quad \underset { \bullet \, \dots } { \text{To } }$$
showing that L and M indeed have di ff erent exponents. To test the statistical significance of this we also fitted a power law A puls ∝ g s T -t (i.e. without an additional mass term and therefore L M / with a single exponent) and find the odds ratio between the two model's global evidence clearly in favour for the scaling relation including a mass term.
In principle Eqs. 6 and 7 are equivalent but from an observational point of view, Eq. 6 should be preferred as it uses easier to determine parameters.
The relative scatter around the fit of about 13% leaves not much space for other properties of the star to significantly a ff ect the pulsation amplitudes. However, there are a number of stars that have substantially lower total amplitudes than what one expect from their M and log g . This could be caused by, for example, binaries or stellar activity damping the oscillations (see, e.g., Huber et al. 2011).
From Fig 10 it appears the low-mass stars are overabundant on the upper giant branch (log g /lessorsimilar 2.0) compared to the stars below the red clump (at log g ∼ 2.3). This is presumably due to the mass loss that stars undergo when evolving up the giant branch (see, e.g., Miglio et al. 2012).
## 6. Granulation versus pulsation amplitudes
Solar-like oscillations are excited by convection, which is also the process responsible for granulation. The basic assumption is that the power of the velocity fluctuations due to p-mode oscillations scales with stellar parameters in the same way as the power of the velocity fluctuations due to granulation. This is roughly supported by solar observations indicating that both the kinetic energy of an oscillation mode and of a single granule are about 10 27 ergs (Korzennik et al. 2012). Many authors (e.g., Kjeldsen & Bedding 2011) have previously argued that amplitudes of solar-like oscillations should scale in proportion to fluctuations due to granulation. This is also supported by our findings, where we show that the pulsation amplitudes scale to nearly the same power of g and M as the granulation intensity fluctuations. It is therefore obvious to correlate these two parameters.
This is done in Fig. 11, where we show the amplitude of the granulation intensity fluctuations as a function of the total pulsation amplitude. Even though there is a tight relation, there is no equality nor is it a linear relation. A power law fit indicates that A gran ∝ A 0 86 . puls . The bottom panel of Fig. 11 shows that while the granulation amplitude is about 3 to 4 times larger than the pulsation amplitude for stars on the main sequence, the ratio drops down to about 1.5 for stars high up on the giant branch.
The fact that there is no linear relation between the granulation and pulsation amplitude is somewhat surprising as one would intuitively expect that a more vigorous convection causes a stronger granulation signal in the same way as larger pulsation amplitudes. On the other hand we have to keep in mind that the amplitude of solar-like oscillations depends on both the excitation (i.e., the amount of energy that is provided by convection) as well as the damping rate (i.e., the mode lifetime). Kjeldsen & Bedding (2011) postulated that the squared p-mode amplitude in velocity scales with the granulation power at ν max times the mode lifetime. E ff orts to define a scaling relations for the mode lifetime have been made (Chaplin et al. 2009; Baudin et al. 2011; Corsaro et al. 2012) but given the fact that measuring mode lifetimes is still quite challenging no consensus has been found so far. It is, however, well established that the mode lifetime is a function of temperature.
To account for this we add a temperature term and find that the granulation intensity fluctuations approximately scale as A gran ∝ A 0 84 . puls T 0 24 . . Even though there is a relatively good correlation between A gran / A s puls and T (see Tab. 2), including the temperature significantly impairs the fit. This is reflected by the increasing rms scatter of the residuals from ∼ 8 to ∼ 10.3% as well as a much lower global model evidence. A much better de-
Fig. 11. Granulation intensity fluctuations A gran as a function of the total pulsation amplitude A puls . Symbol colours are the same as in Fig. 8. The dotted and solid line indicate unity and a power law fit, respectively. Black crosses indicate the ratio A gran / A puls predicted from a power law fit A gran ∝ A s puls g t M u (shifted by 1.5 in y-direction for better visibility). Note that the fit does not include the weak dipole mode stars.
scription of the correlation between the granulation and pulsation amplitude can be given if we include the surface gravity and mass in the fit. We find the granulation amplitude to scale as A gran ∝ A 0 63 . puls g -0 15 . M -0 06 . , which is consistent with what one can expect when comparing the individual scalings for the granulation and pulsation amplitudes.
## 6.1. Weak dipole modes
Fig. 11 shows an interesting feature with a group of stars deviating from the general trend. From a pulsation amplitude of about 200ppm downwards there exists a group of stars that seem to have substantially larger granulation amplitudes than expected. A closer look on the individual power spectra, however, reveals that it is the total pulsation amplitude that is smaller than it should be for 'normal' stars. This is because the dipole modes are much weaker (or even totally suppressed) than the surrounding l = 0 and 2 modes, which is reflected in the integrated amplitude as there is some power 'missing'. The origin of this phenomena is unknown but Mosser et al. (2012a) and only recently Garc´ ıa et al. (2014) found that it is not correlated with, e.g., the temperature or chemical composition of these stars, and Mosser et al. (2013a) argue that it likely results from a very e ffi -cient coupling between pressure and gravity waves.
To check whether this e ff ect is also reflected in the granulation signal that excites the modes, we need to estimate how much pulsation power is missing. We therefore compute the average amplitude of the central two l = 1 modes and the central three l = 0 and 2 modes. Following Mosser et al. (2012a) the individual modeamplitudes are given as A l 2 ( ν ) = δν ∑ δ + δ -[ P ( ν ) -B ( ν )], where δν is the bin width of independent frequency bins with the power density P ( ν ) and background level B ( ν ). In the case of l = 0 and 2 modes, the spectrum is evaluated between δ ± = ν m ± δν 02 / 2 (with ν m and δν 02 being the mode frequency and the small frequency separation, respectively) and in the case of dipole modes, from δ -= ν 0 + δν 02 / 2 to δ + = ν 2 -δν 02 / 2 (with ν 0 and ν 2 being the adjacent l = 0 and 2 mode frequencies). The amplitude ratio A 1 / A 0 is given in Fig. 12 and clearly shows that stars with ν max /greaterorsimilar 60 µ Hz split into two groups . While normal 7 stars have an amplitude ratio of ∼ 1.4 on average, A 1 / A 0 reduces to ∼ 1.0 to 0.5 (depending on ν max) in weak dipole-mode stars. Assuming that the total pulsation amplitude is roughly proportional to A 0 (1 + A 1 / A 0 + A 2 / A 0 ) we can estimate the A puls is underestimated by about 15 to 40% compared to normal stars. Adding this 'missing' pulsation amplitude in Fig. 11 would shift the weak dipole-mode stars into (or at least close to) the population of normal stars. From this we conclude that the granulation signal acts normal in such stars confirming the early result based on KIC 8561221 in which the background was also found to be normal (Garc´ ıa et al. 2014). This needs, however further work, which is beyond the scope of this paper. We therefore excluded these stars from our analysis.
## 7. Testing the scaling relations on the Sun
A good possibility for an independent test of the above deduced scaling relations is given by the Sun. We use a 1-year time series from the green channel of the SOHO VIRGO data / (Frohlich et al. 1997) obtained during the solar activity minimum before Cycle 23 and fitted our model F to the corresponding power density spectrum. The resulting best-fit parameter are listed in Tab. 3, where A gran , /circledot and A puls , /circledot are bolometrically corrected according to Michel et al. (2009). From this analysis it
7 A 2 / A 0 does not show this e ff ect, which is why we do not plot it in Fig. 12. Its average value is about 0.85.
Fig. 12. Ratio between the average amplitude of the central two l = 1 and the central three l = 0 modes, showing a group of stars with atypically small dipole modes ( Al = 1 / Al = 0 < 1). Inserts show the power density spectum of two with red open circles marked stars, where coloured bars label l = 0 (lightred) and 2 (lightgreen) modes.
Table 3. Solar reference values and the corresponding parameter predicted from scaling realtions. The predicted amplitudes a 1 , /circledot and a 2 , /circledot are multiplied by 1.33 to account for the di ff erent passbands of Kepler and SOHO VIRGO. / A gran , /circledot and A puls , /circledot are bolometric values. The units are the same as in Tab. 1 and 2.
| | solar values | scaling relation |
|-------------------------------------------------------------------------------------------------------|---------------------------------------|-------------------------|
| P g , /circledot ν max , /circledot σ /circledot | 0.25(1) 3140(4) 361(4) | |
| a 1 , /circledot b 1 , /circledot a 2 , /circledot b 2 , /circledot τ e ff , /circledot gran , puls , | 36.5(2) 758(8) 35.9(2) 2468(9) 375(4) | 35.5(7) 782(11) 35.5(7) |
| A /circledot A /circledot | 41.0(2) | 340(12) |
| | | 2787(43) |
| | | 38(2) |
| | 12.01(6) | 11.4(6) |
becomes again obvious that the specific treatment of the granulation background has an important impact on the determination of ν max. Whereas Kallinger et al. (2010b) fitted the same model to the same data, they did not arrive at the same ν max. This is presumably due to the fact that we now do also account for the sampling e ff ects described in Sec. 3.3, which redistributes the power in the power excess and shifts ν max from 3120 ± 5 to 3140 ± 4 µ Hz.
In Tab. 3 we also give the values that result from our scaling relations and find them in good agreement with the actual measurements. Note that the predicted amplitudes a 1 , /circledot and a 2 , /circledot are given for the SOHO VIRGO bandpass (centred on 500 nm), / where we assume a simple linear transformation from the Kepler bandpass (centred on 664 nm).
## 8. Summary
In this work we investigated the granulation background spectrum of a large and homogeneous sample of 1364 stars observed by Kepler . The sample includes stars from the main sequence, the sub-giant branch to stars on the ascending giant branch, in the red clump, and on the asymptotic giant branch. Or with other words, stars with a mass ranging from about 0.7 to 2.5 M /circledot that are cooler than the red border of the classical instability strip, that cover a large fraction of the parameter space for which we can expect convective surface layers that exhibit solar-like oscillations and granulation. We used Kepler light curves that span between 90 and 360 days for main-sequence stars and sub-giants and about 1140 days for red giants, corrected them for various instrumental e ff ects and analysed the overall structure of the resulting power density spectra. From this study we find that:
- -The depression in power at about ν max / 2 that is know from the Sun and a few other main-sequence stars is a common feature in all stars of our sample. A plausible explanation for this feature is the presence of a second background component close to the pulsation power excess. The fact that the position of the depression relative to the power excess is roughly constant for all stars indicates that the characteristic timescales and amplitudes of the two components have a fixed ratio.
- -Using two (instead of one) background components significantly improves the estimate of the background underneath the oscillation signal. A single component is only su ffi cient if the instrumental white noise dominates the background (or
- is at least not negligible as for many of the Kepler targets) hiding the specific shape of the granulation signal.
- -In a probabilistic comparison of various functional forms of background models a super-Lorentzian function with a free (but close to 4) exponent turned out to reproduce the observed signal best for many stars. The data do, however, not (yet) provide enough evidence for such a model and a model with an exponent fixed to 4 (which represents a significant simplification for the fitting) can be considered as equally good.
- -The specific choice of the background model influences the determination of ν max. An inappropriate background model can redistribute the power in the pulsation power excess and systematically shift its centre. We find such shifts in a local (around ν max) as well as a global analysis of the power spectrum and note that the systematics are largest between one and two-component models. We confirmed these findings using simulated power spectra.
We then used a Bayesian inference tool to determine the granulation and global oscillation parameters and their uncertainties for our sample of stars. Compared to the previous analysis of Mathur et al. (2011) we extended the sample towards subgiant branch and main-sequence stars and used time series for the red giants that are about three times as long. On the other hand, M giants that show solar-like oscillations (Mosser et al. 2013b) are excluded from our analysis as their granulation signal has such long periods that we cannot derive precise parameters even with the long Kepler time series. The total sample now covers more than 3 order of magnitudes in ν max. From the analysis of the resulting parameters we conclude that:
- -There are tight relations between all granulation parameters and ν max. Whereas the characteristic granulation frequencies scale almost linearly with ν max, the granulation amplitudes approximately scale as ν -0 6 . max . We also confirmed a significant mass dependency of the latter.
- -The defined scaling relations allow to estimate the overall shape of the granulation signal of any solar-type oscillator to within about 15%, and might therefore serve as a good starting point for future large sample studies.
- -In order to allow a comparison with previous measurements and model estimates we also computed e ff ective timescales and total intensity fluctuations. We established that both parameters are predominantly determined by gravity on the stellar surface. They were found to approximately scale as τ e ff ∝ g -0 85 . T -0 4 . and A gran ∝ ( g M 2 ) -1 4 / ∝ R M / 3 4 / . From the rather small rms scatter of the residuals ( < 10% for both parameters) we conclude that no other property of the star plays a significant role.
- -While the theoretical predictions of the characteristic timescales are generally in good agreement with the observation, some discrepancies remain for the intensity fluctuations.
- -There is a statistically significant and surprisingly simple scaling relation for the total pulsation amplitude, which turned out to approximately scale as A puls ∝ ( g M 2 ) -1 3 / ∝ R 4 3 / / M . This implicitly verifies a separate mass and luminosity but no (additional) temperature dependence of the mode amplitudes. Our sample certainly includes stars with diminished pulsation amplitudes due to, e.g., an increased activity level or binary companions. The small rms scatter of the residuals of about 14% therefore indicates that the unperturbed pulsation amplitudes are well approximated by the above scaling relation.
- -The granulation timescale and amplitude as well as the pulsation amplitude can be su ffi ciently described by the mass and surface gravity of a star and the e ff ective temperature has no (or only marginal) additional e ff ect on these parameters.
- -The granulation signal in weak dipole-mode stars is at first order indistinguishable from that of normal stars.
## Appendix A: Power density conversion
Measuring amplitudes of a quasi-stochastic signal (i.e., solarlike oscillations but also the granulation signal) is not straight forward. What we actually measure in a power spectrum is, e.g., the height of an oscillation mode (i.e., the amplitude of the limit spectrum). This does not only scale with the intrinsic mode amplitude and line width but also with the time baseline of the observations. It is therefore convenient to convert the spectral power P PS to spectral power density P PDS , to become independent of the baseline. There are several approaches to do so. A frequently used method (e.g., Appourchaux et al. 2008) is to search for a conversion factor ζ so that
$$\sigma ^ { 2 } & = \zeta \int _ { 0 } ^ { \nu _ { \eta q } } \mathcal { P } _ { P S } ( \nu ) \, d \nu & ( \text{A.1} ) \, \quad \, \quad \, \quad 5999 \\ & \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \ \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \ \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \$$
is satisfied (also known as Parseval's theorem), where σ 2 and ν nq are the variance and Nyqvist frequency 8 of the time series, respectively. The spectral power density is then given as P PDS = ζ P PS . Strictly speaking, this relation is only valid for evenly spaced and continuous data sets. However, introducing gaps in a continuous time series, that contains for example only Gaussian noise, will not change its variance but it will change the integral of the corresponding power spectrum and therefore the conversion factor ζ . This is because the gaps in the time series produce alias peaks in the spectral window function that provide additional power to the spectrum, which is not present in the time domain. To account for this, sort of a filling factor could be introduced in Eq. A.1 but it is more convenient to use the actual spectral window function S PS . We define the conversion factor between spectral power and spectral power density as,
$$\zeta = \left ( \int _ { - \nu _ { n q } } ^ { + \nu _ { n q } } \mathcal { S } _ { P S } ( \nu ) \, d \nu \right ) ^ { - 1 } \,. \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, \\ \text{That said, we note that for most of the Kepler time series the con-} \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad \, & \quad$$
That said, we note that for most of the Kepler time series the conversion factors resulting from Eqs. A.1 and A.2 are almost identical because the spectral window functions of the high-dutycycle time series are close to a Dirac function. However, for stars with large gaps (e.g., due to missing quarters) the di ff erence can be non-negligible.
Acknowledgements. The authors gratefully acknowledge the Kepler Science Team and all those who have contributed to making the Kepler mission possible. Funding for the Kepler Discovery mission is provided by NASA's Science Mission Directorate. TK and JDR are supported by the FWO-Flanders under project O6260 -G.0728.11. TK acknowledges financial support from the Austrian Science Fund (FWF P23608). SH acknowledges support from the European Research Council under the European Communitys Seventh Framework Programme (FP7 2007-2013) / / ERC grant agreement no 338251 (StellarAges) as well as from the Deutsche Forschungsgemeinschaft (DFG) under grant SFB g63 1, 'Astrophysical / flow instabilities and turbulence'. BM acknowledges financial support from the 'Programme National de Physique Stellaire' (PNPS, INSU, France) of CNRS INSU and from the ANR program /
8 In the barycentric reference frame, the median sampling for Kepler LC and SC time series is about 29.38 minutes and 58.85 seconds, respectively, from which we determine ν nq to about 283.6 and 8496 µ Hz.
IDEE 'Interaction Des Etoiles ´ et des Exoplan` etes' (Agence Nationale de la Recherche, France). This work partially used data analysed under the NASA grant NNX12AE17G and under the European Community Seventh Framework Program grant (FP7 2007-2013) ERC grant agreement n. PROSPERITY. SOHO / / is a mission of international collaboration between ESA and NASA.
## References
Aigrain, S., Favata, F., & Gilmore, G. 2004, A&A, 414, 1139
Andersen, B., Leifsen, T., Appourchaux, T., et al. 1998, in ESA Special Publication, Vol. 418, Structure and Dynamics of the Interior of the Sun and
Sun-like Stars, ed. S. Korzennik, 893 Antoci, V., Handler, G., Campante, T. L., et al. 2011, Nature, 477, 570 Appourchaux, T., Michel, E., Auvergne, M., et al. 2008, A&A, 488, 705 Baglin, A., Auvergne, M., Barge, P., et al. 2006, in ESA Special Publication, Vol. 1306, ESA Special Publication, ed. M. Fridlund, A. Baglin, J. Lochard, &L. Conroy, 33
Ballot, J., Barban, C., & van't Veer-Menneret, C. 2011, A&A, 531, A124 Bastien, F. A., Stassun, K. G., Basri, G., & Pepper, J. 2013, Nature, 500, 427 Baudin, F., Barban, C., Belkacem, K., et al. 2011, A&A, 529, A84 Beck, P. G., Montalban, J., Kallinger, T., et al. 2012, Nature, 481, 55 Bedding, T. R., Mosser, B., Huber, D., et al. 2011, Nature, 471, 608 Belkacem, K., Goupil, M. J., Dupret, M. A., et al. 2011, A&A, 530, A142 Belkacem, K., Samadi, R., Goupil, M.-J., et al. 2009, Science, 324, 1540 Borucki, W. J., Koch, D., Basri, G., et al. 2010, Science, 327, 977
Brown, T. M., Gilliland, R. L., Noyes, R. W., & Ramsey, L. W. 1991, ApJ, 368, 599
Brown, T. M., Latham, D. W., Everett, M. E., & Esquerdo, G. A. 2011, AJ, 142, 112
Chaplin, W. J., Elsworth, Y ., Howe, R., et al. 1997, MNRAS, 287, 51
ff
Chaplin, W. J., Houdek, G., Karo
,
C., Elsworth, Y., & New, R. 2009, A&A,
500, L21
Chaplin, W. J., Kjeldsen, H., Bedding, T. R., et al. 2011a, ApJ, 732, 54
Chaplin, W. J., Kjeldsen, H., Christensen-Dalsgaard, J., et al. 2011b, Science,
332, 213
Christensen-Dalsgaard, J. 2002, Reviews of Modern Physics, 74, 1073 Corsaro, E., Fr¨hlich, H.-E., Bonanno, A., et al. 2013, MNR AS, 430, 2313 o Corsaro, E., Stello, D., Huber, D., et al. 2012, ApJ, 757, 190 Deeming, T. J. 1975, Ap&SS, 36, 137 Demarque, P., Guenther, D. B., Li, L. H., Mazumdar, A., & Straka, C. W. 2008,
## ApSS, 316, 31
di Mauro, M. P., Cardini, D., Catanzaro, G., et al. 2011, MNRAS, 415, 3783
Dupret, M., Belkacem, K., Samadi, R., et al. 2009, A&A, 506, 57
Feroz, F., Hobson, M. P., & Bridges, M. 2009, MNRAS, 398, 1601
Frohlich, C., Andersen, B. N., Appourchaux, T., et al. 1997, Solar Physics, 170,
1
Gabriel, M. 1994, A&A, 287, 685
Garc´ ıa, R. A., Hekker, S., Stello, D., et al. 2011, MNRAS, 414, L6
Garcıa, R. A., Mathur, S., Pires, S., et al. 2014, ArXiv e-prints (1405.5374)
Garc´ ıa, R. A., P´rez Hern´ndez, F., Benomar, O., et al. 20 14, A&A, 563, A84
e
a
Gilliland, R. L., Brown, T. M., Christensen-Dalsgaard, J., et al. 2010, PASP, 122,
131
- Gruberbauer, M., Kallinger, T., Weiss, W. W., & Guenther, D. B. 2009, A&A, 506, 1043
Guenther, D. B. 1994, ApJ, 422, 400
- Guenther, D. B., Demarque, P., Kim, Y.-C., & Pinsonneault, M. H. 1992, ApJ, 387, 372
Harvey, J. 1985, in ESA Special Publication, Vol. 235, Future Missions in Solar,
Heliospheric & Space Plasma Physics, ed. E. Rolfe & B. Battrick, 199
ff
Harvey, J. W., Duvall, Jr., T. L., Je eries, S. M., & Pomerantz, M. A. 1993, in
Astronomical Society of the Pacific Conference Series, V ol. 42, GONG 1992.
Seismic Investigation of the Sun and Stars, ed. T. M. Brown, 111
Hekker, S., Elsworth, Y., De Ridder, J., et al. 2011, A&A, 525, A131
Hekker, S., Elsworth, Y., Mosser, B., et al. 2012, A&A, 544, A90
- Herschel, W. 1801, Royal Society of London Philosophical Transactions Series I, 91, 265
- Houdek, G., Balmforth, N. J., Christensen-Dalsgaard, J., & Gough, D. O. 1999, A&A, 351, 582
- Huber, D., Bedding, T. R., Stello, D., et al. 2011, ApJ, 743, 143
Huber, D., Bedding, T. R., Stello, D., et al. 2010, ApJ, 723, 1607
Je ff reys, H. 1998, Theory of probability, Oxford Classic Texts in the Physical Sciences (New York: The Clarendon Press Oxford University Press), xii + 459, reprint of the 1983 edition
Jenkins, J. M., Caldwell, D. A., Chandrasekaran, H., et al. 2010, ApJL, 713, L120
- Kallinger, T., Gruberbauer, M., Guenther, D. B., Fossati, L., & Weiss, W. W. 2010a, A&A, 510, A106
- Kallinger, T., Hekker, S., Mosser, B., et al. 2012, A&A, 541, A51
- Kallinger, T. & Matthews, J. M. 2010, ApJL, 711, L35
- Kallinger, T., Mosser, B., Hekker, S., et al. 2010b, A&A, 522, A1
- Kallinger, T., Weiss, W. W., Barban, C., et al. 2010c, A&A, 509, A77
- Karo ff , C. 2012, MNRAS, 421, 3170
- Karo ff , C., Campante, T. L., Ballot, J., et al. 2013, ApJ, 767, 34
- Kjeldsen, H. & Bedding, T. R. 1995, A&A, 293, 87
- Kjeldsen, H. & Bedding, T. R. 2011, A&A, 529, L8
- Kjeldsen, H., Bedding, T. R., Arentoft, T., et al. 2008, ApJ, 682, 1370
- Koch, D. G., Borucki, W. J., Basri, G., et al. 2010, ApJ, 713, L79
- Korzennik, S. G., Rabello-Soares, M. C., Schou, J., & Larson, T. P. 2012, in Astronomical Society of the Pacific Conference Series, Vol. 462, Progress in Solar Stellar / Physics with Helio- and Asteroseismology, ed. H. Shibahashi, M. Takata, & A. E. Lynas-Gray, 189
Ludwig, H.-G. 2006, A&A, 445, 661
- Ludwig, H.-G., Samadi, R., Ste ff en, M., et al. 2009, A&A, 506, 167
- Mathur, S., Garc´ ıa, R. A., R´gulo, C., et al. 2010, A&A, 511 , A46 e
- Mathur, S., Hekker, S., Trampedach, R., et al. 2011, ApJ, 741, 119
- Michel, E., Samadi, R., Baudin, F., et al. 2009, A&A, 495, 979
- Miglio, A., Brogaard, K., Stello, D., et al. 2012, MNRAS, 419, 2077
- Mosser, B., Belkacem, K., & Vrard, M. 2013a, in EAS Publications Series, Vol. 63, EAS Publications Series, 137-150
- Mosser, B., Dziembowski, W. A., Belkacem, K., et al. 2013b, A&A, 559, A137
- Mosser, B., Elsworth, Y., Hekker, S., et al. 2012a, A&A, 537, A30
- Mosser, B., Goupil, M. J., Belkacem, K., et al. 2012b, A&A, 548, A10
- Mosser, B., Goupil, M. J., Belkacem, K., et al. 2012c, A&A, 540, A143
- Muller, R. 1999, in Astrophysics and Space Science Library, Vol. 239, Motions in the Solar Atmosphere, ed. A. Hanslmeier & M. Messerotti, 35-70
Pinsonneault, M. H., An, D., Molenda- ˙ akowicz, J., et al. 2012, ApJS, 199, 30
Z
- Samadi, R., Belkacem, K., & Ludwig, H.-G. 2013a, A&A, 559, A39
- Samadi, R., Belkacem, K., Ludwig, H.-G., et al. 2013b, A&A, 559, A40
- Samadi, R., Georgobiani, D., Trampedach, R., et al. 2007, A&A, 463, 297
- Smith, J. C., Stumpe, M. C., Van Cleve, J. E., et al. 2012, PASP, 124, 1000 Stello, D., Huber, D., Kallinger, T., et al. 2011, ApJL, 737, L10
- Thygesen, A. O., Frandsen, S., Bruntt, H., et al. 2012, A&A, 543, A160
- Trampedach, R., Asplund, M., Collet, R., Nordlund, Å., & Ste in, R. F. 2013, ApJ, 769, 18
V´ azquez Rami´, H., R´ egulo, C., & Roca Cort´s, T. 2005, A& A, 443, L11 o e
|
10.1051/0004-6361/201424313
|
[
"T. Kallinger",
"J. De Ridder",
"S. Hekker",
"S. Mathur",
"B. Mosser",
"M. Gruberbauer",
"R. A. Garcia",
"C. Karoff",
"J. Ballot"
] |
2014-08-04T20:55:27+00:00
|
2014-08-04T20:55:27+00:00
|
[
"astro-ph.SR"
] |
The connection between stellar granulation and oscillation as seen by the Kepler mission
|
The long and almost continuous observations by Kepler show clear evidence of
a granulation background signal in a large sample of stars, which is
interpreted as the surface manifestation of convection. It has been shown that
its characteristic timescale and rms intensity fluctuation scale with the peak
frequency (\nu_{max}) of the solar-like oscillations. Various attempts have
been made to quantify the observed signal, to determine scaling relations, and
to compare them to theoretical predictions. We use a probabilistic method to
compare different approaches to extracting the granulation signal. We fit the
power density spectra of a large set of Kepler targets, determine the
granulation and global oscillation parameter, and quantify scaling relations
between them. We establish that a depression in power at about \nu_{max}/2,
known from the Sun and a few other main-sequence stars, is also statistically
significant in red giants and that a super-Lorentzian function with two
components is best suited to reproducing the granulation signal in the broader
vicinity of the pulsation power excess. We also establish that the specific
choice of the background model can affect the determination of \nu_{max},
introducing systematic uncertainties that can significantly exceed the random
uncertainties. We find the characteristic background frequency and amplitude to
tightly scale with \nu_{max} for a wide variety of stars, and quantify a mass
dependency of the latter. To enable comparison with theoretical predictions, we
computed effective timescales and intensity fluctuations and found them to
approximately scale as \tau_{eff} \propto g^{-0.85}\,T^{-0.4} and A_{gran}
\propto (g^2M)^{-1/4}, respectively. Similarly, the bolometric pulsation
amplitude scales approximately as A_{puls} \propto (g^2M)^{-1/3}, which
implicitly verifies a separate mass and luminosity dependence of A_{puls}.
|
1408.0818v1
|
## ActiveMonitor: Non-blocking Monitor Executions for Increased Parallelism
Wei-Lun Hung Himanshu Chauhan ∗ Vijay K. Garg
†
The University of Texas at Austin {wlhung,himanshu}@utexas.edu [email protected]
We present a set of novel ideas on design and implementation of monitor objects for multi-threaded programs. Our approach has two main goals: (a) increase parallelism in monitor objects and thus provide performance gains (shorter runtimes) for multi-threaded programs, and (b) introduce constructs that allow programmers to easily write monitor-based multi-threaded programs that can achieve these performance gains. We describe the concepts of our framework, called ActiveMonitor , and its prototype implementation using futures [11]. We evaluate its performance in terms of runtimes of multi-threaded programs on linked-list, bounded-buffer, and other fundamental problems implemented in Java. We compare the runtimes of our implementation against implementations using Java's reentrant locks [29], recently proposed automatic signaling framework AutoSynch [23], and some other techniques from the literature. The results of of the evaluation indicate that monitors based on our framework provide significant gains in runtime performance in comparison to traditional monitors implemented using Java's reentrant locks.
## 1 Introduction
Monitors are the prevalent programming technique for synchronization between threads in sharedmemory parallel programs. They were designed [21] with two primary goals: ( ) i to ensure safety of shared data by enforcing mutual exclusion in critical sections, and ( ii ) to enable conditional synchronization using the wait-notify mechanism. Monitors ensure that while a thread is executing a critical section, other threads that cannot access the critical section wait for it to become available. Waiting threads are notified by some thread exiting the critical section. We argue that this monitor design has certain drawbacks:
- ( a ) For executions on multi-core CPUs, a thread's update to the shared data may cause a cache invalidation for some other thread running on a different core, specially if the thread was waiting for the critical section. Thus, the penalty incurred due to cache misses would slow down the program on multi-core CPUs.
- ( b ) If threads do not require the result of their updates on shared data immediately after performing these updates, then the blocking executions of critical sections for these updates may become a bottleneck, and could reduce the program throughput. For example, consider a thread that inserts items in a linked-list. If this thread does not need the result of the insert operation then blocking insert operations lead to slower runtimes.
∗ Contact Author. Address: ACES 5.432, 201 E 24 Street, Austin, Texas, USA
† Supported in part by the NSF Grants CNS-1346245, CNS-1115808, and Cullen Trust for Higher Education Endowed Professorship
In this paper, we focus on monitors methods that perform updates on shared data. Thus, the solution of using Read/Write locks [29] is not applicable. Observe that the design of monitors to enforce blocking executions was envisioned in 1970 's when processors were single-core, processing power was scarce and saving processor cycles was a primary concern for programmers. Under these settings, making the critical section executions blocking was not a performance issue. The current situation is drastically different: not only multi-core processors are now ubiquitous, but they are also significantly cheaper and faster. Therefore, it is important to explore alternate design and implementation of monitors to exploit the availability of multi-core processors. Given that the blocking executions of monitor methods seem to form a bottleneck for performance, we pose the following question: can monitor design and implementation be altered to allow non-blocking executions and thus improve the throughput of conditionally synchronized multi-threaded programs on multi-core machines? The motivation here is that non-blocking executions can increase parallelism by allowing more instructions to be executed in fixed time from the perspective of the thread(s). For this purpose, we propose two novel techniques:
- 1. A monitor object is instantiated as a thread, and invocation of monitor methods by application (worker) threads is replaced by submissions of equivalent tasks to the monitor thread. The monitor thread assumes the responsibility of executing these tasks and returning the result (if required) to the application threads.
- 2. The monitor methods, replaced by tasks in representation, can be executed in non-blocking manner. The non-blocking executions are explicitly indicated in the code.
These two changes together tackle the drawbacks discussed above. Under our proposed design, the worker threads do not directly access the shared data, only monitor thread does, which helps in improving the data cache locality; and non-blocking executions reduce runtime due to parallelism. This paper presents our experience with design and implementation of these ideas. We propose a framework, called ActiveMonitor , that explores non-blocking executions of monitor methods - possibly at the expense of cheap and abundant processor and memory resources.
Fig. 1 compares the implementation of a bounded-buffer monitor written using: (a) Java's Reentrant locks [29], and (b) our proposed framework and its constructs. We propose two new keywords: activemonitor and nonblocking . The activemonitor keyword, in class declaration at line 1 of Fig. 1(b), is used to declare that this bounded-buffer is implemented as an 'active' monitor object, that is: it would exist as an independent thread. The nonblocking keyword in the method signature of put() , at line 6 , indicates that the execution of this method can be non-blocking. As our experiments show (Section 4), this implementation provides ∼ 33 % faster runtime in comparison to the conventional implementation of Fig. 1(a). Lastly, our implementation is shorter, cleaner, and more intuitive. This is because we use the waituntil construct (at lines 9 15 , in Fig. 1(b)) that provides implicit/automatic signaling mechanism for threads [23]. This keyword and the automatic signaling 1 is not a contribution of this paper; we merely use it in our framework's implementation to exploit the performance benefits it provides.
An immediate criticism of our approach is the need for additional resources for its realization. Creating a new thread for a monitor object consumes additional memory in the form of its stack allocation, as well as increases the demand for CPU time slices. However, it is well known that the processing power of the modern multi-core CPUs remains commonly under utilized . Additionally, with 2 the trends of ever increasing data and computation size, most users tend to favor faster runtimes even at the cost of somewhat increased consumption of processor and memory resources. Our approach provides much faster runtimes for many multi-threaded use cases. In addition, we claim that with careful resource
1 Appendix A briefly covers the automatic signaling mechanism
2 we collected resource usage data, at 10 min. intervals, of Stampede [1] supercomputer over a period of three days. The mean CPU idle time was 86% .
Figure 1: Comparsion of Bounded-Buffer programs written using (a) Java's locks and (b) our approach
management by the means of a dynamic thread creation policy that considers the processor/memory utilization, coupled with the automatic thread signaling mechanism [23], our approach's performance improvements come at an acceptable cost of additional CPU and memory usage. In short, the key contributions of this paper are following:
- · we propose a new monitor design, based on two new keywords: activemonitor and nonblocking , that facilitates non-blocking executions of data updates within their critical sections. We describe the concepts required for this design's implementation in Java.
- · we define rules to ensure correctness in presence of non-blocking operations, and show that with our proposed design, despite additional concurrent executions, the executions are linearizable [20] as per their non-blocking interpretation.
- · we evaluate the runtimes of some fundamental multi-threaded problems using our prototype implementation and Java's Reentrant locks [29]. For majority of these problems, our design and implementation provides ∼ 30 -40 % faster runtimes in comparison to reentrant locks.
The rest of this paper is organized as follows. Section 2 briefly covers the background concepts. Section 3 discusses the key concepts of the framework and correctness of executions. Section 4 presents experimental evaluation of our approach. We discuss some limitations of our approach, and future work in Section 5. Section 6 presents the related work, and Section 7 offers concluding remarks.
## 2 Background
We use the concept of futures [11] to achieve non-blocking executions of critical sections.
Futures: Futures [11] were proposed as a mechanism to facilitate task submissions (to threads) in shared memory parallel programs. In a task submission based model some threads, called executors , provide task execution as a service. Under this model, a thread makes an asynchronous call to some executor thread, and a future object is returned to the caller thread as a pointer to the computation and its possible result. The executor thread performs a task for the caller thread, and updates the returned future object with the result of the computation of the task. The caller thread can collect the result by evaluating the future. Evaluation of a future, i.e. the collection of the result, is performed by invoking a blocking method on the future object. Hence, if needed the thread would have to wait for the task computation to complete. In our framework, we use futures to realize non-blocking executions of monitor methods by treating method invocations as task submissions.
Blocking and Non-Blocking methods : By design, methods provided by monitor objects are usually blocking ( synchronous ). If a thread invokes a blocking method m , then the thread can not execute its next instruction until it has exited m ; whether by normally returning from the method or exiting the method due to an exception. Whereas, if a thread invokes a non-blocking method m ′ then it would be able to execute its next instruction immediately after, while the non-blocking method is being executed in parallel, possibly by another thread.
## 3 ActiveMonitor Concepts
In this section, we discuss the key concepts of the ActiveMonitor framework. The relevant implementation details are presented in Appendix D. We first explain our approach using the bounded-buffer monitor of Fig. 1(b). Observe that apart from using the implicit-signal construct of waituntil , the code uses two additional keywords that we propose: activemonitor and nonblocking . By including the activemonitor keyword in the declaration (eg. line 1 ), the programmer indicates that this monitor is an 'active' object, i.e. it should be instantiated as a thread. However, if there are many monitor objects in a user's program we limit the number of monitor threads - this is an implementation detail, and discussed in Appendix D.2. In the conventional monitor design, worker threads, i.e. threads explicitly created by the user program, execute the instructions of the critical section and update the data protected by the monitor. In contrast, as per our design, once the monitor is instantiated as a thread, it executes its own critical sections on behalf of worker threads. For this, at compile time method calls made to the monitor are treated as task submissions to the monitor thread. Thus, for the program in Fig. 1b, a producer thread's call to the put method, as well as a consumer thread's call to the take method, are converted to submission of equivalent tasks to the monitor thread by our framework. The worker threads now have the option to perform blocking ( synchronous ) or non-blocking ( asynchronous ) executions of monitor methods. Asynchronous executions are made possible by using our second keyword proposal, nonblocking , that allows for non-blocking (asynchronous) execution of instructions of the method using this keyword. We first describe how monitor tasks are generated from monitor methods, and then discuss their asynchronous or synchronous executions.
## 3.1 Monitor Tasks
In our framework, every monitor method call corresponds to an equivalent task. At compile-time, invocations of monitor methods by worker threads are replaced by submissions of the equivalent tasks to the monitor thread.
Definition 1 (Monitor Task) . A monitor task t consists of a boolean predicate P and a set of statements S . If the precondition defined by P is true then t is 'executable' and statements in S can be executed to complete t . Otherwise, t is 'unexecutable'.
Observe that the syntax of a monitor task is similar to that of a guarded command [8]. Consider the put method (lines 8 -13 ) of the bounded-buffer in Fig. 1b. For this monitor method, the equivalent monitor task T is defined by the code of lines 9 -12 . Here, the precondition P is (itemCount < buffer\_size) ; and it checks if the buffer has any space to insert the item. If this condition is false, the waituntil construct ensures that any thread trying to complete this task has to wait unless the buffer is not full. Lines 10 and 11 together form the set of statements S . As indicated by the nonblocking keyword in the method signature at line 8 , the generated task would be submitted for a non-blocking execution to the monitor thread. Removal of this keyword from the method signature makes the execution blocking.
The details of compile time generation of tasks for monitor method calls are in Appendix D.1. Appendix B discusses some special cases of monitor tasks. Our current prototype cannot convert some particular types of monitor methods to equivalent tasks - for example, methods with waituntil not at the beginning of the critical section, or waituntil in a conditional branch. They are also discussed in Appendix D.1.
## 3.2 Monitor Tasks: Execution and Correctness Guarantees
In the ActiveMonitor framework, monitor threads are responsible for executing monitor tasks. Based on the availability of resources, the framework tries to create one thread per monitor object. The implementation details for limiting the monitor thread count are in Appendix D.2. Monitor threads execute monitor tasks while observing the following rules.
Rule 1 (Mutex Invariant) . All the tasks of a single monitor object are executed by the same monitor thread.
Thus, Rule 1 maintains the mutual exclusion of critical sections of a monitor. However, for correctness of executions, we require additional rules. Let proc t ( ) denote the thread that submits the task t to the a monitor; sub t ( ) be the time at which the task t is submitted to the monitor, and exe t ( ) indicate the time when the monitor thread starts executing t .
Rule 2. For a pair of tasks s and t submitted to a monitor M , if proc s ( ) = proc t ( ) , then sub s ( ) < sub t ( ) ⇒ exe s ( ) < exe t ( ) .
Rule 3. Let n 1 , m 2 be two successive method invocations by a worker thread on two different monitors M 1 and M 2 in the user program, and let t 1 , t 2 be the corresponding task submissions at runtime. Then, at runtime t 1 must be completed before t 2 's submission.
The notions of method invocation and response used to define linearizability [20] need a different interpretation under non-blocking executions. In short, invocation now corresponds to submission of the equivalent task to monitor thread, and response corresponds to this tasks completion. With this interpretation, the following result establishes correctness of executions by proving linearizable executions.
Lemma 1. With rules 1, 2, and 3 we get an execution that is equivalent to a lock-based execution.
Proof. Sketched in Appendix C.1.
Observe that Rule 3 forces a worker thread to wait for the previous task to finish even if that task was on a different monitor object. In many applications, a programmer may not require this constraint
on different monitor objects. Hence, to improve performance, we can drop Rule 3 for such applications. We show that the execution is still linearizable with the new interpretation of invocation and response.
Note that in the standard model of concurrent history [20], the thread history is always sequential although the object history may not be sequential. In our model, due to non-blocking executions, we have the dual property: an object history is always sequential whereas a thread history may not be so. We first define the thread order in presence of non-blocking operations. Let s , s 1 2 , ..., s m be m operations performed on a monitor object by a worker thread. Let s i < s j denote that s i was executed before s j . In the standard model, all operations are blocking and we get that s i < s i +1 for all i . In our model, we define the order as follows. If s i is blocking, then s i < s j for all j > i . If s i and s j are operations on the same object and i < j , then s i < s j . If s i is non-blocking and the result of s i is required before s k , then s i < s j for all j ≥ k .
Lemma 2. With Rules 1 and 2, we get an execution that is linearizable.
## Proof. Sketched in Appendix C.2.
Observe that the legal sequential history we get may not preserve the order of invocation of operations, but only the thread order.
## 4 Evaluation
We now present the experimental evaluation of our proposed approach. We evaluate our prototype implementation on five multi-threaded problems. The first problem is a bounded-buffer problem with multiple producers and consumers, the second is a sorted linked-list (of integers) on which threads perform updates. The remaining three are conditional synchronization problems involving varying levels of complexity in their conditional predicates. We implemented the problems using the ActiveMonitor framework, Java's reentrant locks, and AutoSynch [23]. Additional use-case specific implementations (details given in problem descriptions below) are also compared for bounded-buffer, and the linked-list problem. For all the problems, we use a benchmark in which each worker thread performs 512000 operations on shared data protected by monitors. All the experiments are conducted on a machine with 1 socket, 2 AMD Opteron 6180 SE 12 Core, 2 5 . GHz CPUs (= 24 hardware threads) and 64 GBs memory running Linux 2 6 18 . . . We vary the number of worker threads, and measure the time required for all of them to complete their operations. All threads perform a fixed number of warm-up operations on shared data before starting the time measurements. For each problem, we measure runtimes for 25 runs, and report the mean values after removing the highest and the lowest values.
For all the experiments, we restrict the ActiveMonitor framework to create only one monitor thread. We now discuss the setup of experiments and their results in a problem specific manner. Across all results, we denote the implementation techniques with the following notation: LK : Reentrant locks, AS : AutoSynch [23], AM : ActiveMonitor (this paper). For each problem, all the runtimes reported in their results exhibit negligible variance ( < 5% of the mean value) across runs.
## 4.1 Bounded-Buffer (BB) problem [6, 7]
Setup : Items are plain Java objects. Every producer's put invocation is non-blocking, and every consumer's take is blocking. We also compare runtimes of Java's ArrayBlockingQueue [29], denoted by ABQ , based implementation. Number of producers (= number of consumers) is varied from 2 to 24 . We perform three types of experiments, and measure: (a) runtimes wieh varying number of threads for a buffer of fixed size (= 4 ). (b) runtimes with varying buffer-size for fixed number of producers/consumers (= 16 each). (c) runtimes with varying limit on non-blocking tasks allowed on the buffer of size 4 with 16 producers and consumers each.
Figure 2: Runtimes for Bounded-Buffer problem
## Results
Collective results are shown in Fig. 2. Fig. 2a plots the runtimes for experiment (a). Fig. 2b shows the runtimes for experiment (b). This result highlights the benefits of non-blocking executions. Recall that only put calls are non-blocking for AM ; all take calls are blocking. For small size buffers, LK AS , , and ABQ are much slower in comparison to AM due to their blocking insert operations. For larger buffer sizes, LK and ABQ implementations perform slightly faster than AM . On larger buffer sizes, the frequency of threads getting blocked out is much lower, and both LK and ABQ benefit from greedy implementation of Java in lock acquisition; but AM ends up performing slightly more work in enforcing order on tasks (as per Rules 1 2 , ).
Fig. 2c shows the results for experiment (c). 'Limit' on the size of task queue means that whenever the queue maintained by the monitor thread to store tasks, reaches the 'limit' value then even a nonblocking task submission to the monitor is force to block. Making all the invocations blocking (queue size limit = 0 ) leads to an overall runtime (for completion) of ∼ 5 seconds, whereas with no limit on the size of task queue the runtime is ∼ 3 seconds. This observations clearly shows that our non-blocking task based approach is beneficial. However, an unbounded task queue might grow quite large. But as shown by the plot, even for a short limit (around 20 ) on the size of this queue, the runtime performance converges to optimal performance (of unbounded size). Thus, this result establishes that only a small number of non-blocking tasks can provide significant performance benefit.
## 4.2 Sorted Linked-List (SLL) Problem
Setup : A linked-list of integers sorted in non-decreasing order. The number of worker threads is varied from 2 to 24 . The list is pre-populated with 5000 integers. Each worker thread inserts or removes (with equal probability) a random integer. For a fair comparison, seeds are used for randomization so that the threads generate the same sequence of random integers across runs as well as across different monitor implementations. We make every thread execute some local instructions, simple mathematical computations, outside the critical section between any successive calls on the monitor. This is done in attempt to simulate some practical program behavior in which generation of data takes some time . 3 Both insert and remove operations are non-blocking. Additional techniques used for performance comparison: lock-free implementation ( LF ) [18, 13, 32], fine-grained locking ( FG ) [18], and transactional memory ( TM ) implementation in Java [28]. We measured runtimes for two types of experiments: (a) runtimes with varying number of threads, when each thread performed 250 mathematical computations
3 Moreover, The results for the BB problem already showed that we perform better than conventional monitors for saturation tests.
- (a) Runtimes for experiment (a)
- (b) Runtime ratio with varying number of operations outside CS
Figure 3: Results for sorted linked-list problem
(integer additions) outside the critical section between successive calls to the monitor. (b) runtimes with varying number of threads and varying number of operations performed outside the critical section.
## Results
Fig. 3a shows the results for experiment (a). For this experiment, the ActiveMonitor implementation ( AM ) clearly outperforms the reentrant lock based monitor ( LK ), and AutoSynch ( AS ). However, lock-free ( LF ) and transactional memory ( TM ) based implementations perform better. This is because both of them use optimized algorithm for the linked-list data structure. Whereas, we do not use any linked-list specific optimization, and only use non-blocking executions on critical sections.
Fig. 3b plots the ratio of runtimes using LK implementation and AM implementations for experiment (b). The ratio with no operations outside the critical section is close to 1 , i.e. our approach provides no performance benefit. This is expected because no operations can be performed concurrently with the non-blocking updates to the list, and the opportunity of parallelism is effectively absent. As the number of operations outside the critical section is increased, we see significant performance gains with our approach, upto ∼ 30% , due to increased parallelism introduced by non-blocking executions.
## 4.3 Conditional Synchronization Problems
We used three thread synchronization problems that involve different levels of complexity in their conditional predicates. For all three, we measured the time taken for all the worker threads to complete the designated number ( 512000 ) of operations on the monitor.
Round-Robin (RR) Access : Every worker thread accesses the monitor in round-robin order. All operations on the monitor are non-blocking. The number of worker threads is varied from 2 to 24 .
Parametrized Bounded-Buffer (PBB) [6, 7] : Producers put a collection of items into a shared buffer, while consumers remove a number of items from the buffer. For producers, the number of items to be inserted, and for consumers, the count of items to be removed, both are randomly decided at runtime. Similar to the SLL problem, seeds are used for randomization for fair evaluation across implementations. The number of producers (= number of consumers) is varied from 2 to 24 .
Ticketed-Readers/Writers (TRR) [4, 3, 12] : A ticket is used to maintain the access order for readers and writers [3]. Every reader/ writer gets a ticket number indicating its arrival order. Readers/writers wait on the monitor for their turn. On their turn, they enter the monitor but do not
Figure 4: Runtimes for RR, PBB, and TR/W problems: all plots have the same range on y-axis
perform any computation inside the monitor, and immediately exit. Operations for arrival on monitor and waiting for turn to access the monitor are blocking, whereas operations after getting access to critical sections are non-blocking. The number of writer threads is varied from 2 to 24 . The number of readers is kept 5 x the number of writers.
## Results
Fig. 4 presents the runtimes of LK , AS , and AM based implementations for the three problems. For the RR problem, the performance of AM is 6 and 5 times faster than that of AS , and LK , respectively. This is due to better cache locality and fewer context switches. For this problem, after accessing the critical section, every thread is forced to wait unless all other threads have accessed the critical section. This constraint leads to poor cache locality and more context switches for conventional monitor design. However, in our approach, the monitor thread continuously executes all monitor accesses for each thread; therefore, there is minimal number of context switches between each monitor method invocation and that results in excellent cache locality.
For the TR/W problems, AM performs poorly in comparison to LK . This is primarily due to the nature of the problem and its underlying implementation requirements. The problem requires conditional synchronization in the middle of the method calls, and thus our task based approach has to create two tasks per method call. In fact, any family of problems with such characteristics would lead to much higher number of tasks than the actual method calls.
## 4.4 CPU and Memory Consumption
Table 1 presents the results on the CPU and memory usage by LK , AS , and AM for all the evaluated problems. For each problem, the CPU ratio stands for the ratio of an implementation's mean (across runs) CPU consumption and mean (across runs) CPU consumption of LK implementation. As per memory usage, we report the maximum memory (in MBs) consumed across any of the runs performed by the implementations for each problem. As expected, our approach consumes more CPU resources. However, the maximum memory consumed by our approach is only incrementally larger than that of Java's reentrant lock based approach. Thus, we have backed our claim that with careful resource management the cost for the performance benefit is relatively small.
## 5 Discussion & Future Work
The benefits of our approach are manifold: improved runtime performance, better control over desired program behavior, and increased opportunities for applying optimizations such as execution re-ordering,
Table 1: CPU and memory consumption with Reentrant locks ( LK ), AutoSynch ( AS ), and ActiveMonitor ( AM ).
| Problem | BB | BB | BB | SLL | SLL | SLL | RR | RR | RR | PBB | PBB | PBB | TR/W | TR/W | TR/W |
|-----------|------|------|------|-------|-------|-------|------|------|------|-------|-------|-------|--------|--------|--------|
| Approach | LK | AS | AM | LK | AS | AM | LK | AS | AM | LK | AS | AM | LK | AS | AM |
| CPU Ratio | 1.00 | 1.11 | 1.95 | 1.00 | 0.86 | 3.49 | 1.00 | 1.10 | 2.97 | 1.00 | 2.39 | 2.47 | 1.00 | 1.23 | 1.24 |
| Max Mem. | 196 | 236 | 197 | 135 | 135 | 145 | 236 | 224 | 250 | 196 | 227 | 251 | 211 | 232 | 244 |
operation combining, etc. However, some aspects of our framework are open to criticism. First and foremost, based on the design assumptions as well as the experimental results, it is evident that at present, our approach does not suit systems on which processor and memory resources are scarce. Even though our evaluation results indicate slightly increased memory consumption, the increase in CPU consumption is more drastic. Hence, an important future work is an improved implementation of our framework that lowers the CPU usage. Another future work is to address the issue of task generation for problems that require conditional synchronization in the middle of the method call, eg. Ticket-R/W, and causes large number of tasks to be created. It is important to note that our approach extends conventional multi-threaded programs and their constructs, and thus provides a programmer the choice of an implementation she thinks is most suited for her needs.
We showed that we provide linearlizability of executions - interpreted under the non-blocking execution model. Although it is desirable in most of the cases, there are situations when this requirements can be relaxed to certain extents for even faster performance. Such relaxations are very much dependent on the use-cases and the expectations of the programmer. As shown by Kogan et al. in [25], having 'weak' consistency requirements could lead to significant performance gains. In fact, for specific use-cases programmers may only need eventual consistency [37], which requires that any update to shared data is eventually committed, in their programs. However, allowing weak or eventual consistency is not possible for all use-cases. For example, consider multi-threaded implementations (in shared memory model) of Dijkstra's single-source-shortest-path algorithm [5, 31]. Weak or eventual consistency based implementations for this algorithm cannot guarantee correct results. However, for a large number of problems, approximate solutions are good enough, and their parallel implementations exhibit significant gains in performance. Thus, this opens up two interesting future problems for our work. First of them is evaluating performance of our non-blocking execution approach for problems that admit approximate solutions, and the second is to extend our evaluation to the weak consistency model of [25].
When it comes to evaluating the performance of our proposed design and its implementation, the exploration space is large. Not only there is a large number of problems related to multi-threaded updates on data structures, but the set of parameters that affect the runtime for these problems is also large. We also plan to perform experiments by varying more parameters such as load distributions of methods, size of shared data, etc. Another key experiment is to analyze the cache locality of our approach - in terms of cache miss rates per operation and per instruction - for different experiments, and comparing it to cache locality achieved by other techniques.
Lastly, note that our current implementation does not use ideas such as operation combining [15], execution re-ordering, and batch updates [25]. Intuitively, incorporating these strategies in our implementation could lead to even better results than those observed in this paper.
## 6 Related Work
The idea of having monitor objects execute as independent threads is not completely novel. Hoare proposed a similar mechanism, in which all objects are active , of communicating sequential processes (CSP) in [22] long ago. However, the proposal was a formal language for interactions between processes.
Its constructs are still mostly used for formal analysis. Our work focuses on making parallel programs faster by using futures to provide selective non-blocking executions on monitor methods.
Research efforts to increase parallelism in programs have been continuous almost since the advent of computing. With the arrival of multi-core processors, improving multi-threaded programming with syntactic changes, as well as with novel designs of data structures and algorithms, has been the focus of many research efforts in the last decade. Transactional memory [17, 34] is a well-known research effort that proposes modified syntax for ease of writing multi-threaded programs. The prominent contributions in algorithmic direction are: lock-free data structures [13, 32, 16], and wait-free data structures [26, 27, 36]. Some other research efforts have explored increased parallelism for data structures by operation re-ordering and combining complementary operations (insert and remove) [14].
Transactional memory [17, 34] introduced an elegant and much needed design for writing multithreaded programs on shared memory. The promising feature in transactional memory proposal was to provide simplified syntax that guarantees correctness, and delegates the handling of mutual exclusion to the library providing the implementation. Our approach and ideology is similar on these goals. However, constructs for conditional waiting under transactional memory are limited [35, 30, 9]. Hence, writing many conditional synchronization based multi-threaded programs is rather difficult. Also, unlike transactional memory, our approach merely transfers the responsibility of data manipulation to a single monitor thread and does not require any complicated rollback mechanism for resolving conflicting updates on the shared data.
Designing lock-free and wait-free algorithms is a challenging problem for any use-case. In addition, lock-free and wait-free implementations are currently available for a small set of data structures [10, 19]. Even though there is a lot of ongoing research to grow this set, the growth is slow. By using our approach, the end user benefits by an additional level of abstraction and does not have to deal with the complications involved in understanding and correctly implementing wait-free and lock-free algorithms.
We use the concept of futures [11, 29] to realize the idea of non-blocking/asynchronous executions. The work by Kogan et al. [25] is similar to our approach, and makes use of futures for non-blocking executions. However, there are few key differences between our work and [25]. We explore changes to the paradigm of monitors, especially for conditional synchronization problems, whereas [25] focuses on throughputs, under saturation scenarios, of data structures such as stacks, queues, and linked-lists. Hence, our approach provides a more generic tool for performing safe updates and synchronization between threads.
Frameworks such as Legion [2] and Galois [33] focus on improving runtime performance of parallel programs on scientific simulations and large graphs, respectively. Thus, they do not directly relate to thread synchronization and notification based use-cases such as bounded-buffer. Secondly, both of these libraries require advanced skills and familiarity with the features and syntactic details exposed by the libraries. Hence, the performance gains achieved by using them are highly correlated with a programmer's knowledge of the involved optimization parameters needed to tweak the executions for optimal performance.
## 7 Conclusion
Proposing changes to a well established programming paradigm is an ambitious goal. Moreover, convincing programmers to adopt a new programming framework in favor of the conventional one, usually requires the proposed approach to offer gains in terms of runtime performance as well as programming ease/productivity. The ActiveMonitor framework is a step in this direction. In almost all of the problems used in our evaluation, monitors with non-blocking executions outperform Java's reentrant lock based monitors. Additionally, this gain is achieved by a simple mechanism of using only two new
keywords in existing Java programs. At present, our prototype implementation is costly in terms of processor usage. However, with the ubiquity of multi-core processors, optimized implementations of our proposed technique could lead to significant advantages at reduced costs. In this paper, our initial exploration in a new direction, that is: non-blocking executions on monitors, has showed promising results. We hope that with further research in this direction, our approach would lead to even stronger results.
## References
- [1] Texas advanced computing center stampede. https://www.tacc.utexas.edu/stampede/.
- [2] M. Bauer, S. Treichler, E. Slaughter, and A. Aiken. Legion: Expressing locality and independence with logical regions. In Proceedings of the International Conference on High Performance Computing, Networking, Storage and Analysis , SC '12, pages 66:1-66:11, Los Alamitos, CA, USA, 2012. IEEE Computer Society Press.
- [3] P. A. Buhr and A. S. Harji. Implicit-Signal Monitors. ACM Transactions on Programming Languages and Systems , 27(6):1270-1343, Nov. 2005.
- [4] P. J. Courtois, F. Heymans, and D. L. Parnas. Concurrent control with 'readers' and 'writers'. Communications of the ACM , 14(10):667-668, 1971.
- [5] E. W. Dijkstra. A note on two problems in connexion with graphs. Numerische Mathematik , 1(1):269-271, 1959.
- [6] E. W. Dijkstra. Cooperating Sequential Processes, Technical Report EWD-123. Technical report, 1965.
- [7] E. W. Dijkstra. Hierarchical ordering of sequential processes. Acta Informatica , 1(2):115-138, 1971.
- [8] E. W. Dijkstra. Guarded commands, nondeterminacy and formal derivation of programs. Communications of the ACM , 18(8):453-457, Aug. 1975.
- [9] P. Dudnik and M. Swift. Condition variables and transactional memory: Problem or opportunity? In The 4th ACM SIGPLAN Workshop on Transactional Computing , 2009.
- [10] F. Fich, D. Hendler, and N. Shavit. On the inherent weakness of conditional synchronization primitives. In Proceedings of the Twenty-third Annual ACM Symposium on Principles of Distributed Computing , PODC '04, pages 80-87, New York, NY, USA, 2004. ACM.
- [11] R. H. Halstead. Multilisp: A language for concurrent symbolic computation. ACM Trans. Program. Lang. Syst. , 7(4):501-538, Oct. 1985.
- [12] P. B. Hansen. Structured Multiprogramming. Communications of the ACM , 15(7):574-578, 1972.
- [13] T. L. Harris. A Pragmatic Implementation of Non-blocking Linked-Lists. In Proceedings of the 15th International Conference on Distributed Computing , pages 300-314, London, UK, UK, 2001. Springer-Verlag.
- [14] S. Heller, M. Herlihy, V. Luchangco, M. Moir, W. I. Scherer, and N. Shavit. A Lazy Concurrent List-Based Set Algorithm. In J. Anderson, G. Prencipe, and R. Wattenhofer, editors, Principles of Distributed Systems , pages 3-16. Springer Berlin Heidelberg, 2006.
| [15] | D. Hendler, I. Incze, N. Shavit, and M. Tzafrir. Flat Combining and the Synchronization- parallelism Tradeoff. In Proceedings of the Twenty-second Annual ACM Symposium on Parallelism in Algorithms and Architectures , pages 355-364, New York, NY, USA, 2010. ACM. |
|--------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [16] | D. Hendler, N. Shavit, and L. Yerushalmi. A scalable lock-free stack algorithm. In Proceedings of the Sixteenth Annual ACM Symposium on Parallelism in Algorithms and Architectures , SPAA '04, pages 206-215, New York, NY, USA, 2004. ACM. |
| [17] | M. Herlihy and J. E. B. Moss. Transactional memory: Architectural support for lock-free data structures. In Proceedings of the 20th Annual International Symposium on Computer Architecture ISCA '93, pages 289-300, New York, NY, USA, 1993. ACM. |
| [18] | M. Herlihy and N. Shavit. The Art of Multiprocessor Programming . Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 2008. |
| [19] | M. P. Herlihy. Impossibility and universality results for wait-free synchronization. In Proceedings of the Seventh Annual ACM Symposium on Principles of Distributed Computing , PODC '88, pages 276-290, New York, NY, USA, 1988. ACM. |
| [20] | M. P. Herlihy and J. M. Wing. Linearizability: A Correctness Condition for Concurrent Objects. ACM Transactions on Programming Languages and Systems , 12(3):463-492, 1990. |
| [21] | C. A. R. Hoare. Monitors: An Operating System Structuring Concept. Communications of the ACM , 17(10):549-557, 1974. |
| [22] | C. A. R. Hoare. Communicating sequential processes , volume 178. Prentice-hall Englewood Cliffs, 1985. |
| [23] | W.-L. Hung and V. K. Garg. AutoSynch: An Automatic-signal Monitor Based on Predicate Tagging. In PLDI , pages 253-262, 2013. |
| [24] | V. Kodaganallur. Incorporating Language Processing into Java Applications: A JavaCC Tutorial. IEEE Softw. , 21(4):70-77, 2004. |
| [25] | A. Kogan and M. Herlihy. The future(s) of shared data structures. In PODC , 2014. |
| [26] | A. Kogan and E. Petrank. Wait-free queues with multiple enqueuers and dequeuers. In Proceedings of the 16th ACM Symposium on Principles and Practice of Parallel Programming , PPoPP '11, pages 223-234, New York, NY, USA, 2011. ACM. |
| [27] | A. Kogan and E. Petrank. A methodology for creating fast wait-free data structures. In Proceed- ings of the 17th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming PPoPP '12, pages 141-150, New York, NY, USA, 2012. ACM. |
| [28] | G. Korland, N. Shavit, and P. Felber. Noninvasive concurrency with Java STM. In MultiProg 2010: Third Workshop on Programmability Issues for Multi-Core Computers , 2010. |
| [29] | D. Lea. The Java.Util.Concurrent Synchronizer Framework. Sci. Comput. Program. , 58(3):293- 309, 2005. |
| [30] | V. Luchangco and V. J. Marathe. Revisiting condition variables and transactions. In The 6th ACM SIGPLAN Workshop on Transactional Computing , 2011. |
| [31] | U. Meyer and P. Sanders. δ -stepping: A parallelizable shortest path algorithm. J. Algorithms 49(1):114-152, Oct. 2003. |
- [32] M. M. Michael. High Performance Dynamic Lock-free Hash Tables and List-based Sets. In Proceedings of the Fourteenth Annual ACM Symposium on Parallel Algorithms and Architectures , pages 73-82, New York, NY, USA, 2002. ACM.
- [33] K. Pingali, D. Nguyen, M. Kulkarni, M. Burtscher, M. A. Hassaan, R. Kaleem, T.-H. Lee, A. Lenharth, R. Manevich, M. Méndez-Lojo, D. Prountzos, and X. Sui. The Tao of Parallelism in Algorithms. In Proceedings of the 32Nd ACM SIGPLAN Conference on Programming Language Design and Implementation , pages 12-25, New York, NY, USA, 2011. ACM.
- [34] N. Shavit and D. Touitou. Software transactional memory. In Proceedings of the Fourteenth Annual ACM Symposium on Principles of Distributed Computing , PODC '95, pages 204-213, New York, NY, USA, 1995. ACM.
- [35] A. Skyrme and N. Rodriguez. From locks to transactional memory: Lessons learned from porting a real-world application. In The 8th ACM SIGPLAN Workshop on Transactional Computing , 2013.
- [36] S. Timnat, A. Braginsky, A. Kogan, and E. Petrank. Wait-free linked-lists. In Proceedings of the 17th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming , PPoPP '12, pages 309-310, New York, NY, USA, 2012. ACM.
- [37] W. Vogels. Eventually Consistent. Communications of the ACM , 52(1):40-44, 2009.
## Appendix
## A Implicit/Automatic signal monitors:
Monitors can be divided into two categories according to the different implementations of conditional synchronization: explicit-signal monitors and implicit-signal monitors [3]. In explicit-signal monitors, condition variables are used along with wait , and notfiy/notifyAll statements for conditional synchronization between threads. Programmers need to associate assertions with condition variables manually. When a thread wants to execute a set of instructions under a critical section, it first checks some condition variable(s), and waits if the predicate is not true. Once a thread enters the waiting state, another thread on detecting that the conditional predicate has become true, explicitly notifies the waiting thread. The bounded-buffer implementation in Fig. 1(a) that uses Java's reentrant locks is an explicit-signal monitor. In fact, almost all of the prevalent programming languages, including Java, use the explicit-signal design for monitors.
Implicit-signal monitors, also called automatic-signal monitors, require the underlying system to handle thread synchronization and wait/notification. Recently, Hung et al. [23] showed that implicitsignal monitors can be beneficial when implemented on modern multi-core processors. In an implicitsignal design, programmers need to use waituntil statements (line 9 15 , in automatic-signal program in Fig. 1(b)) instead of condition variables for synchronization. In this monitor design, a thread will wait as long as the condition of a waituntil statement is false, and execute the remaining instructions only after the condition becomes true. The responsibility of identifying a waiting thread to signal, and then signaling it, is that of the underlying system/framework rather than of the program logic explicitly put in the thread by a programmer. The details of design and implementations of such monitors is beyond the scope of this paper, and we refer the interested reader to [21, 23].
## B Monitor Task Examples
Here, we highlight some key aspects of our framework's handling of monitor tasks. First, note that the set of statements, S , of a task may be empty. Such a task serves as a barrier; and is legal in our framework. Second, the precondition can either be absent or appear later, and not as the first statement, in the monitor method. When a monitor method has no precondition, our framework creates a task with the precondition as tautology, indicating that the task can be executed at any time. If a monitor method does not start with a waituntil statement but has some such statement in between, then the precondition of the first derived task is a tautology. In Fig. 5, the method bar() has two corresponding tasks T 1 and T 2 , where T 1 has the precondition as tautology and the set of statements S 1 ; T 2 has the precondition P and the set of statements S 2 .
Furthermore, monitor tasks are compositional in nature. Consider the method foo() in Fig. 5. If a method declares n sets of preconditions and statements, then the framework would generate n tasks such that each task T i , 1 ≤ i ≤ n , has a precondition P i and a corresponding set of statements S i . Appendix B presents some detailed examples of these cases.
```
<---------------------------------------------------------------------------------------------------------------------- </----------------------------------------------------------------------------------------------------------------------
void foo() {
&---------------------------------------------------------------------------------------------------------------------- &----------------------------------------------------------------------------------------------------------------------- &---------------------------------------------------------------------------------------------------------------------- &-----------------------------------------------------------------------------------------------------------------------
int +----------------------------------------------------------------------------------------------------------------------
void foo() {
waiituntil(P_n);
&
+
```
Figure 5: Examples of methods leading to compositional tasks and tautology as precondition
## C Lemma Proofs
## C.1 Proof Sketch for Lemma 1
We show that for any execution in our model there exists an equivalent lock-based execution. Since all tasks of any monitor object are executed by a single thread due to Rule 1, mutual exclusion is preserved just as in any lock-based execution. We only need to show that the order of execution of the tasks corresponds to a schedule in which worker threads execute the tasks.
It is sufficient to show that all tasks submitted by a single worker thread execute in the order of submissions. Let s and t be two consecutive tasks submitted by a worker thread. If they are submitted for the same monitor, then the Rule 2 preserves the order. If s is a blocking task, then by definition of blocking task, t cannot be submitted before s is completed. Hence, execution of s precedes execution of t . If s is a non-blocking task and is on a different monitor object from t , then due to Rule 3 we wait for s to finish before submission of t .
## C.2 Proof Sketch for Lemma 2
It is sufficient to show that for every thread history there exists an equivalent sequential thread history that is consistent with the execution by the monitor thread. We get this sequence by considering as linearization point for an operation the instant at which the monitor thread finishes executing the corresponding task. We show that this order is consistent with the thread order.
Let s < t denote that operation(i.e. its corresponding task) s was executed before t . First consider the case when s < t because s is a blocking operation. In this case, task t cannot be submitted before task s is executed. Hence, the order of execution of monitor tasks preserves the thread order. Second, consider the case when s < t because they are operations on the same monitor object and s is invoked before t . In this case Rule 2 guarantees that the order of execution is s followed by t . Finally, consider the case when s and t are tasks submitted by the same thread, s is non-blocking but its results are used by t . In this case, the execution would have to block for collecting the results of s before executing any further. Hence t would be submitted later than the execution of s .
## D Implementation Details
The translation of the code written with our two proposed keywords to an equivalent Java code is performed by using a pre-processor. For our implementation, we use JavaCC [24] pre-processor. We convert all the monitor methods to equivalent tasks, and use a threadpool based executor service framework for monitor threads. Here, we first discuss the pre-processing steps.
## D.1 Pre-processing
We briefly summarize the concepts applied for compiling the code, using the ActiveMonitor framework, in which the two proposed keywords are used.
First, our pre-processor identifies the waituntil statements that capture the preconditions of monitor tasks. Recall that the waituntil keyword is used for automatic signaling between threads, and the predicate provided as its argument forms the precondition for execution of monitor tasks. In the corresponding generated code (in Java) every predicate is created as an inner class with a method called isTrue () , which returns a boolean value as the result of the predicate evaluation. At runtime, our system invokes the isTrue () method to evaluate the predicate when deciding which task should be executed.
Next, the pre-processor creates a Callable [29] or Runnable [29] object that contains the set of statements for each task inside its call () or run () method. If the return value of the original monitor method is void , then we create a Runnable object; otherwise, we create a Callable object. Note that if a non-blocking monitor method has been divided into multiple tasks, the resulting tasks may have some shared variables. To handle these variables, our system creates a data object to store them and all the tasks can access them through this data object.
Finally, we replace the invocations of monitor methods by submitting monitor tasks to threadpool executor (Appendix D.2). The executor returns f , an instance of Java's Future object [29], to the invoker thread when a task is submitted. If the task is blocking, then the thread needs to wait for the computation in f to finish; this is done by evaluating f within the task and returning the result. Otherwise, f is registered with the threadpool executor for exception checking and handling.
In Section 3, we mentioned that our current implementation does not admit a particular family of methods for task based executions. Specifically, it does not admit recursive blocking calls, monitor methods in which conditional synchronization guards using automatic thread signaling ( waituntil ) are not initial statements of the program, or are present in conditional branches. This does not mean that programs requiring such method implementations are not allowed in our framework; just that at present we do not support task generation of such methods. A programmer can implement them using conventional Java syntax, and still use our framework for other methods.
## D.1.1 Discussion: Thread Dependent Variables and Functions
In our current implementation, thread dependent variables and functions within a monitor method cannot be used directly in the Runnable or Callable object that is used in task generation by our approach. This is because the tasks are executed by the monitor thread and not by the worker thread. For example, suppose there is a monitor method that invokes Thread.currentThread() , if we directly add this statement to the generated Runnable object (in the task), then this method's invocation at runtime will return the reference to the monitor thread when it is executed. However, it is obvious that the intent of this call inside the monitor method was to refer to the worker thread. To handle this situation, currently, we require the programmer to perform reference copy and storage and storage in thread-local variables. For read operations of thread dependent variables and functions, the worker thread would need to evaluate them outside the monitor, and store the result with final variables. These final variables can be accessed by the runnable and callable objects. An additional constraint/limitation applies for the case of write operation on thread dependent variables. For write operations, if the monitor method is non-blocking then the results can be stored as intermediate data. The worker thread then writes these results back to its local variable after the task is executed.
## D.1.2 Discussion: Blocking recursive method
Our current pre-processing implementation does not support a b locking recursive method on monitors. This is because the number of the method invocations to be made at the runtime is non-deterministic. Thus, we cannot know how many tasks we need to create at pre-processing time. In addition, since the method is blocking, the monitor thread will get blocked when it recurs.
## D.2 Limiting Monitor Threads using ThreadPools
A threadpool is a collection of threads to execute tasks that are units of computation. These threads are usually created together at the start-up , and remain in the pool to provide executor service. For task execution based programs, the use of a threadpool can significantly improve the runtime performance by having an already existing thread ready to execute the tasks as they arrive. Generally, tasks are stored in a collection, and free threads are responsible for finishing the unexecuted tasks. If there is no task that is eligible for execution, the threads in the pool wait for one to arrive. This approach is especially beneficial when the number of tasks is greater than the number of threads in a pool. Thus, the size of the threadpool is a crucial factor as the timing of creating or destroying a thread may have a significant impact on performance. Given that we spawn a new thread for a monitor object, programs with a large number of monitors could be adversely affected by having too many additional threads in the ActiveMonitor framework.
To keep the overhead of monitor threads relatively low, we control the number of monitor threads created by querying the operating system for available resources and managing the monitor threads dynamically. A component, called Monitor ThreadPool Executor , manages this process. This component instantiates monitor threads as threads of a threadpool at the start of the execution. The pre-processing phase collects the information about number of monitor objects in the program, and the threadpool executor uses this information for creating a threadpool balancing the resource consumption and performance benefit.
## D.2.1 Exception Handling
For a non-blocking method invocation, after submitting its corresponding task to the executor, the invoker does not need to wait for the completion of the task. The task is executed in parallel by a thread of the monitor threadpool. Thus, if an exception occurs during its execution, the thread
that submitted it must be notified of this exception. The ActiveMonitor framework has an exception handler that keeps a log of every exception and provides different mechanisms for programmers to handle exceptions in the non-blocking method. The users may choose to ignore the exceptions or they can specify a maximum number of times a task may be considered for automatic re-tries. Furthermore, our system also provides a hook so that the programmer can write their custom exception handler.
| null |
[
"Weil-Lun Hung",
"Himanshu Chauhan",
"Vijay K. Garg"
] |
2014-08-04T20:58:16+00:00
|
2014-08-04T20:58:16+00:00
|
[
"cs.DC",
"cs.PL"
] |
ActiveMonitor: Non-blocking Monitor Executions for Increased Parallelism
|
We present a set of novel ideas on design and implementation of monitor
objects for multi-threaded programs. Our approach has two main goals: (a)
increase parallelism in monitor objects and thus provide performance gains
(shorter runtimes) for multi-threaded programs, and (b) introduce constructs
that allow programmers to easily write monitor-based multi-threaded programs
that can achieve these performance gains. We describe the concepts of our
framework, called ActiveMonitor, and its prototype implementation using
futures. We evaluate its performance in terms of runtimes of multi-threaded
programs on linked-list, bounded-buffer, and other fundamental problems
implemented in Java. We compare the runtimes of our implementation against
implementations using Java's reentrant locks, recently proposed automatic
signaling framework AutoSynch, and some other techniques from the literature.
The results of of the evaluation indicate that monitors based on our framework
provide significant gains in runtime performance in comparison to traditional
monitors implemented using Java's reentrant locks.
|
1408.0819v1
|
## On the charge transfer between single-walled carbon nanotubes and graphene
Rahul Rao, a) Neal Pierce, Archi Dasgupta #
Honda Research Institute USA, Columbus, Ohio 43212, USA
## Abstract
It is important to understand the electronic interaction between single-walled carbon nanotubes (SWNTs) and graphene in order to use them efficiently in multifunctional hybrid devices. Here we deposited SWNT bundles on graphene-covered copper and SiO2 substrates by chemical vapor deposition and investigated the charge transfer between them by Raman spectroscopy. Our results revealed that, on both copper and SiO2 substrates, graphene donates electrons to the SWNTs, resulting in p-type doped graphene and n-type doped SWNTs.
a) Electronic Mail: [email protected]
# On leave from Penn State University, State College, PA, USA
SWNTs and graphene are unique nanostructured forms of carbon that possess a variety of fascinating electrical, mechanical and chemical properties. It is therefore of 1 great interest to use them in hybrid multifunctional architectures where the unique properties of both materials are utilized synergistically. Indeed there have been several recent reports in the literature on hybrids between nanotubes and graphene in a variety of different configurations, 2-4 and that have exhibited superior performances in applications such as energy storage and optics. 5-7 The nanotube/graphene hybrid devices rely on electronic interactions between the constituents, and it is important to understand these interactions at a fundamental level. Graphene is a semi-metal, whereas SWNTs can be either semiconducting or metallic. Moreover, it is known that 1 substrates and the environment play important roles in regulating the charge densities of both graphene and SWNTs; for example, adsorption of oxygen typically causes doping by positive charge carriers (holes). 8,9
Raman spectroscopy is a powerful tool that has been used extensively to study charge transfer in SWNTs and in graphene. 10-19 Adsorption of electron donating and accepting molecules cause shifts in the Raman peak frequencies, which are proportional to the amount of carriers injected. 10 While charge transfer in SWNTs and graphene due to molecular dopants has been studied separately, there is very little work on the electronic interaction between SWNTs and graphene. One recent study 18 reported the interaction between an individual metallic SWNT and graphene, and found the graphene to be doped n-type by the SWNT. In their study the SWNT was first deposited on a SiO substrate, followed by transfer of chemical vapor deposition (CVD) grown 2
graphene on top of the SWNT. For applications, however, one must consider macroscopic architectures of SWNTs and graphene involving both semiconducting and metallic tubes. Furthermore, the role of the substrate (especially conductive substrates) cannot be neglected when considering charge transfer between SWNTs and graphene.
To this end, here we investigated the electronic interaction between SWNT films and graphene on metal (copper) substrates, an architecture that is attractive for applications such as energy storage. We thus studied the fundamental charge transfer processes that occur due to the matching of Fermi levels between SWNTs, graphene and the Cu substrate in assembled hybrid architectures. Our samples were made by the deposition of SWNT bundles grown via CVD directly on to graphene-coated Cu substrates (obtained from Graphene Laboratories Inc.). The electronic interaction between the SWNTs and graphene was studied by Raman spectroscopy. In addition, the graphene was also transferred from Cu to SiO2 substrates, followed by SWNT deposition in order to compare the effect of an insulating substrate on the charge densities in SWNTs and graphene.
It is important to note is that it is not straightforward to study charge transfer between SWNTs and graphene via Raman spectroscopy. Since both graphene and SWNTs consist of sp -bonded carbon, their Raman peaks overlap, making it difficult to 2 distinguish between their peak frequencies. To circumvent this issue, we used 13 C isotope-labeled graphene (hereafter called 13 C-graphene) on Cu foils. The Raman peaks of 13 C-graphene are red-shifted from non isotope-labeled graphene by a factor equaling
√(12/13), making it possible to unambiguously identify the SWNT and graphene peaks. 2 We were thus able to accurately analyze peak shifts and thereby study the electronic interaction between SWNTs and 13 C-graphene on Cu and SiO2. For comparison, the control samples were as-prepared 13 C-graphene on Cu, and SWNTs deposited on plain Cu foils.
The SWNTs were deposited on the 13 C-graphene by floating catalyst CVD. The 13 C-graphene-coated Cu and SiO2 substrates were loaded downstream at the exhaust end of a 1' diameter quartz tube. The substrates were placed vertically upright so that they faced the flow of gases through the furnace. A syringe pump was connected to the inlet end of the quartz tube for delivery of the precursor. Mixtures of ferrocene (1 wt.%) and ethanol were injected at rates of 12-15 ml/hr into the hot zone of the tube furnace once it reached the growth temperature of 1000 ºC. The precursor evaporated at the inlet of the furnace and the iron and carbon species were carried through the furnace with a mixture of argon (500 sccm) and hydrogen (200 sccm). In the floating catalyst process, the SWNTs grow from iron particles in-flight and are deposited on substrates that are placed downstream. Since the substrates were well outside the hot zone of the furnace, their temperature only went up to ~40 ºC during the growth. The growth time was kept between 5 and 15 minutes so that the deposited SWNT film consisted of sparse individual SWNTs or small bundles. Reducing the amount of the deposited SWNTs allowed us to easily measure the Raman spectra from the graphene underneath. Following SWNT deposition, the samples were analyzed by scanning electron microscopy (SEM) and Raman spectroscopy with multiple excitation wavelengths (633,
532 nm). The laser power was kept low (<2 mW) for the Raman measurements in order to avoid the effects of heating.
Examples of Raman spectra collected with the 633nm excitation from the 13 Cgraphene before (top trace) and after (bottom trace) SWNT deposition are shown in Fig. 1. The Raman spectrum from graphene typically exhibits two intense peaks, labeled the G and G' bands. 20 The G band corresponds to in-plane lattice vibrations of the carbon atoms in the graphene lattice. The G' band is a dispersive second order peak corresponding to scattering of two transverse optic (TO) phonons, and is typically much more intense than the G band in graphene. The G band typically appears at ~1585 cm -1 in graphene on Cu; however it is red-shifted in 13 C-graphene by √(12/13) due to the 13 C isotope. As can be seen in Fig. 1, the G band in the 13 C-graphene appears at ~1520 cm -1 . The G' band undergoes a similar redshift and appears at ~2530 cm -1 (with the 633 nm excitation).
FIG. 1. Typical Raman spectra collected with 633 nm excitation from the 13 C-graphene on Cu before (top trace) and after (bottom trace) SWNT deposition. The G and G' bands have been fitted with Lorentzian peaks. The upper left inset shows the low frequency RBM peaks from the SWNTs on the 13 C-graphene. The upper right inset is an SEM image showing low-density SWNTs on 13 C-graphene.
The G band in a SWNT is more complicated than graphene; it corresponds to multiple vibrational modes tangential to the nanotube axis, with the most intense peak (called G + ) appearing at ~1590 cm -1 20 . Moreover, the Raman spectrum of the SWNT exhibits unique peaks in the low frequency region, called radial breathing modes (RBMs), which correspond to lattice vibrations in the radial direction. The RBM peak
frequencies are inversely proportional to the nanotube diameters. The bottom trace in Fig. 1 shows an example of a Raman spectrum collected after SWNT deposition on the 13 C-graphene. Both the G and G' peaks from the graphene and SWNTs are clearly distinguishable. Also shown in Fig. 1 (upper inset) are the low frequency RBMs from the SWNTs. At least two peaks can be observed at 130 and 153 cm -1 , indicating that the SWNT diameters range between 1-1.5 nm. 20 Similar Raman spectra were obtained from the SWNTs deposited on the 13 C-graphene-covered SiO2 substrates. An SEM image of the SWNTs on the 13 C-graphene is shown in the upper right inset in Fig. 1. As mentioned above, the deposition times were kept short so that the SWNT film did not become too thick to prevent the measurement of the underlying graphene Raman peaks. The SWNTs observed in Fig. 1 appear to be individual or in least small bundles, along with some catalyst particles that inevitably decorate the SWNTs during deposition. The G + band from bundled SWNTs (10 - 20 cm -1 ) is typically broader than for individual SWNTs (<10 cm -1 ). 20 Furthermore, the linewidths of RBMs of individual SWNTs are very narrow (1-3 cm -1 ). 21 In our spectra the average G + band linewidth is 15 cm -1 and the RBM linewidths range from 5-10 cm -1 . The broadened peaks in addition to the observation of multiple RBMs supports the presence of bundles in the SWNTs. In addition to the G and G' bands, a third peak called the D band typically appears at ~1330 cm -1 (with 633 nm excitation) in both the SWNT and graphene Raman spectrum due to disorder. 20 As can be seen in Fig. 1, the spectrum in the D band region exhibits negligible intensity after SWNT deposition, indicating that the SWNT deposition process did not cause any damage to the 13 C-graphene.
The dotted lines in Fig. 1 clearly show that the Raman peaks of graphene under the SWNTs are blueshifted compared to their frequencies before SWNT deposition. These shifts in frequencies are more clearly illustrated in Fig. 2a, which plots the graphene G' peak frequencies against the G peak frequencies from several different samples. The data in Fig. 2a are a comparison of the Raman peaks from the 13 Cgraphene on Cu, before and after SWNT deposition (labeled Graphene/SWNTs in Fig. 2a). In general both the graphene G and G' peaks from graphene are blueshifted after SWNT deposition, indicating hole doping in the graphene. 14 The average G band frequency is blueshifted by 3 cm -1 , while the average G' band frequency is blueshifted by 30 cm -1 due to charge transfer from the SWNTs. It should be noted that the graphene is doped with holes even before SWNT deposition. Pristine un-doped 13 C-graphene exhibits a G band at ~1520 cm -1 (calculated with the assumption that the pristine 12 Cgraphene G band is at ~1580 cm -1 ), which is 3 cm -1 lower than the average G band frequency observed in the 13 C-graphene on the Cu foil (1523 cm -1 ). The dotted lines in Fig. 2a show the expected G and G' band frequencies for pristine (un-doped) 13 Cgraphene. Graphene is known to be hole-doped due to the oxidized Cu substrate following CVD growth, 22 confirming the present observations. The data in Fig. 2a therefore suggests that the 13 C-graphene becomes hole doped due to the interaction with SWNTs. In other words, graphene donates its electrons to the SWNTs.
The Raman peaks from the SWNTs exhibit an equal and opposite behavior compared to the graphene peaks after SWNT deposition on the 13 C-graphene. Fig. 2b shows the G and G' peak frequencies collected from several samples. As mentioned
above, the control samples were SWNT films deposited on plain Cu foils. Unlike the graphene Raman peaks which blueshift after SWNT deposition (Fig. 2a), the Raman peaks from the SWNTs redshift due to interaction with graphene (labeled SWNT/Graphene in Fig. 2b). The average G band frequency in the SWNT/graphene is redshifted by 3 cm -1 compared to the corresponding average value from SWNTs on plain Cu. The G' band frequency is blueshifted by an average of 23 cm -1 . Similar to Fig. 2a, the dotted lines in Fig. 2b denote the frequencies of undoped SWNTs on Cu. Previously it has been shown that doping of SWNTs by electron donor species causes a redshift of the G band frequency. 12,13,15 The observed redshift of the G band in SWNTs/graphene can therefore be attributed to transfer of electrons from the 13 C-graphene to the SWNTs. The n-type doping of the SWNTs is also concomitant with the p-type doping observed in the graphene (Fig. 2).
FIG. 2. (a) Plot of the G' band frequencies against the G band frequencies from the (a) 13 C-graphene with and without SWNTs on Cu substrates, and (b) from the pristine SWNTs and SWNTs on 13 C-graphene coated Cu. The data points indicate the average values in the data sets. The dotted lines in (a) indicate the frequencies of the pristine (un-doped) 13 C-graphene.
We also compared the Raman spectra from SWNTs deposited on SiO2 substrates and graphene-coated SiO2. The graphene was transferred from Cu to SiO2 using the following procedure - A PMMA solution (Microchem 495 PMMA A2) was spin-coated on the graphene/copper foils at 4000 rpm for 60 s and then placed on a ~180 °C hot plate for 1 min. The graphene on the backside of the foil was removed by etching in a nitric acid solution (1 HNO3: 3 H2O) for 3 min. The copper was then removed with a 0.1 M ammonium persulfate solution. After rinsing the graphene film with distilled water, it was placed on an SiO2 substrate. After air-drying, the graphene-on-SiO2 was heated at 180 °C for 1 hour to minimize wrinkling of the graphene film. Finally, the PMMA was removed with an acetone bath.
The Raman spectra from the SiO2 substrates exhibited shifts in the G and G' band frequencies similar to the Cu substrates, with some obvious differences. Figs. 3a and 3b show the G and G' band frequencies from the 13 C-graphene and SWNTs on SiO2. Comparing the peak frequencies from the bare graphene (without SWNTs) on Cu (Fig. 2a) and on SiO2 (Fig. 3a), it is apparent that even before SWNT deposition the doping levels in the graphene are different between the two substrates. Recall that the G band frequency in undoped 13 C-graphene lies at ~1520 cm -1 (dotted vertical line in Fig. 3a). The G band frequencies in the graphene on SiO2 vary from 1510 - 1525 cm -1 , indicating that the doping in graphene on SiO2 varies from n-type to p-type depending on the spot/sample measured. The different doping levels are probably due to the transfer process (described above) from Cu to SiO2, which involves PMMA and other organic solvents that are possibly only partially removed at the end of the process, leaving
behind pockets of doped graphene. However, similar to the case of the Cu substrates, the Raman peaks of the 13 C-graphene under the SWNTs are blueshifted. The average G and G ' band frequencies in the graphene/SWNTs are blueshifted by 10 cm -1 and 30 cm -1 , respectively (Fig. 3a). Thus the graphene is doped with excess holes from the SWNTs. The Raman peaks of the SWNTs redshift concomitantly, although the shift of the G' band is not as severe as on Cu. As can be seen in Fig. 3b, the G band from the SWNTs is redshifted by an average of 5 cm -1 , while the G' band is blueshifted by ~5 cm -1 .
FIG. 3. (a) Plot of the G' band frequencies against the G band frequencies from the (a) 13 C-graphene with and without SWNTs on SiO2 substrates, and (b) from the pristine SWNTs and SWNTs on 13 C-graphene coated SiO2. The data points indicate the average values in the data sets. The dotted lines in (a) indicate the frequencies of the pristine (un-doped) 13 C-graphene.
In spite of the differences in frequencies of the Raman peaks of the graphene and SWNTs on Cu and SiO2, the direction of shifts of the graphene and SWNT peaks on SiO2 is similar to those on Cu. The Raman analysis indicates p-type and n-type doping in the 13 C-graphene and SWNTs, respectively. We also observe similar shifts in peak frequencies with a different laser excitation (532 nm, not shown here). Our results are in contrast to the previous Raman measurements performed on a graphene-covered individual metallic SWNT. 18 In their study, Paulus et al. found the graphene to be ndoped due to the charge transfer from the metallic SWNT. The graphene G band was redshifted by 10 cm -1 , from which they deduced an electron density of 1.12 x 10 13 cm -2 due to electron doping from the SWNT.
In our case the G band of the 13 C-graphene is blueshifted by an average of 3 cm -1 (maximum 9 cm -1 ). Comparing our peak shift with data obtained on gated graphene devices, 14 we estimate the hole concentration in the graphene to be ~0.5 x 10 13 cm -2 (with a maximum of 1 x 10 13 cm -2 ). It is more challenging to determine the charge concentration in SWNTs based on shifts in the Raman spectra. The G band in SWNTs comprises of several peaks that correspond to different types of vibrations tangential to the axis of the SWNT. The two most intense peaks are labeled G and G + for the lower and higher frequency peaks, respectively. The various changes in the Raman spectra due to doping include differences in the intensity ratios between the G and G + peaks, broadening/stiffening of the peaks, and frequency shifts. 14,16,26 Moreover, the asprepared SWNT bundles contain a mixture of semiconducting and metallic SWNTs, 10,15 which have different charges concentrations and charge transfer characteristics with
graphene. Nevertheless, an average redshift of the SWNT G band frequency on Cu (3 cm -1 ) and SiO2 (5 cm -1 ) indicates electron transfer to the SWNTs from the 13 C-graphene.
The p-type and n-type doping of the graphene and SWNTs make sense if one considers their work functions. The work function of graphene has been estimated to be ~4.6 eV 23 while that of SWNTs varies depending on the whether the SWNT is individual (semiconducting or metallic) or in bundles. The peaks in the Raman from bundled SWNTs are typically broader than peaks from individual SWNTs. The In our samples the SWNTs are either individual or in small bundles (Fig. 1). We can therefore use the work function determined for bundles, which is ~4.8 eV. 24 Since the work function is the difference in energy between the vacuum level and the Fermi level, a smaller work function implies a higher lying Fermi level and electrons will flow from this material to the one with the higher work function (lower lying Fermi level). Thus we expect electrons to flow from the graphene to the SWNTs where they contact each other. Moreover, the work functions of Cu and CuO are 4.65 and 5.3 eV, 25 respectively, which are both larger than that of graphene, implying that electrons flow from the graphene to the Cu. This reasoning is in accord with the experimental observation of p-type doping in graphene on Cu prior to SWNT deposition. As mentioned above, the present results contradict the findings of Paulus et al ., where they found n-doping of graphene by the SWNT, 18 rather than p-doping. In their case the interaction of the graphene was with a single metallic SWNT, whose work function was estimated to be 4.5 eV, i.e. lower than that of graphene. The lower work function of the individual SWNT caused the electron doping of graphene.
The differences observed in the electronic interaction between an individual metallic SWNT (Ref. 18) and bundles (present results) highlight the uniqueness of the SWNT and graphene electronic structures, as well as the importance of studying the electronic interactions between graphene and SWNTs in different architectures. With our unique synthesis method, we were able to directly deposit SWNTs on graphenecoated Cu substrates, thereby simulating a hybrid SWNT-graphene device on a metal electrode. Our method can also be easily extended towards the deposition of SWNTs on other 2D layered materials, for example MoS2, whose hybrid with SWNT films has recently been demonstrated as a gate-tuneable p-n junction. 26
## Acknowledgements
AD acknowledges financial support from the Honda Research Institute, USA Inc.
## References
- 1 H.S.P. Wong and Deji Akinwande, Carbon Nanotube and Graphene Device Physics (Cambridge University Press, 2010).
2 R. Rao, G. Chen, L.M.R. Arava, K. Kalaga, M. Ishigami, T.F. Heinz, P.M. Ajayan, and A.R. Harutyunyan, Sci. Rep. 3 , 1891 (2013).
- 3 D.D. Nguyen, N.-H. Tai, S.-Y. Chen, and Y.-L. Chueh, Nanoscale 4 , 632 (2012).
- 4 R.K. Paul, M. Ghazinejad, M. Penchev, J. Lin, M. Ozkan, and C.S. Ozkan, Small 6 , 2309 (2010).
5 Z. Fan, J. Yan, L. Zhi, Q. Zhang, T. Wei, J. Feng, M. Zhang, W. Qian, and F. Wei, Adv. Mater. 22 , 3723 (2010).
6 U.J. Kim, I.L. Lee, J.J. Bae, S. Lee, and G.H. Han, Adv. Mater. 23 , 3809 (2011).
- 7 J. Lin, C. Zhang, Z. Yan, Y. ZHU, Z. Peng, R.H. Hauge, D. Natelson, and J.M. Tour, Nano Lett 13 , 72 (2012).
8 P.G. Collins, Science 287 , 1801 (2000).
- 9 Y. Shi, X. Dong, P. Chen, J. Wang, and L.-J. Li, Phys. Rev. B 79 , 115402 (2009).
10 C.N.R. Rao and R. Voggu, Materials Today 13 , 34 (2010).
11 A. Rao, P. Eklund, S. Bandow, A. Thess, and R. Smalley, Nature 388 , 257 (1997).
- 12 A. Claye, S. Rahman, J.E. Fischer, and A. Sirenko, Chemical Physics 333 , 16 (2001).
13 G. Chen, C.A. Furtado, U.J. Kim, and P.C. Eklund, Phys. Rev. B 72 , 155406 EP (2005).
14 A. Das, S. Pisana, B. Chakraborty, S. Piscanec, S. Saha, U. Waghmare, K. Novoselov, H. Krishnamurthy, A. Geim, and A. Ferrari, Nature Nanotechnology 3 , 210 (2008).
15 R. Voggu, C.S. Rout, A.D. Franklin, T.S. Fisher, and C.N.R. Rao, The Journal of Physical Chemistry C 112 , 13053 (2008).
16 J. Park, K. Sasaki, R. Saito, W. Izumida, M. Kalbac, H. Farhat, G. Dresselhaus, and M. Dresselhaus, Phys. Rev. B 80 , 081402 (2009).
17 D. Fu, W. Fang, Y. Shi, P. Chen, and L. Li, Small 5 , 1422 (2009).
18 G.L.C. Paulus, Q.H. Wang, Z.W. Ulissi, T.P. McNicholas, A. Vijayaraghavan, C.-J. Shih, Z.
Jin, and M.S. Strano, Small 9 , 1954 (2012).
19 A.C. Crowther, A. Ghassaei, N. Jung, and L.E. Brus, ACS Nano 6 , 1865 (2012).
20 A. Jorio, M.S. Dresselhaus, and R. Saito, Raman Spectroscopy in Graphene Related Systems: Carbon Nanotubes, Nanographite and Graphene (Wiley-VCH, 2011).
21 R. Rao, C. Poweleit, J. Menendez, A.M. Rao, Phys. Rev. Lett. 99 , 047403 (2007).
22 J.D. Wood, S.W. Schmucker, A.S. Lyons, E. Pop, and J.W. Lyding, Nano Lett 11 , 4547 (2011).
23 G. Giovannetti, P. Khomyakov, G. Brocks, V. Karpan, J. van den Brink, and P. Kelly, Phys. Rev. Lett. 101 , (2008).
24 S. Suzuki, C. Bower, Y. Watanabe, and O. Zhou, Appl. Phys. Lett. 76 , 4007 (2000).
25 S.C. Lim, J.H. Jang, D.J. Bae, G.H. Han, S. Lee, I.-S. Yeo, and Y.H. Lee, Appl. Phys. Lett. 95 , 264103 (2009).
26 D. Jariwala, V.K. Sangwan, C.-C. Wu, P.L. Prabhumirashi, M.L. Geier, T.J. Marks, L.J. Lauhon, and M.C. Hersam, Proceedings of the National Academy of Sciences (2013).
|
10.1063/1.4893698
|
[
"Rahul Rao",
"Neal Pierce",
"Archi Dasgupta"
] |
2014-08-04T21:14:17+00:00
|
2014-08-04T21:14:17+00:00
|
[
"cond-mat.mtrl-sci"
] |
On the charge transfer between single-walled carbon nanotubes and graphene
|
It is important to understand the electronic interaction between
single-walled carbon nanotubes (SWNTs) and graphene in order to use them
efficiently in multifunctional hybrid devices. Here we deposited SWNT bundles
on graphene-covered copper and SiO2 substrates by chemical vapor deposition and
investigated the charge transfer between them by Raman spectroscopy. Our
results revealed that, on both copper and SiO2 substrates, graphene donates
electrons to the SWNTs, resulting in p-type doped graphene and n-type doped
SWNTs.
|
1408.0821v1
|
## Design and performance of a terahertz photoconductive antenna with nano-crossfinger structure
Jitao Zhang
†
ECE Department,The University of Arizona, Tucson, AZ,85721 † [email protected]
August 6, 2014
Abstract : Improving terahertz(THz) radiation power and/or optics-to-THz efficiency of the photoconductive antenna(PCA) is widely recognized as one of the most attractive and challenging missions in THz community. In this work, the design of a THz PCA with nano-crossfinger structure in the active region is proposed. The THz radiation properties of this PCA was demonstrated by finite-difference-time-domain method based on full-wave model. As a comparison, the PCA with nano-finger structure that promises enhanced THz radiation than conventional PCA was also analyzed numerically. The results indicate that the nano-crossfinger PCA can radiate even higher THz field than the nano-finger PCA, primarily due to the enhanced bias field within the nano-structure.
## 1 Introduction
The existing broadband terahertz(THz) pulse sources can be usually generated through either electronic or optical procedure[1]. On the electronic side, electron accelerator can radiate intense and ultra-broadband THz pulse. However, the bulk facility makes it only available at several places in the world[2, 3]. On the optical side, optical rectification, photoconductive antenna(PCA) and photo-induced gas plasma have been demonstrated before[4]. Among them, the PCA is one of the earliest and most commonly used THz pulse sources due to its simple configuration and promising performance[5]. However, the biggest obstacle of a common PCA is its inefficient radiation, whose averaging power is only in the order of several µW . This problem stems from the low opticsto-THz conversion efficiency, and is eventually determined by the low quantum efficiency. In other words, most of the photo-excited carriers (i.e. electrons
and holes) in the active region of a PCA will be recombined before they can arrive at the PCA's electrodes so that have little contribution to the THz radiation. Many efforts have been made to address this low-conversion-efficiency problem[6, 7, 8, 9, 10]. Very recently, the idea of nano-structure PCA has attracted much attention[11, 12, 13, 14, 15]. The specific design of nano-structures within the bare gap of a common PCA can not only optimize the distribution of the incoming laser beam but also improve the collection efficiency of the electrodes. Thereby, the enhancement of THz power as large as 50 times has been demonstrated successfully[14]. In this work, a PCA with nano-crossfinger structure has been proposed, and its performance was numerically predicated by finite-difference-time-domain(FDTD) method based on full-wave model[16]. Compared with the PCA with nano-finger structure, this PCA can radiate even higher THz field, which makes it a promising candidate of THz pulse source.
## 2 Design and simulation of nano-crossfinger PCA
The structures of the nano-crossfinger and nano-finger PCAs are shown in Fig.1. Different from the nano-finger PCA shown in Fig.1(b), where all of the fingers contact the same electrode, the fingers of the nano-crossfinger PCA shown in Fig.1(a) are divided into half-and-half and contact with different electrodes in an interdigitating way. The fingers of both PCAs have the same width of 100 nm and separation of 100 nm . The finger's length in Fig.1(a) and (b) is 1 7 . µm and 1 µm , respectively. Considering the available memory of the computer, the electrodes of both PCAs are 5 µm by 5 µm and have a gap size of 2 µm in between.
Figure 1: Schematic structure of the (a) nano-crossfinger PCA and (b) nanofinger PCA. The left and right electrodes are anode and cathode, respectively.
In our previous work[16, 17], a full-wave simulation tool has been developed to predicate the radiation properties of the PCA based on FDTD method. The feature of this tool is that the multi-physical phenomena happening in the PCA, such as light-matter interaction, photo-excited carrier dynamics and full-wave propagation of the THz radiation, are considered and embodied in the simulation. Therefore, this tool can fulfill the needs of the comprehensive simulation of the PCA, in which almost all of the parameters that tightly related to the performance of the PCA can be involved, and the THz radiation can be predicated both in the near-field and far-field. This simulation tool was used in this
work. The substrate has a dimension of 12 µm by 5 µm by 2 2 . µm in x y , and z direction, respectively. The incident laser beam illuminated the nano-structure within the gap of the PCA. The distribution of optical absorption on the structure of fingers was simulated by COMSOL, and then was used as input data for FDTD simulation. All of the input parameters of the simulation are kept the same for both PCAs, and summarized in Table.1.
| Parameters | Values |
|--------------------------------------------------|-------------------------------------|
| Material | LT-GaAs electron: 0.1, hole: 0.4 30 |
| Carrier lifetime ( ps ) Mobility ( cm 2 /V · s ) | electron: 200, hole: |
| Permittivity | 12.9 |
| Intrinsic concentration ( cm - 3 | 2.1E6 |
| ) ) | 1E4 |
| Absorption coefficient ( cm - 1 | 800 |
| Laser wavelength ( nm ) | |
| Beam size ( µm ) | 1 |
| Averaging power ( mW ) | 6 |
| Pulse duration ( fs ) | 80 |
| DC voltage ( V ) | 0.5 |
## 3 Simulation results
Figure 2 shows the distribution of the DC bias field of both PCAs. Fig.2(a) to (d) are E fields on the top surface( XOYplane ) of two PCAs. It indicates that E z field of the nano-crossfinger PCA is greatly enhanced, and E x field is enhanced as well. The cut lines drawn in Fig.2(e) and (f) show the enhancement much clearly. The E z field of the nano-crossfinger PCA is approximately 10 times of that of the nano-finger PCA at the peak. Since the bias field is the source of the force that drives the photo-excited carriers to flow to the electrodes, enhanced local bias field will promise larger photo-current and thus higher THz radiation power. One may find in Fig.2(e) that E x is even larger at the bare region of the gap. However, different from the case of the region having finger structure, most of the photo-excited carriers in bare region will have much less contribution to THz radiation since the transport path is too long to be reachable before recombination.
The FDTD simulation was run for 1 ps with time step of 0 033 . fs and mesh size of 20 nm , and the peak of the laser pulse was introduced at 0 15 . ps . The transient E field at 0 15 . ps on the top surface of the PCAs is shown in Fig.3, which clearly indicates the enhancement of the nano-crossfinger PCA. In addition, Figure 4 shows the transient current density on the top surface of the PCAs. It confirmed the prediction above about the function of the bias field. The photo-excited current within the gap region of the nano-crossfinger PCA is larger than that of the nano-finger PCA. This is further confirmed by the
photo-excited current of the cross-section that parallels XOY plane and passes through the center of the laser beam, as shown in Fig.5.
Figure 2: Distribution of the DC bias field. (a) and (c) are E z and E x on the top surface of nano-crossfinger PCA, respectively. (b) and (d) are E z and E x on the top surface of nano-finger PCA, respectively. The horizontal and vertical axises of (a) to (d) represent x and z directions with unit of µm , respectively. The colorbars have unit of V/m . (e) and (f) are cut lines along z axis at the center of x axis. The red and blue solid lines represent the field of nano-crossfinger and nano-finger PCAs, respectively.
The far-field THz radiation was calculated at the point right below the PCA (along z axis) with a distance of 200 mm away from the center of the PCA. Figure 6 shows far-field THz radiations of nano-crossfinger and nano-finger PCAs. The calculated polarized components E φ and E θ of the THz field are along x and y axises, respectively. It is shown that the nano-crossfinger PCA outperforms nano-finger PCA regarding the far-field THz radiation. For example, E φ of nano-crossfinger PCA is 1 2 times of that of nano-finger PCA at peak frequency. .
Moreover, the enhancement of E θ can be as large as 20 times at 10 THz ! This is mainly due to the strongly enhanced bias field in z direction, as shown in Fig.2(f).
Figure 3: Transient E field of (a) nano-crossfinger PCA and (b) nano-finger PCA at 0 15 . ps on the top surface of the PCAs. The x and z aixis are counted by mesh number. The colorbars have unit of V/m .
Figure 4: Transient current density of (a) nano-crossfinger PCA and (b) nanofinger PCA at 0 15 . ps on the top surface of the PCAs. The x and z aixis are counted by mesh number. The colorbars have unit of A/µm 2 .
Figure 5: Time-domain photo-excited current of the cross-section that parallels XOY plane and passes through the center of the laser beam.
Figure 6: Far-field THz radiation. (a) and (b) are time-domain E fields E φ and E θ . (c) and (d) are corresponding spectra calculated by Fast-Fourier-Transform. The red and blue solid lines represent data of nano-crossfinger and nano-finger PCAs, respectively.
## 4 Conclusion
Anano-crossfinger PCA has been designed, and its performance was numerically analyzed and compared with that of another nano-finger PCA. The enhanced THz radiation has been found for this new PCA, which is mainly due to the optimized distribution of the bias field. The proposed PCA is conceptually different from previous interdigitated PCA design. In the interdigitated PCA, the structures have micrometer dimensions, and the whole PCA actually works as an array of common coplanar stripline PCAs. In nano-crossfinger PCA, however, the typical dimension of the fingers is in nanometer. As such, the fingers will not only optimize the distribution of the incoming laser beam by diffraction, but also shorten the transport path of the photo-excited carriers. Therefore, the optics-to-THz efficiency as well as the THz radiation field can be strongly enhanced. Regarding the future work, the far-field pattern and the averaging outpower of the nano-crossfinger PCA will be evaluated numerically, and the experimental verification is also under consideration.
## Acknowledgements
This work can not be done without the contribution of the following colleagues. They are Mingguang Tuo, Min Liang and Hao Xin. They will be listed as co-authors when we consider publishing this work in a journal.
## References
- [1] Y.-S. Lee, Principles of terahertz science and technology . New York, NY: Springer US, 2009.
- [2] G. L. Carr, M. C. Martin, W. R. Mckinney, K. Jordan, G. R. Neil, and G. P. Williams, 'High-power terahertz radiation from relativistic electrons,' Nature , vol. 420, pp. 153-156, 2002.
- [3] Z. Wu, A. S. Fisher, J. Goodfellow, M. Fuchs, D. Daranciang, M. Hogan, H. Loos, and A. Lindenberg, 'Intense terahertz pulses from slac electron beams using coherent transition radiation,' Rev. Sci. Instrum. , vol. 84, pp. 022701-1-10, 2013.
- [4] K. Reimann, 'Table-top sources of ultrashort thz pulses,' Rep. Prog. Phys. , vol. 70, pp. 1597-1632, 2007.
- [5] D. H. Auston, K. P. Chung, and P. R. Smith, 'Picosecond photoconducting hertzian dipoles,' Appl. Phys. Lett. , vol. 45, pp. 284-286, 1984.
- [6] I. Brener, D. Dykaar, A. Frommer, L. N. Pfeiffer, J. Lopata, J. Wynn, and K. West, 'Terahertz emission from electric field singularities in biased semiconductor,' Opt. Lett. , vol. 21, pp. 1924-1926, 1996.
- [7] J. H. Kim, A. Polley, and S. E. Ralph, 'Efficient photoconductive terahertz source using line excitation,' Opt. Lett. , vol. 30, pp. 2490-2492, 2005.
- [8] P. Maraghechi and A. Y. Elezzabi, 'The role of self-similarity in fractal photoconductive thz emitters,' J Infrared Milli Terahz Waves , vol. 32, pp. 1285-1290, 2011.
- [9] S. Winnerl, 'Scalable microstructured photoconductive terahertz emitters,' J Infrared Milli Terahz Waves , vol. 33, pp. 431-454, 2012.
- [10] J. Zhang, 'Enhanced terahertz radiation from an elliptical-beamilluminated sawtooth photoconductive antenna: design and numerical analysis,' arXiv:1407.1095 , vol. XX, pp. 1-7, 2014.
- [11] S.-G. Park, K. H. Jin, M. Yi, J. C. Ye, J. Ahn, and K.-H. Jeong, 'Enhancement of terahertz pulse emission by optical nanoantenna,' ACS Nano , vol. 6, pp. 2026-2031, 2012.
- [12] A. Singh, H. Surdi, V. V. Nikesh, S. S. Prabhu, and G. H. D¨hler, 'Imo proved efficiency of photoconductive thz emitters by increasing the effective contact length of electrodes,' AIP Advances , vol. 3, pp. 122106-1-5, 2013.
- [13] H. Tanoto, J. H. Teng, Q. Y. Wu, M. Sun, Z. N. Chen, S. A. Maier, B. Wang, C. Chum, G. Y. Si, A. J. Danner, and S. J. Chua, 'Nano-antenna in a photoconductive photomixer for highly efficient continuous wave terahertz emission,' Scientific Reports , vol. 3, pp. 1-6, 2013.
- [14] C. W. Berry, N. Wang, M. Hashemi, M. Unlu, and M. Jarrahi, 'Significant performance enhancement in photoconductive terahertz optoelectronics by incorporating plasmonic contact electrodes,' Nature Communication , vol. 4, pp. 1-10, 2013.
- [15] N. Khiabani, Y. Huang, L. E. G.-M. noz, Y.-C. Shen, and A. Rivera-Lavado, 'Significant performance enhancement in photoconductive terahertz optoelectronics by incorporating plasmonic contact electrodes,' Nature Communication , vol. 4, pp. 1-10, 2013.
- [16] J. Zhang, 'Characterization of the terahertz photoconductive antenna by three-dimensional finite-difference time-domain method,' arXiv:1406.3872 , pp. 1-21, 2014.
- [17] J. Zhang, 'Numerical analysis of the emission properties of terahertz photoconductive antenna by finite-difference-time-domain method,' arXiv:1407.2881 , pp. 1-11, 2014.
| null |
[
"Jitao Zhang"
] |
2014-08-04T21:27:00+00:00
|
2014-08-04T21:27:00+00:00
|
[
"physics.optics"
] |
Design and performance of a terahertz photoconductive antenna with nano-crossfinger structure
|
Improving terahertz(THz) radiation power and/or optics-to-THz efficiency of
the photoconductive antenna(PCA) is widely recognized as one of the most
attractive and challenging missions in THz community. In this work, the design
of a THz PCA with nano-crossfinger structure in the active region is proposed.
The THz radiation properties of this PCA was demonstrated by
finite-difference-time-domain method based on full-wave model. As a comparison,
the PCA with nano-finger structure that promises enhanced THz radiation than
conventional PCA was also analyzed numerically. The results indicate that the
nano-crossfinger PCA can radiate even higher THz field than the nano-finger
PCA, primarily due to the enhanced bias field within the nano-structure.
|
1408.0823v3
|
## Quantum Dissipation in a Neutrino System Propagating in Vacuum and in Matter
Marcelo M. Guzzo a , Pedro C. de Holanda , Roberto L. N. Oliveira a a,b a Instituto de F´ ısica Gleb Wataghin Universidade Estadual de Campinas, UNICAMP 13083-970, Campinas, S˜ ao Paulo, Brasil b Northwestern University, Department of Physics & Astronomy, 2145 Sheridan Road, Evanston, IL 60208, USA
## Abstract
Considering the neutrino state like an open quantum system, we analyze its propagation in vacuum or in matter. After defining what can be called decoherence and relaxation e ff ects, we show that in general the probabilities in vacuum and in constant matter can be written in a similar way, which is not an obvious result for such system. From this result, we analyze the situation where neutrino evolution satisfies the adiabatic limit and use this formalism to study solar neutrinos. We show that the decoherence e ff ect may not be bounded by the solar neutrino data and review some results in the literature, in particular the current results where solar neutrinos were used to put bounds on decoherence e ff ects through a model-dependent approach. We conclude explaining how and why these models are not general and we reinterpret these constraints.
Keywords: Dissipation, Decoherence, Neutrino Oscillation, Matter E ff ect PACS: 14.60.St
## 1. Introduction
We present a study on dissipative e ff ects on neutrino evolution, such as the decoherence and relaxation e ff ects, and their consequences in neutrino oscillations. These e ff ects are obtained when we consider neutrinos as an open quantum system [1, 2, 3]. In this approach, neutrinos are considered as a subsystem that is free to interact with the environment that presents a reservoir behavior. 1
neutrinos the decoherence e ff ect could be investigated only using a model-dependent approach, because in general, the quantum coherence is averaged out for solar neutrinos and just relaxation e ff ects can be investigated.
The decoherence e ff ect is the most usual dissipative e ff ect. In the neutrino oscillation phenomenon, the decoherence e ff ect acts only on the quantum interference, dynamically eliminating the oscillating terms in oscillation probabilities. This feature has been investigated in a number of previous studies [7, 8, 9, 10, 11, 12, 13, 14].
The relaxation e ff ect acts in a di ff erent way and it does not affect the oscillating terms. It changes only the pure mixing terms in the probabilities, leading all averaged conversion probabilities to 1 / n , where n is the number of neutrino families. Then, the relaxation e ff ect can change the probability behavior even when the oscillation terms are not important, like the solar neutrino case [3]. The relaxation e ff ect can be confused with the decoherence e ff ect and this can occur in those particular cases where quantum coherence is averaged out in neutrino oscillations. In Ref. [11], the authors analyzed quantum decoherence e ff ect with solar and KamLAND neutrinos. However, for solar
1 Some possible sources of violations of quantum mechanics fundamentals include the spontaneous evolution of pure states into mixed decoherent states [4] induced by interactions with the space-time at Planck scale [5] which unavoidly appear in any formulation of a quantum gravity theory. Such sources of decoherence was first analyzed in Ref. [6] which considered oscillating systems propagating over large distances and the corresponding damping e ff ects in the usual interferometric pattern characterizing the oscillation phenomenon.
There are some experimental bounds on dissipative e ff ects and we will compare some concrete bounds obtained from some experimental data analyses found in the literature. All these limits were obtained for neutrino propagation in vacuum and in two neutrino approximation. For example, in Ref. [14], the analysis was made considering MINOS experiment. There, the decoherence parameter has a superior limit given by γ < 9 11 . × 10 -23 GeV at 95% C.L. and this result agrees with the upper limit found in Ref. [8] where γ < 4 10 . × 10 -23 GeV at 95% C.L., which was obtained for atmospheric neutrino case.
A very interesting upper limit was introduced by Ref. [11] obtained in a model-dependent approach that constrain decoherence e ff ect using solar neutrinos. It was obtained that decoherence parameter is limited to γ < 0 64 . × 10 -24 GeV at 95% C.L. As it is known, the matter e ff ect is important in this case, and we will address this issue later on this article. In [16] an analysis using only reactor neutrinos found di ff erent bounds on the decoherence e ff ect, γ < 6 8 . × 10 -22 eV at 95% at C.L. All bounds presented above can be found in Table 1. 2
In general, bounds on dissipative parameters come from e -γ x /lessorsimilar 1 since this is the kind of damping terms which appear in the oscillation probabilities. This can be checked to
2 Following the arguments of the present article, decoherence e ff ect can be described by one parameter and relaxation e ff ect by another parameter. However, in the case of three neutrino oscillation there are three di ff erent decoherence parameters and two di ff erent relaxation parameters. As we can see in Ref. [15], the decoherence parameters describe the quantum e ff ect between specific families and then, the decoherence bound for accelerator or atmospheric neutrinos can be di ff erent from the one for reactor neutrinos.
work reasonably well for all the limits presented above, with terrestrial experiments with a typical baseline x = 10 20 ∼ 10 22 GeV 1 (20 -∼ 2000 km).
However for the numbers presented in [11], using the bound found for γ < 0 64 . × 10 -24 GeV, the exponential term tends strongly to 1. As it will be clear in this work, the modeldependent approach used in Ref. [11] also constrains the relaxation e ff ect with γ relax . < 10 -25 GeV at 95% C.L. For solarneutrinos x = 10 26 GeV 1 , and the exponential term in this -case makes the survival probability for solar neutrinos to have a unique constant value equal to 1 / 2. This result should spoil the usual solution for solar neutrinos. In our model, the constraint for γ is expected to be two order of magnitude smaller [17].
In the particular case investigated in Ref. [11], where this limit was obtained in a model-dependent approach, the exponential argument depends on other oscillation parameters, including necessarily the neutrino energy, and this makes the bound on γ just suitable in that situation.
In the model-independent approach that we will introduce in this work, the damping term will not depend on any oscillation parameters and the addition of any energy dependence on γ will be an ansatz, as those found in Refs. [8, 11, 14]. Besides, in our model the damping term for solar neutrino does not describe the decoherence e ff ect, but only the relaxation e ff ect. In fact, following the definitions that we will present in this work, the bound found in Ref. [11] can be called of decoherence just because it is proportional to the relaxation e ff ect, which is, in fact, the only dissipative e ff ect that remains after averaging out the solar-neutrino oscillations. Furthermore, our model respects the usual bound condition ( e -x γ /lessorsimilar 1) for the damping terms in the neutrino probabilities.
Our analysis will consider these two non-standard e ff ects. We analyze the propagation in vacuum and in matter. We show that with a careful application of the open quantum system theory it is possible to write the probabilities in vacuum and in constant matter in a similar way, which is not an obvious result in this context. From this result, we analyze the situation where neutrino evolution satisfies the adiabatic limit, and analyze solar neutrinos in two neutrino approximation to show that the decoherence e ff ect cannot be bounded in general using this neutrino source [19]. We discuss the current results [11, 18] where solar neutrinos were used to put limits on decoherence e ff ect through a model-dependent approach. We argue how and why these models are not general and we reinterpret these constraints.
We conclude this work arguing that the decoherence limit in the channel ν e → νµ can be di ff erent from the limit obtained in Ref. [11]. A limit for decoherence parameters can be obtained using a model-independent approach studying neutrinos from sources other than the sun.
## 2. Neutrinos as an Open Quantum System
In open quantum system approach, a global state formed by a subsystem of interest and an environment must be defined.
Table 1: Upper limits on decoherence parameters at 95% C. L. obtained from accelerator, atmospheric, reactor and solar experiments, respectively. These bounds assume that the decoherence parameters are energy independent.
| P ( ν α ν α ) | γ in GeV | baseline / E |
|------------------|-----------------------|-----------------------------------|
| P ( ν µ ν µ ) | 9 . 11 × 10 - 23 [14] | ∼ 730 km / 3 GeV |
| P ( ν µ ν µ ) | 4 . 10 × 10 - 23 [8] | /lessorsimilar 10 4 km / 10 3 GeV |
| P (¯ ν e ¯ ν e ) | 6 . 8 × 10 - 22 [16] | ∼ 200 km / 5 MeV |
| P ( ν e ν e ) | 0 . 64 × 10 - 24 [11] | ∼ 10 8 km / 2 MeV |
As the environment in this approach is a quantum reservoir, it interacts with the subsystem of interest as a whole.
The subsystem of interest can be represented by S states which are associated with the Hilbert space /C0 S , while the quantum reservoir can be represented by R states which are associated with the Hilbert space /C0 R . Basically, those are the fundamental definition about these two di ff erent quantum states. The subsystem of interest may be composed by more than one Hilbert space associated with each element that can be added in the usual quantum description of a system. For instance, when the matter potential is added to mass Hamiltonian in neutrino oscillation in vacuum.
The product tensor from these spaces form the total Hilbert space or the global states space, /C0 G = /C0 S ⊗ /C0 R . This means that we can write a global state as [20, 21]
$$\rho _ { G } = \rho _ { S } \otimes \omega _ { R } \,, \quad \quad \quad ( 1 )$$
where ρ S is the subsystem of interest state, and ω R is the reservoir state. The system evolution is obtained using the following transformation:
$$\rho _ { G } ( t ) = U ( \rho _ { S } \otimes \omega _ { R } ) U ^ { \dagger },$$
such that U = Exp [ -iHtott ] is the unitary operator and the time evolution is governed by the total Hamiltonian that can be defined as Htot = HS + HR + Hint , where HS is the subsystem of interest Hamiltonian, HR is the reservoir Hamiltonian and Hint is the interaction Hamiltonian between reservoir and subsystem of interest.
The subsystem of interest changes its characteristic in time due to the internal dynamic and the interaction with the reservoir [20, 21]. On the other hand, as that reservoir state does not change in time, its dynamics is not important. Then, the dynamic of the subsystem of interest is obtained taking the trace over the reservoir states in Eq. (2) [22, 23, 24, 25], i. e.,
$$\rho _ { S } ( 0 ) \rightarrow \rho _ { s } ( t ) = \Lambda \rho _ { s } ( 0 ) = T r _ { R } U ( \rho _ { S } \otimes \omega _ { R } ) U ^ { \dagger }, \quad ( 3 )$$
where Λ is a dynamic map. Eq. (3) is known as the reduced dynamic of S . Solving the partial trace in Eq. (3), we can rewrite this relation as
$$\Lambda \rho _ { S } ( 0 ) & = \sum _ { \alpha } W _ { \alpha } \rho _ { S } W _ { \alpha } ^ { \dagger }, \quad \quad ( 4 ) \\ \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
where W α ∈ /C0 S and ∑ α W W α † α = /BD [25]. In order to evolve the state, this map must satisfy the complete positivity constraint.
Besides, we need a family of linear maps which must satisfy the semigroup properties [22, 23, 25]. From this, we can obtain a dynamical generator, which can be written as
$$\frac { d \rho _ { \nu } ( t ) } { d t } = - i [ H _ { S }, \rho _ { \nu } ( t ) ] + D [ \rho _ { \nu } ( t ) ] \,. \quad \ \ ( 5 ) \quad \text{ way} \\. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad.$$
This equation has been studied in literature and more information about it and its properties can be found in Refs. [20, 21, 22, 23, 24, 25, 26, 27]. This equation is called Lindblad Master Equation and it is composed by an usual Hamiltonian term and a non-Hamiltonian one which gives origin to dissipative e ff ects. The dissipator in Eq. (5) can be defined as
$$D [ \rho _ { \nu } ] = \frac { 1 } { 2 } \sum _ { k = 1 } ^ { N ^ { 2 } - 1 } \left ( \left [ V _ { k }, \rho _ { \nu } V _ { k } ^ { \dagger } \right ] + \left [ V _ { k } \rho _ { \nu }, V _ { k } ^ { \dagger } \right ] \right ), \quad ( 6 ) \quad \text{$\mathbb{m}$} \quad \text{$\mathbb{m}$,} \\ \text{here } V _ { k } \text{ are dissipative operators which act only on the $N$} \quad \text{ever},$$
where Vk are dissipative operators which act only on the N -dimensional /C0 S space. The trace preservation of ρν occurs only if ∑ k V k † Vk = 1 is satisfied. The Vk operators arise from the interaction of the subsystem of interest with the environment. The propagation through equation (5) leads an initial density matrix state into a new density matrix state [1]. The evolution is complete positive, transforming pure states into mixed states due to dissipation e ff ects [20, 22, 23, 24, 25]. The Von Neumann entropy of the subsystem of interest, S = -Tr [ ρν ln ρν ], must be increasing in time and this is guaranteed if we impose V k † = Vk [28].
Let us start considering only two neutrino families and the relation between the mass and flavor bases in vacuum is given by [29, 30]
$$\rho _ { m } = U ^ { \dagger } \rho _ { f } U \,, \quad \quad \ \ ( 7 ) \quad \ \.$$
where ρ m is written in mass basis, ρ f is written in flavor basis and U is the usual 2 × 2 unitary mixing matrix.
The transformation in Eq. (7) can be used to write the Eq. (5) in the flavor basis or any other basis. Since any unitary transformation over Vk , i. e., AVkA † with AA † = 1, leads to a new matrix of the form:
$$V _ { k } ^ { \prime } = A V _ { k } A ^ { \dagger } = A \left ( \begin{array} { c c } V _ { 1 1 } & V _ { 1 2 } \\ V _ { 1 2 } ^ { * } & V _ { 2 2 } \end{array} \right ) A ^ { \dagger } = \left ( \begin{array} { c c } V _ { 1 1 } ^ { \prime } & V _ { 1 2 } ^ { \prime } \\ V _ { 1 2 } ^ { \prime * } & V _ { 2 2 } ^ { \prime } \end{array} \right ) \,, \quad ( 8 ) \quad \text{isifies} \quad \text{$i$ is $i$} \\ \text{where the new decisionor can be ranaromatic read such that it has} \quad \text{it is $gi$} \end{array}$$
where the new dissipator can be reparametrized such that it has the same form of the old dissipation operator.
Expanding Eqs. (5) and (6) in SU (2) basis matrices we can write Eq. (5) as:
$$\frac { d } { d x } \rho _ { \mu } ( x ) \sigma _ { \mu } = 2 \epsilon _ { i j k } H _ { i } \rho _ { j } ( x ) \sigma _ { \mu } \delta _ { \mu k } + D _ { \mu \nu } \rho _ { \nu } ( x ) \sigma _ { \mu } \,, \quad ( 9 ) \quad \text{subsymbol } \text{th} \\ \sim \epsilon \, \vdots \,.$$
with D µ 0 = D 0 ν = 0 to keep the probability conservation. The matrix Dmn can be parametrized as
$$\Phi _ { m n } = - \left ( \begin{array} { c c c c } \gamma _ { 1 } & \alpha & \beta \\ \alpha & \gamma _ { 2 } & \delta \\ \beta & \delta & \gamma _ { 3 } \end{array} \right ) \,, \quad & \text{enc} \\ \text{complete positivity constraints each parameter in the } \quad 1 \\ \text{form}$$
where the complete positivity constrains each parameter in the following form
$$\begin{array} { c c c c c c c } \cdot & \quad \cdot & \quad \cdot \\ 2 R & \equiv & \gamma _ { 1 } + \gamma _ { 2 } - \gamma _ { 3 } \geq 0 ; \quad & R S - \alpha ^ { 2 } \geq 0 ; \\ 2 S & \equiv & \gamma _ { 1 } + \gamma _ { 3 } - \gamma _ { 2 } \geq 0 ; \quad & R T - \beta ^ { 2 } \geq 0 ; \\ 2 T & \equiv & \gamma _ { 2 } + \gamma _ { 3 } - \gamma _ { 1 } \geq 0 ; \quad & S T - \delta ^ { 2 } \geq 0 ; \end{array}$$
$$2 T \ \equiv \ \gamma _ { 2 } + \gamma _ { 3 } - \gamma _ { 1 } \geq 0 ; \quad S T - \delta ^ { 2 } \geq 0 \, ;$$
$$R S T \geq 2 \alpha \beta \delta + T \delta ^ { 2 } + S \beta ^ { 2 } + R \alpha ^ { 2 } \,. \quad \quad ( 1 1 ) \\. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad.$$
When we take out the reservoir Hamiltonian, HR , and the interaction Hamiltonian, Hint , the quantum evolution return to usual way and then the Eq. (5), which is just the known Liouville quantum equation.
## 2.1. The Subsystem of Interest
Our subsystem of interest will be the neutrinos. As it is well known, many experiments give evidence that neutrinos have mass and mixing, as defined in Eq. (7), such that flavors oscillation can occur [29, 30].
Neutrinos propagate in vacuum or in matter. In both situations it is possible to evolve neutrinos as an open quantum system, through direct application of the Eqs. (5) and (6). However, it is important to take into account in which circumstances these equations were developed and how the subsystem of interest was defined. Then, the definition of neutrinos like a subsystem of interest can change in each case.
We can use a previous knowledge of the Hamiltonian in standard quantum mechanics to define this general subsystem of interest S. As we have seen, the total Hamiltonian in open quantum system approach can be defined as Htot = HS + HR + Hint . In this case, HS is the usual Hamiltonian in closed approach. Then, the more general subsystem of interest is the physical object described by basis in which HS is diagonal.
## 2.2. Quantum Dissipator and the E ff ects in S
It is possible to study how each entry in the matrix in Eq. (10) changes the neutrino probabilities [3]. For simplicity we will work with only two models for quantum dissipator. One with only one new parameter that will describes decoherence e ff ect and another with two new di ff erent parameters that will describe decoherence and relaxation e ff ects.
The most usual dissipator is obtained imposing energy conservation on the subsystem of interest S . This constraint satisfies the following commutation relation: [ HS , Vk ] = 0. This dissipator adds only decoherence to the system of interest S and it is given by
$$D _ { m n } = - d i a g \{ \gamma _ { 1 }, \gamma _ { 1 }, 0 \} \quad \quad \ \ ( 1 2 ) \\ \cdots \cdots \cdots \cdots \cdots$$
where, in this case γ 1 = γ 2 and all other parameters vanish. This statement defines uniquely a particular interaction between the subsystem of interest S and the reservoir.
Therefore, the energy conservation constraint in subsystem of interest S is obtained only if the commutation relation [ HS , Vk ] = 0 is satisfied and the consequence is a quantum dissipator with only one parameter, γ 1 , that describes decoherence e ff ect. In other words, the dynamic evolution is purely decoherent when this specific constraint is applied and no other dissipative e ff ect is present.
To include the relaxation e ff ect we need to violate the above constraint. As the subsystem of interest is free to interact with the reservoir the energy flux can fluctuate and the energy conservation condition imposed over the subsystem of interest can be not satisfied. In this case, the matrix in Eq. (10) can assume its complete form. However, as the matrix in Eq. (10) needs to
be positive, all o ff -diagonal parameters must be smaller than the diagonal parameters. Then, only the diagonal parameter necessarily must be present in case of new physics. For simplicity, we will disregard all o ff -diagonal elements.
By assuming [ HS , Vk ] /nequal 0, a non null D 33 parameter can be included in the dissipator in Eq. (12) and then a new quantum dissipator can be written as
$$D _ { m n } = - d i a g \{ \gamma _ { 1 }, \gamma _ { 1 }, \gamma _ { 3 } \} \,, \quad \quad ( 1 3 ) \quad \text{the} \\. \quad. \. \. \. \. \. \. \quad \end{cases} \quad \text{$\quad$the}$$
where γ 1 continues describing the decoherence e ff ect and γ 3 describes the relaxation e ff ect.
## 2.3. Dissipation in other Specific Subsystem of Interest S'
The quantum dissipator written in Eq. (6) can be defined in many di ff erent ways for neutrinos propagating in vacuum or in constant matter density, but it can have the same form in both cases. It is easy to prove this statement since we can always write HS in Eq. (2) as being diagonal in vacuum or in matter propagation. However, the parameter values in operator Vk are di ff erent in each case.
In the presence of matter the transformation between the effective mass basis and flavor basis can be written changing ρ m → ˜ ρ m and U → ˜ , U where ˜ U is composed by e ff ective mixing angles [30, 29]. This transformation may not bring anything new to the quantum evolution equation in (5) and it can be again parametrized as we made in Eq. (9) with a ˜ Dmn that has the same form of the Dmn that was given by Eq. (10).
In the usual situation in matter propagation we can define HS = Hosc + Hmat and then, the interaction constraints between a specific subsystem of interest S ′ and the reservoir can be imposed in di ff erent ways. Thus, it is possible to define a specific subsystem of interest S ′ that can have a commutation relation with a particular Vk . While, the HS defines the more general subsystem of interest S, the Hosc or Hmat could be used to define other specific subsystems of interest S'.
If we assume, for instance, that [ Hosc , Vk ] = 0, the energy conservation is kept when the propagation is in vacuum and only decoherence can act during the propagation. However, to the same case, when the propagation is in the matter HS /nequal Hosc and therefore this constraint no longer preserve the energy conservation in the subsystem of interest S and we have the situation where [ HS , Vk ] /nequal 0. Thus, the relaxation and decoherence e ff ects may act during the propagation.
Therefore, when one defines HS and its relation with the Vk operators, all the dissipative e ff ects are determined. So, a consequence of the definition of the subsystem of interest S from HS can be summarized as follow: if the subsystem of interest S has its energy conserved then [ HS , Vk ] = 0 and the dissipator has the form of Eq. (12). In this case we are dealing with decoherence e ff ects. If it is not, then [ HS , Vk ] /nequal 0 and the dissipator can be written in its more general form, Eq. (13). Thus, there are both decoherence and relaxation e ff ects taking place during neutrino evolution.
The di ff erence between decoherence and relaxation e ff ect was discussed in this section. Now, we will apply this formalism in neutrino oscillation in vacuum and in constant matter case in order to eliminate any confusion between these two dissipative e ff ects.
## 3. Propagation in Vacuum and in Constant Matter Density
With the Lindblad Master Equation we can study many dissipative e ff ects in neutrino oscillations. Decoherence is the most usual dissipative e ff ect [11, 7, 10, 9, 8, 12, 13, 31], but it is not the only one, as we have seen in previous section. In particular, we are going to study how decoherence and relaxation e ff ects act on the state during its propagation and how these dissipative e ff ects change the oscillation probabilities.
In general, we can calculate the evolution using the dissipator in Eq. (13). We can obtain the evolution using the dissipator given in Eq. (12) just setting γ 3 = 0. The oscillation Hamiltonian in vacuum and in matter is taken in its diagonal form. Usually in vacuum HS is written in the mass basis as HS = diag E { 1 , E 2 } and when the oscillation occurs in constant matter, it is possible to write the Hamiltonian as HS = diag E { ˜ 1 , ˜ E 2 } using the e ff ective mass basis. Note that we have defined two di ff erent subsystems of interest S , one for neutrinos in vacuum and another for neutrinos in constant matter, but both HS are diagonal.
We are going to use the approximation Ei = E + mi / 2 E and ˜ Ei = E + ˜ mi / 2 E . The Eq. (9) can be written as
$$\left ( \begin{array} { c } \dot { \rho } _ { 1 } ( x ) \\ \dot { \rho } _ { 2 } ( x ) \\ \dot { \rho } _ { 3 } ( x ) \end{array} \right ) = \left ( \begin{array} { c c c } - \gamma _ { 1 } & - \Delta & 0 \\ \Delta & - \gamma _ { 1 } & 0 \\ 0 & 0 & - \gamma _ { 3 } \end{array} \right ) \left ( \begin{array} { c } \rho _ { 1 } ( x ) \\ \rho _ { 2 } ( x ) \\ \rho _ { 3 } ( x ) \end{array} \right ) \,, \quad ( 1 4 ) \\ \intertext { where } \Delta \, = \, \Delta m ^ { 2 } / 2 E. \, \text{ If the propagation is in matter, we can} \, \text{ we can} \, \text{
} \, \text{ } \, \text{\vec{ } } \, \text{ } \, \text{ } \, \, } \end{array}$$
where ∆ = ∆ m 2 / 2 E . If the propagation is in matter, we can evoke the e ff ective quantities, which are ∆ → ˜ ∆ = ∆ ˜ m 2 / 2 E , γ i → ˜ γ i by following the Eq. (8) and ρ i → ˜ . ρ i Of course, this changes nothing from the point of view of the equation solution and from now on, we do not mention more this similarity. Further, the component ρ 0 has a trivial di ff erential equation given by ˙0( ρ x ) = 0 and its solution is ρ 0 ( x ) = ρ 0 (0) that in two neutrino oscillation means ρ 0 ( x ) = 1 2. / The Eq. (14) can be written in short form as
$$\dot { R } ( t ) = \mathbb { H } R ( t ) \,, \quad \quad \quad ( 1 5 )$$
$$R ( x ) = \mathbb { M } ( x ) R ( 0 ) \,, \quad \quad \quad ( 1 6 )$$
/C0 where the eigenvalues of /C0 are λ 0 = -γ 3 , λ 1 = -γ 1 -i ∆ and λ 2 = -γ 1 + ∆ i . For each eigenvalue it is possible to obtain a correspondent eigenvector, u 0 , u 1 , u 2 that compose the matrix /BT = [ u 0 , u 1 , u 2 ] that diagonalizes the matrix /C0 by performing the following similarity transformation: /BT † /C0/BT . The solution of the Eq. (15) is given by
/C5 where ( x ) is obtained making
$$\mathbb { M } ( x ) = \mathbb { A }. d i a g \{ e ^ { \lambda _ { 0 } x }, e ^ { \lambda _ { 1 } x }, e ^ { \lambda _ { 2 } x } \}. \mathbb { A } ^ { \dagger } \,. \quad \quad ( 1 7 )$$
/C5
/C5 /BT /BT Furthermore, it is useful to write the propagated state which in this case is given by
$$\rho ( x ) = \left ( \begin{array} { c c } \rho _ { 0 } ( x ) + \rho _ { 3 } ( x ) & \rho _ { 1 } ( x ) - i \rho _ { 2 } ( x ) \\ \rho _ { 1 } ( x ) + i \rho _ { 2 } ( x ) & \rho _ { 0 } ( x ) - \rho _ { 3 } ( x ) \end{array} \right ) \,. \quad ( 1 8 )$$
From the Eq. (14), one can see that the propagated state is written as
$$\rho ( x ) & = \left ( \begin{array} { c c } \frac { 1 } { 2 } + \frac { 1 } { 2 } e ^ { - \gamma _ { 3 } x } \cos 2 \theta & \frac { 1 } { 2 } e ^ { - ( \gamma _ { 1 } - i \Delta ) x } \sin 2 \theta \\ \frac { 1 } { 2 } e ^ { - ( \gamma _ { 1 } + i \Delta ) x } \sin 2 \theta & \frac { 1 } { 2 } - \frac { 1 } { 2 } e ^ { - \gamma _ { 3 } x } \cos 2 \theta \end{array} \right ) \,, \quad ( 1 9 ) \\ \intertext { m o n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k j i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d ik i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i l k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d j i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d j k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d I k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d e n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d i k i n d ier { \Phi } ( 1 9 )$$
where it is possible to identify two unusual behaviors. The o ff -diagonal entries are called coherence elements and it has a damping term that eliminates the quantum coherence during the propagation. This is the exact definition for decoherence e ff ect and we can see clearly that such e ff ect is associated with the matrix elements γ 1 . The diagonal elements in Eq. (19) are known as population elements and they are related to the quantum probabilities of obtaining the eigenvalue E 1 or E 2 of the observable HS .
In the absence of dissipative e ff ects, the observable is diagonal in the mass basis and the diagonal elements of the state are independent of the distance, but in the state in Eq. (19) the probability elements change with the propagation. This dissipative e ff ect implies that the neutrinos may change their flavor without using the oscillation mechanism. As the asymptotic state is a complete mixing, the γ 3 in diagonal elements is called relaxation e ff ect.
The flavor oscillation probabilities can be obtained from the Eq. (7) and ρ f 11 element is the survival probability that is written as
$$& P _ { \nu _ { \alpha } \to \nu _ { \alpha } } = \frac { 1 } { 2 } \left [ 1 + e ^ { - \gamma _ { 3 } x } \cos ^ { 2 } 2 \theta + e ^ { - \gamma _ { 1 } x } \sin ^ { 2 } 2 \theta \cos \left ( \Delta x \right ) \right ]. \ ( 2 0 ) \quad 4. \ \text{Ne} \\ & \text{In Eq. } ( 2 0 ), \text{ the asymptotic probability, } x \, \to \, \infty, \, \text{ goes to} \quad \text{In n}$$
In Eq. (20), the asymptotic probability, x → ∞ , goes to a maximal statistical mixing, P να → να = 1 2, and this happens / for any mixing angle. Thus, by means of this approach, the neutrino may change its flavor and it does not need to use the oscillatory mechanism to this end [1, 3]. In fact, while the decoherence e ff ect, through γ 1 parameter, eliminates the oscillation term , the relaxation e ff ect, γ 3 parameter, eliminates the term in the probability that depends only on the mixings.
When the propagation is performed with the dissipator given in Eq. (12), we obtain some important di ff erences. In this case, /C0 has only two non-trivial eigenvalues which are equal to λ 1 and λ 2 which were derivated before. Then, the matrix /C5 ( x ) is changed to
/C5 /BT and consequently, the state is written as
/BT
$$\mathbb { M } ( x ) = \mathbb { A }. d i a g \{ 1, e ^ { \lambda _ { 1 } x }, e ^ { \lambda _ { 2 } x } \}. \mathbb { A } ^ { \dagger } \,, \quad \ \ ( 2 1 ) \quad \text{$\text{$du}$} \quad \text{$\text{$dep}$}$$
$$\rho ( x ) = \left ( \begin{array} { c c } \frac { 1 } { 2 } + \frac { 1 } { 2 } \cos ^ { 2 } \theta & \frac { 1 } { 2 } e ^ { - ( \gamma _ { 1 } - i \Delta ) x } \sin 2 \theta \\ \frac { 1 } { 2 } e ^ { - ( \gamma _ { 1 } + i \Delta ) x } \sin 2 \theta & \frac { 1 } { 2 } - \frac { 1 } { 2 } \cos ^ { 2 } \theta \end{array} \right ) \,. \quad ( 2 2 ) \quad \text{and it} \\ \Gamma ( x ) \, \text{the state about the same is an in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \,\text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{inin} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \end{array}$$
In the state above, there is only influence of the decoherence e ff ect and only the coherent elements are eliminated during the propagation. In this case, the survival oscillation probability is written as
$$P _ { \nu _ { \alpha } \to \nu _ { \alpha } } = 1 - \frac { 1 } { 2 } \sin ^ { 2 } ( 2 \theta ) \left [ 1 - e ^ { - \gamma _ { 1 } x } \cos ( \Delta x ) \right ]. \quad ( 2 3 ) \quad \text{tive} \quad \\ \text{This probability was discussed in Refs. } [ 1, 3, 8 ] \text{ only in the} \quad \Delta \to.$$
This probability was discussed in Refs. [1, 3, 8] only in the vacuum approach, but we are showing that when the open quantum system approach is applied carefully a similar probability is obtained for the propagation in matter as well.
Then, when there is energy conservation in subsystem of interest, [ HS , Vk ] = 0, the asymptotic probability, x → ∞ , still depends on the mixing angle as
$$P _ { \nu _ { \alpha } \to \nu _ { \alpha } } = 1 - \frac { 1 } { 2 } \sin ^ { 2 } ( 2 \theta ) \,.$$
In this approach, the dynamics is made through Eq. (2) and it depends on how the subsystem of interest interacts with the environment following constraint: [ HS , Vk ] = 0 or [ HS , Vk ] /nequal 0.
From a mathematical point of view, when we consider neutrinos like an open quantum system and taking into account the considerations explored in this section, one can see that there are not significant di ff erences in deriving the quantum evolution in vacuum or in constant matter. This result is trivial in closed approach, but it is not a trivial result in this open approach. In fact, the similarity between these two propagation conditions in open approach is only true when the reservoir interacts in some way with the subsystem of interest represented by S that here, it was defined using mass state in vacuum propagation or effective mass state in matter propagation. Otherwise, there will not be similarities between the vacuum and matter propagation [2, 11].
## 4. Neutrinos in Non-Uniform Matter
In many situations the neutrino propagation occurs where the matter density is not constant. We are going to assume neutrino evolution in non-constant matter only in situations where the adiabatic limit is valid [30, 29]. Thus, the results obtained in this situation are similar to those obtained for propagation in constant matter. The main focus now is to understand which dissipative e ff ects act on neutrinos supposing that the source is far away from the Earth. Solar neutrinos are a great example that we want to study.
Using the same point of view from the previous section, we can write a diagonal Hamiltonian using the e ff ective mass basis. We start with the quantum dissipator written in Eq. (13). Thus, we have to solve the same evolution equation given in Eq. (14), but on the right side, the elements of the first matrix are distance dependent as well. So, the Eq. (15) is written now as
$$\dot { R } ( x ) = \mathbb { H } ( x ) R ( x ) \,,$$
$$\mathbb { M } ( x ) \in d i a g \{ e ^ { \int _ { r } ^ { R _ { 0 } } \lambda _ { 0 } ( x ) d x }, e ^ { \int _ { r } ^ { R _ { 0 } } \lambda _ { 1 } ( x ) d x }, e ^ { \int _ { r } ^ { R _ { 0 } } \lambda _ { 2 } ( x ) d x } \} \,, \quad ( 2 6 )$$
/C0 and it has a solution similar to Eq. (16), but /C5 ( x ) is proportional to
/C5 where r and R /circledot are the creation and detection point, respectively. As /BT is defined in the same way of the previous section, the λ i ( x ) has the same form of λ i defined in Eq. (15), but here ∆ → ˜ ( ∆ x ) and γ i → ˜ γ i may depend on distance. Even for λ 0 the distance dependence may exist [11].
Notice that the energy conservation is given by [ HS , Vk ] = 0, but in general the HS in vacuum propagation is di ff erent from
HS in matter propagation. Consequently, when one imposes energy conservation in matter propagation there is not energy conservation in vacuum propagation and vice-versa. On the other hand, it is possible to obtain a model where the energy conservation is always kept even when HS in vacuum and in matter propagation are di ff erent. In this case, the dissipative quantum operator has a distance dependence such that Vk changes to Vk ( x ) and can be written as
$$V _ { k } = V _ { k } ( x ) = \left ( \begin{array} { c c c c } 2 \sqrt { \gamma _ { 1 } } \cos [ \Theta ( x ) ] & \sqrt { \gamma _ { 1 } } \sin [ \Theta ( x ) ] \\ \sqrt { \gamma _ { 1 } } \sin [ \Theta ( x ) ] & 0 \end{array} \right ) \,, \quad ( 2 7 ) \quad \text{To } \, \text{(the eff} \\ \quad \text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text \text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\tt$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$
$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$\text$ \end{array}$$
where Θ ( x ) = 2( θ -˜ ( θ x )) and the e ff ective angle is given by
$$\tilde { \theta } ( x ) = \frac { 1 } { 2 } \arcsin \left ( \sqrt { \frac { \Delta ^ { 2 } \sin ^ { 2 } [ 2 \theta ] } { ( \Delta \cos [ 2 \theta ] - A ( x ) ) ^ { 2 } + \Delta ^ { 2 } \sin ^ { 2 } [ 2 \theta ] } } \right ) \,. \, \ ( 2 8 ) \quad \text{writer} \\ \text{The off-diagonal elements in vacuum case are null and the} \\ \text{element } \left ( V ( v ) \right ) \,. \, = \, 2 \cdot \sqrt { v } \text{, such that the quantum dissinator in}$$
The o ff -diagonal elements in vacuum case are null and the element { Vk ( x ) } 11 = 2 √ γ 1 such that the quantum dissipator in Eq. (12) is not changed. Supposing the adiabatic limit or constant density matter, we can rewrite the evolution in mass basis into e ff ective mass basis where one considers the addition of the potential matter. In this case, the dissipation operator Vk ( x ) in vacuum changes to ˜ Vk ( x ) = ˜ U UVkU U † † ˜ in matter propagation, such that it is written as
$$\tilde { V } _ { k } ( x ) = \left ( \begin{array} { c c c c } 2 \sqrt { \gamma _ { 1 } } \cos ^ { 2 } [ \Theta ( x ) ] & 0 \\ 0 & - 2 \sqrt { \gamma _ { 1 } } \sin ^ { 2 } [ \Theta ( x ) ] \end{array} \right ) \,, \quad ( 2 9 ) \quad \text{$\text{$in t}$} \\ \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \,.$$
and the dissipator in Eq. (12) continues unchanged as well.
Thus, disregarding models where the operator in Eq. (27) di ff ers by a unitary matrix, this is a unique model where energy conservation constraint in matter propagation and in vacuum propagation are satisfied simultaneously. This occurs due to the fact that energy conservation in matter propagation is given by [ ˜ HS ( x ) , ˜ Vk ( x )] = 0, and this result is valid for any choice of matter potential.
So, as we can mentioned before, if we want that the evolution is purely decoherent, i. e., that the energy conservation, [ ˜ HS ( x ) , ˜ Vk ( x )] = 0, is satisfied during the propagation even when the density matter varies, we must have a dissipation operator like the one in Eq. (27), because it takes into account how much the matter e ff ect could change it.
Returning to the evolution given by Eq. (25), the state evolved using Eq. (25) is written as
$$\tilde { \rho } _ { m } ( x ) = \left ( \begin{array} { c c } \frac { 1 } { 2 } + \frac { 1 } { 2 } e ^ { - \Gamma } \cos 2 \tilde { \theta } & \frac { 1 } { 2 } e ^ { - \Gamma _ { 1 } } \sin 2 \tilde { \theta } \\ \frac { 1 } { 2 } e ^ { - \Gamma _ { 1 } ^ { * } } \sin 2 \tilde { \theta } & \frac { 1 } { 2 } - \frac { 1 } { 2 } e ^ { - \Gamma } \cos 2 \tilde { \theta } \end{array} \right ) \,, \quad ( 3 0 ) \quad \text{$of the} \\. \quad. \quad. \quad. \quad \text{$\quad$ sum}$$
where we have defined
$$\Gamma = - \int _ { r } ^ { R _ { \odot } } \tilde { \gamma } _ { 3 } ( x ) d x \quad \ \ \ ( 3 1 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
and where ˜1( γ x ) = γ 1 if we consider the dissipation operator in Eq. (27).
$$\Gamma _ { 1 } = - \int _ { r } ^ { R _ { \odot } } \tilde { \gamma } _ { 1 } ( x ) d x + i \int _ { r } ^ { R _ { \odot } } \tilde { \Delta } ( x ) d x, \quad ( 3 2 ) \quad \text{can} \\ \text{re} \, \tilde { \nu } _ { 1 } ( x ) = \nu _ { 1 } \text{ if we consider the dissination operator in Fa}. \quad \text{cohe}$$
In general, the second term in Eq. (32) gives rise to fast oscillation terms in the o ff -diagonal elements and it is usually averaged out. Thus, the state has the following form
$$\begin{array} { c }. \\. \\. \\. \end{array} \tilde { \rho } _ { m } ( x ) = \left ( \begin{array} { c c } \frac { 1 } { 2 } + \frac { 1 } { 2 } e ^ { - \Gamma } \cos 2 \tilde { \theta } & 0 \\ 0 & \frac { 1 } { 2 } - \frac { 1 } { 2 } e ^ { - \Gamma } \cos 2 \tilde { \theta } \end{array} \right ) \,, \quad ( 3 3 )$$
where, we conclude that in general we cannot have information about the decoherence e ff ect in this situation.
To obtain the usual adiabatic probability we use the fact that the e ff ective mixing angle changes during the neutrino propagation and then, the mixing angle in the detection point must be di ff erent. We define the initial mass state from the Eq. (7), where in the creation point, we used the e ff ective mixing angles written as ˜ . θ Then, we can change the representation by applying another mixing matrix with another mixing angle. Defining these angles in detection point as ˜ θ d , we have
$$\rho _ { f } ( x ) = U _ { d } \tilde { \rho } _ { m } ( x ) U _ { d } ^ { \dagger } \,,$$
where Ud is the usual mixing matrix, but with mixing angle ˜ θ d . Then, the adiabatic survival probability, ρ f 11 ( x ), is given by
$$P _ { \nu _ { e } \to \nu e } ^ { a d i a b. } = \frac { 1 } { 2 } + \frac { 1 } { 2 } e ^ { - \Gamma } \cos 2 \tilde { \theta } \cos 2 \tilde { \theta } _ { d } \,. \quad \quad ( 3 5 )$$
In the survival probability above, if Γ = 0, we recover the usual survival probability in the adiabatic limit case [30, 29]. The dissipation operator in Eq. (27) is obtained when the energy constraint, [ HS , Vk ] = 0 is imposed and hence only decoherence e ff ect might be described by ˜ γ 1 using the operator in Eq. (12). However, the state more general for solar neutrinos does not hold the ˜1 γ in its description and then, we can conclude that quantum decoherence cannot be limited by solar neutrinos in general. On the other hand, as only ˜3 γ remains in the state (33) and in the probability (35), in general, just the relaxation e ff ect can be limited when one considers solar neutrinos
Now we analyze a situation mentioned in the subsection 2.3 that is, for example, the same supposition that the authors in Ref. [11] used to put limit on decoherence e ff ect using solar neutrinos.
So, we assume neutrinos propagate in matter in the situation where the adiabatic limit is satisfied. As usual, the Hamiltonian is HS = Hosc + Hmat , where Hosc is the oscillation Hamiltonian in vacuum and Hmat is the matter potential. In addition, we assume energy conservation with two di ff erent conditions. One of them is when we suppose energy conservation only with the vacuum piece, [ Hosc , ¯ ] Vk = 0, and another one is when we assume energy conservation only with the matter potential piece, [ Hmat , V k ′ ] = 0. Note that the ¯ Vk and V k ′ follow the definition given by in Eq. (8) and both of them are di ff erent of Vk that may commutate with HS .
These two situations can try to investigate only the decoherence e ff ect. One of them the neutrino state in vacuum can be changed due to the decoherence e ff ect even it is present in the Sun, for instance. With another one, it is possible to study decoherence e ff ect in the Sun environment in order to change the matter e ff ect through a dissipative phenomenon.
As the energy conservation constraint in subsystem of interest was assumed whatever the place that neutrino will go through, for both situations the quantum dissipator used in the propagation in Eq. (5) is given by Eq. (12). However, as we have mentioned, this quantum dissipation includes only quantum decoherence e ff ect in the propagation. So, for the quantum evolution in both situations, the Eq. (25) with /C0 is now given by
$$\mathbb { H } = \left ( \begin{array} { c c c } - \gamma _ { 1 } & - \Delta - A \cos 2 \theta & 0 \\ \Delta + A \cos 2 \theta & - \gamma _ { 1 } & - A \sin 2 \theta \\ 0 & A \sin 2 \theta & 0 \end{array} \right ), \quad ( 3 6 ) \quad \text{which} \\ \text{under} \quad \text{matter} \\ \text{tion is} \\ \text{where} \, \gamma _ { 1 } \, \text{comes from} \, D _ { m n } \, \text{in} \, Eq. \, (10) \, \text{for both cases and in the} \\ \text{amount above} \, \mathbb { H } \, \text{use written in} \, \text{mace basic transaction}$$
/C0 The characteristic polynomial of the above matrix has a complicated solution, but if we consider γ 1 is small such that it can be treated like a perturbation, we obtain in first order approximation the following eigenvalues:
where γ 1 comes from Dmn in Eq. (10) for both cases and in the equation above, was written in mass basis representation.
$$\max \min \min \log \text{gauvalues} \quad \text{ obtain} \quad \text{ obtain} \quad \text{ $\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad\text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$,} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$ \mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf u$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \ \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf w$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{ \mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \ \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbfv$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{#\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \ :: \text{#\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf u$} \ :: \text{#\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{$\mathbf v$} \quad \text{#\mathbf v$} \ :: \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf u$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{$\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \ \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$} \quad \text{#\mathbf v$$$
where A = √ 2 GFne and, for sake of simplicity, we can rewrite in the e ff ective mass basis, such that we get
$$\mathbb { H } = \left ( \begin{array} { c c c } - \tilde { \gamma } _ { 1 } & - \tilde { \Delta } & 0 \\ \tilde { \Delta } & - \tilde { \gamma } _ { 1 } & 0 \\ 0 & 0 & - \tilde { \gamma } _ { 3 } \end{array} \right ), \quad \text{$neu$} \\ \mathbb { H } _ { S } \\ \mathbb { H } \, \text{$her$} \\ \mathbb { H } \, \text{$\gamma_{1}A^{2}$} \sin^{2} 2 \tilde { \theta} / \Delta^{2}$ and $\tilde{\gamma}_{1}$} \, = \, \gamma_{1} - \tilde { \gamma } _ { 3 }. \, \text{From $\mathbb{H}$ given} \\ \mathbb { W } \, \text{we obtain the same state that was given in $E_{1}$} \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb{ A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A } \, \mathbb { A.}$$
/C0
with ˜ γ 3 = γ 1 A 2 sin 2 2 ˜ θ/ ∆ 2 and ˜1 γ = γ 1 -˜ γ 3 . From /C0 given by Eq. (38) we obtain the same state that was given in Eq. (30) where Γ 1 would be defined by ˜1 γ while Γ by ˜3. With the same γ arguments that was given before, Γ 1 becomes null and we obtain the state in Eq. (33). The interpretation is similar that was done before where Γ 1 is not important and only the relaxation e ff ect, Γ ∝ ˜ γ 3 , may change the probability.
In these two situations the constraints are [ HS , ¯ ] Vk /nequal 0 and [ HS , V k ′ ] /nequal 0. Thus, we could expect that the result for these di ff erent constraints, [ Hosc , ¯ ] Vk = 0 and [ Hmat , V k ′ ] = 0, are obtained by an evolution using the dissipator in Eq. (13), as we have seen in subsection 2.3. Besides, this result show that there is not a way to separate the subsystem of interest S in pieces which may or may not interact with the environment and here, as we have [ HS , ¯ ] Vk /nequal 0 and [ HS , V k ′ ] /nequal 0 the relaxation e ff ect appears naturally.
The decoherence and relaxation e ff ects when the propagation in matter may have di ff erent magnitude from the vacuum case. However, independently of we assume [ Hosc , ¯ ] Vk = 0 or [ Hmat , V k ′ ] = 0, we have the same result for the dissipative effects. This looks like an apparent problem because we cannot di ff erentiate between these dissipative models in the solar neutrino case, for example.
In special, the case where [ Hosc , ¯ ] Vk = 0 the Eq. (38) shows relaxation e ff ect is proportional to the decoherence e ff ect for neutrinos propagating in vacuum (the same occurs for the case [ Hmat , V k ′ ] = 0 [18]). This was the result obtained by Ref. [11] and thus, from this model-dependent approach, the decoherence e ff ect in vacuum, γ 1 , was limited by authors in Ref. [11]. Besides, the ¯ wrote there in our notation is written as Vk
$$\bar { V } _ { k } = \left ( \begin{array} { c c } 2 \sqrt { \gamma _ { 1 } } & 0 \\ 0 & 0 \end{array} \right ). \quad \quad ( 3 9 ) \\ \text{want from the 1.7.v. aivan in $E_{a}$ $27\uplus$share}$$
which is di ff erent from the Vk ( x ) given in Eq. (27), where the matter potential becomes important and the energy conservation is always satisfied even when the propagation is through in non-constant matter.
In the Ref. [18] the authors made a microscopic model to the interaction between neutrinos and the solar environment and they reached a dynamic equation similar to Eq. (37), but there the dissipation e ff ect appears as a consequence of this microscopic model where [ Hmat , V k ′ ] = 0 was satisfied. The dynamic obtained in Ref. [11] was also obtained by authors in Ref. [18] even the study propose being di ff erent one another, of course, they reached to same probability as well.
Therefore, the result of the last example is interesting because it has not trivial interpretation. And there is not in the literature a reliable limit for decoherence e ff ect in the channel ν e → νµ obtained from a model-independent approach. Surely, it exists only limits on the relaxation and decoherence e ff ects in the case of a particular model-dependent approach used by Ref. [11] in two neutrino approximation. So, other analysis using a general model-independent approach can be done using neutrinos that come from other sources, where the constraint [ HS , Vk ] = 0 can without any doubt be satisfied and the decoherence e ff ect be limited.
## 5. Comments and Conclusion
The quantum dissipator in Eq. (12) is related to decoherence e ff ects while the quantum dissipator in Eq. (13) is related to decoherence plus relaxation e ff ects. We explicitly relate decoherence e ff ects with a quantum dissipator that conserves energy in the subsystem of interest, a condition that is fulfilled if [ HS , Vk ] = 0. If such condition is violated, then we relate such quantum dissipator with relaxation e ff ects. So, we introduce the unique form in which this condition is satisfied in all points of the evolution since HS is the Hamiltonian that governs the evolution in the usual approach. This means that HS is composed by mass and interaction Hamiltonians in matter propagation and only mass Hamiltonian in the case of the vacuum propagation.
We emphasized the di ff erences and similarities between the /C0 eigenvalues that are obtained when we used the dissipators in Eqs. (12) and (13). We clearly see when the relaxation effect is present in the model and how the behavior of the states is changed in the situation with and without the relaxation effects. We discussed the neutrino evolution in vacuum and in matter with constant density and pointed out how these situations can have similar treatments in open quantum system formalism. We showed that in general the probabilities in vacuum
and in constant matter can be written in similar ways, which is not an obvious result in this approach. It is interesting to note that through the model developed in this article, we do not need to use any method of approximation to obtain the probabilities in all cases. This is di ff erent from what we can find in the literature [2, 11, 18].
We analyzed also the situation where the matter density is not constant. We obtained a dissipation operator in Eq. (27) that conserves energy during the neutrino propagation through a variable matter density. We showed that the decoherence effect from our model-independent analysis cannot be limited in situations where experiments can no longer access the oscillation term in the probabilities, as it is the case when the source is very far away from the detection point. On the other hand, the relaxation e ff ect may still be tested and limited in such situations. Although, as it was made in Ref [11] through a modeldependent approach, it is possible to limit the decoherence in this case because the decoherence e ff ect is connected in some way with the relaxation e ff ect. However, as we have pointed out, the relaxation and decoherence e ff ects are di ff erent phenomena and both bring di ff erent behavior to the neutrinos.
We identified some ambiguities in the definition of decoherence e ff ects present in the literature [11], where there is no clear distinction between decoherence and relaxation e ff ects. In our understanding, the term decoherence is often used to describe a combined e ff ect of decoherence and relaxation when neutrino evolves in a medium with variable density. We described how it would be a dissipative model with only quantum decoherence e ff ects for propagation in matter with non-constant density. From the dissipative operator obtained in Eq. (27), it was possible to see why the decoherence e ff ect was limited in Refs. [11] and mentioned in Ref. [18]. In fact, in those cases it could not exist decoherence e ff ect only, but another e ff ect related in some way with the decoherence e ff ect, because the dissipative operator used by these references, Eq. (39), violates the condition [ HS , Vk ] = 0, when neutrinos propagate in constant or non-constant matter.
Comparing our approach with the ones found in the literature, it is possible to conclude that it avoids all the ambiguities about which kind of dissipative e ff ect is acting on neutrinos. As stated before, the limit for the decoherence e ff ect should be obtained through experiments that access the oscillation pattern in the flavor neutrino probabilities, like KamLAND [16], for instance. The result of the Ref. [11] can be interpreted as an upper limit on the decoherence e ff ect which comes, in fact, from the restriction on the relaxation e ff ect, once that both e ff ects are connected in this model dependent analysis. Our model independent approach is able to put bounds on all dissipation e ff ects in a direct way.
## 6. Acknowledgments
We would like to thank CNPq and FAPESP for several financial supports. R.L.N.O. is grateful to L. Ostrar. and J.A.B. Coelho for instructive discussions and thanks for the support of funding grants 2012 00857-6 and 2013 11651-2 S£o Paulo / / Research Foundation (FAPESP).
## References
- [1] F. Benatti, R. Floreanini, Open system approach to neutrino oscillations, JHEP 02 (2000) 32.
- [2] F. Benatti, R. Floreanini, Massless neutrinos oscillation, Phys. Rev. D 64 (2001) 085015.
- [3] R. L. N. Oliveira, M. M. Guzzo, Quantum dissipation in vacuum neutrino oscillation, Eur. Phys. Jour. C 69 (2010) 493.
- [4] S.W. Hawking, Commun. Math. Phys. 43, 199 (1975); ibidem 87, 395 (1982); Phys. Rev. D 14, 2460 (1976).
- [5] S.B. Giddings and A. Strominger, Nucl. Phys. B 307, 854 (1988); W.H. Zurek, Physics Today 44, No. 10, p. 36 (1991); G. Amelino-Camelia, J. Ellis, N.E. Mavromatos, and D.V. Nanopoulos, Int. J. Mod. Phys. A 12, 607 (1997); L.J. Garay, Int. J. Mod. Phys. A 14, 4079 (1999).
- [6] J. Ellis, J.S. Hagelin, D.V . Nanopoulos, and M. Srednicki, Nucl. Phys. B 241, 381 (1984)
- [7] D. Morgan, E. Winstanley, J. Brunner, L. F. Thompson, Probing quantum decoherence in atmospheric neutrino oscillations with a neutrino telescope, Astrop. Phys. 25 (2006) 311.
- [8] E. Lisi, A. Marrone, D. Montanino, Probing possible decoherence effects in atmospheric neutrino oscillations, Phys. Rev. Lett. 85 (2000) 1166.
- [9] A. M. Gago, E. M. Santos, W. J. C. Teves, R. Z. Funchal, Quantum dissipative e ff ects and neutrinos : current constraints and future perspectives, Phys. Rev. D 63 (2001) 073001.
- [10] G. Barenboim, N. E. Mavromato, Cpt violating decoherence an lsnd: a possible window to planck scale physics, JHEP 01 (2005) 31.
- [11] G. L. Fogli, E. Lisi, A. Marrone, D. Montanino, A.Palazzo, Probing nonstandard decoherence e ff ects with solar and kamland neutrinos, Phys. Rev. D 76 (2007) 033006.
- [12] G. L. Fogli, E. Lisi, A. Marrone, D. Montanino, Status of atmospheric neutrino νµ → ντ oscillations and decoherence after the first k2k spectral data, Phys. Rev. D 67 (2003) 093006.
- [13] T. Ohlsson, Equivalence between gaussian averaged neutrino oscillations and neutrino decoherence, Phys. Lett. B 502 (2001) 159.
- [14] R. L. N. Oliveira, M. M. Guzzo, P. de Holanda, Quantum dissipation and cp violation in minos, Phys. Rev. D 89 (2014) 053002.
- [15] R. L. N. Oliveira, M. M. Guzzo, Dissipation and θ 13 in neutrino oscillations, Eur. Phys. Jour. C 73 (2013) 2434.
- [16] G. Balieiro Gomes, M. M. Guzzo, P. C. de Holanda, R. L. N. Oliveira, Parameter Limits for Neutrino Oscillation with Decoherence in KamLAND, arXiv:1603.04126v1 [hep-ph].
- [17] We are making analyses to put bounds on relaxation for solar neutrinos. This analysis will appear in elsewhere, but our preliminar results are γ < 10 -18 eV at 95% at C.L. for relaxation e ff ect. The result constrains enormously the relaxation e ff ect, such that it becomes unreachable to sensibility of any terrestrial experiments.
- [18] C. P. Burgess, D. Michaud, Neutrino propagation in a fluctuating sun, Annals. Phys. 256 (1997) 1.
- [19] P. Bakhti , Y . Farzan, T. , Revisiting the quantum decoherence scenario as an explanation for the LSND anomaly, JHEP. 2015, 007 (2015).
- [20] R. Alicki, K. Lendi, Quantum dynamical semigrou-ps and applications, Lect. Notes Phys., Springer-Verlag, Berlim, 1987.
- [21] H. P. Breuer, F. Petruccione, The Theory of Open Quantum Systems, Lect. Notes Phys., Oxford University Press, Oxford, 2002.
- [22] G. Lindblad, On the generators of quantum dynamical semigroups, Commun. Phys. 48 (1976) 119.
- [23] E. B. Davies, Markovian master equations, Commun. Phys. 39 (1974) 91.
- [24] R. Dumcke, H. Sponh, The proper form of the gererator in the weak coupling limit, Z. Phys. B 34 (1979) 419.
- [25] K. Kraus, General state changes in quantum theory, Annals. Phys. 64 (1971) 311.
- [26] E. Joos, et al., it Decoherence and the Appearance of Classical World in Qunatum Theoriy, 2ed. Springer, New York, 2003.
- [27] U. Weiss, Quantum Dissipative Systems, Vol. XIII, World ScientificHermann, Singapore, 1993.
- [28] F. Benatti, H. Narnhofer, Entropy behaviour under completely positive maps, Lett. Math. Phys. 15 (1988) 325.
- [29] R. N. Mohapatra, P. B. Pal, Massive Neutrinos in Physics and Astrophysics, Vol. 60, Word Scientific Publishing, Singapore, 1998.
- [30] C. Giunti, C. W. Kim, Fundamentals of neutrino physics and astrophysics, Oxford University Press, New York, 2007.
- [31] D. Hooper, D.Morgan, E. Winstanley, Probing quantum decoherence with high-energy neutrinos, Phys. Lett. B 609 (2005) 206.
| null |
[
"Roberto L. N. Oliveira",
"Marcelo M. Guzzo",
"Pedro C. de Holanda"
] |
2014-08-04T21:40:27+00:00
|
2016-05-10T18:51:19+00:00
|
[
"hep-ph",
"quant-ph"
] |
Quantum Dissipation in a Neutrino System Propagating in Vacuum and in Matter
|
Considering the neutrino state like an open quantum system, we analyze its
propagation in vacuum or in matter. After defining what can be called
decoherence and relaxation effects, we show that in general the probabilities
in vacuum and in constant matter can be written in a similar way, which is not
an obvious result in this approach. From this result, we analyze the situation
where neutrinos evolution satisfies the adiabatic limit and use this formalim
to study solar neutrinos. We show that the decoherence effect may not be
bounded by the solar neutrino data and review some results in the literature.
We discuss the current results where solar neutrinos were used to put bounds on
decoherence effects through a model-dependent approach. We conclude explaining
how and why this models are not general and we reinterpret these constraints.
|
1408.0824v1
|
## Polymer Nanofibers with Outstanding Thermal Conductivity and Thermal Stability: Fundamental Linkage between Molecular Characteristics and Macroscopic Thermal Properties
Teng Zhang, Xufei Wu, Tengfei Luo † † †, ‡,*
† Aerospace and Mechanical Engineering and Center for Sustainable Energy at Notre Dame, University ‡ of Notre Dame, Notre Dame, Indiana 46556, United States
## Abstract
* Tel.:+1 574 631 9683; E-mail: [email protected]. (Corresponding Author) Polymer nanofibers with high thermal conductivities and outstanding thermal stabilities are highly desirable in heat transfer-critical applications such as thermal management, heat exchangers and energy storage. In this work, we unlock the fundamental relations between the thermal conductivity and thermal stability of polymer nanofibers and their molecular characteristics by studying the temperature-induced phase transitions and thermal transport of a series of polymer nanofibers. Ten different polymer nanofibers with systematically chosen molecular structures are studied using large scale molecular dynamics simulations. We found that high thermal conductivity and good thermal stability can be achieved in polymers with rigid backbones, exemplified by π -conjugated polymers, due to suppressed segmental rotations and large phonon group velocities. The low probability of segmental rotation does not only prevent temperature-induced phase transition but also enables long phonon mean free paths due to reduced disorder scattering. Although stronger inter-chain interactions can also improve the thermal stability, polymers with such a feature usually have heavier atoms, weaker backbone bonds, and segments vulnerable to random rotations, which lead to low thermal conductivities. This work elucidates the underlying linkage between the molecular nature and macroscopic thermal properties of polymer nanofibers, which is instrumental to the design of thermally conductive polymer nanofibers with high temperature stabilities.
Keyword: polymer nanofiber, thermal conductivity, thermal stability, molecular dynamics
## 1. Introduction
Amorphous polymers, such as polymer foam and composite polymer, are usually used as thermal insulator due to their low thermal conductivities (~O(0.1) W/mK). However, 1 contrary to common wisdom, recent studies have shown that polymers such as polyethylene (PE) chains are not intrinsically of low thermal conductivities. 2, 3 Using molecular dynamics (MD) simulations, PE single chains were predicted to have very high conductivity (~O(100) W/mK). 2 While a single molecular chain in a strictly straight geometry simulated in Ref. [ ] is 2 impractical in nature, their work shed the light on possible large thermal conductivity in highly aligned polymer fibers, in which the molecular chains are very straight. From single chain to PE fibers, the impact of van der Waals (vdW) forces between polymer chains were also studied by MD simulations. Although the thermal conductivity of PE bulk crystal was found to be 3 reduced to ~45 W/mK due to the vdW-force-induced anharmonic phonon scattering, this 3 value is still larger than many metals.
PE fibers with high thermal conductivity were also reported in experimental studies. 4-8 Oriented PE fibers were fabricated by mechanical drawing process, and the thermal conductivity was improved from ~0.55 W/mK to 18.8 W/mK. PE nanowire arrays of high 4 thermal conductivity (up to ~20 W/mK) were also fabricated by nanoporous template wetting technique, and the ~20 W/mK value for PE nanofiber array was also confirmed by recent 5 time-domain thermoreflectance (TDTR) measurements. High thermal conductivity up to 42 6 W/mK was also reported in high modulus gel-spun PE nanofibers, but this result may suffer 7 from errors due to thermal radiation near room temperature. A very high thermal conductivity of 104 W/mK measured using cantilever method was reported for ultra-drawn PE fibers
recently, leading to the discussion on the upper-limit of polymer thermal conductivity. The 8 availability of thermally conductive polymers can expand the plastic industry, replacing metals and ceramics in heat transfer devices and equipment, leading to energy and cost savings.
The thermal conductivity of polymers is also found to be significantly morphology-dependent. 9-14 From amorphous phase to crystalline structures, the thermal conductivity of PE was predicted to increase from ~0.3 W/mK 15 to ~50 W/mK by MD simulations. Within the crystalline phase, a phase transition of PE at ~400 K can destroy the 9 along-chain segmental order, leading to an abrupt decrease in the thermal conductivity by almost one order of magnitude (MD results). 9, 10 Complex phase transition phenomena in PE nanofiber were observed experimentally, 16 and the corresponding thermal conductivity reduction was also reported by TDTR measurements. Such morphology changes do not 6 necessarily destroy the entity of the polymer nanofibers but significantly impair the thermal conductivity, presenting thermal stability issues in thermal transport. This limits the application of thermally conductive PE nanofibers at high temperatures. It is thus imperative to take the thermal stability into consideration when choosing or designing polymer nanofibers for heat transfer applications.
Experimental efforts in improving thermal conductivity of polymers have been focused almost exclusively on compositing. 17-27 Changing the morphology to enhance thermal conductivity has been studied for a few different polymers by drawing them into nanofibers. 6, 8, 28, 29 However, the selection of thermally conductive nanofibers has not been done under theoretical guidance. Wang et al. made a connection between the modulus of polymer
nanofibers and their thermal conductivities, but the relation is not monotonic. On the other 6 hand, thermal stabilities of the polymer nanofibers are largely ignored when exploring high thermal conductivities. It is thus highly desirable that we can tell whether a polymer has the potential to become thermally conductive with high temperature stability when formed into nanofibers by simply looking at the chemical composition of the molecules. This ability would be instrumental to the design of thermally conductive polymer nanofibers for applications at different temperature ranges. This calls for the elucidation of the fundamental linkage between the thermal properties and the molecular characteristics of polymers.
In this work, we use large scale molecular dynamics (MD) simulations to study the thermal transport in a variety of polymer nanofibers as a function of temperature which influences their morphologies. It is worth noting that classical MD simulations do not include any quantum effect, and all vibrational modes are excited regardless of the temperature in the simulations. In reality, some high frequencies modes are actually not excited when the temperature is lower than the Debye temperature. However, all the simulations presented in this work are at 300K or higher, and the low frequency acoustic modes, which are believed to be the most important contributors to thermal transport, are excited in this temperature range in reality. High frequency modes, especially those related to the light hydrogen atoms, are not excited at room temperatures. However, these modes are usually highly localized and do not contribute much to the thermal transport. 30-33 The simulation details are discussed in the simulation section. We first relate the morphology-influenced thermal conductivity to the inter-chain and intra-chain interatomic interactions through parametric studies of a model PE system. The obtained relations are then generalized by analyzing the temperature-dependent thermal
conductivities of ten different polymers nanofibers featuring weak and strong inter-chain and intra-chain interactions. Finally, we analyze the fundamental relation between the molecular characteristic and phonon properties in these polymer nanofibers. Based on these analyses, we are able to provide a theoretical connection between the thermal properties of polymer nanofibers and their molecular features.
## 2. Results and Discussions
## 2.1. Parametric Study of Thermal Transport in PE
Although PE has high thermal conductivity (~50 W/mK) at 300K, a sharp drop to ~5 W/mK at ~400K due to a phase transition which leads to random segmental rotations is reported previously (Figure 1). 6, 9, 10, 16 Lindemann's criterion predicts that melting initiates when the amplitude of vibration becomes large enough for adjacent atoms to occupy the same space. 34-36 This means that if the atomic movements can be confined, the phase transition can be suppressed. Generally, the two dominant factors that influence the large motions of atoms in a polymer chain are: dihedral angle strength and inter-chain interaction, because they are usually the weakest interactions among the different interatomic interactions ( e.g. , bond length and bond angle) in polymer chains. For example, the energy constants of the dihedral angle of PE backbone is 0.2432 Kcal/mole, and the inter-chain energy constant is 0.054 Kcal/mole, but those for bond length and bond angles are 299.67 Kcal/mole and 39.52 Kcal/mole. 37, 38 If the dihedral angle is weak, the segments in the chain can easily rotate and thus lead to disorder along the chain. Such disorder can be suppressed if the inter-chain interaction is strong enough to bundle the chains tightly together and limit the free space for
segment rotation.
Figure 1. Phase transition from all-trans conformation crystalline phase to aligned phase with random segmental rotations. When temperature increases above the phase transition temperature, the magnitude of polymer segmental vibration will become large enough to break the inter-chain crystalline lattice, leading to disorders along the polymer chains and dramatically lower thermal conductivity.
In this part, we perform parametric studies to control the morphology change of a model PE nanofiber. To suppress segmental rotation and confine chain movement, intrinsically stiff chain backbones and stronger inter-chain confinement are artificially realized by increasing the energy constants of the dihedral angle and inter-chain vdW interaction, respectively. 39, 40 In the condensed-phase optimized molecular potentials for atomistic simulation studies (COMPASS) potential, 37, 38 for PE used in this work, the dihedral angle and the 6/9 Lennard-Jones potential terms adopt the following forms:
$$E _ { d i h e d r a l } = \Sigma _ { n = 1 } ^ { 3 } K _ { n } [ 1 - c o s ( n \Phi ) ]$$
$$E _ { V d w } = \varepsilon \left [ 2 \left ( \frac { \sigma } { r } \right ) ^ { 9 } - 3 \left ( \frac { \sigma } { r } \right ) ^ { 6 } \right ], r < r _ { c }$$
where 𝛷 is the dihedral angle value, 𝐾𝑛 is the dihedral angle energy constant, 𝑟 is the
distance between two atoms, 𝜎 is related to atom equilibrium positions, 𝜀 is the energy constant of the vdW interaction, and 𝑟 𝑐 is the cutoff.
To investigate the relationship between thermal stability and polymer structure, a parametric study on 𝐾𝑛 and 𝜀 is performed. With 𝐾𝑛 and 𝜀 ranging from 0.8 to 1.2 times the original values, polymer structures from 300K to 500K are characterized, 9, 10, 41 and the corresponding thermal conductivities are calculated (Figure 2). The crystalline PE structures are first prepared by running the simulations in NPT ensembles at 300 K and 1 atm for 1 ns. The structures are then heated up to 500K at a rate of 40 K/ns. Figure 2a shows the simulation volume as a function of temperature for different model systems. We see large jumps in the volumes, which are results from phase transitions. Heating rate is known to influence the phase transition temperatures. In a separate set of simulations, we extract the steady state simulation volumes from equilibrium NPT runs at different temperatures which correspond to infinitely slow heating processes (Figure 2a, black dots). The phase transition temperatures from these two sets of data are within 5 K of each other, showing that the 40 K/ns heating rate does not influence the observed morphology change significantly. Since in this parametric study we are interested in the qualitative relations among interatomic interactions, morphology and thermal conductivity, the morphology characterizations are performed on the structures obtained from the simulations with a 40 K/ns heating rate for convenience.
Figure 2. Morphology-thermal conductivity relation from parametric studies on model PE nanofibers. (a) The volume of the model PE systems with different inter-chain vdW interactions and dihedral energies; (b) normalized dihedral angle distribution of original PE at 300K and 500K, and the overlap of the dihedral angle distributions (grey area); (c) dihedral angle distribution of different model PE nanofibers at 300K; (d) the overlap of dihedral angle distributions from 300K to 500K; and (e) thermal conductivity from 300K to 500K for different model PE nanofibers.
Figure 2a shows that phase transitions, which are well indicated by the sharp increase in volume, occur at lower temperatures when the dihedral energy or the vdW energy constant is reduced. The phase transition temperatures shift to higher temperatures (up to 440K), when vdW or dihedral energy is made 20% larger than the original values (Figure 2a, pink and red
lines). Due to the strong bond and angle interactions in polymers, bond length and angle cannot have large thermal fluctuations. 37, 39 Polymer chain structures in different phases differ mainly due to bond rotation, which can be characterized using dihedral angle distribution. 9-11, 42-44 Below the phase transition temperature, the sharp mono-peak dihedral angle distribution around 180 indicates allo trans conformations of the segments along the chain (Figure 2b, 300K and upper inset). The broadening of the dihedral angle distribution peak around 180 o and the appearance of the gauche conformations (dihedral angles of 60 or 300 ) above the o o phase change temperature indicate coiled chains in which segments rotate more freely (Figure 2b, 500K and lower inset). 9-11
Larger vdW energy and dihedral energy lead to stronger confinement for chain motion and segment rotation, and thus sharper peaks are observed (Figure 2c, inset). To compare the morphologies and quantify the segmental rotation at different temperatures, we calculate the overlap of the dihedral angle distribution at two difference temperatures. If the structure remains the same, the overlap is 1. When the difference between two structures becomes larger, the overlap of the dihedral angle distribution decreases. Structures at 300 K, which are highly ordered, are set as the reference, and dihedral angle distributions at other temperatures are compared against this reference ( e.g. , Figure 2b, grey area). In Figure 2d, sharp drops are found in the dihedral angle distribution overlap plot around the phase transition temperatures that are determined from Figure 2a. The much smaller overlaps after the phase transition indicate much more disordered structures, which impair their thermal conductivities due to disorder-phonon scattering. 9-11 Figure 2e shows that the thermal conductivities undergo sharp drops when segmental rotations emerge. The thermal conductivity drops for PE structures
with larger vdW and dihedral energies are found to occur at higher temperatures, thus extending the temperature range for high thermal conductivities (Figure 2e). Even at the same temperature, structures with larger vdW and dihedral energies have higher thermal conductivities (Figure 2e), since these structures have less segmental disorders (Figure 2c inset).
From this parametric study, we understand the following principle: stiffer backbones, which are less susceptible to segmental rotations, and stronger inter-chain interactions, which confine the chain movements and thus further suppress segmental rotations, can enable better thermal stability and higher thermal conductivity.
## 2.2. Polymer with Strong Inter-Chain Interactions
To generalize the principle obtained from Section 2.1 to real materials, we first investigate three other polymer nanofibers which have structures similar to PE but with stronger inter-chain interactions: Nylon 6-6, a polyamide class polymer with strong inter-chain hydrogen bonds (H-bond); Teflon, a fluorocarbon solid with large electron dipole on the C-F bond and thus large Columbic inter-chain interactions; and polyketone (PK), which has a strong chain-chain attraction due to the polar ketone groups. The chemical structures of these polymers are shown in Figure 3 together with other polymers studied in this work. To quantify the strength of the inter-chain interactions, the inter-chain energy density is calculated as the total inter-chain pair energy (summation of vdW and Coulombic energies) divided by the total volume. From PE, Nylon, PK to Teflon, a clear increasing trend is found in the inter-chain energy density (Table 1). As the inter-chain energy gets larger, the
volume expands much less when heated up from 300K to 500K (Figure 4a).
Figure 3. Thermal conductivity and structure of polymer nanofibers at different temperatures.
Π-conjugated polymers are plotted by solid lines with circle markers, while others are plotted by dash lines with rectangular markers.
) and inter-chain energy
Table 1. vdW energy constant ( 𝜀 ) of representative atoms, atomic charge ( q density of PE, Nylon, PK and Teflon. The atom types are included in the brackets.
| | 𝜀 (Kcal/mol) | q (electron charge) | q (electron charge) | Inter-chain Energy Density (Kcal/Å 3 /mol) |
|--------|----------------|-----------------------|-----------------------|----------------------------------------------|
| PE | 0.0540 (C) | 0.053 (H) | -0.106 (C) | 0.01 |
| Nylon | 0.2670 (O) | 0.378 (H) | -0.531 (O) | 0.07 |
| PK | 0.2670 (O) | 0.396 (C) | -0.396 (O) | 0.38 |
| Teflon | 0.0598 (F) | 0.500 (C) | -0.250 (F) | 0.74 |
Figure 4. Structural characteristics of polymer nanofibers with different inter-chain interactions: (a) volume from 300K to 500K, (b) dihedral angle distribution at 300K, and (c) the overlap of dihedral angle distribution from 300K to 500K for PE, Nylon, PK and Teflon.
To further characterize the polymer structures, dihedral angle distributions at 300K of the four polymer nanofibers are calculated (Figure 4b). Similar to PE, both Nylon and PK have narrow distributions, indicating alltrans conformations and straight chains (Figure 4b). For Nylon, segmental rotations are found to emerge above 460K (Figure 4c), and this leads to an abrupt thermal conductivity drop from ~6 W/mK to below 2 W/mK (Figure 3). Experimental data also confirm that phase transition happens around 460K for electrospun Nylon 6-6 nanofibers. 45 It is worth noting that the volume of Nylon shows a discontinuity around 375K (Figure 4a). However, the dihedral angle distribution evolves gradually around this temperature (Figure 4c). It is found that the orientations of some chains in the Nylon nanofiber changed collectively at this temperature, but no random segment rotations along the
chains are observed. The thermal conductivity thus shows no significant change around 375K. For PK nanofiber, the good crystalline structure without much segmental rotations is found to be maintained up to 500K (Figure 4a & 4c). As a result, its thermal conductivity only decreases gradually (shown in Figure 3), likely due to anharmonic phonon scattering. A very high melting point of 551K has also been previously reported in experiment for PK. 46 Although Teflon shows little thermal expansion due to the strong inter-chain interactions (Figure 4a & Table 1), the broader and multi-peak dihedral angle distribution even at 300K indicates intrinsic disorder in segmental orientations (Figure 4b). This lack of long range order in Teflon above 300K has been both predicted by simulation and verified by experiments. 47-49 These disorders scatter phonons and they are inherent in Teflon chains even at room temperature. Therefore, Teflon is found to have low thermal conductivity in the entire 300-500K temperature range (Figure 3).
From these four polymer nanofibers, we see that the thermal stability indeed increases as the inter-chain interaction becomes stronger. However, polymers with stronger inter-chain interactions do not necessarily have higher thermal conductivity, which is related to segmental order.
## 2.3. π-conjugated Polymers with Rigid Backbones
Due to the outstanding electronic-photonic properties and thermochemical stability, π-conjugated materials are applied in many areas, including chemical sensors, light-emitting diode, photovoltaics and cancer treatment. 50-56 In π-conjugated molecules, the overlap of the p -orbitals also imposes strong constraints that suppress bond rotation and thus lead to rigid
backbones. 57-61 Polyacetylene (PA) is the simplest π-conjugated polymer (structure shown in Figure 3). In the COMPASS potentials, the dihedral potential is presented by a sum of three cosine functions (Equation 1). By comparing the potential parameters for PE and PA, much larger dihedral energy constants are found in PA (Table 2). The volume jump of PA at ~480 K shows that the phase transition temperature is ~80 K higher than that of PE (Figure 5a). Since PA and PE are of similar structure and similar inter-chain energy density, higher phase transition temperature in PA thus mainly comes from the stronger dihedral angle. PA dihedral angle distribution at 300K has a sharp mono peak, indicating alltrans conformations and straight chains (Figure 5b & upper inset). Although the overlap of dihedral angle distribution of PA decays as temperature increases, there is no sharp drop from 300K to 500K, indicating that no significant segmental rotation is present even above the ~480 K phase transition temperature. As a result, no sharp thermal conductivity drop is found during the ~480 K phase transition (Figure 3). Similar to Nylon, the volume expansion in PA around 480K is due to the collective change of the orientations of some chains in the nanofiber.
Table 2. Dihedral angle energy constants and pair energy densities of PE, PA, and PBT in
COMPASS potential.
| Type of Dihedral Group | Leading Energy Constant (Kcal/mole) | Pair Energy Density (Kcal /Å 3 /mol) |
|--------------------------|---------------------------------------|----------------------------------------|
| PE: C-C-C-C | 0.1223 | 0.01 |
| PA: C=C-C=C | 8.3667 | 0.01 |
| PBT: C=C-N=C | 6.8517 | 0.24 |
|----------------|----------|--------|
| N=C-C=C | 21.1715 | |
| C=C-S-C | 31.5576 | |
| N-C-S-C | 21.1715 | |
From Table 2, we see even larger dihedral energy constants in π-conjugated polymers with ring structures ( e.g. , PBT) compared to those in PA. We simulated four π-conjugated polymer nanofibers with ring structures, including polythiophene (PT), poly( p -phenylene benzobisthiazole) (PBT), poly( p -phenylene) (PPP) and poly{diimidazo pyridinylene (dihydroxy) phenylene} (PIPD) (structures shown in Figure 3). The very large dihedral energy constants lead to sharp dihedral angle distributions at 300K for all four polymer nanofibers (Figure 5b). The dihedral angles at 0 and 360 indicate o o cis conformations, which come from the ring structures (Figure 5b & lower inset). Due to the existence of ring structures and the inter-chain π-π stacking, 62 large inter-chain energy density (Table 3) and steric effects stabilize the crystalline structure of these four π-conjugated polymer nanofibers. Therefore, no volume jump is observed from 300K to 500K for these polymer nanofibers (Figure 5a), and the dihedral angle distribution overlaps decay slower than PA (Figure 5c). Thermal conductivity calculations show that all five π-conjugated polymers have thermal conductivities higher than 20 W/mK at 300K (Figure 3). Unlike PE, the thermal conductivities of all the π-conjugated polymers only decrease gradually as temperature goes up, which leads to good thermal conductivities (15 ~ 50 W/Km) even up to 500K (Figure 3).
Figure 5. Structural characteristics of polymer nanofibers with different backbone stiffness: (a) volume from 300K to 500K, (b) dihedral angle distribution at 300K, and (c) the overlap of dihedral angle distribution from 300K to 500K for PE, P A, PT, PBT, PPP , PIPD, and Kevlar.
Table 3. Pair energy density, phonon group velociy, volumetric heat capacity, phonon mean free path, and thermal conductivity for the ten folymer nanofibers studied.
| | Pair Energy Density (Kcal/Å 3 /mol) | Phonon Group Velocity (m/s) (averaged) | Volumetric Heat Capacity (J·m c -3 · K -1 ) | Mean free path (nm) | Thermal Conductivity at 300K (W/mK) |
|------|---------------------------------------|------------------------------------------|-----------------------------------------------|-----------------------|---------------------------------------|
| PPP | 0.19 | 7663.86 | 4.3 | 1.37 | 45.13 |
| PA | 0.01 | 7410.13 | 3.37 | 4 | 100.09 |
| PBT | 0.24 | 7241.58 | 3.9 | 1.34 | 37.72 |
| PIPD | 0.16 | 7074.79 | 4.26 | 0.75 | 22.42 |
| PT | 0.06 | 7069.99 | 2.93 | 3.04 | 62.71 |
| PE | 0.01 | 5292.94 | 4.89 | 1.90 | 49.2 |
|--------|--------|-----------|--------|--------|--------|
| PK | 0.38 | 4264.8 | 4.4 | 0.86 | 16.08 |
| Nylon | 0.07 | 3566.04 | 4.89 | 0.46 | 7.99 |
| Kevlar | 0.25 | 3230.32 | 4.16 | 0.6 | 8.05 |
| Teflon | 0.74 | 2801.78 | 4.43 | 0.28 | 3.43 |
We found that the π-conjugated nature is important to achieving high thermal stability and high thermal conductivity at the same time. One important feature of π-conjugated polymer is the existence of alternating single and double bonds. Kevlar also has benzene rings that have alternating single and double bonds, but the delocalization of p -electrons over the whole molecule is not possible due to the presence of the carboxamide group (portion highlighted in blue, Figure 3). Such a characteristic makes Kevlar not a π-conjugated polymer. Figure 5 shows that there is no phase transition from 300K to 500K and no significant random segmental rotations due to the large pair energy density in Kevlar (Table 3). However, the thermal conductivity of Kevlar is below 10 W/mK in the whole temperature range, which is at least 50% smaller than the conjugated polymers. This finding is consistent with a recent measurement which shows that Kevlar has much lower thermal conductivity compared to π-conjugated polymer nanofibers. We found that segmental rotation is intrinsic in Kevlar 6 chains at 300K (Figure 5b, smaller yellow shoulder peaks). However, unlike Teflon which has intrinsic segmental rotations that are very randomized, long range order is still preserved in the rotated Kevlar segments. Therefore, the low thermal conductivity of Kevlar cannot be explained by the rotational disorder-phonon scattering mechanism. We attribute this to the
characteristics of the phonons which are analyzed in the following section.
## 2.4. Phonon Dispersion Relation Analysis
The very high thermal conductivity of PE has been well studied by both simulations and experiments. 2, 3, 8 The high thermal conductivities in π-conjugated polymers, including PA and PPP, have also been reported by studying stand-alone single chains by simulation. 63 Experimental data also show that the thermal conductivities of π-conjugated polymer nanofibers (PIPD and PBT) are much higher than those of other polymers ( e.g. Kevlar and Vectra). 6 Here, to explain the higher thermal conductivities in π-conjugated polymer nanofibers, the following equation for thermal conductivity of one-dimensional systems is analyzed:
$$\kappa = c \cdot \bar { \nu } \cdot \bar { l }$$
where 𝜅 is thermal conductivity along polymer chains, 𝑐 is volumetric heat capacity, 𝑣̅ is the average phonon group velocity, and 𝑙 ̅ is the phonon mean free path. Since our MD simulations are in the classical limit, the volumetric heat capacity can be expressed as 𝑐 = 3𝑘𝐵 𝑉 𝑁 , where 𝑘𝐵 is Boltzmann constant, 𝑁 is the number of atoms, and 𝑉 is volume (Table 3). Although the volumetric heat capacity varies from 2.93 J/(cm K) to 4.89 J/(cm K) 3 3 for the ten polymers, the smallest values are found in PT (2.93 J/(cm K)) and PA (3.37 3 J/(cm K)) - the two π-conjugated polymers with the highest thermal conductivities in all ten 3 nanofibers. Therefore, volumetric heat capacity is not the key to the high thermal conductivity in these nanofibers.
Another factor is the phonon group velocity. To calculate the phonon group velocity,
phonon dispersion relation is first calculated by performing two-dimensional Fourier transforms of the velocities of the backbone atoms obtained from MD simulations (Figure 6, see Appendix B for calculation details). 64-66 Since we believe that acoustic phonons are the dominant heat carriers that contribute to thermal transport, we have only included the low frequency regions including acoustic phonons in Figure 6 for clearer comparison. The complete phonon dispersion relations are shown in Appendix B. For each acoustic branch, the phonon group velocity is estimated as the slope between the original and the point of the max frequency (blue line in PPP, Figure 6). The average phonon group velocity is approximated as the arithmetic mean of all four acoustic branches (one longitudinal, two transverse, and one 2 torsional branch, highlighted by red dots in Figure 6), and the results are listed in Table 3. Although such estimation is relatively rough, it gives us a general idea of the relative magnitude of the phonon group velocities of different polymer fibers so that a comparison among them can be made. It is clear that all the π-conjugated polymers have mean phonon group velocities above 7000 m/s (Table 3, red) - higher than those of the other nanofibers.
Figure 6. Phonon dispersion relation for the ten polymer nanofibers.The acoutic phonon branches are marked with red dots. The average phonon group velocity of one branch is estimated as the slope between the origin and the point of the max frequency (e.g., blue line, in PPP).
At the molecular level, strong bonds and small atomic masses can usually enable high phonon group velocities. In the COMPASS potential, bond strength is defined by the energy constants in the following formula:
$$E _ { b } = K _ { 2 } ( r - r _ { 0 } ) ^ { 2 } + K _ { 3 } ( r - r _ { 0 } ) ^ { 3 } + K _ { 4 } ( r - r _ { 0 } ) ^ { 4 }$$
where 𝑟 is the distance between the two bonded atoms, 𝑟 0 is equilibrium bond distance, 𝐾2 , 𝐾3 , and 𝐾4 are the energy constants defining the strength of the bond. Table 4 shows that larger
bond strength is found in π-conjugated polymers due to p -orbitals conjugation and delocalized electrons. The carboxamide group (in Nylon and Kevlar), on the other hand, will decrease the strength of the bond between carbon and carbonyl carbon in the backbone, leading to lower phonon group velocities. As a result, Nylon and Kevlar have low phonon group velocities (Table 3). Compared to light hydrogen atom, oxygen and fluorine atoms are much heavier, which can also explain the relatively low phonon group velocity in PK, Teflon, and Kevlar.
Table 4. Typical bond parameters in PE, PA, PBT, and Nylon
| | 𝑟 0 | 𝐾 2 | 𝐾 3 | 𝐾 4 |
|----------|--------|---------|----------|---------|
| PE ( ) | 1.533 | 299.67 | -501.77 | 679.81 |
| PA( ) | 1.417 | 470.836 | -627.618 | 1327.63 |
| PBT ( ) | 1.3485 | 508.859 | -977.691 | 1772.51 |
| Nylon( ) | 1.5202 | 253.707 | -423.037 | 396.9 |
With the thermal conductivity, volumetric heat capacity and phonon group velocity, we can calculate the effective phonon mean free path at 300K according to Equation 3. Due to the approximations used in calculating volumetric heat capacity and phonon group velocity, the mean free paths calculated from Equation 3 are not expected to be exactly accurate. However, the comparison of the phonon mean free path among the fibers can offer valuable insights to the phonon transport physics in these nanofibers. For example, Table 3 clearly shows that polymer nanofibers with good along-chain orders, such as the π-conjugated polymers and PE, tend to have larger phonon mean free paths. As discussed previously, the
segments in the π-conjugated polymers are well ordered and thus the disorder-phonon scattering is weak, enabling long phonon mean free paths. It is also interesting to find out that the effective phonon mean free path of Teflon is the smallest. This is believed to be related to the intrinsic segmental disorders in the Teflon chains which scatter phonons.
## 3. Conclusion
Due to the overlap of the p -orbitals in π-conjugated polymers, rigid backbones in PA, PIPD, PBT, PPP and PT suppress segmental rotations and thus phonon scattering. The π-conjugation also enables strong backbones and thus large phonon group velocities. Therfore, π-conjugated polymers are found to have both good thermal stabilities and high thermal conductivities. Strong inter-chain interaction in PK, Nylon, Teflon and Kevlar can suppress thermal expansion and thus stabilize the structures at high temperatures. However, the presence of heavier atoms ( e.g. oxygeon) and weaker backbone bonds will lead to low phonon group velocities and thus low thermal conductivities of these polymers. Polymers with intrinsic segmental rotations ( e.g. , Teflon) will not have high thermal conductivities due to the disorder-phonon scattering. Such a factor also lead to the significant reductions in thermal conductivities of nanofibers which attain segmental rotations at high temperatures. This work made detailed connections between the molecular characteristics and macroscopic thermal properties of polymer nanofibers. It provides a general rule of designing thermally conductive polymer nanofibers with high temperature stabilities: one should choose molecules with intrinsically ordered backbones, strong backbone bonds and strong dihedral angles . According to this rule, π-conjugated polymers are singled out as a category of polymers that are ideal for fabricating nanofibers with high
thermal conductivities for high temperature applications.
## 4. Simulation Section
In the present work, MD simulations with the COMPASS are used to investigate the thermal stabilities and thermal conductivities of 10 different polymer nanofibers, including PA, PT, PPP, Kevlar, Nylon 6-6, PE, PBT, PK, PIPD, and Teflon. The COMPASS potential was created to accurately simulate the structural, vibrational, and thermo-physical properties of polymeric materials in isolated and condensed phases. 37, 38 It is thus chosen for the present study in which the morphology-thermal property relation is important. The initial fiber structures are constructed based on previous literatures (see Appendix A), and then minimized using multiple optimization steps, including steepest descent, conjugate gradient, and Newton minimizations, to guarantee reaching the lowest energy morphology. Periodic boundary conditions in all three spatial dimensions are applied to avoid surface effects. 67 This corresponds to the thick fiber limit where surface effect is not important. A fourth periodic condition which links one end of a chain to the other end of its image is used to simulate infinitely long chains, which mimic the monodisperse ultrahigh molecular weight polymers. 68 The number of chains are chosen so that the cross sectional dimensions are larger than two times of the interaction cutoff (10 Å), avoiding the double counting of the interactions between atoms.
Previous studies have shown that thermal conductivities of polymers chains in the chain length direction can be a function of domain length due to boundary phonon scattering. 9, 11, 63 In order to make fair comparison among different nanofibers, the nanofibers are constructed
by controlling the number of segments in polymer chains so that the simulation domain lengths are all around 50 nm at 300K (see Appendix A). From the minimized structures, the systems are simulated in NPT ensembles for 1 ns to obtain the stable nanofiber structures at different temperatures and 1 atm before any further calculations or characterizations are performed. Such a step effectively eliminates large stresses in the system, especially in the along-chain direction.
Non-equilibrium molecular dynamics (NEMD) simulation is used to calculate the thermal conductivity of different nanofibers. 3, 9-11, 69 In this method, a temperature gradient is created and maintained by Langevin thermostats controlling the temperatures of the two ends of the simulation domain. The temperatures of the heat sink and heat source regions are set to 15 K lower and higher than the average system temperature, respectively. After steady state is reached, temperature gradient ( 𝑑𝑇/𝑑𝑥 ) is obtained by fitting the linear portion of the temperature profile, and heat flux can be calculated using 𝐽 = 𝑑𝑄/𝑑𝑡/𝑆, where 𝑑𝑄/𝑑𝑡 is the average energy change rates of the two Langevin thermostats, and S is the cross sectional area. The thermal conductivity is then calculated by Fourier's law, 𝜅 = -𝐽/(𝑑𝑇/𝑑𝑥) . Figure 7 shows a representative setup and temperature profile of NEMD. For each simulation, four thermal conductivity values are calculated for different time blocks in the steady state, and the final value is the average of them with the error bar being the standard deviation.
Figure 7 NEMD scheme and corresponding temperature profile.
All the simulations are carried out using the large-scale atomic/molecular massively parallel simulator (LAMMPS). 70 The time step is chosen to be 0.25 fs, and a 10 Å cutoff is used for the 6-9 Lennard-Jones interaction, allowing for chain-chain interactions between the third and the fourth nearest-neighboring chains.
## Appendix A: Polymer Nanofiber Structures Construction
The initial structures of ten different polymer nanofibers are built based on the structures documented in literatures. 3, 71-79 The polymers studied include PA, 72 PT, 73 Kevlar, 75 PIPD, 71 PPP, 74 Nylon 6-6 (imported from Material Studio), PE, Teflon, 3 76 PBT, 77 and PK (imported from Material Studio) (Table A1). All the initial structures are minimized using multiple algorithms, including Steepest Descent, Conjugate Gradient, and Newton minimization (Smart Minimizer, Discover module of Material Studio, Accelrys Inc.), before they are used for MD simulations. Runs in NPT ensembles for 1 ns are performed to obtain the stable
structures before the thermal conductivity calculations at different temperatures from 300K to 500K. The structures of crystalline polymer nanofibers can be a strong function of temperature, drawn ratio, and pressure, 16, 80-84 and thus only the supercell dimensions of the stable structures at 300K and 1 atm are listed (Table A2).
Table A1. Structures and dimensions of the unit-cell constructed for different polymer
## nanofibers.
Table A2. The construction and dimensions of the super-cell for different polymer nanofibers.
| | Segments | Chain | Super-cell Dimensions (300K and 1 atm) | Super-cell Dimensions (300K and 1 atm) | Super-cell Dimensions (300K and 1 atm) |
|-----------|------------|---------|------------------------------------------|------------------------------------------|------------------------------------------|
| | per Chain | Number | X (Å) | Y(Å) | Z (Å) |
| PE | 400 | 40 | 506.70 | 24.12 | 33.15 |
| PA | 400 | 56 | 508.07 | 34.82 | 27.19 |
| Nylon 6-6 | 29 | 81 | 501.76 | 43.33 | 34.75 |
| PK | 134 | 40 | 496.65 | 27.98 | 29.03 |
| Teflon | 400 | 56 | 445.64 | 37.45 | 37.74 |
| PT | 160 | 40 | 562.39 | 32.28 | 29.87 |
| PPP | 110 | 40 | 471.91 | 34.14 | 26.28 |
| PBT | 40 | 54 | 478.28 | 35.75 | 32.17 |
| PIPD | 42 | 54 | 496.81 | 36.37 | 32.93 |
| Kevlar | 40 | 40 | 506.70 | 24.12 | 33.15 |
- ● super-cell dimension: X × Y × Z, with X parallel to OA, YZ in cross sectional plane
## Appendix B: Phonon Dispersion Relation Calculation
With the initial unit cell structures, single polymer chains are constructed with the same number of segments as in the nanofibers (Table A1 & A2). For each simulation, only one isolated polymer chain is placed in the supper cell, and the cross sectional dimensions are set to be enough large (50 Å× 50 Å), preventing the interaction with its own image. After the minimization, these single chains are first simulated at 2 K with fixed volume, and then simulated in ensembles of constant volume and energy (NVE) for 50 ps. During one NVE run,
the velocity of every backbone atom is recorded every 5 fs in a two-dimensional matrix. With the velocity matrixes from MD simulations, two-dimensional Fourier transform of the atomic velocity of one certain atom in the unit cell is performed:
$$\Phi ( \omega, k ) = \, \sqrt { \Sigma _ { \alpha } ^ { 3 } \left | \frac { 1 } { N } \Sigma _ { n = 0 } ^ { N - 1 } e ^ { i \frac { n } { N } k } \int \nu _ { \alpha } ( n, t ) e ^ { - i \omega t } d t \right | ^ { 2 } }, \, \alpha = \text{x}, y, z$$
where 𝑣𝛼(𝑛, 𝑡) is the atomic velocity, 𝜔 the frequency, 𝑘 is the wavevector, 𝑛 is the index of repeating unit along chain direction, and 𝑁 is the number of the repeating unit. Then the phonon dispersion relation is calculated by averaging the two-dimensional Fourier transform results of all the backbone atoms in the unit cell (Figure B1).
Wave Vector Reduced by 1/(PA Unit Cell Length)
Figure B1 Phonon dispersion relation for the ten polymer nanofibers.
## Acknowledgements
This research was supported in part by the Notre Dame Center for Research Computing and NSF through TeraGrid resources provided by SDSC Trestles under grant number TG-CTS100078. The authors acknowledge the financial support from American Chemistry Society (PRF# 54129-DNI10) and the startup fund from the University of Notre Dame.
## References
- (1) Hall, J. F. History and Bibliography of Polymeric Insulators for Outdoor Applications. IEEE Trans. Power Delivery 1993 , 8 , 376-385.
- (2) Henry, A.; Chen, G. High Thermal Conductivity of Single Polyethylene Chains using Molecular Dynamics Simulations. Phys. Rev. Lett. 2008 , 101 , 235502.
- (3) Henry, A.; Chen, G.; Plimpton, S. J.; Thompson, A. 1D-to-3D Transition of Phonon Heat Conduction in Polyethylene using Molecular Dynamics Simulations. Phys. Rev. B 2010 , 82 , 144308.
- (4) Choy, C. L.; Luk, W. H.; Chen, F. C. Thermal Conductivity of Highly Oriented Polyethylene. Polymer 1978 , 19 , 155-162.
- (5) Cao, B.; Li, Y.; Kong, J.; Chen, H.; Xu, Y.; Yung, K.; Cai, A. High Thermal Conductivity of Polyethylene Nanowire Arrays Fabricated by an Improved Nanoporous Template Wetting Technique. Polymer 2011 , 52 , 1711-1715.
- (6) Wang, X.; Ho, V.; Segalman, R. A.; Cahill, D. G. Thermal Conductivity of High-Modulus Polymer Fibers. Macromolecules 2013 , 46 , 4937-4943.
- (7) Choy, C. L.; Wong, Y. W.; Yang, G. W.; Kanamoto, T. Elastic Modulus and Thermal Conductivity of Ultradrawn Polyethylene. J. Polym. Sci. B: Polym. Phys. 1999 , 37 , 3359-3367.
- (8) Shen, S.; Henry, A.; Tong, J.; Zheng, R.; Chen, G. Polyethylene Nanofibres with very High Thermal Conductivities. Nat. Nanotechnol. 2010 , 5 , 251-255.
- (9) Zhang, T.; Luo, T. High-Contrast, Reversible Thermal Conductivity Regulation Utilizing the Phase Transition of Polyethylene Nanofibers. ACS Nano 2013 , 7 , 7592-7600.
- (10) Zhang, T.; Luo, T. Morphology-Influenced Thermal Conductivity of Polyethylene Single Chains and Crystalline Fibers. J. Appl. Phys. 2012 , 112 , 094304.
- (11) Luo, T.; Esfarjani, K.; Shiomi, J.; Henry, A.; Chen, G. Molecular Dynamics Simulation of Thermal Energy Transport in Polydimethylsiloxane. J. Appl. Phys. 2011 , 109 , 074321.
- (12) Guo, Z.; Lee, D.; Liu, Y.; Sun, F.; Sliwinski, A.; Gao, H.; Burns, P. C.; Huang, L.; Luo, T. Tuning the Thermal Conductivity of Solar Cell Polymers through Side Chain Engineering. Phys. Chem. Chem. Phys. 2014 , 16 , 7764-7771.
- (13) Algaer, E. A.; Alaghemandi, M.; Bö hm, M. C.; Mü ller-Plathe, F. Anisotropy of the Thermal Conductivity of Stretched Amorphous Polystyrene in Supercritical Carbon Dioxide Studied by Reverse Nonequilibrium Molecular Dynamics Simulations. J. Phys. Chem. B 2009 , 113 , 14596-14603.
- (14) Alaghemandi, M.; Gharib-Zahedi, M.; Spohr, E.; Bö hm, M. C. Thermal Conductivity of Polyamide-6,6 in the Vicinity of Charged and Uncharged Graphene Layers: A Molecular Dynamics Analysis. J. Phys. Chem. C 2012 , 116 , 14115-14122.
- (15) Sperling, L. H. In Introduction to Physical Polymer Science; John Wiley & Sons, Inc.: Hoboken, New Jersey, 2005 .
- (16) Tashiro, K.; Sasaki, S.; Kobayashi, M. Structural Investigation of
- Orthorhombic-to-Hexagonal Phase Transition in Polyethylene Crystal: The Experimental Confirmation of the Conformationally Disordered Structure by X-Ray Diffraction and Infrared/Raman Spectroscopic Measurements. Macromolecules 1996 , 29 , 7460-7469.
- (17) Mamunya, Y. P.; Davydenko, V.; Pissis, P.; Lebedev, E. Electrical and Thermal Conductivity of Polymers Filled with Metal Powders. Eur. Polym. J. 2002 , 38 , 1887-1897.
- (18) Devpura, Patrick E. Phelan, Ravi S. Prasher, Amit Size Effects on the Thermal Conductivity of Polymers Laden with Highly Conductive Filler Particles. Microscale Thermophys. Eng. 2001 , 5 , 177-189.
- (19) Tian, W.; Yang, R. Phonon Transport and Thermal Conductivity Percolation in Random Nanoparticle Composites. CMES 2008 , 24 , 123-141.
- (20) Yu, A.; Ramesh, P.; Itkis, M. E.; Bekyarova, E.; Haddon, R. C. Graphite ,
Nanoplatelet-Epoxy Composite Thermal Interface Materials. J. Phys. Chem. C 2007 111 , 7565-7569.
- (21) Yu, A.; Ramesh, P.; Sun, X.; Bekyarova, E.; Itkis, M. E.; Haddon, R. C. Enhanced Thermal Conductivity in a Hybrid Graphite nanoplatelet-carbon Nanotube Filler for Epoxy Composites. Adv. Mater. 2008 , 20 , 4740-4744.
- (22) Arai, S.; Endo, M.; Sato, T.; Koide, A. Fabrication of Nickel-Multiwalled Carbon Nanotube Composite Films with Excellent Thermal Conductivity by an Electrodeposition Technique. Electrochem. Solid State Lett. 2006 , 9 , C131-C133.
- (23) Terao, T.; Zhi, C.; Bando, Y.; Mitome, M.; Tang, C.; Golberg, D. Alignment of Boron Nitride Nanotubes in Polymeric Composite Films for Thermal Conductivity Improvement. J. Phys. Chem. C 2010 , 114 , 4340-4344.
- (24) Duong, H. M.; Yamamoto, N.; Bui, K.; Papavassiliou, D. V.; Maruyama, S.; Wardle, B. L. Morphology Effects on Nonisotropic Thermal Conduction of Aligned Single-Walled and Multi-Walled Carbon Nanotubes in Polymer Nanocomposites. J. Phys. Chem. C 2010 , 114 , 8851-8860.
- (25) Huang, X.; Iizuka, T.; Jiang, P.; Ohki, Y.; Tanaka, T. Role of Interface on the Thermal Conductivity of Highly Filled Dielectric Epoxy/AlN Composites. J. Phys. Chem. C 2012 , 116 , 13629-13639.
- (26) Huang, X.; Zhi, C.; Jiang, P. Toward Effective Synergetic Effects from Graphene Nanoplatelets and Carbon Nanotubes on Thermal Conductivity of Ultrahigh Volume Fraction Nanocarbon Epoxy Composites. J. Phys. Chem. C 2012 , 116 , 23812-23820.
- (27) Terao, T.; Bando, Y.; Mitome, M.; Zhi, C.; Tang, C.; Golberg, D. Thermal Conductivity Improvement of Polymer Films by Catechin-Modified Boron Nitride Nanotubes. J. Phys. Chem. C 2009 , 113 , 13605-13609.
- (28) Huang, X.; Liu, G.; Wang, X. New Secrets of Spider Silk: Exceptionally High Thermal Conductivity and its Abnormal Change Under Stretching. Adv. Mater. 2012 , 24 , 1482-1486.
- (29) Lu, X.; Wang, C.; Wei, Y. One -Dimensional Composite Nanomaterials: Synthesis by Electrospinning and their Applications. Small 2009 , 5 , 2349-2370.
- (30) Luo, T.; Lloyd, J. R. Non-Equilibrium Molecular Dynamics Study of Thermal Energy Transport in Au-SAM-Au Junctions. Int. J. Heat Mass Transfer 2010 , 53 , 1-11.
- (31) Shenogin, S.; Bodapati, A.; Keblinski, P.; McGaughey, A. J. H. Predicting the Thermal Conductivity of Inorganic and Polymeric Glasses: The Role of Anharmonicity. J. Appl. Phys. 2009 , 105 , 034906.
- (32) Kopidakis, G.; Aubry, S. Discrete Breathers in Realistic Models: Hydrocarbon Structures. Physica B 2001 , 296 , 237-250.
- (33) Savin, A. V.; Manevitch, L. I. Discrete Breathers in a Polyethylene Chain. Phys. Rev. B 2003 , 67 , 144302.
- (34) Lindemann, F. A. The Calculation of Molecular Vibration Frequencies. Physik. Z. 1910 , 11 , 609-612.
(35) Oxtoby, D. W. New Perspectives on Freezing and Melting. Nature 1990 , 347 , 725-730.
- (36) Hansen, J. Phase Transition of the Lennard-Jones System. II. High-Temperature Limit. Phys. Rev. A 1970 , 2 , 221-230.
- (37) Sun, H. COMPASS: an Ab Initio Force-Field Optimized for Condensed-Phase Applications-Overview with Details on Alkane and Benzene Compounds. J. Phys. Chem. 1998 , 102 , 7338-7364.
- (38) Sun, H.; Mumby, S. J.; Maple, J. R.; Hagler, A. T. An Ab Initio CFF93 all-Atom Force Field for Polycarbonates. J. Am. Chem. Soc. 1994 , 116 , 2978-2987.
- (39) Sundararajan, P. R.; Kavassalis, T. A. Molecular Dynamics Study of Polyethylene Chain Folding: The Effects of Chain Length and the Torsional Barrier. J. Chem. Soc. Faraday Trans. 1995 , 91 , 2541-2549.
- (40) Weeks, J. D.; Broughton, J. Q. Van Der Waals Theory of Melting in Two and Three Dimensions. J. Chem. Phys. 1983 , 78 , 4197-4205.
- (41) de Pablo, J. J.; Laso, M.; Suter, U. W. Simulation of Polyethylene Above and Below the Melting Point. J. Chem. Phys. 1992 , 96 , 2395-2403.
- (42) Sumpter, B. G.; Noid, D. W.; Wunderiich, B. Computer Experiments on the Internal Dynamics of Crystalline Polyethylene: Mechanistic Details of Conformational Disorder.
- J. Chem. Phys. 1990 , 93 , 6875-6889.
- (43) Noid, D. W.; Sumpter, B. G.; Wunderlich, B. Molecular Dynamics Simulation of the Condis State of Polyethylene. Macromolecules 1990 , 23 , 664-669.
- (44) Sozzani, P.; Bracco, S.; Comotti, A.; Simonutti, R. Motional Phase Disorder of Polymer Chains as Crystallized to Hexagonal Lattices. Adv. Polym. Sci. 2005 , 181 , 153-177.
- (45) Carrizales, C.; Pelfrey, S.; Rincon, R.; Eubanks, T. M.; Kuang, A.; McClure, M. J.; Bowlin, G. L.; Macossay, J. Thermal and Mechanical Properties of Electrospun PMMA, PVC, Nylon 6, and Nylon 6, 6. Polym. Adv. Technol. 2008 , 19 , 124-130.
- (46) Lommerts, B.; Klop, E.; Aerts, J. Structure and Melting of Perfectly Alternating Ethylene-Carbon Monoxide Copolymers. J. Polym. Sci. B: Polym. Phys. 1993 , 31 , 1319-1330.
- (47) Sprik, M.; Rö thlisberger, U.; Klein, M. L. Structure of Solid Poly (Tetrafluoroethylene):
- A Computer Simulation Study of Chain Orientational, Translational, and Conformational Disorder. J. Phys. Chem. B 1997 , 101 , 2745-2749.
- (48) Hyndman, D.; Origlio, G. NMR Absorption in Teflon Fibers. J. Appl. Phys. 2004 , 31 , 1849-1852.
- (49) Matsushige, K.; Enoshita, R.; Ide, T.; Yamauchi, N.; Taki, S.; Takemura, T. Fine Structure of the III-I Transition and Molecular Motion in Polytetrafluoroethylene. Jpn. J. Appl. Phys. 1977 , 16 , 681-687.
- (50) Brovelli, S.; Cacialli, F. Optical and Electroluminescent Properties of Conjugated Polyrotaxanes. Small 2010 , 6 , 2796-2820.
- (51) Dou, K. P.; De Sarkar, A.; Wang, C. L.; Zhang, R. Q. Intramolecular Torsion Based Molecular Switch Functionality Enhanced in Ï€-Conjugated Oligomolecules by a π-Conjugated Pendant Group. J. Phys. Chem. C 2011 , 115 , 13911-13918.
- (52) Ren, Y.; Biegger, F.; Baumgartner, T. Molecular Engineering of the Physical Properties of Highly Luminescent π-Conjugated Phospholes. J. Phys. Chem. C 2013 , 117 , 4748-4758.
- (53) Chen, Z.; Grumstrup, E. M.; Gilligan, A. T.; Papanikolas, J. M.; Schanze, K. S. Light-Harvesting Polymers: Ultrafast Energy Transfer in Polystyrene-Based Arrays of π-Conjugated Chromophores. J. Phys. Chem. B 2014 , 118 , 372-378.
- (54) Fonseca Deichmann, V. A.; Yakutkin, V.; Baluschev, S.; Akcelrud, L. Optical Tuning of the Fluorescence Spectrum of a π-Conjugated Polymer through Excitation Power. J. Phys. Chem. B 2011 , 115 , 6385-6394.
- (55) Bhattacharyya, S.; Barman, M. K.; Baidya, A.; Patra, A. Singlet Oxygen Generation from Polymer Nanoparticles-Photosensitizer Conjugates using FRET Cascade. J. Phys. Chem. C 2014 , 118 , 9733-9740.
(56) Liu, Y.; Wang, H.; Dong, H.; Jiang, L.; Hu, W.; Zhan, X. High Performance Photoswitches Based on Flexible and Amorphous D-A Polymer Nanowires. Small 2013 , 9 , 294-299.
- (57) Karpfen, A.; Choi, C. H.; Kertesz, M. Single-Bond Torsional Potentials in Conjugated Systems: A Comparison of Ab Initio and Density Functional Results. J. Phys. Chem. A 1997 , 101 , 7426-7433.
- (58) Bré das, J. L.; Heeger, A. J. Theoretical Investigation of Gas-Phase Torsion Potentials Along Conjugated Polymer Backbones: Polyacetylene, Polydiacetylene, and Polythiophene. Macromolecules 1990 , 23 , 1150-1156.
- (59) Sun, G.; Kü rti, J.; Rajczy, P.; Kertesz, M.; Hafner, J.; Kresse, G. Performance of the Vienna Ab Initio Simulation Package (VASP) in Chemical Applications. J. Mol. Struct. : THEOCHEM 2003 , 624 , 37-45.
- (60) Abe, A.; Jernigan, R.; Flory, P. Conformational Energies of n-Alkanes and the Random Configuration of Higher Homologs Including Polymethylene. J. Am. Chem. Soc. 1966 , 88 , 631-639.
- (61) Sancho-Garcí, J. C.; Pé rez-Jimé nez, A. J.; Moscardó , F. Description of C (sp2)-C (sp2) a Rotation in Butadiene by Density Functionals. J. Phys. Chem. A 2001 , 105 , 11541-11548.
- (62) Sinnokrot, M. O.; Valeev, E. F.; Sherrill, C. D. Estimates of the Ab Initio Limit for π-π Interactions: The Benzene Dimer. J. Am. Chem. Soc. 2002 , 124 , 10887-10893.
- (63) Liu, J.; Yang, R. Length-Dependent Thermal Conductivity of Single Extended Polymer Chains. Phys. Rev. B 2012 , 86 , 104307.
- (64) Tamura, S.; Tanaka, Y.; Maris, H. J. Phonon Group Velocity and Thermal Conduction in Superlattices. Phys. Rev. B 1999 , 60 , 2627.
- (65) Shiomi, J.; Maruyama, S. Molecular Dynamics of Diffusive-Ballistic Heat Conduction in Single-Walled Carbon Nanotubes. Jpn. J. Appl. Phys. 2008 , 47 , 2005-2009.
- (66) Mu, X.; Wu, X.; Zhang, T.; Go, D. B.; Luo, T. Thermal Transport in Graphene OxideFrom Ballistic Extreme to Amorphous Limit. Sci. Rep. 2014 , 4 , 3909.
- (67) Zhang, M.; Efremov, M. Y.; Schiettekatte, F.; Olson, E. A.; Kwan, A. T.; Lai, S. L.; Wisleder, T.; Greene, J. E.; Allen, L. H. Size-Dependent Melting Point Depression of Nanostructures: Nanocalorimetric Measurements. Phys. Rev. B 2000 , 62 , 10548-10557.
- (68) Weber, T. A.; Helfand, E. Molecular Dynamics Simulation of Polymers. I. Structure. J. Chem. Phys 1979 , 71 , 4760-4762.
- (69) Macgowan, D.; Evans, D. J. A Comparison of NEMD Algorithms for Thermal Conductivity. Phys. Lett. A 1986 , 117 , 414-416.
- (70) Plimpton, S. Fast Parallel Algorithms for Short-Range Molecular Dynamics. J. Comput.
- Phys. 1995 , 117 , 1-19.
- (71) Takahashi, Y. Crystal Structure of Poly(Pyridobisimidazole), PIPD. Macromolecules 2003 , 36 , 8652-8655.
- (72) Baughman, R. H.; Kohler, B. E.; Levy, I. J.; Spangler, C. The Crystal Structure of Trans,Trans-1,3,5,7-Octatetraene as a Model for Fully-Ordered Trans-Polyacetylene. Synth. Met. 1985 , 11 , 37-52.
- (73) Corish, J.; Morton-Blake, D. A.; Veluri, K.; Bé niè re, F. Atomistic Simulations of the Structures of the Pristine and Doped Lattices of Polypyrrole and Polythiophene. J. Mol. Struct. : THEOCHEM 1993 , 283 , 121-134.
- (74) Ambrosch-Draxl, C.; Majewski, J. A.; Vogl, P.; Leising, G. First-Principles Studies of the Structural and Optical Properties of Crystalline Poly(Para-Phenylene). Phys. Rev. B 1995 , 51 , 9668-9676.
- (75) Tashiro, K.; Kobayashi, M.; Tadokoro, H. Elastic Moduli and Molecular Structures of several Crystalline Polymers, Including Aromatic Polyamides. Macromolecules 1977 , 10 , 413-420.
- (76) Eby, R. K.; Clark, E. S.; Farmer, B. L.; Piermarini, G. J.; Block, S. Crystal Structure of Poly(Tetrafluoroethylene) Homo- and Copolymers in the High Pressure Phase. Polymer 1990 , 31 , 2227-2237.
- (77) Song, H. H.; Hong, S. Preparation of Rigid-Rod Poly(p-Phenylene Benzobisthiazole) Films of Single Crystalline Texture. Polymer 1997 , 38 , 4241-4245.
(78) Satoh, M.; Yamasaki, H.; Aoki, S.; Yoshino, K. Crystal Structure of Drawn Polythiophene Film. Mol. Cryst. Liq. Cryst. 1988 , 159 , 289-295.
- (79) Ferro, D. R.; Porzio, W.; Destri, S.; Ragazzi, M.; Brü ckner, S. Application of Molecular Mechanics to Refine and Understand the Crystal Structure of Polythiophene and its Oligomers. Macromol. Theory Simul. 1997 , 6 , 713-727.
- (80) Ran, S.; Zong, X.; Fang, D.; Hsiao, B. S.; Chu, B.; Phillips, R. A. Structural and Morphological Studies of Isotactic Polypropylene Fibers during Heat/Draw Deformation by in-Situ Synchrotron SAXS/WAXD. Macromolecules 2001 , 34 , 2569-2578.
- (81) Clark, E. S. The Molecular Conformations of Polytetrafluoroethylene: Forms II and IV. Polymer 1999 , 40 , 4659-4665.
- (82) Davies, R. J.; Eichhorn, S. J.; Riekel, C.; Young, R. J. Crystal Lattice Deformation in Single Poly(p-Phenylene Benzobisoxazole) Fibres. Polymer 2004 , 45 , 7693-7704.
- (83) Chatzi, E. G.; Koenig, J. L. Morphology and Structure of Kevlar Fibers: A Review. Polym. Plast. Technol. Eng. 1987 , 26 , 229-270.
- (84) Itoh, T. Change with Temperature in Crystal Structures of Nylons 6, 66 and 610. Jpn. J. Appl. Phys. 1976 , 15 , 2295-2306.
Table of Contents
|
10.1021/jp5051639
|
[
"Teng Zhang",
"Xufei Wu",
"Tengfei Luo"
] |
2014-08-04T21:41:09+00:00
|
2014-08-04T21:41:09+00:00
|
[
"physics.chem-ph",
"cond-mat.mtrl-sci"
] |
Polymer Nanofibers with Outstanding Thermal Conductivity and Thermal Stability: Fundamental Linkage between Molecular Characteristics and Macroscopic Thermal Properties
|
Polymer nanofibers with high thermal conductivities and outstanding thermal
stabilities are highly desirable in heat transfer-critical applications such as
thermal management, heat exchangers and energy storage. In this work, we unlock
the fundamental relations between the thermal conductivity and thermal
stability of polymer nanofibers and their molecular characteristics by studying
the temperature-induced phase transitions and thermal transport of a series of
polymer nanofibers. Ten different polymer nanofibers with systematically chosen
molecular structures are studied using large scale molecular dynamics
simulations. We found that high thermal conductivity and good thermal stability
can be achieved in polymers with rigid backbones, exemplified by
{\pi}-conjugated polymers, due to suppressed segmental rotations and large
phonon group velocities. The low probability of segmental rotation does not
only prevent temperature-induced phase transition but also enables long phonon
mean free paths due to reduced disorder scattering. Although stronger
inter-chain interactions can also improve the thermal stability, polymers with
such a feature usually have heavier atoms, weaker backbone bonds, and segments
vulnerable to random rotations, which lead to low thermal conductivities. This
work elucidates the underlying linkage between the molecular nature and
macroscopic thermal properties of polymer nanofibers, which is instrumental to
the design of thermally conductive polymer nanofibers with high temperature
stabilities.
|
1408.0825v1
|
## Single-Photon Superradiance in Cold Atoms
Rafael A. de Oliveira , 1 Milrian S. Mendes , Weliton S. Martins , 1 1
Pablo L. Saldanha , 2 Jos´ W. R. Tabosa , and Daniel Felinto e 1 1
1 Departamento de F´ ısica, Universidade Federal de Pernambuco, 50670-901 Recife-PE, Brazil
2 Departamento de F´ ısica, Universidade Federal de Minas Gerais, 30161-970 Belo Horizonte-MG, Brazil
The interaction of an ensemble of atoms with common vacuum modes may lead to an enhanced emission into these modes. This phenomenon, known as superradiance, highlights the coherent nature of spontaneous emission, resulting in macroscopic entangled states in mundane situations. The complexity of the typical observations of superradiance, however, masks its quantum nature, allowing alternative classical interpretations. Here we stress how this picture changed with the implementation ten years ago of a new process for single-photon generation from atomic ensembles. We present then the last piece of evidence for the superradiant nature of such process, reporting the observation of an accelerated emission of the photon with a rate that may be tuned by controllably changing the number of atoms in the ensemble. Such investigation paves the way to a new, bottomup approach to the study of superradiance.
## I. INTRODUCTION
Sixty years ago, it was pointed out by Robert Dicke that the full quantum mechanical treatment of spontaneous emission from an ensemble of atoms could lead to enhanced, 'superradiant' emissions in particular modes [1]. Such effect would result in the medium spontaneously radiating in a burst much more directional and faster than if the atoms were emitting independently. This strong cooperativity is originated in the coherent nature of the interaction between atoms and vacuum and in the fact that some vacuum modes are coupled to the whole ensemble. Over time, superradiance has attracted great attention, since it may occur at common situations in many different systems and is the manifestation of a macroscopic entangled state spontaneously generated in the medium.
the Duan-Lukin-Cirac-Zoller (DLCZ) protocol. In the following, we highlight the impact the experimental implementation of the DLCZ protocol had on the study of superradiance, and report an investigation on the dynamics of such photon-pair generation that directly reveals its superradiant character. Our measurements of acceleration in the radiation process complement the known collective enhancement in directionality in the system. In this way, all key aspects of the phenomenology of superradiance have now been identified in connection to the DLCZ photon-pair-generation process, which can be employed henceforth as a framework for the systematic study of superradiance itself.
Even though, superradiance is still an effect hard to characterize and isolate. The interaction between particles and the induction of macroscopic polarizations in the sample, for example, may destroy or mask the observations of spontaneous cooperativity [2]. The first observations of strong superradiance, reported in the 1970s using extended ensembles [3, 4], were mixed with strong propagation and diffraction effects [2, 5]. On top of that, there is the question of the necessity or not of the full quantum mechanical treatment, and the entangled states that naturally comes with it, to understand the experimental results. It is clear that various aspects of superradiance have classical analogues [1-3], like the enhanced decay in a burst, which can be obtained from an ensemble of antennas emitting in phase.
However, a new process for generating photon pairs from atomic ensembles, proposed in 2001 [6] and implemented two years later [7], significantly moves away from the semiclassical views of superradiance and reinforces the central role of macroscopic entanglement for the understanding of the phenomenon and of its potential applications. This process was part of a broad protocol for quantum communication over long distances, known as
Below, in Sec. II we discuss the role of superradiance in the DLCZ protocol, as we introduce the basic process behind single-photon generation in our system. Section III presents our experimental setup, its characterization, and our first results related to superradiance. In this section, we measure the threshold optical depth of the atomic ensemble for superradiant behavior, as revealed by a sharp growth of the probability to detect the emitted single photon in the correct mode, combined with the appearance of nonclassical correlations in the system. We also demonstrate that such detection probability grows, at threshold, with the square of the number of atoms in the ensemble. Section IV focus on our observations of Rabi oscillations for the collective atomic mode and their comparison to an analytical theory for the reading process. This comparison is the basis to extract the decay time of the excited state modified by the condition of superradiance. In Sec. V we demonstrate experimentally, then, that such decay time is decreased as the number of atoms in the ensemble grows in the way expected for a superradiance process. Section VI provides finally our conclusions and perspectives on the subject.
## II. SUPERRADIANCE IN THE DLCZ PROTOCOL
In order to understand its classical analogy, one may view superradiance as a cascade of emissions starting at a state of maximum energy of an ensemble of twolevel atoms [1, 2], see Fig. 1a. Labeling | g 〉 and | e 〉 the ground and excited states, respectively, the initial state | e, e, e · · · e 〉 of the ensemble would carry then no coherence between the atoms. Once a first atom spontaneously decays and emits a photon in the common mode, the system is left in a large symmetrical collective state S {| g, e, e · · · e 〉} ∝ ∑ i | e, e · · · g i · · · e, e 〉 , for which an atom i decayed, but it is not known which one. This large entangled state has a fixed phase between its parts and some coherence, coming from the small probability each atom has of being in the ground state. As the cascade proceeds down the energy ladder, new symmetrical states are formed with more and more atoms in | g 〉 . The amount of coherence then grows in the sample. After many emissions, the system's behavior is dominated by its large coherence and hereafter may be approximated by its classical analogue. In the end, the ensemble is left in state | g, g, g · · · g 〉 , with again no coherence. In the picture presented above, which have been quite successful for explaining experimental observations [2, 3], the full quantum mechanical treatment is only required to describe the spontaneous trigger for an otherwise classical decay process.
FIG. 1: (a) Typical superradiant cascade in an ensemble of atoms with two levels, | g 〉 and | e 〉 . S {| · · · 〉} indicates symmetrical entangled states generated along the cascade. (b) Minimal superradiant cascade of the DLCZ protocol, involving atoms with three levels: | g 〉 , | s 〉 , | e 〉 . Spontaneous emissions on fields 1,2 are observed after excitation of the ensemble by write,read pulses, respectively. (c) Spatial configuration for fields in (b). Write and read are large beams counterpropagating to each other. Fields 1,2 are defined by the optical fibers (in blue) carrying them to detectors D D 1 , 2 . The detected modes, forming a small angle with the excitation fields, define the region in the atomic cloud that stores the collective entangled state. PF and FF denote polarization and frequency filters, respectively.
As for the DLCZ protocol, its building block is the generation of macroscopic entangled states that can be stored over long times. In order to do so, three-level atoms in Λ configuration are employed (Fig. 1b), with an extra ground state | s 〉 added to the above picture. All atoms are initially prepared in | g 〉 . The sample is then excited by a laser pulse ( write ) detuned from the transition | g 〉 → | e 〉 and, with small probability, a photon may be spontaneously emitted in the transition | e 〉 → | 〉 s , with an atom being simultaneously transferred to | s 〉 . If the photon is emitted in a mode ( field 1 ) common to the ensemble, its detection heralds the preparation of the symmetrical entangled state S {| s, g, g · · · g 〉} [6]. A second laser pulse ( read ) may now excite the transition | s 〉 → | 〉 e and, with high probability, a second photon ( field 2 ) is emitted in the transition | e 〉 → | 〉 g . In the end, all atoms are left again in | g 〉 . This last part of the process is the one connected to the usual phenomenology of superradiance, the high probability for the second emission coming from its strong directionality [6]. Differently from previous experiments in superradiance, however, a strong read pulse may 'open' the medium to the outgoing photon, using the effect known as Electromagnetically Induced Transparency (EIT) [8], reducing considerably distortions due to propagation of the photon through a thick, extended sample.
After the first implementations of this process [7, 9, 10], a major development was the introduction in 2005 of a Four-Wave-Mixing configuration (Fig. 1c) for the photon pair generation [11, 12], which solved most complications related to diffraction in the superradiant emission. The write pulse have here a considerably larger waist for its transversal mode than the one for field 1, which is fibercoupled and detected with a small angle to the direction of the write beam. In this case, the stored state with a single excitation in the ensemble would be given by | 1 at 〉 = ∑ i A g,g i | · · · s i · · · g 〉 , where A i gives the probability that the i th atom contributed to the detected mode [13]. In this way, the optical fiber for field 1 defines the spatial shape of the collective state stored in the ensemble. If the read field has also a large waist and is counterpropagating to the write beam, then field 2 is generated in the conjugated mode to field 1 [13]. The result is a superradiant emission of photon 2 in a well defined single mode, which can be coupled to an optical fiber with high efficiency [14]. An indirect observation of the superradiant increase of the spontaneous decay rate of the system was also reported in [13], as part of a detailed study of the saturation and spectrum of the readout process of field 2 under conditions of strong decoherence due to inhomogeneous magnetic fields acting on the atomic ensemble.
The overall process described above amounts then to a minimal superradiant cascade (Fig. 1b), in which a single photon is responsible for the preparation of the initial macroscopic entangled state that later results in the superradiant emission of another single photon. The fundamental importance of a minimal, single-photon su-
perradiance have been emphasized in recent years in a number of publications [15-17], and other experiments at the single-excitation level have been reported in studies of nuclear scattering of synchrotron radiation [18] and of two-photon cascade transitions in cold atoms [19]. The essential quantum-mechanical nature of the effect, in this case, can be directly apprehended from the interplay between its wave-like (collective interference) and particlelike (single-photon detection) aspects. The DLCZ protocol, however, adds to this picture the possibility to generate complex entangled states between different atomic ensembles [20-23] and even to explore these states for practical applications [6].
## III. NONLINEAR ENHANCEMENT AND THRESHOLD
In our experimental setup, the atomic ensemble is a cloud of cold cesium atoms obtained from a magnetooptical trap, with the trap laser tuned 15 MHz below the 6 S 1 / 2 ( F = 4) → 6 P 3 / 2 ( F ′ = 5) transition and the repumper laser resonant with the 6 S 1 / 2 ( F = 3) → 6 P 3 / 2 ( F ′ = 4) transition. Levels | g 〉 , | s 〉 and | e 〉 are given by the hyperfine states | 6 S 1 / 2 ( F = 4) , 〉 | 6 S 1 / 2 ( F = 3) 〉 and | 6 P 3 / 2 ( F ′ = 4) , respectively. 〉 The trap laser is kept on for 20 ms and, together with the trap's quadrupolar magnetic field, turned off for 2 ms, see Fig. 2. During this 2 ms period, the repumper laser is kept on for an extra 0.5 ms, in order to help preparing the atomic ensemble with all atoms initially at | g 〉 . The avalanche photodetectors (APDs) for the photons are then turned on for 1 ms in the last portion of the 2 ms interval. They have 45% detection efficiency for photons around 850 nm.
FIG. 2: Timing for first series of measurements on the threshold for superradiance.
During the 1 ms the APDs are on, a sequence of write and read pulses with durations of 50 and 180 ns, respectively, is sent every 5 µ s to excite the ensemble (Fig. 2). The exciting pulses are cut from cw diode lasers using acousto-optic modulators with ≈ 20 ns rise time. The read pulse arrives in the ensemble 50 ns after the end of the write pulse. The write field was weak and detuned 35 MHz below resonance, to avoid absorption in order to guarantee an uniform excitation of the ensemble. The read pulse was strong and tuned to resonance. In this way, it performed a double role, completely reading out the stored excitation and optically pumping the atoms back to | g 〉 . We employ the geometrical configuration of Fig. 1c, with 400 µ m and 200 µ m for the diameters of the transversal modes of write/read and field 1/field 2, respectively, and a 2 ◦ angle between them. The ensemble is about 3 mm long. Write and read fields have linear polarizations opposite to each other and to the respective photons they generate. Polarization filters were placed then in front of the fibers for fields 1,2 (Fig. 1c) to separate the photons from their respective excitation pulses. For field 1, a frequency filter is employed to eliminate photons from spontaneous decays in the transition | e 〉 → | g 〉 . This filter consists of an in-fiber Fabry-Perot with free spectral range of 20 GHz and linewidth of 400 MHz (FWHM).
In order to characterize the single-photon regime of field 2, both detectors in Fig. 1c were substituted by pairs of detectors connected to the output of fiber beamsplitters [13]. In this way, we are able to measure the integrated quantities P i and P ij giving, respectively, the probability of having single detections in field i and the probability of having joint detections in fields i, j . The total probability of having a detection in field 2 conditioned to one in field 1 is then given by P c = P 12 /P 1 . A direct observation of the onset of collective enhancement in the system as the number of atoms N increases is plotted in Fig. 3, through the dependence of P c (squares) with the sample's Optical Depth OD in the transition | g 〉 → | 〉 e , which is proportional to N .
OD was determined as in Ref. [13], and changed by tuning the power of the trap laser. Our standard way to measure the Optical Depth of the atomic ensemble, then, was to send through it a weak, long pulse (about 0.5µ s long) resonant to the | g 〉 → | 〉 e transition. Comparing the signal for the center of the pulse after the cell with ( V f ) and without ( V i ) the atomic cloud on the pathway, we could calculate directly the optical depth of the cloud through the expression OD = -ln( V /V f i ). We obtained the same results for OD if we tuned the pulse over the resonance and fitted the results with a Lorentzian profile. The measurement of OD was performed typically in the center of the interval the APDs were on (see Fig. 2), without write or read fields acting on the ensemble.
The single-photon character of field 2 was demonstrated by measuring a significant decrease of the quantity P cc = P 122 /P 1 with respect to what is expected for coherent fields. P 122 is the probability for a triple joint detection with two detections in field 2 following one in field 1. In this way, P cc is the conditional probability of having two detections in field 2 after one in field 1. Operationally, the single-photon character results in g c 2 = P cc /P 2 c < 1, with g c 2 the second order
auto-correlation function for the conditioned field 2. For the largest OD s in Fig. 3, we obtain g c 2 = 0 23 . ± 0 06. . The single-photon character may be indicated also by an indirect measurement, through the quantity P /P c 2 , with P 2 the unconditional probability to detect a photon in field 2 [14]. This quantity measures directly the correlation between the photons, since it quantifies how much a detection in field 1 increases the probability of obtaining another in field 2 in the same trial. For our system, P c > 2 P 2 indicates the presence of quantum correlations [7], and P c = P 2 indicates no correlations at all.
FIG. 3: Squares provide the total conditional probability P c for extracting the second photon, once the first was detected, as a function of the ensemble's optical depth. The dashed blue line connects the respective results for P 2 , the unconditional probability to detect a photon in field 2. The employed read pulse was 180 ns long. The inset provides a log-log graph for ∆ P c = P c -P 2 versus OD . The red line is a liner fit of slope s = 1 9 . ± 0 3, such that ∆ . P c ∝ OD s .
There are other straightforward ways to verify the single-photon character of field 2. For the DLCZ photonpair-generation process, the conditional preparation of a single stored excitation would lead to a P c value independent of P 1 [14]. In Fig. 4 we plot then P c versus P 1 for a variation of more than two orders of magnitude in P 1 . We can clearly distinguish region II as the singlephoton regime [14], where all the data in the manuscript was taken. Region I is dominated by the noise in field 1, which decreases P c due to spurious detections in field 1. Region III is the multi-photon regime, where more than one excitation are stored in the ensemble as a result of a strong write pulse.
The plateau on region II indicates ≈ 9% probability to detect a field-2 single photon. In order to obtain an estimation for the probability to extract the single photon from the ensemble in the correct mode, one has to estimate all losses in the detection channel from the face of the ensemble up to the detectors. In our case, we have an 8% loss associated to the non-coated windows of our vacuum chamber, a 45% detection efficiency, and about 55% of other losses in the transmission from the output of the vacuum chamber up to the face of the detectors. After all, we have a net efficiency of about 19% for transforming a single photon in the output of the ensemble into a click of our field-2 detector. The observed plateau at 9% indicates then a probability of ≈ 47% of extracting the single photon from the ensemble in the correct mode.
(%)
FIG. 4: Conditional P c and unconditional P 2 probability of detecting the second photon as a function of the probability P 1 of detecting the first photon.
As pointed out above, a strong indication of the nonclassical nature of the correlations between fields 1,2 is given by G 12 = P /P c 2 > 2. For this reason, all figures plotting P c also present the corresponding P 2 level, so one may check the condition G 12 > 2 for the measured data. A measurement of the dependence of g c 2 with G 12 is provided in [14] for experimental conditions close to ours. There, it is directly verified that typically g c 2 < 1 for G 12 > 2. In Fig. 3, the dashed blue line connects the measured values of P 2 , with the small error bars omitted for clarity, providing then the level to which P c should be compared in order to determine its degree of correlation to field 1. We observe a threshold to quantum correlations for OD ≈ 0 6. .
A log-log graph for the increment of the signal above its background (∆ P c = P c -P 2 ) is shown as an inset in Fig. 3. We note then that ∆ P c ∝ OD s ∝ N s with s = 1 9 . ± 0 3 around threshold, consistent with the expected . s = 2 for a typical superradiant signal. This fast growth of ∆ P c , however, decreases as P c enters the saturation region, where the probability to extract the single stored excitation becomes closer to one (as pointed out above P c ≈ 10% indicates a probability of ≈ 50% to extract the single photon from the ensemble in the correct mode).
## A. Decoherence
The quantities introduced above to characterize the correlation between fields 1,2 may also be used to measure the coherence time between levels | g 〉 and | s 〉 .Three pairs of bias coils in Helmholtz configuration are employed to cancel spurious DC magnetic fields after the trap magnetic field is turned OFF. After this cancelation, a coherence time of more than 500 ns was observed through the measurement of P c (squares) as a function
of the delay between write and read pulses (Fig. 5). This coherence time is much larger than the duration of the wavepacket of the second photon (as measured in the following sections), allowing us to neglect decoherence effects when comparing the experimental results to the theory. In Fig. 5, the blue line plots P 2 , so that we observe that the strong correlations between fields 1 and 2 also live for more than 500 ns.
6
(%)
FIG. 5: Conditional P c and unconditional P 2 probability of detecting the second photon as a function of the delay between write and read pulses.
## IV. COLLECTIVE RABI OSCILLATIONS
The wavepackets of photon 2 are obtained from the quantity p c ( ) = t p 12 ( ) t /P 1 , with p 12 ( ) t the joint probability of detecting an event in field 1 and another in field 2 in a time window ∆ t around t , with t = 0 the moment the read field is turned on. In this way, p c ( ) provides the cont ditional probability of detecting an event in field 2 around t , after an event in field 1 heralding the preparation of the proper collective state. From this definition, we have then P c = ∫ ∞ 0 p c ( ) t dt . The unconditional wavepackets of photon 2 are obtained from p 2 ( ), the probability to det tect an event in field 2 in a time window ∆ t around t . The time dependence of the correlations between fields 1,2 is given by g 12 ( ) t = p c ( ) t /p 2 ( ) t = p 12 ( ) t / p [ 2 ( ) t P 1 ], with g 12 ( ) t > 2 again indicating nonclassical correlations [25].
The wavepackets of photon 2 for OD ≈ 4 8, ∆ . t = 1 ns and various read powers are plotted in Fig. 6 (open circles). In order to decrease the number of free parameters in comparisons to the theory, we normalized p c ( ) by t P c for each curve. The time dependence for the correlations between fields 1,2 can be evaluated (similarly as for Fig. 3) by the ratio of the open circles to the dashed blue curves in Fig. 6 [25]. For these results on the field2 wavepackets, we introduced some modifications on the timing scheme of Fig. 2. The period in which the trials were taken was reduced from 1 ms to 0.5 ms, to improve the uniformity of OD throughout the magnetic-field-off period (see subsection below). The trial period was reduced from 5 µ s to 1 µ s to increase the experiment's repetition rate. Finally, the read pulse duration was increased from 180 ns to 840 ns, to guarantee the depletion of the | s 〉 level even for the lowest read powers employed in the measurements.
The theoretical expression for the wavepacket (solid red curves) can be directly obtained from [13] for a resonant read field in the limit of high intensity and negligible decoherence rate between the ground states:
$$\frac { p _ { c } ( t ) } { P _ { c } } = \frac { \chi \Gamma \Omega ^ { 2 } \Delta t \, e ^ { - \chi \Gamma t / 2 } } { \left ( \Omega ^ { 2 } - \frac { \chi ^ { 2 } \Gamma ^ { 2 } } { 4 } \right ) } \sin ^ { 2 } \left ( \sqrt { \Omega ^ { 2 } - \frac { \chi ^ { 2 } \Gamma ^ { 2 } } { 4 } } \, \frac { t } { 2 } \right ), \ \ ( 1 )$$
with Γ = 5 2 MHz the natural decay rate of the excited . state, Ω the Rabi frequency for the transition | s 〉 → | 〉 e excited by the read laser, and χ a 'cooperativity parameter' leading to the enhanced decay rates characteristic of superradiance. In [13], it was also shown that, for our experimental situation,
$$\chi = 1 + \frac { N } { w _ { 0 } ^ { 2 } k _ { 2 } ^ { 2 } }$$
with w 0 the waist of the photonic transversal mode, k 2 the modulus of its wavevector, and N the number of atoms that interact with this mode. Since both | s 〉 and | e 〉 have Zeeman sublevels, Ω represents only an effective Rabi frequency. As Ω 2 is proportional to the intensity of the read beam, one may write Ω = α √ P Γ, with P the read power and α a fit parameter. In this way, we are left with just two fit parameters for all curves: α and χ .
FIG. 6: Open circles provide the normalized conditional probability as a function of time for six different read powers (in mW): (a) 2.1, (b) 1.2, (c) 0.6, (d) 0.3, (e) 0.15, (f) 0.075. Solid red curves are the theoretical results of Eq. (1) with χ = 3 8 . and α = 9 0. . Dashed blue lines provide the corresponding results for the normalized unconditional probability p 2 ( ) t /P c .
The observed behavior in Fig. 6 for the highest powers can be described then as a damped, forced oscillation, with a straightforward physical interpretation from Eq. (1) [13]. The read beam induces transitions between levels | s 〉 and | e 〉 , forcing the system to perform Rabi oscillations between these levels. The frequency of these oscillations is determined mainly by the strength of the read field, represented by Ω, but also depends on χ Γ.
When an atom is in | e 〉 , it may spontaneously decay to | g 〉 emitting a photon. The spontaneous emission rate is increased due to constructive interference from the emission of different atoms, with such increase represented by the parameter χ . A direct consequence of the superradiant nature of the emission is then the observation of χ > 1. For comparison, the natural decay time of independent atoms for the coherence between excited and ground states is (Γ 2) / -1 ≈ 60 ns, while Fig. 6 shows a decay time of ( χ / Γ 2) -1 ≈ 16 ns. We observe a better agreement between theory and experiment for higher powers, since the theory in [13] was deduced assuming transparency of the sample to the outgoing photon, due to EIT. As power is reduced, the medium becomes more opaque to the photon and propagation effects related to its reabsorption cannot be neglected [13, 26].
## A. OD uniformity
In order to check for the uniformity of OD throughout the whole interval the magnetic field is off, we developed a different method to measure OD in a situation the closest possible to our actual experiment. It consisted in tuning the write field to resonance, considerably decreasing its power, and checking at each trial for the pulse shape after the cell with and without atomic cloud. The distorted pulse shape was then compared to the theoretical result for the propagation of an optical pulse of similar duration through an ensemble of two-level atoms with the same linewidth of the excited state in our experiment. OD was then obtained from the fit of the theory to the experimental results. An example of such measurement is shown in Fig. 7, for the timing employed in the measurements of Fig. 6. Figure 7 shows then the measurement of OD throughout the complete 0.5 ms interval in which the trials occur each time the magneto-optical trap is turned off. Each point in the graph is an average for the fit performed in four trials. In this way, we clearly observe no trend for change in the average value of OD during the period the trap is off and the measurements are performed. The average OD over the whole 0.5-ms interval measured this way, OD = 4 29 . ± 0 01, was also consistent . with the OD obtained from our standard method under the same conditions.
## V. VARIATION OF DECAY TIME WITH OPTICAL DEPTH
Differently from the case of independent atoms, the decay rate in superradiance may vary by tuning the number of atoms in the ensemble, as expected from Eq. (2). Our results for the wavepacket of photon 2 as OD is tuned are plotted in Fig. 8. The symbols and colors in this figure are the same as for Fig. 6. The read power is around 0 3 . mW. The theoretical plots were obtained from independent fittings of the experimental data to Eq. (1), using χ and α as fitting parameters. The values of χ and α obtained for each OD are plotted in Fig. 9. Since OD is proportional to N , following Eq. (2) we finally fit the data for χ versus OD with the curve χ = 1 + β OD , obtaining β = 0 53 . ± 0 02 (see . subsection below). We can see that the increase in the spontaneous decay rate of the atomic ensemble is proportional to its number of atoms, characterizing the superradiant nature of the observed single-photon emission. In Fig. 8 we were able to change the decay time from a minimum of 18 6 . ± 0 8 ns . in panel (a) to about 38 ± 3 ns in (f) (see inset in Fig. 9). As the number of atoms becomes too low, however, the visibility of the Rabi oscillations degrades due to the increase in the noise floor given by p 2 ( ) (dashed blue lines). t This is expected from the behavior of P c as OD decreases (Fig. 3), since it eventually reaches the noise floor given by P 2 once the collective enhancement is lost.
5.0
Optical Depth
FIG. 7: Measurement of optical depth for the outgoing photon throughout the whole 0.5-ms interval at which trials are taken with the trap off.
## A. β parameter
The optical depth can be written as OD = α l 0 , with α 0 the optical density of the sample and l its length. On the other hand, we have α 0 = σ N/V 0 , where σ 0 is the on-resonance scattering cross section of the atom and V is the sample's volume. In this way, we have OD = Nσ /πw 0 2 0 , with w 0 the waist of the photonic gaussian mode defining the ensemble's volume. The interpretation for this expression is straightforward, since each atom will scatter a portion σ /πw 0 2 0 of the incident beam resulting in the total thickness of the sample when the contributions of all atoms are combined.
The value β = 0 53 . ± 0 02 was obtained from a fit of the . function χ = 1 + βOD to the experimental data. From the above relation between OD and N and from Eq. (2), we obtain then
$$\beta = \frac { \pi } { \sigma _ { 0 } k ^ { 2 } }$$
with k the wave vector of field 2. The value of σ 0 can be obtained from σ 0 = glyph[planckover2pi1] ω Γ 2 / I sat , with ω the optical frequency of the transition and I sat its saturation intensity [28]. From this simple analysis, β would be given by
FIG. 8: Open circles provide the normalized conditional probability as a function of time for six different Optical Depths: (a) 4.8, (b) 4.0, (c) 3.4, (d) 2.6, (e) 1.6, (f) 1.0. Solid red curves are the theoretical results for independent fits using Eq. (1). The fit parameters χ and α for each OD are plotted in Fig. 9. Dashed blue lines provide the corresponding results for the normalized unconditional probability p 2 ( ) t /P c . The read power is 0.3 mW.
FIG. 9: Values of χ and α as a function of OD obtained from the fittings in Fig. 8. The red curve is a fit to a function with the same linear dependence as Eq. (2). The inset plot the decay times τ sp = ( χ Γ 2) / -1 .
- [1] R. H. Dicke, Rhys. Rev. 93 , 99 (1954).
- [2] M. Gross and S. Haroche, Phys. Rep. 93 , 301 (1982).
- [3] N. Skribanowitz, I. P. Herman, J. C. MacGillivray, and M. S. Feld, Phys. Rev. Lett. 30 , 309 (1973).
- [4] M. Gross, C. Fabre, P. Pillet, and S. Haroche, Phys. Rev. Lett. 36 , 1035 (1976).
- [5] J. C. MacGillivray and M. S. Feld, Phys. Rev. A 14 , 1169 (1976).
- [6] L.-M. Duan, M. D. Lukin, J. I. Cirac, and P. Zoller, Nature 414 , 413 (2001).
- [7] A. Kuzmich, W. P. Bowen, A. D. Boozer, A. Boca, C.
various well known parameters ( k ω w , , 0 ,Γ) plus the saturation intensity I sat , which will be given as an average over various transitions through different Zeeman sublevels. However, the observed value of β would imply a saturation intensity of I sat ≈ 3 5 mW/cm , a reasonable . 2 value for this specific transition of cesium [28].
## VI. CONCLUSIONS
We presented an investigation on the superradiant nature of the DLCZ photon-pair-generation process, demonstrating the observation of single-photon superradiance in the system. Various aspects of superradiance were then demonstrated in connection with the singlephoton emission. Particularly, we measured the threshold optical depth of the ensemble to start the superradiant process, and demonstrate that the probability to emit the single photon grows with the square of the number of atoms at threshold. We also measure the decrease of the decay time from the excited state, for single-photon emission, as the number of atoms is increased, following the expected behavior for superradiance. Key to our approach is a close comparison of our experimental data for the photonics wavepacket to an analytical theory for the reading process obtained from first principles [13].
Finally, the presented investigation opens the way for a systematic, bottom-up approach to the study of superradiance. The minimal superradiant cascade implemented up to now may be expanded using quantum-optics techniques of conditional generation of states with a larger number of excitations [27]. In this way, it should be possible to generate superradiant cascades with two or more excitations, controllably moving down the typical de-excitation ladder of macroscopic superradiance.
## ACKNOWLEDGEMENTS
We gratefully acknowledge K. N. Cassemiro for her assistance in part of the experiment. This work was supported by CNPq, CAPES, PRPq/UFMG, and FACEPE (Brazilian agencies), particularly through the programs PRONEX and INCT-IQ (Instituto Nacional de Ciˆ encia e Tecnologia de Informa¸ c˜o Quˆntica). a a
- W. Chou, L.-M. Duan, and H. J. Kimble, Nature 423 , 731 (2003).
- [8] K.-J. Boller, A. Imamoglu, and S. E. Harris, Phys. Rev. Lett. 66 , 2593 (1991).
- [9] C.W. Chou, S. V. Polyakov, A. Kuzmich, and H. J. Kimble, Phys. Rev. Lett. 92 , 213601 (2004).
- [10] M. D. Eisaman, L. Childress, A. Andr´ e, F. Massou, A. S. Zibrov, and M. D. Lukin, Phys. Rev. Lett. 93 , 233602 (2004).
- [11] V. Balic, D. A. Braje, P. Kolchin, G. Y. Yin, and S. E. Harris, Phys. Rev. Lett. 94 , 183601 (2005).
- [12] D. N. Matsukevich, T. Chaneli` ere, M. Bhattacharya, S.-Y. Lan, S. D. Jenkins, T. A. B. Kennedy, and A. Kuzmich, Phys. Rev. Lett. 95 , 040405 (2005).
- [13] M. S. Mendes, P. L. Saldanha, J. W. R. Tabosa, and D. Felinto, New J. Phys. 15 , 075030 (2013).
- [14] J. Laurat, H. de Riedmatten, D. Felinto, C.-W. Chou, E. W. Schomburg, and H. J. Kimble, Opt. Express 14 , 6912 (2006).
- [15] M. O. Scully, E. S. Fry, C. H. R. Ooi, and K. W´dkiewicz, o Phys. Rev. Lett. 96 , 010501 (2006).
- [16] M. O. Scully and A. A. Svidzinsky, Science 325 , 1510 (2009).
- [17] T. Bienaim´ e, R. Bachelard, N. Piovella, and Robin Kaiser, Fortschr. Phys. 61 , 377 (2013).
- [18] R. R¨ ohlsberger, K. Schlage, B. Sahoo, S. Couet, and R. R¨ uffer, Science 328 , 1248 (2010).
- [19] T. Chaneli`re, D. N. Matsukevich, S. D. Jenkins, T. A. e B. Kennedy, M. S. Chapman, and A. Kuzmich, Phys. Rev. Lett. 96 , 093604 (2006).
- [20] C. W. Chou, H. de Riedmatten, D. Felinto, S. V. Polyakov, S. J. van Enk, and H. J. Kimble, Nature 438 ,
- 828 (2005).
- [21] C. W. Chou, J. Laurat, H. Deng, K. S. Choi, H. de Riedmatten, D. Felinto, and H. J. Kimble, Science 316 , 1316 (2007).
- [22] Y.-A. Chen, X.-H. Bao, Z.-S. Yuan, S. Chen, B. Zhao, and J.-W. Pan, Phys. Rev. Lett. 104 , 043601 (2010).
- [23] K. S. Choi, A. Goban, S. B. Papp, S. J. van Enk, and H. J. Kimble, Nature 468 , 412 (2010).
- [24] S. V. Polyakov, C. W. Chou, D. Felinto, H. J. Kimble, 'Temporal Dynamics of Photon Pairs Generated by an Atomic Ensemble', Phys. Rev. Lett. 93 , 263601 (2004).
- [25] S. V. Polyakov, C. W. Chou, D. Felinto, and H. J. Kimble, Phys. Rev. Lett. 93 , 263601 (2004).
- [26] R. A. de Oliveira, D. Moretti, D. Felinto, and J. W. R. Tabosa, Phys. Rev. A 86 , 013839 (2012).
- [27] A. Ourjoumtsev, R. Tualle-Brouri, and P. Grangier, Phys. Rev. Lett. 96 , 213601 (2006)
- [28] D. A. Steck, Cesium D Line Data, http://steck.us/alkalidata
|
10.1103/PhysRevA.90.023848
|
[
"Rafael A. de Oliveira",
"Milrian S. Mendes",
"Weliton S. Martins",
"Pablo L. Saldanha",
"José W. R. Tabosa",
"Daniel Felinto"
] |
2014-08-04T21:44:06+00:00
|
2014-08-04T21:44:06+00:00
|
[
"quant-ph",
"physics.optics"
] |
Single-Photon Superradiance in Cold Atoms
|
The interaction of an ensemble of atoms with common vacuum modes may lead to
an enhanced emission into these modes. This phenomenon, known as superradiance,
highlights the coherent nature of spontaneous emission, resulting in
macroscopic entangled states in mundane situations. The complexity of the
typical observations of superradiance, however, masks its quantum nature,
allowing alternative classical interpretations. Here we stress how this picture
changed with the implementation ten years ago of a new process for
single-photon generation from atomic ensembles. We present then the last piece
of evidence for the superradiant nature of such process, reporting the
observation of an accelerated emission of the photon with a rate that may be
tuned by controllably changing the number of atoms in the ensemble. Such
investigation paves the way to a new, bottom-up approach to the study of
superradiance.
|
1408.0827v3
|
a
## Maximum Entropy Principle and the Higgs Boson Mass
Alexandre Alves a , Alex G. Dias , Roberto da Silva b c
Universidade Federal de S˜ ao Paulo, Departamento de Ciˆ encias Naturais e da Terra, Diadema - SP , Brasil b Universidade Federal do ABC, Centro de Ciˆ encias Naturais e Humanas, Santo Andr´ e - SP, Brasil c Universidade Federal do Rio Grande do Sul, Departamento de F´ ısica, Porto Alegre - RS, Brasil
## Abstract
A successful connection between Higgs boson decays and the Maximum Entropy Principle is presented. Based on the information theory inference approach we determine the Higgs boson mass as MH = 125 04 : GLYPH<6> 0 25 GeV, a value fully compatible to the LHC : measurement. This is straightforwardly obtained by taking the Higgs boson branching ratios as the target probability distributions of the inference, without any extra assumptions beyond the Standard Model. Yet, the principle can be a powerful tool in the construction of any model a GLYPH<11> ecting the Higgs sector. We give, as an example, the case where the Higgs boson has an extra invisible decay channel within a Higgs portal model.
Keywords: Higgs boson decays, maximum entropy principle, dark matter
## 1. Introduction
The mechanism of mass generation [1] via spontaneous electroweak symmetry breaking, as realized in the Standard Model (SM) [2], is essential to understand the electroweak interactions and has been almost entirely confirmed with the discovery of the Higgs boson and the observation of its couplings to some of the heavier particles of the SM spectrum [3]. In the form we know it, electroweak symmetry breaking is able to generate the particle masses but cannot fix their values. An exception is the photon which is massless due the unbroken U (1) gauge symmetry of electromagnetic interactions. Several attempts have been made to predict the Higgs boson mass from extensions of the SM, but until now there is no clue whether the best guesses are meant to represent a higher level of evidence for a more fundamental theory and consequently a deeper understanding of subatomic physics.
It has been recognized that entropy can be used as an universal tool for statistical inference in the information theory approach [4, 5, 6]. It is universal in the sense of its wide range of applicability and it is not necessarily tied to physical interpretations, despite the Gibbs-Shannon entropy is equivalent to the usual thermodynamical entropy [4].
In this Letter we show that the Maximum Entropy Principle (MEP) is a powerful inference tool which provides us with the most accurate theoretical determination to date of the Higgs mass. The principle also naturally leads us to assign a theoretical probability density function (PDF) for the Higgs boson mass parameter. It does not assume any beyond the Standard Model hypothesis nor untested physical principles, rather it does only assume the present status of physical knowledge encoded in SM.
Based on this successful mass parameter inference, we propose using MEP in order to furnish further pieces of evidence to favor or not a given theory a GLYPH<11> ecting the Higgs sector. Ac- cording to this proposal, any candidate model is required to maximize a physically well motivated entropy associated to the Higgs bosons decays to all available channels for a given mass. As an illustration of the usefulness of the proposed inference method, we investigate a Higgs-portal scalar dark matter model.
It has been firstly observed in Ref. [7] that the product of the SM Higgs branching ratios is a distribution with a peak close to the experimentally measured value of the Higgs boson mass. Moreover, the author also mentions a possible connection to this observation with entropy arguments in the Higgs decay process, and yielding a potential mean to constrain theoretical extensions of the SM. It was also suggested in Ref. [8] that some fundamental principle might be responsible for the maximization of Higgs branching ratio to photons, a fact that could be used for constraining new physics as well. We shall see that all those observations are consequences of MEP.
Our findings suggest further that entropy production in Higgs decays to massless quanta, namely, to photons and gluons, is maximized nearly the same point of the entropy from all channels. This places the system of a large number of Higgs bosons among the most fundamental ones where it is possible to observe MEP in action.
## 2. MEP inference on the Higgs boson mass
Maximum-entropy distributions are the best estimates that can be made from partial knowledge about the PDF of a given set of random variables. The goal of MEP inference is to determine the least-biased probability density function involved in the evolution of a given system from the computation of the Gibbs-Shannon entropy measure where the partial knowledge is coded as a list of constraints. It is universal in its scope and is known, for example, to lead to the same computation rules of the statistical mechanics, without any further assumptions beyond the usual laws of mechanics (quantum or classical) [4].
Suppose, however, we can compute a parametric family of PDFs from first principles (as the Higgs branching ratios). In this situation, the partial knowledge reflects all the prior information available encoded in the model (the SM, for example) from which one calculates those PDFs. The correspondence between MEP and the Maximum Likelihood Estimate (MLE) is particularly crucial for the inference of the PDF parameters [6].
We are going to show that a physically well motivated entropy measure for an ensemble of Higgs bosons naturally leads us to a MLE for the Higgs mass parameter and the assignment of a PDF for the Higgs mass as well. For that purpose, consider an ensemble of N non-interacting Higgs bosons which decay to SMparticles according to their respective branching ratios BRi , i = GLYPH<13>GLYPH<13>; gg ; Z GLYPH<13>; qq ¯ ; ' + ' GLYPH<0> , WW GLYPH<3> ; ZZ GLYPH<3> , where q = u d ; ; s ; c b t ; ; and ' = e ; GLYPH<22>; GLYPH<28> . There are 14 primary SM decay modes.
The probability that a system evolves from the N Higgs bosons to a bath of final state SM particles due m energetically accessible decay modes is given by a multinomial distribution
$$P ( \{ n _ { k } \} _ { k = 1 } ^ { m } ) \equiv \frac { N! } { n _ { 1 }! \cdots n _ { m }! } \prod _ { k = 1 } ^ { m } ( p _ { k } ) ^ { n _ { k } } \quad \quad ( 1 ) \quad \quad \text{(} \quad \text{of E} \\ \text{up to} \quad \quad \quad$$
where
$$\ a c = p _ { k } ( M _ { H } | \boldsymbol \theta ) \equiv B R _ { k } ( M _ { H }, \boldsymbol \theta ) = \frac { \Gamma _ { k } ( M _ { H }, \boldsymbol \theta ) } { \Gamma _ { H } ( M _ { H }, \boldsymbol \theta ) } \quad \ \ ( 2 )$$
are the Higgs branching ratios as a function of the Higgs boson mass MH and the remaining SM parameters GLYPH<18> related to the electroweak symmetry breaking. The partial decay width to the k -th primary channel is denoted as GLYPH<0> k , and GLYPH<0> H is the total width of the Higgs boson given by the sum GLYPH<0> H = P m k = 1 GLYPH<0> k . Form this definition, P m k = 1 pk = 1, and P m k = 1 nk = N where nk is the number of Higgs bosons decaying to the k th mode.
The partial decay widths of the Higgs boson can be easily computed at leading order in the Standard Model, however, a precise prediction must include higher order corrections of the perturbative series. At leading order, the partial widths are given by
$$\Gamma _ { f \bar { f } } \ = \ \frac { N _ { c } G _ { F } m _ { f } ^ { 2 } M _ { H } } { 4 \pi \sqrt { 2 } } \beta _ { f } ^ { 3 } \Theta ( M _ { H } - 2 m _ { f } ) \ = \ \ \ ( 3 ) \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
$$\Gamma _ { V V } \ & = \ \frac { \delta _ { V } G _ { F } M _ { H } ^ { 3 } } { 1 6 \pi \sqrt { 2 } } \beta _ { V } \left [ 1 - \frac { 4 m _ { V } ^ { 2 } } { M _ { H } ^ { 2 } } + \frac { 3 } { 4 } \left ( \frac { 4 m _ { V } ^ { 2 } } { M _ {$$
$$\Gamma _ { \gamma \gamma } \ = \ \frac { \stackrel { \cdot \cdot \cdot } { G _ { F } m _ { Z } ^ { 2 } } } { 2 \pi \sqrt { 2 } M _ { H } } F _ { \gamma \gamma } ( M _ { H }, \theta ) \quad \ \ ( 5 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
$$\Gamma _ { g g } \ = \ \frac { \alpha _ { s } ^ {$$
$$\Gamma _ { Z \gamma } \ = \ \frac { \dots \cdot$$
Here, Nc = 3(1) for quarks(leptons), GLYPH<14> V = 2(1) for W GLYPH<6> ( Z ), mf is the fermion mass, mV a W or Z mass and GF is the Fermi constant. The velocities GLYPH<12> are defined as GLYPH<12> f = q 1 GLYPH<0> 4 m f 2 = MH 2 , GLYPH<12> V =
$$\sqrt { 1 - 4 m _ { V } ^ { 2 } / M _ { H }$$
if x > 0, and zero otherwise. Higher order corrections and the loop functions of the partial widths to photons, gluons and Z + photon denoted as F GLYPH<13>GLYPH<13> , F GLYPH<13>GLYPH<13> , and FZ GLYPH<13> , respectively, in the formulas above, can be found in Ref. [9] and references therein.
We point out that higher order corrections include, beyond the leading 1 ! 2 Born amplitude, 1 ! 2 loop amplitudes, and 1 ! 3 amplitudes involving additional real emissions. This is the only way to consistently compute finite higher order corrections after renormalizing ultraviolet divergences.
The total entropy is now given by the sum of the multinomial distribution of N Higgs bosons decaying to each possible partition of n 1 particles of type 1, n 2 of type 2, and so on until the m -th mode
$$S _ { N } \ & = \ \sum _ { \{ n \} } ^ { N$$
where P N f n g ( GLYPH<15> ) GLYPH<17> P N n 1 = 0 GLYPH<1> GLYPH<1> GLYPH<1> P N nm = 0 ( GLYPH<15> ) GLYPH<2> GLYPH<14> GLYPH<16> N GLYPH<0> P m i = 1 ni GLYPH<17> .
The number of possible configurations involved in the sum of Eq. (8) is huge for large N . Recently, an asymptotic formula up to order 1 = N has been derived [10] and is given right below at Eq. (9)
$$\begin{array} {$$
We checked the formula up to N = 10 9 using an importance sampling Monte Carlo technique. In practice, for N glyph[greaterorsimilar] 10 7 , the O (1 = N ) term is negligible for the SM Higgs branching ratios pk .
An important remark about S N for the multinomial distribution is due. Calculating the entropy of a system with elementary states grouped into macrostates, with consequent loss of information, is the essence of coarse-graining [11]. The physically motivated entropy S N gathers each type of decay into subgroups counting their occurrence numbers and nothing more. Any additional information about their identities is lost. This way, S N naturally comprises the concept of irreversibility, in obeisance to the second law of thermodynamics [12].
The most important feature of the asymptotic formula of Eq. (9) is the term involving the product of probabilities pk in the thermodynamic limit N ! 1 . This is, up to a normalization factor, a joint PDF to observe a fraction p 1 of decays of type 1, and so on until a fraction pm to observe decays of type m . With fixed probabilities, the normalized product is the likelihood function of the parameter MH . This induces the assignment of the normalized product Q m k = 1 pk ( MH j GLYPH<18> ) to a theoretical Higgs mass PDF. Moreover, requiring that the Higgs mass parameter maximizes S N in the thermodynamic limit is equivalent to a MLE of the Higgs boson mass. In this respect, the physically motivated multinomial distribution is fundamental, once the asymptotic limit of its the entropy (Eq. 9) equals the logarithm of the likelihood function Q m k = 1 pk ( MH j GLYPH<18> ).
A maximum of S N (a concave function in MH ), in the thermodynamic limit, is a solution of the equation
$$\lim _ { N \to \infty } \frac { \partial S _ { N } } { \partial M _ { H } } = \frac { \partial S _ { \infty } } { \partial M _ { H } } = 0 \quad \quad ( 1 0 )$$
where
$$S _ { \infty } ( M _ { H } | \theta ) \equiv \ln \left ( \prod _ { k = 1 } ^ { m } p _ { k } ( M _ { H } | \theta ) \right ) \quad \quad ( 1 1 ) \quad \quad \quad$$
The mass parameter that maximizes S 1 ( MH j GLYPH<18> ), a solution of Eq. (10) that we call ˆ MH , is distributed according to some PDF PH ( ˆ MH ) which is obtained after marginalizing over all GLYPH<18> taking into account their current experimental uncertainties. This should not be confused with the theoretical Higgs mass PDF that Eq. (9) suggests. We calculate PH just to access the theoretical error on the inferred ˆ MH from uncertainties on the remaining SM parameters.
We adopted the SM parameters values, GLYPH<18> S M , recommended by the Higgs Working Group [13] and calculated the SM branching ratios with HDECAY [14] taking into account all the leading EW and QCD corrections available. We do not take double o GLYPH<11> -shell top quark decays into account as its contribution is expected to be a negligible next-to-leading order correction to W W + GLYPH<0> production in the mass region of interest [9], so we effectively have m = 13 decay modes. We checked our results against those quoted in the Higgs Working Group report within less than 1% for the whole mass range from 10 GeV to 1 TeV. The PH PDF was estimated by randomly choosing 2000 points in the SM parameter space, calculating ˆ MH for each point, and fitting the resulting histogram to a gaussian function.
After marginalizing over GLYPH<18> ,
$$P _ { H } ( \hat { M } _ { H } ) = \int _ { \Omega _ { S M } } P _ { H } ( \hat { M } _ { H } | \theta ) d \theta \quad \quad ( 1 2 ) \quad \text{Unc} \\ \quad \$$
where GLYPH<10> S M represents the SM ameters space, we obtained the results shown in Fig. (1). The solid black line ( PH ) is the distribution of the masses that lead to a maximum of S 1 of Eq. (11) while the dashed blue line represents the combination of the most recent CMS and ATLAS measurement of the Higgs mass. The theoretical determination based on MEP inference is
$$\hat { M } _ { H } = ( 1 2 5. 0 4 \pm 0. 2 5 ) \, \text{GeV} \quad \quad ( 1 3 ) \quad \text{uct}$$
to be compared with the latest Higgs mass measurements from CMS [15] and ATLAS [16]
$$M _ { H } \ & = \ \ ( 1 2 5. 0 3 \pm 0. 3 0 ) \, \text{GeV} \quad \text{CMS} \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
Acrude weighted-averaged combination of both results is MH = (125 14 : GLYPH<6> 0 24) GeV. :
In Fig. (1) we also show the evolution of the measured Higgs mass since its discovery in 2012 until its most recent determination. We observe an steady decrease in the value of the mass
124.0 124.5 125.0 125.5 126.0 MH @ GeV D Figure 1: Theoretical and experimental Higgs mass distributions. The solid black line represents the mass distribution obtained from the maximum of the Higgs entropy marginalized over GLYPH<18> , PH MH ( ˆ ). The dashed line is a crude combination of the most recent ATLAS + CMSresults on the Higgs mass. The vertical lines show the experimental evolution of MH towards the theoretical determination. The solid orange line is the Higgs mass PDF obtained from the maximum of the photons decay entropy also marginalized over GLYPH<18> , P GLYPH<13> ( ˆ MH ).
as the measurements get more accurate converging to the theoretical MEP determination. As far as we know, this is the first time an underlying principle is used to determine accurately the Higgs boson mass, without assuming any beyond the Standard Model scenario or untested physical assumptions.
As discussed above, we can interpret the joint PDF given by the normalized product of branching ratios as a theoretical Higgs mass PDF. What kind of distribution should we expect? Under weak conditions, which we checked that all the branching ratios fulfill, a normalized likelihood function L of a parameter GLYPH<24> asymptotically converges to the normal distribution
$$\i d e t r i _ { - } \, \int _ { \substack { \mathcal { N } \left ( \hat { \xi }, 1 / \sqrt { - L ^ { \prime \prime } ( \hat { \xi } ) } \right ) \, \int \, \text{the mass rehion around the maximum } \hat { \xi } } } \,.$$
in the limit of a large number of PDFs in the product [17]. The mean and the standard deviation (s.d.) of the normalized product of the Higgs branching ratios is given by ˆ MH = 125 04 GeV : and 6 13 GeV, respectively. :
In fact, the reduced number of branching ratios in the product leads to a strong departure from the gaussian regime as the Higgs mass gets away from the point of maximum as we can see in Fig. (2). Furthermore, S 1 , the mass dependent term of the multinomial entropy in the thermodynamic limit of Eq. (11), shows the remarkable feature of a fast decreasing in its magnitude at the production thresholds of W and Z pairs as displayed in the shaded red area of Fig. (2). This can be understood in terms of an energy reorganization. In the mass region before the WW and ZZ thresholds, the energy condensed in the Higgs mass is spread away by relativistic light particles, by the way, b -quarks are the heavier decay products of the Higgs in the region MH < 2 MW . If the Higgs mass was large enough to produce two W or Z on-shell, the branching ratios to WW and ZZ would rise steeply and dominate the decays. In that case, the energy stored in the Higgs mass would be just reorganized in
L
¨
H
100 120 140 160 MH @ GeV D Figure 2: The mass dependent term of the multinomial entropy in the thermodynamic limit, Eq. (11), as a function of the Higgs mass. The red shaded area emphasizes the threshold region of W and Z pair production where S 1 drops fast. The dashed line shows a gaussian function with mean 125 04 and s.d. : 6 13. :
their masses making a rapid fall of the entropy.
The nearly gaussian shape of Q m k = 1 pk ( MH ) near the maximum was firstly observed in Ref. [7] with very similar mean and standard deviation (see dashed line in Fig. (2)). It was argued that the maximum of this distribution (in the sense of MLE), which is close to the experimental mass value, could be due some kind of entropy argument and that the Higgs mass is placed in a window of 'maximum opportunity' for its experimental observation. In fact, an information theoretic interpretation can be o GLYPH<11> ered: the log-likelihood is the sum of the information content of the Higgs decays, defined by P m k = 1 GLYPH<0> ln( pk ), which is maximized precisely for the observed Higgs mass. An estimate of the Higgs mass based on the maximum of the photons branching ratio has also been presented in [8].
## 3. MEP inferences on a new physics scenario
Several beyond the Standard Model scenarios add particles the Higgs boson might decay into. Any channel representing a new Higgs decay might potentially move the Higgs mass that maximizes its entropy. In this respect, imposing that ˆ MH remains unaltered, or within certain confidence region around the measured Higgs mass, might narrow the parameters space of the model. Consider then a new channel whose partial width is GLYPH<0> GLYPH<31>GLYPH<31> , a function of MH , m GLYPH<31> the mass of the new particle, and GLYPH<21> its coupling to the Higgs boson.
The branching ratios after including the contribution of the new channel are given by
$$\tilde { p } _ { k } = \frac { p _ { k } } { 1 + p _ { \chi \chi } } \, \ \tilde { p } _ { \chi \chi } = \frac { p _ { \chi \chi } } { 1 + p _ { \chi \chi } } \quad \quad ( 1 5 ) \quad \text{the} \ ( \ )$$
where pk = GLYPH<0> k = GLYPH<0> S M and p GLYPH<31>GLYPH<31> = GLYPH<0> GLYPH<31>GLYPH<31> = GLYPH<0> S M .
Now, the mass dependent part of the entropy of the Higgs decays is calculated with these new branching ratios
$$\tilde { S } _ { \infty } ( M _ { H } | \tilde { \boldsymbol \theta } _ { S M }, m _ { \chi }, \lambda ) \ = \ \ln \left ( \tilde { p } _ { \chi \chi } \prod _ { k = 1 } ^ { m } \tilde { p } _ { k } \right ) \quad \quad ( 1 6 ) \\ = S _ { \infty } ( M _ { H } | \tilde { \boldsymbol \theta } _ { S M } ) \ + \ \ln \left ( \frac { p _ { \chi \chi } } { ( 1 + p _ { \chi \chi } ) ^ { m + 1 } } \right ) \\. \quad \tilde { \boldsymbol \theta } \ \dots \ \dots \ \dots \ \dots \ \dots$$
where S 1 ( MH j GLYPH<18> S M ) is given by Eq. (11).
Deriving both sides of this equation and requiring that dS ˜ dMH = 0 give us the following di GLYPH<11> erential equation
$$\frac { d S _ { \infty } } { d M _ { H } } + \left ( \frac { 1 } { p _ { \chi \chi } } - \frac { m + 1 } { 1 + p _ { \chi \chi } } \right ) \frac { d p _ { \chi \chi } } { d M _ { H } } = 0 \quad \quad ( 1 7 )$$
Now we can seek for solutions of this equation in a mass region around the experimental Higgs mass for a given point in the plane m versus GLYPH<31> GLYPH<21> .
Let us present now an specific example of application of MEPinference in a new physics scenario. A class of interesting models are those where the Higgs boson gains an extra invisible mode, in special, those ones where dark matter couples to the Higgs field as the Higgs-portal models [18]. In Ref. [19], several experimental constraints are used to determine a viable portion of the parameters space for a real scalar dark matter GLYPH<31> interacting with the SM Higgs boson doublet H according to L int = GLYPH<21> 2 GLYPH<31> 2 H H y . The dark matter mass m GLYPH<31> is constrained to lie in a small region of 55-62 GeV, and a coupling GLYPH<21> in the range 10 GLYPH<0> 2 -10 GLYPH<0> 3 in order to evade direct detection and relic abundance bounds plus the LHC bound on an invisible Higgs decay ( ˜ p GLYPH<31>GLYPH<31> = BRinv < 0 19). In order to place constraints on the : model, we mapped the allowed region found in Ref. [19] and searched for ˆ MH including the new dark matter decay channel p 14 = ˜ p GLYPH<31>GLYPH<31> = BR H ( ! GLYPH<31>GLYPH<31> ).
In Fig. (3) we show the region allowed by direct detection and relic density measurements inside the solid green quadrilateral [19], the shaded areas represent the Higgs branching ratio to scalar dark matter in this model, and region between the dashed red curves are those points respecting ˆ MH 2 [124 67 125 61] : ; : GeV, the 95% CL region of our combined LHC results of the Higgs mass. The yellow dotted lines represent the solution to the Eq. (17) for the central Higgs mass of 125 14 GeV. :
This result can be seen as prediction of MEP actually. If an invisible decay channel is found in the Run2 of LHC, for example, and an scalar dark matter is confirmed, the experimental result on the coupling to the Higgs and the dark matter mass can be compared to MEP inference. Moreover, as the Higgs mass measurement gets more accurate, the region between lines will shrink to a narrower area leading to a more sharply prediction. On the other, if MEP is correct, no scalar dark matter from a Higgs portal can exist inside the green quadrilateral but outside the dashed lines. We see, by the way, that nearly half the current allowed region would be already ruled out by MEP.
D
@
Figure 3: The allowed region by direct and indirect (relic density) as found in Ref. [19] is inside the green quadrilateral. Also shown are the isocontours of the Higgs branching ratio to dark matter. The points inside the region between the dashed red lines are those which maximize the Higgs decays entropy within the 95% CL experimental region. The yellow dotted lines represent the solution to the Eq. (17) for the central Higgs mass of 125 14 GeV. :
## 4. Entropy production in Higgs decays to massless quanta
From the thermodynamic point of view, each decay of the Higgs boson increases the entropy of the bath consisting of SM particles as the energy concentrated in the Higgs boson mass is dissipated into an increased number of quanta. In this respect, photons and gluons can be considered the more e GLYPH<14> -cient channels to increase entropy once they are massless particles [20, 21]. In Fig. (1) we show P GLYPH<13> ( ˆ MH ), obtained exactly as PH , but maximizing the Gibbs-Shannon entropy of the photons channel decay S GLYPH<13> . The maximum points of both PDFs are very close. In fact, the gluonic peak is also near, GLYPH<24> 119 GeV, which lead us to hypothesize that the massless channels play an important role in the explanation of why S H displays a maximumin first place. As a function of the Higgs mass, S H is small wherever the Higgs decays to photons and gluons are negligible and strongly increases around the peaks of p GLYPH<13>GLYPH<13> = BR H ( ! GLYPH<13>GLYPH<13> ) and pgg = BR H ( ! gg ). The creation of massive quanta in a co-moving volume of space is less e GLYPH<11> ective in increasing entropy as they organize part of the released energy in their own masses. This is, in part, why entropy is taken as the photon density when we consider the thermodynamic evolution of the Universe: radiation is the best way to spread energy across the space and increase entropy.
To confirm these intuitions, we calculated ln GLYPH<16> Q 11 k = 1 pk GLYPH<17> taking into account all channels but the massless ones (photons and gluons) multiplying the remaining branching ratios by 1 = (1 GLYPH<0> p GLYPH<13>GLYPH<13> GLYPH<0> pgg ). The maximum point of the total entropy barely changes to ˆ MH = 125 02 GeV, a negligible shift compared to : the value obtained with the full product.
Withdrawing just the gluonic decay channel, consistently changing the remaining ones, not only marginally changes the point of maximum, now at ˆ MH = 125 00 GeV, but it shortens : the distance to the point of maximum of the entropy created in photons decays which now is given by 124 85 GeV. :
From the observations made above we conclude that: (1) an experimentally compatible inference on the Higgs mass can be made considering only tree-level Higgs couplings; (2) entropy creation from decays to massless gauge bosons, especially the photon, is maximized close to the same Higgs mass of the full collectively determined entropy; (3) if photons were the only massless fields the Higgs could decay into, the agreement of the maximal points would improve considerably.
All these findings suggest that the decay of a system of scalar particles participating an electroweak symmetry breaking mechanism, as in SM, respects MEP concerning the scalars mass in two independent ways: collectively including all particles the Higgs couples to generate masses and individually for the massless gauge bosons related to the unbroken symmetries. Nevertheless, the interplay between massive and massless modes is necessary to accurately determine the Higgs mass.
## 5. Discussions and perspectives
The reason why the massless modes peak around the measured Higgs mass might be due the interplay of all the other SM parameters and perhaps by the structure of its symmetry group and the form of the Higgs sector [8]. Di GLYPH<11> erent parameters, for example, lead to another Higgs mass according to the MEP constraint. Removing a channel of the product of branching ratios may perturb ˆ MH . In particular, the light fermion decays, whose observation is beyond the LHC capabilities, cannot be neglected and the inference ˆ MH = 125 04 GeV would change a : lot if they were withdrawn.
From the point of view of the information theory, MEP inference applies to any new model a GLYPH<11> ecting the Higgs boson decays apart from physical considerations. Its scope of applications is wide, for example, extended Higgs sectors, beyond the Standard Model scenarios predicting new SM Higgs decays and or deviations of SM Higgs couplings, models for elec-/ troweak baryogenesis and Higgsogenesis, Higgs-inflaton proposals and new physics solutions to absolute stabilization of the vacuum. By the way, it is worthy mentioning that MEP should apply to any new Higgs boson from a new physics scenario. Requiring maximization of entropy might bring information about decay rates and its mass as we have delineated in this work.
Testing the validity of MEP in Higgs decays is also straightforward. Any new information that could be related to the Higgs sector can be checked to keep the maximum of S H compatible with the measured Higgs mass. We expect that as the experimental errors on the SM parameters get smaller, the gaussian PDFs of Fig. (1) get narrower with aligned means.
We conclude this Letter stating that the Maximum Entropy Principle seems to be an innate inference tool for a system composed of a large number of Higgs bosons undergoing a collective decay into SM particles. According to this interpretation, such a system can be placed amongst the most fundamental ones where MEP applies.
## Acknowledgments
This research was partially supported by the Conselho Nacional de Desenvolvimento Cient´ ıfico e Tecnol´gico (CNPq), o by grants 303094 2013-3 (A.G.D.) and 11862 2012-8 (R.S.). / / The work of A.A. and A.G.D. was supported by Fundac ¸˜ ao de Amparo ` a Pesquisa do Estado de S˜ ao Paulo (FAPESP), under the grant 2013 22079-8. We also want to thank Adriano Natale, / Yan Levin and David D'Enteria for helpful discussions and suggestions.
## References
## References
- [1] F. Englert and R. Brout, Phys. Rev. Lett. 13 , 321 (1964); P. W. Higgs, Phys. Lett. 12 (1964) 132; P. W. Higgs, Phys. Rev. Lett. 13 , 508 (1964); G. S. Guralnik, C. R. Hagen and T. W. B. Kibble, Phys. Rev. Lett. 13 (1964) 585; P. W. Higgs, Phys. Rev. 145 (1966) 1156; T. W. B. Kibble, Phys. Rev. 155 (1967) 1554.
- [2] S. Weinberg, Phys. Rev. Lett. 19 , 1264 (1967).
- [3] G. Aad et al. , Phys. Lett. B 716 , 1 (2012); S. Chatrchyan et al. , Phys. Lett. B 716 , 30 (2012).
- [4] E. T. Jaynes, Phys. Rev. 106 , 620 (1957).
- [5] P. F. Fougre (editor), Maximum Entropy and Bayesian Methods , Fundamental Theories of Physics Series, vol. 39, Kluwer Academic Publishers, 1989.
- [6] T. M. Cover and J. A. Thomas, Elements of Information Theory , Second Edition, Wiley-Interscience, 2006.
- [7] D. d'Enteria, arXiv:1208.1993 [hep-ph].
- [8] A. Alves, A. G. Dias, E. R. Barreto, arXiv:1312.5333 [hep-ph].
- [9] A. Djouadi, Phys. Rept. 459 , 1 (2008) [hep-ph 0503173]. /
- [10] J. Cichon and Z. Golebiewski, 23rd International Meeting on Probabilistic, Combinatorial and Asymptotic Methods for the Analysis of Algorithms , http: // luc.devroye.org AofA2012-CfP.html. /
- [11] A. Lesne, Mathematical Structures in Computer Science 24 , 4 (2014).
- [12] H. E. Kandrup, Phys. Rev. D 37 , 3505 (1988).
- [13] The Higgs Working Group, arXiv:1310.8361 [hep-ex].
- [14] A. Djouadi, J. Kalinowski and M. Spira, Comput. Phys. Commun. 108 , 56 (1998) [hep-ph 9704448]. /
- [15] The CMS Collaboration, CMS-PAS-HIG-14-009.
- [16] The ATLAS Collaboration, ATLAS-HIGGS-2013-12, arXiv:1406.3827 [hep-ex].
- [17] D. A. S. Fraser and P. McDunnough, The Canadian Journal of Statistics, 12 , 183 (1984).
- [18] A. Djouadi et al , Phys. Lett. B 709 , 65 (2012).
- [19] J. M. Cline, P. Scott, K. Kainulainen, and C. Weniger, Phys. Rev. D 88 , 055025 (2013). See Figure 6, left panel.
- [20] S. Frautschi, Science 217 , 593 (1982).
- [21] J. D. Bekenstein, Phys. Rev. 23 , 287 (1981).
|
10.1016/j.physa.2014.10.084
|
[
"Alexandre Alves",
"Alex G. Dias",
"Roberto da Silva"
] |
2014-08-04T21:53:32+00:00
|
2014-11-18T17:09:34+00:00
|
[
"hep-ph"
] |
Maximum Entropy Principle and the Higgs Boson Mass
|
A successful connection between Higgs boson decays and the Maximum Entropy
Principle is presented. Based on the information theory inference approach we
determine the Higgs boson mass as $M_H= 125.04\pm 0.25$ GeV, a value fully
compatible to the LHC measurement. This is straightforwardly obtained by taking
the Higgs boson branching ratios as the target probability distributions of the
inference, without any extra assumptions beyond the Standard Model. Yet, the
principle can be a powerful tool in the construction of any model affecting the
Higgs sector. We give, as an example, the case where the Higgs boson has an
extra invisible decay channel. Our findings suggest that a system of Higgs
bosons undergoing a collective decay to Standard Model particles is among the
most fundamental ones where the Maximum Entropy Principle applies.
|
1408.0828v1
|
## Pre-Reduction Graph Products: Hardnesses of Properly Learning DFAs and Approximating EDP on DAGs
Parinya Chalermsook ∗
Bundit Laekhanukit †
Danupon Nanongkai ‡
## Abstract
The study of graph products is a major research topic and typically concerns the term f ( G ∗ H ), e.g., to show that f ( G ∗ H ) = f ( G f H ) ( ). In this paper, we study graph products in a non-standard form f ( R G [ ∗ H ] where R is a 'reduction', a transformation of any graph into an instance of an intended optimization problem. We resolve some open problems as applications.
glyph[negationslash]
The first problem is minimum consistent deterministic finite automaton (DFA) . We show a tight n 1 -glyph[epsilon1] -approximation hardness, improving the n 1 / 14 -glyph[epsilon1] hardness of [Pitt and Warmuth, STOC 1989 and JACM 1993], where n is the sample size. (In fact, we also give improved hardnesses for the case of acyclic DFA and NFA.) Due to Board and Pitt [Theoretical Computer Science 1992], this implies the hardness of properly learning DFAs assuming NP = RP (the weakest possible assumption). This affirmatively answers an open problem raised 25 years ago in the paper of Pitt and Warmuth and the survey of Pitt [All 1989]. Prior to our results, this hardness only follows from the stronger hardness of improperly learning DFAs, which requires stronger assumptions, i.e., either a cryptographic or an average case complexity assumption [Kearns and Valiant STOC 1989 and J. ACM 1994; Daniely et al. STOC 2014]. The second problem is edge-disjoint paths (EDP) on directed acyclic graphs (DAGs). This problem admits an O ( √ n )-approximation algorithm [Chekuri, Khanna, and Shepherd, Theory of Computing 2006] and a matching Ω( √ n ) integrality gap, but so far only an n 1 / 26 -glyph[epsilon1] hardness factor is known [Chuzhoy et al., STOC 2007]. ( n denotes the number of vertices.) Our techniques give a tight n 1 / 2 -glyph[epsilon1] hardness for EDP on DAGs, thus resolving its approximability status.
As by-products of our techniques: (i) We give a tight hardness of packing vertex-disjoint k -cycles for large k , complimenting [Guruswami and Lee, ECCC 2014] and matching [Krivelevich et al., SODA 2005 and ACM Transactions on Algorithms 2007]. (ii) We give an alternative (and perhaps simpler) proof for the hardness of properly learning DNF, CNF and intersection of halfspaces [Alekhnovich et al., FOCS 2004 and J. Comput.Syst. Sci. 2008]. Our new concept reduces the task of proving hardnesses to merely analyzing graph product inequalities, which are often as simple as textbook exercises. This concept was inspired by, and can be viewed as a generalization of, the graph product subadditivity technique we previously introduced in SODA 2013. This more general concept might be useful in proving other hardness results as well.
∗ Max-Planck-Institut f¨r Informatik, Germany. u Work partially done while at IDSIA, Switzerland. Supported by the Swiss National Science Foundation project 200020 144491/1
† McGill University, Canada. Supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) grant no. 28833 and by Dr&Mrs M.Leong fellowship. Istituto Dalle Molle di Studi sull'Intelligenza Artificiale (IDSIA). Supported by ERC starting grant 279352 (NEWNET).
‡ University of Vienna, Austria. Work partially done while at Nanyang Technological University (NTU), Singapore, and ICERM, Brown University, USA.
## Contents
| 1 Introduction | 1 Introduction | 1 Introduction | 1 |
|------------------|-------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------|-----|
| | 1.1 . . . . . . . . . . . . . . . . . . . . . | The Concept of Pre-Reduction Graph Product | 1 |
| | 1.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . | Problems and Our Results . . . . . | 2 |
| 2 | Overview | | |
| | | | 6 |
| | 2.1 . . . . . . . . . . . . . . . . . . . . | Example of Reduction R : Vertex-Disjoint Paths . | 6 |
| | 2.2 . . . . . . . . . . . . . . . . . . . . . . . . . | The Use of Pre-Reduction Products . . | 7 |
| | 2.3 . . . . . . . . . | Toward A Tight Hardness of the Minimum Consistent DFA Problem | 8 |
| | 2.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . | Related Concept . . . . . . . . . . | 9 |
| | 2.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . | Organization . . . . . . . . . . . . | 10 |
| 3 | Preliminaries | | |
| | | | 10 |
| | 3.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . | Terms . . . . . . . . . . . . . . . . | 10 |
| | 3.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . | Problems . . . . . . . . . . . . . . | 11 |
| 4 | Meta | Theorems | 12 |
| | 4.1 | Proof of Theorem 4.1 (Meta Theorem for Maximization | |
| | . . . . . . . . . . | Problems) | 14 |
| | 4.2 . . . . . . . . . . | Proof of Theorem 4.2 (Meta Theorem for Minimization Problems) | 14 |
| | 4.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . | Overview of Applications . . . . . | 15 |
| 5 | Hardnesses of Finite Automata Problems: Minimum Consistency and Proper | | |
| | PAC | Learning | 15 |
| | 5.1 . . . . . . . . . . . . . . . . | Hardness of MinCon (ADFA, NFA) via Graph Products | 15 |
| | 5.2 . . . . . . . . . . . . . . . . . . . . . . | Tight Hardness for MinCon (ADFA, DFA) . | 19 |
| | 5.3 . . . . . . . . . . . . . . . . . . . . . . . . | Hardness of Proper PAC-Learning . . . . | 22 |
| | Hardness | | |
| 6 | | of EDP on DAGs | 23 |
| | 6.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . | Reduction R . . . . . . . . . . . . | 23 |
| | 6.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . | α -Projection Property . . . . . . . | 24 |
| | 6.3 . . . . . . . . . . . | Geometry of Paths: Regions, switching boxes, and configurations | 25 |
| | 6.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . | Proof . . . . . . . . . . . . . . . . . | 26 |
| | Other | Problems | |
| 7 | | | 28 |
| | 7.1 . . . . . . . . . . . . . . . . . . . . . . . . . | Hardness of k -Cycle Packing for Large k | 29 |
| | 7.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . | Learning CNF Formula . . . . . . | 30 |
| 8 | Conclusion and Open Problems | Conclusion and Open Problems | 31 |
| A | List of Bad Examples | List of Bad Examples | 35 |
| | A.1 Edge Disjoint Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . | A.1 Edge Disjoint Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . | 35 |
## 1 Introduction
## 1.1 The Concept of Pre-Reduction Graph Product
Background: Graph Product and Hardness of Approximation. Graph product is a fundamental tool with rich applications in both graph theory and theoretical computer science. It is, roughly speaking, a way to combine two graphs, say G and H , into a new graph denoted by G ∗ H . For example, the following lexicographic product , denoted by G · H , will be particularly useful in this paper.
$$( \text{Lexicographic Product} ) & \quad V ( G \cdot H ) = V ( G ) \times V ( H ) = \{ ( u, v ) \, \colon u \in V ( G ) \, \underset { \L } { \text{and} } \, v \in V ( H ) \}. \\ & E ( G \cdot H ) = \{ ( u, a ) ( v, b ) \, \colon u v \in E ( G ) \, \text{or} \, \left ( u = v \, \underset { \L } { \text{and} } \, a b \in E ( H ) \right ) \}. \quad ( 1 )$$
A common study of graph product aims at understanding how f ( G ∗ H ) behaves for some function f on graphs denoting a graph property. For example, if we let α G ( ) be the independence number of G (i.e., the cardinality of the maximum independent set), then α G ( · H ) = α G α H . ( ) ( )
Graph products have been extremely useful in boosting the hardness of approximation. One textbook example is proving the hardness of n glyph[epsilon1] for approximating the maximum independent set problem (i.e., approximating α G ( ) of an input graph G ): Berman and Schnitger [BS92] showed that we can reduce from Max 2SAT to get a constant approximation hardness c > 1 for the maximum independent set problem, and then use a graph product to boost the resulting hardness to n glyph[epsilon1] for some (small) constant glyph[epsilon1] . To illustrate how graph products amplify hardness, suppose we have a (1 001)-gap reduction . R I [ ] that transforms an instance I of SAT into a graph G . Since α ( ) · is multiplicative, if we take a product R I [ ] k for any integer k , the hardness gap immediately becomes (1 0001) . k = 2 Ω( ) k . Choosing k to be large enough gives 2 log 1 -glyph[epsilon1] n hardness. Therefore, once we can rule out the PTAS, graph products can be used to boost the hardness to almost polynomial. This idea is also used in many other problems, e.g., in proving the hardness of the longest path problem [KMR97].
Our Concept: Pre-Reduction Graph Product. This paper studies a reversed way to apply graph products: instead of the commonly used form of ( R I [ ]) k = ( R I [ ] ∗ R I [ ] ∗ . . . ) to boost the hardness of approximation, we will use R I [ k ] = R I [ ∗ I ∗ . . . ]; here, I is a graph which is an instance of a hard graph problem such as maximum independent set or minimum coloring. We refer to this approach as pre-reduction graph product to contrast the previous approach in which graph product is performed after a reduction (which will be referred to as post-reduction graph product). The main conceptual contribution of this paper is the demonstration to the power and versatility of this approach in proving approximation hardnesses. We show our results in Section 1.2 and will come back to explain this concept in more detail in Section 2.
We note one conceptual difference here between the previous post-reduction and our prereduction approaches: While the previous approach starts from a reduction R that already gives some hardness result, our approach usually starts from a reduction that does not immediately provide any hardness result; in other words, such reduction alone cannot be used to even prove NP-hardness. (See Section 2 for an illustration.) Moreover, in contrast to the previous use of ( R I [ ]) k which requires R I [ ] to be a graph, our approach allows us to prove hardnesses of problems whose input instances are not graphs. Also note that our approach gives rise to a study of graph products in a new form: in contrast to the usual study of f ( G ∗ H ), our hardness results crucially
Table 1: Summary of our hardness results.
| Problems | Upper Bounds | Prev. Hardness | New Hardness |
|-----------------------------------------------------------------------|-------------------------------------------------------|-------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------|
| MinCon( DFA , DFA ) EDP on DAGs k -cycle packing | O ( n ) ˜ O ( n 1 / 2 ) [CKS06] O (min( k, n 1 / 2 )) | n 1 / 14 - glyph[epsilon1] [PW93] n 1 / 26 - glyph[epsilon1] [CGKT07] Ω( k ) [GL14] | n 1 - glyph[epsilon1] n 1 / 2 - glyph[epsilon1] O (min( k, n 1 / 2 - glyph[epsilon1] )) |
| MinCon( CNF , CNF ), MinCon( DNF , DNF ), MinCon(Halfspace,Halfspace) | O ( n ) | n 1 - glyph[epsilon1] | n 1 - glyph[epsilon1] (Alternative proof) |
rely on understanding the behavior of f ( R G [ ∗ H ]) for some function f , reduction R , and graph product ∗ (which happens to always be the lexicographic product in this paper). Another feature of this approach is that it usually leads to simple proofs that do not require heavy machineries (such as the PCP-based construction) - some of our hardness proofs are arguably simplifications of the previous ones; in fact, most of our hardness results follow from the meta-theorem (see Section 4) which shows that a bounds of f ( R G [ ∗ H ]) in a certain form will immediately lead to hardness results. We list some bounds of f ( R G [ ∗ H ]) in Theorem 2.1.
## 1.2 Problems and Our Results
## 1.2.1 Minimum Consistent DFA and Proper PAC-Learning DFAs
In the minimum consistent deterministic finite automaton (DFA) problem, denoted by MinCon( DFA , DFA ) , we are given two sets P and N of positive and negative sample strings in { 0 1 , } ∗ . We let the sample size , denoted by n , be the total number of bits in all sample strings. Our goal is to construct a DFA M (see Section 3 for a definition) of minimum size that is consistent with all strings in P ∪ N . That is, M accepts all positive strings x ∈ P and rejects all negative strings y ∈ N .
This problem can be easily approximated within O n ( ). Due to its connections to PAC-learning automata and grammars (e.g. [DlH10, Pit89]), the problem has received a lot of attention from the late 70s to the early 90s. The NP-hardness of this problem was proved by Gold [Gol78] and Angluin [Ang78]. Li and Vazirani [LV88] later provided the first hardness of approximation result of (9 / 8 -glyph[epsilon1] ). This was greatly improved to n 1 14 / -glyph[epsilon1] by Pitt and Warmuth [PW93]. Our first result is a tight n 1 -glyph[epsilon1] hardness for this problem, improving [PW93]. In fact, our hardness result holds even when we allow an algorithm to compare its result to the optimal acyclic DFA (ADFA), which is larger than the optimal DFA. This problem is called MinCon( ADFA DFA , ) ; see Section 3 for detailed definitions.
Theorem 1.1. Given a pair of positive and negative samples ( P N , ) of size n where each sample has length O (log n ) , for any constant glyph[epsilon1] > 0 , it is NP-hard to distinguish between the following two cases of MinCon(ADFA,DFA):
- · Yes-Instance: There is an ADFA of size n glyph[epsilon1] consistent with ( P N , ) .
- · No-Instance: Any DFA that is consistent with ( P N , ) has size at least n 1 -glyph[epsilon1] .
In particular, it is NP-hard to approximate the minimum consistent DFA problem to within a factor of n 1 -glyph[epsilon1] .
The main motivation of this problem is its connection to the notion of properly PAC-learning DFAs. It is one of the most basic problems in the area of proper PAC-learning [DlH10, Pit89, PW93]. Roughly speaking, the problem is to learn an unknown DFA M from given random samples, where a learner is asked to output (based on such random samples) a DFA M ′ that closely approximates M (see, e.g., [Fel08] for details). The main question is whether DFA is properly PAC-learnable.
glyph[negationslash]
This question was the main motivation behind [PW93]; however, the n 1 14 / -glyph[epsilon1] hardness in [PW93] was not strong enough to prove this. Kearns and Valiant [KV94] showed that a proper PAC-learning of DFAs is not possible if we assume a cryptographic assumption stronger than P = NP . In fact, their result implies that even improperly PAC-learning DFAs (i.e., the output does not have to be a DFA) is impossible. Very recently, Daniely et al. [DLSS14] obtained a similar result by assuming a (fairly strong) average-case complexity assumption generalizing Feige's assumption [Fei02].
glyph[negationslash]
glyph[negationslash]
glyph[negationslash]
The question whether the cryptographic assumption could be replaced by the RP = NP assumption (which would be the weakest assumption possible) was asked 25 years ago in [Pit89, PW93]. In particular, the following is the first open problem in [Pit89]: (i) Can it be shown that DFAs are not properly PAC-learnable based only on the assumption that RP = NP ? (ii) Stronger still, can the improper learnability result of [KV94] be strengthened by replacing the cryptographic assumptions with only the assumption that RP = NP ?
glyph[negationslash]
Applebaum, Barak and Xiao [ABX08] showed that proving lower bounds for improper learning using many standard ways of reductions from NP-hard problems will not work unless the polynomial hierarchy collapses, suggesting that an answer to the second question is likely to be negative. For the first question, some hardnesses of proper PAC-learning assuming RP = NP were already known at the time (e.g. [PV88]) and there are many more recent results (see, e.g., [Fel08] and references therein). Despite this, the basic problem of learning DFAs (originally asked in the above question) has remained open. Theorem 1.1 together with a result of Board and Pitt [BP92] immediately resolve this problem.
## Corollary 1.2. Unless NP = RP , the class of DFAs is not properly PAC-learnable.
We also note an amusing connection between this type of result and Chomsky's 'Poverty of the Stimulus Argument', as noted by Aaronson [Aar08]: 'Let's say I give you a list of n -bit strings, and I tell you that there's some nondeterministic finite automaton M , with much fewer than n states, such that each string was produced by following a path in M . Given that information, can you reconstruct M (probably and approximately)? It's been proven that if you can, then you can also break RSA!' Our Corollary 1.2 implies that for the case of deterministic finite automaton, being able to reconstruct M will imply not only that one can break RSA but also solve, for instance, traveling salesman problem (TSP) probabilistically.
## 1.2.2 Edge-Disjoint Paths on DAGs
In the edge-disjoint paths problem ( EDP ) problem, we are given a graph G = ( V, E ) (which could be directed or undirected) and k source-sink pairs s t 1 1 , s 2 t 2 , . . . , s k t k (a pair can occur multiple times). The objective is to connect as many pairs as possible via edge-disjoint paths. Throughout, we let n and m be the number of vertices and edges in G , respectively. Approximating EDP has been extensively studied. It is one of the major challenges in the field of approximation algorithms. The problem has received significant attention from many groups of researchers, attacking the problem from many angles and considering a few variants and special cases (see, e.g., [RS95, Chu12, CL12, CKS09, CKS05, Kle05, KT98, KK10] and references therein).
Table 2: The current status of EDP .
| Cases | Upper Bounds | Integrality Gap | Prev. Hardness |
|------------|--------------------------------------------------|-------------------|------------------------------------------|
| Undirected | O ( n 1 / 2 ) [CKS06] | Ω( n 1 / 2 ) | log 1 / 2 - glyph[epsilon1] n [ACG + 10] |
| DAGs | ˜ O ( n 1 / 2 ) [CKS06] | Ω( n 1 / 2 ) | n 1 / 26 - glyph[epsilon1] [CGKT07] |
| Directed | O (min( m 1 / 2 , n 2 / 3 )) [Kle96, CK07, VV04] | Ω( n 1 / 2 ) | n 1 / 2 - glyph[epsilon1] [GKR + 03] |
glyph[negationslash]
In directed graphs, EDP can be approximated within a factor of O (min( m 1 2 / , n 2 3 / )) [Kle96, CK07, VV04]. The O m ( 1 2 / ) factor is tight on sparse graphs since directed EDP is NP-hard to approximate within a factor of n 1 2 / -glyph[epsilon1] , for any glyph[epsilon1] > 0 [GKR + 03]. In contrast to the directed case, undirected EDP is much less understood: The approximation factor for this case is O n ( 1 2 / ) [CKS06] with a matching integrality gap of Ω( n 1 2 / ) for its natural LP relaxation, suggesting an n 1 2 / -glyph[epsilon1] hardness. Despite these facts, we only know a log 1 2 / -glyph[epsilon1] n hardness of approximation assuming NP ⊆ ZPTIME( n polylog( n ) ). Even in special cases such as planar graphs (or, even simpler, brickwall graphs, a very structured subclass of planar graphs), it is still open whether undirected EDP admits an o n ( 1 2 / ) approximation algorithm. This obscure state of the art made undirected EDP one of the most important, intriguing open problems in graph routing. (Table 2 summarizes the current status of EDP .)
One problem that may help in understanding undirected EDP is perhaps EDP on directed acyclic graphs (DAGs). This case is interesting because (i) its complexity seems to lie somewhere between the directed and undirected cases, (ii) it shares some similar statuses and structures with undirected EDP , and (iii) it has close connections to directed cycle packing [KNS + 07] (i.e. hard instances for EDP on DAGs are used as a gadget in constructing the hard instance for directed cycle packing). In particular, on the upper bound side, the technique in [CKS06] gives an O n ( 1 2 / poly log n ) upper bound not only to undirected EDP but also to EDP on DAGs. Moreover, the integrality gap of Ω( n 1 2 / ) applies to both cases, suggesting a hardness of n 1 2 / -glyph[epsilon1] for them. However, previous hardness techniques for the case of general directed graphs [GKR + 03] completely fail to give a lower bound on both DAGs and undirected graphs . On the other hand, subsequent techniques that were invented 1 in [AZ06, ACG + 10] to deal with undirected EDP can be strengthened to prove the currently best hardness for DAGs [CGKT07] , which is 2 n 1 26 / -glyph[epsilon1] . These results suggest that the complexity of DAGs lies between undirected and directed graphs. In this paper, we show that our techniques give a hardness of n 1 2 / -glyph[epsilon1] for this case, thus completely settling its approximability status. Our result is formally stated in the following theorem.
Theorem 1.3. Given an instance of EDP on DAGs, consisting of a graph G = ( V, E ) on n vertices and a source-sink pairs ( s , t 1 1 ) , . . . , ( s k , t k ) , for any glyph[epsilon1] > 0 , it is NP-hard to distinguish between the following two cases:
- · Yes-Instance: There is a collection of edge disjoint paths in G that connects 1 /n glyph[epsilon1] fraction of the source-sink pairs.
- · No-Instance: Any collection of edge disjoint paths in G connects at most 1 /n 1 2 / -glyph[epsilon1] fraction of the source-sink pairs.
1 The result in [GKR + 03] crucially relies on the fact that EDP with 2 terminal pairs is hard on directed graphs. This is not true if the graph is a DAG or undirected.
2 Their result is in fact proved in a more general setting of EDP with congestion c for any c ≥ 1
In particular, it is NP-hard to approximate EDP on DAGs to within a factor of n 1 2 / -glyph[epsilon1] .
## 1.2.3 Other Results
Minimum Consistent NFA. Our techniques also allow us to prove a hardness result for the minimum consistent NFA problem as stated formally in the following theorem.
Theorem 1.4. Given a pair of positive and negative samples ( P N , ) of size n where each sample has length O (log n ) , for any constant glyph[epsilon1] > 0 , it is NP-hard to distinguish between the following two cases of MinCon(ADFA,NFA):
- · Yes-Instance: There is an ADFA of size n glyph[epsilon1] consistent with ( P N , ) .
- · No-Instance: Any NFA that is consistent with ( P N , ) has size at least n 1 2 / -glyph[epsilon1] .
In particular, it is NP-hard to approximate the minimum consistent NFA problem to within a factor of n 1 2 / -glyph[epsilon1] .
This improves upon the n 1 14 / -glyph[epsilon1] hardness of Pitt and Warmuth [PW93]. We note that this hardness result is not strong enough to imply a PAC-learning lower bound for NFAs. Such hardness was already known based on some cryptographic or average-case complexity assumptions [KV94, DLSS14]. We think it is an interesting open problem to remove these assumptions as we did for the case of learning DFAs.
k -Cycle Packing. Our reduction for EDP can be slightly modified to obtain hardness results for k -Cycle Packing, when k is large. In the k -cycle packing problem, given an input graph G , one wants to pack as many disjoint cycles as possible into the graph while we are only interested in cycles of length at most k . An O (min( k, n 1 2 / ))-approximation algorithm for this problem can be easily obtained by modifying the algorithm of Krivelevich et al. [KNS + 07]). Very recently, Guruswami and Lee [GL14] obtained a hardness of Ω( k ), assuming the Unique Game Conjecture, when k is a constant. This matches the upper bound of Krivelevich et al. for small k . In this paper, we compliment the result of Guruswami and Lee by showing a hardness of n 1 2 / -glyph[epsilon1] for some k ≥ n 1 2 / , matching the upper bound of Krivelevich et al. for the case of large k .
Theorem 1.5. Given a directed graph G , for any glyph[epsilon1] > 0 and some k ≥ | V ( G ) | 1 2 / , it is NP-hard to distinguish between the following cases:
- · There are at least | V ( G ) | 1 2 / -glyph[epsilon1] disjoint cycles of length k in G .
- · There are at most | V ( G ) | glyph[epsilon1] disjoint cycles of length at most 2 k -1 in G .
In particular, for some k ≥ n 1 2 / , the k -cycle packing problem on n -vertex graphs is hard to approximate to within a factor of n 1 2 / -glyph[epsilon1] .
Alternative Hardness Proof for Minimum Consistent CNF, DNF, and Intersections of Halfspaces. Our techniques for proving the DFA hardness result can be used to give an alternative proof for the hardness of the minimum consistent DNF, CNF, and intersections of thresholded halfspaces problems. In the minimum consistent CNF problem, we are given a collection of samples of size n , and our goal is to output a small CNF formula that is consistent with all such
samples. Alekhnovich et al. [ABF + 08] previously showed tight hardnesses for these problems, which imply that the classes of CNFs, DNFs, and the intersections of halfspaces are not properly PAClearnable. Our techniques give an alternative proof (which might be simpler) for these results. More specifically, we give an alternative proof for the following theorem and corollary (stated in terms of CNF, but the same holds for DNF and intersection of halfspaces ). 3
Theorem 1.6. Let glyph[epsilon1] > 0 be any constant. Given a pair of positive an negative samples ( P N , ) of size n where each sample has length at most n glyph[epsilon1] , it is NP-hard to distinguish between the following two cases:
- · Yes-Instance : There is a CNF formula of size n glyph[epsilon1] consistent with ( P N , ) .
- · No-Instance : Any CNF consistent with ( P N , ) must have size at least n 1 -glyph[epsilon1] .
In particular, it is NP-hard to approximate the minimum consistent CNF problem to within a factor of n 1 -glyph[epsilon1] .
Corollary 1.7. Unless NP = RP , the class of CNF is not properly PAC-learnable.
## 2 Overview
## 2.1 Example of Reduction R : Vertex-Disjoint Paths
To illustrate the pre-reduction graph product concept, consider the vertex-disjoint path ( VDP ) problem. The objective of VDP is the same as that of EDP except that we want paths to be vertex-disjoint instead of edge-disjoint. The approximability statuses of EDP and VDP on DAGs and undirected graphs are the same, and we choose to present VDP due to its simpler gadget construction. Our hardness of VDP can be easily turned into a hardness of EDP .
Our goal is to show that this problem has an approximation hardness of n 1 2 / -glyph[epsilon1] , where n is the number of vertices. We will use the following reduction 4 R which transforms a graph G (supposedly an input instance of the maximum independent set problem) into an instance R G [ ] of the vertexdisjoint paths problem with Θ( | V ( G ) | 2 ) vertices. We start with an instance R G [ ] as in Figure 1a where there are k source-sink pairs (Figure 1a shows an example where k = 6) and edges are oriented from left to right and from top to bottom. Let us name vertices in G by 1, 2, . . . , k . For any pair of vertices i and j , where i < j , such that edge ij does not present in G , we remove a vertex v ij from R G [ ], as shown in Figure 1b (this means that two edges that point to v ij will continue on their directions without intersecting each other). See Section 6 for the full description of R in the context of EDP .
To see an intuition of this reduction, define a canonical path be a path that starts at some source s i , goes all the way right, and then goes all the way down to t i (e.g., a thick (green) path in Figure 1b). It can be easily seen that any set of vertex-disjoint paths in R G [ ] that consists only of canonical paths can be converted to a solution for the maximum independent set problem. Conversely any independent set S in G can be converted to a set of | S | vertex-disjoint paths. For example, canonical paths between the pairs ( s , t 1 1 ) and ( s , t 2 2 ) in R G [ ] in Figure 1b can be converted
3 +
It is noted in [ABF 08] that one only needs to prove the hardness of CNF, since this problem is a special case of the intersection of thresholded halfspaces problem, and the proof for DNF would work similarly.
4 We thank Julia Chuzhoy who suggested this reduction to us (private communication).
Figure 1: The reduction R for the vertex-disjoint paths problem. The thick (green) path in Figure 1b shows an example of a canonical path.
to an independent set { 1 2 , } in G and vice versa. In other words, if we can force the VDP solution to consist only of canonical paths, then we can potentially use the | V | 1 -glyph[epsilon1] hardness of maximum independent set to prove a tight | V | 1 -glyph[epsilon1] = | V ( R G [ ]) | 1 2 / -glyph[epsilon1] hardness of VDP . This intuition, however, cannot be easily turned into a hardness result since the VDP solution can use non-canonical paths, and it is possible that VDP ( R G [ ]) is much larger than α G ( ); see Section A.1 for an example where α G ( ) = O (1) and VDP ( R G [ ]) = Ω( | V ( G ) ). | Thus, the reduction R by itself cannot be used even to prove that VDP is NP-hard!
## 2.2 The Use of Pre-Reduction Products
The above situation is very common in attempts to prove hardnesses for various problems. A usual way to obtain hardness results is to modify R into some reduction R ′ . This modification, however, often blows up the size of the reduction, thus affecting its tightness. For example, VDP and EDP on DAGs are only known to be n 1 26 / -glyph[epsilon1] -hard, as opposed to being potentially n 1 2 / -glyph[epsilon1] -hard, as suggested by the integrality gap. Moreover, the reduction R ′ is usually much more complicated than R . In this paper, we show that for many problems the above difficulties can be avoided by simply picking an appropriate graph product ∗ and understanding the structure of R G [ ∗ G ∗ . . . ]. To this end, it is sometimes easier to study f ( R G [ ∗ H ]) for any graphs G and H , although we eventually need only the case where G = H . This gives rise to the study of the behavior of f ( R G [ ∗ H ]) which is a non-standard form of graph product in comparison with the standard study of f ( G ∗ H ). In fact, most results in this paper follow merely from bounding f ( R G [ ∗ H ]) in the form
$$g ( G * H ) \leq f ( R [ G * H ] ) \leq g ( G ) f ( H ) + \text{poly} ( | V ( G ) | ),$$
where g is an objective function of a problem whose hardness is already known (in this paper, g is either maximum independent set or minimum coloring), and f is an objective function of a problem that we intend to prove hardness. Our bounds for functions f corresponding to problems that we want to solve, e.g. the minimum consistent DFA (function dfa ) and maximum edge-disjoint paths (function edp ), are listed in the theorem below. (Recall that G H · denotes the lexicographic product as defined in Eq. 1.)
Theorem 2.1 (Bounds of graph products; informal) . There is a reduction R 1 (respectively R 2 ) that transforms a graph G into an instance of the minimum consistent DFA problem of size ˜ ( Θ | V ( G ) | 2 ) (respectively the maximum edge-disjoint paths problem of size Θ( | V ( G ) | 2 ) ) such that, for any graphs G and H ,
$$\chi ( G \cdot H ) \leq d \text{fa} ( R _ { 1 } [ G \cdot H ] ) \leq \chi ( G ) \text{d} \text{fa} ( R _ { 1 } [ H ] ) + O ( | V ( G ) | ^ { 2 } )$$
$$\alpha ( G \cdot H ) \leq \text{edp} ( R _ { 2 } [ G \cdot H ] ) \leq \alpha ( G ) \text{edp} ( R _ { 2 } [ H ] ) + O ( | V ( G ) | ^ { 2 } )$$
See Section 5.1 (especially, Corollary 5.3 and Lemma 5.4) and Section 6 (especially, Lemma 6.3) for the details and proofs of Eq. 3 and Eq. 4, respectively. It only requires a systematic, simple calculation to show that these inequalities imply hardnesses of approximation; we formulate this implication as a 'meta theorem' (see Section 4) which roughly states that for large enough k ,
$$f ( R [ G ^ { k } ] ) \approx g ( G ^ { k } )$$
where G k is G ∗ G ∗ G ∗ . . . ( k times). (For an intuition, observe that when k is large enough, the term poly( | V ( G ) ) in Eq. 2 will be negligible and an inductive argument can be used to show that | g G ( k ) ≤ f ( R G [ k ]) ≤ g G ( ) O k ( ) (recall that, in our case, g is multiplicative)). This means that the hardness of f 5 is at least the same as the hardness of g on graph product instances G k . For the case of DFA and EDP , R 1 ( G ) and R 2 ( G ) increase the size of input size to | V ( G ) | 2 while α and χ have the hardness of | V ( G ) | 1 -glyph[epsilon1] . Thus, we get a hardness of n 1 2 / -glyph[epsilon1] where n is the input size of DFA and EDP . This immediately implies a tight hardness for EDP and an improved hardness of DFA . How this translates to a hardness of f depends on how much instance blowup the reduction R G [ k ] causes. For our problems of DFA and EDP , it is a well known result that the hardness of α and χ stays roughly the same under the lexicographic product, i.e., α and χ on G k have a hardness of | V ( G k ) | 1 -glyph[epsilon1] . The meta theorem and Theorem 2.1 say that this hardness also holds for DFA and EDP . Since R G 1 [ k ] and R G 2 [ k ] increase the size of input instances by a quadratic factor - from | V ( G k ) | to n = | V ( G k ) | 2 -we get a hardness of n 1 2 / -glyph[epsilon1] where n is the input size of DFA and EDP . This immediately implies a tight hardness for EDP and an improved hardness for DFA .
## 2.3 Toward A Tight Hardness of the Minimum Consistent DFA Problem
To get the tight n 1 -glyph[epsilon1] hardness for DFA , we have to adjust R 1 in Theorem 2.1 to avoid the quadratic blowup. We will exploit the fact that, to get a result similar to Eq. 5, we only need a reduction R defined on the k -fold graph product G k instead of on an arbitrary graph G as in the case of R 1 . We modify reduction R 1 to R 1 ,k that works only on an input graph in the form G k and produces an instance R 1 ,k [ G k ] of size almost linear in | V ( G k ) | while inequalities as in Theorem 2.1 still hold, and obtain the following.
Lemma 2.2. For any k , there is a reduction R 1 ,k that reduces a graph G k = G · G · . . . into an instance of the minimum consistent DFA problem of size O k ( · | V ( G k ) | · | V ( G ) | 2 ) such that
$$\chi ( G ) ^ { k } \leq \text{dfa} ( R _ { 1, k } [ G ^ { k } ] ) \leq \chi ( G ) ^ { 2 k } | V ( G ) | ^ { 4 }$$
The description of reduction R 1 ,k and the proof of Theorem 2.2 can be found in Section 5.2. Observe that the size O k ( · | V ( G k ) | · | V ( G ) | 2 ) of R 1 ,k ( G k ) is almost linear (almost O V G ( | ( k ) )) |
5 For conciseness, we will use g and f to refer to problems and their objective functions interchangeably.
as the extra O k V ( | ( G ) | 2 ) is negligible when k is sufficiently large. Similarly, the term | V ( G ) | 4 in Eq. 6 is negligible and thus the value of dfa ( R 1 ,k [ G k ]) is sandwiched by χ G ( ) k and χ G ( ) 2 k . This means that if χ G ( ) is small (i.e., χ G ( ) ≤ | V ( G ) | glyph[epsilon1] ), then dfa ( R 1 ,k [ G k ]) will be small (i.e., dfa ( R 1 ,k [ G k ]) ≤ | V ( G k ) | 2 glyph[epsilon1] ), and if χ G ( ) is large (i.e., χ G ( ) ≥ | V ( G ) | 1 -glyph[epsilon1] ), then dfa ( R 1 ,k [ G k ]) will be also large (i.e., dfa ( R 1 ,k [ G k ]) ≥ | V ( G k ) | 1 -glyph[epsilon1] ). The hardness of n 1 -glyph[epsilon1] for DFA thus follows.
We note that in Theorem 2.1, we can replace DFA by NFA , a function corresponds to the minimum consistent NFA problem, thus getting a hardness of n 1 2 / -glyph[epsilon1] for this problem as well. This is, however, not yet tight. We would get a tight hardness if we can replace DFA by NFA in Theorem 2.2, which is not the case. We also note that the proof for the tight hardness for the minimum consistent CNF problem follows from the same type of inequalities: We show that there exists a near-linear-size reduction R 3 ,k from the minimum coloring problem to the minimum consistent CNF problem (with function cnf ) such that
$$\chi ( G ) ^ { k } \leq \text{cnf} ( R _ { 3, k } [ G ^ { k } ] ) \leq \chi ( G ) ^ { k } | V ( G ) | ^ { O ( 1 ) }.$$
The proofs of the bounds of graph products (Eq. 3, Eq. 4,Eq. 6 and Eq. 7) are fairly short and elementary; in fact, we believe that they can be given as textbook exercises. These proofs can be found in Section 5, Section 6 and Section 7.
## 2.4 Related Concept
Our pre-reduction graph product concept was inspired by the graph product subadditivity concept we previously introduced in [CLN13a] (some of these ideas were later used in [CLN13b, CLN14]). There, we prove a hardness of approximation using the following framework. As before, let f be an objective function of a problem that we intend to prove hardness and g be an objective function of a problem whose hardness is already known. We show that there are graph products ⊕ ∗ , e , and ∗ such that
- · We can 'decompose' f ( G ∗ e J ): g G ( ) ≤ f ( G ∗ e J ) ≤ g G ( ) + f ( G ∗ J ), and
- · f (( G ⊕ H ) ∗ J ) is 'subadditive': f (( G ⊕ H ) ∗ J ) ≤ f ( G ∗ J ) + f ( H ∗ J ).
We then use the above inequalities to show that if we let G k = G ⊕ G ⊕ . . . ( k times), then
$$g ( G ^ { k } ) \leq f ( G ^ { k } * _ { e } J ) \leq g ( G ^ { k } ) + k f ( G * J ).$$
For large enough k , the term kf G ( ∗ J ) is negligible and thus f ( G k ∗ e J ) ≈ g G ( k ). We use this fact to show that the approximation harness of f is roughly the same as the hardness of g . Observe that if we let R G [ ] = G ∗ e J , the above inequalities can then be used to show that
$$g ( G \oplus H ) \leq f ( R [ G \oplus H ] ) \leq g ( G \oplus H ) + f ( R [ G ] ) + f ( R [ H ] ).$$
In the problems considered in [CLN13a], one can easily bound f ( R G [ ]) and f ( R H [ ]) by | V ( G ) | and | V ( H ) , | respectively. So, our meta theorem will imply that f ( G k ∗ J ) ≈ g G ( k ), which leads to the approximation hardness of f . This means that the previous concept in [CLN13a] can be viewed as a special case of our new concept where we restrict the reduction R to be a graph product R G [ ] = G ∗ e J . The way we use the reduction R in this paper goes beyond this. For example, our reduction R 2 for EDP as illustrated in Figure 1 cannot be viewed as a natural graph product.
Moreover, our reduction R 1 reduces a graph G to an instance of DFA which has nothing to do with graphs. (This is possible only when we abandon viewing reduction R as a graph product.) Our meta theorem also shows that bounds of graph products in a much more general form can imply hardness results. Finally, the way we exploit graph products using the reduction R 1 ,k has never appeared in [CLN13a].
## 2.5 Organization
After giving necessary definitions in Section 3, we prove meta theorems in Section 4. These theorems show that bounding f ( R G [ ∗ H ]) in a certain way will immediately imply a hardness result. They allow us to focus on proving appropriate bounds in later sections. In Section 5, we prove such bounds for the consistency problems and their implications to the hardness of proper PAC-learning. In Section 6, we prove such bounds of the edge-disjoint paths problem on DAGs. Bounds for other problems can be found in Section 7.
## 3 Preliminaries
## 3.1 Terms
Given two graph G and H , the lexicographic product of G and H , denoted by G H · , is defined as
$$V ( G \cdot H ) & = V ( G ) \times V ( H ) = \{ ( u, v ) \, \colon u \in V ( G ) \, \underline { \text{and} } \, v \in V ( H ) \}. \\ E ( G \cdot H ) & = \{ ( u, a ) ( v, b ) \, \colon u v \in E ( G ) \, \underline { \text{or} } \, ( u = v \, \underline { \text{and} } \, a b \in E ( H ) ) \}.$$
Since the lexicographic product is the only graph product concerned in this paper, later on, we will simply use the term graph product to mean the lexicographic product. We define the k -fold graph product of G , denoted by G k , as
$$G ^ { k } = G \cdot G ^ { k - 1 } \text{ for any integer } k > 1 \text{ \ and \ } G ^ { 1 } = G$$
The properties of the lexicographic product that makes it becomes an import tools in proving hardness of approximation is that it multiplicatively increases the independent and chromatic numbers of graphs, without creating an overly dense resulting graph (the OR product also satisfies multiplicativity of independent and chromatic numbers, but it does not serve our purpose).
Theorem 3.1. Let G and H be any graphs. The followings hold on G H · .
- · α G ( · H ) = α G α H ( ) ( ) .
- · χ G χ H ( ) ( ) log | V ( G ) | ≤ χ G ( · H ) ≤ χ G ( ) · χ H ( ) .
In particular, for any k ≥ 1 , α G ( k ) = α G ( ) k and χ G ( ) k -o (1) ≤ χ G ( k ) ≤ χ G ( ) k .
A deterministic finite automaton (DFA) is defined as a 5-tuple ( Q, Σ , δ, q 0 , F ) where Q is the set of states, Σ is the set of alphabets, δ : Q × Σ → Q is a transition function, q 0 is initial state, and F ⊆ Q is the set of accepting states. One can naturally extend the transition function δ into δ ∗ : Q × Σ ∗ → Q by inductively defining δ ∗ ( q, x 1 , . . . , x glyph[lscript] ) as δ ∗ ( δ q, x ( 1 ) , x 2 , . . . , x glyph[lscript] ) and δ ∗ ( q, null ) = q . We say that M accepts x if and only if δ ∗ ( q , x 0 ) ∈ F . The size of DFA M is
glyph[negationslash]
measured by the number of states of M , i.e., | Q | . We say that a DFA is acyclic if there is no state q ∈ Q and string x such that δ ∗ ( q, x ) = q . For NFA, the transition is defined by δ : Q × Σ → 2 Q instead, i.e., each transition possibly maps to several states. An NFA M accepts a string x ∈ Σ ∗ if and only if the transition δ ∗ ( q , x 0 ) contains an accepting state, i.e. δ ∗ ( q , x 0 ) ∩ F = . ∅
## 3.2 Problems
In this section, we list all problems considered in this paper.
Minimum Consistency: In the Minimum Consistency problem, denoted by MinCon( H F , ) , we are given collections P and N of positive and negative sample strings in { 0 1 , } ∗ , for which we are guaranteed that there is a hypothesis h ∈ H that is consistent with all samples in P ∪ N , i.e., h x ( ) = 1 for all x ∈ P and h x ( ) = 0 for all x ∈ N . Our goal is to output a function f ∈ F that is consistent with all these samples, while minimizing | f | . In other words, H and F are the classes of the real hypothesis that we want to learn and those that our algorithm outputs respectively. This notion of learning allows our algorithm to output the hypothesis that is outside of the hypothesis class we want to learn.
Now we need a slightly modified notion of approximation factor. For any instance ( P N , ), we denote by OPT H ( P N , ) the size of the smallest hypothesis h ∈ H consistent with ( P N , ). Let A be any algorithm for MinCon ( H F , ), i.e., A always outputs the hypothesis in F . The approximation gauranteed provided by A is:
$$\sup _ { \mathcal { P }, \mathcal { N } } \frac { | \mathcal { A } ( \mathcal { P }, \mathcal { N } ) | } { \text{OPT} _ { \mathcal { H } } ( \mathcal { P }, \mathcal { N } ) }$$
With this terminology, the problem of learning DFA can be abbreviated as MinCon( DFA , DFA ) .
Edge Disjoint Paths: In the edge-disjoint paths (EDP) problem, given a graph G = ( V, E ) and a set of source-sink pairs { ( s , t 1 1 ) , . . . , ( s k , t k ) } , our goal is to find a collection of paths P = { P i 1 , P i 2 , . . . , P i glyph[lscript] : i j ∈ [ k , P ] i j connects s i j to t i j } that are edge disjoint while maximizing |P| . That is, we want to connects as many source-sink pairs as possible using a collection of edge-disjoint paths. Our focus is on the special case of EDP where G is a directed acycle graph (DAG).
Bounded-Length Edge-Disjoint Cycles: Given a graph G = ( V, E ), the cycle packing number of G , denoted by ν G ( ), is the maximum integer glyph[lscript] such that there exist cycles C , . . . , C 1 glyph[lscript] which are pairwise edge-disjoint in G . The edge-disjoint cycle problem (EDC) asks to compute the value of ν G ( ). If we are additionally given an integer k , the k -cycle packing number of G , denoted by ν k ( G ), is the maximum integer glyph[lscript] for which there exist pairwise edge-disjoint cycles C , . . . , C 1 glyph[lscript] where each cycle C j contains at most k vertices. In the k -edge-disjoint cycle problem ( k -EDC), we are asked to compute ν k ( G ) given an input ( G,k ).
Maximum Independent Set: Given a graph G = ( V, E ), a subset of vertices S ⊆ V is independent in G if and only if G has no edge joining any two vertices in S . The independence number of G , denoted by α G ( ), is the size of a largest independent set in G . In the maximum independent set problem, we are asked to compute an independent set S in G with maximum size.
The following is the hardness results of the maximum independent set problem by H˚stad , a 6 which will be used to obtain the hardness of EDP on DAGs.
Theorem 3.2 ([H˚s96]+[Zuc07]) a . Let glyph[epsilon1] > 0 be any constant. Given graph G = ( V, E ) , it is NP-hard to distinguish between the following two cases:
- · ( Yes-Instance :) α G ( ) ≤ | V ( G ) | glyph[epsilon1]
- · ( No-Instance :) α G ( ) ≥ | V ( G ) | 1 -glyph[epsilon1]
glyph[negationslash]
Chromatic Number: Given a graph G = ( V, E ), a proper coloring σ : V ( G ) → [ c ] is a function that assigns colors to vertices of G so that any two adjacent vertices receive different colors assigned by σ (i.e., uv ∈ E = ⇒ σ u ( ) = σ v ( )). The chromatic number of G , denoted by χ G ( ), is the minimum integer c such that a proper coloring σ : V ( G ) → [ c ] exists, i.e., G can be properly colored by c colors. In the graph coloring problem, we are asked to compute a proper coloring σ : V ( G ) → [ c ] while minimizing c . We will be using the following hardness of approximation result by Feige and Kilian [FK98] . 6
Theorem 3.3 ([FK98]+[Zuc07]) . Let glyph[epsilon1] > 0 be any constant. Given graph G = ( V, E ) , it is NP-hard to distinguish between the following two cases:
- · ( Yes-Instance :) χ G ( ) ≤ | V ( G ) | glyph[epsilon1]
- · ( No-Instance :) χ G ( ) ≥ | V ( G ) | 1 -glyph[epsilon1]
## 4 Meta Theorems
In this section, we prove general theorems that will be used in proving most hardness results in this paper. These theorems give abstractions of the (graph product) properties one needs to prove in order to obtain hardness of approximation results. Our techniques can be used to derive hardnesses for both minimization and maximization problems. For the former, the reduction is from minimum coloring, while the latter is obtained via a reduction from maximum independent set.
Let us start with maximization problems. Suppose we have an optimization problem Π such that any instance I ∈ Π is associated with an optimal function OPT Π ( I ). We consider a transformation R that maps any graph G into an instance R G [ ] of the problem Π. We say that a transformation R satisfies a low α -projection property with respect to a maximization problem Π if and only if the following two conditions hold:
- · (I) For any graph G = ( V, E ), OPT Π ( R G [ ]) ≥ α G ( ).
- · (II) There are universal constants c 1 , c 2 > 0 (independent of the choices of graphs) such that, for any two graphs G and H ,
$$\text{OPT} _ { \Pi } ( R [ G \cdot H ] ) \leq | V ( G ) | ^ { c _ { 1 } } + \alpha ( G ) ^ { c _ { 2 } } \text{OPT} _ { \Pi } ( R [ H ] ).$$
glyph[negationslash]
6 The hardness results of the maximum independent set problem [H˚s96] and the graph coloring problem [FK98] a hold under the assumption NP = ZPP. The results were later derandomized by Zuckerman in [Zuc07] and thus hold under the assumption P = NP.
glyph[negationslash]
- · (III) There is a universal constant c 0 > 0 such that
$$\text{OPT} _ { \Pi } ( R [ G ] ) \leq c _ { 0 } | R [ G ] |.$$
Intuitively, the transformation R with the low α -projection property tells us that there are relationships between the optimal solution of the problem Π on R G [ ] and the independence number of G . Instead of looking for a sophisticated construction of R , we focus on a 'simple' transformation R that establishes a connection on one side, i.e., OPT Π ( R G [ ]) ≥ α G ( ), and the 'growth' of OPT Π is 'slow' with respect to graph products. Property (III) of the low α -projection property says that the optimal is at most linear in the size of the instance, which is the case for almost every natural combinatorial optimization problem.
Next, we turn our focus to a minimization problem. In this case, we relate the optimal solution to the chromatic number of an input graph. Specifically, one can define the low χ -projection property with respect to a minimization problem Π as follows.
- · (I) For any graph G = ( V, E ), OPT Π ( R G [ ]) ≥ χ G ( ).
- · (II) There are universal constants c 1 , c 2 > 0 (independent of the choices of graphs) such that, for any two graphs G and H , we have
$$\text{OPT} _ { \Pi } ( R [ G \cdot H ] ) \leq | V ( G ) | ^ { c _ { 1 } } + \chi ( G ) ^ { c _ { 2 } } \text{OPT} _ { \Pi } ( R [ H ] ).$$
- · (III) There is a universal constant c 0 > 0 such that
$$\text{OPT} _ { \Pi } ( R [ G ] ) \leq c _ { 0 } | R [ G ] |.$$
We observe that the existence of such reductions is sufficient for establishing hardness of approximation results, and the hardness factors achievable from the theorems depend on the size of the reduction.
Theorem 4.1 (Meta-Theorem for Maximization Problems) . Let Π be a maximization problem for which there is a reduction R for Π that satisfies low α -projection property with | R G [ ] | = O V G ( | ( ) | d ) . Then for any glyph[epsilon1] > 0 , given an instance I of Π , it is NP-hard to distinguish between the following two cases:
- · ( Yes-Instance :) OPT Π ( I ) ≥ | I | 1 /d -glyph[epsilon1]
- · ( No-Instance :) OPT Π ( I ) ≤ | I | glyph[epsilon1]
Theorem 4.2 (Meta-Theorem for Minimization Problems) . Let Π be a minimization problem for which there is a reduction R for Π that satisfies low χ -projection property with | R G [ ] | = O V G ( | ( ) | d ) , for some constant d ≥ 0 . Then for any glyph[epsilon1] > 0 , given an instance I of Π , it is NP-hard to distinguish between the following two cases:
- · ( Yes-Instance :) OPT Π ( I ) ≤ | I | glyph[epsilon1]
- · ( No-Instance :) OPT Π ( I ) ≥ | I | 1 /d -glyph[epsilon1]
## 4.1 Proof of Theorem 4.1 (Meta Theorem for Maximization Problems)
Consider a reduction R that transforms a graph G into an instance of Π that satisfies the low α -projection property. We analyze how the optimal value changes over glyph[lscript] -fold lexicographic products.
$$\text{Lemma } 4. 3. \text{ For any positive integer } \ell, \text{ OPT} _ { \Pi } ( R [ G ^ { \ell } ] ) \leq \ell c _ { 0 } | V ( G ) | ^ { c _ { 1 } + d + 1 } \alpha ( G ) ^ { 2 c _ { 2 } \ell }$$
Proof. This is proved by induction on a positive integer glyph[lscript] . The base case glyph[lscript] = 1 holds because OPT Π ( R G [ ]) ≤ c 0 | R G [ ] | ≤ c 0 | V ( G ) | d . Assume that the induction hypothesis holds for any glyph[lscript] > 1, and consider OPT Π ( R G [ glyph[lscript] +1 ]). By writing G glyph[lscript] +1 = G G · glyph[lscript] and applying the low α -projection property, we have
$$O P T _ { \Pi } ( R [ G ^ { \ell + 1 } ] ) \leq | V ( G ) | ^ { c _ { 1 } } + \alpha ( G ) ^ { c _ { 2 } } O P T _ { \Pi } ( R [ G ^ { \ell } ] )$$
Then, by applying induction hypothesis, we have
$$\text{OPT} _ { \Pi } ( R [ G ^ { \ell + 1 } ] ) \ & \leq \ | V ( G ) | ^ { c _ { 1 } } + \alpha ( G ) ^ { c _ { 2 } } \left ( \ell c _ { 0 } | V ( G ) | ^ { c _ { 1 } + d + 1 } \alpha ( G ) ^ { 2 c _ { 2 } \ell } \right ) \\ & \leq \ | V ( G ) | ^ { c _ { 1 } } + \alpha ( G ) ^ { c _ { 2 } + 2 c _ { 2 } \ell } \ell c _ { 0 } | V ( G ) | ^ { c _ { 1 } + d + 1 } \\ & \leq \ ( \ell + 1 ) c _ { 0 } | V ( G ) | ^ { c _ { 1 } + d + 1 } \alpha ( G ) ^ { 2 c _ { 2 } ( \ell + 1 ) }$$
We note that the exponent of the term α G ( ) depends on glyph[lscript] (the number of times the product is applied), while that of | V ( G ) | does not. Intuitively speaking, this is why the contribution of the term | V ( G ) | c 1 vanishes after taking graph products.
Hardness of Approximation. Now we prove the hardness of approximation result claimed in Theorem 4.1. Start from graph G as given by Theorem 3.2. Then construct an instance R G [ glyph[lscript] ] with glyph[lscript] = 1 glyph[ceilingleft] /glyph[epsilon1] glyph[ceilingright] . This results in the instance R G [ glyph[lscript] ] of the problem Π of size N = | R G [ glyph[lscript] ] | = O V G ( | ( ) | glyph[lscript]d ). Yes-Instance
In the , we have
$$O P T _ { \Pi } ( R [ G ^ { \ell } ] ) \geq \alpha ( G ^ { \ell } ) = \alpha ( G ) ^ { \ell } \geq | V ( G ) | ^ { ( 1 - \epsilon ) \ell } = N ^ { 1 / d - O ( \epsilon ) }.$$
In the No-Instance , we have
$$\text{OPT} _ { \Pi } ( R [ G ^ { \ell } ] ) \leq O ( | V ( G ) | ^ { d + c _ { 1 } + 1 } \alpha ( G ) ^ { 2 c _ { 2 } \ell } ).$$
Since α G ( ) ≤ | V ( G ) | glyph[epsilon1] in this case, we have
$$\alpha ( G ) ^ { 2 c _ { 2 } \ell } \leq | V ( G ) | ^ { 2 c _ { 2 } } = | V ( G ) | ^ { O ( 1 ) } = N ^ { O ( \epsilon ) }.$$
This implies that OPT Π ( R G [ glyph[lscript] ]) ≤ | V ( G ) | O (1) N O glyph[epsilon1] ( ) = N O glyph[epsilon1] ( ) , and the gap between Yes-Instance and No-Instance is N 1 /d -O glyph[epsilon1] ( ) . This completes the proof.
## 4.2 Proof of Theorem 4.2 (Meta Theorem for Minimization Problems)
Similarly to the case of maximization problems, we can prove the following lemma by induction on integers glyph[lscript] . We shall skip the proof as it is the same as that of Lemma 4.3 except that α G ( ) is replaced by χ G ( ).
Lemma 4.4. For any positive integer glyph[lscript] , OPT Π ( R G [ glyph[lscript] ]) ≤ glyph[lscript]c 0 | V ( G ) | c 1 + +1 d χ G ( ) 2 c 2 glyph[lscript] .
Hardness of Approximation Take the instance R G [ glyph[lscript] ] with glyph[lscript] = 1 glyph[ceilingleft] /glyph[epsilon1] glyph[ceilingright] .
In the Yes-Instance , we have the following bound, which is slightly different from the case of the maximization problem.
$$O P T _ { \Pi } ( R [ G ^ { \ell } ] ) \geq \chi ( G ^ { \ell } ) \geq ( \chi _ { f } ( G ) ) ^ { \ell } \geq | V ( G ) | ^ { \ell ( 1 - 2 \epsilon ) } = N ^ { 1 / d - O ( \epsilon ) }.$$
In the No-Instance , Lemma 4.4 gives OPT Π ( R G [ glyph[lscript] ]) ≤ N O glyph[epsilon1] ( ) . Thus, we have the desired gap, completing the proof.
## 4.3 Overview of Applications
Most of the reductions in this paper are direct applications of the above two meta theorems. That is, we design the following reductions.
- · A reduction R EDP for EDP such that | R edp [ G ] | = O V G ( | ( ) | 2 ) and satisfies α -projection property. This implies a tight n 1 2 / -glyph[epsilon1] hardness of approximating EDP on DAGs.
- · A reduction R fa for MinCon such that | R fa [ G ] | = O V G ( | ( ) | 2 ) and satisfies χ -projection property. This gives n 1 2 / -glyph[epsilon1] hardness of approximating MinCon(NFA,ADFA).
Notice that the reduction R fa above is not tight. To obtain a tight result, we need | R fa [ G ] | = O V G ( | ( ) ), and it seems difficult to obtain such a reduction. | We instead exploit the further structure of graph products and prove bounds of the form
$$\chi ( G ^ { k } ) \leq \text{OPT} ( R ^ { \prime } _ { f a } [ G ^ { k } ] ) \leq \chi ( G ) ^ { O ( k ) } | V ( G ) | ^ { O ( 1 ) }$$
Now our reduction size is smaller, i.e., | R ′ fa [ G k ] | = | V ( G ) | (1+ (1)) o k as opposed to | R fa [ G k ] | = | V ( G ) | 2 k . Moreover, the reduction R ′ fa exploits the fact that the input graph is written as a k -fold product of graphs. This more restricted form of graph products allows us to prove tight hardness (and PAC impossibility result) of DFA and DNF/CNF Minimization.
## 5 Hardnesses of Finite Automata Problems: Minimum Consistency and Proper PAC Learning
We show in this section the hardness of the consistency problems for finite automata, as well as the implications on impossibility results for PAC learning. We start our discussion by proving the hardness for MinCon (ADFA,NFA), which includes the minimum consistent NFA problem ( MinCon (NFA,NFA)) as a special case. Then we proceed to prove the tight hardness of approximating MinCon (ADFA,DFA), which implies the tight hardness of approximating the minimum consistent DFA problem and also implies the impossibility result for proper PAC-learning DFA.
## 5.1 Hardness of MinCon (ADFA, NFA) via Graph Products
In this section, we show an N 1 2 / -glyph[epsilon1] hardness for MinCon (ADFA,NFA). Formally, we prove the following theorem.
Theorem 5.1. Let glyph[epsilon1] > 0 be any positive constant. Given two sets of positive and negative sample strings P N , over alphabet Σ = { 0 1 , } with a total length of N bits, it is NP-hard to distinguish the following two cases:
- · There is an acyclic deterministic finite automata of size N glyph[epsilon1] that is consistent with all strings in P ∪ N .
- · Any non-deterministic finite automata consistent with P∪N must have at least N 1 2 / -glyph[epsilon1] states.
This is done by designing a reduction R G [ ] with χ -projection property and | R G [ ] | = O V G ( | ( ) | 2 ). Our proof in fact shows that the projection properties hold for both optimal DFA and NFA functions.
## 5.1.1 The Reduction R
We will be working with binary strings, i.e., the alphabet set Σ = { 0 1 , } . Given a graph G = ( V, E ), we construct two sets P N , of positive and negative samples, which encode vertices and edges of the graph. We assume w.l.o.g. that | V ( G ) | = 2 k for some integer k . Therefore, each vertex u ∈ V ( G ) can be associated with a k -bit string 〈 u 〉 ∈ { 0 1 , } k .
Now our reduction R G [ ] is defined as follows. The positive samples are given by
$$\mathcal { P } = \left \{ \langle u \rangle 1 \langle u \rangle ^ { R } \colon u \in V ( G ) \right \}$$
and the negative samples are
$$\mathcal { N } = \left \{ \langle u \rangle 1 \langle v \rangle ^ { R } \colon u v \in E ( G ) \right \}$$
We denote this instance of the consistency problem by an ordered pair ( P N , ). Now we proceed to prove property (I), that any NFA consistent with ( P N , ) must have at least χ G ( ) states.
Lemma 5.2. Let M = ( Q, Σ , δ, q 0 , F ) be an NFA that is consistent with ( P N , ) . Then for any vertex u ∈ V ( G ) , glyph[negationslash]
$$\delta ^ { * } ( q _ { 0 }, \langle u \rangle ) \not \subseteq \bigcup _ { v \colon u v \in E ( G ) } \delta ^ { * } ( q _ { 0 }, \langle v \rangle ).$$
glyph[negationslash]
Proof. Assume for contradiction that δ ∗ ( q , 0 〈 u 〉 ) ⊆ ⋃ v uv : ∈ E G ( ) δ ∗ ( q , 0 〈 v 〉 ). Since 〈 u 〉 1 〈 u 〉 R is a positive sample, there is a state q ∈ δ ∗ ( q , 0 〈 u 〉 ) that leads to an accepting state (i.e., δ ∗ ( q, 1 〈 u 〉 R ) ∩ F = ∅ ). By the assumption, the state q also belongs to another set δ ∗ ( q , 0 〈 v 〉 ) for some v : uv ∈ E G ( ).
glyph[negationslash]
Now consider the string 〈 v 〉 1 〈 u 〉 R , which is a negative sample because vu ∈ E G ( ). Since q ∈ δ ∗ ( q , 0 〈 v 〉 ) and δ ∗ ( q, 1 〈 u 〉 R ) ∩ F = , the string ∅ 〈 v 〉 1 〈 u 〉 R must be accepted by M , a contradiction.
Lemma 5.2 implies in particular that, for each vertex u ∈ V ( G ), the set δ ∗ ( q , 0 〈 u 〉 \ ) ( ⋃ v uv : ∈ E G ( ) δ ∗ ( q , 0 〈 v 〉 ) ) is not empty. Now denote by OPT DFA ( R G [ ]) and OPT NFA ( R G [ ]) the number of states in the minimum DFA and NFA that are consistent with the samples R G [ ] = ( P N , ) G , respectively.
Corollary 5.3. Any NFA M that is consistent with ( P N , ) G must have at least χ G ( ) states. Therefore, OPT DFA ( R G [ ]) ≥ OPT NFA ( R G [ ]) ≥ χ G ( ) for all G .
Proof. For each state q ∈ Q , define a set C q = { u ∈ V ( G ) : q ∈ δ ∗ ( q , 0 〈 u 〉 ) \ ( ⋃ v uv : ∈ E G ( ) δ ∗ ( q , 0 〈 v 〉 ) } . It is easy to see that C q is an independent set and thus form a proper color class of G . Lemma 5.2 implies that each vertex u ∈ V ( G ) belongs to at least one class. So, { C q } q ∈ Q gives a proper | Q | -coloring of G , implying that | Q | ≥ χ G ( ).
## 5.1.2 χ -Projection Property
We will consider a specific class of DFA M = ( Q, Σ , δ, q 0 , F ), which we call canonical DFA . Specifically, we say that a DFA is canonical if it has the following properties.
- · The state diagram has exactly glyph[lscript] layers for some glyph[lscript] , and each path from q 0 to any sink has length exactly glyph[lscript] .
- · All accepting states are in the last layer.
Denote shortly by OPT ( R G [ ]) the number of states in the minimum canonical DFA consistent with R G [ ]. So we have that OPT ( R G [ ]) ≥ OPT DFA ( R G [ ]) ≥ OPT NFA ( R G [ ]). The following lemma gives the χ -projection property for OPT ( ) ·
## Lemma 5.4. OPT ( R G [ · H ]) ≤ χ G ( )( OPT ( R H [ ]) + O V G ( | ( ) )) |
To prove this lemma, we show how to construct, given a canonical DFA for R H [ ], a 'compact' canonical DFA for R G [ · H ]. We note that one key idea here is to avoid exploiting the DFA for R G [ ] but instead tries to use the color classes of G in its optimal coloring to 'compress' the DFA for R G [ · H ].
Proof. Let M H = ( Q , H { 0 1 , } , δ H , q H , F H ) be the minimum DFA for the instance R H [ ] whose number of states is s = OPT ( R H [ ]) and has glyph[lscript] H = 2 h +1 layers for h = log glyph[ceilingleft] | V ( H ) |glyph[ceilingright] . Let C , . . . , C 1 B be the color classes of G defined by the optimal coloring, so B = χ G ( ). Let f : V ( G ) → [ B ] be the corresponding coloring function. We will also be using several copies of a directed complete binary tree with 2 k leaves, where each leaf corresponds to a string in { 0 1 , } k and is associated with a vertex in V ( G ). Call this directed binary tree T k .
We will use M H and T k to construct a new acyclic DFA M that have at most B s ( + O V G ( | ( ) )) | states and exactly glyph[lscript] = 2( k + glyph[lscript] H ) + 1 layers. Now we proceed with the description of machine M = ( Q, { 0 1 , } , δ, q, F ). We start by taking a copy of directed tree T k , and call this copy T (0) k . The starting state q is defined to be the root of T (0) k . This is the first phase of the construction. Notice that there are k layers in the first phase, so exactly k positions of any input string will be read after this phase. Each state in the last layer is indexed by state ( 〈 v 〉 ) for each v ∈ V ( G ).
In the second phase , we take B copies of the machines M H where the j th copy, denoted by M ( j ) H = ( Q ( j ) H , { 0 1 , } , δ ( j ) H , q ( j ) H , F ( j ) H ), is associated with color class C j defined earlier. For each vertex v ∈ V ( G ), we connect the corresponding state state ( 〈 v 〉 ) in the last layer of Phase 1 to the starting state q ( f ( v )) H . This transition can be thought of as a 'null' transition which can be removed afterward, but keeping it this way would make the analysis simpler. Since each copy of M H has 2 glyph[lscript] H +1 layers, now our construction has exactly 2 glyph[lscript] H + k +1 layers.
In the final phase, we first extend all rejecting states in M ( j ) H by a unified path until it reaches layer 2( glyph[lscript] H + k ) + 1. This is a rejecting state rej 0 . Now, for each j = 1 , . . . , B , we connect each accepting state in the last layer of M ( j ) H to the root in the copy T ( j ) k again by a 'null' transition, so we reach the desired number of layers now (notice that each root-to-leaf path has 2( k + glyph[lscript] H ) + 1 states.) The states in the last layer of T ( j ) k are indexed by state j, ( 〈 v 〉 ). The accepting states of M are defined as F = ⋃ B j =1 { state j, ( 〈 u 〉 R ) : u ∈ C j } , and the rest of the states are defined as rejecting. This completes our construction. See Figure 2 for illustration.
Figure 2: The illustration of the construction in the proof of Lemma 5.4.
The size of the construction is | V ( G ) | + Bs + O V G B ( | ( ) | ) = B s ( + O V G ( | ( ) )). | The next claim shows that the machine M is consistent with samples obtained from the product of G and H , which thus finish the proof.
Claim 5.5. Given a machine M H that is consistent with samples R H [ ] , the machine M constructed as above is consistent with samples R G [ · H ] .
Proof. First we check the positive sample. For each vertex ( u, a ) ∈ V ( G · H ), the corresponding string 〈 ( u, a ) 1 ( 〉 〈 u, a ) 〉 R can be thought of as x = 〈 u 〉〈 a 〉 1 〈 a 〉 R 〈 u 〉 R . After the first k transitions, the machine M will stop at the state q ( f ( u )) H . Then the substring 〈 a 〉 1 〈 a 〉 R will lead to an accepting state in F ( f ( u )) H (since M H is consistent with samples in R H [ ]). Now, at the current state, we are at the root of the tree T ( f ( u )) k , and we are left with the substring 〈 u 〉 R . Since u ∈ C f ( u ) , the substring 〈 u 〉 R leads to an accepting state. This proves that the machine M always accepts positive samples.
Next, consider a negative sample 〈 ( u, a ) 1 ( 〉 〈 v, b ) 〉 R generated by the edge ( u, a )( v, b ) ∈ E G ( ). Again, this can be thought of as 〈 u 〉〈 a 〉 1 〈 b 〉 R 〈 v 〉 R . There are two possible cases:
glyph[negationslash]
- · If uv ∈ E G ( ), then the machine will enter M ( f ( u )) H after reading the substring 〈 u 〉 . Next, the machine reads the substring 〈 a 〉 1 〈 b 〉 R . If it manages to reach the third phase without rejection (i.e., M H accepts 〈 a 〉 1 〈 b 〉 R ), then it will enter the tree T ( f ( u )) k . Note that there is no edge joining two vertices in C f ( u ) because it is a color class. Thus, the substring 〈 v 〉 leads to a rejection because uv ∈ E G ( ) implies that v ∈ C f ( u ) .
(Notice that the rejection does not depend on what happens inside M H .)
- · If u = v and ab ∈ E G ( ), then after the first k transitions, the machine enters M ( f ( u )) H with the input string 〈 a 〉 1 〈 b 〉 R for ab ∈ E H ( ). Since M H is consistent with samples in R H [ ], this would lead to a rejection in M ( f ( u )) H and therefore in M .
We will also need the base case condition as required by the low α -projection property.
Lemma 5.6. For any graph G , OPT ( R G [ ]) ≤ O V G ( | ( ) | 2 log | V ( G ) ) |
Proof. We simply use the tree T 2 k +1 with the initial state q at the root of T 2 k +1 , where each vertex at the leaf can be associated with a string in { 0 1 , } 2 k +1 . We simply define the accepting states to be those that correspond to the strings of the form 〈 u 〉 1 〈 u 〉 R . The size of the construction is 2 2 k +1 = O V G ( | ( ) | 2 log | V ( G ) ). |
## 5.2 Tight Hardness for MinCon (ADFA, DFA)
Notice that the construction in the previous section is not tight because the size of the negative samples in R G [ ] is large compared to the number of vertices in graph G , i.e., |N| = Θ( | V ( G ) | 2 ). To handle this problem, we take into account the structure of the lexicographic product and 'encode' negative samples in a more compact form, i.e. we ideally want the construction size to be nearly linear on | V ( G ) , | i.e., |N| = O V G ( | ( ) | 1+ (1) o ), instead of quadratic.
To this end, we construct a reduction R G k [ k ] We remark that, while the construction in this section gives tighter results for DFA, ADFA, and OBDD, it does not apply to NFA.
## 5.2.1 The Reduction R G k [ k ]
We show a reduction R G k [ k ] of size | R G k [ k ] | = | V ( G ) | k (1+ (1)) o . Consider a graph H = G k (the k -fold lexicographic product of G ). We will encode the edge structures of H into the positive and negative samples as follows.
Positive Samples: For each glyph[vector] u = ( u , . . . , u 1 k ) ∈ V ( H ), define a positive sample
$$\ p o s ( \vec { u }, i ) = \langle u _ { 1 } \rangle \dots \langle u _ { k } \rangle 1 \langle u _ { 1 } \rangle \dots \langle u _ { i } \rangle.$$
The set of all positive samples is denoted by
$$\mathcal { P } = \{ p o s ( \vec { u }, i ) \, \colon u \in V ( H ), i = 1 \dots k \}$$
Negative Samples: For each a pair of vertices glyph[vector] u ∈ V ( H ) and v ∈ V ( G ) such that u v i ∈ E G ( ), define a negative sample
$$\ n e g ( \vec { u }, v, i ) = \langle u _ { 1 } \rangle \dots \langle u _ { k } \rangle 1 \langle u _ { 1 } \rangle \dots \langle u _ { i - 1 } \rangle \langle v _ { i } \rangle$$
The set of all negative samples is denoted by
$$\mathcal { N } = \{ n e g ( \vec { u }, v, i ) \, \colon \vec { u } \in V ( H ), v \in V ( G ), u _ { i } v \in E ( G ), i = 1, \dots, k \}$$
Intuitively, an edge in the the input graph represents a conflict between two vertices. Negative samples are thus defined to capture a conflict (an edge) in the product of graphs between vertices glyph[vector] u and glyph[vector] v at coordinate i . Notice that the size of positive and negative samples are |P| = kn k and |N| = kn k | E G ( ) | ≤ kn k +2 .
Let OPT ADFA ( ) denote the number of states in the optimal acyclic DFA that is consistent with · the samples. We will prove the following lemma.
$$\text{Lemma 5.7.} \ \chi ( H ) \leq \text{OPT} _ { D F A } ( \mathcal { P }, \mathcal { N } ) \leq \text{OPT} _ { A D F A } ( \mathcal { P }, \mathcal { N } ) \leq \chi ( G ) ^ { 2 k } | V ( G ) | ^ { 4 }.$$
The bound OPT DFA ( P N , ) ≤ OPT ADFA ( P N , ) is trivial. For the other bounds, we will prove the left and right-hand side inequalities of Lemma 5.7 in Section 5.2.2 and Section 5.2.3, respectively. The hardness result then follows trivially from Theorem 3.3 and Theorem 3.1. In particular, taking the hard instance of the graph coloring problem as in Theorem 3.3, we have that
Yes-Instance
: OPT DFA ( P N , ) ≤ χ G ( ) 2 k n 4 ≤ n O (1) .
No-Instance
: OPT DFA ( P N , ) ≥ χ H ( ) ≥ χ G ( ) k - o (1) ≥ n ( k - O (1)) .
Since |P| + |N| = O kn ( k +2 ), this implies the hardness gap of ( |P| + |N| ) 1 - ε , for any ε > 0.
## 5.2.2 The Lower Bound of OPT DFA
First, we show the lower bound for OPT DFA ( P N , ). Let M = ( Q, Σ , δ, q 0 , F ) be a DFA consistent with ( P N , ). We construct from M a | Q | -coloring of H : For each state q ∈ Q , we define a color class C q = { glyph[vector] u : δ ∗ ( q , 0 〈 glyph[vector] u 〉 ) = q } . Since M is deterministic, each vertex must get at least one color.
glyph[negationslash]
Lemma 5.8. For any vertices glyph[vector] u,glyph[vector] v ∈ C q , glyph[vector] uglyph[vector] v ∈ E H ( ) . That is, C q is a proper color class of H .
Proof. Suppose to a contrary that there is a pair of vertices glyph[vector] u,glyph[vector] v ∈ C q such that glyph[vector] uglyph[vector] v ∈ E H ( ). Since H is obtained by the lexicographic product, there exists a coordinate i in which glyph[vector] u and glyph[vector] v conflict, i.e., u j = v j for all j < i and u v i i ∈ E G ( ). We know that δ ∗ ( q, 1 〈 u 1 〉 . . . 〈 u i 〉 ) ∈ F because pos glyph[vector] , ( u i ) = 〈 glyph[vector] u 〉 1 〈 u 1 〉 . . . 〈 u i 〉 is a positive sample. Since δ ∗ ( q , 0 〈 glyph[vector] u 〉 ) = δ ∗ ( q , 0 〈 glyph[vector] v 〉 ) = q , we must also have δ ∗ ( q , 0 〈 v 〉 1 〈 u 1 〉 . . . 〈 u i 〉 ) = δ ∗ ( q, 1 〈 u 1 〉 . . . 〈 u i 〉 ) ∈ F . But, this contradicts the fact that neg glyph[vector] , ( v glyph[vector] , u i ) = 〈 glyph[vector] v 〉 1 〈 v 1 〉 . . . 〈 v i -1 〉〈 u i 〉 = 〈 glyph[vector] v 〉 1 〈 u 1 〉 . . . 〈 u i 〉 is a negative sample.
## 5.2.3 The Upper bound of OPT ADFA
Now we need to argue that there is an acyclic DFA M of size χ G ( ) 2 k n 4 . Suppose V ( G ) = { 0 1 , } glyph[lscript] . Let c = χ G ( ), and σ : V ( G ) → [ c ] be an optimal coloring of G . Our construction has two steps. First, we construct a complete rooted c -ary tree with 2 k level, namely S . Note that S is a directed tree whose edges are oriented toward leaves. Each vertex in S except the root is associated with one color class from σ . In particular, for each internal vertex a of S , each child x of a is associated with a distinct color from [ c ]. We define the coloring of S by ρ : V ( S ) → [ c ]. Second, we replace each vertex a of S by a complete binary tree T with n leaves; we denote this copy of T by T a . Each leaf q of T a is associated with a vertex u of G and thus has a color σ u ( ) assigned. (We abuse σ q ( ) = σ u ( ) to mean a color of q .) For any vertex x in S that is a child of a , we join every leaf q of T a with color σ q ( ) = ρ a ( ) to the root r of T x . The transition edge qr is a null transition unless a is a vertex at level k in S ; for the case that a is at level k , the transition edge qr is labeled '1'. (Note that a null transition edge qr means that we will merge q and r in the final construction. It is easy to see that this results in a DFA (not NFA) because S is a tree.) It can be seen that the
Figure 3: The illustration of the construction in the proof of Lemma 5.7.
constructed directed graph has a single source vertex (i.e., a vertex with no incoming edges), which we define as a starting state q 0 .
To finish the construction, we define accepting states. Let a 0 be the root of S . Consider a vertex a i at level i > k in S and its corresponding tree T a i . Since S is a tree, there is a unique path from a 1 to a i , namely, P = a , . . . , a 0 k , a k +1 . . . , a i . For each leaf q of T a i , we define q as an accepting state if and only if σ q ( ) = ρ a ( i -k +1 ), i.e., q and a i -k +1 receive the same color. See Figure 3 for illustration.
Each copy of T has at most 2 n vertices, and S has at most c 2 k vertices. Thus, the size of the DFA M is at most 2 nc 2 k . Also, observe that M is acyclic.
## Lemma 5.9. The DFA M is consistent with ( P N , ) .
Proof. Consider any sample glyph[vector] u ∈ P ∪ N , which must be of the form:
$$\vec { u } = \langle u _ { 1 } \rangle \dots \langle u _ { k } \rangle 1 \langle u _ { k + 1 } \rangle \dots \langle u _ { k + i } \rangle \quad \text{for $i\,\colon1\leq i\leq k$ and $u_{j}\in V(S)$ for all $j=1,\dots,k+i$.}$$
Note that u j = u k + j for all j = 1 2 , , . . . , i -1. The transition δ ∗ ( q , glyph[vector] 0 u ) forms a path P in M , which traverses from the starting state q 0 to some state q j . (That is, q j = δ ∗ ( q , glyph[vector] 0 u ).) By construction, P corresponds to the path a , . . . , a 1 i in S and thus must visit a leaf q j of tree T a j , for j = 1 , . . . , i . Moreover, each q j is associated with vertex u j ∈ V ( G ). Notice that ρ a ( j ) = σ u ( j -1 ) because we have an edge from q j -1 to T a j if and only if q j -1 and a j receive the same color.
If glyph[vector] u is a positive sample in P , then we have u i -k = u i . (Note that q j and u j receive the same color for all j = 1 2 , , . . . , k .) Since a i -k +1 has the same color as q i -k (and so does u i -k = u i ), we have ρ a ( i -k +1 ) = σ q ( i ). Thus, q i is an accepting state.
glyph[negationslash]
If glyph[vector] u is a negative sample in N , then we must have an edge u i -k u i ∈ E G ( ). So, u i -k and u i receive different colors. Since ρ a ( i -k +1 ) = σ u ( i -k ), it follows that ρ a ( i -k +1 ) = σ q ( i ). Thus, q i is not an accepting state. This proves that M is consistent with both positive and negative samples.
## 5.3 Hardness of Proper PAC-Learning
Here we show that DFAs are not PAC-learnable. That is, we prove Corollary 1.2. We will use the connection between PAC learning and the existence of an Occam algorithm , defined as follows.
Definition 5.10. An Occam algorithm for a hypothesis class H in terms of function classes F is an algorithm A that for some constant k ≥ 0 and α < 1 , the following guarantee holds. Let h ∈ H has size n and represents some language L h ( ) . Then on any input of s samples of L r ( ) , each of length at most m , the algorithm A outputs an element h ∈ H of size at most n m s k k α that is consistent with each of the s samples.
Therefore, an Occam algorithm for DFA is the case when H = DFA = F , and the measure of the size of each hypothesis h ∈ DFA is the number of states. It is known that PAC learnability of DFA implies the existence of an Occam algorithm for the same hypothesis class as stated formally in the following theorem.
Theorem 5.11 ([BP92], statement from [Pit89]) . If DFAs are properly PAC-learnable, then there exists a randomized Occam algorithm for DFA that runs in polynomial time.
Theorem 5.11 implies that, to prove Corollary 1.2, it suffices to rule out the existence of a randomized Occam algorithm for DFA, which is shown in the next Theorem.
Theorem 5.12. Unless NP = RP , there is no polynomial time randomized Occam algorithm for DFA.
Proof. We prove by contrapositive. Assume that there is a randomized Occam algorithm A for DFA with parameters ( k, α ) for some constants k ≥ 0 and 0 ≤ α < 1. Then we argue that there would exist and an algorithm that distinguishes between the Yes-Instance and No-Instance given in Theorem 1.1. To see this, take an instance of MinCon(ADFA,DFA) as in Theorem 1.1. So, we have a pair of sets ( P, N ) of N samples, each of length O (log N ). The parameters of the Occam algorithm A are thus s = N and m = O (log N ).
We choose the parameter glyph[epsilon1] in Theorem 1.1 to be glyph[epsilon1] = (1 -α / ) (2 k +1).
In the Yes-Instance , there is a DFA of size N glyph[epsilon1] consistent with the samples. Thus, our hypothesis class has size n = N glyph[epsilon1] . By definition, the Occam algorithm A gives us a DFA M of size
$$| M | \leq N ^ { \epsilon k } \cdot ( \log N ) ^ { k } \cdot N ^ { \alpha } \leq N ^ { \alpha + ( 2 k ) \epsilon } \leq N ^ { \alpha + ( 2 k ) \frac { 1 - \alpha } { 2 k + 1 } } = N ^ { 1 - \frac { 1 - \alpha } { 2 k + 1 } } = N ^ { 1 - \epsilon }$$
In the No-Instance , any DFA M consistent with ( P, N ) has size | M > N | 1 -glyph[epsilon1] . Therefore, the randomized Occam algorithm A can distinguish between the Yes-Instance and No-Instance in Theorem 1.1, implying that NP = RP. This completes the proof.
glyph[negationslash]
A similar but weaker theorem can be proven for the case of NFAs. Indeed, we rule out the existence Occam algorithm for NFA with parameter 0 ≤ α ≤ 1 2, assuming that NP = RP. /
Theorem 5.13. Unless NP = RP , there is no polynomial time randomized Occam algorithm for NFA with parameter 0 ≤ α ≤ 1 2 / .
Proof. The proof is essentially the same as that of Theorem 5.12 with slightly different parameters. We prove by contrapositive. Assume that there is a randomized Occam algorithm A for NFA with parameters ( k, α ) for some constants k ≥ 0 and 0 ≤ α ≤ 1 2. We will show that the algorithm / A can be used to distinguishes between the Yes-Instance and No-Instance given in Theorem 1.4 and thus implying that NP = RP .
Take an instance of MinCon(ADFA,NFA) as in Theorem 1.1. So, we have a pair of sets ( P, N ) of N samples, each of length O (log N ). The parameters of the Occam algorithm A are thus s = N and m = O (log N ).
We choose the parameter glyph[epsilon1] in Theorem 1.1 to be glyph[epsilon1] = (1 / 2 -α / ) (2 k +1).
In the Yes-Instance , there is an NFA of size N glyph[epsilon1] consistent with the samples. Thus, our hypothesis class has size n = N glyph[epsilon1] . By definition, the Occam algorithm A gives us a NFA M with size
$$| M | \leq N ^ { \epsilon k } \cdot ( \log N ) ^ { k } \cdot N ^ { \alpha } \leq N ^ { \alpha + ( 2 k ) \epsilon } = N ^ { \alpha + ( 2 k ) \frac { 1 / 2 - \alpha } { 2 k + 1 } } = N ^ { 1 / 2 - \frac { 1 / 2 - \alpha } { 2 k + 1 } } = N ^ { 1 / 2 - \epsilon }$$
In the No-Instance , any NFA M consistent with ( P, N ) has size | M > N | 1 2 / -glyph[epsilon1] . Therefore, the randomized Occam algorithm A can distinguish between the Yes-Instance and No-Instance in Theorem 1.4, implying that NP = RP. This completes the proof.
Corollary 5.14. Unless NP = RP , there are no Occam algorithms for the following hypothesis classes:
- · Deterministic Finite Automata (DFA)
- · Acyclic Deterministic Finite Automata (ADFA)
- · Ordered Branching Decision Diagram (OBDD)
In particular, for any glyph[epsilon1] ∈ (0 , 1) , k > 0 , the minimum consistent hypothesis problems for these classes are N 1 -glyph[epsilon1] OPT k -hard to approximate unless NP = RP .
## 6 Hardness of EDP on DAGs
In this section, we prove the | V ( G ) | 1 2 / -glyph[epsilon1] hardness of approximating EDP on DAGs and packing vertex-disjoint bounded length cycles. We will first show the construction for EDP, and later we argue that a slight modification of the construction yields the hardness of packing vertex-disjoint bounded length cycles.
## 6.1 Reduction R
We first define the canonical reduction -→ R G [ ] formally. Given a graph G = ( V, E ) on n vertices, the switching graph of G , denoted by -→ R G [ ], is a graph defined on a plane and constructed in two steps as follows. The coordinates of graph -→ R G [ ] lie in the box formed by the corners (0 0) and ( , n, n ).
First Step: For each vertex i ∈ V ( G ), we draw a line segment glyph[lscript] i on the plane connecting vertices s i and t i as shown in Figure 4. To be precise, the line glyph[lscript] i goes from the coordinate ( n +1 -i, 0)
to the coordinate ( n + 1 -i, i ) of the grid and then goes to the coordinate (0 , i ). For each pair of vertices i, j ∈ V ( G ), we have an intersection point y i,j at the crossing point of lines glyph[lscript] i and glyph[lscript] j . Some of these intersection points will be later defined as vertices in the switching graphs whereas others are just a crossing points in the plane embedding. We call this graph ̂ R G [ ] which will also be crucial in our analysis. Edges in ̂ R G [ ] are directed from left to right and top to bottom.
glyph[negationslash]
Second Step: For each edge ij ∈ E G ( ), we split y i,j into two vertices x in i,j and x out i,j and have a directed edge e i,j = x in i,j x out i,j in the graph -→ R G [ ]. Otherwise, if ij ∈ E G ( ), the intersection point y i,j is replaced by an uncrossing as in Figure 4.
glyph[negationslash]
Figure 4: The graph -→ R G [ ] where (3 4) , ∈ E G ( ) but (4 , 5) ∈ E G ( )
First, the following lemma establishes a (simple) connection between EDP and the maximum independent set problem.
Lemma 6.1. For any graph H , edp ( -→ R H [ ]) ≥ α H ( ) .
Proof. Let S ⊆ V ( H ) be any independent set in H . We define the collection of paths P S = { P i } i ∈ S in graph -→ R H [ ]. Since S is an independent set, any pair of paths P i and P j for i, j ∈ S are disjoint by construction.
Unfortunately, the converse of this inequality does not hold within any reasonably small factor. In fact, there is a graph H for which α H ( ) = 2 but edp ( -→ R H [ ]) = Ω( n ); see Appendix A. Therefore, we focus on proving the low α -projection property.
## 6.2 α -Projection Property
For technical reasons, we will need to analyze a slightly different measure from the optimal value edp ( -→ R G [ ]). This notion will be a weaker notion of feasible solutions for EDP. We say that a
glyph[negationslash]
collection of disjoint paths P = { P , . . . , P 1 glyph[lscript] } is orderly feasible if for any pair P = ( s , . . . , t i j ) and P ′ = ( s i ′ , . . . , t j ′ ) such that i < i ′ , then it must be the case that j < j ′ ; for instance, in an orderly feasible set, if we connect s 1 to t 3 , it must be the case that s 2 is connected to t j for j > 3. Intuitively, in an orderly feasible set P , a path is allowed to start from s i and ends at some sink t j for j = , but every pair of paths in i P is forced to 'cross' at some point. Observe that any collection of feasible edge disjoint paths must also be orderly feasible. As a consequence, if we define ˜ ˜ edp ( -→ R G [ ]) as the maximum cardinality of all orderly feasible collections of paths, then we have that edp ( -→ R G [ ]) ≥ edp ( -→ R G [ ]).
The following observation is more or less obvious.
$$\text{Observation $6.2$. $ For any graph $G$, $\widetilde{edp}(\overrightarrow{R}[G]) \leq | \overrightarrow { R }[G]|$}$$
Next, the following lemma will finish the proof of the low α -projection property.
Lemma 6.3. For any two graphs G and H ,
$$\widetilde { \text{edp} } ( \overrightarrow { R } [ G \cdot H ] ) \leq 3 | V ( G ) | ^ { 2 } + \alpha ( G ) \widetilde { \text{edp} } ( \overrightarrow { R } [ H ] )$$
We will spend the rest of this section to prove the lemma.
## 6.3 Geometry of Paths: Regions, switching boxes, and configurations
This section discusses the structure of the graph -→ R G [ · H ] and a feasible solution for EDP in -→ R G [ · H ]. We define some terminologies that will be needed in the analysis.
Ordering of Paths We need a notion of 'ordering' of edge-disjoint paths with respect to certain curve. We think of graph ̂ R G [ ] as being drawn on the plane with standard x and y coordinates. All sources and sinks are on y and x axes respectively. -→
For any collection of edge-disjoint paths P in R G [ ], one can naturally map these paths on the graph ̂ R G [ ] and think of them as curves on the plane. A continuous curve C : [0 , 1] → R 2 is said to be good if for all t < t ′ , point C t ( ) is dominated by point C t ( ′ ) in the plane and the curve C does not go through any intersection point y i,j (informally, the curve is directed to the top and right). Let C be any good curve. The ordering glyph[precedesequal] C is defined on the set of paths P ′ intersecting C as follows: Paths P ≺ C P ′ if and only if C intersects P before it intersects P ′ . Since C does not intersect point y i,j , either P ≺ C P ′ or P ′ ≺ C P .
Regions and Switching Boxes. In -→ R G [ · H ], we have canonical paths P ia for i ∈ V ( G ) and a ∈ V ( H ). For each i ∈ V ( G ), we define a region R i on the plane that contains all paths P ( i,a ) for a ∈ [ r ]. For i, j ∈ V ( G ), the intersection between regions R i and R j is called a bounding box B i, j ( ) which contains | V ( H ) | 2 virtual vertices of the form Y ( i, a ) , ( j, b ) for a, b ∈ V ( H ). Notice that a canonical path P ( i,a ) is completely contained inside region R i , and as we walk on the path from s i, a ( ) to t ( i, a ), we will visit the bounding boxes B i, ( 1) , . . . , B ( i, n ) in this order. For convenience, the region in R i between B i, i ( -1) and B i, i ( +1) is called B i, i ( ). See Figure 6 for illustration.
glyph[negationslash]
Proposition 6.4. Consider any box B i, j ( ) for i = j . One of the following two cases holds:
- · For all a, b ∈ V ( H ) , the virtual vertex y ( i,a ) ( , j,b ) is a directed edge e ( i,a ) ( , j,b ) . This happens when ij ∈ E G ( ) , and we say that the box B i, j ( ) is a non-switching box.
glyph[negationslash]
Figure 5: Both C 1 and C 2 are good curves that originated from (0 , 0) (illustrated by dotted lines). We have Q ′ ≺ C 1 Q while Q ≺ C 2 Q ′ .
- · For all a, b ∈ V ( H ) , the virtual vertex is an uncrossing, in which case, we say that the box B i, j ( ) is a switching box.
The term switching box is coined from an intuitive reason: Consider a switching box B i, j ( ) and a collection of edge-disjoint paths that are routed in a solution. Let P top and P left be the paths in the solution that enter this box from the top and left respectively, so paths in P top (resp. P left ) must leave the box from the bottom (resp. right). Define the curves C in (and C out ) as the union of left and top boundaries of B i, j ( ) (resp. the union of right and bottom boundaries). With respect to the curve C in all paths in P top are ordered after paths in P left , while this becomes the opposite for C out . In other words, the box B i, j ( ) 'switches' the order of these paths.
## 6.4 Proof
Now we prove Lemma 6.3. Let I be the set of indices of edge-disjoint paths in -→ R G [ · H ] where, for each ( i, a ) ∈ I , there is a path Q ( i,a ) connecting s i, a ( ) to t ( ψ i, a ( )) and the paths Q ( i,a ) are edge-disjoint and orderly feasible (recall that orderly feasible solutions may connect s i, a ( ) to some other sink t ( i , a ′ ′ )). We say that a path Q ( i,a ) is semi-canonical if it is completely contained in region R i . Let I ′ ⊆ I be the set of semi-canonical paths.
$$\text{Lemma } 6. 5. \ | I ^ { \prime } | \leq \alpha ( G ) \widetilde { \text{edp} } ( \overrightarrow { R } [ H ] )$$
glyph[negationslash]
Proof. Wefirst define the partition of I ′ by the first coordinates of paths. Define I ′ i = ( { i, a ) : ( i, a ) ∈ I ′ } , so we will have I ′ = ⋃ i ∈ V ( G ) I ′ i . We count the number of indices i ∈ V ( G ) such that I ′ i = ∅ . Define Λ = { i ∈ V ( G ) : I ′ i = ∅} . We claim that Λ is an independent set and therefore | Λ | ≤ α G ( ): Suppose i, j ∈ V ( G ) such that I , I ′ i ′ j = ∅ . Let ( i, a ) ∈ I ′ i and ( j, b ) ∈ I ′ j be any two paths. Due to the fact that this is an orderly feasible solution, these two paths must cross at some virtual vertex y ( i,a ′ ) ( , j,b ′ ) inside box B i, j ( ), and they must share an edge e ( i,a ′ ) ( , j,b ′ ) , a contradiction. This implies that | Λ | ≤ α G ( ).
glyph[negationslash]
Figure 6: Regions and switching boxes in glyph[vector] R G [ ]. There are two paths routed inside region R 2 .
glyph[negationslash]
Next, we argue that | I ′ i | ≤ ˜ edp ( -→ R H [ ]) for all i ∈ Λ, which will complete the proof of the lemma. If we consider the box B i, i ( ), we see an isomorphic copy H ′ of graph -→ R H [ ], in which each path Q ( i,a ) corresponds to another path Q ′ ϕ a ( ) that connects some 'source' s ′ ( ϕ a ( )) to 'sink' t ′ ( ξ a ( )). We claim that the collection of paths Q ′ ϕ a ( ) is orderly feasible in the instance ( H , ′ { ( s ′ ( a , t ) ′ ( a ) } a ∈ V ( H ) ): Assume otherwise that some paths Q ′ ϕ a ( ) and Q ′ ϕ b ( ) do not cross inside H ′ , so the paths Q ( i,a ) and Q ( i,b ) must cross at some other box B i, j ( ) for i = . If such box is a switching box, it is impossible j for these two paths to cross because they must enter and leave the box from the same direction; otherwise, if box B i, j ( ) is not a switching box, it is also impossible for them to cross.
Let I ′′ = I \ I ′ . For convenience, let us renumber I ′′ such that I ′′ = { 1 , . . . , t } such that the source of path 1 is above that of path 2 and so on. Now we will show that | I ′′ | ≤ 3 | E G ( ) | + 1. Our proof will rely on the notion of configurations. We first define the order of boxes B i, j ( ) for i > j such that B (2 , 1) is the first box, which precedes B (3 1) (the second box), and so on. , More formally, the box B i, j ( ) precedes B i , j ( ′ ′ ) if and only if i < i ′ or i = i ′ and j < j ′ ; in short, this is simply a lexicographic order of boxes. This defines a total order over boxes.
We define a number of good curves C , . . . , C 1 z for z = ( | V ( G ) | 2 ) , where the curve C h is any good curve such that (i) C h (0) = (0 , 0), (ii) C h (1) = ( x max , y max ) and (iii) the first h -1 boxes are above C h , while z -h +1 curves are below it (notice that C h partitions the region [0 , n +1] × [0 , n +1] into two parts, i.e. one above the curve and the other below it).
Observation 6.6. For each h = 1 , . . . , z and path i ∈ I ′′ , the curve C h intersects path i ∈ I ′′ .
Proof. This is just because any path in the orderly feasible solution starts from the region above the curve C h , while it ends in the region below the curve.
For each h = 1 , . . . , z , a curve C h can be used to define a configuration σ h = ( x , . . . , x 1 t ) of paths in I ′′ where x j ∈ I ′′ is the index of the j th path that intersects with the curve C h (this order is well-defined because the curve has directions). Notice that σ 1 = ( t, . . . , 1), and σ z = (1 , . . . , t ).
The number of reversals of a configuration σ h is the number of locations j such that σ j > σ j +1 . Denote this number by rev σ ( h ), so we have that rev σ ( 1 ) = t -1, and rev σ ( z ) = 0. Our proof proceeds by analyzing how the number of reversals changes over configurations σ , . . . , σ 1 z . We will show that, for any h , we have rev σ ( h ) -rev σ ( h +1 ) ≤ 3, which implies that t -1 = rev σ ( 1 ) -rev σ ( z ) = ∑ z -1 h =1 ( rev σ ( h ) -rev σ ( h +1 )) ≤ 3( z -1); in other words, | I ′′ | = t ≤ 3 z ≤ 3 | V ( G ) | 2 . So the last thing we need to prove is the following lemma:
Lemma 6.7. For any h = 1 , . . . , z -1 , we have rev σ ( h ) -rev σ ( h +1 ) ≤ 3 .
Proof. Let B i, j ( ) be the h th box and J ⊆ I ′′ be the indices of paths entering this box. If the box B i, j ( ) is a non-switching box, then it must be the case that σ h +1 = σ h due to the fact that paths cannot cross inside region B i, j ( ). This implies that rev σ ( h ) = rev σ ( h +1 ) in this case.
Now we consider the other case when B i, j ( ) is a switching box. We write J = J top ∪ J left where J top (and J left ) is the set of indices of paths entering box B i, j ( ) from the top (and left respective). It is clear that paths coming out of the bottom and right of the box are exactly J top and J left respectively. Notice that, while the curve C h crosses J top after J left , the curve C h +1 would cross paths in J left before those in J top . The configurations σ h and σ h +1 can be written as σ h = σ ′ ◦ σ left ◦ σ top ◦ σ ′′ and σ h +1 = σ ′ ◦ σ top ◦ σ left ◦ σ ′′ respectively. See Figure 7 for illustration.
Figure 7: Configurations of curve C h and C h +1 are σ h = (4 2 1 3) and , , , σ h +1 = (2 1 4 3) respec-, , , tively. In this case, σ top = (2 1) and , σ left = (4). Box B i, j ( ) is a switching box.
## 7 Other Problems
In this section, we prove the hardness of k -Cycle Packing and DNF/CNF Minimization. As noted previously, our proof for CNF minimization is an alternative proof of Aleknovich et al. [ABF + 08].
## 7.1 Hardness of k -Cycle Packing for Large k
glyph[negationslash]
We consider the problem of packing edge-disjoint k -cycles in which our goal is to pack as many cycles of length at most k as possible. We only need to slightly change the reduction -→ R G [ ] as used in Section 6 in the following way: In the second step, for each pair i, j ∈ V ( G ), if ij ∈ E G ( ), we do the same, but for ij ∈ E G ( ) (including the case when i = ), we make two new vertices on each line j before and after the jump (see Figure 8). Also, we have a back edge from t i to s i for each i ∈ V ( G ).
glyph[negationslash]
Figure 8: The graph -→ R G [ ] where (3 4) , ∈ E G ( ) but (4 , 5) ∈ E G ( ). The differences between this gadget and EDP gadget only lies in the new vertices a in i,j , a out i,j , b in i,j , b out i,j
With this reduction, any 'canonical' cycle between source s i to sink t i (and taking back edge to s i ) must have length exactly 2 n +2, so we choose the value k = 2 n +2. Let edc ( -→ R G ′ [ ]) denote the optimal value of k -cycle packing. We now establish the connection between the optimal value of EDP solution in -→ R G [ ] and the k -EDC solution in -→ R G ′ [ ].
Notice that for any cycle that uses only one back edge, there is a corresponding path from some s i to t i . The number of these cycles corresponds exactly to EDP ( -→ R G [ ]), so we can write
$$\text{EP} ( \overrightarrow { R } [ G ] ) \leq e d c ( \overrightarrow { R } ^ { \prime } [ G ] ) \leq \text{EP} ( \overrightarrow { R } [ G ] ) + \widetilde { e d c } ( \overrightarrow { R } ^ { \prime } [ G ] )$$
where ˜ edc ( -→ R G ′ [ ]) is the number of cycles that use more than one back edge. The following lemma says that these cycles must be longer than k , i.e. ˜ edc ( -→ R G ′ [ ]) = 0. In other words, this implies that edc ( -→ R G ′ [ ]) = EDP ( -→ R G [ ]).
Lemma 7.1. Let C be a cycle in -→ R G ′ [ ] that uses more than one back edge. Then | C > k | .
Proof. Any cycle C must start at some s i and ends at s i . Let i = i 0 , i 1 , . . . , i glyph[lscript] = i be the indices of the source-sink pairs visited in the cycle C , i.e. the cycle goes through s i 0 → t i 1 s i 1 → s i 2 → . . . →
t i glyph[lscript] s i glyph[lscript] . Observe that a path that goes from s j to t j ′ visit exactly ( n -j + j ′ ) vertices of the form y i,j , because such path must go right j ′ times and go down n -j times (in arbitrary order). Combining this with s j and t j ′ , such path would visit 2( n -j + j ′ +1) vertices. Therefore, the total length of the cycle C is
$$\cdots \omega \sum _ { j = 0 } ^ { \ell - 1 } 2 ( n - i _ { j } + i _ { j + 1 } + 1 ) = 2 \ell ( n + 1 ) + 2 ( i _ { \ell } - i _ { 0 } ) = 2 \ell ( n + 1 )$$
So this cycle would have been longer than the threshold k if glyph[lscript] > 1.
## 7.2 Learning CNF Formula
We present an alternative proof for the hardness of properly learning CNF using our framework. Our reduction is quite similar to Alekhnovich et al.'s (see [ABF + 08]), but our proof highlights the role of graph products in the proof (while their construction cannot be seen as a standard graph product in any way).
Let G be any graph. We think of a vertex u ∈ V ( G ) as an integer in { 1 , . . . , n } . For each vertex u ∈ V ( G ), we define an encoding 〈 u 〉 = 0 u -1 10 n -u . For each edge uv ∈ V ( G ), the encoding of an edge uv has two 1s at the positions corresponding to u and v . Our reduction encodes the k -fold graph product H = G k into samples as follows. For each glyph[vector] u = ( u , . . . , u 1 k ) ∈ V ( H ), we define a negative sample neg glyph[vector] ( u ) = 〈 u 1 〉 . . . 〈 u k 〉 . For each glyph[vector] u ∈ E H ,i ( ) ∈ [ k ] and u v i ∈ E G ( ), we define a positive sample pos glyph[vector] , ( u v, i ) = 〈 u 1 〉 . . . 〈 u i -1 〉〈 u v i 〉〈 u i +1 〉 . . . 〈 u k 〉 .
Notice that the total number of variables is nk , where we think of them as k blocks; in each of which, there are n variables. Denote by z i, u ( ) the variable in block i ∈ [ k ] that corresponds to a vertex u ∈ V ( G ).
## Lemma 7.2. OPT ( R H [ ]) ≥ χ H ( )
Proof. Suppose ∧ M j =1 C j be a CNF formular that is consistent with all samples. We claim that the number of clauses M is at least χ H ( ). For each glyph[vector] u ∈ V ( H ), let σ glyph[vector] ( u ) be the index such that C σ glyph[vector] ( u ) evaluates to false on sample neg glyph[vector] ( u ); this clause must exist since this is a negative sample (if there are many such indices for vertex glyph[vector] u , we choose any arbitrary one). Now for each s ∈ [ M ], we define the set of vertices Q s ⊆ V ( H ) as Q s = { glyph[vector] u : σ glyph[vector] ( u ) = s } .
## Claim 7.3. Q s is an independent set for all s ∈ [ M ] .
Proof. Assume otherwise that some glyph[vector] uglyph[vector] v ∈ E H ( ) such that glyph[vector] u,glyph[vector] v ∈ Q s . Let i be the index such that u j = v j for all j < i and u v i i ∈ E G ( ). Let X,Y ⊆ [ k ] × V ( G ) be the subset of variable indices that appear positively and negatively in clause C s , so we can rewrite C s = ( ∨ ( j,u ) ∈ X z j, u ( ) ) ∨ ( ∨ ( j,u ) ∈ Y z j, u ( ) ) . Since C s evaluates to false on both neg glyph[vector] ( u ) and neg glyph[vector] ( v ), we can neither have variable z i, u ( i ) nor z i, v ( i ) in the clause C s : Suppose otherwise that ( i, u i ) ∈ X or ( i, u i ) ∈ Y , then either neg glyph[vector] ( u ) or neg glyph[vector] ( v ) would have evaluated to true on clause C s (contradicting glyph[vector] u ∈ Q s ). Similarly, if ( i, v i ) ∈ X or ( i, v i ) ∈ Y , then either neg glyph[vector] ( v ) or neg glyph[vector] ( u ) would have been true in clause C s .
glyph[negationslash]
glyph[negationslash]
In other words, ( i, u ) , ( i, v ) ∈ X ∪ Y . But notice that pos glyph[vector] , ( u v , i i ) ( i ,u ′ ′ ) = neg glyph[vector] ( u ) ( i ,u ′ ′ ) for all ( i , u ′ ′ ) ∈ { ( i, u i ) , ( i, v i ) } , so C s must evaluate to false on input pos glyph[vector] , ( u v , i i ), a contradiction.
We have just shown that { Q s } s ∈ [ M ] is a valid M -coloring for graph H , so we must have M ≥ χ H ( ), as desired.
Next, we prove the upper bound.
Lemma 7.4. OPT ( R H [ ]) ≤ χ G ( ) k n O (1)
glyph[negationslash]
Proof. We construct the same formula as in Aleknovich et al. That is, let I , . . . , I 1 M be color classes of G and σ : V ( G ) → [ M ] be the corresponding coloring function. Define the formula f i ( z ) = ∧ M c =1 ∨ u ∈ I c z i, u ( ), for i = 1 2 , , . . . , k , and define f ( z ) = ∨ k i =1 f i ( z ). This formula can be turned into a CNF of size at most χ G ( ) k | V ( G ) | O (1) .
Claim 7.5. The formula f is consistent with all the samples.
glyph[negationslash]
Proof. Consider each negative sample neg glyph[vector] ( u ) for glyph[vector] u = ( u , . . . , u 1 k ) ∈ V ( H ). For each i ∈ [ k ], notice that f i ( 〈 u i 〉 ) evaluates to false because ∨ u ∈ I σ u i ( ) 〈 u i 〉 u is false (since u i ∈ I σ u ( i ) is the only bit of 〈 u i 〉 that is '1'). This implies that ∨ k i =1 f i ( neg glyph[vector] ( u )) is false.
glyph[negationslash]
glyph[negationslash]
Now consider, for each glyph[vector] u ∈ V ( H ), v : u v i ∈ E G ( ) and i ∈ [ k ], a positive sample pos glyph[vector] , ( u v, i ). We claim that f i ( pos glyph[vector] , ( u v, i )) is true, which causes f ( pos glyph[vector] , ( u v, i )) to be true: Assume to the contrary that f i is false, so some term ∨ w I ∈ c 〈 u v i 〉 w is false for some c ; notice that it can be false only if 〈 u v i 〉 w = 0 for all w ∈ I c ; since we have 〈 u v i 〉 u i = 〈 u v i 〉 v = 1, it must be the case that both u i and v belong to I c , contradicting the fact that I c is a color class.
## 8 Conclusion and Open Problems
We have shown applications of pre-reduction graph product techniques in proving hardness of approximation. For some applications, such as EDP, proving α -projection property implies tight hardness, but for some others, we need a more careful reduction of the form R G [ glyph[lscript] ] (taking into account the fact that the input is an glyph[lscript] -fold product of graphs).
There are many open problems on edge-disjoint paths. Most notably can one narrow down the gap of undirected EDP between O ( √ n ) upper bound and log 1 2 / -glyph[epsilon1] n lower bound? For directed EDP, there is still a (relatively large) gap in the case of low congestion routing, between the upper bound of n 1 /c [KS04] and the lower bound of n 1 (3 +23) / c [CGKT07] if we allow routing with congestion c . We believe that our techniques are likely to work there (in a much more sophisticated manner), and it would potentially close this gap. This would resolve an open question in Chuzhoy et al. [CGKT07].
Another interesting problem is the cycle packing problem. For this problem, the approximability is pretty much settled on undirected graphs with an upper bound of O (log 1 2 / n ) and a lower bound of log 1 2 / -glyph[epsilon1] n [FS11, KNS + 07]. On directed graphs, there is still a large gap between n 1 2 / and Ω( log n log log n ). For k -cycle packing problem, it is interesting to see whether our technique gives k 1 -glyph[epsilon1] hardness for small k .
Acknowledgement: We thank Julia Chuzhoy for suggesting the EDP reduction and for related discussions when the first author was still at the University of Chicago.
## References
| [Aar08] | Scott Aaronson. 6.080/6.089 Great Ideas in Theoretical Computer Sci- ence, Spring 2008, Lecture 21. MIT OpenCourseWare, 2008. Available at http://stellar.mit.edu/S/course/6/sp08/6.080/courseMaterial/topics/ topic1/lectureNotes/lec21/lec21.pdf . |
|------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [ABF + 08] | Michael Alekhnovich, Mark Braverman, Vitaly Feldman, Adam R. Klivans, and Toniann Pitassi. The complexity of properly learning simple concept classes. J. Comput. Syst. Sci. , 74(1):16-34, 2008. Announced at FOCS 2004. |
| [ABX08] | Benny Applebaum, Boaz Barak, and David Xiao. On basing lower-bounds for learning on worst-case assumptions. In FOCS , pages 211-220, 2008. |
| [ACG + 10] | Matthew Andrews, Julia Chuzhoy, Venkatesan Guruswami, Sanjeev Khanna, Kunal Talwar, and Lisa Zhang. Inapproximability of edge-disjoint paths and low congestion routing on undirected graphs. Combinatorica , 30(5):485-520, 2010. |
| [Ang78] | Dana Angluin. On the complexity of minimum inference of regular sets. Information and Control , 39(3):337-350, 1978. |
| [AZ06] | Matthew Andrews and Lisa Zhang. Logarithmic hardness of the undirected edge-disjoint paths problem. J. ACM , 53(5):745-761, 2006. Also, in STOC'05. |
| [BP92] | Raymond Board and Leonard Pitt. On the necessity of occam algorithms. Theoretical Computer Science , 100(1):157 - 184, 1992. |
| [BS92] | Piotr Berman and Georg Schnitger. On the complexity of approximating the indepen- dent set problem. Inf. Comput. , 96(1):77-94, 1992. Also, in STACS'89. |
| [CGKT07] | Julia Chuzhoy, Venkatesan Guruswami, Sanjeev Khanna, and Kunal Talwar. Hardness of routing with congestion in directed graphs. In STOC , pages 165-178, 2007. |
| [Chu12] | Julia Chuzhoy. Routing in undirected graphs with constant congestion. In STOC , pages 855-874, 2012. |
| [CK07] | Chandra Chekuri and Sanjeev Khanna. Edge-disjoint paths revisited. ACM Transac- tions on Algorithms , 3(4), 2007. |
| [CKS05] | Chandra Chekuri, Sanjeev Khanna, and F. Bruce Shepherd. Multicommodity flow, well-linked terminals, and routing problems. In STOC , pages 183-192, 2005. |
| [CKS06] | Chandra Chekuri, Sanjeev Khanna, and F. Bruce Shepherd. An O ( √ n ) approximation and integrality gap for disjoint paths and unsplittable flow. Theory of Computing , 2(1):137-146, 2006. |
| [CKS09] | Chandra Chekuri, Sanjeev Khanna, and F. Bruce Shepherd. Edge-disjoint paths in planar graphs with constant congestion. SIAM J. Comput. , 39(1):281-301, 2009. |
| [CL12] | Julia Chuzhoy and Shi Li. A polylogarithmic approximation algorithm for edge-disjoint paths with congestion 2. In FOCS , pages 233-242, 2012. |
| [CLN13a] | Parinya Chalermsook, Bundit Laekhanukit, and Danupon Nanongkai. Graph products revisited: Tight approximation hardness of induced matching, poset dimension and more. In SODA , pages 1557-1576, 2013. |
|------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [CLN13b] | Parinya Chalermsook, Bundit Laekhanukit, and Danupon Nanongkai. Independent set, induced matching, and pricing: Connections and tight (subexponential time) approxi- mation hardnesses. In FOCS , pages 370-379, 2013. |
| [CLN14] | Parinya Chalermsook, Bundit Laekhanukit, and Danupon Nanongkai. Coloring graph powers: Graph product bounds and hardness of approximation. In LATIN , pages 409- 420, 2014. |
| [DlH10] | Colin De la Higuera. Grammatical inference: learning automata and grammars . Cam- bridge University Press, 2010. |
| [DLSS14] | Amit Daniely, Nati Linial, and Shai Shalev-Shwartz. From average case complexity to improper learning complexity. In STOC , pages 441-448, 2014. |
| [Fei02] | Uriel Feige. Relations between average case complexity and approximation complexity. In STOC , pages 534-543, 2002. |
| [Fel08] | Vitaly Feldman. Hardness of proper learning. In Encyclopedia of Algorithms . Springer, 2008. |
| [FK98] | Uriel Feige and Joe Kilian. Zero knowledge and the chromatic number. J. Comput. Syst. Sci. , 57(2):187-199, 1998. Also, in CCC 1996. |
| [FS11] | Zachary Friggstad and Mohammad R. Salavatipour. Approximability of packing disjoint cycles. Algorithmica , 60(2):395-400, 2011. |
| [GKR + 03] | Venkatesan Guruswami, Sanjeev Khanna, Rajmohan Rajaraman, F. Bruce Shepherd, and Mihalis Yannakakis. Near-optimal hardness results and approximation algorithms for edge-disjoint paths and related problems. J. Comput. Syst. Sci. , 67(3):473-496, 2003. Also, in STOC 1999. |
| [GL14] | Venkatesan Guruswami and Euiwoong Lee. Inapproximability of feedback vertex set for bounded length cycles. Electronic Colloquium on Computational Complexity (ECCC) , 21:6, 2014. |
| [Gol78] | E. Mark Gold. Complexity of automaton identification from given data. Information and Control , 37(3):302-320, 1978. |
| [H˚s96] a | Johan H˚stad. a Clique is hard to approximate within n 1 - glyph[epsilon1] . In FOCS , pages 627-636, 1996. |
| [KK10] | Kenichi Kawarabayashi and Yusuke Kobayashi. The edge disjoint paths problem in eulerian graphs and 4-edge-connected graphs. In SODA , pages 345-353, 2010. |
| [Kle96] | Jon Michael Kleinberg. Approximation algorithms for disjoint paths problems . PhD thesis, Citeseer, 1996. |
| [Kle05] | Jon M. Kleinberg. An approximation algorithm for the disjoint paths problem in even- degree planar graphs. In FOCS , pages 627-636, 2005. |
|------------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [KMR97] | David R. Karger, Rajeev Motwani, and G. D. S. Ramkumar. On approximating the longest path in a graph. Algorithmica , 18(1):82-98, 1997. |
| [KNS + 07] | Michael Krivelevich, Zeev Nutov, Mohammad R. Salavatipour, Jacques Yuster, and Raphael Yuster. Approximation algorithms and hardness results for cycle packing prob- lems. ACM Transactions on Algorithms , 3(4), 2007. |
| [KS04] | Stavros G. Kolliopoulos and Clifford Stein. Approximating disjoint-path problems using packing integer programs. Math. Program. , 99(1):63-87, 2004. |
| [KT98] | Jon M. Kleinberg and ´ Eva Tardos. Approximations for the disjoint paths problem in high-diameter planar networks. J. Comput. Syst. Sci. , 57(1):61-73, 1998. |
| [KV94] | Michael J. Kearns and Leslie G. Valiant. Cryptographic limitations on learning boolean formulae and finite automata. J. ACM , 41(1):67-95, 1994. Announced at STOC 1989. |
| [LV88] | Ming Li and Umesh V. Vazirani. On the learnability of finite automata. In COLT , pages 359-370, 1988. |
| [Pit89] | Leonard Pitt. Inductive inference, DFAs, and computational complexity . Springer, 1989. |
| [PV88] | Leonard Pitt and Leslie G. Valiant. Computational limitations on learning from exam- ples. J. ACM , 35(4):965-984, 1988. |
| [PW93] | Leonard Pitt and Manfred K. Warmuth. The minimum consistent DFA problem cannot be approximated within any polynomial. J. ACM , 40(1):95-142, 1993. Announced at STOC 1989. |
| [RS95] | Neil Robertson and Paul D. Seymour. Graph minors .xiii. the disjoint paths problem. J. Comb. Theory, Ser. B , 63(1):65-110, 1995. |
| [VV04] | Kasturi R. Varadarajan and Ganesh Venkataraman. Graph decomposition and a greedy algorithm for edge-disjoint paths. In SODA , pages 379-380, 2004. |
| [Zuc07] | David Zuckerman. Linear degree extractors and the inapproximability of max clique and chromatic number. Theory of Computing , 3(1):103-128, 2007. Also, in STOC 2006. |
## Appendix
## A List of Bad Examples
In this section, we provide the evidences that all the reductions considered in this paper are neither trivially 'working' nor subadditive in the sense of our previous SODA paper [CLN13a]. Therefore, we will need the new conceptual ideas introduced in this paper.
## A.1 Edge Disjoint Paths
We show a graph G in which α G ( ) = 2 but EDP ( -→ R G [ ]) = n/ 3. To ensure that G does not have an independent set of size 3, we define graph G by defining a triangle-free graph H and let G = K n \ H . We consider a set of vertices V ( G ) = A ∪ B ∪ C , where | A | = { 1 , . . . , n/ 3 } , B = { n/ 3 + 1 , . . . , 2 n/ 3 } and C = { 2 n/ 3 + 1 , . . . , n } . Graph H only have edges between A and B in such a way that, for any i ∈ A and j ∈ B , we have ij ∈ E H ( ) if and only if i -j > n/ 3. It is obvious that H is bipartite, so if we define G = K n \ H , we have α G ( ) = 2.
Figure 9: Bad Example for EDP reduction
Now we check that EDP ( -→ R G [ ]) = n/ 3. For each i ∈ [ n ], let E i denote the set of edges on the canonical path from s i to t i in -→ R G [ ]. For each i ∈ B , we define a path P i that:
- · Start at s i and go straight until it meets with an edge in E i -n/ 3 , at which the path turns downward (the turning is possible because we have an edge i ( i -n/ 3) ∈ E G ( )).
- · The path goes downward until it meets with an edge in E n -i + n/ 3 , at which the path turns again towards the right.
- · Once path P i meets with an edge in E i again, it takes a turn downward and remains so until it reaches t i .
Observation A.1. For any i, j ∈ B P , i ∩ P j = ∅ .
| null |
[
"Parinya Chalermsook",
"Bundit Laekhanukit",
"Danupon Nanongkai"
] |
2014-08-04T22:01:59+00:00
|
2014-08-04T22:01:59+00:00
|
[
"cs.CC"
] |
Pre-Reduction Graph Products: Hardnesses of Properly Learning DFAs and Approximating EDP on DAGs
|
The study of graph products is a major research topic and typically concerns
the term $f(G*H)$, e.g., to show that $f(G*H)=f(G)f(H)$. In this paper, we
study graph products in a non-standard form $f(R[G*H]$ where $R$ is a
"reduction", a transformation of any graph into an instance of an intended
optimization problem. We resolve some open problems as applications.
(1) A tight $n^{1-\epsilon}$-approximation hardness for the minimum
consistent deterministic finite automaton (DFA) problem, where $n$ is the
sample size. Due to Board and Pitt [Theoretical Computer Science 1992], this
implies the hardness of properly learning DFAs assuming $NP\neq RP$ (the
weakest possible assumption).
(2) A tight $n^{1/2-\epsilon}$ hardness for the edge-disjoint paths (EDP)
problem on directed acyclic graphs (DAGs), where $n$ denotes the number of
vertices.
(3) A tight hardness of packing vertex-disjoint $k$-cycles for large $k$.
(4) An alternative (and perhaps simpler) proof for the hardness of properly
learning DNF, CNF and intersection of halfspaces [Alekhnovich et al., FOCS 2004
and J. Comput.Syst.Sci. 2008].
|
1408.0829v1
|
## A GLIMPSE AT QUASAR HOST GALAXY FAR-UV EMISSION, USING DLAS AS NATURAL CORONAGRAPHS
Zheng Cai 1,2 , Xiaohui Fan 1 , Pasquier Noterdaeme 3 , Ran Wang 1 , Ian McGreer 1 , Bill Carithers 4 , Fuyan Bian 5,13 , Jordi Miralda Escude 6 , Hayley Finley 3 , Isabelle P ˆ a ris 7 , Donald P. Schneider 8, 9 , Nadia L. Zakamska 12 , Jian Ge 10 , Patrick Petitjean 3 , Anze Slosar 11
Steward Observatory, University of Arizona, Tucson, AZ 85721
Physics Department, University of Arizona, Tucson, AZ 85721
Institut d'Astrophysique de Paris, CNRS-UPMC, UMR7095, 98bis bd Arago, 75014 Paris, France
Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA, USA
Research School of Astronomy and Astrophysics, Australian National University, Canberra, Weston Creek, ACT, 2611
Institut de Ci`ncies del Cosmos, Universitat de Barcelona, Spain e
Departamento de Astronomia, Universidad de Chile, Casilla 36-D
Department of Astronomy and Astrophysics, The Pennsylvania State University, University Park, PA 16802 Institute for Gravitation and the Cosmos, The Pennsylvania State University, University Park, PA 16802 Department of Astronomy, University of Florida, USA
Brookhaven National Laboratory, Upton, NY 11973, USA
Department of Physics & Astronomy, Johns Hopkins University, 3400 N. Charles St., Baltimore, MD 21218, USA and
Stromlo Fellow
Accepted for publication in The Astrophysical Journal
## ABSTRACT
In merger-driven models of massive galaxy evolution, the luminous quasar phase is expected to be accompanied by vigorous star formation in quasar host galaxies. In this paper, we use high column density Damped Lyman Alpha (DLA) systems along quasar sight lines as natural coronagraphs to directly study the far-UV (FUV) radiation from the host galaxies of luminous background quasars. We have stacked the spectra of ∼ 2,000 DLA systems ( N HI > 10 20 6 . cm -2 ) with a median absorption redshift 〈 z 〉 = 2 6 selected from quasars observed in the SDSS-III Baryon Oscillation Spectroscopic . Survey. We detect residual flux in the dark troughs of the composite DLA spectra. The level of this residual flux significantly exceeds systematic errors in the SDSS fiber sky subtraction; furthermore, the residual flux is strongly correlated with the continuum luminosity of the background quasar, while uncorrelated with DLA column density or metallicity. We conclude that the flux could be associated with the average FUV radiation from the background quasar host galaxies (with medium redshift 〈 z 〉 = 3 1) that is not blocked by the intervening DLA. Assuming all of the detected flux originates . from quasar hosts, for the highest quasar luminosity bin ( 〈 L 〉 =2.5 × 10 13 L glyph[circledot] ), the host galaxy has a FUV intensity of 1 5 . ± 0 2 . × 10 40 erg s -1 ˚ A -1 ; this corresponds to an unobscured UV star formation rate of 9 M glyph[circledot] yr -1 .
## 1. INTRODUCTION
It has been shown that, in the local universe, most luminous and massive spheroidal galaixes have central super-massive black holes (SMBHs) (e.g. McLure et al. 1999), and that fundamental relationships exist between the black hole mass and bulge stellar mass (e.g. Magorrian et al. 1998; Ferrarese & Merritt 2000; Gebhardt et al. 2000; Marconi & Hunt 2003; H¨ring & Rix 2004). a Likelywise, many high-redshift quasars are associated with massive host galaxies (e.g. Aretxaga et al. 1998; Carilli et al. 2001). The study of high redshift quasar host galaxies opens up an important avenue to study the assembly and evolution of massive galaxies, in particular the relationship to the nuclear black holes (e.g. Schramm et al. 2008).
ultra-luminous infrared galaxy (ULIRG) phase and a dust-ejection phase (e.g., Hopkins et al. 2006). The study of SFR and dust obscuration in quasar hosts would provide strong tests of this merger-driven evolutionary model, and cast light on the relationship between quasar hosts and ULIRGs. However, progress in this field has been hindered by the difficulty of isolating quasar host galaxies, since such observations are invariably hindered by the immense brightness of the quasar nuclei in the rest-frame UV/optical regime.
In the merger-driven evolutionary model, quasars are triggered by interactions or mergers of gas-rich galaxies (e.g. Sanders et al. 1988; Hopkins et al. 2006). The interaction of galaxies produces inflows of gas, which simultaneously fuels both intense star formation in the host galaxies and black hole growth (Hopkins et al. 2006). The optically luminous quasar phase follows a dusty
Observations at millimeter (mm) and sub-mm wavelengths, which correspond to the rest-frame far-IR (FIR), are crucial for probing high-redshift quasar hosts, because the FIR signal traces the large molecular gas reservoirs which fuel star formation (e.g. Carilli & Wang 2006), and the contribution from the star formation can be disentangled from the AGN activity in the FIR regime (e.g. Yun et al. 2000). Recent observations with the Atacama Large Millimiter/submillimeter Array (ALMA) suggest that the z ∼ 6 quasar hosts have dynamical masses M dyn = 10 10 -10 11 M , one order of magnitude glyph[circledot] higher than that of typical local galaxies (Wang et al. 2013); and that the star formation rate (SFR) of the quasar hosts ranges from a few hundred to one thousand
M glyph[circledot] yr -1 (e.g. Wang et al. 2011).
At shorter wavelengths, recent surveys of quasar hosts have concentrated on the near-infrared, at rest-frame wavelengths greater than the Balmer break, and where the fraction of the observed light due to the host galaxy is larger than that in the rest-frame UV (e.g. Guyon et al. 2006). Hubble Space Telescope (HST) and ground-based observations with excellent seeing conditions ( < 0 4 . ′′ ) or near-IR adaptive optics (AO) have been used to successfully detect the rest-frame optical light of the quasar host galaxies (Kukula et al. 2001; Ridgway et al. 2001; Peng et al. 2006; Targett et al. 2013; Young et al. 2013). However, at z > 1, obtaining reliable measurements of the host galaxy luminosities remains extremely difficult, because of the severe effect of surface brightness dimming at high redshift and the technical difficulties associated with the construction of a precise PSF (e.g. Hutchings et al. 2002; Targett et al. 2013).
There has been much interest in directly tracing the formation of massive young stars in high redshift galaxies in the UV (e.g. Shapley et al. 2005; Bian et al. 2012). However, it is even more difficult to resolve and detect the high redshift quasar hosts at the rest-frame UV regime with current generation telescopes, especially in the farUV (FUV) regime ( λ ∼ 1000 ˚ ). A The contrast between the bright nuclear point source and the host galaxy increases dramatically beyond z > 1, and the FUV light from host galaxies is inevitably swamped by the UVbright nuclei. Reliable measurements of the FUV emission of the host galaxies are extremely difficult using the traditional PSF subtraction method. Recent deep HST observations did not detect the host galaxy of one z = 6 quasar in the rest-frame near-UV regime, even with delicate observations designed to minimize PSF variations and provide careful PSF subtraction (Mechtley et al. 2012). New techniques, such as coronagraphs, must be developed to block out the UV light from central AGN, and in doing so to reveal the UV emission from host galaxies (e.g. Martel et al. 2003; Finley et al. 2013).
Damped Ly α absorption (DLAs) systems at the redshift of the quasar, hereafter called associated DLAs, have recently been developed as a technique for probing quasar hosts in the UV regime (Hennawi et al. 2009; Zafar et al. 2011). The idea is that some associated DLAs completely absorb the strong Ly α emission from the central AGN, but do not fully obscure the larger-scale extended Ly α emission or Lyman continuum from the host galaxy. Therefore, associated DLAs can act as natural coronagraphs and enable the study of the Ly α emission from the quasar hosts (Finley et al. 2013). Utilizing longslit spectroscopy, Zafar et al. (2011) reported the detection of the Lyman continuum of the host galaxy in the dark trough of the DLA in front of quasar Q0151+048A at z = 1 9. .
In this paper, we demonstrate a novel techique for probing the FUV emission from quasar host galaxies. By stacking a large number of intervening DLAs in the foreground of quasars from the SDSS-III Baryon Oscillation Spectroscopic Survey (BOSS), we are able to detect the FUV emission of the quasar hosts within the stacked DLA dark trough. The HI clouds in galactic or circumgalactic environments are likely to be clumpy and the HI volume filling factor is expected to be signifi- cantly smaller than unity (e.g. Braun & Walterbos 1992; Borthakur et al. 2011). Therefore the clumpy HI gas may only block the quasar continuum, whereas much of the host remains unobscured. In § 2, we discuss the observations and our DLA sample selection. The results of the composite spectra are presented in § 3. We discuss the implication of our results on the origins of the flux in the composite DLA trough in § 4. Throughout this paper, we adopt a cosmology based upon the fifth year Wilkinson Microwave Anisotropy Probe (WMAP) data (Komatsu et al. 2009): Ω Λ = 0 72, Ω . m = 0 28, Ω . b = 0 046, and . H = 70 km s 0 -1 Mpc -1 .
## 2. METHOD AND SAMPLE
## 2.1. Using intervening DLAs as Coronagraphs
DLAs are characterized by a wide flat trough with zero transmitted flux (i.e., dark trough or dark core), resulting from the natural broadening of the absorption at high HI column densities. In this study, we define the DLA dark trough as a region where the flux is negligibly small; specifically, regions where the ratio of the expected flux to the continuum flux is < 1 × 10 -4 . Intervening DLAs can act as natural coronagraphs, because the quasar continuum, emitted from the centrally bright nucleus, is completely absorbed within the DLA dark trough. However, it is quite likely that regions of the extended emission from the quasar hosts is unobstructed by the intervening clumpy HI absorbers, and the extended stellar light thus leaks through and can be detected within the dark trough of the foreground DLAs.
## 2.2. The Clean DLA (CDLA) sample
Our sample is selected from the Baryon Oscillation Spectroscopic Survey (BOSS) (Dawson et al. 2013; Ahn et al. 2013). BOSS is a spectroscopic survey that will measure redshifts of 1.5 million luminous red galaxies and Ly α absorption towards 160,000 high redshift quasars (Eisenstein et al. 2009). 1,000 optical fibers with 2 arcsec diameters are plugged into the 1,000 holes in an aluminum plate to receive light from 160-200 quasars, 560630 galaxies, 100 ancillary science targets and standard stars (e.g., Dawson et al. 2013). Also, each plate contains at least 80 sky bers which placed at blank area on the sky to model the background for all the science fibers. The distribution of the sky fibers is constrained to cover the entire focal plane. This designs allow the sampling of the varying sky background over the focal plane the spectrograph (e.g., Dawson et al. 2013). The BOSS spectra have the moderate resolution of R ∼ 2000 from 3700 ˚ - 10,400 ˚ . With the typical exposure time A A of 1-hour for each plate, the BOSS spectra have a modest median S/N of ∼ 2 -3 per resolution element for quasars at rest-frame wavelength λ = 1041 -1185 ˚ (Lee et al. A 2012).
The Data Release 10 Quasar (DR10Q) catalog (Pˆris a et al. 2013) includes 166,798 quasars detected over 6,000 deg , of which 117,774 are at 2 z > 2 15. . A combination of target selection metods (Ross et al. 2012; Bovy et al. 2011) achieved this high surface density of high-redshift quasars. Using a fully automated procedure based on profile recognition using correlation anaysis, Noterdaeme et al. (2012) constructed a catalog of DLAs and subDLAs with N HI > 10 20 0 . cm -2 detected from an auto-
matic search along DR9Q lines-of-sight. Using a similar technique, Noterdaeme et al. (2014) generated the DLA catalog for Data Release 10 (DR10).
The DLA detection technique (Noterdaeme et al. 2012, 2009) pays special attention to the low-ionization metal lines in order to remove possible blends or misidentifications. The redshift measurement is mainly based on the profile fit the DLA. Some of the DLA redshift measurements have been refined based on the low-ionisation metal lines. The equivalent width of the metal lines is obtained automatically by locally normalising the continuum around each line of interest and then modeling the absorption lines with a Voigt profile. For non-detections, upper limits are obtained from the noise around the expected line positions.
The final BOSS DR10 DLA catalog includes 11,030 DLAs with N HI > 10 20 3 . cm -2 , continuum-to-noise ratio (CNR) > 2, and absorption redshift z abs > 2 15. . We avoided proximate DLAs at velocities less than 5,000 km s -1 from the quasar as these have a high probability of being physically associated with the local quasar environment. Strong proximate DLAs from BOSS are studied in Finley et al. (2013).
In our investigation, the DLA purity is crucial. Guided by this consideration, we include only DLAs satisfying the following stringent criteria: (a) spectra with median CNR > 4 0; . (b) > 3 σ low-ionization metal line detections; (c) column densities N HI > 10 20 6 . cm -2 ; which requires the width of the DLA trough to be greater than four times the spectral resolution; (d) Spearman's correlation coefficient of the Voigt profile fitting is greater than 0.6, which ensures an accurate estimate of the DLA column density (see details in Noterdaeme et al. 2009). Criterion (b) ensures that it is a DLA that is responsible for absorption rather than an unresolved blend of multiple narrow absorption features, while criteria (a), (c) and (d) enable an accurate determination of the column density and redshift. ∼ 2,300 out of 11,030 DLAs in the catalog satisfy these four criteria. Further, we visually inspect the 2,300 systems to confirm that they are DLAs. We eliminate ∼ 40 systems close to quasar OVI+Ly β region. Also, the visual inspection enables us to find a small number of DLAs with inaccurate redshift measurements, and we paid careful attention to refine the redshifts for some of the DLAs according to Voigt profile fitting and metal lines if necessary. We also drop a number of DLAs with blending of Lyman limit systems or strong Ly α forest absorption in both DLA wings, which makes the measurements of the DLA column density and redshift difficult. A sample of 2,138 DLAs pass the visual inspection. Further, we constrain the redshift of DLAs in the moderate redshift range of 2 2 . < z < 3 2. . A sample of 1,940 DLAs within the redshift range are finally selected and defined. In the remainder of the paper, we refer to these 1,940 DLAs as the clean DLA sample ( CDLA sample). In Fig. 1, the right panel shows the redshift distribution of these 1,940 DLAs with different quasar luminosity bins. The orange histogram represent the highest luminosity bins, while the red histogram represents the lowest luminosity bin. The left panel shows the redshift distribution for the corresponding background quasars. The redshift distribution of DLAs in different luminosity bins are similiar within the redshift range of 2 2 . < z < 3 2. .
Fig. 1.The right panel shows the redshift distribution of SDSS DR10 DLAs with different quasar luminosity bins. The orange histogram represents the highest quasar luminosity bin, while the red histogram represents the lowest quasar luminosity bin. The left panel presents the redshift distribution of the background quasars. The DLA redshift range we consider is z = 2 2 . -3 2. . The redshift distribution of the DLAs and corresponding background quasars in different quasar luminosity bins are similiar in the DLA redshift range of z = 2 2 . -3 2. .
DLAs with N HI > 10 20 6 . cm -2 have a dark trough with a rest-frame velocity width extending at least ± 340 km s -1 from the center, much larger than the DLA redshift uncertainties. In addition to refining the DLA redshifts from Voigt profile fitting on Ly α part, we use multiple low-ionization metal lines to fit the redshifts ( z ion ) of ∼ 500 DLAs. We find that the largest difference between the redshift fitted by low-ionization metal lines and redshift fitted by Ly α Voigt profile ( c × ∆ z/ (1+ z )) is ∼ 150 km s -1 . We also checked the DLA trough of the composite DLA spectra (see § 3.1 and § 3.3, and also the inset in Fig. 3), and the central ± 250 km s -1 is consistent with a flat absorption trough within 1σ . All of this evidences support that the DLA redshift uncertainties are much smaller than the width of DLA dark trough; and we conservatively define the dark trough in the composite spectra as the region within ± 150 km s -1 about the DLA center.
## 3. RESIDUAL FLUX IN THE DARK TROUGH
## 3.1. Initial stacking
The SDSS spectra are shifted to the rest-frame of the DLA while conserving the flux, i.e., f i ( λ res ) = (1 + z DLA ) × f i , sdss ( λ obs ), where λ res is the DLA rest-frame wavelength; λ obs is the wavelength in the SDSS spectra ( λ obs = λ res × (1 + z DLA )); f i ( λ ) is the flux density of an individual spectrum redshifted to the rest frame; and f i , sdss ( λ ) is the flux density of an individual SDSS spectrum in the observed frame. In the dark trough, the flux density in each pixel (wavelength) of the composite spectra is calculated by two methods: (1) the 3 σ -clipped mean of all the spectra at the same wavelength; (2) median of all the spectra at the same wavelength. Outside the dark trough with λ < 1213 ˚ A or λ > 1218 ˚ , we take the median value to stack the specA tra. We calculate the error at each pixel (wavelength) in the stacked spectrum by propagating the errors of the pixels at the same position in every individual spec-
$$\underset { \text{$m$ the} } { \text{$m$ the} } \quad \text{$m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$, the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$. $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ and $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ that $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ to the the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $m$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $m$ the $m$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $m$ the $p$ the $m$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $p$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $p$. $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$. the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m' the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m. $ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m` the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m\$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$ the $m$$$
mainly removes the large outliers due to noise, but not clipping away the Ly α emission from DLA galaxy. On the one hand, even assuming most of DLA host galaxies are as bright as L ∗ galaxy at z = 2 -3 (e.g., Shapley et
al. 2003), the Ly α emission from DLA host is still submerged to the noise of the individual SDSS-III spectrum. The detailed discussion is included in § 4.1. On the other hand, from visual inspection of the entire CDLA sample, we do not find DLAs with strong Ly α emission in the dark trough.
In the stacking process, we do not introduce any additional scaling. It is true that different quasars are different in luminosity, quasar host flux, and DLA host emission. However, in the DLA dark trough, quasar continuum is completely blocked in the DLA dark trough. Also, for SDSS-III spectra, the average flux densities of the DLA hosts and quasar hosts are generally about or more than one order of magnitude lower than the noise level (see more details in § 4.1 and § 4.2). For SDSS-III spectrum, the noise is mainly due to the sky background and the CCD read noise (e.g., Dawson et al. 2013), and in the dark trough of DLAs at z = 2 6 . ± 0 3, . different systems generally have similar noise level. Therefore, following previous similar studies, we just simply stacking the spectra without any additional scaling (e.g., Rahmani et al. 2010; Noterdaeme et al. 2014).
After the initial stacking, we find that the dark trough of the stacked DLA shows a positive offset. We calculate the mean flux density by averaging the flux density in the dark trough region. Following the discussions in § 2.2, we conservatively define the stacked dark trough region within ± 150 km s -1 about the center. For the stacking of 3 σ -clipped mean, the average dark trough flux density is ¯ F dark = 6 5 . ± 0 6 . × 10 -19 ergs cm -2 A -1 s -1 ; and for the median stacking, the average dark trough flux density is ¯ F dark = 7 0 . ± 0 6 . × 10 -19 ergs cm -2 A -1 s -1 . A general flux residual in the stacked DLA dark trough has been documented in previous work with much smaller sample size (Rahmani et al. 2010; Pˆris et al. 2012). a The large DLA database of SDSS-III/BOSS DR10 enables us, for the first time, to carefully study and test the origin of the positive residual flux in the dark core. This residual flux can arise from three sources: (a) sky subtraction residual (systematic sky subtraction errors); (b) Ly α emission around λ ∼ 1216 ˚ from the DLA galaxies; (c) FUV conA tinua emission at (1 + z DLA ) / (1 + z QSO ) × 1216 ˚ A ∼ 1100 ˚ from the quasar host galaxies which is not blocked by A the foreground DLAs.
## 3.2. The determination of the sky subtraction residual
In order to calculate the sky subtraction residual, we select a group of 150 DLAs at z > 3 8 . with a median redshift < z > = 4 0, and examine the flux in the DLA . Lyman limit region ranging from rest-frame λ = 800 ˚ to A 900 ˚ , where the optical depth A τ ( λ ) is expected to be over 1000. For the sample of z > 3 8 DLAs, the Lyman limit . region at rest-frame λ = 820 ˚ to 900 ˚ corresponds to an A A observed wavelength from 4100 ˚ to 4500 ˚ , which covA A ers the average wavelength of DLA trough in our CDLA sample. The expected flux of quasar continua should be negligible small in the DLA Lyman limit region because of the large optical depth. The flux in this region can only be contributed by: (a) sky-subtraction residuals; (b) unobscured escaping ionizing photons with λ glyph[lessorsimilar] 900 ˚ from A DLA galaxies and (c) photons from quasar hosts at even shorter wavelength. Therefore, the average flux in DLA Lyman limit regions can be regarded as an upper limit
Fig. 2.Composite spectra of the Lyman limit region (restframe 820 A- 900 A) observed by stacking a group of 150 ˚ ˚ z ≥ 3 8 . DLAs with a median redshift of z ∼ 4 0. . The inset figure shows a zoom-in of the DLA absorption for the composite spectrum of the group of z ≥ 3 8 DLAs. . The stacked Lyman limit region corresponds to observed frame λ obs ∼ 4100- 4500 ˚ . The upper A limit of the sky-subtraction can be determined by averaging the flux in the composite Lyman-limit region (see Section 3.2). The average flux of the Lyman limit region (red dot) is 3 5 . ± 0 008 . × 10 -19 ergs cm -2 ˚ A -1 s -1 , and the median flux at the composite DLA dark trough (blue dot with errorbar) of the CDLA sample is 7 6 . ± 0 6 . × 10 -19 ergs cm -2 ˚ A -1 s -1 , which is significantly larger than the SDSS sky-subtraction residual. The observed wavelength of blue point is determined by DLA center at the median redshift of CDLA sample.
for the sky-subtraction residual. A number of studies have found that the escape fraction is small, glyph[lessorsimilar] 5%- 10% relative to photons escaping at 1500 A (Malkan et al. ˚ 2003; Siana et al. 2007; Bridge et al. 2010). This small escape fraction suggests that the ionizing photons could only have a minor or even negligible contribution to the average flux in the DLA Lyman limit region.
The 150 DLAs at z > 3 8 we have selected from the . DLA catalog all have CII and/or SiII detections and passed the visual inspection to ensure the DLA nature. We use the median to stack the spectra at rest-frame λ = 820 - 900 ˚ in the stacked spectra. A Then, we average the flux from λ = 820 ˚ - 900 ˚ . The average flux density A A is equal to 3 502 . ± 0 008 . × 10 -19 ergs cm -2 A -1 s -1 , and this value can be regarded as an upper limit of the skysubtraction (see red dot in Fig. 2). We double checked our results using the average sky subtraction residual as a function of wavelength which is determined by the SDSS pipeline group (Bolton et al. 2012 (Schlegel et al. 2014 in prep., private communication). The sky-subtraction residual is ∼ 3 × 10 -19 ergs cm -2 A -1 s -1 at ∼ 4300 ˚ , A consistent with the upper-limit on the sky-subtraction residual we have derived. In Fig. 2, the inset shows a zoom-in of the DLA absorption for the composite spectrum of the group of z ≥ 3 8 . DLAs. Also, in Fig. 2, we present the median flux in the composite DLA dark trough (blue dot with errorbar) of the CDLA sample is 6 5 . ± 0 6 . × 10 -19 ergs cm -2 ˚ A -1 s -1 , which is significantly larger than the SDSS sky-subtraction residual (red dot in Fig. 2). The observed wavelength of blue dot in Fig. 2 is determined by the center of DLA trough at the median redshift of CDLA sample.
In § 3.1, we initially stacked all the spectra of the CDLA sample, and find the average dark trough flux density is ¯ F dark = 6 5 . ± 0 6 . × 10 -19 ergs cm -2 A -1 s -1 , significantly greater than the upper limit of the sky subtraction residual determined in § 3.2 (Fig. 2). This dark trough flux density indicates that in addition to the sky subtraction residual, flux must be contributed by other sources. In this section, we will examine and test the possible origin of the flux residual in the DLA dark trough.
In order to examine the origin of the flux residual in the DLA dark core, we will group the quasars according to their luminosities, and check if the residual intensities in the dark trough correlate with the quasar luminosities. The motivation for this experiment is that if the intensity in the stacked dark trough is mainly contributed by the Lyman continua of the foreground DLA, then this intensity should be uncorrelated with the background quasar luminosity, and we expect the three luminosity bins to yield similar intensities in the dark trough. We divide the quasar into three bins according to their luminosity bins: (1) the highest luminosity bin contains quasars with the luminosity density at 1450 ˚ ( A L 1450 ) > 9 × 10 42 erg s -1 ˚ A -1 ; (2) middle luminosity bin with 5 × 10 42 erg s -1 ˚ A -1 < L 1450 < 9 × 10 42 erg s -1 ˚ A -1 ; (3) low luminosity bin with L 1450 < 5 0 . × 10 42 erg s -1 ˚ A -1 . Each luminosity bin contains similar number of quasars.
The detailed stacking procedure is as follows: we first calculate the quasar intensity for each quasar using the following expression:
$$I _ { i, \text{dark} } ( \lambda _ { \text{res} } ) = & \, ( f _ { i, \text{sdss} } ( \lambda _ { \text{obs} } ) - f _ { \text{sky} } ) \times 4 \pi D _ { L } ^ { 2 } ( z _ { \text{QSO} } ) \times ( 1 + z _ { \text{DLA} } ), \text{compare t} \\ \,, \, \quad, \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots \quad \, \cdots$$
where λ obs = λ res × (1 + z QSO ), f i , sdss is the flux density for an individual spectrum, f sky = 3 5 . × 10 -19 erg s -1 cm -2, which is determined in § 3.2 and further supported by SDSS calibration group (Schlegel et al. 2014, in prep.). Following the stacking method described in § 3.1, we stack the quasar intensity I i, dark ( λ res ) for each luminosity bin by taking the 3 σ -clipped mean and median value in the dark trough. Outside the dark trough, we just take the median value to stack the spectra.
Note that the 3 σ -clipped mean is only a legitimate method when we stacking the spectra within the dark trough. In the dark trough, the residual flux, if exists, is only in a very low level. It is necessary to first rule out the large outliers due to the noise before taking the average. However, outside the dark trough, the different quasar continua have large scatter in luminosity. The 3 σ -clipping will reject large number of bins in the continua simply because the quasars differ in luminosity. Therefore, the 3 σ -clipped mean is no longer a legitimate way to stack the quasar continua outside the dark trough. Fig. 3 further demonstrates this point. In Fig. 3, one can see that in the DLA dark trough, the number of rejected pixels is small (only ∼ 0.5% of the sample size). However, for the quasar continua at rest-frame wavelength λ < 1214 ˚ or A λ > 1218 ˚ , A the number of the rejected pixels for each wavelength significantly become larger than 1% of the sample size. This is because outside the dark trough, the 3 σ -clipping meanly rejects bins in the continua becasue quasars differ in luminosity. We have checked that, for the stacked quasar bolometric luminosity, the median value is 10% smaller than the 3 σ -clipped mean, 25%
,
Fig. 3.We present the number of 3 σ rejected pixels as a function of rest-frame wavelength for the highest quasar luminosity bin. From this figure, one can see that in the dark trough region from λ ∼ 1214 -1218 ˚, the number of 3A σ rejected bins is small. On average, 4 pixels at specific wavelength are rejected in the dark trough region, about 0.5% of the sample size (this sumple contains ∼ 750 DLAs). This is because in the dark trough, the 3 σ -clipping only rejects large outliers due to the noise. However, for the quasar continua at λ < 1214 ˚ or A λ > 1218 ˚, the number of the rejected A pixels is significantly larger than 1% of the sample size. Outside the dark trough, the 3 σ -clipping rejects large number of bins in the continua simply because the quasars differ in luminosity. Therefore, outside the DLA dark trough, the 3 σ -clipped mean is no longer a legitimate way to stack the quasar continua. Outside the DLA dark trough at λ < 1214 ˚ and A λ > 1218 ˚, we just take the A median value to stack the quasar continua (see Section 3.3).
lower than the average without sigma-clipping. Outside the DLA dark trough, we take the median to stack the quasar continua. In the next Section ( 4.2), we will be § consistent to just use the median composite spectrum to compare the quasar bolometric luminosity with the dark trough intensities (see details in § 4.2).
We summarize our results in Table 1 and Fig. 4. The upper panel in Fig. 4 presents the composite DLA spectra in three quasar luminosity bins using the CDLA sample. The middle panel presents the 3 σ -clipped mean of the DLA dark trough after removing the sky-subtraction residual (Eq. (1)). The stacked dark trough, which lies between the two vertical dashed lines, is defined within ± 150 km s -1 from the center (see § 2.2). The lowest panel presents the median value of the flux residual in the DLA dark trough. The zoom-in insets in the middle and lowest panel presents the 3 σ -clipped mean and median value of the overall CDLA sample, centered on the DLA absorption to show the non-detection of Ly α emission from the DLA hosts. The middle and lower panels demonstrate that the dark trough flux density is non-zero for the two higher luminosity bins, and moreover, the dark core intensity increases for the higher quasar luminosity bin. We have compared the results derived from 3 σ -clipped mean and median value. We found the residual intensities derived using 3 σ -clipped mean are consistent with the values obtained from the median stacking (Table 1).
Fig. 5 shows that the correlation between the dark trough intensity and quasar luminosity still holds for DLAs with the largest dark troughs in CDLA sample. DLAs with N HI > 10 20 9 . cm -2 have an expected dark trough width within ± 580 km s -1 about the center. Again, we conservatively define the dark trough region to be ± 250 km s -1 about the center. The upper panel of Fig. 5 shows the composite DLA ( N HI > 10 20 9 . cm -2 ) spectra for the same three luminosity bins with that in Fig. 4. The middle panel shows the 3 σ -clipped mean
Fig. 4.The upper panel shows the composite DLA ( N > 10 20 6 . cm -2 ) spectra for three different luminosity bins of our CDLA sample. The middle panel shows the 3 σ -clipped mean of the expanded DLA dark core region after removing the sky-subtraction residual. The middle and the lower panels demonstrate the statistically significant results that the higher dark core intensity corresponds to the higher quasar luminosity. As stated in ( § 2.2), we conservatively define the dark trough within ± 150 km s -1 from the center, which region is between the two vertical dashed lines in the middle and lower panel. The insets in the middle and lower panels present the composite spectrum of the full CDLA sample, zoom-in the region around wavelength λ ∼ 1216 ˚ to show the non-detection of the Ly A α emission from the DLA host galaxies. The lowest panel shows the median value of the expanded DLA dark core region after subtracting the sky-subtraction residual.
of the expanded DLA dark core region after subtracting the sky-calibration residual, and the lowest panel shows the median value for the same dark trough region. The insets in the middle and lower panel present the composite spectra of the N HI > 10 20 9 . cm -2 CDLA sample, centered on the composite DLA absorption to show the non-detection of the Ly α emission from the DLA host galaxies. The sample size of N HI > 10 20 9 . cm -2 CDLA sub-sample is only half of the overall CDLA sample, yet it is just evident that the higher quasar luminosity bin still corresponds to a higher dark trough intensity, i.e., the dark trough intensity with highest quasar luminosity (yellow in Fig. 5) is 3σ higher than that with lowest quasar luminosity (red in Fig. 5).
The relation between the quasar luminosities and the intensities in the stacked dark trough is summarized in
Fig. 6. The CDLA sample with N HI > 10 20 6 . cm -2 is shown in red, and the luminosity distribution is shown in the lower panel in red. A sub-group of DLAs in the CDLA sample with multiple metal lines are shown in black, where the redshifts are obtained from the low ionization lines. The blue points represent DLAs in the CDLA sample with N HI > 10 20 9 . cm -2 . The comparison between N HI > 10 20 6 . cm -2 and N HI > 10 20 9 . cm -2 DLAs demonstrates that the dark core intensities are not correlated with the DLA column densities. We also compare DLAs in two bins of equivalent width, which are denoted by triangles in red and yellow color in Fig. 6. We find that the DLA dark core intensities do not correlate with the DLA metal line equivalent width. Fig. 6 clearly suggests a > σ 3 detection of the correlation between the mean DLA dark core intensity and quasar luminosity.
Fig. 5.The upper panel shows the composite DLA ( N > 10 20 9 . cm -2 ) spectra for three different luminosity bins of the CDLA sample for stronger DLA systems. The middle panel presents the 3 σ -clipped mean of the expanded DLA dark core region after subtracting the sky-calibration residual, and the lowest panel shows the median value of the same dark trough region. In the middle and lower panels, the dark trough is defined between the two vertical dashed lines ( ± 250 km s -1 from the center, see § 2.2). The insets in the middle and lower panel present the composite spectrum of N col > 10 20 9 . cm -2 CDLA sample, centered on the composite DLA absorption to show the non-detection of the Ly α emission from the DLA host galaxies.
The histogram in lower panel shows the distribution of the quasar luminosities for CDLA samples.
## 4. DISCUSSION
Using the CDLA sample of 1,940 DLAs, we demonstate that the dark trough intensity is significantly greater than the sky-subtraction residual, and the intensity in the composite dark trough correlates with quasar luminosity. Further, the intensities of the dark trough do not correlate with DLA metal equivalent width or DLA column density. In this section, we will discuss two possible scenarios of the physical origin for the dark-trough intensity and its relation to the quasar luminosity.
## 4.1. DLA galaxy emission
More than hundreds of DLAs at z > 2 have been identified over the last twenty years. However, only a hand- ful of DLAs have been confirmed to have Ly α emission originated from the DLA host galaxies. Here, we first give a brief summary of such studies. Moller & Warren (1993) confirmed a Ly α emission line associated with a DLA with column density N col =10 21 cm -2 at z = 2 8. . The luminosity of Ly α emission is about 2 × 10 42 erg s -1 ( ∼ 0 4 . × L ∗ at z = 3) (Ciardullo et al. 2012). The velocity offset of the Ly α emission is about +50 ± 100 km s -1 . Djorgovski et al. (1996) confirmed a Ly α emission associated with a sub-DLA of column density N col = 10 20 0 . cm -2 at z = 3 1. . The luminosity of the Ly α emission is 5 × 10 42 erg s -1 ( ∼ L ∗ ), and the velocity offset is about +200 km s -1 . Leibundgut & Robertson (1999) confirmed a Ly α emission from galaxy associated with a DLA with N col = 10 20 85 . cm -2 at z = 3 1. . The Ly α luminosity is about L ∗ , and the velocity offset is about 490 km s -1 .
Fig. 6.A summary of the relation between the quasar luminosities and the intensities in the stacked DLA trough of the CDLA sample. The comparison between DLAs with N HI > 10 20 6 . (red dots) and N HI > 10 20 9 . (blue dots) demonstrate that the dark core intensities are not correlated with the DLA column densities. The red and yellow triangles show that the DLA dark core intensities do not correlate with the DLA metal line equivalent width. There is a glyph[greaterorsimilar] 3 σ detection of the correlation between the mean DLA dark core intensities and the quasar luminosities. The histogram in the lower panel shows the distribution of the quasar luminosities for CDLA samples in each of the three luminosity bins.
Moller et al.(2002) presented a GRB DLA with a ∼ 0.5 L ∗ Ly α emission of DLA galaxy at z = 2 3, . and the Ly α emission is redshifted by 530 km s -1 . Moller et al. (2004) detect a ∼ 0.5 L ∗ Ly α emission in the dark trough of a DLA with large column density N col ∼ 10 21 8 . cm -2 at z = 2 0. . The velocity offset of this Ly α emission is about +10 ± 150 km s -1 . Kulkarni et al. (2012) presents a super-DLA with column density of 10 22 0 . cm -2 associated with a L ∗ Ly α emission at z=2.2. In the high S/N spectrum, the Ly α emission is consistent with a velocity offset of +300 km s -1 . By stacking a sample of 99 DLAs with N HI > 10 21 7 . cm -2 , Noterdaeme et al. (2014) present the detection of a Ly α emission line with a luminosity of 0 65 . × 10 42 erg s -1 in the composite DLA trough, corresponding to 0.1 × L ∗ galaxies at z = 2 -3, and this composite Ly α emission has a velocity offset of 130 km s -1 .
After stacking the spectra using the DLAs in the CDLA sample, we did not detect Ly α emission in the dark trough of the composite spectra (e.g., see the insets in the middle and lower panels of Fig. 4 and Fig. 5).
In Fig. 7, we present the composite spectra of 2,000 simulated DLAs with Ly α emission being added, and the stacking method is 3 σ -clipped mean. The simulated DLAs have same column density distribution with that of our CDLA sample. The noise has been included according to the realistic BOSS spectra. The Ly α emission is assumed to have the same luminosity, FWHM (750 km s -1 ) and velocity offset (445 km s -1 ) with the composite spectrum of L ∗ LBGs at z = 2 -3 (Shapley et al. 2003). Brown line shows the 3 σ -clipped mean of simulated DLAs by assuming that 100% DLA host galaxies have Ly α emission similar to that of L ∗ LBGs, and all the Ly α emission entered the SDSS fibers. Blue line
Fig. 7.The composite spectra of the 2,000 simulated DLAs with realistic galaxy Ly α emission being added. The simulated DLAs are stacked using the 3 σ -clipped mean. Simulated DLAs have same column density distribution with that of our CDLA sample. The noise has been included according to the realistic observed spectra. The Ly α emission of DLA host is assumed to have the similar luminosity, FWHM and velocity offset with that of L ∗ LBGs at z = 2 -3 (Shapley et al. 2003). Brown line shows the stacking of the simulated DLAs by assuming that 100% of DLA hosts have such strong Ly α emission entering the SDSS fibers. Blue (red) line assumes that 50% (10%) of the L ∗ Ly α emission entering the SDSS fibers for DLA host galaxies. Yellow line indicates none of the emission from the DLA hosts contribute to the residual flux in the DLA dark trough, either because the emission from DLA host galaxies are below the detection limit, or because the impact parameter between the DLA cloud and Ly α emitting region is bigger than 10 kpc (see § 4.1). The region between two vertical dashed lines indicate the conservative dark trough we defined (same with Fig. 4). From the region between 1214 ˚ - 1218 ˚ , one can see A A that the strong Ly α emission will significantly affect the trough profile of the stacked DLAs. With strong Ly α emission, the flux will increase with the wavelength, yet our observed flux (black line) does not. The observed results are generally consistent with the non-detection of the strong Ly α emission from DLA galaxies. Also, this figure demonstrates that the 3 σ -clipped stacking does not clip away the Ly α emission from the DLA host galaxies.
assumes 50% of the simulated DLA host galaxies have L ∗ Ly α emission. Red line assumes 10% of the simulated DLA hosts have such strong Ly α emission. Yellow line indicates that the Ly α emission from the DLA hosts do not contribute to the residual flux in the DLA dark trough, either because the emission from DLA host galaxies are below the detection limit (see first probility in § 4.1), or because the impact parameter between the DLA clouds and Ly α emitting region are bigger than 10 kpc (see third probability in § 4.1). From the region between two vertical dashed lines, one can see that the strong Ly α emission will significantly affect the profile of the stacked DLA trough. In the region of the stacked dark trough, with the present of strong Ly α emission, the stacked flux will increases with the wavelength, yet our observed flux (black curve) does not. The observed results are generally consistent with the non-detection of the strong Ly α emission from DLA galaxies. Also, this figure demonstrates that the 3 σ -clipped stacking does not clip away the Ly α emission from the DLA host galaxies.
Fig. 8 presents a similar plot with Fig. 7. In this figure, we assume that the Ly α emission from DLA host galaxies have the same luminosity with 0.1 × L ∗ starforming galaxies at z = 2 -3 (Noterdaeme et al. 2014). The luminosity, FWHM (300 km s -1 ) and velocity offset (130 km s -1 ) of the Ly α emission are followed by that described in Noterdaeme et al. (2014). Brown color presents the 3 σ -clipped mean of simulated DLAs by assuming that 100% simulated DLA host galaxies have Ly α emission with the luminosity of 0.1 L ∗ , and all the Ly α
Fig. 8.Similar plot with Fig. 7. In this figure, the fainter Ly α emission of DLA host galaxies is assumed. We assume the Ly α emission have the similar luminosity, FWHM and velocity offset with that of 0.1 × L ∗ star-forming galaxies at z = 2 -3 (Noterdaeme et al. 2014). Brown color presents the 3 σ -clipped mean of simulated DLAs by assuming that 100% simulated DLA host galaxies have Ly α emission with the luminosity of 0.1 × L ∗ . Blue (red) line assumes 50% (10%) of DLAs have 0.1 × L ∗ Ly α emission. Yellow line indicates that the flux from the DLA hosts did not contribute to the residual flux in the DLA dark trough (see § 4.1). From the observed spectra (black line), we did not detect the Ly α emission from DLA host galaxies. Our results are consistent with the following conclusions ( § 4.1): (1) DLA host galaxies are generally hosted by faint galaxies in relatively low-mass halos; (2) for DLAs with column density of N col > 10 20 6 . cm -2 , the impact parameter between the DLA clouds and Ly α emitting regions may be generally bigger than 10 kpc (see § 4.1).
emission entered the SDSS fibers. Blue (red) line assumes 50% (10%) of the simulated DLA host galaxies have 0.1 × L ∗ Ly α emission entering the fibers. Yellow line indicates that the flux from the DLA hosts did not contribute to the residual flux in the DLA dark trough (see § 4.1). From the observed spectra (black line), we did not detect the Ly α emission from DLA host galaxies. Our results are consistent with the above discussions: (1) DLAs are generally hosted by faint galaxies in relatively low-mass halos; (2) for column density of N col > 10 20 6 . cm -2 , most of the DLAs have their impact parameters between the DLA clouds and Ly α emitting regions bigger than 10 kpc.
Fig. 9 shows a similar plot with Fig. 8. In this figure, we assume that the Ly α emission from DLA host galaxies have the typical stellar mass of 10 8 5 . M glyph[circledot] (Møller et al. 2013). Assuming the DLA hosts follow the galaxy main sequence, the typical luminosity of DLA host galaxies should be about 0.03 × L ∗ galaxies at z = 2 -3. In this figure, we further assume that the DLA host galaxies have luminosity of 0.03 × L ∗ . Brown color presents the 3 σ -clipped mean of simulated DLAs by assuming that 100% simulated DLA host galaxies have Ly α emission with the luminosity of 0.03 L ∗ . Blue (red) line assumes 50% (10%) of the simulated DLA host galaxies have 0.03 × L ∗ Ly α emission. Yellow line indicates that the flux from the DLA hosts did not contribute to the residual flux in the DLA dark trough. From the figure, if only 10% of the Ly α emission similar to 0.03 × L ∗ , the stacked Ly α emission will below our current detection limit. Black line shows that we did not detect the Ly α emission from DLA host galaxies, which is consistent with the red and yellow lines.
We estimate the 3σ upper limit ( F 3 σ (Ly α )) by using the gaussian fitting errors, assuming the rest-frame FWHM of the expected Ly α emission is 400 km s -1 . The 3σ upper limit of the Ly α emission line flux is
Fig. 9.In this figure, we assume that the Ly α emission from DLA host galaxies have smaller stellar mass of 10 8 5 . M glyph[circledot] (Møller et al. 2013). Further, if we assume the DLA hosts follow the galaxy main sequence, the typical luminosity of DLA host galaxies should be about 0.03 × L ∗ galaxies at z = 2 -3. In this figure, we assume that the DLA host galaxies have luminosity of 0.03 × L ∗ . Again, brown color presents the 3 σ -clipped mean of simulated DLAs by assuming that 100% simulated DLA host galaxies have 0.03 L ∗ Ly α emission entering the SDSS fibers. Blue (red) line assumes that 50% (10%) of the simulated DLA host galaxies have Ly α emission with the luminosity of 0.03 L ∗ . Yellow line indicates that the flux from the DLA hosts did not contribute to the residual flux in the DLA dark trough (see § 4.1). From the figure, if only 10% of the DLA hosts have Ly α emission with the luminosity of 0.03 L ∗ , the stacked Ly α emission is below our current detection limit. From the observed spectra (black), we did not detect the Ly α emission from DLA host galaxies.
F 3 σ (Ly α ) = 7 5 . × 10 -19 erg s -1 cm -2 . Considering the 2 ′′ SDSS-III fiber diameter, this limit translates to a limiting 1σ surface brightness limit of 0.8 × 10 -19 ergs s -1 cm -2 arcsec -2 . This depth is about a factor of 2 deeper than Rahmani et al. (2010) and comparable to the depth of the long-slit spectroscopic observations of Rauch et al. (2008). At the mean redshift ( 〈 z 〉 = 2 6), the 3-. σ limit on the Ly α emission corresponds to a Ly α luminosity of 4.0 × 10 40 ergs s -1 . This upper limit on the Ly α luminosity reaches a depth of ∼ 0.01 L ∗ , where L ∗ ∼ 4 × 10 42 erg s -1 at z = 3 (Cassata et al. 2011; Ciardullo et al. 2012). This corresponds to a 3σ upper limit on the Ly α -based SFR (SFR Ly α ) of 0.04 M glyph[circledot] yr -1 (Kennicutt 1998; Dijkstra & Westra 2010).
The non-detection of the Ly α emission from the DLA galaxies in our CDLA sample can be explained by the following possibilities: (1) DLAs are mostly harboured by faint (subL ∗ ) galaxies in relatively low-mass dark matter halos; (2) A large fraction of DLA clouds have impact parameters much larger than the stellar half light radius of the DLA host galaxy and exceed the 1 ′′ fiber radius, corresponding to ∼ 6 kpc at 〈 z 〉 = 2 65, such that a large . number of the fibers are not exposed to the DLA galaxy stellar light; (3) systemic offsets and redshift uncertainties of the Ly α emission lines in the DLA galaxies.
The first possibility is supported by an analysis of the mass-metallicity relation in a sample of 110 DLAs from z = 0 1 . -5 1 . conducted by Møller et al. (2013). They suggest that the typical DLA has a stellar mass of log( M /M ∗ glyph[circledot] ) ∼ 8.5, in agreement with earlier results that most massive DLA galaxies only correspond to the least massive LBGs (Fynbo et al. 1999). Also, the low-mass DLA halos have been suggested by the simulations (e.g. Mo et al. 1999; Haehnelt et al. 2000; Nagamine et al. 2004). However, a few works favor more massive DLA dark matter halos on the order of 10 12 h -1 M glyph[circledot] (e.g. Font-Ribera et al. 2012).
The second possibility is supported by several previous studies. The fibers may not be exposed to the DLA galaxy light due to the large impact parameter of the DLA clouds with respect to the center of the galaxy. Zwaan et al. (2005) present a comprehensive HI 21cm absorption survey of nearby galaxies, which contains 40 HI absorption features with column densities N HI > 10 20 0 . cm -2 . Half of the DLAs were found to have an impact parameter b > 8 kpc from the center of the galaxy, indicating that the stellar emission from a large fraction of DLA-hosting galaixes would not enter the 2 ′′ BOSS fiber, if the DLAs at z = 2 -3 have the same impact parameter as that of the nearby galaxies. Krogager et al. (2012) studied all of the 10 known DLAs with the identified galaxy counterparts at z = 2 -3, and found that DLAs with impact parameter ranges from 0.1 ′′ -3.0 ′′ . Pontzen et al. (2008) are able to reproduce several properties of DLAs in their simulations, and the impact parameters at z = 3 are expected to be b glyph[lessorsimilar] 30 kpc. Further, Fynbo et al. (2008) supports that higher metallicity DLAs have larger disks. In our CDLA sample, we only selected DLAs with metal line detections which biases our DLAs to higher metallicity, which may have larger HI disks based on the simulations. It is true that diffuse halo Ly α emission (Steidel et al. 2011) may enter the fiber. However, the surface brightness of the diffuse Ly α halo emission significantly drops by a factor of > 3 when the impact parameter is greater than 15 kpc.
The third possibility is also supported by numerous observations. Using a sample of 89 Lyman Break Galaxies (LBGs) with 〈 z 〉 = 2 3 . ± 0 3, . Steidel et al. (2010) find that the Ly α lines have systemic velocity offsets of ∆ v Ly α = 445 ± 27 km s -1 . Assuming the Ly α emission of DLA galaxy hosts has the same systemic velocity offsets as LBGs, the Ly α emission would be located outside of the dark trough of DLAs in our CDLA sample. The three possibilities we have discussed likely work together to cause the non-detection of DLA Ly α emission.
By stacking a sample of 99 DLAs with N HI > 10 21 7 . cm -2 , Noterdaeme et al. (2014) present the detection of a Ly α emission line with a luminosity of 0 65 . × 10 42 erg s -1 in the composite DLA trough, corresponding to 0.1 × L ∗ galaxies at z = 2 -3. The idea is that DLAs with large column densities are expected to have smaller galaxy impact parameters. This detection of Ly α emission from DLA galaxies does not contradict our result, because the very large DLAs with N HI > 10 21 7 . cm -2 constitute only ∼ 2% of our CDLA sample. On the contrary, this further supports the conclusion that most DLAs with N HI < 10 21 7 . cm -2 are generally not associated with close galaxy counterparts. Our non-detection of Ly α emission supports the hypothesis that the lower column density DLAs could have larger impact parameters (up to a few tens of kpc); thus a large number of damped Ly α absorbers in our sample are located > 10 kpc from the central galaxies.
Besides the contribution of Ly α emission from DLA galaxies, the dark core flux may also be contributed by the DLA Lyman continua at λ ∼ 1216 ˚ . However, if the A dark core flux comes from the DLA Lyman continua, it is difficult to explain why the dark core flux correlates with quasar luminosity. Therefore, our results favor the interpretation that the flux residual in the dark trough is highly unlikely to be contributed by Lyman continua or Ly α emission from the DLAs.
## 4.2. The FUV light from quasar host galaxies
Theoretically, that the luminous quasar phase naturally coincides with intense star formation was been proposed early on, and has the natural interpretation that both processes rely on reservoirs of gas brought to the center by gas-rich mergers and disk instabilities (e.g. Sanders et al. 1988; Hopkins et al. 2006; Lutz et al. 2008). Several observational studies in the mid-infrared and sub-mm regime (e.g. Lutz et al. 2008; Serjeant & Hatziminaoglou 2009) confirmed that there exists a link between the quasar host SFR and the black hole accretion rate. Angl´ es-Alc´zar a et al. (2013) further propose that galaxy-scale torque-limited accretion (Hopkins & Quataert 2011) naturally and robustly yields black holes and galaxies evolving along the observed scaling relations, and that AGN feedback does not need to couple with galaxy-scale gas and regulate black hole growth. Therefore, FUV stellar light from the quasar hosts is the most plausible interpretation for the flux residual in the composite DLA dark trough.
It is possible that the DLA absorption, while blocking the quasar continuum region, only partially blocks the host galaxies. The SDSS fiber is exposed to the regions of the quasar host galaxy which are not obscured by the intervening DLA and may contribute residual flux. This is consistent with models of clumpy HI distributions in DLA galaxies as suggested by recent observations (e.g. Kashikawa et al. 2013). Kanekar et al. (2009) used the Very Long Baseline Array (VLBA) to image 18 DLAs in redshifted HI 21cm, and suggest that for quasar radio emission with source sizes > 100pc, the median DLA HI covering factor for DLAs at z = 1 5 . -3 5 is . f med ∼ 0 6. . If the 21-cm and Ly α (UV) absorption do arise in the same cloud complexes, this result indicates that the emission from quasar hosts is not entirely covered by the foreground DLAs. Recent simulations and observations also support that in the galactic and circumgalactic environment, the covering factor of optically thick systems is glyph[lessorsimilar] 40 -50% for ∼ 0 1 . L ∗ galaxies at z = 2 -3 (Kanekar et al. 2009). The half-light radius of quasar host galaxies at z = 2 -4, typically 3-5 kpc (Peng et al. 2006), is fully covered by the SDSS fiber ( ∼ 12 kpc). The relation between the dark core residual flux and quasar luminosity supports a correlation between the SFR in the quasar hosts and the quasar luminosity.
After correcting for the sky-subtraction residual in the highest luminosity bins in the CDLA sample, the median intensity in the dark core is ∼ 16 4 . ± 2 1 . × 10 40 erg s -1 ˚ A -1 . We take this value as the average intensity of a quasar host galaxy at a rest-frame wavelength λ ∼ 1100 ˚ . This value corresponds to an intensity of A ∼ 7 5 . ± 1 0 . × 10 40 erg s -1 ˚ A -1 at λ ∼ 1700 ˚ for a A galaxy with a constant star formation history and an age of 0.5 Gyr, after correcting for intergalactic medium absorption (Bruzual & Charlot 2003; Schaye et al. 2003). At the mean redshift 〈 z 〉 = 3 1, this corresponds to 0.3 . L ∗ , where L ∗ is determined from > 2,000 spectroscopic LBGs from z = 1 9 . -3 4 (Reddy et al. 2008). . The inferred SFR based on the UV continua (SFR UV ) is ∼ 9 M glyph[circledot] yr -1 (Madau et al. 1998). Following the same proce-
dure, the middle-luminosity bin corresponds to SFR UV = 5 M glyph[circledot] yr -1 , and the non-detection in lowest luminosity bin translates to a 3 -σ upper limit of SFR UV =7.5 M glyph[circledot] yr -1 . If we assume DLAs have HI covering fraction of ∼ 0 5 as discussed in the last paragraph, the SFRs . UV for each luminosity bin will be a factor of 2 × higher than our measured value. We further derive the bolometric luminosities ( L bol ) for the three luminosity bins from the absolute magnitudes at 1450 ˚ , assuming a bolometA ric conversion factor ζ 1450A = L bol νL ν, 1450A = 4 4 (Richards . et al. 2006). We obtain L bol = 25 0 . , 13 0 . , 7 0 . × 10 12 L glyph[circledot] , for our three quasar bins; respectively. The properties of these three subsamples are summarized in Table 1 based on our measurements.
Aretxaga et al. (1998) report that the SFR of quasar host galaxies is > 100 -200 M glyph[circledot] yr -1 from a sample of three quasars at z = 2 using R and I band measurements. Jahnke et al. (2004) estimated a SFR ∼ 2 -30 M glyph[circledot] yr -1 for z = 1 8 . -2 6 quasar host galaxies using HST . ACS F606W and F850LP bands. Villforth et al. (2008) found a moderate SFRs of ∼ 33 M glyph[circledot] yr -1 for quasar host galaxies using multi-band data from U to K . Our SFR UV values from partially obscured quasar hosts are generally consistent with these previous observations.
Using the long-slit spectroscopy, Zafar et al. (2011) reported the detection of the host galaxy in the dark trough of the DLA in front of quasar Q0151+048A at z = 1 9. The detected positive residual flux well matches . the reported brightness of the host galaxy (Fynbo et al. 2000), and the measured counts in the dark trough is about a factor of 27 les than the quasar continuum in B-band. In Fig. 10, we compare our results with the detection of Zafar et al. (2011). The Ly α mean optical depth at z = 2 6 ( . 〈 τ 〉 ( z = 2 6)) is 0.3; and . 〈 τ 〉 ( z = 1 9) . =0.1. After correcting the mean Ly α absorption due to IGM, the result in Zafar et al. (2011) is a factor of 2 × higher than our results. Note Q0151+48A is one order of magnitude higher than BOSS quasars. Also, the detected residual flux for Q0151+48A is the Lyman continuum at λ ∼ 1216 ˚ , while the residual flux for our BOSS quasar A is the Lyman continuum at 〈 λ 〉 ∼ 1100 ˚ . Overall, the A difference between our results and that of Zafar et al. (2011) can be well interpreted by the large scatter of the covering fraction among different DLA clouds, and plus the uncertainties of the dust extinction among different quasar hosts.
Extended scattered nuclear light from the surrounding nebulosity could also be considered as a source for the emission, in addition to light from star formation (Young et al. 2009). The signature of scattered nuclear light is suggested by a few polarimetric mesurements (e.g. Smith et al. 2004; Letawe et al. 2007; Borguet et al. 2008). From Table 1, Fig. 4 and Fig. 5, the residual luminosity we detected is about 1.5% of quasar luminosity for the highest quasar luminosity bin. Observations of type 2 quasars suggest that scattering efficiencies can be at above or at 1% level (Zakamska et al. 2005). With the assumption that any broad Balmer line emission that is visible in the off-nuclear spectrum arises from scattered quasar light, Miller & Sheinis (2003) estimated the scattered light emission from four quasar host galaxies using Keck longslit spectroscopy, and found that the fraction
Fig. 10.We compare our results (red dots with errorbars) with the detection of Zafar et al. (2011) (black dot with errorbar). The plot shows the ratio of the dark trough residual intensity to the quasar luminosity density at rest-frame 1450 ˚ ( A L 1450 ). The dark trough intensity can be regarded as the unobscured FUV emission from the quasar host galaxies (see Section 4.2). After correcting the Ly α absorption due to IGM, the result in Zafar et al. (2011) is a factor of 2 × higher than our results. Note Q0151+48A is one order of magnitude higher than BOSS quasars. Also, the detected residual flux for Q0151+48A is the Lyman continuum at λ ∼ 1216 ˚, while the residual flux for our BOSS quasar is the Lyman conA tinuum at 〈 λ 〉 ∼ 1100 ˚. The difference between our results and A that of Zafar et al. (2011) can be well interpreted by the large scatter of the covering fraction among different DLA clouds, and plus the uncertainties of the dust extinction among different quasar hosts.
of the scattered nuclear light is generally small compared with the quasar host galaxies' stellar light (Miller & Sheinis 2003). However, Young et al. (2009) also report that based on their theoretical models, scattered light should be a concern for host characterization in high-redshift observations. From our current observations, any quantitative estimate of the quasar scattered light is difficult, and thus, extended scattered nuclear light remains a possibility for some of the residual flux. Thus SFR UV we derived could be regarded as a stringent upper limit for the quasar host galaxies.
## 5. SUMMARY AND FUTURE PROSPECTS
We define a Clean-DLA ( CDLA ) sample from quasars in SDSSIII-BOSS DR10 and stacked all of the spectra in three luminosity bins. In the composite spectra, we do not detect the Ly α emission coming from the DLA host galaxies. We find that the residual intensities in the DLA dark troughs correlate with the quasar luminosities (Fig. 4 and Fig. 5). We further discussed possible scenarios for the origin of the residual intensity (see details § 4), and conclude that it could be mainly contributed by FUV emission from the quasar host galaxies, rather than sky-subtraction residuals or DLA galaxy emission. For the highest luminosity bin with the median quasar bolometric luminosity of 2.5 × 10 13 L glyph[circledot] , the median dark trough residual intensity is = 16 4 . ± 2 1 . × 10 40 erg s -1 ˚ A -1 , correspnding to a 0.3 L ∗ galaxy at z ∼ 3 and an observed SFR UV of ∼ 9 M glyph[circledot] yr -1 . For the middle luminosity bin with the median quasar bolometric luminosity of 1.3 × 10 13 L glyph[circledot] , the median residual intensity in the stacked dark trough is = 8 8 . ± 2 3 . × 10 40 erg s -1 ˚ A -1 , correspnding to a 0.2 L ∗ galaxy at z ∼ 3 and an observed SFR UV of ∼ 5 M glyph[circledot] yr -1 . We do not detect the residual intensity in the stacked dark trough for the lowest bolometric lu-
minosity. The median quasar bolometric luminosity for the lowest luminosity bin is 7.0 × 10 12 L glyph[circledot] , and we put a 3σ upper limit on dark trough intensity of 7.5 × 10 40 erg s -1 ˚ A -1 , and a 3σ upper limit on SFR of 4.5 M glyph[circledot] yr -1 .
The more comprehensive study of the quasar host galaxies should combine the data from the rest-frame UV to rest-frame FIR. The SFR UV we derived may not represent a complete measure of the energy generated from high-redshift quasar hosts due to dust obscuration. Models suggest an evolutionary link between the optical quasar and local ULIRGs (SMGs at high redshift) (e.g. Sanders et al. 1988), where quasars are proposed to emerge from SMGs. Thus, it is possible that quasar host galaxies may still be heavily enshrouded by dust, and that a considerable amount of the UV emission from star formation is dust-obscured. FIR emission should provide a more comprehensive measure of the total SFR in dusty circumnuclear starbursts, and a direct comparison between SFR UV and SFR FIR provides an estimate for the dust obscuration in the quasar host galaxies.
However, quasars at z = 2 -5 with the similar bolometric luminosities with BOSS quasars have not been observed in the rest-frame FIR. With ALMA, one can probe a carefully-selected sample of BOSS quasars at z ∼ 2 -3 with reasonable exposure times. The previous studies have determined a relation between the bolometric luminosities ( L bol ) and rest-frame FIR luminosity ( L FIR ), using quasars at higher redshifts ( z glyph[greaterorsimilar] 5) and quasars at z = 2 -4 but with much higher bolometric luminosities than BOSS quasars (Wang et al. 2011; Omont et al. 2003, 2001; Carilli et al. 2001; Priddey et al. 2003). Assuming this relation between L bol and L FIR can be applied to quasars with similar luminosity to our sample at z = 2 -3, we can predict that for our highest luminosity bins, the L bol -inferred L FIR is ∼ 10 12 4 . L glyph[circledot] . Using FIR/sub-mm spectral energy distribtuions (SEDs), previous studies suggest that a significant fraction of the rest-frame FIR emission comes from massive star formation, possibly indicating the formation of early galactic bulges (Leipski et al. 2012; 2014). Several studies also simply suggest 50% of the FIR emission is powered by the star formation (e.g., Bertold et al. 2003; Wang et al. 2008, 2011). If we also assume 50% of the L bol -inferred
L FIR ∼ 10 12 4 . L glyph[circledot] is powered by star formation, the median FIR-derived star formation rate (SFR FIR ) for our highest quasar luminosity bin should be ∼ 300 M glyph[circledot] yr -1 . This expected SFR FIR is already more than one orders of magnitude higher than the SFR UV we derived.
Future detailed studies from rest-frame IR to sub-mm will constrain better dust heating models of host galaxies for BOSS quasars. The next generation spaced-based and ground-based telescope adaptive optics will provide significantly higher spatial resolution and enable much more precise PSF subtraction, and make it possible to fully resolve and accurately measure the UV emission from quasar hosts at z glyph[greaterorsimilar] 2. A direct comparison between the observed SFR FIR and SFR UV will directly probe the dust obscuration in the quasar host galaxies, and can be used to compare with the quasar evolutionary models.
Acknowledgement: We thank the anonymous referee for insightful comments which have significantly improved the paper. ZC thanks George Becker and J. Xavier Prochaska for useful discussions. ZC, XF acknowledge support from NSF grants AST 08-06861 and AST 11-07682. Funding for SDSS-III has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, and the U.S. Department of Energy Office of Science. The SDSS-III web site is http: //www.sdss3.org/. SDSSIII is managed by the Astrophysical Research Consortium for the Participating Institutions of the SDSSIII Collaboration including the University of Arizona, the Brazilian Participation Group, Brookhaven National Laboratory, University of Cambridge, Carnegie Mellon University, University of Florida, the French Participation Group, the German Participation Group, Harvard University, the Instituto de Astrofisica de Canarias, the Michigan State/Notre Dame/JINA Participation Group, Johns Hopkins University, Lawrence Berkeley National Laboratory, Max Planck Institute for Astrophysics, Max Planck Institute for Extraterrestrial Physics, New Mexico State University, New York University, Ohio State University, Pennsylvania State University, University of Portsmouth, Princeton University, the Spanish Participation Group, University of Tokyo, University of Utah, Vanderbilt University, University of Virginia, University of Washington, and Yale University.
-
## REFERENCES
Ahn, C. P., Alexandroff, R., Allende Prieto, C., Anders, F., Anderson, S. F., Anderton, T., Andrews, B. H., Aubourg, E., ´ Bailey, S., Bastien, F. A., & et al. 2013, ArXiv e-prints Angl´ es-Alc´zar, D., Ozel, F., & Dav´, R. 2013, ApJ, 770, 5 a ¨ e Aretxaga, I., Terlevich, R. J., & Boyle, B. J. 1998, MNRAS, 296, 643 Bian, F., Fan, X., Jiang, L., Dey, A., Green, R. F., Maiolino, R., Walter, F., McGreer, I., Wang, R., & Lin, Y.-T. 2012, ApJ, 757, 139 Bolton, A. S., Schlegel, D. J., Aubourg, E., Bailey, S., Bhardwaj, ´
V., Brownstein, J. R., Burles, S., Chen, Y.-M., Dawson, K., Eisenstein, D. J., Gunn, J. E., Knapp, G. R., Loomis, C. P., Lupton, R. H., Maraston, C., Muna, D., Myers, A. D., Olmstead, M. D., Padmanabhan, N., Pˆris, I., Percival, W. J., a Petitjean, P., Rockosi, C. M., Ross, N. P., Schneider, D. P., Shu, Y., Strauss, M. A., Thomas, D., Tremonti, C. A., Wake, D. A., Weaver, B. A., & Wood-Vasey, W. M. 2012, AJ, 144, 144
- Borguet, B., Hutsem´ ekers, D., Letawe, G., Letawe, Y., & Magain, P. 2008, A&A, 478, 321
Borthakur, S., Tripp, T. M., Yun, M. S., Bowen, D. V., Meiring,
- J. D., York, D. G., & Momjian, E. 2011, ApJ, 727, 52 Bovy, J., Hennawi, J. F., Hogg, D. W., Myers, A. D., Kirkpatrick,
- J. A., Schlegel, D. J., Ross, N. P., Sheldon, E. S., McGreer,
- I. D., Schneider, D. P., & Weaver, B. A. 2011, ApJ, 729, 141 Braun, R. & Walterbos, R. A. M. 1992, ApJ, 386, 120 Bridge, C. R., Teplitz, H. I., Siana, B., Scarlata, C., Conselice,
- C. J., Ferguson, H. C., Brown, T. M., Salvato, M., Rudie,
- G. C., de Mello, D. F., Colbert, J., Gardner, J. P., Giavalisco,
- M., & Armus, L. 2010, ApJ, 720, 465
Bruzual, G. & Charlot, S. 2003, MNRAS, 344, 1000
- Carilli, C. L., Bertoldi, F., Omont, A., Cox, P., McMahon, R. G., & Isaak, K. G. 2001, AJ, 122, 1679
- Carilli, C. L. & Wang, R. 2006, AJ, 132, 2231
Cassata, P., Le F`vre, O., Garilli, B., Maccagni, D., Le Brun, V., e Scodeggio, M., Tresse, L., Ilbert, O., Zamorani, G., Cucciati, O., Contini, T., Bielby, R., Mellier, Y., McCracken, H. J., Pollo, A., Zanichelli, A., Bardelli, S., Cappi, A., Pozzetti, L., Vergani, D., & Zucca, E. 2011, A&A, 525, A143
- Ciardullo, R., Gronwall, C., Wolf, C., McCathran, E., Bond, Padilla, N., Francke, H., Matkovi´, A., Altmann, M., & c Herrera, D. 2012, ApJ, 744, 110
- N. A., Gawiser, E., Guaita, L., Feldmeier, J. J., Treister, E., Dawson, K. S., Schlegel, D. J., Ahn, C. P., et al. 2013, AJ, 145, 10 Dijkstra, M. & Westra, E. 2010, MNRAS, 401, 2343 Ferguson, H. C., Dickinson, M., Giavalisco, M., Kretchmer, C., Ravindranath, S., Idzi, R., Taylor, E., Conselice, C. J., Fall, S. M., Gardner, J. P., Livio, M., Madau, P., Moustakas, L. A., Papovich, C. M., Somerville, R. S., Spinrad, H., & Stern, D. 2004, ApJ, 600, L107 Ferrarese, L. & Merritt, D. 2000, ApJ, 539, L9 Finley, H., Petitjean, P., Pˆris, I., Noterdaeme, P., Brinkmann, a J., Myers, A. D., Ross, N. P., Schneider, D. P., Bizyaev, D., Brewington, H., Ebelke, G., Malanushenko, E., Malanushenko, V., Oravetz, D., Pan, K., Simmons, A., & Snedden, S. 2013, ArXiv e-prints Font-Ribera, A., Miralda-Escud´, J., Arnau, E., Carithers, B., e Lee, K.-G., Noterdaeme, P., Pˆris, I., Petitjean, P., Rich, J., a Rollinde, E., Ross, N. P., Schneider, D. P., White, M., & York, D. G. 2012, JCAP, 11, 59 Fynbo, J. P. U., Prochaska, J. X., Sommer-Larsen, J., Dessauges-Zavadsky, M., & Møller, P. 2008, ApJ, 683, 321 Fynbo, J. U., Møller, P., & Warren, S. J. 1999, MNRAS, 305, 849 Gebhardt, K., Bender, R., Bower, G., Dressler, A., Faber, S. M., Filippenko, A. V., Green, R., Grillmair, C., Ho, L. C., Kormendy, J., Lauer, T. R., Magorrian, J., Pinkney, J., Richstone, D., & Tremaine, S. 2000, ApJ, 539, L13 Guyon, O., Sanders, D. B., & Stockton, A. 2006, ApJS, 166, 89 Haehnelt, M. G., Steinmetz, M., & Rauch, M. 2000, ApJ, 534, 594 Hao, C. N., Xia, X. Y., Mao, S., Wu, H., & Deng, Z. G. 2005, ApJ, 625, 78 H¨ aring, N. & Rix, H.-W. 2004, ApJ, 604, L89 Hennawi, J. F., Prochaska, J. X., Kollmeier, J., & Zheng, Z. 2009, ApJ, 693, L49 Hopkins, P. F. & Quataert, E. 2011, MNRAS, 415, 1027 Hopkins, P. F., Somerville, R. S., Hernquist, L., Cox, T. J., Robertson, B., & Li, Y. 2006, ApJ, 652, 864 Hutchings, J. B., Frenette, D., Hanisch, R., Mo, J., Dumont, P. J., Redding, D. C., & Neff, S. G. 2002, AJ, 123, 2936 Jahnke, K., S´nchez, S. F., Wisotzki, L., Barden, M., Beckwith, a S. V. W., Bell, E. F., Borch, A., Caldwell, J. A. R., H¨ussler, a B., Heymans, C., Jogee, S., McIntosh, D. H., Meisenheimer, K., Peng, C. Y., Rix, H.-W., Somerville, R. S., & Wolf, C. 2004, ApJ, 614, 568 Kanekar, N., Lane, W. M., Momjian, E., Briggs, F. H., & Chengalur, J. N. 2009, MNRAS, 394, L61 Kashikawa, N., Misawa, T., Minowa, Y., Okoshi, K., Hattori, T., Toshikawa, J., Ishikawa, S., & Onoue, M. 2013, ArXiv e-prints Kennicutt, Jr., R. C. 1998, ApJ, 498, 541 Kanekar, N., Lane, W. M., Momjian, E., Briggs, F. H., & Chengalur, J. N. 2009, MNRAS, 394, L61 Komatsu, E., Dunkley, J., Nolta, M. R., Bennett, C. L., Gold, B., Hinshaw, G., Jarosik, N., Larson, D., Limon, M., Page, L., Spergel, D. N., Halpern, M., Hill, R. S., Kogut, A., Meyer, S. S., Tucker, G. S., Weiland, J. L., Wollack, E., & Wright, E. L. 2009, ApJS, 180, 330 Krogager, J.-K., Fynbo, J. P. U., Møller, P., Ledoux, C., Noterdaeme, P., Christensen, L., Milvang-Jensen, B., & Sparre, M. 2012, MNRAS, 424, L1 Kukula, M. J., Dunlop, J. S., McLure, R. J., Miller, L., Percival, W. J., Baum, S. A., & O'Dea, C. P. 2001, MNRAS, 326, 1533 Leitherer, C. & Heckman, T. M. 1995, ApJS, 96, 9 Letawe, G., Magain, P., Courbin, F., Jablonka, P., Jahnke, K., Meylan, G., & Wisotzki, L. 2007, MNRAS, 378, 83 Lutz, D., Sturm, E., Tacconi, L. J., Valiante, E., Schweitzer, M., Netzer, H., Maiolino, R., Andreani, P., Shemmer, O., & Veilleux, S. 2008, ApJ, 684, 853
- Madau, P., Pozzetti, L., & Dickinson, M. 1998, ApJ, 498, 106 Magorrian, J., Tremaine, S., Richstone, D., Bender, R., Bower, G., Dressler, A., Faber, S. M., Gebhardt, K., Green, R., Grillmair, C., Kormendy, J., & Lauer, T. 1998, AJ, 115, 2285
Malkan, M., Webb, W., & Konopacky, Q. 2003, ApJ, 598, 878 Marconi, A. & Hunt, L. K. 2003, ApJ, 589, L21
Martel, A. R., Ford, H. C., Tran, H. D., Illingworth, G. D., Krist, J. E., White, R. L., Sparks, W. B., Gronwall, C., Cross, N. J. G., Hartig, G. F., Clampin, M., Ardila, D. R., Bartko, F., Ben´ ıtez, N., Blakeslee, J. P., Bouwens, R. J., Broadhurst, T. J., Brown, R. A., Burrows, C. J., Cheng, E. S., Feldman, P. D., Franx, M., Golimowski, D. A., Infante, L., Kimble, R. A., Lesser, M. P., McCann, W. J., Menanteau, F., Meurer, G. R., Miley, G. K., Postman, M., Rosati, P., Sirianni, M., Tsvetanov, Z. I., & Zheng, W. 2003, AJ, 125, 2964
- McLure, R. J., Kukula, M. J., Dunlop, J. S., Baum, S. A., O'Dea, C. P., & Hughes, D. H. 1999, MNRAS, 308, 377
Mechtley, M., Windhorst, R. A., Ryan, R. E., Schneider, G.,
Cohen, S. H., Jansen, R. A., Fan, X., Hathi, N. P., Keel, W. C.,
Koekemoer, A. M., R¨ttgering, H., Scannapieco, E., Schneider,
o
D. P., Strauss, M. A., & Yan, H. J. 2012, ApJ, 756, L38
Miller, J. S. & Sheinis, A. I. 2003, ApJ, 588, L9
- Mo, H. J., Mao, S., & White, S. D. M. 1999, MNRAS, 304, 175 Møller, P., Fynbo, J. P. U., Ledoux, C., & Nilsson, K. K. 2013, MNRAS, 430, 2680
- Nagamine, K., Springel, V., & Hernquist, L. 2004, MNRAS, 348, 435
- Noterdaeme, P., Petitjean, P., Carithers, W. C., Pˆris, I., a Font-Ribera, A., Bailey, S., Aubourg, E., Bizyaev, D., Ebelke, G., Finley, H., Ge, J., Malanushenko, E., Malanushenko, V., Miralda-Escud´ e, J., Myers, A. D., Oravetz, D., Pan, K., Pieri, M. M., Ross, N. P., Schneider, D. P., Simmons, A., & York, D. G. 2012, A&A, 547, L1
- Noterdaeme, P., Petitjean, P., Ledoux, C., & Srianand, R. 2009, A&A, 505, 1087
- Noterdaeme, P., Petitjean, P., Pˆris, I., et al. 2014, A&A, 566, a A24
- Omont, A., Beelen, A., Bertoldi, F., Cox, P., Carilli, C. L., Priddey, R. S., McMahon, R. G., & Isaak, K. G. 2003, A&A, 398, 857
- Omont, A., Cox, P., Bertoldi, F., McMahon, R. G., Carilli, C., & Isaak, K. G. 2001, A&A, 374, 371
Pˆ aris, I., Petitjean, P., Aubourg, E., Bailey, S., Ross, N. P., ´ Myers, A. D., Strauss, M. A., Anderson, S. F., Arnau, E., Bautista, J., Bizyaev, D., Bolton, A. S., Bovy, J., Brandt, W. N., Brewington, H., Browstein, J. R., Busca, N., Capellupo, D., Carithers, W., Croft, R. A. C., Dawson, K., Delubac, T., Ebelke, G., Eisenstein, D. J., Engelke, P., Fan, X., Filiz Ak, N., Finley, H., Font-Ribera, A., Ge, J., Gibson, R. R., Hall, P. B., Hamann, F., Hennawi, J. F., Ho, S., Hogg, D. W., Ivezi´, Z., c ˇ Jiang, L., Kimball, A. E., Kirkby, D., Kirkpatrick, J. A., Lee, K.-G., Le Goff, J.-M., Lundgren, B., MacLeod, C. L., Malanushenko, E., Malanushenko, V., Maraston, C., McGreer, I. D., McMahon, R. G., Miralda-Escud´ e, J., Muna, D., Noterdaeme, P., Oravetz, D., Palanque-Delabrouille, N., Pan, K., Perez-Fournon, I., Pieri, M. M., Richards, G. T., Rollinde, E., Sheldon, E. S., Schlegel, D. J., Schneider, D. P., Slosar, A., Shelden, A., Shen, Y., Simmons, A., Snedden, S., Suzuki, N., Tinker, J., Viel, M., Weaver, B. A., Weinberg, D. H., White, M., Wood-Vasey, W. M., & Y` eche, C. 2012, A&A, 548, A66
Pˆ aris, I., Petitjean, P., Aubourg, E., Ross, N. P., Myers, A. D., ´ Streblyanska, A., Bailey, S., Hall, P. B., Strauss, M. A., Anderson, S. F., Bizyaev, D., Borde, A., Brinkmann, J., Bovy, J., Brandt, W. N., Brewington, H., Brownstein, J. R., Cook, B. A., Ebelke, G., Fan, X., Filiz Ak, N., Finley, H., Font-Ribera, A., Ge, J., Hamann, F., Ho, S., Jiang, L., Kinemuchi, K., Malanushenko, E., Malanushenko, V., Marchante, M., McGreer, I. D., McMahon, R. G., Miralda-Escud´ e, J., Muna, D., Noterdaeme, P., Oravetz, D., Palanque-Delabrouille, N., Pan, K., Perez-Fournon, I., Pieri, M., Riffel, R., Schlegel, D. J., Schneider, D. P., Simmons, A., Viel, M., Weaver, B. A., Wood-Vasey, W. M., Y`che, C., & e York, D. G. 2013, ArXiv e-prints
- Peng, C. Y., Impey, C. D., Rix, H.-W., Kochanek, C. S., Keeton, C. R., Falco, E. E., Leh´r, J., & McLeod, B. A. 2006, ApJ, 649, a 616
- Pontzen, A., Governato, F., Pettini, M., Booth, C. M., Stinson, G., Wadsley, J., Brooks, A., Quinn, T., & Haehnelt, M. 2008, MNRAS, 390, 1349
Priddey, R. S., Isaak, K. G., McMahon, R. G., & Omont, A. 2003, MNRAS, 339, 1183 Rahmani, H., Srianand, R., Noterdaeme, P., & Petitjean, P. 2010, MNRAS, 409, L59 Rauch, M., Haehnelt, M., Bunker, A., Becker, G., Marleau, F., Graham, J., Cristiani, S., Jarvis, M., Lacey, C., Morris, S., Peroux, C., R¨ttgering, H., & Theuns, T. 2008, ApJ, 681, 856 o Reddy, N. A., Steidel, C. C., Pettini, M., Adelberger, K. L., Shapley, A. E., Erb, D. K., & Dickinson, M. 2008, ApJS, 175, 48 Richards, G. T., Lacy, M., Storrie-Lombardi, L. J., Hall, P. B., Gallagher, S. C., Hines, D. C., Fan, X., Papovich, C., Vanden Berk, D. E., Trammell, G. B., Schneider, D. P., Vestergaard, M., York, D. G., Jester, S., Anderson, S. F., Budav´ri, T., & a Szalay, A. S. 2006, ApJS, 166, 470 Ridgway, S. E., Heckman, T. M., Calzetti, D., & Lehnert, M. 2001, ApJ, 550, 122 Ross, N. P., Myers, A. D., Sheldon, E. S., Y`che, C., Strauss, e M. A., Bovy, J., Kirkpatrick, J. A., Richards, G. T., Aubourg, ´ ., E Blanton, M. R., Brandt, W. N., Carithers, W. C., Croft, R. A. C., da Silva, R., Dawson, K., Eisenstein, D. J., Hennawi, J. F., Ho, S., Hogg, D. W., Lee, K.-G., Lundgren, B., McMahon, R. G., Miralda-Escud´ e, J., Palanque-Delabrouille, N., Pˆris, I., Petitjean, P., Pieri, M. M., Rich, J., Roe, N. A., a Schiminovich, D., Schlegel, D. J., Schneider, D. P., Slosar, A., Suzuki, N., Tinker, J. L., Weinberg, D. H., Weyant, A., White, M., & Wood-Vasey, W. M. 2012, ApJS, 199, 3 Sanders, D. B., Scoville, N. Z., & Soifer, B. T. 1988, ApJ, 335, L1 Schaye, J., Aguirre, A., Kim, T.-S., Theuns, T., Rauch, M., & Sargent, W. L. W. 2003, ApJ, 596, 768 Schramm, M., Wisotzki, L., & Jahnke, K. 2008, A&A, 478, 311
- Serjeant, S. & Hatziminaoglou, E. 2009, MNRAS, 397, 265
- Shapley, A. E., Steidel, C. C., Erb, D. K., Reddy, N. A.,
- Adelberger, K. L., Pettini, M., Barmby, P., & Huang, J. 2005, ApJ, 626, 698 Siana, B., Teplitz, H. I., Colbert, J., Ferguson, H. C., Dickinson, M., Brown, T. M., Conselice, C. J., de Mello, D. F., Gardner, J. P., Giavalisco, M., & Menanteau, F. 2007, ApJ, 668, 62 Smith, J. E., Robinson, A., Alexander, D. M., Young, S., Axon, D. J., & Corbett, E. A. 2004, MNRAS, 350, 140 Steidel, C. C., Erb, D. K., Shapley, A. E., Pettini, M., Reddy, N., Bogosavljevi´, M., Rudie, G. C., & Rakic, O. 2010, ApJ, 717, c 289 Targett, T. A., Dunlop, J. S., Cirasuolo, M., McLure, R. J., Bruce, V. A., Fontana, A., Galametz, A., Paris, D., Dav´, R., e Dekel, A., Faber, S. M., Ferguson, H. C., Grogin, N. A., Kartaltepe, J. S., Kocevski, D. D., Koekemoer, A. M., Kurczynski, P., Lai, K., & Lotz, J. 2013, MNRAS, 432, 2012 Villforth, C., Heidt, J., & Nilsson, K. 2008, A&A, 488, 133 Wang, R., Wagg, J., Carilli, C. L., Neri, R., Walter, F., Omont, A., Riechers, D. A., Bertoldi, F., Menten, K. M., Cox, P., Strauss, M. A., Fan, X., & Jiang, L. 2011, AJ, 142, 101 Wang, R., Wagg, J., Carilli, C. L., Walter, F., Lentati, L., Fan, X., Riechers, D. A., Bertoldi, F., Narayanan, D., Strauss, M. A., Cox, P., Omont, A., Menten, K. M., Knudsen, K. K., Neri, R., & Jiang, L. 2013, ArXiv e-prints Young, J. E., Eracleous, M., Shemmer, O., Netzer, H., Gronwall, C., Lutz, D., Ciardullo, R., & Sturm, E. 2013, ArXiv e-prints Young, S., Axon, D. J., Robinson, A., & Capetti, A. 2009, ApJ, 698, L121 Yun, M. S., Carilli, C. L., Kawabe, R., Tutui, Y., Kohno, K., & Ohta, K. 2000, ApJ, 528, 171 Zafar, T., Møller, P., Ledoux, C., et al. 2011, A&A, 532, A51 Zakamska, N. L., Schmidt, G. D., Smith, P. S., Strauss, M. A., Krolik, J. H., Hall, P. B., Richards, G. T., Schneider, D. P., Brinkmann, J., & Szokoly, G. P. 2005, AJ, 129, 1212 Zwaan, M. A., van der Hulst, J. M., Briggs, F. H., Verheijen, M. A. W., & Ryan-Weber, E. V. 2005, MNRAS, 364, 1467
TABLE 1
## Dark Trough Intensities for different Quasar Luminosity Bins
| I QSO ( λ = 1450A) ˚ (erg s - 1 ˚ A - 1 ) | L bol (10 12 L glyph[circledot] ) | I trough (median) (erg s - 1 ˚ A - 1 ) | I trough (3 σ -clipped mean) (M glyph[circledot] yr - 1 ) | SFR UV (M glyph[circledot] yr - 1 ) |
|---------------------------------------------|-------------------------------------|------------------------------------------|-------------------------------------------------------------|---------------------------------------|
| 1 . 3 × 10 43 | 25 | 16 . 4 ± 2 . 1 × 10 40 | 15 . 2 ± 2 . 1 × 10 40 | 9 |
| 6 . 6 × 10 42 | 13 | 8 . 8 ± 2 . 3 × 10 40 | 8 . 2 ± 2 . 3 × 10 40 | 5 |
| 4 . 4 × 10 42 | 7 | 3- σ < 7 . 5 × 10 40 | 3- σ < 7 . 5 × 10 40 | 3- σ < 4 . 5 |
a The intensities I trough are calculated using Eq. (1) which mean the sky-subtraction residual has been corrected
|
10.1088/0004-637X/793/2/139
|
[
"Zheng Cai",
"Xiaohui Fan",
"Pasquier Noterdaeme",
"Ran Wang",
"Ian McGreer",
"Bill Carithers",
"Fuyan Bian",
"Jordi Miralda Escude",
"Hayley Finley",
"Isabelle Paris",
"Donald P. Schneider",
"Nadia L. Zakamska",
"Jian Ge",
"Patrick Petitjean",
"Anze Slosar"
] |
2014-08-04T22:02:06+00:00
|
2014-08-04T22:02:06+00:00
|
[
"astro-ph.GA"
] |
A Glimpse at Quasar Host Galaxy Far-UV Emission, Using DLAs as Natural Coronagraphs
|
In merger-driven models of massive galaxy evolution, the luminous quasar
phase is expected to be accompanied by vigorous star formation in quasar host
galaxies. In this paper, we use high column density Damped Lyman Alpha (DLA)
systems along quasar sight lines as natural coronagraphs to directly study the
far-UV (FUV) radiation from the host galaxies of luminous background quasars.
We have stacked the spectra of $\sim$2,000 DLA systems (N_HI>10^{20.6} cm^{-2})
with a median absorption redshift < z > = 2.6 selected from quasars observed in
the SDSS-III Baryon Oscillation Spectroscopic Survey. We detect residual flux
in the dark troughs of the composite DLA spectra. The level of this residual
flux significantly exceeds systematic errors in the SDSS fiber sky subtraction;
furthermore, the residual flux is strongly correlated with the continuum
luminosity of the background quasar, while uncorrelated with DLA column density
or metallicity. We conclude that the flux could be associated with the average
FUV radiation from the background quasar host galaxies (with medium redshift <
z > = 3.1) that is not blocked by the intervening DLA. Assuming all of the
detected flux originates from quasar hosts, for the highest quasar luminosity
bin (< L >= 2.5x 10^{13} L_sun), the host galaxy has a FUV intensity of 1.5 +/-
0.2 x 10^{40} erg s^{-1} A^{-1}; this corresponds to an unobscured UV star
formation rate of 9 M_sun/yr.
|
1408.0830v2
|
## FORMAL GEOMETRY FOR NONCOMMUTATIVE MANIFOLDS
## HENDRIK OREM
Abstract. This paper develops the tools of formal algebraic geometry in the setting of noncommutative manifolds, roughly ringed spaces locally modeled on the free associative algebra. We define a notion of noncommutative coordinate system, which is a principal bundle for an appropriate group of local coordinate changes. These bundles are shown to carry a natural flat connection with properties analogous to the classical Gelfand-Kazhdan structure.
Every noncommutative manifold has an underlying smooth variety given by abelianization. A basic question is existence and uniqueness of noncommutative thickenings of a smooth variety, i.e., finding noncommutative manifolds abelianizing to a given smooth variety. We obtain new results in this direction by showing that noncommutative coordinate systems always arise as reductions of structure group of the commutative bundle of coordinate systems on the underlying smooth variety; this also explains a relationship between D -modules on the commutative variety and sheaves of modules for the noncommutative structure sheaf.
## Contents
| 1. | Introduction | 2 |
|------------|----------------------------------------------------------------|-----|
| 1.1. | Acknowledgements | 5 |
| 2. | Preliminaries | 5 |
| 2.1. | Definitions and Notation | 5 |
| 2.2. | Recollections on Formal Geometry | 7 |
| 3. | Noncommutative Formal Geometry | 7 |
| 3.1. | Automorphisms of the Noncommutative Disk | 7 |
| 3.2. | The Functor of Noncommutative Jets | 8 |
| 3.3. | The Frame Bundle Functor | 8 |
| 3.4. | The Noncommutative Gelfand-Kazhdan Structure | 9 |
| 4. | NC-Thickenings, NC-Coordinate Systems, and NC-Connections | 11 |
| 4.1. | Relation with NC-Smooth Thickenings | 11 |
| 4.2. | Relation with NC-Connections | 11 |
| 5. | NC-Thickenings as Reductions of Structure Group | 12 |
| 5.1. | Existence Results | 13 |
| 5.2. | Obstruction Classes and the NC-Atiyah Class | 13 |
| 6. | Degeneration to the Associated Graded and Lower Central Series | 14 |
| 6.1. | The Lower Central Series Filtration | 14 |
| 6.2. | Degeneration | 14 |
| 6.3. | Categorical Geometry of Lower Central Series Algebras | 14 |
| 7. | Relation to D -Modules | 15 |
| 7.1. | Construction of Functor | 15 |
| 7.2. | Comparison to the Functor of [PT14] | 16 |
| References | References | 16 |
## 1. Introduction
Throughout X is a smooth variety over C of dimension n .
Definition 1.1. Let A be an associative C -algebra. Define the lower central series filtration as follows. Set L 1 ( A ) = A , and L k ( A ) = [ A, L k -1 ( A )]. Then the lower central series ideals are
$$M _ { k } ( A ) = A L _ { k } ( A ) A = A L _ { k } ( A ).$$
We say that A is NC-complete if A is complete for the filtration M A k ( ). The NC-completion of an algebra is its completion with respect to the filtration M k .
The remarkable paper of Kapranov [Kap98] puts forward a framework of noncommutative geometry in which the local objects are NC-complete algebras with a smoothness property.
Definition 1.2. Let A be a NC-complete C -algebra, π : A → A ab the abelianization map, and x ∈ Spec A ab . Denote by S = π -1 ( S ) ⊂ A the multiplicative subset corresponding to a multiplicative subset S ⊂ A ab .
- 1 The stalk of A at x is the direct limit of all localizations A S [ -1 ], where S ⊂ A ab runs over multiplicative subsets of functions not vanishing at x .
- 2 An NC-complete algebra A is NC-smooth of dimension n if all of its completed stalks are isomorphic to A n = k 〈〈 x , . . . , x 1 n 〉〉 .
- ̂ 3 An affine NC-manifold is a pair (Spec A ,A ab ) of a smooth affine variety and NC-smooth algebra A abelianizing to it.
- 4 An NC-manifold is a smooth variety X of dimension n and a sheaf A of associative C -algebras which is locally an affine NC-manifold of dimension n .
- 5 An NC-thickening of a smooth variety X is a sheaf of algebras A such that ( X, A ) is an NC-manifold. NC-thickenings of X form a category NC -Th X .
Remark 1.3 . NC-manifolds were first defined using a different filtration by commutator ideals; however, this filtration induces the same topology as M k (see claim 2.4 of [JO14]), and therefore the notion NC-complete is the same.
Example 1.4. Affine examples are the (NC-completions of) the locally free algebras in [JO14]:
- 1 The completed free algebra ̂ A n is an NC-thickening of the formal disk. 2 For any smooth complete intersection X = C [ x , . . . , x 1 n + m ] / f , . . . , f 〈 1 m 〉 , the algebra A = A n + m / f , . . . , f 〈 ˜ 1 ˜ m 〉 , where each ˜ f i ∈ f i + M 2 , is locally free. Its NC-completion is an NC-thickening of X .
- 3 Let A be locally free, and recall the notation of definition 1.2. If ¯ is any multiplicative S subset of A/M 2 without zero divisors, then A S [ -1 ] is locally free.
- 4 Any formally smooth (sometimes called quasi-free) algebra is locally free, by the formal tubular neighborhood theorem ([CQ95], Section 6, Theorem 2).
Section 5 of [Kap98] contains non-affine examples:
- 1 There is an NC-thickening of projective space P n given by gluing n + 1 copies of the NC-completion of C 〈 x , . . . , x 1 n 〉 .
- 2 NC-thickenings of Grassmannians and flag varieties arise as natural noncommutative variants of the classical functors of points.
- 3 Under certain conditions, Kapranov constructs an NC-thickening of the moduli of vector bundles on a projective variety.
This paper builds on ideas of [Kap98], [JO14], and [PT14], which had a unifying theme of studying NC-manifolds via commutative algebraic geometry on the abelianization. Kapranov constructs an NC-thickening of arbitrary smooth affine varieties X , which are unique but not functorial, so this does not give the existence of NC-thickenings of non-affine X . One goal of this paper is to give a new geometric criterion for the existence of NC-thickenings; we accomplish this by developing an analogue of formal algebraic geometry for NC-manifolds.
The bundle of coordinate systems M of X is defined by its fiber over x ∈ X being the space of isomorphisms between the formal neighborhood of x and the abstract formal disk Spec ̂ O n = Spec C /llbracket x , . . . , x 1 n /rrbracket . It is a principal bundle for the pro-algebraic group G + n of augmented algebra automorphisms of ̂ O n ; we denote by G n the full group of algebra automorphisms. Formal geometry, the study of the bundle M , goes back to [GK71], and has been used to great effect for example in [BZF04], [BD04], [BK04], [Gai11], and [Kap99].
The bundle of coordinate systems carries many rich structures, encapsulating information about sheaves of jets and closely related to the L ∞ -spaces of [Cos11]. Our goal is to reformulate questions about NC-manifolds in terms of analogous objects we call noncommutative coordinate systems . These are principal bundles for H + n = Aut aug ̂ A n , augmented automorphisms of the formal noncommutative disk , arising from NC-thickenings roughly as frame bundles, and form a category NC -Coord X .
A key aspect of classical formal geometry is the Gelfand-Kazhdan structure on M . This is a splitting of the Atiyah sequence of M , valued in Lie G n , where G n is the full group of automorphisms of ̂ O n . As explained in [Kap99], this splitting carries within it the L ∞ structure on T X [ -1], as well as the flat connection on all jet bundles. In section 3.4, we discover a noncommutative Gelfand-Kazhdan structure on any noncommutative coordinate system N .
Theorem 4.1. The functor Coord : NC -Th X → NC -Coord X is an equivalence of categories. The inverse is given by taking flat sections of Assoc N ( ̂ A n ) under the NC-Gelfand Kazhdan structure on N .
To a certain extent this perspective is visible in section 4.4 of [Kap98], but the foundations and implications of noncommutative formal geometry are not systematically developed.
A recent paper [PT14] makes several remarkable observations about NC-thickenings of a given variety X . The authors define a notion of NC-connection:
Definition 1.5. An NC-connection is a degree one, square zero derivation D of
$$\mathcal { A } _ { X } \coloneqq \Omega _ { X } ^ { \bullet } \otimes _ { \mathcal { O } _ { X } } \widehat { T } _ { \mathcal { O } _ { X } } \Omega _ { X } ^ { 1 } \\ \text{ntial on } \Omega _ { X } ^ { \bullet } \text{ and acting on } \Omega _ { X } ^ { 1 }$$
$$D ( 1 \otimes \alpha ) = \alpha \otimes 1 + \nabla _ { 1 } ( \alpha ) + \nabla _ { 2 } ( \alpha ) + \cdots$$
extending the de Rham differential on Ω · X and acting on Ω 1 X ⊂ ̂ T O X Ω 1 X as where ∇ i ( α ) ∈ Ω 1 X ⊗ O X T i (Ω 1 X ).
A twisted NC-connection is a sheaf of algebras J and a sheaf of ideals I such that ( J I , ) is locally isomorphic to ( ̂ T O X (Ω 1 X ) , T ̂ ≥ 1 O X (Ω 1 X )), equipped with an analogue of the derivation D (see definition 2.6).
NC-thickenings are shown to be equivalent to NC-connections:
Theorem 2.3.23 in [PT14] . Given a twisted NC-connection with derivation D , ker D is an NC-thickening of X , and this induces an equivalence of categories NC -Conn X ∼ → NC -Th X .
Second, the authors construct a functor from the category D ( X ) of D -modules on X to modules over A for any NC-thickening A of X . We explain how both the notion of NCconnection and the functor from D ( X ) arises naturally from the perspective of NC-coordinate systems:
Theorem 4.4. Let A be an NC-thickening of X , and N an NC-coordinate system. Then the NC-Gelfand-Kazhdan structure on N induces a twisted NC-connection on Assoc N ̂ A n ; this construction induces an equivalence of categories NC -Coord X ∼ → NC -Conn X .
This is fully analogous to the commutative situation, where an L ∞ -algebra structure on T X [ -1] arises from the geometry of the bundle of coordinate systems [Kap98]; this analogy is discussed further at the end of the introduction.
Let X dR denote the de Rham stack of X [Sim96]; recall that QC( X dR ) = D ( X ). The classical Gelfand-Kazhdan structure descends the map X → BG + n classifying M to a map X dR → BG n . The appearance of the de Rham stack in noncommutative formal geometry gives the following relation between D ( X ) and modules over NC-thickenings A .
Proposition 7.1 (Informal). The (commutative) bundle of coordinate systems M and an NC-coordinate system N induce a commutative diagram
/d123
/d123
/d35
/d35
/d47
/d47
which naturally gives functors
Note that, in particular, N is a reduction of structure group M from G + n to H + n . This perspective has a number of applications. For example, we give a new description of the existence of NC-thickenings of a variety X .
$$\text{which naturally gives functors} \\ \text{ind} _ { q } \left ( Q C ( B G _ { n } ) \right ) & \cong \mathcal { D } ( X ) \rightarrow \text{ind} _ { p } \left ( \widehat { A } _ { n } \text{-mod} ( Q C ( B H _ { n } ) ) \right ) \rightarrow \mathcal { A } \text{-mod} \,. \\ \text{Note that, in particular, $\mathcal{N}$ is a reduction of structure group $\mathcal{M}$ from $G_{n}^{+}$ to $H_{n}^{+}$. This}$$
Corollary 5.2. Asmooth variety X admits an NC-thickening if and only if Assoc M ( G /H + n + n ) has global sections.
Note that the quotient G /H + n + n is taken as an algebraic stack, and thus Assoc M ( G /H n n ) is a sheaf of stacks over X . The perspective of NC-coordinate systems also provides a globalization of local statements about H n . In the same way that representations of G n give via formal geometry a universal construction of natural sheaves on smooth varieties, the representation theory of H n is a source for natural sheaves on NC-manifolds.
Finally, we study how NC-coordinate systems degenerate to the associated graded along the lower central series filtration and give a categorification of the main result of [JO14]. In particular, the associated graded algebra gr M k ( ̂ A n ) carries an action of G n . Thus gr M k ( ̂ A n ) -mod is a G n -category. The main result of [JO14] is that if X = Spec A ab is affine with NCthickening A , then
$$g r _ { M _ { k } } ( A ) \cong \Gamma ( X, \text{Assoc} _ { \mathcal { M } } ( g r _ { M _ { k } } ( \widehat { A } _ { n } ) ) ^ { \nabla } ),$$
where ( -) ∇ denotes the sheaf of flat sections. We now state an informal analogue of this statement for categories of modules.
## Theorem 6.6 (Informal). There is an equivalence of categories
$$\begin{array} { c } \cdot & \cdot & \cdot & \cdot \\ g r _ { M } ( A ) \text{-mod} ( Q C ( X ) ) \xrightarrow { \sim } \Gamma ( X, \text{Assoc} _ { \mathcal { M } } ( g r _ { M } ( \widehat { A } _ { n } ) \text{-mod} ) ^ { \nabla } ). \\ \text{conclude the introduction with an explanation of how this paper and [PT14] } \text{non-qual} \text{ how this non-ratio more than}. \end{array}$$
We conclude the introduction with an explanation of how this paper and [PT14] relate to Kapranov's seminal paper [Kap99], as well as how this paper relates more broadly to noncommutative geometry and Koszul duality. Among other things, Kapranov constructed a (cochain level) equivalence between JO X (with its natural flat connection) and ̂ Sym O Ω , 1 with structure maps dual to an L ∞ -structure on T X [ -1]. Equivalently, Kapranov constructed an isomorphism between a natural enhancement of the formal neighborhood of the diagonal in X × X and a natural enhancement of the formal neighborhood of the zero section in the tangent bundle. The L ∞ -structure maps on T X [ -1] were constructed via the geometry of the bundle of coordinate systems on X .
Analogously, in [PT14] the authors recover A (the 'structure sheaf' of a NC-manifold) from the kernel of a collection of structure maps on ̂ T O Ω . This paper shows how to construct 1 these structure maps from the bundle of NC-coordinate systems associated with A . The interpretation of these structure maps in an analogous way as a Koszul dual homotopy algebra structure will be the subject of future work.
From the perspective of Koszul duality, this paper can be viewed as the case of 'manifolds over the associative operad' in the following philosophy: the structure maps of the Koszul dual split the Atiyah sequence of the bundle of coordinate systems. Kapranov's result explained above is the commutative version (where the Koszul dual is the L ∞ -algebra T X [ -1]); here we expect to interpret the splitting maps for noncommutative manifolds in terms of the Koszul dual A ∞ algebra. A version of this is already visible in sections 5.3, 5.4 and 7.1 of [PT14], in an analytic setting. This suggests a deep connection between Gelfand-Kazhdan structures and Koszul duality of operads.
- 1.1. Acknowledgements. The author would like to thank David Ben-Zvi and David Jordan for many helpful conversations about formal geometry and noncommutative algebra, and Alexander Polishchuk, Travis Schedler, and Michel Van den Bergh for remarks on earlier drafts of this paper. The author thanks the NSF for support via DMS 1148490 Research and Training Grant and DMS 1103525.
## 2. Preliminaries
This section collects definitions and facts from the theories of formal algebraic geometry and noncommutative manifolds which we will use.
- 2.1. Definitions and Notation. Throughout X is a smooth variety of dimension n over C .
Definition 2.1. We define the local objects of study in this paper.
1 Let ̂ A n = C 〈〈 x , . . . , x 1 n 〉〉 denote the completed free associative algebra on n generators. 2 Let ̂ O n = C /llbracket x , . . . , x 1 n /rrbracket denote the completed free commutative algebra on n generators. 3 Let H n (resp. H + n ) denote the group of algebra automorphisms (respectively augmented C -algebra automorphisms) of ̂ A n .
- 4 Let h n (resp. h + n ) denote the Lie algebra of H n (resp. H + n ).
- ̂ 6 Let g n (resp. g + n ) denote the Lie algebra of G n (resp. G + n ). Concretely, this is the Lie algebra of vector fields on the formal disk (resp. those vector fields vanishing at the origin).
- 5 Let G n (resp. G + n ) denote the group of algebra automorphisms (respectively augmented C -algebra automorphisms) of O n .
Definition 2.2. The de Rham stack X dR associated to X is given on affine schemes S by X dR ( S ) = X S ( red ), where S red is the reduced subscheme of S .
Definition 2.3. Let F be a sheaf on X with flat connection ∇ . We denote sheaf of flat sections ( F ) ∇ .
Definition 2.4. Let U ⊂ X be open. Define NC -Th U to be the category whose objects are NC-manifolds A with abelianization O U , and morphisms given by morphisms of algebras identical on O U .
↦
Proposition 2.5 ([Kap98] 4.3.1) . The assignment U → NC -Th U is a stack of groupoids locally trivial in the Zariski topology.
Definition 2.6 ([PT14], 2.3.22) . We define the category NC -Conn X of twisted NCconnections on X as follows. Objects are tuples ( T , J , ϕ, D ) where T is a sheaf of O X -algebras, J ⊂ T is a sheaf of two-sided ideals Zariski locally isomorphic to the pair ( ̂ T O X Ω 1 X , T ̂ ≥ 1 O X Ω 1 X ) , ϕ is an isomorphism of graded algebras
$$\varphi = ( \oplus \varphi _ { k } ) \colon \bigoplus _ { k \geq 0 } \mathcal { J } ^ { k } / \mathcal { J } ^ { k + 1 } \stackrel { \sim } { \longrightarrow } T _ { \mathcal { O } _ { X } } \Omega _ { X } ^ { 1 },$$
and D is a differential of degree 1 on Ω · X ⊗ O X T such that D restricts to the de Rham differential on Ω · X and the diagram
/d31
/d47
/d47
/d47
/d47
/d47
/d47
/d47
/d47
/d47
/d47
/d43
/d43
/d43
/d43
commutes.
A morphism between ( T , J , ϕ, D ) and ( T ′ , J ′ , ϕ , D ′ ′ ) is a morphism ϕ : T →T ′ of O X -algebras sending J to J ′ which induces the identity on T O X Ω 1 X under ϕ and ϕ ′ , and such that
$$i d \otimes f \, \colon \Omega _ { X } ^ { \bullet } \otimes _ { \mathcal { O } _ { X } } \mathcal { T } \rightarrow \Omega _ { X } ^ { \bullet } \otimes _ { \mathcal { O } _ { X } } \mathcal { T } ^ { \prime }$$
is compatible with D and D ′ .
One of the main results of [PT14], theorem 2.3.23 in loc. cit., is the following
Theorem 2.7. Let ( T , J , ϕ, D ) be a twisted NC-connection on X . Then ker D ⊂ T = (Ω · X ⊗ O X T ) 0 is an NC-thickening of X , and this extends to an equivalence of categories
$$N C - C o n n _ { X } \stackrel { \sim } { \to } N C - T h _ { X }.$$
/d127
/d53
/d53
- 2.2. Recollections on Formal Geometry. Our (greatly abridged) treatment has been heavily influenced by [BK04] and [BZF04], but most closely follows [Gai11]; details can also be found in [JO14].
Every smooth variety X of dimension n has a canonical G + n -torsor whose fiber at x ∈ X is the set of isomorphisms between the formal neighborhood of x and Spec ̂ O n . This bundle has the structure of a scheme of infinite type over X (see [BZF04] for the scheme structure), and we denote it by M ∼ . The bundle M is locally trivial over X in the Zariski topology.
Notice that M /G + n = X . However, sections of M admit an action of the larger group G n ; this is known classically as the Gelfand-Kazhdan structure on M , or the Harish-Chandra connection in the language of [JO14]; see section 3.4 for further discussion. Section 3 of [Gai11] contains the following results.
Proposition 2.8. We have an isomorphism M /G n → X dR . Moreover, the maps X → BG + n and X dR → BG n classifying M over X (resp. over X dR ) fit into a Cartesian diagram
/d47
/d47
/d15
/d15
/d15
/d15
/d47
/d47
This diagram allows us to construct many natural sheaves on X (resp. X dR ) from modules for G + n (resp. G n ). For example, pulling back the G n -module ̂ O n gives the sheaf JO of jets of functions; other jet sheaves arise similarly. Additionally, an algebra object A of QC( BG + n ) (resp. QC( BG n )) pulls back to an algebra object p A ∗ of QC( X ) (resp. QC( X dR )).
## 3. Noncommutative Formal Geometry
We develop analogues of fundamental results about the bundle M and related structures recalled in section 2.2 with ̂ O n = k x , ..., x /llbracket 1 n /rrbracket replaced by ̂ A n = k 〈〈 x , ..., x 1 n 〉〉 . This leads us to the notion of a bundle of NC-coordinate systems.
- 3.1. Automorphisms of the Noncommutative Disk. The surprisingly strong parallel between commutative and noncommutative formal geometry arises from the following local statements.
Proposition 3.1. We have a diagram
/d47
/d47
/d15
/d15
/d15
/d15
/d15
/d15
/d47
/d47
/d47
/d47
inducing an isomorphism of the quotients.
Proof. Algebra automorphisms automatically preserve the commutator filtration, and the identification ̂ A /M n 2 ∼ O = ̂ n gives the vertical arrows. To see that the third vertical arrow is an isomorphism, consider the action of H n on ̂ O n . The action is transitive, and the stabilizer of the origin is precisely H + n . /square
The following is standard.
/d47
/d47
Lemma 3.2. Write ̂ A n = T k ( V ) for a vector space V . Then
3.2. The Functor of Noncommutative Jets. Consider the sheaf O X /squaremultiply A on X × X . By proposition 2.1.5, the pre-image of the complement of the ideal of the diagonal I ∆ ⊂ O X /squaremultiply O X is an Ore set in O X /squaremultiply A . We can then restrict to open neighborhoods of the diagonal, and take a limit.
$$\widehat { A } _ { n } = \widehat { T } _ { k } ( V ) \text{ for a vector space } V. \text{ Then} \\ \mathfrak { h } _ { n } = \text{Der} _ { k } ( \widehat { A } _ { n } ) \cong \widehat { T } _ { k } ( V ) \otimes _ { k } V ^ { * }, \text{ and} \\ \mathfrak { h } _ { n } ^ { + } = \text{Der} _ { k } ^ { + } ( \widehat { A } _ { n } ) \cong \widehat { T } _ { k } ^ { \geq 1 } ( V ) \otimes _ { k } V ^ { * }. \\ \text{if Noncommutative Jets. } \text{ Consider the sheaf } \mathcal { C } \\ \text{$\text{$pre-image of the complement of the ideal of the di}$}.$$
Definition 3.3. Let ( X, A ) be an NC-manifold. Then O X /squaremultiply A is a sheaf on X × X (see above). Let p : ̂ X × X → X be the projection from the formal neighborhood of the diagonal to X , and i : ̂ X × X → X × X the inclusion. We define J NC A = p i ∗ ∗ ( O X /squaremultiply A ).
We have the following key fact due to Kapranov [Kap98], proposition 4.4.1.
Proposition 3.4. The sheaf J NC A is a sheaf of O X -algebras locally isomorphic to and its abelianization is isomorphic to JO X .
$$\mathcal { O } _ { X } \otimes _ { \underline { \mathbb { C } } } \underline { \hat { A } _ { n } }, \\ \mathcal { J O } _ { X }.$$
Proof. The first statement follows from the local triviality of the formal neighborhood of X ⊂ X × X as a bundle on X . The second follows from the fact that A abelianizes to O X . /square
The following is clear by construction.
Proposition 3.5. The fibers of J NC A are canonically identified with the completed stalks of A .
The following definition appears in section 4.4 of [Kap98].
Definition 3.6. Define the category NC -Jet X with objects given by pairs ( J , ψ ) where J is an O X -algebra locally isomorphic to O X ⊗ C ̂ A n and ψ : J → JO X is a surjection with kernel given locally by the commutator ideal. Morphisms ( J , ψ ) → J ( ′ , ψ ′ ) are given by isomorphisms f : J → J ′ such that ψ ′ ◦ f = ψ .
The following equivalence of categories is proposition 4.4.2 in [Kap98]. In section 3.4, we will construct the quasi-inverse functor.
Proposition 3.7. The functor J NC : NC -Th X → NC -Jet X is an equivalence of categories.
- 3.3. The Frame Bundle Functor. The functor of noncommutative jets yeilds a sheaf of O X -algebras locally isomorphic to O X ⊗ C ̂ A n . From this, give the following
Definition 3.8. Let A be an NC-thickening of X . Define frame ( J NC A ) = N to be the H + n -torsor of augmented O X -algebra isomorphisms from J NC A to O X ⊗ C ̂ A n . We write Coord ( A ) = frame ( J NC A ) = N for the bundle of NC-coordinate systems associated to A .
Remark 3.9 . With the above definition, it is clear that the fibers of N = Coord ( A ) are isomorphic to the H + n -torsor of augmented C -algebra isomorphisms from A x to ̂ A n .
Proposition 3.10. We have equivalences of sheaves and
$$\ v e \ e q u v a l e n c e s \ o f \ s h e a v e s \\ f r a m e \left ( \text{Assoc} _ { \mathcal { N } } ( \widehat { A } _ { n } ) \right ) & \stackrel { \sim } { \to } \mathcal { N } \\ \\ \bullet & \widehat { \lambda } \ \sim \ \mathcal { N } C \ \bullet$$
$$\text{Assoc} _ { f r a m e } ( \mathcal { J } ^ { N C } \mathcal { A } ) ( \widehat { A } _ { n } ) \stackrel { \sim } { \to } \mathcal { J } ^ { N C } \mathcal { A }. \\ \text{gument showing the equivalence between $GL$} \\ \text{$c_{\sim} \sim \dots \, \max_{1} \leq \dots \, \max_{1}$}$$
Proof. The standard argument showing the equivalence between GL m -bundles and rank m -vector bundles applies after obvious modifications. /square
Definition 3.11. Define the category NC -Coord X with objects given by principal H + n -bundles N locally trivial in the Zariski topology, equipped with isomorphisms N × H + n G + n ∼ → M and morphisms given by compatible isomorphisms of bundles over X .
We may now strengthen the statement of proposition 3.10
Proposition 3.12. The functor frame : NC -Jet X → NC -Coord X is an equivalence of categories. The quasi-inverse is given by N ↦→ Assoc N ( ̂ A n ) . In summary, we presently have equivalences
/d47
/d47
/d36
/d36
/d43
/d43
/d107
/d107
The following subsection will give us the tools required to construct a quasi-inverse to J NC , and thus also to Coord .
## 3.4. The Noncommutative Gelfand-Kazhdan Structure. In usual formal geometry, we have a diagram
/d47
/d47
/d47
/d47
/d47
/d47
/d47
/d47
/d15
/d15
/d116
/d116
/d47
/d47
where
The map θ is called the Harish-Chandra connection , or the Gelfand-Kazhdan structure . Rather than being a connection on M as a G + n -bundle, it is a connection on M as a ( g n , G + n )-module; that is, it carries an action of g n which is integrable along the subalgebra g + n (see [JO14] for details). Extending the action from G + n to g n gives a flat connection on jet bundles, whose flat sections undo the functor of jets; equivalently, it induces the diagram in proposition 2.8.
$$\mathcal { E } _ { \mathcal { M } } = \left ( \mathcal { T } _ { \mathcal { M } } \right ) ^ { G _ { n } ^ { + } } \cong \text{Der} _ { \mathcal { O } _ { X } - a l g } \left ( \mathcal { J } \mathcal { O } _ { X }, \mathcal { O } _ { X } \otimes _ { \underline { \mathbb { C } } } \widehat { \mathcal { O } } _ { n } \right ). \\ \text{is called the Harish-Chandra connection, or the $Gelfand-Kazhd$} \\ \text{heino a connection on $M$ as $G^{+}$-hundle it is a connection on $M$}$$
Definition 3.13. Given A ∈ NC -Th X and N = Coord ( A ), define
$$\text{iven } \mathcal { A } \in N C - 1 h _ { X } \text{ and } \mathcal { N } = C o o r d ( \mathcal { A } ), \text{ define} \\ \mathcal { E } _ { \mathcal { N } } = \text{Der} _ { \mathcal { O } _ { X } - a l g } \left ( \mathcal { J } ^ { N C } \mathcal { A }, \mathcal { O } _ { X } \otimes _ { \underline { \complement } } \underline { \widehat { A } _ { n } } \right ).$$
We wish to give an analogous diagram:
/d47
/d47
/d15
/d15
/d117
/d117
/d47
/d47
In the commutative case, θ is an isomorphism, and is dual to the action map of vector fields on the formal disk on functions on the formal disk.
$$( 2 ) & & 0 \xrightarrow { \sim } \underset { \overset { \sim } { \sim } } { \longrightarrow } \underset { \overset { \sim } { \longrightarrow } } { \overset { \sim } { \longrightarrow } } \underset { \overset { \theta ^ { N C } } { \longrightarrow } } { \overset { \sim } { \longrightarrow } } \underset { \overset { \theta ^ { N C } } { \longrightarrow } } { \overset { \sim } { \longrightarrow } } \underset { \overset { \theta ^ { N C } } { \longrightarrow } } { \overset { \sim } { \longrightarrow } } \text{$\mathcal{ } } 0 \xrightarrow { \sim } \underset { \overset { \sim } { \longrightarrow } } { \longrightarrow } \text{$\mathcal{ } } \text{Assoc} _ { \mathcal { N } } ( \mathfrak { h } _ { n } ) & \xrightarrow { \sim } \underset { \overset { \sim } { \longrightarrow } } { \longrightarrow } \text{Assoc} _ { \mathcal { N } } ( \mathfrak { h } _ { n } ) \\ & & \text{In the commutative case, $\theta$ is an isomorphism, and is dual to the action map of vector file} \\ & & \text{on the formal disk on functions on the formal disk.}$$
Proposition 3.14. Let N be an NC-coordinate system. On a trivialization of J NC A , we have descriptions of the sheaves:
$$y \ u n c { \ s n c { a v c { s } } }. \\ a d _ { \mathcal { N } } = \text{Assoc} _ { \mathcal { N } } ( \mathfrak { h } _ { n } ^ { + } ) & \cong \widehat { T } _ { \mathcal { O } } ^ { \geq 1 } ( \Omega _ { X } ^ { 1 } ) \otimes _ { \mathcal { O } _ { X } } \mathcal { T } _ { X } \\ \text{Assoc} _ { \mathcal { N } } ( \mathfrak { h } _ { n } ) & \cong \widehat { T } _ { \mathcal { O } } ( \Omega _ { X } ^ { 1 } ) \otimes _ { \mathcal { O } _ { X } } \mathcal { T } _ { X } \\ \text{Assoc} _ { \mathcal { N } } ( \mathfrak { h } _ { n } / \mathfrak { h } _ { n } ^ { + } ) & \cong \mathcal { T } _ { X }. \\ \text{halization of $3$ $1$ $1$ $1$ for future from noncoincidence $3$ $4$ an}.$$
Proof. This is a globalization of 3.2. It follows from propositions 3.4 and 3.5 that the fiber of J NC A at x ∈ X , in turn isomorphic to the stalk A x , is canonically identified with ̂ T V k , where V is the cotangent space to X at x . Equivalently, the generators of A x abelianize to the generators of ̂ O X,x . /square
Remark 3.15 . This is analogous to the decomposition of ad M , where the tensor powers above are replaced with symmetric powers of the tangent bundle; see [Kap98], section 4.2.
Proposition 3.16. The sheaf of Lie algebras Assoc N ( h n ) acts naturally on J NC A , hence there is a natural map Assoc N ( h n ) →E N . This map is an isomorphism.
Proof. The fibers of E N are given by Der C -alg ( A x , A ̂ n ) . Thus the natural action map induces an isomorphism on stalks. /square
Remark 3.17 . The action map defining the splitting is a map of Lie algebras; therefore the connection is flat.
Definition 3.18. We call the ( h n , H + n )-valued connection θ NC = α -1 the noncommutative Gelfand-Kazhdan structure on N .
Remark 3.19 . The map θ induces a connection on any associated bundle Assoc N ( M ), M ∈ H + n -mod. This connection can be written
$$\nabla _ { \xi } ( s ) = d s ( \xi ) + \theta ( \tilde { \xi } ),$$
where ˜ is any lift of ξ ξ ∈ T X to E N .
Remark 3.20 . The above discussion extends the action of H + n on N ∈ NC -Coord X to an action of h n = Lie H n . As in the commutative setting (see 2.2), this can also be phrased in terms of extending the action to H n . We will often use the language of H n rather than h n in order to phrase results in terms of classifying spaces.
Corollary 3.21. Given an NC-coordinate system N on X , we have N /H n ∼ = X dR .
Proof. This follows from proposition 3.14.
/square
Remark 3.22 . In the language of [GR11], an NC-coordinate system N is naturally a crystal on X , i.e., it is equipped with descent data from X to X dR .
/d47
/d47
/d47
/d47
/d47
/d47
## 4. NC-Thickenings, NC-Coordinate Systems, and NC-Connections
This section establishes equivalences of categories between NC-thickenings of a variety X , bundles of NC-coordinate systems over X , and NC-connections on X .
- 4.1. Relation with NC-Smooth Thickenings. We will show an equivalence of categories between noncommutative thickenings of X and noncommutative systems of coordinates on X .
Theorem 4.1. The functor Coord : NC -Th X → NC -Coord X is an equivalence of categories. The inverse is given by taking flat sections of Assoc N ( ̂ A n ) under the NC-Gelfand Kazhdan structure on N .
Proof. The fact that Coord is an equivalence follows from propositions 3.7 and 3.12. The second claim follows from proposition 4.2 below. /square
Proposition 4.2. There is a natural isomorphism
$$\ u u \, \iota \sigma \iota \iota \iota \Pi \iota \sigma \iota \iota \rho \iota \sigma \iota \rho \iota \sigma \iota \rho \\ \mathcal { A } & \stackrel { \sim } { \to } \left ( \mathcal { J } ^ { N C } \mathcal { A } \right ) ^ { \nabla }. \\ \left ( \mathcal { T } ^ { N C } \mathcal { A } \right ) ^ { \nabla } \text{ as follows.}$$
Proof. We define a map ϕ : A → ( J NC A ) ∇ as follows. A section of A determines one of A /squaremultiply A via the diagonal map, hence of O X /squaremultiply A via projection. For any U ⊂ X ,
$$\left ( \mathcal { O } _ { X } \not \in \mathcal { A } \right ) ( U \times U ) \rightarrow \mathcal { J } ^ { N C } \mathcal { A } ( U ).$$
We claim that the image of this composition lands in the subsheaf of flat sections of J NC A . The formula in remark 3.19 shows that flatness of a section is equivalent to the derivative along the base being equal to the derivative long the fiber. The above map factors through the diagonal map A → A /squaremultiply A , hence this is satisfied.
The proposition now follows from
Lemma 4.3. The stalks of ( J NC A ) ∇ are naturally isomorphic to A x .
Proof.
$$P r o o f. \text{ This can be deduced from } [ J O 1 4 ], \text{ proposition } 4. 2 3.$$
We may now add another equivalence of categories to diagram (1):
/d43
/d43
/d36
/d36
/d43
/d43
/d106
/d106
/d107
/d107
- 4.2. Relation with NC-Connections. The aim of this section is to interpret the connection ∇ in terms of the NC-connections of [PT14]. An equivalence of categories follows formally from the preceding section and the main theorem of [PT14], but we wish to illustrate directly how NC-connections relate to NC-coordinate systems.
The noncommutative Gelfand-Kazhdan structure on N induces a flat connection on J NC A (see remarks 3.17 and 3.19). This flat connection can be equivalently described by
$$\omega \in \Omega _ { X } ^ { 1 } \otimes _ { \mathcal { O } _ { X } } \mathcal { E } n d ( \mathcal { J } ^ { N C } \mathcal { A } ).$$
/square
/square
Note that E nd ( J NC A ) is locally identified with
$$( \mathcal { J } \cdots \mathcal { A } ) & \text{ is locally apparent} \, \text{win} \\ \prod _ { k } \mathcal { H } o m ( \mathcal { T } _ { X } ^ { \otimes k }, \mathcal { T } _ { X } ) & \cong \prod _ { k } \mathcal { H } o m \left ( \Omega _ { X } ^ { 1 }, \left ( \Omega _ { X } ^ { 1 } \right ) ^ { \otimes k } \right ). \\ \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon$$
Thus locally ω yields an untwisted NC-connection (see definition 1.5) as follows.
A flat connection ω ∈ Ω 1 ⊗E nd ( F ) on a sheaf F defines a differential D on Ω · X ⊗ O X F extending the de Rham differential. For F = ̂ T O X Ω , the sheaf of endomorphisms decomposes 1 X into a product as above, and so does the differential D . Restricting D to gives an operator of the form
$$\Omega _ { X } ^ { 1 } \subset \widehat { T } _ { \mathcal { O } _ { X } } \Omega _ { X } ^ { 1 } \subset \Omega _ { X } ^ { \bullet } \otimes _ { \mathcal { O } _ { X } } \widehat { T } _ { \mathcal { O } _ { X } } \Omega _ { X } ^ { 1 } \\ \therefore \text{form}$$
$$\Pi \, \text{in form} \\ \sum _ { k } \nabla _ { k }, \quad \nabla _ { k } \colon \Omega _ { X } ^ { 1 } \to \Omega _ { X } ^ { 1 } \otimes _ { \mathcal { O } _ { X } } T ^ { k } \left ( \Omega _ { X } ^ { 1 } \right ), \\ \nu _ { k } \quad \nu \sim 1 \quad \nu _ { k } \quad \dots \cdots \cdots \cdots \cdots \cdots \cdots.$$
where the first tensor factor of Ω 1 on the right-hand side above is in graded degree 1.
The above discussion works locally in order to identify J NC A with ̂ T O X Ω . Without this 1 X the flat connection on J NC A still induces a degree 1 derivation D on Ω · X ⊗ O X J NC A , which satisfies D 2 = 0 due to the flatness of the connection. We summarize the discussion in the following
Theorem 4.4. The preceding construction of D endows Assoc N ( ̂ A n ) = J NC A with the structure of a twisted NC-connection. This induces an equivalence of categories
$$\mathcal { G K } \colon N C - C o o r d _ { X } \stackrel { \sim } { \to } N C - C o n n _ { X }$$
and the following diagram of equivalences commutes.
/d43
/d43
/d91
/d107
/d107
/d107
/d107
/d91
/d113
/d113
Proof. The sheaf of algebras J NC A has a natural ideal sheaf locally identified with ̂ T ≥ 1 O X Ω . 1 X The preceding discussion gives a construction of D . Recalling definition 2.6, it remains to construct an isomorphism of graded algebras
$$\text{orpins} & \text{ or gracea aigepras} \\ & \varphi \colon \bigoplus _ { k \geq 0 } \left ( \mathcal { J } ^ { N C } \mathcal { A } \right ) ^ { k } / \left ( \mathcal { J } ^ { N C } \mathcal { A } \right ) ^ { k + 1 } \simeq T _ { \mathcal { O } _ { X } } \Omega _ { X } ^ { 1 }. \\ & \wedge \quad \wedge$$
The map of H n -modules ̂ A n → O ̂ n induces a map J NC A → JO X , as required in the definition of an object of NC -Jet X . The filtered components of this map give the desired maps ϕ i . /square
## 5. NC-Thickenings as Reductions of Structure Group
This section gives a new criterion for the existence of NC-thickenings of a given variety X , and studies natural obstructions.
/d43
/d43
/d37
/d37
- 5.1. Existence Results. We establish a new criterion of existence of NC-thickenings of X .
Theorem 5.1. Every N ∈ NC -Coord X arises as a reduction of structure group of M from G + n to H + n , and this reduction respects the natural action of H n on N and G n on M . Equivalently, NC-coordinate systems correspond to commutative diagrams
/d123
/d123
/d35
/d35
/d47
/d47
where q classifies M .
Proof. The statement that N is a reduction of M is immediate from the definition of NCcoordinate system. This reduction intertwines the H n and G n actions by the commutativity of the diagram
/d47
/d47
/d15
/d15
/d15
/d15
/d47
/d47
where the vertical arrows are the NC and classical Gelfand-Kazhdan structures. /square
Corollary 5.2. A smooth variety X admits an NC-thickening if and only if Assoc M ( G /H + n + n ) has global sections.
Remark 5.3 . Note that the quotient G /H + n + n is taken stack-theoretically; thus the associated bundle above gives a canonically defined sheaf of stacks on all smooth varieties capturing the existence of NC-thickenings.
Remark 5.4 . This corollary gives a simple geometric reason for the existence of NC-thickenings of affine schemes: Assoc M ( G /H + n + n ) will always have global sections.
- 5.2. Obstruction Classes and the NC-Atiyah Class. The notion of noncommutative Atiyah class α A ∈ H 1 ( X, Hom( T ⊗ 2 X , T X )) associated to an NC-thickening A is due to Kapranov [Kap98].
Definition 5.5. The sheaf of noncommutative (or NC) 1-jets J NC 1 A is the quotient of J NC A by the pre-image of higher-order jets in JO .
Definition 5.6. The NC-Atiyah class α A of an NC-thickening A is the class of
$$0 \to \mathcal { T } _ { X } \to \mathcal { J } _ { 1 } ^ { N C } \mathcal { A } \to \mathcal { T } _ { X } ^ { \otimes 2 } \to 0$$
in Ext 1 X ( T ⊗ 2 X , T X ). Equivalently, α A is the class obtained by locally identifying J NC 1 A with T ≤ 2 O X Ω . 1 X
An equivalent way of specifying of the map θ of section 3.4 is to give a 1-form ω ∈ Ω 1 N /X ⊗ p ∗ Assoc N ( h n ), where p : N → X . Using the decomposition by degree of N , we can form the affine bundle N (2) whose fibers are isomorphisms A x / m 3 → ̂ A / n m 3 which are the identity in degree 1 (cf. section 4.2 of [Kap99]); the form ω then yields ω 2 ∈ Ω 1 N (2) /X ⊗ p ∗ H om ( T ⊗ 2 X , T X ).
By [Kap98], lemma 4.1.1, the form ω 2 gives a cohomology class
$$[ \omega _ { 2 } ] \in H ^ { 1 } ( X, \mathcal { H } o m ( \mathcal { T } _ { X } ^ { \otimes 2 }, \mathcal { T } _ { X } ) ) \cong \text{Ext} ^ { 1 } ( \mathcal { T } _ { X } ^ { \otimes 2 }, \mathcal { T } _ { X } ).$$
Comparing definitions, we see that J NC 1 A is globally trivial when pulled back to N (2) , and ω 2 classifies the action of invariant vector fields on N (2) on this trivial sheaf O N (2) ⊗ ̂ A / n m 3 . Since this action gives descent data for J NC 1 A , we must have the following.
Proposition 5.7. The cohomology classes [ ω 2 ] = α A are equal.
Remark 5.8 . This is fully analogous to the commutative situation: in [Kap99], example 4.2.2, it is observed that the usual Atiyah class of T X arises as the 'degree 2' part of the commutative Gelfand-Kazhdan structure.
Remark 5.9 . Theorem 4.5.3 in [Kap98] says roughly that 1-smooth thickenings extend to 2-smooth thickenings precisely when the second A ∞ -compatibility condition is satisfied by the Atiyah class. See also the discussion of A ∞ -spaces in section 5.4 of [PT14], analogous to the L ∞ -spaces of [Cos11] and [Kap99]. The broader relationship between coordinate systems and Koszul duality suggested by this pattern will be the topic of future work.
## 6. Degeneration to the Associated Graded and Lower Central Series
This section recalls some results of the paper [JO14] and gives a categorification of the main theorem of that paper.
- 6.1. The Lower Central Series Filtration. We restate a stronger version of lemma 4.25 of [JO14], which follows from the original argument:
Proposition 6.1. The graded algebra gr M ( ̂ A n ) is an algebra object in QC( BG n ) . For A the NC-thickening of a smooth affine variety X = Spec A ab , theorem 4.26 of [JO14] shows
Moreover, the fact that this isomorphism is canonical gives the sheaf-theoretic analogue for non-affine X and NC-thickenings A .
$$& \text{shows} \\ & \text{Theorem $6.2$} \text{ } \text{ } \text{there is an isomorphism } \Gamma \left ( X, \left ( \text{Assoc} _ { \mathcal { M } } ( N _ { k } ( \widehat { A } _ { n } ) ) \right ) ^ { \nabla } \right ) \cong N _ { k } ( A ). \\ & \text{ Moreover, the fact that this isomorphism is canonical gives the sheaf-theoretic analogue} \\ & \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{$$
## 6.2. Degeneration.
Definition 6.3. Let K n = Aut( gr M ( ̂ A n )). Define the K n -bundle ˜ M as the bundle of graded O X -algebra isomorphisms from gr M ( A ) to gr M ( ̂ A n ) . Combining theorems 6.2 and 5.1, we have the following
- Corollary 6.4. Assoc N ( K n ) = ∼ M ˜ for any N ∈ NC -Coord X . 6.3. Categorical Geometry of Lower Central Series Algebras. We will show that the category of modules over the associated graded algebra gr M ( A ) is associated to the bundle of coordinate systems from gr M ( ̂ A n )-mod. Throughout this section we consider differential graded categories of modules.
It is a general fact that if a group G acts on an algebra A , then A -mod is a G -category [FG06]. The following result appears in section 3.2.5 of [Gai11].
Lemma 6.5. Let A ∈ Alg (QC( BG )) , and P → X a G -bundle. Denote A = Assoc P ( A ) . Then
$$\Gamma ( X, \text{Assoc} _ { \mathcal { P } } ( A \text{-mod} ) ) \cong \mathcal { A } \text{-mod} ( Q C ( X ) ).$$
Proof. It is easy to see that A -mod(QC( X )) equalizes the appropriate cotensor product diagram:
/d47
$$\text{gram:} \\ E q \underset { \mathbb { S } \sim } { \longrightarrow } Q C ( \mathcal { P } ) \times ( A \text{-mod} ) \underset { \mathbb { S } \sim } { \longrightarrow } \overset { a c t ^ { * } } { \text{c} } ( \mathcal { P } ) \otimes Q C ( G ) \otimes ( A \text{-mod} )$$
/d47
/d47
/d47
/d47
/d47
/d103
/d103
/d79
/d79
/square
We will need a slight extension of the notion of associated bundle, owing to the natural action of G n on the G + n -bundle M . Denote by M× G n ( N ) the quotient of M× N by the diagonal G n -action.
Theorem 6.6. There is an equivalence of categories
$$\Gamma ( X, \mathcal { M } \times _ { G _ { n } } ( g r _ { M } ( \widehat { A } _ { n } ) \text{-mod} ) ) \cong g r _ { M } ( A ) \text{-mod} \,. \\ \text{cell that } \mathcal { M } \times _ { G _ { n } ^ { + } } g r _ { M } ( \widehat { A } _ { n } ) \cong \mathcal { J } g r _ { M } ( \widehat { A } _ { n } ) \text{ by theorem } 6. 2, \text{ and the} \,.$$
Remark 6.7 . This can be viewed as a categorical analogue of theorem 4.26 of [JO14]. Intuitively, we think of M× G n ( gr M ( ̂ A n ) -mod) as the flat sections of the sheaf of categories Assoc M ( gr M ( ̂ A n ) -mod).
Proof. First recall that M× G + n gr M ( ̂ A n ) = ∼ J gr M ( ̂ A n ) by theorem 6.2, and therefore M× G n gr M ( ̂ A n ) = ∼ gr M ( ̂ A n ). The theorem then follows from lemma 6.5. /square
## 7. Relation to D -Modules
In section 3.2 of [PT14], it is observed that there is a functor from the category of D -modules on X to the category of modules over any NC-thickening A of X . We wish to interpret this in terms of the appearance of the de Rham stack above.
- 7.1. Construction of Functor. Let ( N ∇ , ) be an NC-coordinate system on X . Then recall 3.4 that N /H n ∼ = X dR . Results of section 5 imply that we have a commutative diagram
/d123
/d123
/d35
/d35
/d47
/d47
Recall [Gai13] that, for a map f : X → Y , one has the functor
$$\text{ind} _ { f } \, \colon Q C ( Y ) \text{-mod} & \to Q C ( X ) \text{-mod} \\ \mathcal { L } & \mapsto Q C ( X ) \otimes _ { Q C ( Y ) } \mathcal { C }$$
so in particular ind f (QC( Y )) = QC( ∼ X ). Thus we can write
$$\mathcal { D } ( X ) = Q C ( X _ { d R } ) \cong \text{ind} _ { q } ( Q C ( B G _ { n } ) ) \cong \text{ind} _ { p } \left ( \text{ind} _ { \pi } ( Q C ( B G _ { n } ) ) \right ).$$
Moreover, by lemma 6.5,
Note that Γ( X dR , QC( X dR )) = QC( ∼ X dR ). We now have the following
$$& \text{loreover, by lemma 6.5}, \\ & \mathcal { J } ^ { N C } \mathcal { A } \text{-mod} ( Q C ( X _ { d R } ) ) \cong \Gamma \left ( X _ { d R }, \text{ind} _ { p } \left ( \widehat { A } _ { n } \text{-mod} ( Q C ( B H _ { n } ) ) \right ) \right ). \\ & \text{ote that } \Gamma ( X _ { d R }, Q C ( X _ { d R } ) ) \cong Q C ( X _ { d R } ). \text{ We now have the following}$$
## Proposition 7.1. There are functors
such that applying Γ( X dR , ind p ( -)) induces a functor
$$& \text{ind} _ { \pi } \text{ QC} ( B G _ { n } ) & \stackrel { \prime } { \rightarrow } \text{ QC} ( B H _ { n } ) \rightarrow \widehat { A } _ { n } \text{-mod} ( \text{QC} ( B H _ { n } ) ) \\ \text{applying} \text{in} g & \Gamma ( X _ { d R }, \text{ind} _ { p } ( - ) ) & \text{induces a functor} \\ & \text{$/\sim}$$
$$\text{such that applying} \, \Gamma ( X _ { d R }, \, \text{ind} _ { p } ( - ) ) \, \text{induces a functor} \\ \text{QC} ( X _ { d R } ) \cong \Gamma \left ( X _ { d R }, \, \text{ind} _ { p } \left ( \text{ind} _ { \pi } \, \text{QC} ( B G _ { n } ) \right ) \right ) & \to \text{ind} _ { p } \left ( \widehat { A } _ { n } \text{-mod} ( \text{QC} ( B H _ { n } ) ) \right ) \\ & \cong \mathcal { J } ^ { N C } \mathcal { A } \text{-mod} ( \text{QC} ( X _ { d R } ) ). \\ \text{--} \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdots$$
Moreover, the functor of flat sections ( -) ∇ : QC( X dR ) → Shv C ( X ) induces a functor J NC A -mod(QC( X dR )) →A -mod( Shv C ( X )) , where Shv C ( X ) denotes the category of sheaves of C -vector spaces on X .
Proof. The H n -linear map A n → C induces the desired functor
̂ It remains to see that A acts on flat sections of a J NC A -module. But this follows from proposition 4.2 and the Leibniz rule. /square
$$\text{The } H _ { n } \text{-linear map } \widehat { A } _ { n } \rightarrow \mathbb { C } \text{ indices the desired functor} \\ \mathbb { C } \text{-mod} \left ( \mathbb { Q } \mathbb { C } ( B H _ { n } ) \right ) & \cong \mathbb { Q } \mathbb { C } ( B H _ { n } ) \rightarrow \widehat { A } _ { n } \text{-mod} ( \mathbb { Q } \mathbb { C } ( B H _ { n } ) ). \\. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad..$$
Corollary 7.2. For every NC-thickening A of X , we obtain a functor
$$\mathcal { D } ( X ) \to \mathcal { A } \text{-mod} \,.$$
- 7.2. Comparison to the Functor of [PT14] . The functor D -mod →A -mod in [PT14] is as follows. Given M ∈ D -mod, we can write M as a Ω · X -comodule via M ⊗ O X Ω . · X Also, J NC A has a flat connection so we get a de Rham-comodule J NC A⊗ O X Ω . We then form · X
This is a dg-module over J NC A⊗ Ω . Taking the kernel of the differential in degree 0, we · X get that H 0 ( K M ( )) is a module over H 0 ( J NC A⊗ Ω ) = · X ∼ A , as desired.
$$\text{connection so we get a de Kham-comodule } \mathcal { J } \stackrel { \cdot \, \cdot \, \cdot } { \lambda } \otimes _ { \mathcal { O } _ { X } } \Omega \\ K ( M ) = ( M \otimes \Omega _ { X } ^ { \bullet } ) \otimes _ { \Omega _ { X } ^ { \bullet } } \left ( \mathcal { J } ^ { N C } \mathcal { A } \otimes \Omega _ { X } ^ { \bullet } \right ). \\ \text{le over } \mathcal { J } ^ { N C } \mathcal { A } \otimes \Omega _ { X } ^ { \bullet }. \text{ Taking the kernel of the different} i$$
In proposition 7.1, we have the augmentation ̂ A n → C . This induces a functor on H n -equivariant modules
$$\text{QC} ( B H _ { n } ) \rightarrow M o d _ { \widehat { A } _ { n } } ( \text{QC} ( B H _ { n } ) ).$$
which, associating categories to N , gives a functor
$$\mathcal { D } ( X ) \rightarrow M o d _ { \mathcal { J } ^ { N C } \mathcal { A } } ( \mathcal { D } ( X ) ).$$
We then claim that A acts naturally on the sheaf of flat sections of any J NC A -module in D ( X ), i.e., we have a composition
$$\mathcal { D } ( X ) \rightarrow M o d _ { \mathcal { J } ^ { N C } \mathcal { A } } ( \mathcal { D } ( X ) ) \rightarrow M o d _ { \mathcal { A } } ( S h v _ { \mathbb { C } } ( X ) ).$$
These two functors manifestly coincide.
## References
[BD04] Alexander Beilinson and Vladimir Guerchonovitch Drinfeld. Chiral algebras , volume 51. American Mathematical Society Providence, RI, 2004.
[BK04] Roman Bezrukavnikov and Dmitry Kaledin. Fedosov quantization in algebraic context. Mosc. Math. J , 4(3):559-592, 2004.
[BZF04] D Ben-Zvi and E Frenkel. Vertex algebras and algebraic curves. Mathematical surveys and monographs , 88, 2004.
[Cos11] Kevin J Costello. A geometric construction of the Witten genus, II. arXiv preprint arXiv:1112.0816 , 2011.
[CQ95] Joachim Cuntz and Daniel Quillen. Algebra extensions and nonsingularity. J. Amer. Math. Soc. , 8(2):251-289, 1995.
[FG06] Edward Frenkel and Dennis Gaitsgory. Local geometric Langlands correspondence and affine KacMoody algebras. In Algebraic geometry and number theory , pages 69-260. Springer, 2006.
[Gai11] Dennis Gaitsgory. Universal Constructions of Crystals: The Categorical Meaning of the Sugawara Construction. 2011.
[Gai13] Dennis Gaitsgory. Sheaves of categories and the notion of 1-affineness. arXiv preprint arXiv:1306.4304 , 2013.
[GK71] IM Gelfand and DA Kazhdan. Some problems of differential geometry and the calculation of cohomologies of lie algebras of vector fields. In Soviet Math. Dokl , volume 12, pages 1367-1370, 1971.
[GR11] Dennis Gaitsgory and Nick Rozenblyum. Notes on geometric langlands: crystals and d-modules. arXiv preprint arXiv:1111.2087 , 2011.
[JO14] David Jordan and Hendrik Orem. An Algebro-geometric Construction of Lower Central Series of Associative Algebras. International Mathematics Research Notices , 2014.
[Kap98] Mikhail Kapranov. Noncommutative geometry based on commutator expansions. Journal f¨r die u reine und angewandte Mathematik (Crelles Journal) , 1998(505):73-118, 1998.
[Kap99] Mikhail Kapranov. Rozansky-Witten invariants via Atiyah classes. Compositio Mathematica , 115(01):71-113, 1999.
[PT14] Alexander Polishchuk and Junwu Tu. DG-resolutions of NC-smooth thickenings and NC-FourierMukai transforms. Mathematische Annalen , 360(1-2):79-156, 2014.
[Sim96] Carlos Simpson. Homotopy over the complex numbers and generalized de rham cohomology. Lecture Notes in Pure and Applied Mathematics , pages 229-264, 1996.
| null |
[
"Hendrik Orem"
] |
2014-08-04T22:14:27+00:00
|
2014-11-04T01:05:50+00:00
|
[
"math.AG",
"math.RA"
] |
Formal Geometry for Noncommutative Manifolds
|
This paper develops the tools of formal algebraic geometry in the setting of
noncommutative manifolds, roughly ringed spaces locally modeled on the free
associative algebra. We define a notion of noncommutative coordinate system,
which is a principal bundle for an appropriate group of local coordinate
changes. These bundles are shown to carry a natural flat connection with
properties analogous to the classical Gelfand-Kazhdan structure.
Every noncommutative manifold has an underlying smooth variety given by
abelianization. A basic question is existence and uniqueness of noncommutative
thickenings of a smooth variety, i.e., finding noncommutative manifolds
abelianizing to a given smooth variety. We obtain new results in this direction
by showing that noncommutative coordinate systems always arise as reductions of
structure group of the commutative bundle of coordinate systems on the
underlying smooth variety; this also explains a relationship between
$\mathcal{D}$-modules on the commutative variety and sheaves of modules for the
noncommutative structure sheaf.
|
1408.0831v3
|
## The problem of dynamic cavitation in nonlinear elasticity
(In: S´ eminaire Laurent Schwartz - EDP et applications (2012-2013) , Exp. 14, 1-17. DOI: 10.5802/slsedp.41.)
Jan Giesselmann ∗
Alexey Miroshnikov †
Athanasios E. Tzavaras ‡
## Abstract
The notion of singular limiting induced from continuum solutions (slic-solutions) is applied to the problem of cavitation in nonlinear elasticity, in order to re-assess an example of non-uniqueness of entropic weak solutions (with polyconvex energy) due to a forming cavity.
## 1 Introduction
The equations describing radial motions of isotropic elastic materials admit a special weak solution describing a cavity that emerges at a point from a homogeneously deformed state. Cavitating solutions are self-similar in nature and were introduced by Ball [3] and by Pericak-Spector and Spector [11, 12]. They turn out to decrease the total mechanical energy and provide a striking example of non-uniqueness of entropy weak solutions for polyconvex energies [11].
There is a class of problems in solid mechanics, such as fracture, cavitation or shear bands, where discontinuous motions emerge from smooth motions via a mechanism of material instability. Any attempt to study such solutions that lie at the limits of continuum modeling needs to reckon with the problem of giving a proper definition for such solutions. Once the material breaks or a shear band forms the motion can no longer be described at the level of continuum modeling and microscopic modeling or higher-order regularizing mechanisms have to be taken into account. Still, as such structures develop there is expected an intermediate time scale where both types of modeling apply. In these lecture notes we present the notion of singular limiting induced from continuum solution (or slic-solution) [8], stating that a discontinuous motion is a slic-solution if its averages are a family of smooth approximate solutions to the problem.
The lecture notes are based on two recent works on cavity formation [10] and [8]. [10] complements [11, 12] and establishes various further properties of weak solutions describing cavitation, in particular indicating that cavity formation is necessarily associated with a unique precursor shock. In [8] the notion of singular limiting induced from continuum solution is introduced and applied to the problem of cavity formation. It is shown that there is an energetic cost for creating the cavity, which is captured by slic-solutions but neglected by the usual entropic weak solutions. Once this surface energy cost is accounted for, the paradox of nonuniqueness is removed, in the sense that
∗ Weierstrass Institute, Berlin, Germany, [email protected]
† Department of Mathematics and Statistics, University of Massachusetts Amherst, USA, [email protected]
‡ Department of Applied Mathematics, University of Crete, Heraklion, Greece and Institute for Applied and Computational Mathematics, FORTH, Heraklion, Greece, [email protected]
the cavitating solution together with the surface energy of the cavity has higher energy than the homogeneous deformation.
The structure of these notes is as follows: In section 2 we present the system of elasticity and the general requirements imposed by mechanical considerations and discuss various notions of relaxed convexity hypotheses associated to elasticity. In section 3 we introduce the equations of radial elasticity applicable to isotropic elastic materials. In section 4 we present the problem of cavitation and outline the construction and properties of weak solutions describing cavity formation from a homogeneously deformed state. In section 5 we introduce an illuminating example for fracture in 1-d and in section 6 the concept of singular limiting induced from continuum solutions which gives an interpretation to this example as a solution of the equations of one-dimensional elasticity. It is shown that there is an additional contribution from the opening crack to the resulting energy of the fracturing solution. In section 7, the notion of slic-solution is extended for the cavitation problem and the effect of this concept to the energy of the cavitating solution is calculated. It turms out that the slic-solution provides a more discriminating concept of solution for strong singularities.
## 2 The equations of elasticity
The equations of elasticity is the system of nonlinear partial differential equations
$$\frac { \partial ^ { 2 } y } { \partial t ^ { 2 } } = \text{div} \, \frac { \partial W } { \partial F } ( \nabla y )$$
where y x, t ( ) : R d × R + → R d describes the motion. It is customary to introduce the velocity v = ∂y ∂t and the deformation gradient F = ∇ y and to write (2.1) as a first order hyperbolic system
$$\begin{cases} & \partial _ { t } F _ { i \alpha } = \partial _ { \alpha } v _ { i } \\ & \partial _ { t } v _ { i } = \partial _ { \alpha } \frac { \partial W } { \partial F _ { i \alpha } } ( F ) \\ & \partial _ { \alpha } F _ { i \beta } - \partial _ { \beta } F _ { i \alpha } = 0 \end{cases}$$
The last equation stands for a constraint that ensures that F is a gradient. This constraint does not cause problems as it is an involution [4], that is if it is satisfied for the initial data then it is propagated by (2.2) 1 to the solutions. In (2.1) the hypothesis of hyperelasticity, S = ∂W ∂F , is used stating that the Piola-Kirchhoff stress S is derived as a gradient of a stored energy function W : M d × d → R . This hypothesis makes the theory consistent with the second law of thermodynamics.
The principle of frame indifference dictates that the stored energy satisfies the invariance
$$W ( Q F ) = W ( F ) \ \forall \, Q \in S O ( d )$$
for any matrix Q describing a proper rotation. In order to interpret a geometric map y as a physical motion, y has to be globally one-to-one so that matter does not interpenetrate. A necessary condition for achieving that is to cost infinite energy whenever a finite volume is compressed down to a zero volume, what dictates the natural assumption for the stored energy
$$W ( F ) \to \infty \ \text{ as } \ \det F \to 0 \,.$$
These two requirements are in general inconsistent with convexity of the stored energy as can be seen in the schematic picture of Figure 1. It is thus too restrictive to take W as a convex function.
Figure 1: A depiction of the stored energy
The system (2.2) is a system of conservation laws. It is equipped for smooth solutions with the identity
$$\partial _ { t } \left ( \frac { 1 } { 2 } | v | ^ { 2 } + W ( F ) \right ) - \partial _ { \alpha } \left ( v _ { i } \frac { \partial W } { \partial F _ { i \alpha } } ( F ) \right ) = 0$$
expressing the conservation of mechanical energy. In the parlance of conservation law η = 1 2 | v | 2 + W F ( ) is an entropy with associated entropy flux q α = -v i ∂F ∂W iα . The issue in the background is that, due to the lack of convexity of W , the entropy η is not convex. On the other hand the standard theory of systems of conservation laws is intricately connected to convexity and is inapplicable in a direct way. Hyperbolicity of (2.2) is equivalent to the condition on W :
glyph[negationslash]
$$\frac { \partial ^ { 2 } W } { \partial F _ { i \alpha } \partial F _ { j \beta } } ( F ) \ \xi _ { i } \xi _ { j } \nu _ { \alpha } \nu _ { \beta } > 0 \quad \forall \xi \neq 0 \,, \ \nu \in \mathcal { S } ^ { d - 1 }$$
called Legendre-Hadamard conditions and associated to positive-definiteness of the acoustic tensor.
A lot of effort has been devoted into finding suitable notions to replace the assumption of convexity, starting from works on the calculus of variations associated to the minimization of potential energy. The following notions have emerged: W F ( ) is called (strictly) rank-1 convex if
$$W ( \theta F + ( 1 - \theta ) G ) < \theta W ( F ) + ( 1 - \theta ) W ( G ) \ \ 0 < \theta < 1, \ \forall F, G \ \text{ with } F - G = \xi \otimes \nu.$$
Strict rank-1 convexity is equivalent to the Legendre-Hadamard conditions for W sufficiently smooth. W F ( ) is called polyconvex if it is expressed in the form
$$W ( F ) = g ( F, \text{co} F, \text{det} \, F ) = g \circ \Phi ( F ) \quad \text{with } g ( \Xi ) \text{ convex}.$$
The latter definition is intimately connected to the notion of null-Lagrangean which is the analog of constant functions in the calculus of variations. An integrand Φ( F ) is a null-Lagrangean iff
$$\int _ { \Omega } \Phi ( \nabla y + \nabla \phi ) \, d x = \int _ { \Omega } \Phi ( \nabla y ) d x \ \forall \, y \in W ^ { 1, p } \,, \, \phi \in C _ { c } ^ { \infty } \,.$$
The following equivalences hold:
Φ( F ) is a null-Lagrangean
$$\psi ( \lambda ) & \infty \alpha \text{ in } \Psi ( F + \nabla \phi ) \, d x = \Phi ( F ) \left | \Omega \right | \quad \forall \, F \in M ^ { d \times d } \,, \, \phi \in C _ { c } ^ { \infty } \\ & \iff \quad \Phi ( F ) \text{ is rank-1 affine} \\ & \iff \quad \Phi ( F ) = \alpha \, F + \beta \, \text{cof} \, F + c \det F \\ & \iff \quad \partial _ { \alpha } \left ( \frac { \partial \Phi } { \partial F _ { i \alpha } } ( \nabla y ) \right ) = 0 \quad \text{in } \mathcal { D } ^ { \prime } \\ & \quad \cdot \quad \cdot$$
The reader is referred to [1, 2] for precise statements and their proofs, The introduction of nullLagrangeans into the theory of elasticity is due to Ericksen [7] and Edelen [6] and their role in the mathematical theory of elasticity is pointed out by Ball [1]. Their importance arises from the fact that null-Lagrangeans are weakly continuous in W 1 ,p [2], and from their role in the existence theory of equilibrium elasticity [1, 2] and in obtaining symmetrizable extensions for polyconvex elastodynamics [13, 5, 16].
Here, we explore their role in the problem of cavity formation.
## 3 Radial motions in 3-d elasticity
A motion of the form y x, t ( ) = w R,t ( ) x R , where R = | x | , x ∈ R d , is called radial. For an elastic material to support radial motions it must be isotropic and as always frame indifferent, that is the stored energy satisfies
$$W ( Q F ) = W ( F ) = W ( F Q ) \ \forall Q \in S O ( d ) \,.$$
The class of such stored energies has been characterized to be of the form W F ( ) = Φ( v , v 1 2 , ..., v d ), where Φ : R d ++ → R is a symmetric function of the eigenvalues v i of + √ F T F the so called principal stretches (see [15]).
For reasons that will be explained later we will restrict attention here to the case of three space dimensions d = 3 and the special example of stored energy
$$\Phi ( v _ { 1 }, v _ { 2 }, v _ { 3 } ) = \frac { 1 } { 2 } ( v _ { 1 } ^ { 2 } + v _ { 2 } ^ { 2 } + v _ { 3 } ^ { 2 } ) + h ( v _ { 1 } v _ { 2 } v _ { 3 } ) \,,$$
with h : R + → R + being a C 3 convex function that satisfies h δ ( ) → + ∞ as δ → 0+. This stored energy is of the polyconvex class, and the behavior as δ → 0+ is placed to avoid that a finite volume is compressed down to zero and guarantee that the solutions avoid infinite compression and can thus be interpreted as elastic motions.
## 3.1 The equations of radial elasticity.
For radial motions y x, t ( ) = w R,t ( ) x R , with R = | x | , x ∈ R d , the deformation gradient is of the form
$$F = w _ { R } \hat { x } \otimes \hat { x } + \sum _ { j = 2 } ^ { d } \frac { w } { R } \hat { x } _ { j } ^ { \perp } \otimes \hat { x } _ { j } ^ { \perp } \,,$$
where ˆ = x x R , ˆ x ⊥ j the perpendicular unit vectors, and the principal stretches are w R in the radial direction and w R of multiplicity d -1 in the orthogonal directions. The Piola-Kirchhoff stress tensor is computed as
$$S = \frac { \partial W } { \partial F } = \frac { \partial \Phi } { \partial v _ { 1 } } \hat { x } \otimes \hat { x } + \sum _ { j = 2 } ^ { d } \frac { \partial \Phi } { \partial v _ { j } } \hat { x } _ { j } ^ { \perp } \otimes \hat { x } _ { j } ^ { \perp } \,.$$
To represent a physically realizable motion, we impose that det F = w R ( w/R ) d -1 > 0 where F = ∇ y . This is equivalent to w R > 0 a condition that also suffices to avoid interpenetration of matter for radial motions.
A cumbersome but straightforward computation shows that w R,t ( ) satisfies the partial differential equation of second-order
$$w _ { t t } = \frac { 1 } { R ^ { d - 1 } } \, \partial _ { R } \left ( R ^ { d - 1 } \frac { \partial \Phi } { \partial v _ { 1 } } ( w _ { R }, \frac { w } { R }, \dots, \frac { w } { R } ) \right ) - \frac { d - 1 } { R } \frac { \partial \Phi } { \partial v _ { 2 } } ( w _ { R }, \frac { w } { R }, \dots, \frac { w } { R } ) \,.$$
It is instructive to give a short alternative derivation of (3.1). Consider the action functional L = K -P , where K is the kinetic and P the potential energy:
$$\mathcal { L } [ w ] = K - P = \int _ { 0 } ^ { T } \int _ { 0 } ^ { 1 } \frac { 1 } { 2 } w _ { t } ^ { 2 } R ^ { d - 1 } - \Phi ( w _ { R }, \frac { w } { R }, \dots, \frac { w } { R } ) R ^ { d - 1 } d R d t \,.$$
The critical points of the functional L are computed by setting to zero the variational derivative d dδ ∣ ∣ ∣ δ =0 L ( w + δψ ) = 0. This leads to the Euler-Lagrange equations
$$\int _ { 0 } ^ { T } \int _ { 0 } ^ { 1 } R ^ { d - 1 } w _ { t } \psi _ { t } - R ^ { d - 1 } \frac { \partial \Phi } { \partial v _ { 1 } } \psi _ { R } - ( d - 1 ) \frac { \partial \Phi } { \partial v _ { 2 } } \frac { \psi } { R } R ^ { d - 1 } \ d R d t = 0$$
which are precisely the weak-form of (3.1).
## 3.2 Radial elasticity features
Upon introducing the velocity, radial strain, and transversal strain, respectively
$$v = w _ { t } \ \ a = w _ { R } \ \ b = \frac { w } { R }$$
## (3.1) is expressed as the first order system
$$\ell & \quad \ell \\ \partial _ { t } v & = \frac { 1 } { R ^ { d - 1 } } \, \partial _ { R } \left ( R ^ { d - 1 } \frac { \partial \Phi } { \partial v _ { 1 } } ( a, b, \dots, b ) \right ) - \frac { d - 1 } { R } \frac { \partial \Phi } { \partial v _ { 2 } } ( a, b, \dots, b ) \\ \partial _ { t } a & = \partial _ { R } v & \quad \ell \\ \partial _ { t } b & = \frac { 1 } { R } v & \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \ell \quad \e$$
subject to the constraint ∂ R ( Rb ) = a , which is again an involution propagating from the initial data and causes no problems as constraint. For Φ 11 := ∂ 2 Φ ∂v 2 1 > 0, the system (3.2) is hyperbolic with wave speeds λ ± = ± √ Φ 11 ( a, b, ..., b ), λ 0 = 0.
A comparison with the system of elasticity in one-space dimension (5.1) indicates the following differences
- (i) The system (3.2) has lower order terms and a geometric singularity at the origin.
- (ii) As contrasted to the case of one-dimensional elasticity the wave speeds λ ± depend on the lower order terms.
- (iii) The system has an involution.
- (iv) Following the usual theory of conservation laws a weak entropy solution will be defined to satisfy the entropy inequality
$$\partial _ { t } \left ( \frac { 1 } { 2 } v ^ { 2 } + \Phi ( a, b, \dots, b ) \right ) R ^ { d - 1 } + \partial _ { R } \left ( R ^ { d - 1 } v \frac { \partial \Phi } { \partial v _ { 1 } } ( a, b, \dots, b ) \right ) \leq 0$$
expressing the dissipation of mechanical energy. The assumption (H ) implies that, 1 in a deviation to the usual theory, the entropy here is merely polyconvex.
Little is known at present concerning the existence of weak solutions for the radial elasticity system (3.2). One additional difficulty is that solutions of (3.2) have to be constructed to satisfy the constraint det F > 0 so as to be interpreted as mechanical motions. This constraint is difficult to preserve even in approximating schemes, with the exception of the variational approximation scheme in [9] which preserves the positivity of Jacobians and produces iterates that decrease the mechanical energy.
## 4 The problem of cavitation
The objective is to consider the equations of elasticity
$$\frac { \partial ^ { 2 } y } { \partial t ^ { 2 } } = \text{div} \, \frac { \partial W } { \partial F } ( \nabla y )$$
and to examine under what conditions the equations admit solutions with cavities. As already noted such solutions are radial, y x, t ( ) = w R,t ( ) x R with R = | x | , and there will be a cavity provided w (0 , t ) > 0, see Figure 2 for a depiction of such a solution.
The natural energy bound for the elasticity system is
$$\int \frac { 1 } { 2 } | y _ { t } | ^ { 2 } + | \nabla y | ^ { p } d x < \infty$$
with the exponent p determined by the growth of the stored energy W . Accordingly, if one is interested to explore cavitating solutions with finite energy, the Sobolev embedding theorem dictates to consider growth exponents p < d , where d is the space dimension. Cavitating solutions lie outside the realm of the usual continuum modeling, and one might question if continuum mechanics can account for their presence. Nevertheless, Ball [3] in a groundbreaking work examined whether the equations of equilibrium elasticity admit radial solutions with cavities. He noted that
$$y ( \cdot, t ) \in W _ { l o c } ^ { 1, p } \Longleftrightarrow \begin{cases} w ( \cdot, t ) & \text{ absolutely continuous on } ( 0, 1 ) \\ \int _ { 0 } ^ { 1 } \left ( | w _ { R } | ^ { p } + \frac { | w | ^ { p } } { R ^ { p } } \right ) R ^ { d - 1 } d R < \infty \end{cases}$$
Figure 2: A cavitating solution
and that for y ∈ W 1 1 , loc and for d ≥ 2 there is no delta-mass at the origin associated to the cavity, i.e.
$$\nabla y = \frac { w } { R } I + ( w _ { R } - \frac { w } { R } ) \hat { x } \otimes \hat { x } \quad \text{in } \mathcal { D } ^ { \prime } \text{ and a.e.}$$
This situation should be contrasted with the case d = 1 where a jump discontinuity induces a delta mass at the origin.
## 4.1 Cavitation in equilibrium elasticity
Ball [3] studied the cavitation problem from two perspectives. Using direct methods of the calculus of variations, he studies the minimization problem
$$\min \int _ { 0 } ^ { 1 } \Phi ( w ^ { \prime } ( R ), \frac { w } { R }, \dots, \frac { w } { R } ) \, R ^ { d - 1 } d R$$
over the set of admissible functions A λ . He also considered the associated Euler-Lagrange equations
$$\partial _ { R } \left ( R ^ { d - 1 } \frac { \partial \Phi } { \partial v _ { 1 } } \right ) - ( d - 1 ) R ^ { d - 2 } \frac { \partial \Phi } { \partial v _ { 2 } } & = 0 \\ w ( 1 ) & = \lambda$$
and carried out a bifurcation analysis for this system. The analysis in [3] shows:
- · There is an absolute minimum in the admissible set
$$\mathcal { A } _ { \lambda } = \left \{ w \in W ^ { 1, 1 }, w ( 0 ) \geq 0, w ^ { \prime } ( R ) > 0, w ( 1 ) = \lambda \right \}.$$
- · The minimizer w λ satisfies the Euler-Lagrange equations for R > 0 and
- a. There is a critical value λ cr such that for λ > λ cr there is a unique minimizer w λ with w λ (0) > 0.
- b. The trivial solution w = λR is stable (in a suitable sense) if λ ≤ λ cr and unstable if λ > λ cr . If λ > λ cr then w λ is stable.
The reader is referred to Ball [3] for the details (and precise statements) of these results and to Sivaloganathan and Spector [14] for an account of later results on the problem of cavitation in elastostatics. The problem of cavitation has also been studied in a context of incompressible elasticity [3] which lies outside the realm of our discussion.
## 4.2 Dynamic radial elasticity - Cavitation
Consider next the problem of dynamic cavitation in compressible, isotropic elastic materials. Such materials have stored energies W F ( ) = Φ( v , v 1 2 , v 3 ) and support an ansatz of radial motions with w R,t ( ) satisfying the equations of radial elasticity (3.1). The latter admit the special solution w R,t ( ) = λR which corresponds to a homogeneous elastic deformation of stretching λ and is a special solution emanating from initial data w 0 ( R ) = λR .
Pericak-Spector and Spector in a remarkable work [11, 12] postulated a self-similar ansatz of solutions
$$y ( x, t ) = w ( R, t ) \frac { x } { R } = t \varphi ( \frac { R } { t } ) \frac { x } { R }$$
and constructed a second self-similar solution for the dynamic elasticity system associated to cavity formation ϕ (0) > 0 provided the stretching λ is bigger than some critical value, λ > λ cr .
## Theorem 4.1 (Pericak-Spector and Spector [11]) For a stored energy function
$$\Phi ( v _ { 1 }, v _ { 2 }, v _ { 3 } ) = \frac { 1 } { 2 } ( v _ { 1 } ^ { 2 } + v _ { 2 } ^ { 2 } + v _ { 3 } ^ { 2 } ) + h ( v _ { 1 } v _ { 2 } v _ { 3 } ) \,,$$
with h : R + -→ R + satisfying the hypotheses
$$h ^ { \prime \prime } > 0 \,, \quad h ^ { \prime \prime \prime } < 0 \quad \lim _ { v \to 0 } h ( v ) = \lim _ { v \to \infty } h ( v ) = \infty \,,$$
and for λ > λ cr there exist cavitating solutions (4.1) for dimension d = 3 satisfying
$$w _ { t t } & = \frac { 1 } { R ^ { 2 } } \partial _ { R } \left ( R ^ { 2 } \frac { \partial \Phi } { \partial v _ { 1 } } ( w _ { R }, \frac { w } { R }, \frac { w } { R } ) \right ) - \frac { 1 } { R } \left ( \frac { \partial \Phi } { \partial v _ { 2 } } + \frac { \partial \Phi } { \partial v _ { 3 } } \right ) ( w _ { R }, \frac { w } { R }, \frac { w } { R } ) \, \ R > 0 \, \ t > 0 \\ w ( R, t ) & = \lambda R \quad \text{for $R>\bar{ }r$} t \text{ for some $\bar{ }r>0$}$$
and w (0 , R ) = λR .
An initial version of this theorem was proved in [11] for (H ) and 1 d ≥ 3 and it was extended in [12] to a far more general class of polyconvex constitutive functions Φ. A variant also holds for dimension d = 2 but the growth of Φ is there restricted to be slightly superlinear. Additional regularity properties for the cavitating solutions and a bifurcation study is provided in [10].
We give an outline of the main ideas in the construction and the properties of the constructed solutions. Our exposition follows [11] and [10] and we refer there for the details. The starting point is the self-similar ansatz
$$w ( R, t ) = t \varphi ( \frac { R } { t } ) \quad \ s = \frac { R } { t } \,.$$
Note that w (0 , t ) = tϕ (0) and thus ϕ (0) stands for the velocity of the cavity. Introducing (4.2) to (3.1) leads to the ordinary differential equation
$$( s ^ { 2 } - \Phi _ { 1 1 } ) \ddot { \varphi } = \frac { 2 } { s } ( \dot { \varphi } - \frac { \varphi } { s } ) \underbrace { \left [ \Phi _ { 1 2 } + \frac { \Phi _ { 1 } - \Phi _ { 2 } } { \dot { \varphi } - \frac { \varphi } { s } } \right ] } _ { P ( \dot { \varphi }, \frac { \varphi } { s }, \frac { \varphi } { s } ) \geq 1 }.$$
An important property is that due to (H ) the function 1 P defined above satisfies P ≥ 1.
Figure 3: (a) A typical trajectory of (4.4); (b) The cavity solution
It is instructive to rewrite (4.3) as a first order system for the variables
$$a = \dot { \varphi } \,, \ \ b = \frac { \varphi } { s }$$
It takes the form
$$( s ^ { 2 } - \Phi _ { 1 1 } ( a, b, b ) ) \dot { a } & = \frac { 2 } { s } ( a - b ) P ( a, b, b ) \\ \dot { b } & = \frac { 1 } { s } ( a - b )$$
A simple examination of the system (4.4) reveals the following properties:
- · The system is singular with a free boundary type of singularity at the sonic line s 2 -Φ 11 = 0.
- · The system has one class of special solutions, the states a = , corresponding to a homogeb neous deformation.
- · A second class of solutions consists of jump discontinuities. These satisfy the RankineHugoniot conditions
$$\sigma ^ { 2 } [ a ] - [ \Phi _ { 1 } ] & = 0 \\ [ b ] & = 0$$
and give rise to shocks. Note that along a shock, b is continuous but a jumps. For functions of class (H 1 ), hypothesis (H ) implies that Φ 2 111 < 0 and it can be worked out that the Lax shock admissibility criterion implies that the admissible shocks satisfy b -= b + and a -< a + = λ .
- · The main item is to study the third class of solutions to (4.4), that is continuous solutions. The analysis is based on a monotonicity property that can be read directly from (4.4). So long as the solution does not cross the equilibrium diagonal a = b or the sonic singularity, solutions are monotone functions, see Figure 3(a).
The analysis of the continuous solution of (4.3) is based on a de-singularization of the problem. One may introduce the variables ( ϕ, v ) where v = ˙( ϕ ϕ s ) 2 and rewrite (4.4) in the form
$$\dot { \varphi } & = \hat { \Phi } ( s, \varphi, v ) \\ \dot { v } & = \hat { U } ( s, \varphi, v )$$
Under (H ) the functions Φ and 1 ˆ ˆ U are non-singular at the origin s = 0 and the standard existence theory can be applied to the last system. At the origin there hold the following properties for the Cauchy stress T (0) (see [11])
$$\text{either} \quad T ( 0 ) & = 0 \Leftrightarrow v ( 0 ) = H \quad \text{stress free cavity} \\ \text{or} \quad T ( 0 ) & = G ( \varphi ( 0 ) ) \quad \text{cavity with content.}$$
The second condition is analogous to the kinetic relations familiar from the theory of phase transitions. We refer to [10] for the corresponding analysis of the singularity at dimension d = 2. The constructed solution has the following properties:
- (i) a = ˙ ϕ ↗ b = ϕ s ↘ a -b ↗ .
- (ii) It can be extended on a maximal interval of existence (0 , T )
$$Q = s ^ { 2 } - \Phi _ { 1 1 } \to 0 \quad a - b \to c < 0 \quad \text{as } s \to T.$$
- (iii) If Φ 111 < 0 the solution is connected to a uniformly deformed state through a single shock that is admissible via the Lax shock admissibility condition.
We refer to [11] and [10] for the proofs. It should be noted that the constructed cavitating solution consists of a single precursor shock that connects a convex function ϕ s ( ) to a uniform deformation, see Figure 3(b). It is necessary to have a precursor shock, that is, it is not possible for ϕ s ( ) to connect in a C 1 -fashion to a uniformly deformed state via a sonic singularity (see [10]). In summary, ϕ s ( ) looks like:
- (a) At σ there is a shock.
- (b) ϕ s ( ) convex for s < σ .
- (c) ϕ s ( ) = λs for s > σ .
The following remarkable property was proved in [11, Thm 7.2]: The cavitating solution decreases the mechanical energy, namely, if
$$E ( y, B _ { \rho } ) = \int _ { B _ { \rho } } \frac { 1 } { 2 } | y _ { t } | ^ { 2 } + W ( \nabla y ) d x$$
y h = λx is the homogeneous solution and y c is the solution with the cavity, then
$$\Gamma - \text{$w$ was non-convolution} \, \Lambda \, y _ { c } \, \Lambda \, \text{$w$ was non-convolution} \, \Lambda \, y _ { c } \, \Lambda \, \text{$w$ was} \\ E ( y _ { c }, B _ { \rho } ) - E ( y _ { h }, B _ { \rho } ) & = ( t \sigma ) ^ { 3 } \frac { 4 \pi } { 3 } \left [ \Phi ( a _ { - }, \lambda, \lambda ) - \Phi ( \lambda, \lambda, \lambda ) \\ & \quad + \frac { 1 } { 2 } \left ( \Phi _ { 1 } ( a _ { - }, \lambda, \lambda ) + \Phi _ { 1 } ( \lambda, \lambda, \lambda ) \right ) ( \lambda - a _ { - } ) \right ] \\ & < 0 \quad \text{ whenever} \quad a _ { - } \coloneqq \varphi ( \sigma - ) < \lambda \\ \cdot \quad \cdot \quad \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \end{cases}$$
where the right hand side in (4.5) corresponds to the dissipation at the precursor outgoing shock and is strictly negative.
This result states nonuniqueness for entropy weak solutions (with polyconvex energy) because of the singularity at the cavity. As already noted in [11], the paradox arises that by opening a cavity the energy of the material decreases, what induces an autocatalytic mechanism for failure. From a perspective of mechanics the problem appears to be that there is no surface energy cost for opening the cavity that is accounted for by the weak solution.
## 5 Fracture in 1-d
Insight into the non-uniqueness issue can be obtained by studying the equations of one-dimensional elasticity or the equivalent form of the elasticity system,
$$y _ { t t } = \partial _ { x } \tau ( y _ { x } ) \quad \longleftrightarrow \quad \begin{cases} u _ { t } - v _ { x } = 0 \\ v _ { t } - \tau ( u ) _ { x } = 0 \end{cases},$$
where v = y t is the velocity and u = y x is the strain. This equation admits the special solution y h ( x, t ) = λx corresponding to a homogeneous deformation with strain u h = λ and velocity v h = 0. We assume that the stress function τ ( u ) satisfies the hypotheses
$$\tau ^ { \prime } ( u ) > 0 \,, \quad \tau ^ { \prime \prime } ( u ) < 0 \, & & ( a _ { 1 } )$$
$$\tau ( u ) \to - \infty \ \text{ as } u \to 0 \ \text{ and } \ \int _ { 1 } ^ { u } \tau ( s ) d s \to + \infty \ \text{ as } u \to 0. \ \text{ (a_2)}$$
Under ( a 1 ) the wave speeds λ 1 2 , ( u ) = ± √ τ ′ ( u ) are real and (5.1) is hyperbolic. The hypothesis τ ′′ ( u ) < 0 is appropriate for an elastic material exhibiting softening elastic response and plays an important role in the forthcoming analysis. The hypothesis ( a 2 ) is applicable in the case of longitudinal motions and is placed to exclude that a finite volume is compressed down to zero. In the sequel we will consider only tensile deformations and this hypothesis will not play any significant role. (In the case of shearing motions τ ( u ) is defined for u ∈ R and ( a 2 ) is removed).
Motivated from the problem of cavitation we introduce the ansatz
$$y ( x, t ) = t \varphi \Big { ( } \frac { | x | } { t } \Big { ) } \frac { x } { | x | } = t Y ( \frac { x } { t } )$$
with Y (0) > 0. This ansatz is similar to the one used for the solution of the Riemann problem for (5.1) except at the origin x = 0. A calculation shows that
$$\partial _ { x } y = Y ^ { \prime } ( \xi ) + 2 Y ( 0 ) \delta _ { \xi = 0 }$$
and such an ansatz could conceivably provide a cavitating (in fact fracturing) solution, except that at the origin there is a delta mass and a suitable interpretation to the term τ ( ∂ y x ) should be supplied. Of course this is the main difference between the case d = 1 and the dimensions d ≥ 2 where a delta mass in the origin does not appear.
In [8], the following function
$$\nu ^ { \prime } \quad & \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\quad y ( x, t ) = t Y ( \frac { x } { t } ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\ \quad \quad y ( \xi ) = \begin{cases} \lambda \xi & \xi < - \sigma \\ - Y ( 0 ) + \alpha \xi & - \sigma < \xi < 0 \\ Y ( 0 ) + \alpha \xi & 0 < \xi < \sigma \\ \lambda \xi & \sigma < \xi \end{cases} \quad \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \\ \quad \quad \quad$$
is tested as a candidate for solution of (5.1). The constants α, λ and Y (0) are selected to sastisfy the Rankine-Hugoniot jump conditions at the two outgoing shocks at ξ = ± σ
$$Y ( 0 ) & = \sigma ( \lambda - \alpha ) \\ \sigma ^ { 2 } & = \frac { \tau ( \lambda ) - \tau ( \alpha ) } { \lambda - \alpha }$$
Figure 4: A fracturing motion in Lagrangian coordinates
Figure 5: A shear band in Eulerian coordinates
One easily checks that
τ ′′ ( u ) < 0 and α < λ = ⇒ both shocks are Lax-admissible.
The question arises if (5.3) can be given a suitable interpretation as a solution of (5.1). Note that if the equation (5.1) is interpreted as longitudinal motions then such a solution would correspond to fracture , see Figure 4. If (5.1) is interpreted as shear motions then (5.3) corresponds to a shear band , see Figure 5. The difference among the two cases is that the material splits apart in the former case, whereas it remains in contact in the latter. A more precise statement of the underlying question is whether the model (5.1) with the hyperbolic and strain softening stress-strain relation ( a 1 ) may support such motions.
## 6 Slic-solutions
There is a class of problems in material science where structures with discontinuous displacement fields emerge via a material instability mechanism. Typical examples are development of cracks in fracture, cavitation in elastic response, or formation of shear bands in plastic deformations. Such problems lie at the limits of applicability of continuum modeling and the usual concept of weak solutions is in any case inadequate to describe these motions. Nevertheless, as the material
transitions from a regime where continuum modeling is applicable to a regime that the model has to be modified (or perhaps atomistic modeling has to be employed), it is expected that at the interface both types of modeling have to apply in an intermediate regime. It is further expected that such structures should appear in a small parameter limit of more complex models that incorporate 'higher-order physics', and that their appearance occurs in a stable way.
The concept of singular limiting induced from continuum solution (in short slic -solution) [8] is an attempt to give meaning to such discontinuous solutions. Roughly speaking a discontinuous solution of (5.1) will be a slic-solution if it can be obtained as the limit of approximate smooth solutions that are an averaging of y . Due to the self-similar form of (5.2) averaging in space entails averaging in time. In [8], the following definition is introduced and used to test the example (5.3).
Definition 6.1 Let y ∈ L ∞ loc ( ( -∞ ∞ , ) ; L 1 loc ( R ) ) satisfy for some ε > 0 the monotonicity condition:
$$y ( x _ { 1 }, t ) - y ( x _ { 2 }, t ) > \varepsilon ( x _ { 1 } - x _ { 2 } ) \quad \text{for } x _ { 1 }, x _ { 2 }, t \in \mathbb { R } \text{ with } x _ { 1 } > x _ { 2 }. \quad \text{(mc)}$$
For φ a mollifier, φ ∈ C ∞ c ( R ) , φ ≥ 0 , supp φ ⊂ B 1 (the ball of radius 1), ∫ φ = 1 , we let φ n = nφ nx ( ) and define the averaged function
$$y ^ { n } ( x, t ) = \phi _ { n } \underset { x } { \ast } y = \int \phi _ { n } ( x - z ) y ( z, t ) d z.$$
The function y is called a singular limiting induced from continuum (slic-) solution of (5.1) provided for any symmetric mollifier
$$\partial _ { t t } y ^ { n } - \partial _ { x } \tau ( \partial _ { x } y ^ { n } ) = \colon f ^ { n } \to 0 \quad i n \ \mathcal { D } ^ { \prime }$$
as n →∞ .
The example (5.3) is extended to negative values of t by setting y = λx for t < 0. The resulting function is given in explicit form by
$$y ( x, t ) = \begin{cases} \lambda x \mathbb { 1 } _ { x < - \sigma t } + ( - t Y ( 0 ) + \alpha x ) \mathbb { 1 } _ { - \sigma t < x < 0 } + ( t Y ( 0 ) + \alpha x ) \mathbb { 1 } _ { 0 < x < \sigma t } + \lambda x \mathbb { 1 } _ { \sigma t < x } & t > 0 \\ \lambda x & t < 0 \,. \end{cases}$$
Using the explicit form, we calculate the averagings in (6.1). A lengthy computation in [8] shows that for ψ x, t ( ) ∈ C ∞ c ( R × R )
$$\int _ { \mathbb { R } } \int _ { \mathbb { R } } y ^ { n } \psi _ { t t } + \tau ( y _ { x } ^ { n } ) \psi _ { x } \, d x d t & = \int _ { 0 } ^ { \infty } \int _ { - \frac { 1 } { n } } ^ { \frac { 1 } { n } } \tau ( \alpha + 2 \phi _ { n } ( x ) t Y ( 0 ) ) \psi _ { x } d x d t \, + \, o ( 1 ) \\ & \to 2 t Y ( 0 ) L \int _ { 0 } ^ { \infty } \psi _ { x } ( 0, t ) d t \\ \dots$$
where L = lim u →∞ τ ( u ) u . One thus concludes:
If L > 0 then y x, t ( ) is not a slic-solution.
If L = 0 then y x, t ( ) is a slic-solution.
Consequently, a meaning to (5.3) as a slic-solution can only be given for sublinear growth of τ ( u ).
Next, we consider the energy balance. Let B = ( -r, r ) contain the entire wave fan of the approximate solution (6.1) at time t . The velocity at the boundary of such a domain vanishes, v n ∣ ∣ ∂B = 0 and the total energy of the wave fan
$$E [ y ^ { n } ; B ] = \int _ { B } \frac { 1 } { 2 } ( \partial _ { t } y ^ { n } ) ^ { 2 } + W ( \partial _ { x } y ^ { n } ) \, d x$$
evolves according to the energy balance equation
$$\frac { d } { d t } \int _ { B } \frac { 1 } { 2 } ( \partial _ { t } y ^ { n } ) ^ { 2 } + W ( \partial _ { x } y ^ { n } ) d x = \int _ { B } f ^ { n } v ^ { n } \, d x \,.$$
A calculation again shows
$$\dots \dots \dots \dots \dots \dots \dots \\ \frac { d } { d t } \int _ { B } \frac { 1 } { 2 } ( \partial _ { t } y ^ { n } ) ^ { 2 } & + W ( \partial _ { x } y ^ { n } ) d x = \int _ { B } f ^ { n } v ^ { n } d x \\ & = Y ( 0 ) ^ { 2 } \sigma - 2 \sigma ( W ( \alpha ) - W ( \lambda ) ) + 2 \int _ { 0 } ^ { \frac { 1 } { n } } \tau ( \alpha + 2 \phi _ { n } ( x ) t Y ( 0 ) ) 2 Y ( 0 ) \phi _ { n } ( x ) d x \\ & \to \mu _ { - \sigma } + \mu _ { \sigma } + p _ { c } = \colon T \\, \quad, \quad, \quad \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \end{array}$$
where the total energy production T is split into the energy dissipation of the two shocks µ ± σ and the contribution to the energy by the surface energy of the cavity p c . One easily also checks that
$$T & = \mu _ { - \sigma } + \mu _ { \sigma } + 2 ( \tau _ { \infty } - \tau ( \alpha ) ) Y ( 0 ) \\ & = \sigma Y ( 0 ) ^ { 2 } - 2 \sigma ( W ( \lambda ) - W ( \alpha ) ) + 2 \tau _ { \infty } Y ( 0 ) \\ & = \sigma Y ( 0 ) ^ { 2 } + 2 Y ( 0 ) \Big { ( } \tau _ { \infty } - \frac { W ( \lambda ) - W ( \alpha ) } { \lambda - \alpha } \Big { ) } > 0 \\. \quad \Big { - } \quad \dots \. \. \. \. \. \. \. \.$$
so that if τ ∞ = ∞ then T = + ∞ while if τ ∞ < ∞ then 0 < T < ∞ .
## 7 Cavitation in 3-d
Next, we consider solutions of the isotropic elasticity equations (3.1), with stored energy satisfying (H ) and (H ), of the form 1 2
$$y ( x, t ) = w ( | x |, t ) \frac { x } { | x | } = t \varphi ( s ) \frac { x } { | x | } \,, \ \ s = \frac { | x | } { t }$$
where ϕ s ( ) is the cavitating self-similar solution outlined in section 4. The objective is to examine how the intuition from the example in section 5 transfers to the problem of cavitation.
The following extension of the notion of slic-solution - adapted to the radial case - can be introduced.
Definition 7.1 Let y x, t ( ) = w x ,t ( | | ) x | x | with w ∈ L ∞ loc ( R ; L 1 loc ( R )) and w ,t ( · ) monotone increasing satisfy y x, t ( ) = λx for t ≤ 0 and for | x | > rt , t ¯ > 0 or some ¯ r > 0 . The function y is called a singular limiting induced from continuum (slic)-solution of (2.1) if
$$y ^ { n } = w ^ { n } ( | x |, t ) \frac { x } { | x | } \,, \quad w i t h \quad w ^ { n } = \phi _ { n } \ast w \,,$$
satisfies det ∇ y n ≥ ε n > 0 for all ∈ N , and
$$\frac { \partial ^ { 2 } y ^ { n } } { \partial t ^ { 2 } } - d i v \frac { \partial W } { \partial F } ( \nabla y ^ { n } ) = \colon f ^ { n } \to 0 \quad i n \ \mathcal { D } ^ { \prime }$$
as n →∞ , for all φ ∈ C ∞ c ( R ) positive, symmetric mollifiers with φ (0) > 0 .
The convolution w n = φ n glyph[star] R w is defined by first extending w R,t ( ) antisymmetrically in R and the definition takes into account the layer structure at the cavity. In agreement with the definition of slic-solution (and of the energy in one space dimension) we define the energy of a multidimensional slic-solution.
Definition 7.2 Let w ∈ W 1 , ∞ loc ( R ; L 1 loc ( R )) , The energy of a slic-solution in a domain B ⊂ R d containing the entire wave fan and for a.e. t ∈ R is defined as
$$E [ y, B ] ( t ) \coloneqq \lim _ { n \to \infty } \int _ { B } \frac { 1 } { 2 } | y _ { t } ^ { n } ( x, t ) | ^ { 2 } + W ( \nabla y ^ { n } ( x, t ) ) d x$$
with y n defined by (7.1) .
The following is proved in [8, Thm 3.8, Prop 3.10] based on detailed estimations of w n = φ n glyph[star] R w on the entire wave fan but mainly around the cavity:
Theorem 7.3 (i) The weak solution constructed by Pericak-Spector and Spector is a slic-solution if and only if
$$\lim _ { u \to \infty } \frac { h ^ { \prime } ( u ^ { 3 } ) } { u } = 0 \,.$$
$$\left ( i i \right ) \, I f \, L = \lim _ { u \rightarrow \infty } \frac { h ( u ) } { u } \ t h e n$$
$$\cdot \, \cdot \, \cdot \, \quad & \quad \cdots \quad \ u \\ \lim _ { n \to \infty } \int _ { B } \frac { 1 } { 2 } & | v ^ { n } | ^ { 2 } + W ( \nabla y ^ { n } ) - W ( \lambda ) d x \\ & = \underbrace { \int _ { B } \left [ \frac { 1 } { 2 } | v | ^ { 2 } + W ( \nabla y ) - W ( \lambda ) \right ] d x } _ { < 0 } + \underbrace { \left ( t \varphi ( 0 ) \right ) ^ { 3 } \frac { 4 \pi } { 3 } L } _ { > 0 } = \colon P _ { w f } > 0 \,. \\ \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \,.$$
In particular, (ii) shows that for a slic-solution the energy of the solution with cavity is larger than the energy of the uniform deformation.
A rough explanation of the discrepancy . We refer to [8] for the proof, but outline the main calculation of the discrepancy between a weak and a slic solution. Recall that
$$y ( x, t ) = w ( R, t ) \frac { x } { R } & = t \varphi \left ( \frac { R } { t } \right ) \frac { x } { R } \\ v ( R, t ) & = w _ { R } \left ( \frac { w } { R } \right ) ^ { 2 } \\ \quad \, \quad \,.$$
where ϕ s ( ) is the self-similar solution of (4.3) in section 4, and that the specific volume v ( · , t ) is strictly monotone increasing and obeys the bounds
$$v _ { 0 } \leq v ( R, t ) \leq \lambda ^ { 3 } \,.$$
The approximate solution and the corresponding approximation of the specific volume are
$$\mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { V } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ {\Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \mathcal { \Psi } _ { \Psi } \\ v ^ { n } & = w _ { R } ^ { n } \left ( \frac { w ^ { n } } { R } \right ) ^ { 2 }$$
where φ n symmetric mollifier, φ (0) > 0 and w ,t ( · ) is an odd extension of w . It can be proved that v n satisfies the bounds
$$c _ { 1 } & \leq v ^ { n } ( R, t ) \leq c _ { 2 } \quad \text{ for } R > \frac { 1 } { n } \\ c _ { \phi } n ^ { 3 } w ( 0, t ) ^ { 3 } & \leq v ^ { n } ( R, t ) \leq c _ { 3 } ( 1 + t ^ { 3 } n ^ { 3 } ) \quad \text{ for } R < \frac { 1 } { n }$$
where c φ a positive constant depending on the shape of the mollifier. Note that the approximate specific volume v n detects that there is a cavity forming, while v stays bounded away from zero.
To show we have a slic-solution we need to show that
$$\frac { \partial ^ { 2 } y ^ { n } } { \partial t ^ { 2 } } - \text{div} \, \frac { \partial W } { \partial F } ( \nabla y ^ { n } ) = \colon f ^ { n } \to 0 \quad \text{in} \, \mathcal { D } ^ { \prime }.$$
The discrepancy D between the weak and the slic-solution lies in the behavior near the cavity, in the ball | x < | 1 n , and can be computed by the following heuristic 'calculation'
$$\dots \sim \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \ \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \$$
Therefore, D is related to the limit lim u →∞ h ′ ( u 3 ) u . For the details and the calculation of the discrepancy in the energy, see [8].
In conclusion, two issues may arise when a weak solution is viewed as limit of continuous solutions in a context of strong singularities (like a cavity or fracture or shear band)
- a. New terms might appear in the momentum balance equation.
- b. Even when such forces vanish in the limit, their effect can be felt in the energy balance and their contribution might affect the total energy as the singularity forms.
Acknowledgements This research was supported by the EU FP7-REGPOT project 'Archimedes Center for Modeling, Analysis and Computation'. AET is partially supported by the 'Aristeia' program of the Greek Secretariat for Research.
## References
- [1] J.M. Ball , Convexity conditions and existence theorems in nonlinear elasticity, Arch. Rational Mech. Anal. 63 (1977), 337-403.
- [2] J.M. Ball, J.C. Currie and P.J. Olver Null Lagrangians, weak continuity, and variational problems of arbitrary order J. Functional Analysis 41 (1981), 135-174.
- [3] J.M. Ball , Discontinuous equilibrium solutions and cavitation in nonlinear elasticity, Philos. Trans. Roy. Soc. London Ser. A , 306 , (1982) 557-611.
- [4] C. Dafermos , Quasilinear hyperbolic systems with involutions, Arch. Rational Mech. Anal. 94 (1986), 373-389.
- [5] S. Demoulini, D.M.A. Stuart, A.E. Tzavaras , A variational approximation scheme for three-dimensional elastodynamics with polyconvex energy, Arch. Rational Mech. Anal. 157 (2001), 325-344.
- [6] D.G.B. Edelen , The null set of the Euler-Lagrange operator Arch. Rational Mech. Anal. 11 (1962), 117-121.
- [7] J.L. Ericksen , Nilpotent energies in liquid crystal theories, Arch. Rational Mech. Anal. 10 (1962), 189-196.
- [8] J. Giesselmann and A.E. Tzavaras , Singular limiting induced from continuum solutions and the problem of dynamic cavitation. (submitted), (2013), arXiv preprint arXiv:1306.6084.
- [9] A. Miroshnikov and A.E. Tzavaras , A variational approximation scheme for polyconvex elastodynamics that preserves the positivity of Jacobians. Comm. Math. Sciences 10 (2012), 87-115.
- [10] A. Miroshnikov and A.E. Tzavaras , On the construction and properties of weak solutions describing dynamic cavitation. (preprint).
- [11] K.A. Pericak-Spector and S.J. Spector , Nonuniqueness for a hyperbolic system: cavitation in nonlinear elastodynamics. Arch. Rational Mech. Anal. 101 (1988), 293 - 317.
- [12] K.A. Pericak-Spector and S.J. Spector , Dynamic cavitation with shocks in nonlinear elasticity. Proc. Royal Soc. Edinburgh Sect A 127 (1997), 837 - 857.
- [13] T. Qin , Symmetrizing nonlinear elastodynamic system, J. Elasticity 50 (1998), 245-252.
- [14] J. Sivaloganathan and S.J. Spector , Myriad radial cavitating equilibria in nonlinear elasticity. SIAM J. Appl. Math. 63 (2003), 1461 - 1473.
- [15] C. Truesdell, W. Noll , The non-linear field theories of mechanics , Handbuch der Physik III , 3 (Ed. S.Fl¨gge), Springer Verlag, Berlin, 1965. u
- [16] D.H. Wagner , Symmetric hyperbolic equations of motion for a hyper-elastic material, J. Hyper. Differential Equations 6 (2009), 615-630.
|
10.5802/slsedp.41
|
[
"Jan Giesselmann",
"Alexey Miroshnikov",
"Athanasios E. Tzavaras"
] |
2014-08-04T22:25:03+00:00
|
2017-09-01T20:00:21+00:00
|
[
"math.AP",
"35L67, 35L70, 74B20, 74H20, 74H60"
] |
The problem of dynamic cavitation in nonlinear elasticity
|
The notion of singular limiting induced from continuum solutions
(slic-solutions) is applied to the problem of cavitation in nonlinear
elasticity, in order to re-assess an example of non-uniqueness of entropic weak
solutions (with polyconvex energy) due to a forming cavity.
|
1408.0832v1
|
## SUSY-inspired one-dimensional transformation optics
Mohammad-Ali Miri,* Matthias Heinrich, and Demetrios N. Christodoulides
CREOL, College of Optics and Photonics, University of Central Florida, Orlando, Florida 32816-2700, USA * [email protected]
Transformation optics aims to identify artificial materials and structures with desired electromagnetic properties by means of pertinent coordinate transformations. In general, such schemes are meant to appropriately tailor the constitutive parameters of metamaterials in order to control the trajectory of light in two and three dimensions. Here we introduce a new class of one-dimensional optical transformations that exploits the mathematical framework of supersymmetry (SUSY). This systematic approach can be utilized to synthesize photonic configurations with identical reflection and transmission characteristics, down to the phase, for all incident angles, thus rendering them perfectly indistinguishable to an external observer. Along these lines, low-contrast dielectric arrangements can be designed to fully mimic the behavior of a given high-contrast structure that would have been otherwise beyond the reach of available materials and existing fabrication techniques. Similar strategies can also be adopted to replace negative-permittivity domains, thus averting unwanted optical losses.
Introduction. The problem of reconstructing the shape of a potential distribution solely from information carried by its scattering pattern has a long-standing history in a number and diverse disciplines of science and technology [1-5]. Naturally, in such inverse problems, the question of uniqueness is of crucial importance: Are the properties of an object fully determined by its corresponding far-field scattering data? In general, the answer is no. In quantum mechanical settings, for example, one can always identify an N-parameter family of different potentials that support the same discrete set of N bound-state eigenvalues and exhibit similar scattering characteristics [4]. Closely related to this subject is the idea of supersymmetry (SUSY) [6-10]. This mathematical framework emerged in quantum field theory as a means to treat fermions and bosons on equal footing [9]. Subsequently, notions of supersymmetry were utilized to obtain isospectral and phaseequivalent potentials within the context of non-relativistic quantum mechanics [10].
On the other hand, recent developments in transformation optics and optical conformal mapping have brought about novel methodologies to address inverse problems [11-16]. By virtue of coordinate transformations, such schemes can in principle provide the spatial distribution of electric permittivities and magnetic permeabilities that would perform a desired task such as cloaking [17-22]. As one would expect, the material properties required to implement such configurations might not always be available in practice. Clearly of interest would be to develop alternative strategies that allow one to judiciously control the scattering properties of an object, while at the same time reducing the complexity of the structures involved.
As we will see, supersymmetry can provide a new avenue for one-dimensional transformation optics that would have been otherwise impossible using other multi-dimensional approaches (see Fig. 1). Along these lines we introduce appropriate optical transformations in 1D refractive index landscapes and explore their implications in terms of their far-field response. In addition to finding superpartners with similar scattering behavior, systematic SUSY deformations allow us to design systems that exhibit identical complex reflection and transmission coefficients for all incident angles. As a result, two such dielectric objects, however dissimilar, become virtually
indistinguishable. Remarkably, the proposed formalism can be employed to synthesize photonic configurations that behave exactly the same way as high refractive index contrast devices - by only utilizing low-contrast dielectric media. Similar methodologies can be employed to substitute negative-permittivity inclusions with purely dielectric media as a means to obtain the intended functionality without introducing any additional loss.
Fig. 1. Schematic overview of the different SUSY optical transformations. Starting from a given fundamental structure 𝜖 , supersymmetric partners 𝜖𝑝 can be constructed. Whereas the broken SUSY system 𝜖𝑝 ( 𝑏𝑟 ) preserves all bound modes, unbroken SUSY ( 𝜖𝑝 ( 𝑢𝑏 ) ) removes the fundamental mode. Regardless, in both cases the intensity reflection and transmission coefficients of the superpartners are identical to those of the fundamental system. In order to maintain the full complex scattering characteristics, a family 𝜖𝑓 of iso-phase structures can be synthesized. Finally, a hierarchical sequence of higher-order superpartners 𝜖 𝑝 2 , … 𝑁 ( 𝑢𝑏 ) may be utilized to obtain a scattering-equivalent structure, which requires a substantially lower refractive index contrast than that involved in the original system 𝜖 .
Supersymmetric optical transformations. In one-dimensional inhomogeneous settings, the propagation of TE polarized waves is known to obey the Helmholtz equation [23] �𝜕𝑥𝑥 + 𝜕𝑦𝑦 + 𝑘0𝜖 𝑥 �𝐸𝑧 2 ( ) ( 𝑥 𝑦 , ) = 0 where 𝑘0 is the vacuum wavenumber and 𝜖 𝑥 ( ) is the relative permittivity of a given (fundamental) structure to be emulated via SUSY transformations (see Fig.1). The analysis of TM waves can be carried out in a similar manner (see the supplementary information). In the TE case, the spatial dependence of the electric field 𝐸𝑧 can be described via 𝐸𝑧 ( 𝑥 𝑦 , ) = 𝜓 𝑥 𝑒 ( ) 𝑖𝛽𝑦 . Here, 𝛽 = 𝑘0𝑛0 sin 𝜃 represents the y-component of the wave vector for an incidence angle 𝜃 , and 𝑛0 = �𝜖 𝑥 → ( ± ∞ ) is the background refractive index. By employing the normalized quantities 𝑋 = 𝑘0𝑥 𝑌 , = 𝑘0𝑦 and 𝛺 = 𝛽 2 / 𝑘0 2 , the Helmholtz equation then reduces to a 1D Schrödinger-like equation
$$H \psi ( X ) = \Omega \psi ( X ). \quad \ ( 1 )$$
The resulting Hamiltonian 𝐻 = 𝜕𝑋𝑋 + ( 𝜖 𝑋 ) can be factorized as 𝐻 = 𝐵𝐴 + 𝛼 where the operators 𝐴 and 𝐵 are defined as 𝐴 = 𝜕𝑋 + 𝑊 𝑋 ( ) , 𝐵 = 𝜕𝑋 -𝑊 𝑋 ( ) , and 𝛼 is an auxiliary constant of the problem. Here 𝐵 = -𝐴 † , where ' † ' represents the Hermitian conjugate. The superpotential 𝑊 can then be obtained as a solution of the Riccati equation [7]:
$$\epsilon ( X ) = + W ^ { \prime } - W ^ { 2 } + \alpha, \quad \ \ ( 2 )$$
in terms of the fundamental permittivity profile 𝜖 𝑋 ( ) . Once 𝑊 has been determined, one can establish a partner Hamiltonian 𝐻𝑝 = 𝐴𝐵 + 𝛼 , which corresponds to a new distribution in the electric permittivity
$$\epsilon _ { p } ( X ) = - W ^ { \prime } - W ^ { 2 } + \alpha. \quad \ \ ( 3 )$$
As a direct consequence of this construction, the modes 𝜓𝑝 of the partner potential 𝜖𝑝 are related [24] to the ones of the fundamental through the following expressions 𝜓𝑝 ∝ ( 𝜕𝑋 + 𝑊 𝜓 ) and 𝜓 ∝ ( 𝜕𝑋 -𝑊 𝜓𝑝 ) . These latter relations hold for guided waves as well as for radiation modes, and each such pair of states is characterized by a common eigenvalue. We note that two options for choosing 𝛼 exist: (a) Assuming that the structure support s at least one bound state, one may opt to set α equal to the fundamental mode's eigenvalue, i.e., 𝛼 = 𝛺0 . (b) The other possibility is to choose 𝛼 > 𝛺0 , irrespective of whether the system supports bound states or not. The first case corresponds to an unbroken SUSY: The two potentials share the guided wave eigenvalue spectra, except for that of the fundamental mode, which does not
have a corresponding state in the partner. In the second case, however, SUSY is broken, and the two arrangements share an identical eigenvalue spectrum, including that of the fundamental mode. As an example, Fig. 2(a) depicts the relative permittivity distribution 𝜖 𝑋 ( ) = 1 + exp[ - 𝑋/ ( 5) ] 8 corresponding to a step-index-like waveguide; its unbroken and broken SUSY partners are shown in Figs. 2(b,c) respectively. 𝑊 can also be found analytically [7] via 𝑊 = -𝜕𝑋 ln( 𝜓0 ) in the unbroken SUSY case, i.e. for 𝛼 = 𝛺0 , when 𝜖 supports at least one bound state 𝜓0 . In either regime, Eq. (2) can always be solved numerically to obtain the superpotential 𝑊 . An alternative approach is to start with an arbitrary superpotential and construct the two superpartner structures 𝜖 and 𝜖𝑝 according to Eqs. (2,3). In this scenario, it still remains to be determined whether SUSY is unbroken or broken. This question can be resolved by the so-called Witten index [6]. In general if, 𝑊 𝑋 ( ) approaches 𝑊 ± at 𝑋 → ± ∞ , unbroken SUSY requires 𝑊+ = -𝑊, while a broken SUSY demands that 𝑊+ = 𝑊-.
Fig. 2. Relative permittivity distributions of the original and the transformed potentials, (a) The fundamental system has a step-like profile 𝜖 𝑋 ( ) = 1 + exp[ - 𝑋 ( /5) ] 8 . (b) The superpartner in the unbroken SUSY regime, (c) The superpartner in the broken SUSY case, and (d) phase-equivalent structures. (e) Scattering geometry. (f-h) Superpotentials 𝑊 corresponding to panels (b-d). (j) Identical reflectivity 𝑅 (solid line) and transmittivity 𝑇 (dashed line) corresponding to Figs. 1(a-d). (k-m) Relative phases of the reflection ( 𝛥𝛷𝑟 , solid line) and transmission ( 𝛥𝛷𝑡 , dashed) coefficients of the structures in (b-d) compared to the fundamental system (a) as a function of the incident angle 𝜃 . The scattering characteristics were evaluated by means of the differential transfer matrix method [23].
It is important to note that more than one superpotential can exist for any given distribution 𝜖 𝑋 ( ) . As shown in the supplementary information, one can actually find a parametric family 𝑊𝑓 of viable superpotentials that satisfy Eq. (3). Whereas all members of this family lead to the same superpartner 𝜖𝑝 , each of them describes a different permittivity distribution 𝜖 according to Eq. (2). The resulting parametric family [10,25] of structures 𝜖𝑓 ( 𝑋 𝐶 ; ) is associated with the fundamental distribution 𝜖 and its ground state 𝜓0 as follows:
$$\epsilon _ { f } ( X ; C ) = \epsilon ( X ) + 2 \partial _ { X X } \ln \left ( C + \int _ { - \infty } ^ { X } \psi _ { 0 } ^ { 2 } ( X ^ { \prime } ) d X ^ { \prime } \right ), \quad ( 4 )$$
where 𝐶 represents a free parameter. Figure 2(d) depicts such family members for the fundamental structure ϵ shown in Fig. 2(a) when 𝐶 = 0.1 , 0.5 or 2.0 , respectively. Note that the original permittivity distribution ϵ is in itself a member of this family, since 𝜖𝑓 →𝜖 for
𝐶 → + ∞ . All the modes 𝜓𝑓 of any other member are related to its states 𝜓 according to 𝜓𝑓 ∝ ( 𝜕𝑋 -𝑊𝑓 )( 𝜕𝑋 + 𝑊 𝜓 ) (see supplementary information). As a result, all family members share the same guided wave characteristics, e.g. they have identical sets of eigenvalues as in the case of broken SUSY. Figures 2(f-h) provide an overview of the different superpotential functions in the regimes of unbroken and broken SUSY (Figs. 2(b,c)) as well as for the family members shown in Fig. 2(d).
Scattering characteristics . Let us now turn our attention to the scattering characteristics of structures connected by SUSY transformations. Consider a plane wave exp ( 𝑖𝑋𝑛0 cos 𝜃 + 𝑖𝑌𝑛0 sin 𝜃 ) incident from the left, i.e. 𝑋 → -∞ , as shown schematically in Fig. 2(e). For reasons of simplicity we assume a uniform background medium, 𝑛+ = 𝑛-= 𝑛0 at 𝑋 → ± ∞ ; the general case of 𝑛+ ≠ 𝑛is discussed in the supplementary information. Here, the reflected and transmitted waves in the far field are given as 𝑟 exp( -𝑖𝑋𝑛0 𝑐𝑜𝑠 𝜃 + 𝑖𝑌𝑛0 𝑠𝑖𝑛 𝜃 ) and 𝑡 exp ( 𝑖𝑋𝑛0 cos 𝜃 + 𝑖𝑌𝑛0 sin 𝜃 ) in terms of the complex reflection and transmission coefficients 𝑟 and 𝑡 [24]. By adopting similar solutions for the partner scatterer 𝜖𝑝 , its respective reflection and transmission coefficients 𝑟 𝑝 and 𝑡 𝑝 can readily be found (see Table 1 and supplementary information). Interestingly, the SUSY transformation yields a partner structure with exactly the same absolute values in reflection and transmission, as illustrated in Fig. 2(j). Evidently, all the permittivity distributions from Figs. 2(a-d) display identical reflectivities 𝑅 = | 𝑟 | 2 = �𝑟 � 𝑝 2 and transmittivities 𝑇 = 1 -𝑅 for all angles of incidence. In contrast, the scattering phases depend on whether supersymmetry is broken or not (see Table 1). In the case of unbroken SUSY, both reflection and transmission coefficients acquire additional phases with respect to the fundamental scattering potential. If on the other hand SUSY is broken, the transmission coefficient is the same in both amplitude and phase. Finally, one can show that each member of the parametric family 𝜖𝑓 directly inherits all scattering properties of the original structure, i.e. they are phase-equivalent to 𝜖 . Figures 2(k-m) illustrate these relations.
Table 1 . Reflection and transmission coefficients for the different SUSY transformations. 𝑊-= 𝑊 -∞ ( ) designates the asymptotic value of the superpotential on the left side of the structure, and 𝑟 𝑡 , are the coefficients of the original structure.
| Coefficient | Unbroken SUSY | Broken SUSY | Iso-phase |
|---------------|------------------------------------------------|-----------------------------------------------|-------------|
| Reflection | 𝑟 𝑝 = 𝑟 ⋅ exp �- 2 𝑖 tan -1 � 𝑛 0 cos 𝜃 𝑊 - �� | 𝑟 𝑝 = 𝑟 ⋅ exp �- 2 𝑖 tan -1 � 𝑛 0 cos 𝜃 𝑊 - � | 𝑟 𝑓 = 𝑟 |
| Transmission | 𝑡 𝑝 = 𝑡 ⋅ exp �- 2 𝑖 tan -1 � 𝑛 0 cos 𝜃 � � | 𝑡 𝑝 = 𝑡 | 𝑡 𝑓 = 𝑡 |
𝑊-
Wavelength dependence . So far the performance of these systems has been examined at a given operating wavelength 𝜆0 . Of importance would be to investigate to what extend their supersymmetric properties persist when the wavelength 𝜆 varies around 𝜆0 . As one would expect, even if two dissimilar profiles exhibit the same phases at a given wavelength, their internal light dynamics may gradually undergo different changes with 𝜆 . To elucidate this structural dispersion, we provide the spectral dependence of the difference in transmittivities 𝛥𝑇 (or reflectivities 𝛥𝑅 ) between the fundamental structure (Fig. 2(a)) and its superpartners (Fig. 2(b-d)) as a function of the incidence angle 𝜃 , as shown in Figs. 3(a-c). As these figures indicate, this difference only becomes notable in the unbroken SUSY regime (Fig. 3(a)), while it is almost absent under broken SUSY and iso-phase conditions (Figs. 3(b,c)). The difference in the corresponding reflection phases is similarly presented in Figs. 3(d-f). The dashed lines trace the abrupt phase jumps of 𝜋 , which mark the resonances in the two partners and intersect at the design wavelength 𝜆0 . Evidently, the iso-phase design displays the greatest resilience with respect to spectral deviations. Note that resonances play no role in the transmission phases, as can be seen in Figs. 3(g-j). In this latter case, the iso-phase system again proves to be the least susceptible to spectral deviations. These results demonstrate that SUSY transformations can be robust over a broad spectral range around the design wavelength.
Fig. 3. Reflection/transmission characteristics of structures obtained by SUSY transformations depicted in Fig. 2 as function of wavelength 𝜆 and angle of incidence 𝜃 . (a-c) Intensity difference in transmission. (d-f) Relative phases in reflection and (g-j) Relative phases in transmission. The dashed lines follow the resonance-induced 𝜋 phase jumps in fundamental structure and unbroken-SUSY partner. Top row: Unbroken SUSY, Middle row: Broken SUSY, bottom row: Iso-phase case ( 𝐶 = 0.5 ).
Index-contrast reduction . One of the main challenges in designing optical systems is the limited dynamic range of refractive indices associated with available materials. This issue becomes particularly acute when high contrast arrangements are desirable. For example, the number of grating unit cells required to achieve a certain diffraction efficiency grows with the inverse logarithm of the index contrast 𝑛2 / 𝑛1 between the individual layers [23]. As it turns out, SUSY optical transformations can be utilized to reduce the index contrast needed for a given structure. This can be done through a hierarchical ladder of superpartners, i.e. sequentially removing the bound states of the original high-contrast arrangement (Fig. 4a). As a general trend, each successive step demands less contrast in the corresponding index landscape than the previous one (Fig. 4b). The ultimate result is a low-contrast equivalent structure that fully inherits the reflectivity and transmittivity of the original configuration (Figs. 4c,d).
Fig. 4. (a) Hypothetical high-contrast dielectric layer arrangement that supports 𝑁 = 9 guided modes. (b) Hierarchical sequence of partner structures obtained through iterative SUSY transformations. (c) Despite the general trend towards lower-contrast configurations, each intermediate step inherits the reflectivity and transmittivity of the fundamental system (a). (d) The resulting low-contrast structure is free of bound states and faithfully mimics the intensity scattering characteristics of the original highcontrast configuration for all angles of incidence.
Replacing negative-permittivity features . Finally, SUSY transformations can provide a possible avenue in replacing negative-permittivity inclusions (typically accompanied by losses) by purely dielectric materials. In this respect, inverse SUSY transformations, which now add modes with certain propagation constants to a given structure, can instead be used to locally elevate the permittivity (see supplementary information). Along similar lines, it is possible to find superpotentials that relate a structure with metallic or negative permittivity regions to an equivalent arrangement with entirely positive ϵ, as depicted in Fig. 5. Here we make use of the fact that in a broken -SUSY transformation, the spatial average of ϵ happens to be a conserved quantity. Therefore, changes in the broader vicinity of the original metal-dielectric structure can be used to achieve this goal.
Fig. 5. (a) A metal-dielectric grating arrangement comprising five layers of negative electrical permittivity (red sections). (b) An entirely dielectric superpartner grating constructed in the broken SUSY regime, using the respective superpotential (c). (d) Despite the absence of any metallic regions, the equivalent structure exhibits identical reflectivities/transmittivities.
Discussion. In conclusion, we have introduced a new type of supersymmetric optical transformations for arbitrary one-dimensional refractive index landscapes. Compared to conventional transformation optics, our approach poses significantly less stringent requirements on the constituent parameters, and does not involve any modifications to the magnetic response of the materials involved. This method can be utilized to construct photonic arrangements that faithfully mimic the behavior of high-index-contrast or even metal-dielectric structures. SUSY transformation optics may have potential applications in a wide range of scenarios that rely on engineered scattering and transmission properties, such as for example optical metasurfaces, anti-reflection coatings and diffraction gratings. Of interest will be to explore how the aforementioned strategies could be paired up with recently developed transformation schemes for guided-wave photonics based on dielectric materials [26,27].
## Acknowledgment
We acknowledge the financial support from National Science Foundation (NSF) (1128520), and from Air Force Office of Scientific Research (AFOSR) (FA9550-12-1-0148, FA9550-14-1-0037). M.H. was supported by the German National Academy of Sciences Leopoldina (LPDS 2012-01).
## References
- 1. I. M. Gel'fand, and B. M. Levitan, "On the determination of a differential equation from its spectral function," Izvest. Akad. Nauk. 15 , 309-360 (1951).
- 3. K. Chadan, and P. C. Sabatier, " Inverse Problems in Quantum Scattering Theory ," (Springer, New York, 1977).
- 2. E. Wolf, "Three-dimensional structure determination of semi-transparent objects from holographic data," Opt. Commun. 1 , 153-156 (1969).
- 4. P. Deift, and E. Trubowitz, "Inverse scattering on the line," Comm. Pure Appl. Math. 32, 121-251 (1979).
- 6. E. Witten, "Dynamical breaking of supersymmetry," Nucl. Phys. B185, 513-554 (1981).
- 5. I. Kay, and H. E. Moses, "Reflectionless Transmission through Dielectrics and Scattering Potentials," J. Appl. Phys. 27, 15031508 (1956).
- 7. F. Cooper, A. Khare, and U. Sukhatme, "Supersymmetry and quantum mechanics, " Phys. Rep. 251, 267-385 (1995).
- 8. M. M. Nieto, "Relationship between supersymmetry and the inverse method in quantum mechanics," Phys. Lett. B 145, 208210 (1984).
- 9. A. Neveu, and J. H. Schwarz, "Factorizable dual model of pions," Nucl. Phys. B31, 86-112 (1971).
- 11. J. B. Pendry, D. Schurig, and D. R. Smith, "Controlling Electromagnetic Fields", Science 312, 1780-1782 (2006).
- 10. A. Khare, and U. Sukhatme, "Phase-equivalent potentials obtained from supersymmetry," J. Phys. A: Math. Gen. 22, 28472860 (1989).
- 12. U. Leonhardt, "Optical Conformal Mapping, " Science 312, 1777-1780 (2006).
- 14. V. M. Shalaev, "Transforming Light," Science 322, 384-386 (2008).
- 13. A. Vakil, and N. Engheta, "Transformation Optics Using Graphene," Science 332, 1291-1294 (2011).
- 15. U. Leonhardt, and T. G. Philbin, "Transformation Optics and the Geometry of Light," Prog. Opt. 53, 69-152 (2009).
- 17. A. Alù, M. Silveirinha, A. Salandrino, and N. Engheta, "Epsilon-near-zero metamaterials and electromagnetic sources: Tailoring the radiation phase pattern," Phys. Rev. B 75, 155410 (2007).
- 16. H. Chen, C. T. Chan, and P. Sheng, "Transformation optics and metamaterials," Nat. Mater. 9, 387-396 (2010).
- 18. G. W. Milton, M. Briane, and J. R. Willis, "On cloaking for elasticity and physical equations with a transformation invariant form," New J. Phys. 8, 248-248 (2006).
- 20. L. H. Gabrielli, J. Cardenas, C. B. Poitras, and M. Lipson, "Silicon nanostructure cloak operating at optical frequencies," Nat. Photon. 3, 461-463 (2009).
- 19. J. Valentine, J. Li, T. Zentgraf, G. Bartal, and X. Zhang, "An optical cloak made of dielectrics," Nat. Mater. 8, 568-571 (2009).
- 21. Y. Lai, J. Ng, H. Y. Chen, D. Z. Han, J. J. Xiao, Z.-Q. Zhang, and C. T. Chan, "Illusion optics: The optical transformation of an object into another object," Phys. Rev. Lett. 102, 253902 (2009).
- 23. P. Yeh, A. Yariv, and C. S. Hong, "Electromagnetic propagation in periodic stratified media. I. General theory," J. Opt. Soc. Am. 67, 423-438 (1977).
- 22. B. Zhang, Y. Luo, X. Liu, and G. Barbastathis, "Macroscopic Invisibility Cloak for Visible Light," Phys. Rev. Lett. 106, 033901 (2011).
- 24. M.-A. Miri, M. Heinrich, R. El-Ganainy, and D. N. Christodoulides, "Supersymmetric Optical Structures," Phys. Rev. Lett. 110, 233902 (2013).
- 26. H. Xu, B. Zhang, T. Yu, G. Barbastathis, and H. Sun, "Dielectric waveguide bending adapter with ideal transmission: practical design strategy of area-preserving affine transformation optics," J. Opt. Soc. Am. B 29, 1287-1290 (2012).
- 25. M.-A. Miri, M. Heinrich, and D. N. Christodoulides, "Supersymmetry-generated complex optical potentials with real spectra", Phys. Rev. A. 87 , 043819 (2013).
- 27. D. Liu, L. H. Gabrielli, M. Lipson, and S. G. Johnson, "Transformation inverse design," Opt. Express 21, 14223-14243 (2013).
## Supplementary information for
## SUSY-inspired one-dimensional transformation optics
Mohammad-Ali Miri, Matthias Heinrich, and Demetrios N. Christodoulides
## S1. Iso-phase potential families
For any given distribution of the relative electrical permittivity 𝜖 𝑋 ( ) , an infinite number of viable superpotentials exist. To show this, let us start from Eq. (3), which relates the superpartner 𝜖𝑝 to the superpotential 𝑊 . Starting from a particular 𝑊 , this solution can be generalized by adopting the form 𝑊𝑓 = 𝑊 +1/ 𝑣 , in which case the unknown function 𝑣 satisfies ( 𝜕𝑋 2 𝑊 𝑣 ) = 1 . Direct integration readily leads to 𝑣 = 𝑒 +2∫𝑊𝑑𝑋 �𝐶 + ∫𝑒 -2∫𝑊𝑑𝑋 𝑑𝑋� , where 𝐶 is an arbitrary constant, giving rise to a parametric family 𝑊𝑓 of superpotentials
$$W _ { f } ( X ; C ) = W + \partial _ { x } \ln ( C + \int e ^ { - 2 \int w d x } d X ). \quad \ \ ( S 1 )$$
Whereas all members of this family correspond to the same superpartner 𝜖𝑝 , each of them describes a different distribution according to Eq. (2). The resulting family of structures 𝜖𝑓 is related to the fundamental one 𝜖 according to
$$\epsilon _ { f } ( X ; C ) = \epsilon ( X ) + 2 \, \partial _ { X X } \ln \left ( C + \int e ^ { - 2 \int w d x } d X \right ), \quad ( S 2 )$$
If the superpotential 𝑊 has been specifically obtained from the bound state 𝜓0 , one obtains Eq. (4). Note that since 𝜖𝑓 → 𝜖 for 𝐶 → + ∞ , the fundamental structure is itself a member of this family. The modes 𝜓𝑓 of any other 𝜖𝑓 can be found by transforming the states 𝜓 of 𝜖 according to
$$\psi _ { f } \varpropto \left ( \partial _ { x } - W _ { f } \right ) ( \partial _ { x } + W ) \psi. \quad \text{(S3)}$$
Based on this relation, which holds for all bound modes as well as scattered states, one can show that the family 𝜖𝑓 of iso-phase potentials exhibit identical reflection- and transmission coefficients, down to the phase, as the fundamental structure 𝜖 .
## S2. Reflection/transmission coefficients of superpartner structures
In order to relate the scattering characteristics of a superpartner to that of its associated fundamental structure, let us first consider an incident plane wave described by exp( 𝑖𝑋𝑛0 cos 𝜃 + 𝑖𝑌𝑛0 sin 𝜃 ) impinging on both structures from the left side. The reflected and transmitted waves in the fundamental system are then described by 𝑟 exp( -𝑖𝑋𝑛0 cos 𝜃 + 𝑖𝑌𝑛0 sin 𝜃 ) and 𝑡 exp( 𝑖𝑋𝑛0 cos 𝜃 + 𝑖𝑌𝑛0 sin 𝜃 ) , respectively. Accordingly, the corresponding waves in the superpartner geometry are given by 𝑟 𝑝 exp( -𝑖𝑋𝑛0 cos 𝜃 + 𝑖𝑌𝑛0 sin 𝜃 ) and 𝑡 𝑝 exp( 𝑖𝑋𝑛0 cos 𝜃 + 𝑖𝑌𝑛0 sin 𝜃 ) . By applying the relation 𝜓𝑝 ∝ ( ∂𝑋 + 𝑊 𝜓 ) (that holds between the wave functions of any state pairs in the two structures) to these radiation states, one finds
$$\left ( e ^ { + i X n _ { 0 } \cos \theta } + r _ { p } e ^ { - i X n _ { 0 } \cos \theta } \right ) \in \left ( ( + i n _ { 0 } \cos \theta + W _ { - } ) e ^ { + i X n _ { 0 } \cos \theta } + ( - i n _ { 0 } \cos \theta + W _ { - } ) r e ^ { - i X n _ { 0 } \cos \theta } \right ),$$
for 𝑋 → -∞ , and
$$t _ { p } e ^ { + i X n _ { 0 } \cos \theta } \in ( + i n _ { 0 } \cos \theta + W _ { + } ) t e ^ { + i X n _ { 0 } \cos \theta }, \quad ( S 5 )$$
for 𝑋 → + ∞ . In these two equations 𝑊 ± denotes the limit of 𝑊 at 𝑋 → ± ∞ respectively. Taking into account that both equations should have the same proportionality constant, one can show that the reflection and transmission coefficients are related via:
$$r _ { p } = \frac { W _ { - } - i n _ { 0 } \cos \theta } { W _ { - } + i n _ { 0 } \cos \theta } r, \quad \ \ ( S 6. a )$$
$$t _ { p } = \frac { W _ { + } + i n _ { 0 } \cos \theta } { W _ { - } + i n _ { 0 } \cos \theta } t. \quad \quad ( S 6. b )$$
Obviously, the reflectivity 𝑅 = | 𝑟 | 2 = | 𝑟 𝑝 | 2 as well as the transmittivity 𝑇 = 1 -𝑅 = | | 𝑡 2 = | 𝑡 𝑝 | 2 of the superpartner structures is identical. The specific relations for the complex coefficients given in Table 1 in turn follow from 𝑊+ = -𝑊in the case of unbroken supersymmetry and 𝑊+ = 𝑊in the broken supersymmetry regime.
To derive similar expressions for the family of iso-phase structures 𝜖𝑓 ( 𝑋 𝐶 ; ) , we label reflected and transmitted waves as 𝑟 𝑓 exp( -𝑖𝑋𝑛0 cos 𝜃 + 𝑖𝑌𝑛0 sin 𝜃 ) and 𝑡 𝑓 exp( 𝑖𝑋𝑛0 cos 𝜃 + 𝑖𝑌𝑛0 sin 𝜃 ) respectively. One can then relate 𝑟 𝑓 and 𝑡 𝑓 to 𝑟 and 𝑡 via the transformation 𝜓𝑓 ∝ �∂𝑋 -𝑊𝑓� ∂𝑋 ( + 𝑊 𝜓 ) . On the other hand, one can write the family of associated superpotentials as
$$W _ { f } ( X ; C ) = W + \partial _ { X } \ln \left ( C + \int _ { - \infty } ^ { X } \psi _ { 0 } ^ { 2 } ( X ^ { \prime } ) d X ^ { \prime } \right ), \quad ( S 7 )$$
by applying Eq. (5) of the main text and using the fact that 𝑊 can be written as 𝑊 = -∂𝑋 ln( 𝜓0 ) in terms of the ground state of the fundamental structure. Consequently, 𝑊𝑓 and 𝑊 have the same asymptotic behavior, i.e., 𝑊𝑓 ,± = 𝑊 ± . In the far field 𝑋 → ± ∞ , the transformation simplifies to 𝜓𝑓 ∝ ( ∂𝑋 - 𝑊 )( ∂𝑋 + 𝑊 𝜓 ) . Note that ( ∂𝑋 - 𝑊 )( ∂𝑋 + 𝑊 ) = ( ∂𝑋𝑋 + 𝑊 ′ -𝑊 2 ) = 𝐻 - 𝛼 , and therefore 𝜓𝑓 ∝ ( 𝐻 - 𝛼 𝜓 ) = ( 𝛺 - 𝛼 𝜓 ) . Since the scattered waves depend on 𝑌 according to exp( 𝑖𝑌𝑛0 sin 𝜃 ) , the corresponding eigenvalue in the Helmholtz equation is given by 𝛺 = 𝑛0 2 sin 2 𝜃 . Hence, 𝜓𝑓 ∝ ( 𝑛0 2 sin 2 𝜃 - 𝛼 𝜓 ) , and therefore:
$$( e ^ { + i X n _ { 0 } \cos \theta } + r _ { f } e ^ { - i X n _ { 0 } \cos \theta } ) \in ( n _ { 0 } ^ { 2 } \sin ^ { 2 } \theta - \alpha ) ( e ^ { + i X n _ { 0 } \cos \theta } + r e ^ { - i X n _ { 0 } \cos \theta } ),$$
for 𝑋 → -∞ , and
$$t _ { f } e ^ { + i X n _ { 0 } \cos \theta } \in ( n _ { 0 } ^ { 2 } \sin ^ { 2 } \theta - \alpha ) t e ^ { + i X n _ { 0 } \cos \theta },$$
for 𝑋 → + ∞ . Given that both equations should be normalized with respect to the same constant, it directly follows that in the iso-phase scenario
$$r _ { f } = r, \quad \ \ ( S 1 0. a )$$
$$t _ { f } = t. \quad \ \ ( S 1 0. b )$$
## S3. Configurations with dissimilar backgrounds
Even though the formalism of the main text was developed by assuming that the background medium is the same ( 𝑛-= 𝑛+ = 𝑛0 ), superpartners obeying Eqs. (2,3) can be generated even if 𝑛- ≠ 𝑛+ (here 𝑛 ± = �𝜖 (± ∞ ) and ). Figure S1(a) shows an example of a step-like distribution in the relative permittivity; the corresponding superpartner 𝜖𝑝 is
depicted in Fig. S1(b). The superpotential mediating between them no longer has the same absolute value at 𝑋 → ± ∞ , but rather is shifted by a certain offset related to the height of the potential step. Neither of the two partner structures does support any guided modes, therefore supersymmetry is necessarily broken. As depicted in Fig. S1(c), numerical results obtained by differential transfer matrix method [S2] show that the angle-dependent power reflectivities 𝑅 = | 𝑟 | 2 = | 𝑟 𝑝 | 2 and transmittivities 𝑇 = 1 -𝑅 = | | 𝑡 2 ⋅ ( 𝑛+ / 𝑛)(cos 𝜃+ / cos 𝜃-) = | 𝑡 𝑝 | 2 ⋅ ( 𝑛+ / 𝑛)(cos 𝜃+ / cos 𝜃-) remain identical for both systems (Fig. S1(c)), while the phases do not (Fig. S1(d)).
Analytical relations of the scattering coefficients can be found by following analogous steps to those in the previous section. The results are
$$r _ { p } = \frac { W _ { - } - i n _ { - } \cos \theta _ { - } } { W _ { - } + i n _ { - } \cos \theta _ { - } } r,$$
$$r _ { p } = & \, \frac { W _ { - } - i n _ { - } \cos \theta _ { - } } { W _ { - } + i n _ { - } \cos \theta _ { - } } r, \quad \ ( S 1 1. a ) \\ t _ { p } = & \, \frac { W _ { + } + i n _ { + } \cos \theta _ { + } } { W _ { - } + i n _ { - } \cos \theta _ { - } } t, \quad \ ( S 1 1. b ) \\ \intertext { t and trancmitted wave on alloc related from}$$
where 𝜃and 𝜃+ represent the incident and transmitted wave angles related through Snell's law 𝑛sin 𝜃-= 𝑛+ sin 𝜃+ . It follows that | 𝑟 | 2 = | 𝑟 𝑝 | 2 and | 𝑡 | 2 = | 𝑡 𝑝 | 2 .
Fig. S1. (a) Step-like potential 𝜖 𝑋 ( ) = 2 + tanh(X/0.3) and (b) its superpartner 𝜖𝑝 ( 𝑋 ) . (c) Despite the potential offset, the intensities of reflected and transmitted waves are identical for both structures, (d) but the phases remain different as expected for the broken SUSY regime.
## S4. Inverse SUSY transformation
In the unbroken symmetry regime, the conventional SUSY transformation may remove a mode from a given fundamental structure 𝜖 . In doing so, the total area of the relative permittivity is reduced (see Fig. 2(a,b)). This can be shown easily by noting that in the unbroken supersymmetry regime the two superpartners are related via
$$\epsilon _ { p } - \epsilon = - 2 W ^ { \prime }. \quad \text{(S12)}$$
After integrating both sides of this equation we get
$$\int _ { - \infty } ^ { + \infty } \epsilon _ { p } ( X ) d X - \int _ { - \infty } ^ { + \infty } \epsilon ( X ) d X = - 2 ( W _ { + } - W _ { - } ) \,. \quad \ \ ( S 1 3 )$$
Since in the unbroken supersymmetry regime 𝑊- ≠ 𝑊 + , the SUSY transformation cannot preserve the total area of the relative permittivity distribution.
On the other hand, one can utilize an inverse SUSY transformation and add a bound state to a given structure 𝜖 , and in doing so elevate the total area of a given permittivity distribution. We factorize the fundamental Hamiltonian as 𝐻 = 𝐴𝐵 + 𝛼 and define the partner Hamiltonian as 𝐻elev = 𝐵𝐴 + 𝛼 . Consequently, the two superpartner permittivity distributions can be written as:
$$\epsilon ( X ) = - W ^ { \prime } _ { e l e v } - W ^ { 2 } _ { e l e v } + \alpha, \quad ( S 1 4. a )$$
$$\epsilon _ { \text{elev} } ( X ) = + W _ { \text{elev} } ^ { \prime } - W _ { \text{elev} } ^ { 2 } + \alpha. \ \ ( S 1 4. b )$$
Equation (S14.a) can be solved numerically to obtain the superpotential 𝑊elev , and from that the partner structure 𝜖elev can be constructed through Eq. (S14.b). As expected in this case the partner structure 𝜖elev exhibits all the guided mode eigenvalue spectrum of the fundamental structure 𝜖 , as well as an additional guided mode, which takes the place of its previous ground state. As it turns out, the eigenvalue of this additional state is given by the factorization parameter 𝛼 . Note that any value 𝛼 > 𝛺0 can be chosen, where 𝛺0 represents the ground state eigenvalue of the fundamental structure. Figure S2 illustrates the inverse SUSY transformation for the specific example of the fundamental structure discussed in Fig. 2 of the main text.
Fig. S2. Inverse SUSY transformation. (a) Step-index-like fundamental structure 𝜖 𝑋 ( ) supporting 𝑁 = 3 guided modes. (b) Its inverse superpartner 𝜖elev ( 𝑋 ) , supporting a phase matched mode associated with each of the fundamental structure's states as well as an additional with the desired eigenvalue α = 2.5 . (c) Superpotential relating the two structures.
## S5. SUSY formalism for the TM polarization
Under TM polarization conditions the magnetic field component satisfies the equation:
$$\left ( \partial _ { x x } + \partial _ { y y } - \partial _ { x } ( \ln \epsilon ) \partial _ { x } + k _ { 0 } ^ { 2 } \epsilon ( x ) \right ) H _ { z } ( x, y ) = 0. \quad ( S 1 5 )$$
In this case again by considering stationary solutions of the form 𝐻𝑧 = 𝜓 𝑥 𝑒 ( ) 𝑖𝛽𝑧 and after using normalized coordinates 𝑋 = 𝑘0𝑥 𝑌 , = 𝑘0𝑦 and assuming 𝛺 = 𝛽 2 / 𝑘0 2 we reach at:
$$( \partial _ { X X } - \partial _ { X } ( \ln \epsilon ) \partial _ { X } + \epsilon ) \psi = \Omega \psi. \quad ( S 1 6 )$$
As is, this is not a Schrödinger-like equation, and hence the factorization technique cannot be directly applied. On the other hand, by using the transformation 𝜓 𝑋 ( ) = √𝜖𝑓 𝑋 ( ) , this equation can be converted to the desired form:
$$( \partial _ { X X } + V _ { \text{eff} } ) f = \Omega f, \quad ( S 1 7 )$$
where 𝑉eff is an effective potential that can be expressed in terms of the relative permittivity 𝜖 as 𝑉eff = 𝜖 3 4 ( 𝜖 /𝜖 ′ ) 2 + 1 2 ( 𝜖 ′′ 𝜖 / ) . This relation can also be rewritten as:
$$V _ { \text{eff} } = \epsilon + \left ( \frac { \epsilon ^ { \prime } } { 2 \epsilon } \right ) ^ { \prime } - \left ( \frac { \epsilon ^ { \prime } } { 2 \epsilon } \right ) ^ { 2 }. \quad ( S 1 8 )$$
Following the SUSY formalism, the two superpartner effective potentials can now be written in terms of the superpotential 𝑊 via
$$V _ { \text{eff} } ( X ) = + W ^ { \prime } - W ^ { 2 } + \alpha, \quad \ \ ( S 1 9. a )$$
$$V _ { \text{eff}, p } ( X ) = - W ^ { \prime } - W ^ { 2 } + \alpha \,. \quad ( S 1 9. b )$$
One can then reconstruct the relative permittivity of the partner structure 𝜖𝑝 from its corresponding effective potential 𝑉eff 𝑝 , by numerically solving the nonlinear equation
$$V _ { \text{eff}, p } = \epsilon _ { p } + \left ( \frac { \epsilon _ { p } ^ { \prime } } { 2 \epsilon _ { p } } \right ) ^ { \prime } - \left ( \frac { \epsilon _ { p } ^ { \prime } } { 2 \epsilon _ { p } } \right ) ^ { 2 }. \quad ( S 2 0 )$$
Of importance will be to show that these transformations preserve the reflection/transmission properties of SUSY partners in the TM case. For this reason note that for effective potentials described in Eqs. (S19), we have the intervening relations 𝑓 𝑝 ∝ 𝜕𝑋 ( + 𝑊 𝑓 ) and 𝑓 ∝ ( 𝜕𝑋 - 𝑊 𝑓 𝑝 ) therefore:
$$\psi _ { p } / \sqrt { \epsilon _ { p } } \, \in \, ( \partial _ { X } + W ) \, \psi / \sqrt { \epsilon }, \quad ( S 2 1. a )$$
$$\psi / \sqrt { \epsilon } \, \infty \, ( \partial _ { X } - W ) \, \psi _ { p } / \sqrt { \epsilon _ { p } } \,. \quad ( S 2 1. b )$$
Obviously presence of the relative permittivities in the denominator makes these last equations different than those of the TE case. However for calculating the reflection/transmission coefficients (see Supplementary section S2) only the asymptotic behavior of these equations at 𝑋 → ± ∞ are of our interest. On the other hand, according to Eqs. (S18-S20), the SUSY partners have the same asymptotic behaviors. Therefore one can use the exact same analysis of the Supplementary section S2 based on Eqs. (S21) to show that the reflection/transmission coefficients of SUSY partners constructed for the TM polarization are related via equations given in Table. 1 of the main text.
## Supplementary references
S1. F. Cooper, A. Khare, and U. Sukhatme, "Supersymmetry and quantum mechanics," Phys. Rep. 251 , 267-385 (1995). S2. P. Yeh, " Optical Waves in Layered Media " (Wiley, New York, 1988).
|
10.1364/OPTICA.1.000089
|
[
"Mohammad-Ali Miri",
"Matthias Heinrich",
"Demetrios N. Christodoulides"
] |
2014-08-04T22:28:04+00:00
|
2014-08-04T22:28:04+00:00
|
[
"physics.optics",
"math-ph",
"math.MP"
] |
SUSY-inspired one-dimensional transformation optics
|
Transformation optics aims to identify artificial materials and structures
with desired electromagnetic properties by means of pertinent coordinate
transformations. In general, such schemes are meant to appropriately tailor the
constitutive parameters of metamaterials in order to control the trajectory of
light in two and three dimensions. Here we introduce a new class of
one-dimensional optical transformations that exploits the mathematical
framework of supersymmetry (SUSY). This systematic approach can be utilized to
synthesize photonic configurations with identical reflection and transmission
characteristics, down to the phase, for all incident angles, thus rendering
them perfectly indistinguishable to an external observer. Along these lines,
low-contrast dielectric arrangements can be designed to fully mimic the
behavior of a given high-contrast structure that would have been otherwise
beyond the reach of available materials and existing fabrication techniques.
Similar strategies can also be adopted to replace negative-permittivity
domains, thus averting unwanted optical losses.
|
1408.0833v1
|
A temporal perspective on the rate of convergence in first-passage percolation under a moment condition
Daniel Ahlberg ∗†
August 6, 2014
## Abstract
We study the rate of convergence in the Shape Theorem of first-passage percolation, obtaining the precise asymptotic rate of decay for the probability of linear order deviations under a moment condition. Our results are stated for a given time and complements recent work by the same author, in which the rate of convergence was studied from the standard spatial perspective.
## 1 Introduction
Consider first-passage percolation on the Z d nearest-neighbour graph for d ≥ 2. Large deviations were first studied in the context of first-passage percolation in the 1980s, in a pioneering work of Grimmett and Kesten [3]. In this work, together with the subsequent work of Kesten [4], the authors investigate the rate of convergence of travel times to the so-called time constant, and provide necessary and sufficient conditions for exponential decay for the probability of linear order deviations. To obtain the exponential decay one requires finite moment of exponential order of the passage times.
It was only recently, in [1], that large deviations in the regime of polynomial decay of the probability tails were studied. This regime is highly interesting since it is in this regime that strong laws such as the Subadditive Ergodic Theorem, due to Kingman [5], and the Shape Theorem, whose precise conditions were found by Cox and Durrett [2], cease to hold. In this paper we build on this recent work and offer a temporal perspective on the rate of convergence in the Shape Theorem. We complement the results of [1] with estimates on the probability of linear order deviations at a given time. The results are sharp in the regime of polynomial decay.
We will assume throughout that the edges of the Z d lattice are assigned independent non-negative random weights from a distribution satisfying F (0) < p c ( d ). Here p c ( d ) denotes the critical probability for bond percolation on Z d . The weights induce a random metric structure to Z d , and we denote the distance metric by T x, y ( ) for x, y ∈ Z d . Let
$$\mu ( z ) \coloneqq \lim _ { n \to \infty } \frac { T ( 0, n z ) } { n },$$
∗ IMPA, Estrada Dona Castorina 110, 22460-320 Rio de Janeiro, Brazil.
† Department of Mathematics, Uppsala University, SE-75106 Uppsala, Sweden.
which is known to exist in probability without further assumption on the edge weights. The limit µ ( ) is usually known as the · time constant . Richardson [6] was the first to realize that the above convergence holds in all directions simultaneously, in that B t := { z ∈ Z d : T (0 , z ) ≤ } t obeys a law of large numbers.
/negationslash
Our assumption on the edge weights are known to imply that µ z ( ) = 0 for all z = 0 (see [4, Theorem 6.1]). Consequently, the deterministic set B µ t := { z ∈ Z d : µ z ( ) ≤ } t is finite for all t ≥ 0. Cox and Durrett [2] provided the precise condition for Richardson's result to hold. Let Y denote the minimum of the 2 d weights assigned to the edges adjacent to the origin. The result of Cox and Durrett says that if E [ Y d ] < ∞ , then, almost surely, for every ε ∈ (0 1) ,
/negationslash
$$\mathcal { B } _ { ( 1 - \varepsilon ) t } ^ { \mu } \subset \mathcal { B } _ { t } \subset \mathcal { B } _ { ( 1 + \varepsilon ) t } ^ { \mu }$$
for large enough t . The given moment condition is also necessary for this conclusion.
In this paper we aim to investigate in further detail the probability that (1) fails, and that B t deviates significantly from its asymptotic rate of growth. Let
$$\mathcal { T } _ { \varepsilon } \coloneqq \{ t \geq 0 \, \colon \text{either inclusion in } ( 1 ) \, \text{ fails} \}.$$
We will study the behaviour of P ( t ∈ T ε ) for large t and fixed ε ∈ (0 1). , Cox and Durrett's result implies that T ε is almost surely bounded if and only if E [ Y d ] < ∞ . In fact, P ( t ∈ T ε ) may even be bounded away from zero unless E [ Y d ] < ∞ , as we remark upon at the very end. Instead we assume that E [ Y d ] < ∞ holds, and provide the following results. The first result shows that in the regime of polynomial decay of the tails, P ( t ∈ T ε ) is governed by the tails of Y .
Theorem. Assume that d ≥ 2 , F (0) < p c ( d ) and E [ Y d ] < ∞ . Then, for every ε > 0 and q ≥ 1 , there is M = M ε,d,q ( ) such that
$$M ^ { - 1 } \, t ^ { d } \, \mathbb { P } ( Y > t ) \, \leq \, \mathbb { P } ( t \in \mathcal { T } _ { s } ) \, \leq \, M \, t ^ { d } \, \mathbb { P } ( Y > t / M ) + M \, t ^ { - q } \ \text{ for } t \geq 1.$$
Using Markov's inequality we find that for each α > 0, the condition E [ Y d + α ] < ∞ implies that P ( t ∈ T ε ) = O t ( -α ). Another consequence of the theorem is the following characterization.
Corollary. Assume that F (0) < p c ( d ) . For every α ≥ 0 , ε > 0 and d ≥ 2 ,
$$\mathbb { E } [ Y ^ { d + \alpha } ] < \infty \quad \Leftrightarrow \quad \int _ { 0 } ^ { \infty } t ^ { \alpha - 1 } \, \mathbb { P } ( t \in \mathcal { T } _ { \varepsilon } ) \, d t < \infty.$$
Using Fubini's theorem we see that
$$\mathbb { E } \left | \mathcal { T } _ { \varepsilon } \right | \, = \, \mathbb { E } \int _ { 0 } ^ { \infty } 1 _ { \{ t \in \mathcal { T } _ { \varepsilon } \} } \, d t \, = \, \int _ { 0 } ^ { \infty } \mathbb { P } ( t \in \mathcal { T } _ { \varepsilon } ) \, d t,$$
where | · | denotes Lebesgue measure. Hence the conclusion of the corollary was for α = 1 known already in [1, Theorem 2], but the statement for general α > 0 was previously unknown. The proofs we present below will be based on results and ideas from [1].
## 2 Proof
We first recall some results that will be required for the analysis: For every ε > 0 and d ≥ 2 there exists constants M = M ε,d ( ) and γ = γ ε, d ( ) such that
$$\mathbb { P } ( T ( 0, z ) - \mu ( z ) < - \varepsilon x ) \, \leq \, M \, e ^ { - \gamma x } \quad \text{for all } z \in \mathbb { Z } ^ { d } \, \text{and } x \geq | z |. \quad \text{(2)} \\ \intertext { f in addition } \mathbb { R } [ V ^ { \alpha } ] \, \times \, \text{for comma } \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \kappa \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \alpha \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \kappa \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda. \quad \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \eta \, \kappa \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \,
\lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \eta \, \kappa \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda. \quad \, \lambda. \quad \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda. \quad \, \lambda
\, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \xi \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \varphi \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \lambda \, \skip \, 2 } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } _ } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } { } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } }, } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } \ } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } ; } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } }$$
If, in addition, E [ Y α ] < ∞ for some α > 0 and q ≥ 1, then we may choose M = M α,ε,d,q ( ) such that
$$\mathbb { P } ( T ( 0, z ) - \mu ( z ) > \varepsilon x ) & \leq M \, \mathbb { P } ( Y > x / M ) + M \, t ^ { - q } \quad \text{for all $z\in\mathbb{Z}^{d}$ and $x\geq |z|$.} \quad (3) \\ \intertext { T h o f m o w e c t o m o n t m o n t s o f r e t m o n d b r o f C i m m o n t t $ o f $ a $ a $ f o r o c o n d i n t o }$$
The former statement was first proved by Grimmett and Kesten [3, 4] for coordinate directions, and later extended in [1]. The latter statement is original in [1].
Fix ε > 0 and q ≥ 1. We wish to estimate the decay of P ( t ∈ T ε ). Define thus
$$A ( z ) \, \coloneqq \, \{ t \geq 0 \, \colon T ( 0, z ) > t \text{ and } \mu ( z ) \leq ( 1 - \varepsilon ) t \},$$
$$B ( z ) \, \coloneqq \, \{ t \geq 0 \, \colon T ( 0, z ) \leq t \text{ and } \mu ( z ) > ( 1 + \varepsilon ) t \}.$$
$$\begin{matrix} 2 \\ 3 \end{matrix}$$
Note that t ∈ T ε if and only if t ∈ A z ( ) ∪ B z ( ) for some z ∈ Z d .
## 2.1 The lower bound
We begin with the lower bound. Let D denote the set of all z ∈ Z d such that µ z ( ) ≤ (1 -ε t ) and whose /lscript 1 -distance from the origin is even. Let Y ( z ) denote the minimum weight among the 2 d edges adjacent to z , and note that the Y ( z )'s are independent for z ∈ D , as points in D are at /lscript 1 -distance at least 2. Clearly A z ( ) ⊂ T ε for every z ∈ Z d . Since also T (0 , z ) ≥ Y ( z ) we obtain
$$\mathbb { P } ( t \in \mathcal { T } _ { \varepsilon } ) & \, \geq \, \mathbb { P } ( T ( 0, z ) > t \, \text{for some} \, z \in D ) \\ & \, \geq \, \mathbb { P } ( Y ( z ) > t \, \text{for some} \, z \in D ). \\ \text{lication of Cachv-Schwarz's inequality shows that any non-negative random}$$
An application of Cachy-Schwarz's inequality shows that any non-negative random variable X satisfies P ( X > 0) ≥ E [ X / ] 2 E [ X 2 ], and if X is binomially distributed with parameters n and p , then a further lower bound is given by np/ (1 + np ). Applying this to (4) leaves us with
$$\Phi ( t \in \mathcal { T } _ { \varepsilon } ) \, \geq \, \frac { | D | \, \mathbb { P } ( Y > t ) } { 1 + | D | \, \mathbb { P } ( Y > t ) }.$$
By assumption E [ Y d ] < ∞ , which implies that t d P ( Y > t ) ≤ E [ Y d ] via Markov's inequality. As the set D grows as t d , we obtain the required lower bound on P ( t ∈ T ε ).
## 2.2 The upper bound
We now continue with the upper bound. The union bound leaves us with the upper bound
$$\mathbb { P } ( t \in \mathcal { T } _ { \varepsilon } ) \, \leq \, \sum _ { \mu ( z ) \leq ( 1 - \varepsilon ) t } \mathbb { P } ( T ( 0, z ) > t ) \, + \, \sum _ { \mu ( z ) > ( 1 + \varepsilon ) t } \mathbb { P } ( T ( 0, z ) \leq t ).$$
We treat the two sums separately. Using (2) we find M 1 = M ε,d 1 ( ) and γ 1 = γ 1 ( ε, d ) such that
$$\sum _ { \mu ( z ) > ( 1 + \varepsilon ) t } \mathbb { P } ( T ( 0, z ) \leq t ) \ & \leq \sum _ { \mu ( z ) \geq ( 1 + \varepsilon ) t } \mathbb { P } \left ( T ( 0, z ) - \mu ( z ) \leq - \frac { \varepsilon } { 1 + \varepsilon } \mu ( z ) \right ) \\ & \leq \sum _ { \mu ( z ) \geq t } M _ { 1 } \, e ^ { - \gamma _ { 1 } \mu ( z ) },$$
which is at most M e 2 -γ t 2 for t ≥ 1, for some constants M 2 and γ 2 . Using (3) we may find M 3 = M ε,d,q 3 ( ) such that
$$\sum _ { \mu ( z ) \leq ( 1 - \varepsilon ) t } \mathbb { P } ( T ( 0, z ) > t ) & \leq \sum _ { \mu ( z ) \leq ( 1 - \varepsilon ) t } \mathbb { P } ( T ( 0, z ) - \mu ( z ) > \varepsilon t ) \\ & \leq \sum _ { \mu ( z ) \leq t } M _ { 3 } \, \mathbb { P } ( Y \geq t / M _ { 3 } ) + \frac { M _ { 3 } } { t ^ { d + q } }.$$
Since the cardinality of the set B µ t grows at the order of t d we obtain, for some M 4 ,
$$\mathbb { P } ( t \in \mathcal { T } _ { \varepsilon } ) \, \leq \, M _ { 4 } \, t ^ { d } \, \mathbb { P } ( Y \geq t / M _ { 3 } ) + M _ { 4 } \, t ^ { - q } + M _ { 2 } \, e ^ { \gamma _ { 2 } t },$$
for all t ≥ 1, which gives the required upper bound.
Remark. Finally, we mention that the condition E [ Y d ] < ∞ cannot be relaxed in general. Consider for instance the case when P ( Y > t ) = t -d for t ≥ 1. Then E [ Y α ] is finite for all α < d and infinite for α = d , but the bound in (5) shows that P ( t ∈ T ε ) is bounded away from zero.
## References
- [1] D. Ahlberg. A Hsu-Robbins-Erd˝ os strong law in first-passage percolation. Ann. Probab. , to appear.
- [2] J. T. Cox and R. Durrett. Some limit theorems for percolation processes with necessary and sufficient conditions. Ann. Probab. , 9(4):583-603, 1981.
- [3] G. Grimmett and H. Kesten. First-passage percolation, network flows and electrical resistances. Z. Wahrsch. Verw. Gebiete , 66(3):335-366, 1984.
- [4] H. Kesten. Aspects of first passage percolation. In ´ cole d'´t´ de probabilit´s de SaintE e e e Flour, XIV-1984 , volume 1180 of Lecture Notes in Math. , pages 125-264. Springer, Berlin, 1986.
- [5] J. F. C. Kingman. The ergodic theory of subadditive stochastic processes. J. Roy. Statist. Soc. Ser. B , 30:499-510, 1968.
- [6] D. Richardson. Random growth in a tessellation. Proc. Cambridge Philos. Soc. , 74:515528, 1973.
| null |
[
"Daniel Ahlberg"
] |
2014-08-04T22:48:02+00:00
|
2014-08-04T22:48:02+00:00
|
[
"math.PR",
"60K35"
] |
A temporal perspective on the rate of convergence in first-passage percolation under a moment condition
|
We study the rate of convergence in the Shape Theorem of first-passage
percolation, obtaining the precise asymptotic rate of decay for the probability
of linear order deviations under a moment condition. Our results are stated for
a given time and complements recent work by the same author, in which the rate
of convergence was studied from the standard spatial perspective.
|
1408.0834v1
|
## Multichannel quantum defect theory for ro-vibrational transitions in ultracold molecule-molecule collisions
Jisha Hazra
Department of Chemistry, University of Nevada Las Vegas, Las Vegas, Nevada 89154
Brandon P. Ruzic
JILA, NIST, and Department of Physics, University of Colorado, Boulder, Colorado 80309, USA
## N. Balakrishnan
Department of Chemistry, University of Nevada Las Vegas, Las Vegas, Nevada 89154
John L. Bohn
University of Colorado, Boulder,
JILA, NIST, and Department of Physics, Colorado 80309, USA (Dated: August 6, 2014)
## Abstract
Multichannel quantum defect theory (MQDT) has been widely applied to resonant and nonresonant scattering in a variety of atomic collision processes. In recent years, the method has been applied to cold collisions with considerable success, and it has proven to be a computationally viable alternative to full-close coupling (CC) calculations when spin, hyperfine and external field effects are included. In this paper, we describe a hybrid approach for molecule-molecule scattering that includes the simplicity of MQDT while treating the short-range interaction explicitly using CC calculations. This hybrid approach, demonstrated for H -H 2 2 collisions in full-dimensionality, is shown to adequately reproduce cross sections for quasi-resonant rotational and vibrational transitions in the ultracold (1 µ K) and ∼ 1-10 K regime spanning seven orders of magnitude. It is further shown that an energy-independent short-range K -matrix evaluated in the ultracold regime (1 µ K) can adequately characterize cross sections in the mK-K regime when no shape resonances are present. The hybrid CC-MQDT formalism provides an alternative approach to full CC calculations at considerably less computational expense for cold and ultracold molecular scattering.
PACS numbers: 34.50.-s, 67.85.-d
## I. INTRODUCTION
Molecules in a translationally cold gas present a particular perspective on collisions and chemistry. One the one hand, atoms in the colliding molecules exchange energy on the scale of tens to thousands of Kelvin, driven by deep potential energy surfaces. On such surfaces occur rotational, vibrational, and chemical transformations. On the other hand, the ability of the molecules to initiate this activity is strongly dependent on behavior at the µ K - mK translational energy scales of the gas. The slowly moving molecules, to get close enough to react, must first negotiate their way through the long-range forces acting between them. These forces, negligible at room temperature, loom large in the ultracold. The dominance of long-rage forces had led to control over chemical reaction, by, for example, the simple expedient of applying a modest electric field to alter the dipole moments of molecules [1]. For this reason cold molecules are seen as novel tools for probing and controlling chemistry with unprecedented resolution [2].
This dichotomy of energy scales presents a unique point of view for theories of molecules interacting at ultracold temperatures, which must now account for dynamics occurring over many orders of magnitude in energy. Luckily, the energy dichotomy relates in a direct way to motion on disparate spatial scales. Specifically, the full, energy-sharing dynamics of atoms in the collision complex occurs where all participating atoms are close together, whereas the long-range dynamics occurs between well-delineated collision partners that are far apart. The business of cold collision theory is to accurately account for the relatively straightforward long-range dynamics, while incorporating, to the extent desirable or reasonable, the shortrange dynamics.
The separation into short- and long-range physics finds its natural expression in the multichannel quantum defect theory (MQDT), whose origins go back to understanding spectra of Rydberg atoms [3-5], but which has been successfully extended to more general contexts [6, 7], including cold collisions of atoms [8-12], atoms and ions [13, 14], atoms and molecules [15, 16], and molecule-molecule reactive scattering [17-20]. In all cases, long-range wave functions are carefully constructed and then matched to a wave function that is a suitable representation of the short-range physics. Depending on the context, the short-range physics can be successfully treated in a schematic way by (for example) positing absorbing boundary conditions to represent chemical reactions [17, 18, 20] or, in the case of alkali atom cold
collisions, by means of simple spin-dependent phase shifts [8, 21].
In this article we tackle head-on the complete short-range dynamics of molecule-molecule scattering for the comparatively straightforward case of H +H 2 2 collisions at collision energies ≤ 10 K, where comparison with numerically accurate scattering calculations can be made. A main finding is that the MQDT approach can be accurate and considerably more efficient numerically, provided sufficient care is taken in constructing the long-range wave functions. Thus short-range and long-range dynamics can be successfully welded together in this important prototype case where energy can be exchanged between rotational and vibrational degrees of freedom of two molecules. The calculations presented here represent a first, necessary step toward adapting MQDT methods to the broader problem of cold chemistry, which should ultimately lead to understanding how to manipulate reaction dynamics in realistic ultracold gases.
The paper is organized as follows. In section II we present in detail the close-coupling (CC) and MQDT formalisms for non-reactive scattering in collisions between two 1 Σ molecules. In section III we provide numerical illustration of the method for quasi-resonant rotational and vibrational transitions in H -H 2 2 collisions, including both ortho and para symmetries. Conclusions and future directions are presented in section IV.
## II. THEORY
## A. Quantum close-coupling approach for molecule-molecule collisions
The molecule-molecule scattering theory for collisions of two 1 Σ diatomic molecules has been well established and described in detail in many prior works [23-27]. Only a brief description to introduce the key terminologies and set the stage for the MQDT formalism is given here. The full close coupling (CC) [28] methodology based on the solution of the time-independent Schr¨dinger equation is used to solve the four-body scattering problem o in Jacobi coordinates. After elimination of center-of-mass motion, the Hamiltonian for the relative motion of two H 2 molecules in space-fixed coordinates may be written as
$$\hat { H } = - \frac { \hbar { ^ { 2 } } } { 2 \mu } \frac { \partial ^ { 2 } } { \partial R ^ { 2 } } + \frac { \hat { \ell } ^ { 2 } } { 2 \mu R ^ { 2 } } + \sum _ { i = 1 } ^ { 2 } \hat { h } _ { i } ( r _ { i } ) + U ( \mathbf r _ { 1 }, \mathbf r _ { 2 }, \mathbf R )$$
where R is the vector joining the center of mass of the two H 2 molecules, µ and ˆ are glyph[lscript] the reduced mass and orbital angular momentum of the two colliding H 2 molecules and U ( r 1 , r 2 , R ) is the interaction potential. The terms, ˆ ( h i r i ) , i = 1 2 are the Hamiltonians of , the two isolated H 2 molecules:
$$\hat { h } _ { i } ( r _ { i } ) = - \frac { \hslash ^ { 2 } } { 2 \mu _ { i } } \frac { \partial ^ { 2 } } { \partial r _ { i } ^ { 2 } } + \frac { \hat { j } _ { i } ^ { 2 } } { 2 \mu _ { i } r _ { i } ^ { 2 } } + V ( r _ { i } )$$
where r i , µ i , and ˆ j i are the internuclear separation, reduced mass, and the rotational angular momenta of the two separated H 2 molecules. The H -H 2 2 interaction potential is expanded in terms of coupled spherical harmonics [24]
$$U ( { \mathbf r } _ { 1 }, { \mathbf r } _ { 2 }, { \mathbf R } ) = \sum _ { \lambda } A _ { \lambda } ( { r } _ { 1 }, { r } _ { 2 }, { R } ) Y _ { \lambda } ( \hat { r } _ { 1 }, \hat { r } _ { 2 }, \hat { R } )$$
with
$$Y _ { \lambda } ( \hat { r } _ { 1 }, \hat { r } _ { 2 }, \hat { R } ) = \sum _ { m _ { \lambda } } \langle \lambda _ { 1 } m _ { \lambda _ { 1 } } \lambda _ { 2 } m _ { \lambda _ { 2 } } | \lambda _ { 1 2 } m _ { \lambda _ { 1 2 } } \rangle \times Y _ { \lambda _ { 1 } m _ { \lambda _ { 1 } } } ( \hat { r } _ { 1 } ) Y _ { \lambda _ { 2 } m _ { \lambda _ { 2 } } } ( \hat { r } _ { 2 } ) Y _ { \lambda _ { 1 2 } m _ { \lambda _ { 1 2 } } } ^ { * } ( \hat { R } ) \quad ( 4 )$$
where λ ≡ { λ , λ 1 2 , λ 12 } and m λ ≡ { m ,m λ 1 λ 2 , m λ 12 } . The indices λ 1 , λ 2 and λ 12 are nonnegative integers and the sum of these three quantities must be an even integer. The homonuclear symmetry of H 2 requires that λ 1 and λ 2 must be even. The quantity in angular brackets of the above equation is a Clebsch-Gordan coefficient, and Y λm λ are spherical harmonics. The Sch¨ rodinger equation is conveniently formulated by introducing the total angular momentum representation [28]. The total angular momentum -→ J = - → j 12 + -→ glyph[lscript] is the vector sum of total rotational angular momentum - → j 12 = -→ j 1 + -→ j 2 of the two molecules and orbital angular momentum -→ glyph[lscript] . Note that all molecules remain in singlet electronic spin states, so we suppress this notation in the following. For collisions between two indistinguishable molecules, the total wave function Ψ may be expanded in terms of rotational and vibrational wave functions of the two H 2 molecules, Φ JMglyph[epsilon1] I glyph[epsilon1] P vjglyph[lscript] , in the total angular momentum representation [28]:
$$\Psi ( { \mathbf r } _ { 1 }, { \mathbf r } _ { 2 }, { \mathbf R } ) = \frac { 1 } { R } \sum _ { v, j, \ell, J, M } F _ { v j \ell } ^ { J M \epsilon _ { I } \epsilon _ { P } } ( R ) \Phi _ { v j \ell } ^ { J M \epsilon _ { I } \epsilon _ { P } } ( { \mathbf r } _ { 1 }, { \mathbf r } _ { 2 }, { \mathbf R } )$$
where F JMglyph[epsilon1] I glyph[epsilon1] P vjglyph[lscript] ( R ) are the radial expansion coefficients, v ≡ v , v 1 2 represents the vibrational quantum numbers and j ≡ j , j 1 2 specifies the rotational quantum numbers of the two diatomic fragments. The quantity glyph[epsilon1] I = ( -1) j 1 + j 2 + glyph[lscript] is the eigenvalue of the spatial inversion
operator, and glyph[epsilon1] P is the eigenvalue of the exchange permutation symmetry operator for two H 2 molecules (for the indistinguishable case, e.g., para-para or ortho-ortho). The explicit expression for Φ JMglyph[epsilon1] I glyph[epsilon1] P vjglyph[lscript] is given in Eqs. (6), (8) and (15) of Ref. [27]. The radial expansion coefficients F JMglyph[epsilon1] I glyph[epsilon1] P vjglyph[lscript] are evaluated by solving the close-coupled radial equations in R ,
$$\Phi _ { \Phi } \Phi _ { \Phi } \left ( - \frac { \hbar { 2 } } { 2 \mu } \frac { d ^ { 2 } } { d R ^ { 2 } } & + \frac { \hbar { 2 } \ell ( \ell + 1 ) } { 2 \mu R ^ { 2 } } + \epsilon _ { v j } - E \right ) F _ { v j \ell } ^ { J M \epsilon _ { I } \epsilon _ { P } } ( R ) \\ & + \sum _ { v ^ { \prime }, j ^ { \prime }, \ell ^ { \prime } } \mathcal { U } _ { v j \ell, v ^ { \prime } j ^ { \prime } \ell ^ { \prime } } ^ { J M \epsilon _ { I } \epsilon _ { P } } ( R ) F _ { v ^ { \prime } j ^ { \prime } \ell ^ { \prime } } ^ { J M \epsilon _ { I } \epsilon _ { P } } ( R ) = 0 \\ & \quad \dots \quad \dots$$
resulting from substitution of Eqs. (1) and (5) in the time-independent Schr¨dinger equation o H Ψ = E Ψ. Here E is the total energy of the system, and we define the collision energy to be E c = E -glyph[epsilon1] v 1 j 1 + glyph[epsilon1] v 2 j 2 = E -glyph[epsilon1] vj . The symbol glyph[epsilon1] v j i i ( i = 1 2) denotes the asymptotic , ro-vibrational energies of the two H 2 molecules. Under molecule permutation the interaction potential, U JMglyph[epsilon1] I glyph[epsilon1] P vjglyph[lscript],v ′ j glyph[lscript] ′ ′ ( R ), is given by
$$\mathcal { U } _ { v j \ell, v ^ { \prime } j ^ { \prime } \ell ^ { \prime } } ^ { J M \epsilon _ { 1 } \epsilon _ { P } } ( R ) = \Delta _ { v j _ { 1 } j _ { 2 } } \Delta _ { v ^ { \prime } j _ { 1 } ^ { \prime } j _ { 2 } ^ { \prime } } [ \mathcal { U } _ { v j \ell, v ^ { \prime } j ^ { \prime } \ell ^ { \prime } } ^ { J M \epsilon _ { 1 } } ( R ) + \epsilon _ { P } ( - 1 ) ^ { j _ { 1 } ^ { \prime } + j _ { 2 } ^ { \prime } + j _ { 1 2 } ^ { \prime } + \ell ^ { \prime } } \mathcal { U } _ { v j \ell, \vec { v } ^ { \prime } j ^ { \prime } \ell ^ { \prime } } ^ { J M \epsilon _ { 1 } } ( R ) ],$$
where ¯ = v v v 2 1 and ¯ = j j j j 2 1 12 , glyph[epsilon1] P = ± 1, and ∆ vj 1 j 2 = [2(1 + δ v 1 v 2 δ j 1 j 2 )] -1 2 / . The matrix elements of the interaction potential, U JMglyph[epsilon1] I vjglyph[lscript],v ′ j glyph[lscript] ′ ′ ( R ), are defined as
$$\mathcal { U } _ { v j \ell, v ^ { \prime } j ^ { \prime } \ell ^ { \prime } } ^ { J M \epsilon _ { I } } ( R ) = \sum _ { \lambda } B _ { v j _ { 1 } j _ { 2 }, v ^ { \prime } j ^ { \prime } j ^ { \prime } _ { 2 } } ^ { \lambda \epsilon _ { I } } ( R ) f _ { j l, j ^ { \prime } l ^ { \prime } } ^ { J ; \lambda }$$
where the radial elements B λglyph[epsilon1] I vj 1 j 2 ,v j ′ ′ 1 j ′ 2 ( R ) are given by
$$B _ { v j _ { 1 } j _ { 2 }, v ^ { \prime } j _ { 1 } ^ { \prime } j _ { 2 } } ^ { \lambda \epsilon _ { I } } ( R ) = \int _ { 0 } ^ { \infty } \int _ { 0 } ^ { \infty } \chi _ { v j _ { 1 } j _ { 2 } } ^ { \epsilon _ { I } } ( r _ { 1 }, r _ { 2 } ) A _ { \lambda } ^ { \epsilon _ { I } } ( r _ { 1 }, r _ { 2 }, R ) \chi _ { v ^ { \prime } j _ { 1 } ^ { \prime } j _ { 2 } ^ { \prime } } ^ { \epsilon _ { I } } ( r _ { 1 }, r _ { 2 } ) d r _ { 1 } d r _ { 2 },$$
and the function f J λ ; jl,j ′ l ′ is given in terms of 3 -j , 6 -j , and 9 -j symbols:
$$\Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gammega \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma \Gamma
\begin{pmatrix} f _ { j _ { i }, j _ { i } } ^ { \prime } & \lambda _ { 1 } \\ 0 & 0 & 0 & 0 \end{pmatrix} \begin{pmatrix} j _ { 2 } & j _ { 2 } ^ { \prime } & \lambda _ { 2 } \\ 0 & 0 & 0 & 0 \end{pmatrix} \begin{pmatrix} l & l ^ { \prime } & \lambda _ { 1 2 } \\ 0 & 0 & 0 & 0 \end{pmatrix} \begin{pmatrix} j _ { 2 } ^ { \prime } & j _ { 2 } ^ { \prime } & j _ { 1 } ^ { \prime } \\ 0 & 0 & 0 & 0 \end{pmatrix} \begin{pmatrix} j _ { 1 2 } & j _ { 2 } & j _ { 2 } & j _ { 1 } \\ j _ { 1 2 } & j _ { 1 2 } & J & J \end{pmatrix} \begin{pmatrix} \begin{pmatrix} 1 0 \\ j _ { 1 2 } & j _ { 2 } & j _ { 2 } & j _ { 1 } \\ \lambda _ { 1 2 } & \lambda _ { 2 } & \lambda _ { 1 } \end{pmatrix}, \ \ ( 1 0 ) \\ \lambda _ { 1 2 } & \lambda _ { 2 } & \lambda _ { 1 } \end{pmatrix}$$
with the notation
$$[ x _ { 1 }, x _ { 2 }, \dots, x _ { n } ] = ( 2 x _ { 1 } + 1 ) ( 2 x _ { 2 } + 1 ) \dots ( 2 x _ { n } + 1 ).$$
In the coupled-channel formalism, either the wave function F ( R ) and its derivative F ′ ( R ) or the log-derivative matrix Y = F F ′ -1 is propagated from a point in the classically forbidden region near the origin, R ∼ 0, to where the interaction potential becomes negligible, R ∞ . In the present case, the CC equation for each value of R is solved by propagating the log-derivative matrix Y by following the methods of Johnson and Manolopoulos [29, 30]. The scattering matrix S for specific values of J , glyph[epsilon1] I and glyph[epsilon1] P is evaluated by matching the Y matrix to known asymptotic solutions of the CC equations at R ∞ . The boundary condition is
$$( \Upsilon J - J ^ { \prime } ) = ( \Upsilon N - N ^ { \prime } ) \, \text{K}.$$
The matrices J (not to be confused with the total angular momentum) and N are diagonal matrices of asymptotic functions. For convenience, the total number of coupled-channels N is partitioned into N o open channels (with E > 0) and N c closed channels (with E ≤ 0) such that N = N o + N c . For the open channels N o these functions are known as Riccati-Bessel functions, and for the closed channels N c they are modified spherical Bessel functions of the first and third kinds [31]. J ′ and N ′ are the derivative matrices of J and N , respectively. For an N channel problem the scattering S matrix is easily calculated by considering only the open-open sub-block of K matrix by the following expression
$$\mathbf S = ( 1 + i \mathbf K _ { o o } ) ^ { - 1 } ( 1 - i \mathbf K _ { o o } ).$$
Finally, the state-to-state cross section is obtained from the S matrix. For indistinguishable molecule collisions one must symmetrize the cross-section with the statistically weighted sum of the exchange-permutation symmetry components. Explicit expressions for state-to-state cross section with and without exchange symmetry have been given in prior publications [27, 32]. For completeness, and for the ease of comparisons with MQDT results, the expressions for the symmetrized cross sections are reproduced below:
$$\sigma _ { v _ { 1 } j _ { 1 } v _ { 2 } j _ { 2 } \to v _ { 1 } ^ { \prime } j _ { 1 } ^ { \prime } v _ { 2 } ^ { \prime } j _ { 2 } ^ { \prime } } ( E _ { c } ) = W ^ { + } \sigma ^ { \epsilon _ { P } = + 1 } + W ^ { - } \sigma ^ { \epsilon _ { P } = - 1 }$$
with
$$\sigma ^ { \epsilon _ { P } } \, = \, \frac { \pi ( 1 + \delta _ { v _ { 1 } v _ { 2 } } \delta _ { j _ { 1 } j _ { 2 } } ) ( 1 + \delta _ { v ^ { \prime } _ { 1 } v ^ { \prime } _ { 2 } } \delta _ { j ^ { \prime } _ { 1 } j ^ { \prime } _ { 2 } } ) } { ( 2 j _ { 1 } + 1 ) ( 2 j _ { 2 } + 1 ) k ^ { 2 } }$$
$$\times \sum _ { j _ { 1 2 } j _ { 1 2 } ^ { \prime } \ell \ell ^ { \prime } J \epsilon _ { I } } ^ { \mu \mu } ( 2 J + 1 ) | \delta _ { v j \ell, v ^ { \prime } j ^ { \prime } \ell ^ { \prime } } - S _ { v j \ell, v ^ { \prime } j ^ { \prime } \ell ^ { \prime } } ^ { J \epsilon _ { I } \epsilon _ { P } } ( E _ { C } ) | ^ { 2 }$$
where k 2 = 2 µE /h c ¯ . 2 In the case of collisions of two ortho-H 2 molecules having nuclear spin I = 1 and weight factors W + = 2 3 and / W -= 1 3, one must consider both exchange / permutation symmetries glyph[epsilon1] P = ± 1 for the calculation of state-to-state cross sections. For collisions between two para-H 2 molecules with nuclear spin I = 0 and weight factors W + = 1 and W -= 0, only one exchange-permutation symmetry glyph[epsilon1] P = +1 is required for evaluating the cross-section. To describe the state-to-state cross-section between two H 2 molecules, we use the term 'combined molecular state', CMS, which denotes the combined ro-vibrational quantum numbers of the two molecules. In this notation, the collision of the first H 2 molecule having the ro-vibrational state ( v , j 1 1 ) with the second H 2 molecule in state ( v , j 2 2 ) is denoted by a unique term ( v , j 1 1 , v 2 , j 2 ). This CMS is the quantum state which characterizes the molecule-molecule system before or after the collision.
It should be emphasized that the CC method described here is a numerically exact calculation that incorporates the complete four-body physics of the collision complex, provided that sufficiently many channels are included in the calculation (which is certainly possible for light molecules such as H ). 2 This method does, however, require the complete calculation to be performed separately for each collision energy of interest. The number of such calculations may be large, say in the case where cross sections vary with energy due to resonances or (at ultracold temperature) due to the Wigner threshold laws. Restricting this requirement of calculations at many energies is a main accomplishment of the MQDT method, to which we now turn.
## B. The MQDT Formalism
The MQDT formalism modifies the scattering calculations in several ways. First, it acknowledges that, beyond a certain interparticle spacing R m , the scattering channels become independent from one another, and their wave functions can be constructed in each channel individually. This leads to a reduction in computational time since the number of arithmetic operations is proportional to N 3 for the CC calculation. Whereas, this number is only proportional to N for the MQDT calculation. Second, it notes that this distance R m can often be chosen small enough that all channels are 'locally open,' meaning that the kinetic energy at R m is positive in each channel. In this circumstance, boundary conditions in closed channels need not yet be applied, and the wave function at R m will not have the
sensitive energy dependence required near resonances.
Third, the asymptotic wave functions to which one matches the short-range wave function are themselves chosen to exhibit weak energy dependence, so that the resulting short-range K -matrix, K sr , is only weakly dependent on energy and magnetic field. This allows for efficient calculations over a wide range of energy and field. Features such as resonances and Wigner threshold laws are then recovered at a later stage via relatively simple algebraic procedures. This method has proven useful and economical in molecular scattering [6] and in ultracold collisions [8]. Here we describe its application to the H -H 2 2 cold collision problem.
K sr is defined by writing the matrix wavefunction, M , in terms of MQDT reference functions, ˆ and ˆ, f g
$$M _ { i j } = \hat { f } _ { i } \delta _ { i j } - \hat { g } _ { i } K _ { i j } ^ { s r } \quad \text{for $R\geq R_{m}$}$$
where M is a N × N matrix that contains N wavefunctions with physical boundary conditions at the origin. K sr is obtained by matching the log-derivative of M to the log-derivative matrix Y at R m
$$M ^ { ^ { \prime } } M ^ { - 1 } = Y$$
$$K ^ { s r } = ( \mathbf Y \hat { g } - \hat { g ^ { \prime } } ) ^ { - 1 } ( \mathbf Y \hat { f } - \hat { f ^ { \prime } } ).$$
To achieve a weakly energy and field dependent K sr , we let ˆ and f ˆ have WKB-like g boundary conditions well within the classically allowed region at R = R x ≤ R m [6, 33],
$$\hat { f } _ { i } ( R ) = \frac { 1 } { \sqrt { k _ { i } ( R ) } } \sin \left ( \int _ { \underset { \mathbb { R } } { R _ { x } } } ^ { R } k _ { i } ( R ^ { \prime } ) d R ^ { \prime } + \phi _ { i } \right ) \text{ at } R = R _ { x }$$
$$\hat { g } _ { i } ( R ) = - \frac { \cdot \cdot \cdot \cdot } { \sqrt { k _ { i } ( R ) } } \cos \left ( \int _ { R _ { x } } ^ { R } k _ { i } ( R ^ { \prime } ) d R ^ { \prime } + \phi _ { i } \right ) \text{ at } R = R _ { x }$$
where φ i denotes an energy independent phase described by Ruzic et al. [22]. Here, k i ( R ) = √ 2 µ ¯ h 2 ( E -glyph[epsilon1] v j i i -V ii ( R )), and the derivatives of ˆ and ˆ at f g R x are defined by the full, radial derivatives of equations (20) and (21).
One obtains ˆ and ˆ at all f g R by solving a 1-D Schr¨dinger equation o
$$\left ( - \frac { \hbar { ^ { 2 } } } { 2 \mu } \frac { d ^ { 2 } } { d R ^ { 2 } } + \frac { \hbar { ^ { 2 } } \ell _ { i } ( \ell _ { i } + 1 ) } { 2 \mu R ^ { 2 } } + V ^ { l r } + \epsilon _ { v _ { i } j _ { i } } - E \right ) \binom { \hat { f } _ { i } } { \hat { g } _ { i } } = 0$$
subject to the boundary conditions (20) and (21). For H -H 2 2 scattering, beyond the strong interaction region, one only needs to deal with the weak, attractive van der Waals forces.
Hence, the MQDT reference functions can be obtained by choosing a long-range expansion for the reference potential, V lr = -C 6 R 6 -C 8 R 8 -C 10 R 10 .
The matrix K sr and the linearly independent solutions, ˆ and ˆ, carry all the information f g required to obtain the scattering observables. To obtain the physical scattering matrix, S phys , four MQDT parameters A G , , η and γ are required in each channel [22]. These four quantities correctly describe the asymptotic behavior of the reference wave functions ˆ and f ˆ. g Explicit expressions for these parameters are given in Eqs (12a) to (12d) in [22].
By partitioning K sr into energetically open (o) and closed (c) channels, we eliminate the unphysical growth inherent in M by the following transformation
$$\tilde { K } = K _ { o o } ^ { s r } - K _ { o c } ^ { s r } \left ( \cot \gamma + K _ { o c } ^ { s r } \right ) ^ { - 1 } K _ { c o } ^ { s r }$$
where cot γ is a diagonal matrix of dimension N c × N c . Hence, ˜ , K represents the N o wavefunctions with physical boundary conditions both at the origin and asymptotically. Roots of det( K cc +cot γ ) approximate the locations of resonances in the cross section.
In order to relate ˜ K to S phys , another set of energy-normalized, linearly independent solutions is required. For each energetically open channel, the reference functions f and g are defined as
$$f _ { i } ( R ) \xrightarrow { R \to \infty } k _ { i } ^ { - 1 / 2 } \sin ( k _ { i } R - \ell _ { i } \pi / 2 + \eta _ { i } )$$
$$g _ { i } ( R ) \xrightarrow { R \to \infty } - k _ { i } ^ { - 1 / 2 } \cos ( k _ { i } R - \ell _ { i } \pi / 2 + \eta _ { i } )$$
These functions are related to ˆ and ˆ through the following expressions, f g
$$f _ { i } ( R ) = \mathcal { A } _ { i } ^ { 1 / 2 } \hat { f } _ { i } ( R )$$
$$g _ { i } ( R ) = \mathcal { A } _ { i } ^ { - 1 / 2 } \mathcal { G } _ { i } \hat { f } _ { i } ( R ) + \mathcal { A } _ { i } ^ { - 1 / 2 } \hat { g } _ { i } ( R ).$$
Hence, S phys is obtained by the following series of simple transformations
$$K = \mathcal { A } ^ { 1 / 2 } \tilde { \mathbf K } \left ( I + \mathcal { G } \tilde { \mathbf K } \right ) ^ { - 1 } \mathcal { A } ^ { 1 / 2 }$$
$$\text{S} ^ { \text{phys} } = e ^ { i \eta } \left ( I + i \text{K} \right ) \left ( I - i \text{K} \right ) ^ { - 1 } e ^ { i \eta }$$
where A and G and η are diagonal matrices of order N o × N o and I is the identity matrix.
## III. RESULTS
Our main goal in this article is to demonstrate the power of the MQDT method for ultracold, non-resonant and quasi-resonant molecular scattering. To this end we wish to
FIG. 1: Effective potential energy curves for para-H 2 scattering as defined in eq.(30). The labels for nearly degenerate curves are separated by a colon, where the highest lying threshold is on the right. The panel on the right shows the CMSs included in the basis set.
establish two criteria for H . 2 First, that the separation between long and short-range physics is reasonable; and second, that the energy-dependent scattering may be easily described via an essentially energy-independent short-range wave function. We will also examine the sensitivity of results to the choice of the reference potential. To address these criteria, we will focus on H 2 collisions in which energy is nontrivially transferred among vibrational and rotational degrees of freedom between the two molecules.
## A. Quasi-resonant scattering: convergence with respect to matching distance R m
To establish the use of MQDT as a reasonable separation between short- and long-range behavior, we choose a problem where nontrivial energy exchange occurs in the short-range physics. The specific example is one of 'quasi-resonant' energy transfer in para-para H 2 scattering, whereby two units of rotational angular momentum are transferred from a vibrationless molecule to a vibrating molecule [32, 34]. To illustrate this process, diabatic potential energy curves for two interacting para-H 2 molecules versus the intermolecular separation R are shown in Fig. 1. These are effective potentials, defined as
$$V _ { e f f } ( R ) = \epsilon _ { v j } + \mathcal { U } _ { v j \ell, v j \ell } ^ { J M \epsilon _ { f } \epsilon _ { P } } ( R ) + \frac { \hbar { ^ { 2 } } \ell ( \ell + 1 ) } { 2 \mu R ^ { 2 } }.$$
The particular quasi-resonant process of interest takes the initial channel ( v , j 1 1 , v 2 , j 2 ) = (1 0 0 2) to (1 , , , , 2 0 0). The name quasi-resonant pertains to the fact that the thresholds of , ,
these channels are nearly degenerate, as seen in Fig. 1. Specifically, their energy separation, 25.45 K, is comparable to the well depth of the isotropic part of the H -H 2 2 interaction ∼ 31.7 K. The slightly different centrifugal distortion of the vibrational levels v = 0 and v = 1 is responsible for this small energy gap between the two CMSs. In this process the total rotational angular momentum is conserved by the collision. These kinds of transitions which have a small internal energy and internal rotational angular momentum gaps are found to be very efficient and highly state-selective and have been referred to as quasi-resonant rotationrotation (QRRR) transfer [34]. Note that alternative final states are also possible, but the one we have selected is known to be the dominant one. Indeed, as illustrated in [35, 36], to accurately describe such quasi-resonant energy transfer, one does not need to couple any other v, j levels in the basis set. One can restrict the basis set in the CC calculations to just those involved in the quasi-resonant transition yet still get results comparable to those from a larger basis set that includes many other CMSs. Thus, for the purpose of simplicity, we have resorted to a small basis (1,0,0,0),(1,0,0,2),(1,2,0,0) and (1,2,0,2) (as shown in the right panel of Fig. 1) that primarily includes the quasi-resonant channels in the CC calculations.
Using the restricted basis set described above, we have computed scattering cross sections for this process using the full CC calculation and the MQDT formalism. In the full CC calculation asymptotic matching to free-particle wave functions is carried out at R ∞ = 100 a 0 . In the MQDT approach, reference functions are determined from both the isotropic parts of the diagonal elements of the long-range diabatic potential coupling matrix, U JMglyph[epsilon1] I glyph[epsilon1] P vjglyph[lscript],vj glyph[lscript] ( R ), and also the long-range approximation for the reference potential, V lr = -C 6 R 6 -C 8 R 8 -C 10 R 10 . As discussed in the next subsection, we find that more accurate results are obtained when the long-range part of the diabatic potential curves are employed.
First, we establish convergence of elastic and inelastic cross sections as a function of the short-range matching distance R m . Results of these studies are shown in Figure 2, for elastic (left panel) and inelastic QRRR (right panel) scattering. The QRRR transition is the dominant inelastic channel that corresponds to total angular momentum J = 2 and s -wave scattering in the incident channel. The solid black curve refers to results from the full closecoupling calculation, while the other curves correspond to the MQDT results for various values of the matching radius R m . The convergence with respect to R m is excellent and occurs as soon as R m exceeds the region of the van der Waals well. The MQDT calculations are converged and nearly quantitatively reproduce the results from the full CC calculation
FIG. 2: Elastic (left) and inelastic (right) cross sections for (1,0,0,2) to (1,2,0,0) quasi-resonant scattering in para-H . 2 The black curves correspond to the full CC calculation, while the other curves correspond to the MQDT result with different matching radii, R m .
for a matching distance of R m =9.2 a.u . Note that the van der Waals length, r vdw = ( 2 µC 6 ¯ h 2 ) 1 4 / for H -H 2 2 is 14.5 a 0 . The agreement is also nearly quantitative for elastic scattering cross sections shown in the left panel. Although, for elastic scattering, the results are less sensitive to the matching distance. It should be emphasized that these convergence tests involve a single short-range K sr matrix computed at an initial collision energy of 1 µ K, which is able to capture the dynamics at other collision energies in 1 µ K-1 K range. This is an important aspect of MQDT calculations as discussed in more detail below.
## B. Choice of reference potential and energy independent parameters
A key aspect of MQDT is to simplify the calculation and description of scattering, especially at ultracold temperatures. We have already demonstrated this for the para-para case where an energy independent short range K sr is able to reproduce cross sections over a wide range of energies when resonances are absent. An equally important issue is the choice of reference potential for the evaluation of MQDT reference functions. While for atom-atom scattering the obvious choice is the long-range expansion, for complex systems such as the present case, a more accurate choice given by the effective potential of Eq.(30) may be adopted. To explore these issues we consider similar quasi-resonant energy transfer in ortho-ortho and ortho-para collisions. Further, we extend the energy range of these calculations to 10 K to capture a d -wave shape resonance reported in an earlier work [35].
Similar to the case of para-H , 2 here we consider a QRRR process in which an orthoH ( 2 v = 1 , j = 1) molecule hits another vibrationless ortho-H ( 2 v = 0 , j = 3) molecule and takes away two units of rotational angular momentum. The process is described as ( v , j 1 1 , v 2 , j 2 ) = (1 1 0 3) , , , → (1 , 3 0 1). , , In this case the energy difference between the two threshold channels is 45 5 K. Full CC calculations have been reported for this process . previously [35, 36], but we show results for the positive exchange symmetry to compare with MQDT in Fig.3. The elastic cross section is calculated according to equation (16). Only cross-sections for total angular momentum J = 2 that include s-wave scattering in the incident channel are shown. The solid black curve denotes the full CC result for both elastic (left panel) and inelastic (right panel) collisions obtained by matching to free particle wave functions at R ∞ = 100 a 0 .
MQDT results are obtained using a short-range matching distance of 9.5 a 0 . The different curves for MQDT refer to different choices for the reference potential. The red curve corresponds to using a simplified potential V lr = -C /R 6 6 , while the green curve corresponds to the diagonal elements of the isotropic part of diabatic coupling matrix discussed above. Both results correspond to a single K sr matrix evaluated at 1 µ K for the entire energy regime. Both reference potentials are able to identify the d -wave resonance near 1 K, but do not find its energy position particularly accurately. In addition, the very simplest reference potential struggles to reproduce the non-resonant elastic cross section near threshold since this potential does not deliver an accurate phase shift in the region R m < R < R ∞ . The blue curves in Figure 3 are also obtained using the diabatic potential coupling matrix for the reference potential, but here K sr matrix has been evaluated at various energy values in the 1 µ K-10 K range followed by interpolation in a fine energy grid. This grid consists of three different regions: (i) in the ultralow energy range, E c = 1 µ K - 100 mK, K sr matrix is evaluated at 1 µ K, 1 mK and 100 mK; (ii) at low energies in the range E c =200 mK-1 K, K sr is evaluated at 9 points with 100 mK separation; (iii) from 1 K - 10 K, an energy spacing of 1 K was employed. In this case the MQDT reproduces the full CC result quite well. MQDT identifies the resonance position and lineshape and matches the background cross sections at the percent level.
The ability to interpolate the short-range K -matrix and still get quantitative results stems from the smoothness of this quantity in energy, as shown in Fig. 4. Only elements corresponding to the quasi-resonant channels (1,1,0,3 and 1,3,0,1) are shown. Although the
1
FIG. 3: This figure shows the elastic cross section for H 2 ( v = 1 , j = 1)+H ( 2 v = 0 , j = 3) collisions (left panel) and the inelastic cross section for H 2 ( v = 1 , j = 1)+H ( 2 v = 0 , j = 3) → H ( 2 v = 1 , j = 3)+H ( 2 v = 0 , j = 1) quasi-resonant process (right panel).
FIG. 4: Diagonal elements of the short-range K-matrix corresponding to the quasi-resonant transition [(1,3,0,1) → (1,1,0,3)] as a function of the collision energy. The isotropic part of the diabatic potential matrix elements is used for the reference potential and the matching radius R m = 9 5 . a 0 .
partial waves, glyph[lscript] = 2 and 4 are present in both the initial and final channels, the dominant contribution comes from glyph[lscript] = 0 for E c = 1 µ K -1 K. It is clear from Fig.4 that up to 1 K the short-range K-matrix is independent of energy, but it becomes a smooth function of energy beyond 1 K. Thus, an energy independent short-range K-matrix evaluated at 1 µ K cannot be expected to adequately reproduce a shape resonance near 1 K.
Also, we note that the well depth of the isotropic part of the diabatic potential is only ∼ 31.7 K. When the collision energy becomes a significant fraction of the potential well depth the K sr matrix becomes a strong function of energy, and its energy dependence needs to be taken into account. Thus, for systems such as H +H 2 2 characterized by a relatively shallow van der Waals well, the energy dependence of the short-range K-matrix becomes important in describing the resonances supported by the van der Waals well. For other systems with deeper potential wells, one may be able to use an energy independent K sr matrix over a wider range of scattering energies. It is also possible that the MQDT treatment will need modification to handle resonances already present in K sr [16].
Next, we demonstrate the ability of the MQDT method to describe physics driven by vibrational, rather than rotational, dynamics. Specifically, we consider a second type of H 2 scattering, wherein a vibrationless ortho-H 2 hits a para-H 2 molecule that carries one quantum of vibration, transferring this vibration to the ortho molecule [36]. In our notation, paraH ( 2 v = 1 , j = 0)+ortho-H ( 2 v = 0 , j = 1) → para-H ( 2 v = 0 , j = 0) +ortho-H ( 2 v = 1 , j = 1) i.e, (1 , 0 0 1) , , → (0 , 0 1 1). Despite a small energy gap (8.5 K) between the initial and final , , states, one observes a surprisingly small inelastic cross section. This is because, due to the homonuclear symmetry of the H 2 molecules, the transfer of rotational energy between j = 0 and j = 1 states is symmetry forbidden. Hence the entire process is driven by vibrational energy transfer, which is generally less efficient than rotational energy transfer [34].
The resulting cross sections, computed according to the same three degrees of approximation as in Figure 3, are presented in Fig. 5, along with that from full CC calculations. The left panel depicts the elastic cross sections, and the right panel shows the inelastic counterparts. These cross sections were computed for total angular momentum J = 1 to account for the dominant s -wave scattering in the incident channel at ultralow energies. A matching radius of R m = 9 5 . a 0 is used for the MQDT calculations.
The solid black curve in Fig. 5 denotes the full close-coupling results computed on an energy grid of 10 µ K. The blue curve corresponds to MQDT calculations in which an interpolation scheme similar to that of the ortho-ortho case is adopted for the short range K sr matrix. The green curve represents the same calculation, assuming K sr (computed at E c = 1 µ K) is valid at all energies. Like the ortho-ortho case both the blue (interpolated) and green (non-interpolated) curves almost exactly follow the black curve in the Wignerthreshold region, but the green curve begins to deviate above 100 mK as the resonance
FIG. 5: Elastic (left) and inelastic (right) cross sections for ortho-para scattering of H 2 . The black curves represent the full CC calculation, while the other curves denote different MQDT results as described in the text below.
region is approached.
It is abundantly clear from the above discussion that while a single K sr matrix is not capable of reproducing the dynamics over the entire energy range, including the resonance region, it is able to accurately describe the dynamics in the Wigner threshold regime. The accuracy of both the elastic and inelastic cross sections in the three cases considered validates the key idea of MQDT: the energy dependence of scattering observables is entirely tractable within the simple behavior of the long-range physics. We also emphasize that the method remains numerically tractable all the way down to 1 µ K, about seven orders of magnitude lower in energy than the height of the centrifugal barriers. At 1 µ K (and throughout the Wigner threshold regime) the elastic cross sections from the MQDT and full CC calculations agree to within 0.2-4% for all three initial states considered here. The corresponding inelastic cross sections agree to witin 3-10%. Finally, we note that the MQDT calculations are also accurate for partial waves glyph[lscript] = 1 , and 2 arising from other values of J (e.g., J = 0 , 1 for ortho-ortho and J = 0 , 2 for ortho-para) although they do not contribute significantly to the cross sections and hence not shown.
## IV. CONCLUSIONS
We have presented a hybrid approach that combines full close-coupling calculations at short-range with the MQDT formalism at long-range to evaluate cross sections for elastic and quasi-resonant inelastic scattering in collisions of H 2 molecules. It is found that the full
CC calculation can be restricted to a relatively short-range, ∼ 9.0 a.u , which is just outside the region of the van der Waals potential well. Beyond this region, the scattering process is described within the MQDT formalism. Further, it is found that one may use a single short-range K -matrix, computed at say 1 µ K, to evaluate cross sections at energies all the way into the mK range, leading to significant savings in computational time. This works as long as scattering resonances are absent. When resonances are present, the short-range K -matrix becomes sensitive to energy and an interpolation of its elements computed on a relatively sparse grid in energy may be employed to yield reliable results.
The choice of H -H 2 2 for the present work is in part motivated by the possibility of fulldimensional CC calculations with no approximations (other than basis set truncation). However, due to its shallow van der Waals potential well, it is probably not the system for which MQDT provides the most accurate values. This is because, at the short range matching distance, for collision energies in the Kelvin range, the interaction potential becomes comparable to the scattering energy and determination of an energy independent short-range K-matrix is no longer possible. Furthermore, extending the calculations to energies beyond 1 K becomes difficult as the effective potential for glyph[lscript] > 2 becomes positive for all values of R and computation of MQDT reference functions becomes difficult. Nevertheless, the hybrid approach seems to be very promising, and the savings in computations will be more dramatic when considering open shell systems with spin, hyperfine levels, and magnetic field effects.
While the results demonstrate the relevance of the MQDT approach in ultracold moleculemolecule scattering, there is still much to be developed. The treatment of resonances has become mundane in atom-atom scattering, but remains to be adequately developed for cold molecules [16]. Also, the long-range PES for H -H 2 2 remains reasonably isotropic, so that complete isotropy could be assumed when constructing the reference functions. Potentials with stronger anisotropies may necessitate a different long-range treatment, owing to strong interchannel couplings. Finally, for truly reactive systems, such as F+H , the short-range 2 calculation is more conveniently carried out in hyperspherical, rather than Jacobi coordinates, in which case the application of MQDT needs to be modified to accommodate the transition between short- and long-range coordinate systems, as well as between short- and long-range wave functions. Such calculations are in progress.
## V. ACKNOWLEDGEMENTS
This work was supported in part by NSF grant PHY-1205838 (N.B.) and ARO MURI grant No. W911NF-12-1-0476 (N.B. and J.L.B). Computational support by National Supercomputing Center for Energy and the Environment at UNLV is gratefully acknowledged. JH is grateful to Brian Kendrick for many helpful discussions.
- [1] S. Ospelkaus, K.-K. Ni, G. Qu´m´ ener, B. Neyenhuis, D. Wang, M. H. G. de Miranda, J. L. e Bohn, J. Ye, and D. S. Jin, Phys. Rev. Lett. 104 , 030402 (2010).
- [2] For example, R. V. Krems, W. C. Stwalley, B. Friedrich, Eds., Cold Molecules: Theory, Experiment, Applications (Boca Raton, CRC Press, 2009).
- [3] M. J. Seaton, Proc. Phys. Soc. London 88 , 801 (1966).
- [4] M. J. Seaton, Rep. Prog. Phys. 46 , 167 (1983).
- [5] U. Fano and A. R. P. Rau, Atomic Collisions and Spectra (Academic Press, Orlando, FL, 1986).
- [6] F. H. Mies, J. Chem. Phys. 80 , 2514 (1984).
- [7] C. H. Greene, A. R. P. Rau, and U. Fano, Phys. Rev. A 30 , 3321 (1984).
- [8] J. P. Burke, C. H. Greene, and J. L. Bohn, Phys. Rev. Lett. 81 , 3355 (1998).
- [9] F. H. Mies and M. Raoult, Phys. Rev. A 62 , 012708 (2000).
- [10] M. Raoult and F. H. Mies, Phys. Rev. A 70 , 012710 (2004).
- [11] B. Gao, E. Tiesinga, C. J. Williams, and P. S. Julienne, Phys. Rev. A 72 , 042719 (2005).
- [12] T. M. Hanna, E. Tiesinga, and P. S. Julienne, Phys. Rev. A 79 , 040701 (2009).
- [13] Z. Idziaszek, T. Calarco, P. S. Julienne, and A. Simoni, Phys. Rev. A 79 , 010702 (2009).
- [14] B. Gao, Phys. Rev. Lett. 104 , 213201 (2010).
- [15] J. F. E. Croft, A. O. G. Wallis, J. M. Hutson, and P. S. Julienne, Phys. Rev. A 84 , 042703 (2011).
- [16] J. F. E. Croft, J. M. Hutson, and P. S. Julienne, Phys. Rev. A 86 , 022711 (2012).
- [17] Z. Idziaszek and P. S. Julienne, Phys. Rev. Lett. 104 , 113202 (2010).
- [18] Z. Idziaszek, G. Qu´m´ner, J. L. Bohn, and P. S. Julienne, Phys. Rev. A 82, 020703 (2010). e e
- [19] B. Gao, Phys. Rev. Lett. 105 , 263203 (2010).
- [20] G. R. Wang, T. Xie, Y. Huang, W. Zhang, and S.L. Cong, Phys. Rev. A 86 , 062704 (2012).
- [21] R. Pires, M. Repp, J. Ulmanis, E. D. Kuhnle, M. Weidem¨ller, T. G. Tiecke, C. H. Greene, u B. P. Ruzic, J. L. Bohn, and E. Tiemann, Phys. Rev. A (accepted, 2014).
- [22] B. P. Ruzic, C. H. Greene, and J. L. Bohn, Phys. Rev. A 87 , 032706 (2013).
- [23] K. Takayanagi, Adv. At. Mol. Phys. 1 , 149 (1965).
- [24] S. Green, J. Chem. Phys. 62 , 2271 (1975).
- [25] M. H. Alexander and A. E. DePristo, J. Chem. Phys. 66 , 2166 (1977).
- [26] S. K. Pogrebnya and D. C. Clary, Chem. Phys. Lett. 363 , 523 (2002).
- [27] G. Qu´ em´ner and N. Balakrishnan, J. Chem. Phys. e 130 , 114303 (2009).
- [28] A. M. Arthurs and A. Dalgarno, Proc. R. Soc. London, Ser. A 256 , 540 (1960).
- [29] B. R. Johnson, J. Comput. Phys. 13 , 445 (1973).
- [30] D. E. Manolopoulos, J. Chem. Phys. 85 , 6425 (1986).
- [31] M. Abramowitz and I. S. Stegun, Handbook of Mathematical Functions, U.S. Govern- ment Printing Office, Washington, DC, (1964).
- [32] N. Balakrishnan, G. Qu´ em´ner, R.C. Forrey, R. J. Hinde and P. C. Stancil, J. Chem. Phys. e 134 , 014301 (2011).
- [33] C. H. Greene, A. R. P. Rau, and U. Fano, Phys. Rev. A 30 , 3321 (1984).
- [34] G. Qu´ em´ner, N. Balakrishnan, and R. V. Krems, Phys. Rev. A e 77 , 030704(R) (2008).
- [35] S. Fonseca dos Santos, N. Balakrishnan, S. Lepp, Qu´m´ner, R. C. Forrey, R. J. Hinde, and e e P. C. Stancil, J. Chem. Phys. 134 , 214303 (2011).
- [36] S. Fonseca dos Santos, N. Balakrishnan, R. C. Forrey, and P. C. Stancil, J. Chem. Phys. 138 , 104302 (2013).
- [37] B. Gao, Phys. Rev. A 58 1728 (1998).
|
10.1103/PhysRevA.90.032711
|
[
"Jisha Hazra",
"Brandon P. Ruzic",
"N. Balakrishnan",
"John L. Bohn"
] |
2014-08-04T22:50:04+00:00
|
2014-08-04T22:50:04+00:00
|
[
"physics.atom-ph",
"physics.chem-ph"
] |
Multichannel quantum defect theory for ro-vibrational transitions in ultracold molecule-molecule collisions
|
Multichannel quantum defect theory (MQDT) has been widely applied to resonant
and non-resonant scattering in a variety of atomic collision processes. In
recent years, the method has been applied to cold collisions with considerable
success, and it has proven to be a computationally viable alternative to
full-close coupling (CC) calculations when spin, hyperfine and external field
effects are included. In this paper, we describe a hybrid approach for
molecule-molecule scattering that includes the simplicity of MQDT while
treating the short-range interaction explicitly using CC calculations. This
hybrid approach, demonstrated for H$_2$-H$_2$ collisions in
full-dimensionality, is shown to adequately reproduce cross sections for
quasi-resonant rotational and vibrational transitions in the ultracold
(1$\mu$K) and $\sim$ 1-10 K regime spanning seven orders of magnitude. It is
further shown that an energy-independent short-range $K$-matrix evaluated in
the ultracold regime (1$\mu$K) can adequately characterize cross sections in
the mK-K regime when no shape resonances are present. The hybrid CC-MQDT
formalism provides an alternative approach to full CC calculations at
considerably less computational expense for cold and ultracold molecular
scattering.
|
1408.0835v1
|
## The Design and Operation of The Keck Observatory Archive
G. Bruce Berriman *a , Christopher R. Gelino , Robert W. Goodrich a b , Jennifer Holt b , Mihseh Kong a , Anastasia C. Laity a , Jeffrey A. Mader b , Melanie Swain a , Hien D. Tran b a NASA Exoplanet Science Institute, Infrared Processing and Analysis Center, California Institute of Technology, Pasadena, CA 91125; b W. M. Keck Observatory, 65-1120 Mamalahoa Hwy, Kamuela, HI 96743.
## ABSTRACT
The Infrared Processing and Analysis Center (IPAC) and the W. M. Keck Observatory (WMKO) operate an archive for the Keck Observatory. At the end of 2013, KOA completed the ingestion of data from all eight active observatory instruments. KOA will continue to ingest all newly obtained observations, at an anticipated volume of 4 TB per year. The data are transmitted electronically from WMKO to IPAC for storage and curation. Access to data is governed by a data use policy, and approximately two-thirds of the data in the archive are public.
Keywords: Archives, ground-based telescopes, software development, data management, metrics, software architecture.
## 1. INTRODUCTION
Managed by a consortium of partner institutions, the W. M. Keck Observatory operates two 10-m telescopes at the summit of Mauna Kea, and in 2013 celebrated its 20 th year of operations. It was fitting, then, that at the end of 2013, the Keck Observatory Archive (KOA) (https://koa.ipac.caltech.edu) completed ingestion of all observations from the eight instruments in active use. KOA itself opened for business to the public in 2006, serving data from the High Resolution Echelle Spectrometer (HIRES). Data from the Near-Infrared Spectrograph (NIRSPEC) and the Near-Infrared Camera 2 (NIRC2) instruments followed between 2010 and 2012. As part of a plan for a rapid expansion of the archive, five more instruments have been added since 2012 and a sixth, the decommissioned Near-Infrared Camera (NIRC), will be added in July 2014. A second decommissioned instrument, the Long Wave Spectrometer (LWS), is planned for September 2014. The full complement of instruments offers wavelength coverage from the blue to the infrared, and includes imaging and spectrographic capabilities. In addition, data from the now decommissioned Keck Interferometer are accessible through a dedicated interface. The growth of the archive is reflected in the growth in scientific publications resulting from its use, with 60 papers citing KOA as of March 2014.
The KOA is a collaboration between the Observatory and the NASA Exoplanet and Science Institute (NExScI) at the Infrared Processing and Analysis Center (IPAC), and takes advantage of the Observatory staffÕs expertise with the instrumentation, operations and data organization, and of NExScIÕs expertise and infrastructure for archiving and serving large and complex data sets. Access to the data is governed by a data access policy, signed by the partner institutions and implemented by KOA. Newly acquired data are prepared for ingestion into the archive at the Observatory, and then electronically transferred for ingestion into the archive data storage and database system at NExScI. KOA has its own dedicated hardware, but re-uses the Science Information software system developed originally for the NASA/IPAC Infrared Science Archive (IRSA). A rigorous development model and rigorous test plan assures the robustness and performance of the software. Astronomers access the archive through its web page, where they launch queries to discover and visualize data, and download them for analysis.
This paper describes the design and operations of KOA, and provides a more thorough description of the archive than provided in earlier papers. 1, 2 A companion paper describes the metadata management approach that enables KOA to 3 deliver uniform, self-describing data sets across all instruments.
* [email protected]; phone 1-626-395-1817; https://koa.ipac.caltech.edu
## 2. SCIENTIFIC CONTENT OF THE KECK OBSERVATORY ARCHIVE
## 2.1 Science And Calibration Files
As of March 31, 2014, KOA housed over 2.15 million (nearly 30 TB) of raw (level 0) science and calibration data files recorded by the eight active instruments during 12,000 full and shared nights of observing. The data were stored locally over the operating life of the Observatory. Table 1 breaks down the science and calibration content by instrument, and Figure 1 shows how the number of science files has grown over the lifetime of the archive; the steps in this figure represent the data from newly ingested instruments.
## Running Total of Science Files
Figure 1: Growth In The Number Of Science Files In KOA (2004 To 2014).
The number of science files increased dramatically since September 2012, when the 600,000 science files (7 TB) made with HIRES, NIRC2 and NIRSPEC were augmented with data from Low Resolution Imaging Spectrometer (LRIS), the Multi-Object Spectrometer For Infra-Red Exploration (MOSFIRE), the DEep Imaging Multi-Object Spectrograph (DEIMOS), the Echellette Spectrograph and Imager (ESI), and the OH-Suppressing Infrared Imaging Spectrometer (OSIRIS). We anticipate that this volume will increase by approximately 4 TB per year with newly acquired data. Observations with two decommissioned instruments will be made available to Principal Investigators (PIs) later in 2014: the Near-Infrared Camera (NIRC), in July 2014, and the Long Wave Spectrometer (LWS) in September 2014. Data acquired by new instruments commissioned at the Observatory will be ingested as they begin science operations. The next instrument is the Near-Infrared Echelle Spectrometer (NIRES), expected to be available in late 2014.
Table 1: Data Volume and Release Dates of Active Keck Instruments Archived at KOA (March 2014). The shaded boxes denote data sets yet to be made public.
| Instrument | Nights | Volume (TB) | PI Access | Science Files | Calibration Files | Public Access |
|--------------|----------|---------------|----------------|-----------------|---------------------|-----------------|
| HIRES | 3,033 | 4.9 | July 2004 | 136,400 | 206,846 | July 2006 |
| NIRSPEC | 1,691 | 0.7 | May 2010 | 334,874 | 110,908 | May 2010 |
| NIRC2 | 1,700 | 1.8 | May 2012 | 405,998 | 160,071 | May 2012 |
| LRIS | 3,530 | 6.8 | September 2012 | 208,524 | 226,228 | September 2013 |
| MOSFIRE | 355 | 1.4 | January 2013 | 55,361 | 32,947 | January 2014 |
| DEIMOS | 1,293 | 10.4 | March 2013 | 38,309 | 71,881 | March 2014 |
|----------|---------|--------|---------------|----------|----------|---------------|
| ESI | 820 | 0.7 | July 2013 | 23,143 | 29,848 | July 2014 |
| OSIRIS | 891 | 2.9 | December 2013 | 56,007 | 53,166 | December 2014 |
## 2.2 Browse And Extracted Products
KOA derives Òquick lookÓ browse products, served as JPEG images, of raw science and calibration data for all instruments, intended to offer an overview of the content of each file. These browse products are usually created simply from the data files, but in the case of DEIMOS, components of the Montage image mosaic engine accelerated the 4 creation of the JPEGs, which required stitching of images written to multi-Header Unit (HDU) Flexible Image Transportation System (FITS) files.
KOA has automated existing pipelines to create and serve extracted and calibrated browse-quality products for HIRES, NIRC2 and OSIRIS, as summarized in Table 2. Extractions of HIRES spectra exploited the HIRES makee 5 package to create and signal-to-noise spectra as a function of wavelength for each order flux, an estimate of the extraction quality, diagnostic information, and a Pass/Fail grade, assigned by KOA, to inform the user of the quality of the automatic reduction.
Table 2: Number of extracted or reduced files archived in KOA (March 2014)
| Instrument | Number of Files |
|--------------|-------------------|
| HIRES | 51,707,346 |
| NIRC2 | 372,221 |
| OSIRIS | 85,040 |
NIRC2 images are dark-subtracted and flat-field corrected, but are neither sky subtracted nor mosaicked when the raw images are dithered. The OSIRIS pipeline delivered by the instrument team has been automated for use by KOA 6, 3 , and the pipeline includes the important step of identifying the sky frame for background subtraction.
## 2.3 Weather Information
KOA preserves ancillary weather data for 1221 nights, including observatory-generated plots of weather data and webpages produced by the CONCAM weather monitoring project, which was discontinued after 2007. It provides links to the Mauna Kea Weather Center's seeing data (since September 24, 2009) and night movie (available for 3 months after date of observation), and to the Atmospheric Attenuation plots from SkyProbe@CFHT (available since April 2, 2001).
## 3. DATA RELEASE POLICY
Access to the KOA data is governed by a Data Release Policy, agreed and signed by representatives of each of the Observatory partner institutions. The policy applies to newly acquired data and to previously acquired data, stored at the Observatory since individual instruments were commissioned. Under the terms of this policy, PIÕs and their collaborators are guaranteed exclusive access to their data and associated metadata for at least 18 months after the date of the observations, when the data become public immediately. For the five instruments ingested after 2012, the policy included a clause that PIs had exclusive access for one year after the data were served through KOA regardless of whether the proprietary period had expired.
PIs may request extensions to the 18-month proprietary period, if for example, observations are part of a long-term program or if a science program had been impeded by bad weather. For new observations, PIÕs may request extensions when making biannual proposals and reviewed by a designated selecting official (SO) assigned at each member institution; the decision of the SO is final. For previously acquired data, PIÕs are informed by KOA when their data are in the archive and are given an opportunity to request extensions. SOs enter approvals on a dedicated web page built by
KOA to track proprietary periods of all observing programs. Figure 2 shows how the number of public science files has grown from 2004 to March 2014. More than 900,000 science files are now public, approximately 65% of the total.
Figure 2: Growth in the number of science files in KOA from August 2004 to March 2014.
## 4. PREPARATION OF DATA FOR ARCHIVING, TRANSFER AND INGESTION
Data are prepared for archiving at the W. M. Keck Observatory, and then sent electronically to NExScI for ingestion into the physical archive; that is, the database and file storage system where data are stored, managed and curated.
## 4.1 Preparation Of Data For Archiving
Data Preparation Software. Data acquired from each instrument have been stored on digital tape at the observatory since the commissioning of each instrument extending as far back as 1994 in some cases. These data were not intended for long-term archiving or public release through an archive and, consequently, the attributes of the data sets are often heterogeneous and incomplete. Thus, KOA prepares fully self-describing data sets for ingestion into the archive. It developed the Data Evaluation and Preparation (DEP) software package to perform this task: it runs at the Observatory and then sends the data to NExScI. In what follows, it should be understood that the KOA team verified the ranges, data types and descriptions of all the metadata for HIRES, NIRSPEC and NIRC2, but to allow rapid incorporation of subsequent instruments, it validated only the most commonly queried attributes. KOA pays particular attention to ensuring the data type of each file is correctly identified; that is, whether it is on-sky science target or calibration file . 3
For ease in updating and testing changes, DEP has a modular design where the modules perform a single step in the processing flow. First, the DEP locates all FITS files for each instrument used on the telescopes in the previous 24-hour period. It then retrieves the Principal Investigator (PI), program ID, and title from the telescope schedule database and assigns the correct files to the PI(s). DEP is able to identify multiple PIs if the nights were shared . DEP then calls the 7 Data Quality Assessment (DQA) process, described in the next paragraph. Scaled JPEGs of each file are then created as a quick-look browse product. Next, the DEP calls data reduction processes, as needed, for three instruments: to produce extracted spectra from HIRES, to calibrate imaging data from NIRC2, and to produce data cubes for OSIRIS. Ancillary data such as weather and seeing estimates are also added. Lastly, the data are transferred electronically from WMKO to NExScI, where the data are stored and curated.
DQA plays a particularly important role in the data preparation process. It ensures metadata integrity and data quality, appends new FITS keywords necessary for archiving, and renames the file to a unique identifier filename (KOAID), which encodes the instrument name, date, and time of the observations. Checks are performed for filename integrity, instrument name, date and time, pixel data integrity, and duplicate data files. DQA distinguishes science files from calibration files, and identifies the type of calibration file for each instrument using, for example, information such as azimuth, elevation and dome position to determine if the file was measured on-sky. The data type is recorded as a keyword in the FITS file. Finally, DQA creates an md5sum of each file, used to ensure the integrity of the data transfer to NExScI.
Data Preparation Hardware Architecture. DEP and DQA are run on Sparc Solaris machines for MOSFIRE, HIRES, NIRC2, and NIRSPEC, and on i86pc Solaris for DEIMOS, ESI, and OSIRIS. The OSIRIS data reduction pipeline is run on a Red Hat Enterprise Linux quad-core virtual machine on solid-state storage, resulting in an increased processing speed of a factor of three compared with the quick-look data reduction machine. Both NIRC and LWS processing was performed on a dual processor, 8-core i86pc Solaris virtual environment on solid state storage. Running six LWS processing threads resulted in speedup of a factor of 18 over using a single thread on the existing i86pc machine. In the future, KOA plans to transfer all processing and reduction to virtual environments.
## 4.2 Data Transfer And Ingestion Into The Archive
The data are transmitted electronically from WMKO to a staging area at IPAC, from which the data are ingested into the archive, usually the afternoon after the observations were obtained. ÒIngestionÓ in this context means the data files are validated and transferred to archive mass storage disks, and the metadata used in querying the archive are ingested into the database, KOA applies rigorous methodology for ensuring the integrity of the data and accurate ingestion into the archive. Briefly, DEP computes the MD5 checksum of every file that is transmitted to NExScI, where the checksums are recalculated and compared to those sent: if any pair of values differs, the ingestion is rejected and data are retransmitted after correction of the problem. The transmitted data must be organized within a predefined file and directory structure; any deviations from this structure will again cause the nightÕs ingestion to be rejected, and the data are retransmitted after investigation. Finally, the data types and ranges of metadata that will be ingested into the database are checked. The database itself contains tables for each instrument. There are three other operational tables: a status table tracking nightly files delivered from WMKO, including weather data; a table of users recording PI login information; and a table for controlling PI access to protected data.
Figure 3: The frequency of data transmission and ingestion defects for the NIRC2 instrument, from January 2012 to March 2014.
All transmission and ingestion defects are recorded and tracked in an issue tracking system. KOA maintains a permanent time-stamped record of the results of transmission and ingestion of nightly data, and a record of the number of ingestion and transfer defects with time for all instruments. Fig 3 shows the rate of defects for one instrument, NIRC2. The high frequency of defects in Spring 2012 corresponds to the ingestion of previously acquired NIRC2 data that had been stored at the Observatory.
An Interface Control Document (ICD) specifies the structure and organization of the transmitted data, the response to error conditions and the content of the log file. The bullet list below describes the contents of the ICD in more detail:
- ∞ Identification of the directory in which the data will reside in the archive and the structure of all the subdirectories for raw science and calibration files, quick look files, and (as needed) level 1 files and quality assurance information
- ∞ Identification of all files in all subdirectories and their naming conventions, including data files and MD5 checksums.
- ∞ The format and content of all tables (e.g. tables of metadata).
- ∞ Specification of all keywords, their definitions, data types and ranges. These keywords include those created at the telescope, and those added by the DEP process.
- ∞ Specifications of all log files and their contents that will track the status of transfer and ingestion.
- ∞ Response to error conditions during transfer and ingestion.
## 5. THE ARCHITECTURE OF THE PHYSICAL ARCHIVE
## 5.1 Software Architecture
KOA exploits a sustainable and extensible archive software science information system, built originally to support the NASA Infrared Science Archive (IRSA) at IPAC, which began operations in 1999 . This information system exploits a 8 component-based architecture to enable strong re-use and adaptation, and avoid as far as possible special, usually vendor-specific, functionality that may become unsupported and may not be portable between platforms. The components, written largely in American National Standards Institute (ANSI)-compliant C, usually perform one generic function such as coordinate transformations, composing web pages, or downloading data. Third-party components are used only if supported and in wide use. Perhaps the best example is the Apache web server, maintained by the Apache Software Foundation, which keeps apace with changing platforms and the evolution of the Hyper Text Transport Protocol (HTTP) that underpins the World Wide Web. An example of components specific to astronomy are the libraries for reading files that comply with the Flexible Image Transport System (FITS), the international standard for serving astronomical data in a machine-independent fashion.
The architecture is inherently extensible because new applications and archive projects inherit all existing functionality. Each component is a module with a standard interface that communicates with other components through an executive library, which starts components as child services and parses return values. Web-based user interfaces and program interfaces are thin layers atop the architecture, and they can usually evolve in response to user recommendations without substantial changes to the underlying components. Development uses text editors rather than third-party integrated development environments and web-page creation tools. New applications plug together existing components and new components are added as needed; these may be new generic modules or custom modules for a particular data set or service. Thus the IRSA architecture has been extended to support the NASA Exoplanet Archive, (formerly the NASA Star and Exoplanet Database) the Cosmic Evolution Survey (COSMOS) archive, the Kepler Science Analysis System (KSAS), the Large Binocular Telescope Interferometer (LBTI) archive and the Las Cumbres Global Telescope Observatory (LCOGT) archive . Functionality developed to support a new archive is then automatically available across 9 the board. Thus, while a module for finding calibration files in KOA is specific to it , data-ingestion, data-validation and 3 data-packaging modules developed for KOA have been adopted by the NASA Exoplanet Archive and by the LBTI archive.
A particular example of software re-use is the collaborative development of the Interactive Correlation and Examination (ICE) toolkit, intended for immersion, interaction and visualization of archive data, including tabular, image and spectral data, and used by KOA for interaction with the table of results returned by the query engine, and for interactive plotting of HIRES extracted spectra. It is in active use by the Exoplanet Archive, the LBTI archive, the LCOGT archive, and the Virtual Astronomical Observatory (VAO) catalog cross-comparison service. These projects pool resources to determine requirements, delivery schedules, and to develop and test the components. ICE is scalable, portable, fast and re-usable, and represents an investment in a long-term solution for data immersion. Briefly, it separates computationally intensive tasks from their presentation and visualization in the browser. Processing-intensive tasks are performed on the server side: examples are the generation of large-scale images, background image generation and manipulation of large-scale tables containing millions of records. Presentation and interaction takes place in the browser via JavaScript GUIs, and the connections between that browser-side and the server-side processes are provided by CGI calls.
## 5.2 KOA Hardware Architecture
The KOA instance of the science information system software resides on dedicated platforms. A development server, ingestion server, and a public-facing webserver are all operated as virtual machines, hosted on one of two Dell
PowerEdge TM R720 servers. The data are stored on multiple Nexsan TM storage arrays, which use a 10-gig network connection to a SunFire TM x4250 file server. KOA uses an Oracle database, available at no direct charge through an Institutional site license, and shares a Sun X3-2L TM server with the NASA Exoplanet Archive.
KOA shares hardware and services with those archives using the common software architecture just described. They share a code repository and software configuration management system (the commercial tool Accurev TM) and the Seapine Software TestTrack TM issue system for managing user requests, bug reports, and ingestion defects. Accurev TM and TestTrack TM are housed on dedicated servers.
Backups of KOA data, software and database tables are performed in accordance with IPAC system policies. The backup schedule depends on how often data sets changes, but the policies ensure there are always at least two full backups on tape onsite and offsite.
## 6. SCIENCE SERVICES
Users access the archive through query forms on the archive web page at https://koa.ipac.caltech.edu. They may launch searches for data from multiple or single instruments, and query by attributes such as date and time, position, object name, program identifiers and by observation parameters. Entries in a list of public programs launch a query to recover the program data. PIs are given secure password-protected access to all their protected data sets. The query engines enable users to return for HIRES, NIRSPEC and NIRC2, the best set of calibration data for optimum extraction or reduction of the science files . 3
The web page provides descriptions of the instrumentation (usually the instrument pages at Keck or an adaptation of them), and the raw and calibrated or extracted data products (including for interactive visualization of HIRES extracted spectra), and tutorials on the services and how to use them. The query forms include brief descriptions of the formats of the parameters, as well as links to more detailed pages.
Queries launched from any of these services return a table of the metadata records that satisfy the search queries, with links to science and calibration files, browse products and calibrated or extracted products. These records include program parameters, instrument parameters, with context sensitive help on column names. The table is interactive and offers sorting and filtering of records to provide a more compact and manageable table (this is invariably more efficient than attempting to devise a new query).
## 7. DEVELOPMENT APPROACH AND TESTING
The science services described above, and all components that underpin them (calibration association, packaging data for download etc.), have been delivered through application of the Evolutionary Delivery lifecycle model 10 , a combination of staged delivery and evolutionary prototyping. This model combines feedback on previous releases and the needs of major deliveries, such as the deployment of a new instrument, to develop a schedule for a new release. Once established, the release date is considered a hard deadline.
As part of this development cycle, KOA has a rigorous test plan that was developed originally for the first instrument served, HIRES, and since extended and automated to support new instruments, new functionality and changes in platforms and development. During the HIRES delivery, test cases were derived to exercise the paths through each component and all the error conditions. A test was deemed passed if the results matched those expected, and failed otherwise. All test results were documented, and failures are tracked until closed in the issue tracking system. The HIRES test bed became the basis for a regression test bed that was extended as needed when new functionality and instruments were brought online. Many of the test cases for individual components have since been automated with Perl scripts. Table 2 summarizes the tests performed for the ingestion of OSIRIS observations, in December 2013. A total of 1,187 tests were completed, with 161 new tests added to support OSIRIS; the italicized components modules that were built or modified to support OSIRIS. Preparation of the OSIRIS data for the archiving, transfer and ingestion of the data was considered to comprise an end-to-end system test.
Table 3: Summary of Component-Based Testing for the OSIRIS data ingestion. The italics denote components that were revised or edited to support OSIRIS.
Component
Tests
| OSIRIS data ingestion | 68 |
|---------------------------------|------|
| Calibration Association by date | 39 |
| User Interface Ð Basic Search | 65 |
| User Interface Ð Full Search | 649 |
| Get a FITS file by file ID | 23 |
| Download from User Interface | 29 |
| Login/Data Access | 35 |
| PI Interface Released Programs | 9 |
| Quicklook Previews Ð raw | 45 |
| Quicklook Previews - Extracted | 16 |
| Master calibration association | 123 |
| Download by streaming | 43 |
| Login Library | 18 |
| Keck Interferometer | 25 |
| Total Completed | 1187 |
The deployment of an interactive results table in place of a static web-based table and deployment of interactive visualization of HIRES extracted spectra led to a major additions to testing of the JavaScript-driven functions on the web page, previously limited to testing radio buttons and options menus. A total of 200 tests were added to support the table visualization, plus a total of over 60 tests of the filtering actions of the table returned by the query engine. These tests are performed on a total of 22 platform and browser combinations, with each KOA team member performing a part of the testing. The platform and browser combinations are revised with every major deliverable as new platforms are introduced and older ones are deprecated.
Finally, small groups of external users are asked to evaluate KOA services and documentation, especially prior to the release of a new instrument. Their feedback has proven invaluable in developing interfaces that provide a low barrier of entry for new users and return sufficient information for interpretation by astronomers.
## 8. SCIENTIFIC IMPACT
The growing volume of data available to the public is reflected in a growing number of citations to KOA in peerreviewed journals. Figure 4 shows a cumulative plot of the number of citations to KOA up to the end of March 2014. The publication rate in 2014 appears to be accelerating, with a quarterly publication rate in 2014 that is three times that in 2013.
Figure 4: Cumulative Citations to KOA in peer-reviewed Journals, from 2009 to 2014Q1.
While the research performed with KOA reflects the range of investigations enabled by KeckÕs instrumentation, KOA has proven particularly valuable in studies of exoplanets and in follow-up to Kepler observations, and in studies of the intergalactic medium. Examples of published research include the discoveries include the discoveries of a four-planet system orbiting HD 141399 11 and of a Neptune-sized planet around Gliese 687b 12 ; and of an analysis of the atmosphere of ROXs 42Bb 13 ; and investigations of OVI as a tracer of large-scale stellar feedback at 2<z<4 14 .
KOA underwrote the extension of the HIRES reduction pipeline, makee , 5 to extract spectra from an instrument upgrade that replaced the single CCD camera with a mosaic of three CCDs, to provide wider wavelength coverage. KOA uses the pipeline to extract the browse spectra it serves, but the code has been made available for download since 2008 and is in 5 active use by astronomers. As of March 21, 2014, a total of 99 papers cite this version of makee 5 . Recent papers include abundance studies of the atmospheres of exoplanets. 15, 16.
## 9. ARCHIVE METRICS
This paper has described metrics that monitor the growth of the archive, robustness of data ingestion, and its scientific impact. These metrics are in fact part of a systematic program of metrics, measured quarterly since KOA began operations. There are a total of 50 metrics measured in total, and Table 4 summarizes the categories of metrics and some of the quantities measured. Broadly speaking, these metrics are intended to measure take-up by the community, assess usage to inform archive development efforts, and measure archive performance.
Table 4: Summary of the categories of metrics gathered quarterly by KOA and Examples of Measured Quantities
| Category | Measurement |
|-----------------------------|---------------------------------------------------------------------------------------------------------------------------------------------------|
| Growth in Scientific Impact | Citations in peer-reviewed literature |
| Growth in usage | Measured cumulatively and per instrument, and by anonymous users and by PI: ∞ Parameters searched on ∞ Number of queries ∞ Data volume downloaded |
| | ∞ Number of IP addresses |
|------------------------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| Growth in public and protected data sets | Measured per instrument: ∞ Number of public nights, volume, data files ∞ Number of private nights, volume, data files ∞ Extensions to proprietary periods and their values |
| Support for users | ∞ Number of issues open ∞ Time for resolution of issues |
| Data ingestion stability and efficiency | ∞ Time to ingest data. ∞ Number of data transmission failures and ingestion defects with time |
Two examples show how these metrics have been used by the archive. Measurements of the attributes queried by users show that much of the most common searches are by object name and position and by UT time, and were the most commonly searched, and this informed the development of the basic search functionality. Over 90% of the data downloaded are by anonymous users rather than by PIs accessing their protected data. While this is in part because PIs access data in near-real time at the Observatory, it is also because KOA used unique file names to serve data rather than file names used at the telescope. Accordingly, KOA is providing a script for PIs to recover file names.
## 10. BEST PRACTICES FOR BUILDING AN ARCHIVE
KOA recommends the following best practices for building and operating an archive:
- ∞ The most efficient development model is collaboration between instrument specialists and archive specialists.
- ∞ Take advantage of existing archive infrastructure to reduce development costs.
- ∞ Ensure that archiving operations do not restrict observing practices.
- ∞ Work with instrument staff to develop a complete and consistent set of keywords across all instruments, as far as is possible.
- ∞ Define and document procedures for transmitting data to the archive and for ingesting data into it.
- ∞ Define a set of metrics for measuring attributes of archive operations.
- ∞ Keep records of all archive procedures.
- ∞ Ensure that prospective end-users evaluate services and solicit their comments on the usefulness of archive services.
## 11. FUTURE DEVELOPMENT
The long-term plan is to make KOA an indispensable tool for the astronomical community. This plan is currently under development, but the services under consideration include: serving reduced products and providing access to data reduction pipelines for all instruments; supporting interoperability with existing archives, including VO-compliant services; supporting moving object searches and supporting more complex queries and interactive user navigation. Finally, the KOA architecture serves as a model for ground-based archives. It is in active use in developing archives for the Large Binocular Telescope Interferometer (LBTI) archive and the Las Cumbres Observatory Global Telescope (LCOGT).
Acknowledgements: The Keck Observatory Archive (KOA) is collaboration between the W. M. Keck Observatory (WMKO) and the NASA Exoplanet Science Institute (NExScI). Funding for KOA is provided by the National Aeronautics and Space Administration (NASA). WMKO is operated as a scientific partnership among the California Institute of Technology, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. NExScI is sponsored by NASAÕs Origins Theme and Exoplanet Exploration Program, and operated by the California Institute of Technology in coordination with the Jet Propulsion Laboratory. This material is based in part upon work supported by NASA under Grant and Cooperative agreement No. NNX13AH26A.
## REFERENCES
- [1] Berriman, G. B., Ciardi, D. R., Laity, A. C., Tahir-Kheli, N. D., Conrad, A., Mader, J., Tran, H. D., Bida, T., ÒThe Design of the W. M. Keck Observatory ArchiveÓ, Astronomical Data Analysis Software and Systems XIV, 347, 627 (2005).
- [2] Berriman, G. B., Ciardi, D., Abajian, M., et al., ÒThe NASA Exoplanet Science Institute Archives: KOA and NStEDÓ, Astronomical Data Analysis Software and Systems XIX, 434, 119 (2010).
- [3] Tran, H. D., Holt, J., Goodrich, R. W., Mader, J. A., Swain, M., Laity, A., Kong, M., Gelino, C.R., Berriman, G. B. Ò Metadata and data management for the Keck Observatory archive.' SPIE Paper 9152-92 (2014).
- [4] The Montage Image Mosaic Engine, http://montage.ipac.caltech.edu.
- [5] Makee User Guide and Technical Documentation: http://www.astro.caltech.edu/~tb/makee.
- [6] Holt, J., Tran, H. D., Goodrich, R., Berriman, G. B., Gelino, C. R., and the KOA Team, ÒAutomating OSIRIS Data Reduction for the Keck Observatory ArchiveÓ, Astronomical Data Analysis Software and Systems XXIII, In press (2014).
- [7] Tran, H. D., Mader, J. A., Goodrich, R. W., Tsubota, M., ÒTechnical Solutions in Preparing Data for The Keck Observatory Archive (KOA)Ó, Proc. SPIE 8451, 8451E-29 (2012).
- [8] Berriman G. B.,Good, J. C., Deelman, E., and Alexov, A. ÒTen years of software sustainability at the Infrared Processing and Analysis Center.Ó Phil. Trans. R. Soc., 369, 334 (2011).
- [9] Brown, T. M., and 51 co-authors. ÒLas Cumbres Observatory Global Telescope Network.Ó PASP, 125, 1031.
- [10] The Evolutionary Delivery development http://www.informit.com/articles/article.aspx?p=102256&seqNum=10
model,
- [11] Vogt, S. S., Butler, R. P., Rivera, E. J., Kibrick, R., Burt, J., Hanson, R., Meschiari, S., Henry, G. W., Laughlin, G. ÒA Four-planet System Orbiting The K0V Star HD 141399.Ó Ap J, 787, 97 (2014).
- [12] Burt, J., Vogt, S. S., Butler, R. P., Hanson, R., Meschiari, S., Rivera, E. J., Henry, G. W., Laughlin, G. ' The LickCarnegie Exoplanet Survey: Gliese 687b: A Neptune-Mass Planet Orbiting a Nearby Red Dwarf.Ó Ap J, submitted. (2014).
- [13] Currie, T., Burrows, A., Daemgen, S. ÒA First-look Atmospheric Modeling Study of the Young Directly Imaged Planet-mass Companion, ROXs 42Bb.Ó Ap J, 787, 104 (2014).
- [14] Lehner, N., O'Meara, J. M., Fox, A. J., Howk, J. C., Prochaska, J. X., Burns, V., Armstrong, A. A. ' Galactic and Circumgalactic OVI and its Impact on the Cosmological Metal and Baryon Budgets at 2<z<3.5.Ó AP J, in press, (2014)
- [15] Teske, J. K., Cunha, K., Smith, V. V., Schuler, S. C., Griffith, C. A.' C/O Ratios of Stars with Transiting Hot Jupiter Exoplanets.Ó Ap J, in press (2014).
- [16] Liu, F., Asplund, M., Ramirez, I., Yong, D., Melendez, J. ÒA high precision chemical abundance analysis of the HAT-P-1 stellar binary: constraints on planet formation.Ó Ap J, in press (2014).
|
10.1117/12.2054772
|
[
"G. Bruce Berriman",
"Christopher R. Gelino",
"Robert W. Goodrich",
"Jennifer Holt",
"Mihseh Kong",
"Anastasia C. Laity",
"Jeffrey A. Mader",
"Melanie Swain",
"Hien D. Tran"
] |
2014-08-04T23:08:11+00:00
|
2014-08-04T23:08:11+00:00
|
[
"astro-ph.IM"
] |
The Design and Operation of The Keck Observatory Archive
|
The Infrared Processing and Analysis Center (IPAC) and the W. M. Keck
Observatory (WMKO) operate an archive for the Keck Observatory. At the end of
2013, KOA completed the ingestion of data from all eight active observatory
instruments. KOA will continue to ingest all newly obtained observations, at an
anticipated volume of 4 TB per year. The data are transmitted electronically
from WMKO to IPAC for storage and curation. Access to data is governed by a
data use policy, and approximately two-thirds of the data in the archive are
public.
|
1408.0836v1
|
## ESSAYS ON ECLIPSES, TRANSITS AND OCCULTATIONS AS TEACHING TOOLS IN THE INTRODUCTORY ASTRONOMY COLLEGE COURSE
Noella L. Dcruz 1
Abstract: We occasionally include projects in our learner-centered introductory astronomy college course to enable non-science major students explore some astronomical concepts in more detail than otherwise. Such projects also highlight ongoing or upcoming astronomical events. We hope that students will feel more interested in astronomy through projects tied to astronomical events. In Spring 2012, we offered short essays focused on eclipses, transits and occultations to promote the rare transit of Venus that occurred on June 5 th , 2012. We asked students to write two short essays from three that were offered. The essays contained descriptive and conceptual parts. They were meant to serve as teaching tools. 62% of 106 essays from 55 students earned A, B or C grades. 21% of 47 feedback survey respondents felt the essays increased their interest in astronomy. 49% of respondents felt that the essays were not educationally beneficial and should not be offered again. The most common written response to our survey indicated that students need more guidance and better preparation in writing successful essays. Since students found the conceptual parts of the essays difficult, in the future we will provide relevant activities prior to essay deadlines to help students create successful essays.
Keywords: College non-majors; Research into teaching/learning; Teaching approaches; Essays; Solar system; Transits.
## ENSAIOS SOBRE ECLIPSES, TRÂNSITOS E OCULTAÇÕES COMO FERRAMENTAS DE ENSINO EM UM CURSO UNIVERSITÁRIO INTRODUTÓRIO DE ASTRONOMIA
Resumo: Nós ocasionalmente incluímos projetos em nosso curso universitário introdutório centrado no aluno para permitir aos estudantes que pertencem às carreiras não científicas explorar alguns conceitos astronômicos em mais detalhes do que o normal. Tais projetos também enfatizam eventos astronômicos em curso ou futuros. Esperamos que os alunos se sintam mais interessados na astronomia através de projetos ligados a eventos astronômicos. No termo de Primavera de 2012 (EUA), propomos ensaios curtos focados em eclipses, trânsitos e ocultações para promover o raro trânsito de Vênus que ocorreu no dia 5 de junho de 2012. Pedimos aos alunos que escrevessem dois ensaios curtos dentre três que foram propostos. Os ensaios continham partes descritivas e conceituais. Eles foram feitos para servir como ferramentas de ensino. 62% de 106 ensaios de 55 alunos ganhou graus A, B ou C. 21% dos 47 entrevistados que responderam ao levantamento posterior sentiu que os ensaios aumentaram seu interesse na astronomia. 49% dos inquiridos consideraram que os ensaios não eram benéficos para a educação e que não devem ser propostos novamente. As respostas escritas mais comuns da nossa pesquisa indicaram que os alunos precisam de mais orientação e melhor preparação para escrever ensaios bem sucedidos. Como os alunos consideraram difíceis os aspectos conceituais dos ensaios, no futuro iremos fornecer atividades relevantes antes dos ensaios para ajudar os alunos a escrevê-los com sucesso.
Palavras-chave: Carreiras não científicas universitárias; Investigação sobre ensino/aprendizagem; Abordagens de ensino; E nsaios ; Sistema Solar; Trânsitos.
1 Department of Natural Sciences, Joliet Junior College. E-mail: <[email protected]>.
## ENSAYOS SOBRE ECLIPSES, TRÁNSITOS Y OCULTACIONES COMO HERRAMIENTAS DE ENSEÑANZA EN EL CURSO UNIVERSITARIO INTRODUCTORIO A LA ASTRONOMÍA
Resumen: Ocasionalmente, incluimos proyectos en nuestro curso de introducción a la astronomía universitario centrado en el alumno para permitir que los estudiantes de carreras no científicas exploren algunos conceptos astronómicos en más detalle que lo habitual. Estos proyectos también ponen en relevancia eventos astronómicos en curso o futuros. Esperamos que los estudiantes se sientan más interesados en la astronomía a través de proyectos vinculados a eventos astronómicos. En el período de primavera de 2012 (EUA), propusimos breves ensayos centrados en los eclipses, tránsitos y ocultaciones para promover el raro tránsito de Venus que se produjo el 5 de junio de 2012. Le pedimos a los estudiantes que escribieran dos ensayos cortos de tres que se proponían. Los ensayos contenían partes descriptivas y conceptuales. Los mismos estaban destinados a servir como herramientas de enseñanza. 62% de los 106 ensayos de 55 estudiantes obtuvo grados A, B o C. 21% de los 47 encuestados que respondieron al cuestionario posterior consideró que los ensayos aumentaron su interés por la astronomía. 49% de los encuestados consideró que los ensayos no eran educacionalmente útiles y que no deben ser propuestos de nuevo. Las respuestas escritas más comunes a nuestra encuesta indicaran que los estudiantes necesitan más orientación y una mejor preparación en la redacción de ensayos exitosos. Dado que los estudiantes encontraron las piezas conceptuales de los ensayos difíciles, en el futuro vamos a ofrecer actividades pertinentes antes de los plazos de redacción para ayudar a los estudiantes a crear ensayos de mayor calidad.
Palabras clave: Carreras universitarias no científicas; Investigación en la enseñanza/aprendizaje, Enfoques de enseñanza; Ensayos; Sistema Solar; Tránsitos.
## 1. Introduction
Joliet Junior College (JJC), a community college in Joliet, Illinois, USA, offers an introductory astronomy course aimed at non-science majors. Our introductory astronomy course involves a survey of astronomy in one semester via lecture only. In order to sign up for the introductory astronomy class, students need to be able to do basic arithmetic and to be able to be competent enough in English that they can enroll in the first course in college level English writing at JJC. Students who take this course have a wide spread in age, ranging from traditional age college students (18 - 22 years old) to older students. This course is one of several that students can choose from in order to fulfill their physical science requirement for their Associate's degree. (Community colleges in USA typically offer associate degrees, after completion of two years of a specific set of college courses. Students complete their Bachelor's degree at another institution.) Students can choose courses in other disciplines such as chemistry, physics, geography and geology besides astronomy to fulfill the physical science course requirement.
JJC offers the introductory astronomy course in face-to-face and online formats. In this paper, we are concerned with only the face-to-face format. We use a variety of teaching strategies to keep students engaged during class such as lecture-tutorials, thinkpair-share questions, small group discussions and ranking tasks to keep students engaged during class. These have been proven to help students improve their understanding and mastery of astronomy concepts (SLATER; ADAMS, 2003). Outside of class, students do online quizzes and homework that mostly involves ranking tasks to further their learning.
Occasionally, we include projects in our astronomy course, so students can explore certain topics and associated concepts in more detail. In the past, we have offered group poster projects. Our students were not always able to meet with their group outside of class, and they felt this was the main concern with group projects (D'CRUZ, 2009). Because of this, in Spring 2012, we chose to offer individual short essays. Our essays were designed to have both descriptive and conceptual portions, and were intended to be teaching tools rather than projects that mainly involved gathering and presenting information.
We usually structure our projects around sky events and astronomy events that are in the news because students and the public tend to be more interested in the promoted topics. For example, total solar eclipses capture the attention of people all around the world, not just in the zone of totality (PASACHOFF, 2009; PASACHOFF, 2010). Both solar and lunar eclipses capture the interests of the general public and students because such events can be observed easily and they are highlighted in the news (BENNETT, et al. , 2012; LITTMANN; ESPENAK; WILLCOX, 2008). An astronomy event that was in the news a few years ago was the purported 400 th anniversary of the invention of the telescope in 2008. We offered group projects focused on telescopes to commemorate this anniversary, in Spring 2008 (D'CRUZ, 2009).
In Spring 2012, we decided to focus our two sections of introductory astronomy around eclipses (which we always cover) and two other similar astronomical events called transits and occultations. The motivation for this was the rare event, 'the transit of Venus,' that occurred on 5 th June 2012 for the USA (ESPENAK, 2011; also available at <http://eclipse.gsfc.nasa.gov/OH/transit12.html>; PASACHOFF, 2012). A transit occurs when an object that looks small on the sky passes in front of an object that looks larger on the sky (WINN, 2011). The smaller object appears as a black disk against the face of the larger object. It occurs when the Earth (or the observer) and two other astronomical objects happen to be in a straight line as they move in space. For example, an annular solar eclipse seen from the Earth can also be called the transit of the Moon across the Sun's disk. When an observer sees an apparently larger object obscure an apparently smaller one, the event is called an occultation (WINN, 2011; <http://www.occultations.org/>).
The previous transit of Venus occurred on Jun 8 th 2004 (universal time), and the next one will occur on 11 Dec 2117 (ESPENAK, 2011), hence this is indeed a rare event. It is rare because Venus' orbit is inclined by 3.4 degrees to the Earth's orbital plane (ESPENAK, 2011). Venus travels through the Earth's orbital plane in early June and December, but the Earth does not line up with the Sun and Venus at those times because of the differing speeds with which Earth and Venus travel in their orbits around the Sun. The combination of the non-zero inclination of Venus' orbit and the differing orbital speeds of Earth and Venus causes the transit of Venus to recur with gaps of 8 years, 105.5 years, 8 years and 121.5 years currently (ESPENAK, 2011). Because the next transit of Venus will not occur in our lifetime, we felt it was essential to bring this exciting and rare event to the attention of our students in Spring 2012. The event could also be viewed from North America, weather permitting, so students could witness it for themselves if they wished.
The astrophysical significance of the transit of Venus is historical -it was used to measure the distance between the Earth and Sun (ESPENAK, 2011; PASACHOFF, 2012). Currently, radar measurements are far more accurate in determining this
distance; hence this transit event appears not be as scientifically important now. Pasachoff (2012) describes some current scientific studies that can be carried out during a transit such as learning about Venus' atmosphere.
Transits occur in other situations too, such as when one of Jupiter's moons passes in front of its disk e.g. http://www.hubblesite.org/newscenter/archive/releases /2004/30/. Another situation that involves transits is in the detection of planets orbiting other stars (e.g. the Kepler mission, BORUCKI, 2010; <http://kepler.nasa.gov>). Such planets are called exoplanets or extrasolar planets. Exoplanets and their parent stars are too far from Earth to image the smaller planet passing in front of the larger parent star (which is the transit detection method discussed so far). Instead, the star's brightness is measured frequently, and the transit is detected photometrically when the star's brightness dims slightly as its planet passes in front of it (WINN, 2011). This dimming repeats after a time interval of one orbital period, indicating the presence of a planet. The frequent detection of transits these days, especially in the discovery of exoplanets, was another motivating factor in choosing the course theme.
To enable students to explore the course theme, we covered the topics in lecture, we included theme-related questions on the tests and final exam and we asked students to write two theme-related essays. Essays and writing assignments that focus on astronomy concepts are used in introductory astronomy classes to enable students to demonstrate mastery of the topics being taught and encourage students to think more deeply about these topics (GREENSTEIN, 2013; SLATER, 2008). Slater (2008) indicates that writing within a particular discipline is important because it helps with learning the content and it helps to develop the ability to communicate successfully within that discipline. Slater (2008) also mentions that students need substantial guidance with writing essays in order to be successful, and that writing assignments that take 5 minutes or less are easier to implement and therefore have been more successful than essays.
We use multiple-choice questions extensively in our class. Having theme-related essays allowed us to gain deeper insight into how students think about the theme because our essays included conceptual parts in addition to descriptive ones. We would not have been able to gather such extensive information about student learning and preconceptions regarding the theme concepts solely through multiple-choice questions.
At the beginning of the Spring 2012 semester, we had asked our students to sign a statement, which said that they allowed their responses to all parts of the class to be used and shared in studies such as this one. The statement also mentioned their responses would be kept confidential. Students were given the option of not signing the statement, and students who chose not to sign have not been included in this study.
This paper is organized as follows: Section 2 covers the astronomy and general education goals of this project. Section 3 explains the essay assignment details. Section 4 provides information on the submitted essays and suggestions for possible future modifications. Section 5 details feedback received from students regarding the essays. Section 6 explains how our students fared on test and final exam questions related to the course theme. Section 7 contains the implications for future essay offerings, and Section 8 contains our conclusions.
## 2. Astronomy and general education goals
Since the sky event we were focused on in Spring 2012 was the transit of Venus, we had three major goals for this project:
- Could students explain what is meant by the 'transit of Venus?' (Goal Ia)
- Could students explain is meant by a transit, an occultation and an eclipse via lecture and essays? (Goal IIa)
- Would essays on transits, occultations and eclipses increase students' interest in astronomy? (Goal III)
The first major goal of this project, Goal Ia, is tied to why the transit of Venus occurs rarely. So an associated goal was for students to know why the transit of Venus occurs rarely (Goal Ib).
Since the transit of Venus and transiting exoplanets were the main motivating factors behind the course theme, we decided to have our students learn about some of the factors that affect transits. Students were expected to learn the following in connection with our second major goal:
Goal IIb: what is meant by angular size of a spherical object and how it depends on the object's radius, and distance from Earth.
Goal IIc: predict whether a transit will occur if the rotation period of the transiting object is altered or if the rotation period of the background object is altered, assuming all other parameters stayed unchanged; and to predict whether the duration of the transit will be altered if a transit does occur. The duration of the transit is illustrated in Figure 1.
Large circle = background object Black dot = transiting object
Figure 1: We define the duration of the transit to be the time taken for the transiting object to travel from point P to point Q of the background object as shown above.
Goal IId: predict whether a transit will occur if the angular size of the transiting object is altered or if the angular size of the background object is altered, assuming all other parameters stay unchanged; and to predict whether the duration of the transit will be altered if a transit does occur.
Goal IIe: determine how the orientation of the orbit affects whether a transit occurs and how long it lasts if it does occur (see Figure 2).
Figure 2: This figure shows a transiting object's original orbit on the left, and three altered orbits on the right. The small circle is the transiting object for all panels. The large hatched circle is the background object in all panels except for middle right, where both objects are the same distance from the observer. An edge-on orbit refers to one seen in side-view; a face-on orbit refers to one seen from above. Students were asked to predict if a transit would still occur for one of the altered orbits in Essay 1.
Goal IIf: predict whether changing a transiting exoplanet's rotation period, orbital radius or orbital period will alter the duration of the transit.
Goal IIg: predict whether changing a transiting exoplanet's rotation period, orbital radius or orbital period will alter the amount dimming of its parent star when the planet is completely in front of the star.
We aimed to achieve goal I through lecture, and goal II through background information provided during lecture and through essay assignments. We informed students that besides the essay assignments, a few test questions would be related to these goals in addition to our usual slate of questions on eclipses and exoplanets.
In order to predict how successful our students would be in correctly knowing and explaining the terms listed in Goals Ia and IIa, we relied on results from Prather et al. (2005). These authors studied the efficacy of learning in the introductory astronomy class via lecture only or lecture-tutorials. A lecture-tutorial is a structured, conceptual, workbook activity, that students complete during class and therefore can receive immediate feedback from their instructor. Prather, et al., find that 52% of questions in a 68-question survey are answered correct when students are taught solely through lecture, and that 72% of answers are correct when lecture is shortened and combined with in-class lecture tutorials. Since our students learned about Goal 1a through lecture and about Goal IIa through lecture, think-pair-share questions and out-of-class essays, we felt that if students scored at least 52% when explaining these two goals, we would be successful. Lightman and Sadler (1993) show that teachers tend to vastly overestimate the learning achieved by their students when they are asked to make similar predictions. This is why we tied our prediction to a robust study of learning in the introductory astronomy class, but we cannot tell how valid this cut-off is for this paper. There is no similar study on the efficacy of learning through essays in the introductory astronomy class as mentioned in Section 1. Therefore, we are unable to predict how well students would fare on Goal IIa through essays alone.
The third major goal of the project was to increase student interest in astronomy via essays. We were unable to predict the outcome of this goal because it is usually not possible to predict attitudes. But we obviously wanted a majority of students to have an increased interest in astronomy.
Goals I and II, and the standard goals of the introductory astronomy class (SLATER; ADAMS, 2003), help to develop students' logical and critical thinking skills. Critical thinking is one of the institutional general education goals of a science class for non-science majors at JJC. Essay assignments enable students to communicate via writing, which is another institutional general education goal of a class like this one. We do not assess these two general education goals specifically in this project, but students were able to practice the associated skills through the essays and conceptual questions outlined here. (General education goals at the college level in the US typically refer to skills associated with oral and written communication, quantitative literacy, scientific literacy and critical thinking. Some of these skills are common to multiple disciplines and are used frequently outside of formal educational settings. Hence, they are emphasized at the Associate's and Bachelor's degree levels in the US.)
Individual essay assignments had additional goals that are detailed in Section 3.
## 3. Details of Essay Assignments
Students were asked to submit two of the three essays described below. This allowed them some choice in essay topics and deadlines, and staggered our grading. We have found that assigning students at least two essays or two projects usually helps students develop the skills needed to gather and organize information for such assignments (D'CRUZ, 2009; SLATER, 2009). Since grading individual submissions requires considerable time commitment, we told students we would grade the first two essays they submitted.
We chose to assign individual essays because these do not rely on group effort. We have assigned group projects in the past, and have given students up to 90 minutes of class time per project to facilitate working in a group. However, students still needed to meet outside of class to complete their project, and it was difficult for many groups to find a common meeting time (D'CRUZ, 2009; D'CRUZ, 2010).
We also chose to assign short essays so that a small part of the course grade would be devoted to essays. These were not meant to be high stakes assignments. We actively encouraged all students to submit essays as related questions would appear on tests and the final exam.
Students were expected to submit essays that were 3 - 5 pages long, with double-spaced 12 pt font text, inclusive of bibliography (around 1200 - 1500 words). An additional page or two could be used for images. Each essay was worth 2.5% of the course grade. All essays had grading rubrics associated with them and detailed guidelines of what was expected of students. Students were given two weeks to complete an essay.
Students were informed that all essays would be checked for originality using turnitin, an anti-plagiarism service that JJC subscribes to (http://www.turnitin.com). They were asked to submit electronic versions of their essays directly to the turnitin dropbox available through the college's online learning management system.
The English pre-requisite for this course requires that students should have sufficient English skills to be able to sign up for the first English writing class required for an associate in arts and associate in science degrees. (JJC does not offer Bachelor's degrees.) We took this into consideration when designing the assignments. These assignments were not expected to produce research papers, however we did expect students to have a well-written introduction and conclusion, to break up their writing into paragraphs that flowed smoothly and to use correct grammar, spelling and punctuation. About one-third of each essay's grade was devoted to these basic English skills. Appendix A explains how the grading of English skills was separated from the grading of content.
Students were expected to provide a bibliography of their sources in the Modern Language Association (MLA) format, and to cite their sources within their essay. To assist students with this we directed them to Purdue University's Online Writing Lab guidelines on the MLA format (http://owl.english.purdue.edu/owl/resource/747/01/). When grading, we focused only on the bibliography, and assigned it full points if all sources were listed in sufficient detail whether or not they were in MLA format. We did not penalize students for not citing sources within their essays.
We were unable to grade drafts because there was insufficient time during the term to grade and return drafts while offering three essays. But we offered to read Essays 1 and 2 prior to submission so students could know whether or not their essay needed further work. Our handouts informed students that since no grade was being assigned at this time, there was no guarantee of a high grade if we said that their essay draft was good. We also did not check for plagiarism at this reading. Two of our students took advantage of this, and we feel this helped them to improve their essays.
However, during through the Spring 2012 semester, we encountered a student in our Life in the Universe class, where similar short essays were offered, who was displeased with this approach. We had told him that his essay was fine prior to submission, but we found the essay to be substantially plagiarized when we graded it, and therefore assigned it zero points. As a result, the student complained to our department chair and dean that our comments on his draft were misleading. Because of this student's complaint, we did not offer to read Essay 3 drafts from our introductory astronomy students, though we answered specific questions about the essay.
We also directed students to the college's writing and reading center for help with their essays. We did not ask students if they used this service.
## 3.1 Description of essays
## 3.1.1 Essay 1: Transits and Occultations involving solar system objects
The first essay was offered immediately after solar and lunar eclipses and angular size were covered in lecture. This assignment enabled students to become familiar with the terms transit, occultation, and angular size (goals IIa, b). For this essay, students were asked to describe a transit and an occultation event, excluding solar and lunar eclipses and transiting exoplanets. They were asked to search the 'Astronomy Picture of the Day' website for these events and their descriptions (http://apod.nasa.gov) and to include photographs of the events in their essay. They could use other websites if they wished. We provided links to examples of each event from Astronomy Picture of the Day within the college's online learning management system, and explained how students could use the search feature on the Astronomy Picture of the Day website to search for transits and occultations.
The conceptual portion of the essay was twofold. Students were asked to predict whether a transit would still occur and whether its duration would be longer, shorter or remain unchanged if the angular size or rotation periods of the foreground and background objects were changed while everything else stayed unchanged (they needed to address only one of these four). The duration of the transit was defined as shown in Figure 1. They were also asked to predict whether a transit would still occur and whether its duration would be longer, shorter or remain unchanged if the orbital path of the apparently smaller object were altered as shown in Figure 2 (they needed to address only one of three options shown). These two 'stand-alone' conceptual questions were designed to help them to master the geometry of transits and are related to goals IIa,b,c,d,e.
We also asked students to write about what might have not been clear to them (so we could help them with improving their understanding of these topics) and what more they would want to know about such events. This, along with the predictions, provided some uniqueness to submissions, which reduced the chance of plagiarized essays.
## 3.1.2 Essay 2: Spacecraft and Transits, Occultations and Eclipses
The second essay was offered after gravity, orbits and the electromagnetic spectrum were covered. This assignment asked students to describe a transit, occultation or eclipse event that was recorded by a spacecraft. Their event had to include at least one solar system object besides the spacecraft i.e. they could not choose a transiting exoplanet for this essay. They were asked to state in which part of the electromagnetic spectrum the event was recorded as part of their description.
Students were also asked to describe the scientific mission of the spacecraft, its orbital path and how it was maneuvered into its orbital path. The goals of this assignment involved goal IIa from Section 2 above (including identifying the three objects that were aligned in their event), learning about a spacecraft mission, the region of the electromagnetic spectrum that the spacecraft operates in and how a spacecraft is launched and directed to its destination. We explained how orbital paths are altered when this essay was given to students, and we directed them to the part of our textbook where this is covered (BENNETT et al. , 2012). We also asked them to include concepts associated with gravity such as gravitational force, acceleration, speed, velocity, etc. in this part of the essay to demonstrate their understanding of these concepts. This last part formed the conceptual portion of Essay 2.
We asked students to write about what may have been unclear to them about the science of their spacecraft and achieving the orbit of their spacecraft and/or what more they might wish to know about achieving the orbit of their spacecraft. This was done to deter plagiarism and for us to be aware of what difficulties students may have had.
We provided students with a list of transits, occultations and eclipses recorded by spacecraft, so students could be sure they picked an appropriate event for their essay. We provided links to the spacecraft websites, but allowed the students to search for information on the events on their own. This information was available through JJC's online learning management system.
## 3.1.3 Essay 3: Discovering Exoplanets using the Transit Method
The third essay assignment involved transiting exoplanets, and was offered after this topic was covered in lecture. Transiting exoplanets cannot be imaged like the events that Essays 1 and 2 involved, because the planet and its star are too far from Earth. Instead they are detected by the dip in the star's brightness when the planet is in front of the star (WINN, 2011). Since students had been introduced to transits during lecture and by doing at least one of the first two essays, we hoped they would feel comfortable exploring transiting exoplanets even though it is impossible with current instruments to obtain a photograph of the event.
Students were told to imagine that they had just discovered a transiting exoplanet and were excited to tell everyone about its discovery and properties. We asked students to choose a transiting exoplanet from the Exoplanet Encyclopedia (http://www.exoplanet.eu). In order to assist students, we provided a link to the Exoplanet Encyclopedia in JJC's learning management system and we showed students how to choose transiting exoplanets from this database.
Students were expected to briefly describe the experiment responsible for their planet's discovery and to identify the part of the electromagnetic spectrum that the experiment used in the discovery. We directed students to the list of exoplanet experiment websites maintained by the Exoplanet Encyclopedia website, so they could find the pertinent information for their exoplanet.
The assignment asked them to indicate whether the exoplanet is in the Milky Way galaxy or outside of it. They also had to include the exoplanet's orbital and physical properties; a table of this information was sufficient. They were asked to compare the shape and size of their exoplanet's orbit to the orbit of a solar system planet, so that the reader could have some insight into the exoplanet's orbital size and shape. This gave students a chance to revise Kepler's first and second laws of orbital motion, and properties of an ellipse such as eccentricity, focus, variation of orbital speed within an elliptical orbit, etc. (BENNETT et al. , 2012). The assignment asked whether the chosen exoplanet was jovian (Jupiter-like) or terrestrial (Earth-like) or neither, and to explain how they arrived at their choice. Properties of jovian and terrestrial planets had been discussed in detail when we covered our solar system during lecture (BENNETT et al. , 2012). This was an opportunity for students to explore these types of planets further.
In order to help students grasp the transit technique in the context of exoplanets, we asked students to predict how the transit duration and the parent star's brightness during the transit would change if the orbital period, orbital radius or rotation period of the planet were changed (they needed to address only one of the latter three). These questions were related to goals IIb,e,g, and Kepler's third law of orbital motion (BENNETT et al ., 2012). They were also asked to write about what more they would like to know about their exoplanet, its star and its discovery and what may not have been clear to them about the discovery of the exoplanet. This last part, together with the conceptual questions, also served to reduce the chance of plagiarized submissions.
## 3.2 Grading of Essays
Grading of each required item in an essay was done on a very good, satisfactory, unsatisfactory basis, usually using a 3, 2, 1 point scale. With this scale, it was not possible to earn a zero for it unless the essay was substantially plagiarized or a student did not turn in an essay. Also, if an essay contained one or two minor errors, it was possible for the student to earn 100% with this grading scale. As mentioned above, about one-third of the points were for appropriate English language usage including writing a good introduction and conclusion using the same scale. Detailed rubrics are available on request.
## 4. Overview of essay submissions and possible future modifications
63 students completed the course. We received 106 essays from 55 students. Of these, 23 were Essay 1 submissions, 37 were Essay 2 submissions and 46 were Essay 3 submissions. 8 students did not submit essays, and 4 submitted one essay only.
The grade distribution of the average essay score (the average of two essay scores) was as follows: A = 21.7%, B = 21.7%, C = 19%, D = 15% and F = 22.6%. The distribution of letter grades for each essay is shown in Table 1. The course grade distribution is shown in Table 2 for students who did each essay and for all students who completed the course. For all of these distributions we used the following grade scale: A = 88% to 100%; B = 78 to 87.99%; C = 68 to 77.99%; D = 58 to 67.99%; F < 58%. We elaborate further on these distributions in Section 4.5 below.
Table 1: Percentage of students scoring A, B, C, D, F grades for Essays 1, 2 and 3. The number of students who submitted each essay is shown in brackets in row 1. The conversion of percentages to letter grades is shown in brackets in the leftmost column and at the beginning of Section 4. The percentages in each column may not add up to 100% because of rounding.
| Essay Letter Grade | Essay 1 (n=23) | Essay 2 (n=37) | Essay 3 (n=46) |
|----------------------|------------------|------------------|------------------|
| A (88% to 100%) | 56.5% | 16.2% | 17% |
| B (78 to 87.99%) | 13% | 32.4% | 9% |
| C (68 to 77.99%) | 21.7% | 13.5% | 24% |
| D (58 to 67.99%) | 8.7% | 16.2% | 17% |
| F (< 58%) | 0% | 21.6% | 33% |
Table 2: The distribution of course grades for students who did Essays 1, 2 and 3 and for all students who completed the course are shown as percentages. The number of students involved in each column is shown in brackets. The percentages in each column may not add to 100% because of rounding. The letter grade ranges are given in Table 1 and at the beginning of Section 4.
| Course Letter Grade | Course grade for Essay 1 submitters (n=23) | Course grade for Essay 2 submitters (n=37) | Course grade for Essay 3 submitters (n=46) | Course grade for all students (n=63) |
|-----------------------|----------------------------------------------|----------------------------------------------|----------------------------------------------|----------------------------------------|
| A | 26.1% | 13.5% | 10.9% | 12.7% |
| B | 26.1% | 21.6% | 21.7% | 20.6% |
| C | 30.4% | 35.1% | 37.0% | 36.5% |
| D | 17.4% | 27.0% | 26.1% | 23.8% |
| F | 0% | 2.7% | 4.3% | 6.3% |
Two essays were substantially plagiarized i.e. over 60% of the essay content and bibliography matched sources according to turnitin.com, and a third essay was 48% plagiarized. The two students involved were warned that their first essay submissions violated JJC's Academic Honor Code. The substantially plagiarized essay earned a zero, while the 48% plagiarized essay was penalized 48%. The students were informed that a second plagiarized submission would be reported to the college. They were also given a
copy of the turnitin.com match so they were aware of why their essay was considered plagiarized. One of these two students submitted a second essay that was substantially plagiarized, and was reported to the college for violating JJC's Academic Honor Code. This essay was also given a score of zero.
We note that the English language writing skills (grammar, spelling, punctuation, flow, formatting, introduction and conclusion, etc.) of our students averaged 85% in Essay 1, 73% in Essay 2 and 70% in Essay 3. While the scores for Essays 2 and 3 are satisfactory, these could be improved. However, within the scope of this course, it would not be possible for us to spend class time or provide detailed guidance on this.
All essays contained descriptive and conceptual portions. The descriptive parts had concepts embedded in them, but for the details presented below and in the rest of the paper, when we mention conceptual questions, we are referring to the 'stand-alone' conceptual questions in Essays 1 and 3 detailed above, not the ones that were part of the descriptive portion. We also refer to writing about gravity related concepts in Essay 2 as the 'conceptual' part of this essay.
## 4.1 Review of Essay 1 submissions
Essay 1 met with the greatest success. The average score was 86% for this essay. 22% of the essay score was for conceptual questions, and students averaged 75% on this. Students did extremely well on the descriptive part of the essay (Goal IIa dominated this part), averaging 90%, indicating that they felt very confident explaining what is a transit and an occultation.
91% of students (21 of 23) were able to easily understand how the orientation of the orbit affects whether a transit occurs or not (see Figure 2 and goal IIe), though they did not all address how the transit duration would change. 55% of those who did were correct (6 of 11).
The effect of rotation period or angular size on the transit and its duration (goals IIc, IId) had very few correct responses. 14% of responses (3 of 22) correctly indicated that rotation period does not affect the transit or its duration, and two of these three students included excellent descriptions of how to distinguish rotation from orbital motion. 27% of responses (6 of 22) addressed how the angular size of either object affects the transit and all were correct. Half of these students correctly explained how the transit duration changes with angular size; the other half did not address this part. 27% of responses (6 of 22) showed that students were unable to distinguish between rotational and orbital motion, and therefore were not able to respond correctly. We had not anticipated this; otherwise we would have provided some pre-emptive guidance ahead of time. 27% of responses were unclear (6 of 22). 4.5% of responses (1 of 22) attempted to incorrectly use the angle between Earth's orbit and the transiting object's orbit as the angular size of the transiting object. One student, out of the 23 who did this essay, did not address this question.
In the future, in addition to explaining the difference between rotational and orbital motions, we could have an in-class activity covering the conceptual questions of Essay 1 prior to the deadline. We could follow this up with an online quiz administered through the college's online learning management system that closes at least a week
before the essay deadline. This would allow students to practice and develop confidence with these concepts before turning in the essay.
## 4.2 Review of Essay 2 submissions
The average score for Essay 2 was 70%. Students had considerable difficulty with including concepts related to gravitational force, acceleration, speed and velocity in their essay. This formed 19% of the essay score, and students averaged 57% for this, mainly because several of them did not include some examples of gravity or speed in their essays. They were able to do well on the rest of the essay and their scores averaged 74% here. Students averaged 73% when describing their event i.e. the portion of the essay that involved Goal IIa. Hence Goal IIa met with success in Essay 2.
We covered gravity concepts through lecture, accompanied by existing activities (lecture tutorial, think-pair-share questions, ranking tasks). However, it is clear that this coverage is not sufficient in helping students to write about the concepts. Here is an example of what we expect if a student chose the Cassini spacecraft: 'The gravitational force between Earth and the Cassini spacecraft when 10 km separates them is smaller than between Saturn and Cassini when they are 10 km apart because Earth is lighter than Saturn.'
In the future, we will spend some class time helping students demonstrate their understanding of these concepts through writing. We could provide students with examples. Then students could draft what they plan to include in their essay. We could spend 60 minutes of class time, divided over two classes to review these drafts so students are appropriately prepared to address this part of the essay. Helping students with writing scientific concepts could help them to more deeply understand and retain these concepts, so it would be worthwhile to spend class time on this (GREENSTEIN, 2013; SLATER, 2008).
## 4.3 Review of Essay 3 submissions
The average score for Essay 3 was 67%. Students averaged 74% on the descriptive parts of the essays, which is good. They averaged 70% on the part that included Goal IIa, indicating that this goal met with success. On concepts related to Kepler's laws, jovian and terrestrial planets and transits, students averaged 56%. These conceptual questions made up 34% of the essay grade.
Around half the students addressed how the shape and size of their planet's orbit compared to that of a solar system planet (Kepler's 1 st Law for shape and size, and Kepler's 2 nd Law for shape). Most of these gave excellent comparisons. 72% (33 of 46) addressed whether or not their planets is terrestrial or jovian. About 76% (25 of 33) of these students could correctly classify their planet as terrestrial or jovian, and around half of these provided compelling explanations for their choice using the size and masses of their planets.
The above numbers indicate that students need some help in order to be more successful with these two parts of the essay. Perhaps they could be given examples of what would be suitable to write and/or some prior practice with this. Alternatively, this part could be re-worded so that the student is asked to imagine visiting their exoplanet.
Then they can describe what they experience as the planet rotates and orbits, whether they experience seasons, whether they could stand on the surface of the planet, whether their planet has an atmosphere and whether they could withstand the temperature at the surface of the planet. As long as they correctly addressed at least three of these very well and compare them to the Earth or a solar system planet of their choosing, they would receive full points. Such an approach might appeal more to students and could help them with turning in better essays. It would also serve to keep the chance of plagiarized submissions low.
With the transit duration related question in Essay 3, there were 48 responses from 41 students; 4 students addressed more than one of the options presented and 5 did not address this. 66% of responses (35 of 48) for this part of the essay were correct, which is very encouraging. 62.5% of responses (30 of 48) addressed how the transit duration is affected by increasing the orbital period (Goal IIf). 73% of these answers (22 of 30) were correct - the transit duration increases because the orbital speed decreases when the orbital period increases according to Kepler's 3 rd Law. All responses explaining the effect a decrease in orbital radius has on transit duration (17% or 8 of 48) were correct - the transit takes less time because the planet orbits faster when closer to the star according to Kepler's 3 rd Law (Goal IIf). 12.5% of responses (6 of 48) involved the option related to the planet's rotation period (Goal IIf). 83% (5 of 6) of these correctly answered that increasing the rotation period of the planet does not affect transit duration. The remaining 8% of responses (4 of 48) contained answers that were unclear.
When addressing what affects the dimming of the star during the transit for the options provided, 39 students provided one response each, while 15% of students who did this essay (7 of 46) did not attempt this question. 36% of the responses were correct (14 of 39), which is disappointing (Goal IIg). 13% of responses (5 of 39) were unclear. 23% of responses (9 of 39) addressed the increase in rotation period of the planet, and all of these students correctly noted that rotation period does not affect the dimming of the star. The other two options made up 64% of responses (25 of 39). 21% of responses related to the effect of decreasing orbital radius on the star's dimming during a transit, (3 of 14), correctly explained that the star's brightness drops less when the planet's orbital radius is decreased because the planet's angular size decreases for the observer. 18% of responses involving the effect of an increase in orbital period on stellar dimming during a transit, (2 of 11), correctly indicated that the star's brightness is dimmed more when the planet's orbital period increases because the planet's angular size increases for the observer.
This last question is possibly the least successful of the conceptual questions. 21% of responses (8 of 39) indicated that the question was interpreted as asking for how much time the star's dimming is affected, not how the intensity of the dimming changes, despite having a figure included in the handout to accompany this question. Such an interpretation allowed students to arrive at the correct answer for the wrong reasons - something that would not have been revealed if this were presented as a multiple-choice question. Perhaps this question should be removed from the essay assignment and offered as an in class activity, and/or re-worded.
As with the transit related concepts in Essay 1, if we provided students with practice with the concepts in Essay 3 ahead of time, it is likely that more of them would
be able to arrive at the correct conclusions to include in their essays. We could provide examples in the essay handout and/or we can have students do similar questions in class so students know what is expected of them. We could offer quizzes administered through the course learning management system at least a week prior to the essay deadline to follow up on activities done in class. We could also couple these with reducing the number of conceptual questions in the essays (e.g. remove the one about the dimming of the star in Essay 3). Given that other course material needs to be covered during the semester, it may not be possible to devote class time to all of the concepts we have included in Essays 1 and 3, and we would adjust the essays accordingly.
## 4.4 What did students learn from the essays?
From these essays, we find that students successfully learned about what is a transit, what is an eclipse and what is an occultation, and they recognized that two objects and the Earth (or a spacecraft) need to be aligned in order to observe such events (Goal IIa). They also understood how the orientation of the orbit on the sky affects the occurrence of transits (Goal IIe).
Around a quarter of the students were unable to distinguish between rotational motion and orbital motion in Essay 1, though not all students chose to address the concept of rotation period, so this is a lower limit. After Essay 1 submissions were returned to students, we explained to all students the difference in these two motions with demonstrations, including asking students to stand and spin in place to emphasize rotation. It is encouraging to note that this confusion was reduced to 2% in Essay 3, though again this is a lower limit since not all students chose to address the effects of rotation period on transit duration and decrease in stellar brightness during the transit.
## 4.5 Determining the difficulty level of the essays
Table 1 shows the break down of essay grades for the three essays, and Table 2 shows the distribution of course grades for students who did each essay as well as the overall course grade distribution. The course grade distributions for students doing Essays 2 and 3 are similar, and these are similar to the course grade distribution of all 63 students who completed the course. The course grade distribution for Essay 1 contains more As and Bs compared to the same distributions of those who did Essays 2 and 3. Therefore, it is not unexpected that Essay 1 has the highest average score. However, we cannot conclude that Essay 1 is the easiest essay based on the essay grade distribution.
Since students found the conceptual portion of each essay the most difficult, we will use the fraction of the essay score allocated to 'stand-alone' concepts as the criterion that determines the difficulty level of each essay. 22% of Essay 1's score was assigned to concepts, and 19% of Essay 2's score was assigned to concepts. Since almost the same percentage of both essays was allocated to concepts, we can assume that the difficulty level of these two essays is comparable. The fact that the average score of Essay 1 submitters was much higher than that of Essay 2 submitters can therefore be attributed to the difference in abilities between the groups of students who did these essays. We clearly see this difference in their course grade distribution in
Table 2. There were 8 students who did both essays, and their average scores were comparable, at 87%. However, this is not a large enough sample from which to draw conclusions about how the essays compare in difficulty. We will conclude solely from the fraction of the score allocated to concepts that the two essays are likely to be comparable in difficulty.
Essay 3 had 34% of its score assigned to concepts, which is a lot more than for either of the other essays. This would indicate that Essay 3 was the hardest of the essays, though we did not intentionally make this essay the hardest. We had not anticipated that students would find all of the concepts in this essay so challenging, otherwise we have reduced the number of concepts before offering this essay. The distribution of course grades for students doing Essay 3 is similar to those who did Essay 2. However, Essay 3 scores show that 50% of the scores are As, Bs and Cs, compared to 62% for Essay 2 scores. This is another indication that Essay 3 was more difficult than the others.
In the case of Essays 2 and 3, 29 students did both, and their average scores were 67% for Essay 2 and 66% for Essay 3. This does not agree with the argument that the essays have differing difficulty. But the averages are close because some of the students who did both essays improved from Essay 2 to 3, while others did worse on Essay 3 than Essay 2, and the rest stayed unchanged. We had 14 students who did Essays 1 and 3, and their scores averaged 85% and 72% respectively. This does point to Essay 3 being more difficult, but since the sample size is small, it would not be reliable to draw conclusions from.
Hence, we use the fact that Essay 3 had a larger number of concepts and that fewer students earned As, Bs and Cs in this essay compared to Essay 2 (with both groups of students performing similarly in the course) to conclude that Essay 3 was the hardest of the three essays. We did not intend to make this essay more difficult than the others.
## 5. Students' attitudes towards essays
We constructed a feedback survey on the essay assignments to find out how students felt about them, so that future essay assignments could be better structured. The survey contained five statements, which requested responses of 'Yes,' 'No', or 'Maybe.' The survey contained a sixth and final question requesting a written response. 47 students responded to the survey, including two who did not submit essays.
We told students that participation in the survey was voluntary, and that their responses would not be tied to their names. We offered a small amount of extra credit for participating as the survey was done outside of class, through the college's online learning management system.
45 out of 55 students who submitted essays provided us with feedback. The course and essay grade distribution of these 45 students are almost identical to the same distributions for all those who submitted essays. This, combined with the fact the sample size is reasonably large, leads us to conclude that the responses received do not have any a priori bias.
The results of how students felt about the five statements are shown in Table 3.
Table 3: Student responses to the first five questions of the essay feedback survey. Students were asked how they felt about the five statements listed above.
| How do you feel about the statements in questions 1 - 5? | Yes | No | Maybe |
|---------------------------------------------------------------------------------------------------------------------------------------------------------------------|-------|-------|---------|
| Q1. The essay handouts contained sufficient details about what your instructor expected. | 66% | 17% | 17% |
| Q2. The essay grading rubrics were fair. | 70.2% | 23.4% | 6.4% |
| Q3. I felt that the essays increased my interest in astronomy. | 21.3% | 57.4% | 21.3% |
| Q4. The educational value of essays is such that I would recommend them in future offerings of this course. | 27.7% | 48.9% | 23.4% |
| Q5. I would like to witness the transit of Venus on 5th June 2012 if the weather and my schedule permit. The transit starts at 5:10pm. The Sun sets before it ends. | 63.8% | 14.9% | 21.3% |
Two-thirds of students said the handouts provide sufficient information to create their essays, and 70% find the grading rubrics to be fair. This is encouraging.
We feel that our most significant results from these questions is that 21% of students felt the essays increased their interest in astronomy, and that 49% of students felt that the essays were not educationally beneficial to offer in the future. From Table 1, we see that students who did Essay 1 earned a greater fraction of As, Bs and Cs in this essay compared to those who did Essays 2 and 3. We also can tell from comparing Tables 1 and 2, that for Essays 2 and 3, more students scored lower on essays than in other parts of the course, indicating that the essays were more challenging than other parts of the course. Since there were more students who did Essays 2 and 3, it is likely that 28% of students would find the essays educationally beneficial, 21% would feel that the essays increased their interest in astronomy, and a majority would feel the opposite way.
Table 4, shows the average essay grade distribution of our 47 survey respondents both as percentage and raw numbers in the second row. The table also shows the percentage of students responding 'Yes,' 'No' and 'Maybe' to questions 3 and 4 of the survey for each essay letter grade.
In the case of question 3, the influence of essays on student interest in astronomy, 21% of students report an increase in interest in astronomy across all grades i.e. students who scored higher in essays did not feel more interested in astronomy than those who scored lower. This probably indicates that it might be difficult to increase interest in astronomy through this particular set of essays, or through essays that emphasize concepts.
Around 40% of those with higher grades find the essays to be educationally beneficial according to Table 4. 22% of those earning D grades find the essays to be educationally beneficial and this drops to zero for the group of students who earned failing essay grades. If more students could earn higher grades for their essays, it is possible that a majority of students would find the essays educationally beneficial and recommend them to be offered to future introductory astronomy students instead of just 28%.
Table 4: The distribution of average essay grades for two essays for all survey respondents is shown in the second row as percentages, along with the number of respondents in each grade. The breakdown of Yes, No, Maybe responses for questions 3 and 4 from the essay feedback survey for each essay letter grade are shown as percentages in the third and fourth rows respectively. The percentages in each cell of rows 3 and 4 may not add to 100% because of rounding.
| Average Essay Grade | A | B | C | D | F |
|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------|---------------------|--------------------|---------------|---------------------|------------------|
| Percentage of survey respondents, with number of respondents in brackets. | 19% (n=9) | 28% (n=13) | 11% (n=5) | 19% (n=9) | 23% (n=11) |
| Yes, No, Maybe percentages for Q3 of survey, 'I felt that the essays increased my interest in astronomy,' for each essay grade | 22.2%, 33.3%, 44.4% | 23%, 38.5%, 38.5% | 20%, 60%, 20% | 22%, 78%, 0% | 18%, 82%, 0% |
| Yes, No, Maybe percentages for Q4 of survey, 'The educational value of essays is such that I would recommend them in future offerings of this course,' for each essay grade | 44.4%, 22.2%, 33.3% | 38.5%, 46.2% 15.4% | 40%, 20%, 40% | 22.2%, 66.6%, 11.1% | 0%, 72.7%, 27.2% |
In Spring 2008, when we asked our students to do two group poster projects on telescopes, a majority of students, around 45%, felt the projects were educationally beneficial and should be offered again. This compares well with how those who earned As, Bs and Cs on essays in Spring 2012 feel. A majority of our Spring 2008 students, around 40%, also felt that the projects increased their interest in astronomy. The projects in Spring 2008 were less conceptually oriented than the ones presented here and they featured a different topic, so at least these two factors would have affected their interest in astronomy (D'CRUZ, 2009). Perhaps working in groups was also beneficial to students versus working individually.
Our survey indicates that 64% of students would have liked to view the 2012 transit of Venus, which is encouraging. When the event neared, we informed students about the opportunity to view the transit at Joliet Junior College. However, since the semester had ended a few weeks earlier, only one student came to the viewing event.
## 5.1 Responses to the sixth survey question
The sixth and last survey question was:
'Q6. What is the most important thing the instructor needs to know before assigning similar essays next semester in order to improve the assignments?'
44 respondents wrote 68 comments for this question. The main trends of the comments are:
- A. 28% of the comments indicated that expectations need to be made clearer, there needs to be less requirements, instructions need to be simplified and students need to be prepared better to meet the expectations. 66% of students felt the handouts clearly explained the expectations according to the first question in the survey, which is inconsistent with these comments. However, it is clear that some students
needed help in constructing their essays, and that some students found the essays to be more difficult than other parts of the course. In Section 4, we mentioned that we would provide students with suitable activities to prepare them for at least the conceptual parts of the essays in the future. Two comments suggested that example essays be provided so that students have a better idea of expectations. This can also be done in the future. We hope that these alterations will help students to be better prepared to write essays.
- B. 10% of the comments indicated that the essays are fine as is and expectations are clear. This is reassuring.
- C. 6% of comments indicated that other essay topics should be considered. The particular theme we chose for Spring 2012 resulted in essay topics that were somewhat narrowly focused. We can certainly provide a wider range of topics for future essays.
- D. 6% of comments indicated that our grading should be more lenient. We reported in Table 3 that 70% of students found the grading rubrics to be fair, which is at odds with these comments. We do not feel the grading was biased against students as over 62% of essays were assigned A, B, or C grades. However, as mentioned in Section 4, Essay 3 unintentionally turned out to be the hardest of the three, with the other two being comparable. Since more students did Essay 3 than the other two, it is likely that some students would feel the grading was not lenient. A comment related to this concern mentioned that we should remind students of essay requirements so they can check whether their essays contain the necessary pieces prior to submission. This student also wrote that the reminder could reduce the displeasure some felt upon receiving their essay grade. Since the handouts contain the requirements and grading rubrics, providing such a reminder can be easily done in the future.
- E. The remaining 50% of comments covered a wide variety of topics. Perhaps the most relevant of these to future essay offerings were to make each essay worth more than 2.5% of the course grade and to provide more time to complete the essays.
## 6. Test and final exam questions related to course theme
## 6.1 Test questions related to course theme
We wanted all students to be able to master goals I and II, that were described in Section 2. But since goal I was not included in the essays, and since not all students did all the essays, we felt that the inclusion of a few related questions on tests and final exams would be a reasonable way to assess their mastery of these goals. We put theme related questions on our tests only after returning graded essays and providing students with feedback on them. As a result, we could include four theme-related multiplechoice questions on only the second test, because we had graded and returned only Essay 1 at this time. (Our tests and final exam contain only multiple-choice questions.)
The test questions addressed goals IIa and IIb. We always have questions on solar and lunar eclipses, and on exoplanets on our tests, so we report here only on the additional questions that were connected to the course theme. The questions and the percentage of students who answered them correctly are given below:
- 1. The Moon's orbital distance from the Earth is slowly increasing. The Moon's angular size in the future will be larger/smaller/unchanged compared to currently. (Goal IIb)
- 48% answered this question correctly, stating the angular size will be smaller.
- 2.
- Which of the following best describes an occultation? (Goal IIa) (a) an object of smaller angular size passes in front of an object of larger angular size
- (b) an object of larger angular size passes in front of an object of smaller angular size
- (c) a lighter object passes in front of a heavier object
- (d) a heavier object passes in front of a lighter object
- 67% correctly chose (b)
- 3. During an annular solar eclipse, the Moon has a smaller angular size than the Sun, and is unable to completely block out the Sun. This type of eclipse could also be called (Goal IIa)
- (a) a transit of the Moon
- (b) an occultation of the Moon
- (c) a transit of the Sun
- (d) an occultation of the Sun
- 49% correctly chose (a)
- 4. What effect or effects would be most significant if the Moon's orbital plane were exactly the same as the Earth's orbital plane? (Goal Ib)
- (a) Solar eclipses would be much rarer.
- (b) Solar eclipses would be much more frequent.
- (c) Solar eclipses would occur with the same frequency as currently.
- 88% of students correctly answered (b).
## 6.2. Transit related final exam questions
Since we wished to highlight the transit of Venus during the course, we felt that the final exam should contain questions regarding this topic (goal I). We also included questions related to goal II. The final exam was comprehensive, which meant we were limited to including five multiple-choice questions addressing these items. These were in addition to our regular coverage of eclipses and exoplanets. To help students prepare for this part of the exam, we provided a practice quiz with similar multiple-choice questions through our learning management system. Students could attempt this quiz as many times as they wished before the exam. There were 6 submissions from 4 students for this quiz. 62 students did the final exam, out of 63 who completed the course.
Between 72% and 82% of students answered the following final exam questions correctly (the exact wording of the questions is available upon request):
- (a) What is meant by the transit of Venus? (Goal Ia)
- (b) How does Venus' angular size change as its distance from Earth decreases? (Goal IIb)
- (c) How does the time taken for Jupiter's moon, Io, to transit across Jupiter's diameter change when Jupiter's rotation period is increased, assuming everything else stays unchanged? (Goal IIc)
- (d) If Io's orbital radius increased while Io is centered on Jupiter, will it block more/less/the same amount of Jupiter's disk, assuming everything else stays unchanged? (Io is a moon of Jupiter.) (Goal IIg)
52% of students were able to correctly answer why the transit of Venus occurs so rarely (Goals Ib). The question focused on the fact that Venus' orbit is inclined to the Earth's orbital plane, not on the relative orbital speeds of Earth and Venus (we did not emphasize the latter much in our class). Since students had done very well on the test question concerning the effect of the inclination of the Moon's orbit on the frequency of solar eclipses, we had hoped that they would extend that knowledge to explain the rarity of the transit of Venus. However, this was not the case.
The test and final exam questions showed that students were easily able to comprehend what is meant by a transit and occultation and that their understanding of how the angular size of an object depends on distance improved over the course of the semester.
## 7. Implications for future essay offerings
Sections 4 and 5 have shown that writing about concepts at a level that demonstrates deep conceptual learning is difficult. The stand-alone transit concepts included in the essays required more guidance and practice than was provided through our brief lectures in order for students to master them. We will provide suitable activities in the future to help students feel more comfortable and confident with these concepts when constructing their essays. Concepts like Kepler's Laws and gravitational force were covered through lecture tutorials, think-pair-share questions, and ranking tasks. When students were asked to demonstrate their understanding of these through writing, there were several who were not ready for these, as most of our usual assessments do not involve writing in such a manner. In the future, we would offer students more guidance and practice with writing about scientific concepts.
Based on the results presented, we feel that we can suitably modify the essay assignments so more future students find them to be educationally beneficial and more can be persuaded to increase their interest in astronomy through the essays. We feel that while essays can be challenging, offering such writing assignments helps students to think about astronomy topics and concepts more deeply (GREENSTEIN, 2013; SLATER, 2008) and they complement the other learner-centered activities that our students do (lecture tutorials, think-pair-share questions, ranking tasks, etc.). However, one has to consider how much course time can be devoted to a course theme versus the rest of the topics that need to be covered, and this would limit the number of themerelated concepts that could be explored.
In the future, when such essays are offered, if more class time is to be spent on concepts and how to incorporate them into essays, it may be better to offer only two essays instead of three, and have all students do both. Fewer additional in-class activities will be needed and regular coverage of course topics will be minimally impacted if only two essays are offered. Another advantage is that students can be given more than two weeks to work on an essay. Pushing back the deadline will ensure sufficient time for relevant in-class activities prior to submission.
We also recognize that we unintentionally made Essay 3 more difficult than Essays 1 and 2. (Essays 1 and 2 are probably comparable in difficulty.) In the future, we will attempt to keep the difficulty level of the essays comparable by keeping around 20 - 25% of the essay score or less for conceptual questions, and not ask students to address too many of these within a single essay.
We will also allocate more of the course grade to essays so that students recognize the need to devote sufficient time and effort to them in order to earn a score of C or higher. They will feel better rewarded for the time and effort they put into such projects if they form a larger part of the course grade. Since we have mentioned that the essays could be simplified, it is not clear how much more they would be weighted in the future. We would possibly double the weighting to 5%. This would cause a reduction in the weighting of our tests.
We consider the above modifications as the most important ones for future offerings of this set of essays. In the broader context, if other astronomy course themes were adopted for conceptually oriented essays, we would accordingly develop in-class activities or use existing activities for concepts similar to those in Essays 1 and 3. We would help students with writing about concepts similar to those in Essay 2. We would likely offer two essays only and ensure that the number of concepts they cover not overwhelm our students.
Other changes that would be easy to implement and possibly improve student success with essays would be:
- a) to make sample essays available so students have an idea of what is expected for an A or a B grade. In the past, with group projects, we have shown students a high quality and a low quality assignment and asked them which one they would assign the higher grade to. This helps them to be aware of what to include and what to avoid in a project.
- b) to remind students about the details expected and how these are connected to the grading rubrics, as suggested by one of our students.
- c) to ask JJC's writing and reading center's tutors to visit our introductory astronomy classes when essays are offered to provide guidance and possibly a sample essay. In Spring 2013 and Fall 2013, we asked JJC's writing and reading center director to visit our class to provide guidance on writing essays to our Life in the Universe students. The director also had one of the center tutors create a sample that was given to all students.
- d) to modify our handouts for each essay so fewer students feel confused and overwhelmed by them. We may make them more closely aligned with the essay guidelines suggested by JJC's writing and reading center, or those used in English classes.
Transits, occultations and eclipses will continue to be the news, and we would like to cover these through lecture and essays (or suitable projects) in the future as we have done here. The next two opportunities related to the theme presented here will be when the transit of Mercury occurs on 9 May 2016 (http://eclipse.gsfc.nasa.gov/transit/transit.html) and when the total solar eclipse of 21 Aug 2017 occurs (http://www.mreclipse.com/Special/SEnext.html; http://eclipse.gsfc.nasa.gov/SEplot/SEplot2001/SE2017Aug21T.GIF). The former will be entirely visible from Joliet, IL from around 6:12am till 1:42pm local time, weather permitting (http://eclipse.gsfc.nasa.gov/transit/transit.html). A telescope with a solar filter will be needed to see the event. The latter will be visible (weather permitting again) from Illinois and other parts of North America. The eclipse will have the greatest duration close to Hopkinsville, IL, with the Sun being completely blocked for 2 minutes 40.1 seconds, at this location. This will occur around 1:15pm on 21 Aug 2017 (http://eclipse.gsfc.nasa.gov/SEplot/SEplot2001/SE2017Aug21T.GIF). Solar filters will be needed to view the eclipse safely.
In the broader context, we would like to have other course themes focusing on upcoming or ongoing astronomical events or on some broad aspect of the course like stars, galaxies, etc. so we can help non-science major students to learn some of the associated concepts more deeply. We feel that devoting class time to such a theme is worthwhile, provided there is sufficient time to cover the rest of the course material with minimal impact, and if a majority of students feel that writing such essays is educationally beneficial.
## 8. Conclusions
Through theme-related questions on one test and the final exam, we found that students averaged 63% when asked what is meant by a transit, an occultation and the 'transit of Venus.' Since this average is higher than the 52% we predicted in Section 2, it indicates that two of our main goals, Ia and IIa, were met through lecture.
The test and final exam questions showed that students' understanding of how an object's angular size is affected by its distance from Earth improved as the semester progressed. They also did well on questions related to how rotation period affects the duration of a transit and how much of a background object is blocked when the distance of the foreground object is changed in a transit.
We also used short theme-related essays to help students further explore the theme and related concepts. The essays were meant to be more of a teaching tool than primarily an information gathering exercise for students. The essays showed that students scored at least 70% when explaining what is a transit, an occultation and an eclipse and the need for two objects and an observer or spacecraft to be in a straight line for these events to occur. This means that our second major goal, IIa, was met through essays, in addition to lecture.
Students' understanding that the rotation period of an object does not affect a transit improved from Essay 1 to Essay 3. However, the 'stand alone' transit concepts in Essay 3 were harder to grasp than the concepts in Essay 1. Students did well with the more descriptive parts of the essays.
Students felt they needed more guidance and preparation in order to write successful essays. In the future, we will provide the necessary scaffolding, such as inclass activities and online quizzes, to help students be more successful with the 'standalone' conceptual portions of essays, as those were what students found to be the most difficult. We will also allocate around 20 - 25% or less of each essay score to deep conceptual learning, and limit the number of such concepts in the future. This will ensure that the essays are not so difficult that they overwhelm students, and will increase the chance for student success.
The third major goal, to increase student interest in astronomy through short essays, did not meet with success. 21% of students felt that the essays increased their interest in astronomy. 28% of students felt that the educational value of the essays was such that they could be offered again, while 49% disagreed.
The fraction of students whose interest in astronomy increased due to these essays was essentially independent of their average essay score. This indicates that it may not be easy to increase interest in astronomy through the set of essays presented here.
40% of students who had average essay scores in the A, B or C range viewed the essays as being educationally beneficial and recommended that they be offered again. However, only 20% of those whose essays earned Ds and none of those whose essays earned Fs felt positively towards the essays. If more students earned As, Bs or Cs on their essays, then it is possible that a majority of students would view them as educationally beneficial and as worthwhile assignments for future students.
In conclusion, we feel that using short thematic essays in the introductory astronomy college course as a teaching tool is worthwhile provided a majority of students feel they are educationally beneficial. We hope that future essay offerings and associated activities will be more rewarding to students, and will motivate more students to be interested in astronomy.
## Acknowledgements
We thank our Spring 2012 face-to-face ASTR 101 students for generously providing feedback on the essay assignments, and for agreeing to let us use their essays and test answers in this project. We are very grateful to Tim Slater for reading an earlier draft and providing valuable comments that have improved the paper. We thank Vikram Dwarkadas, whose comments on a subsequent draft improved the paper too. We are also very grateful to the valuable comments and suggestions of the two anonymous referees, which have resulted in a much better version of this paper.
## References
BENNETT, J.; DONAHUE, M.; SCHNEIDER, N.;VOIT, M. The Essential Cosmic Perspective . 6.ed. San Francisco, CA: Pearson Addison-Wesley, 2012.
BORUCKI, W.; et al . Kepler Planet Detection Mission: Introduction and First Results. Science, Washington D.C., v. 327, p. 977-980, 2010.
D'CRUZ, N. Non-Science Major Students' Attitudes towards Group Projects in Astrobiology . In: 216. Meeting of the American Astronomical Society; Miami, FL, 2010.
D'CRUZ, N. Students' Reactions to Group Projects on Telescopes in ASTR 101 . In: 213. Meeting of the American Astronomical Society; Long Beach, CA, 2009.
ESPENAK, F. 2012 Transit of Venus. In: CHAPMAN, D. (Ed.). Royal Astronomical Society of Canada Observer's Handbook 2012 . Toronto: Royal Astronomical Society of Canada, 2011. p.144-152.
GREENSTEIN, G. Writing is Thinking: Using Writing to Teach Science. Astronomy Education Review , Washington D.C., v . 12, p. 010401, 2013. Available from: <http://aer.aas.org/resource/1/aerscz/v12/i1/p010401\_s1>, Accessed on 18 Aug. 2013 .
LIGHTMAN, A.; SADLER, P. Teacher Predictions versus Actual Student Gains. Physics Teacher , v. 31, p. 162-167, 1993.
LITTMANN, M.; ESPENAK, F.; WILLCOX, K. Totality : Eclipses of the Sun. 3. ed. New York: Oxford University Press, 2008. 341p.
PASACHOFF, J. M . Solar Eclipses and the International Year of Astronomy . In: 214. Meeting of the American Astronomical Society, Pasadena, CA, 2009.
PASACHOFF, J. M. Transit of Venus: Last Chance to See. Nature , London, v. 485, 303-304, 2012.
PASACHOFF, J. Solar Eclipses for Science and for Public Education. In: DE LEÓN, M.; DE DIEGO, D. M.; ROS, R. M. (Eds). AIP Conference Proceedings. New York, NY, v. 1283, p. 40-45, 2010.
PRATHER, E. E.; SLATER, T.F.; ADAMS, J.P.; BAILEY, J.M.; JONES, L.V.; DOSTAL, J.A. Research on a Lecture-Tutorial Approach to Teaching Introductory Astronomy for Nonscience Majors. Astronomy Education Review , v. 3 n. 2, p. 122136, 2005.
SLATER, T. F. Engaging Student Learning in Science Through Writing Tasks. The Physics Teacher , College Park, MD, v. 46, p. 123-125, 2008.
## SLATER, T. F.; ADAMS, J. P. Learner Centered Astronomy Teaching : Strategies for ASTRO 101. Upper Saddle River, NJ: Pearson, 2003. 167p.
WINN, J. N. Exoplanet Transits and Occultations. In: SEAGER, S. (Ed.) Exoplanets . Tucson, AZ: University of Arizona Press, 2011, p. 55-77.
## APPENDIX A: Grading essays for English language skills versus science content.
The science content expected to be in each essay has been detailed in Section 3. The English language skills that were graded for each essay were:
- 1) a well-written introduction
- 2) a well-written conclusion
- 3) spelling, punctuation, grammar, ease of reading (or flow), font size, text spacing, essay length, formatting (is the text broken into reasonably sized paragraphs?) were graded together.
About one-third of each essay's grade was devoted to English language skills.
The main error with writing an introduction was that the introductory paragraph would start either abruptly or with a very short 'introduction' (one or two sentences) followed by the main essay content. This means the introduction is essentially missing.
The most frequent error with the conclusion was that either the concluding paragraph would be missing or it would be so short it would not sufficiently serve as a conclusion to the essay.
Both these errors were easy to identify and separate from the scientific content. If there were content errors in either introduction or conclusion, then these was graded as part of the content. The chance that a student had points deducted twice (i.e. for science content and English language) for the same error in the introduction and conclusion was very small.
Spelling, punctuation, font size, text spacing, essay length, formatting errors can be easily identified as separate errors from content, so there was no ambiguity in grading these.
The content asked for was specific, so errors in flow and grammar were not easily confused with errors in content. The flow of the writing can affect the content presentation, but choppiness in the writing is more of a distraction to the reader than a problem with identifying what message the writing attempts to convey. Similarly, it was usually clear whether or not the required content was present and satisfactorily addressed even if there were grammatical errors.
If confusing or unclear sentences were in the essay, these were usually graded as content related.
One point (out of 30 - 35 total points for each essay) was usually deducted for a combination of three errors associated with spelling, punctuation and grammar.
## Noella L. Dcruz
Font size and text spacing are related to essay length, so one point (out of 30 35 total points for each essay) was deducted for an essay that was around half a page too short. An essay with multiple unusually long paragraphs had 1 point deducted for this error.
Based on the above, it is unlikely points were deducted twice for the same error. Therefore it is unlikely that essay scores were lower than they should have been because both science content and English language skills were graded.
The average scores for English language skills in Essays 2 and 3 were higher, but not much higher than the average scores for these essays, indicating that the science content of these essays was somewhat more challenging for these students than their English language skills. The average essay scores for Essay 1 and its English skills were comparable.
| null |
[
"Noella L. Dcruz"
] |
2014-08-04T23:14:55+00:00
|
2014-08-04T23:14:55+00:00
|
[
"physics.ed-ph",
"astro-ph.IM"
] |
Essays on Eclipses, Transits and Occultations as Teaching Tools in the Introductory Astronomy College Course
|
We occasionally include projects in our learner-centered introductory
astronomy college course to enable non-science major students explore some
astronomical concepts in more detail than otherwise. Such projects also
highlight ongoing or upcoming astronomical events. We hope that students will
feel more interested in astronomy through projects tied to astronomical events.
In Spring 2012, we offered short essays focused on eclipses, transits and
occultations to promote the rare transit of Venus that occurred on June 5th,
2012. We asked students to write two short essays from three that were offered.
The essays contained descriptive and conceptual parts. They were meant to serve
as teaching tools. 62% of 106 essays from 55 students earned A, B or C grades.
21% of 47 feedback survey respondents felt the essays increased their interest
in astronomy. 49% of respondents felt that the essays were not educationally
beneficial and should not be offered again. The most common written response to
our survey indicated that students need more guidance and better preparation in
writing successful essays. Since students found the conceptual parts of the
essays difficult, in the future we will provide relevant activities prior to
essay deadlines to help students create successful essays.
|
1408.0837v1
|
## J -Trace Identities and Invariant Theory
Allan Berele Department of Mathematics ∗ DePaul University Chicago, IL 60614
August 6, 2014
## Abstract
We generalize the notion of trace identity to J -trace. Our main result is that all J -trace identities of M n,n are consequence of those of degree 1 2 n n ( +3). This also gives an indirect description of the queer trace identities of M E n ( ).
KeywordsTrace identities, J -trace-identities, queer trace identities, invariant theory, Sergeev algebra
## 1 Introduction
## 1.1 Traces, Supertraces and J -Traces
Let F be a field. A trace function on an F -algebra A is a linear map from A to a commutative F -algebra, typically the center of A , such that tr ( ab ) = tr ( ba ) for all a, b ∈ A . Even if the image of the trace map is not in the center of A , we will always take it to be in a commutative algebra that acts on A . One generalization that has been studied is that in which A is a Z 2 -graded algebra, or superalgebra, A = A 0 ⊕ A 1 . In this case one studies supertraces, which are linear, homogeneous (degree preserving) maps from A to a supercommutative ring which has a graded action on A such that str ab ( ) = ( -1) αβ str ba ( ) for all homogeneous a ∈ A α , b ∈ A β . By the homogenuity of the supertrace we also have a · str b ( ) = ( -1) αβ str b a ( ) . A basic example is A = M E n ( ), where E is the infinite Grassmann algebra with its usual grading making it into a supercommutative algebra, and A i =
∗ Partially supported by a Faculty Research Grant from DePaul University
M E n ( i ) for i ∈ Z 2 . If a = ( a ij ) ∈ M E n ( ) then the supertrace of can be defined as str a ( ) = a ii .
∑ There are other gradings and supertraces defined on M E n ( ). Let ω be the involution on E defined by ω e ( ) = ( -1) α e for all e ∈ E α . Given k + = /lscript n , we embed E into M E n ( ) by mapping each e to the diagonal matrix
$$\omega _ { k, \ell } ( e ) = d i a g ( \underbrace { e, \dots, e } _ { k }, \underbrace { \omega ( e ), \dots, \omega ( e ) } _ { \ell } ), \\ M ( k, \ell ) _ { 0 } \text{ to be the elements of } M _ { n } ( E ) \text{ which comm} \\ \cdots \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{} \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{$$
and define M k,/lscript ( ) 0 to be the elements of M E n ( ) which commute with E , and M k,/lscript ( ) 1 to be the ones that supercommute. It is not hard to see that M k,/lscript ( ) 0 consists of all n × n matrices of the form ( A B C D ) in which A ∈ M E k ( 0 ), D ∈ M E /lscript ( 0 ), and B and C have entries in E 1 ; and M k,/lscript ( ) 1 is the opposite. The algebra M k,/lscript ( ) 0 is generally denoted M k,/lscript . The supertrace on M k,/lscript ( ) is defined on homogeneous elements of degree d as
$$\ s t r ( a _ { i j } ) & = \sum _ { i = 1 } ^ { k } a _ { i i } - ( - 1 ) ^ { d } \sum _ { i = k + 1 } ^ { n } a _ { i i }, \\. \quad \dots \quad \dots \quad \dots \quad.. \quad.. \quad \dots \quad \dots$$
and extended to all of M k,/lscript ( ) by linearity. It is useful to consider M k,/lscript ( ) to be an E -algebra by letting each e ∈ E act as ω k,/lscript ( e ). With this action we have that the supertrace is E -linear in the natural sense that str ea ( ) = e · str a ( ) for all a .
In this paper we will be interested in algebras A with supertrace with a degree one element j such j 2 = 1. The principal example is M n,n ( ) and j = ( 0 I n I n 0 ) . For any such algebra A 1 = jA 0 = A j 0 and A 0 = jA 1 = A j 1 . Moreover, if a ∈ A 0 commutes with j then
$$s t r ( a ) = s t r ( j ^ { 2 } a ) = - s t r ( j a j ) = - s t r ( a )$$
and unless the characteristic is 2 this implies str a ( ) = 0.
## 1.2 Universal Algebras
Let X be an infinite set. Define F X,tr 〈 〉 to be the commutative F -algebra generated by the symbols tr ( u ) for u in the free algebra F X 〈 〉 with the relations tr ( αu ) = αtr u ( ), tr ( u + v ) = tr ( u ) + tr ( v ) and tr ( uv ) = tr ( vu ) for all α ∈ F , u, v ∈ F X 〈 〉 . There is an obvious trace map from F X 〈 〉 to F X,tr 〈 〉 with the universal property that given any algebra homomorphism F X 〈 〉 → A , where A is an algebra with trace A → C , there is a unique map F X,tr 〈 〉 → C making the following diagram commute:
Let F X,tr 〈 〉 be the tensor product F X,tr 〈 〉 ⊗ F X 〈 〉 . There are two traditional approaches to defining a trace function on F X,tr 〈 〉 , the issue being how to define tr ( tr ( u )) = tr ( u tr ) (1). Either one defines a new commuting variable γ = tr (1), or else chooses an integer n and sets tr (1) = n in order to study algebras with tr (1) = n . At any rate, if A is an F -algebra with trace function to the central subalgebra C , (and with tr (1) = n if we are in the latter case), then the above universal property takes the simpler form that given any set theoretic map X → A there is a unique extension to a homomorphism F X,tr 〈 〉 → A which preserves trace. Elements of F X,tr 〈 〉 are called trace polynomials or mixed trace polynomials, and ones which are zero under every trace preserving homomorphism F X,tr 〈 〉 → A , for a given A , are called trace identities for A ; and elements of F X,tr 〈 〉 are called pure trace polynomials and ones that are identities for some A are called pure trace identities of A . The set of trace identities for A will be denoted I ( A,tr ) and it is an ideal of F X,tr 〈 〉 . The quotient will be denoted U A,tr ( ) and it has the expected universal properties. Likewise ¯ ( I A,tr ) = I ( A,tr ) ∩ F X,tr, 〈 〉 and ¯ ( U A,tr ) = F X,tr /I A,tr 〈 〉 ¯ ( ).
Given two sets of variables X and Y , considered degree 0 and degree 1, respectively, there are easy generalizations of these concepts to F X,Y,str 〈 〉 and F X,Y,str 〈 〉 , the free superalgebra with supertrace, including the notations of supertrace identities or pure supertrace identities. Note that F X,Y,str 〈 〉 is required to be supercentral in F X,Y,str 〈 〉 so that if u ∈ F X,Y,str 〈 〉 is degree one, then it anticommutes with all degree one elements. For example, the Grassmann algebra has a supertrace map str : E → E given by the function ω defined above. With respect to this supertrace E satisfies the mixed supertrace identities
$$\ s t r ( x ) = x \text{ and } s t r ( y ) = - y$$
and all supertrace identities of E are consequences of these two.
For our purposes we will need another generalization. We will be considering superalgebras with supertrace, which have a degree one element j whose square is equal to 1. In such a case we say that the algebra has a J -trace.
We define F X,J 〈 -trace 〉 to be the free supertrace algebra F X, 〈 { J } , str 〉 modulo the relation that J 2 = 1. Note that all variables are even and
J is odd. Elements of F X,J 〈 -trace 〉 are called J -trace polynomials, and elements of ¯ F X, 〈 { J } , J -trace 〉 are called pure J -trace polynomials. And, for algebras A with supertrace and designated odd idempotent j , one defines I ( A,J -trace) as the ideal of J -trace identities for A and U A,J ( -trace) the corresponding universal algebra. Likewise, ¯ ( I A,J -trace) will be the pure J -trace identities, and ¯ ( U A,J -trace) will be the quotient F X,J 〈 -trace 〉 /I ¯ ( A,J
-trace). For example, M (1 1) satisfies the mixed , J -trace identity
$$2 [ J x _ { 1 } + x _ { 1 } J, J x _ { 2 } + x _ { 2 } J ] = s t r ( J x _ { 1 } + x _ { 1 } J ) s t r ( J x _ { 2 } + x _ { 2 } J ) \quad \ ( 1 )$$
From one point of view J -trace identities are identities of A 0 since each variable x i is degree zero and substituted by elements of A 0 only. For this reason we may speak of the J -trace of A 0 . On the other hand, if we wanted to describe a graded J -trace identity f ( x , . . . , x 1 k , y 1 , . . . , y /lscript ) in both even and odd variables we could simply let
$$g ( x _ { 1 }, \dots, x _ { k + \ell } ) = f ( x _ { 1 }, \dots, x _ { k }, J x _ { k + 1 }, \dots, J x _ { k + \ell } ).$$
↦
↦
Then f will vanish under all graded substitutions x i → a i ∈ A 0 , y j → a j ∈ A 1 if and only if g vanishes under all substitutions x i → a i ∈ A 0 .
## 1.3 J -trace-identities of M n,n
Our main result is the determination of the J -trace-identities of M n,n ( ) in charcteristic zero. Following in the footsteps of Procesi, see [6], our approach is based on invariant theory. Procesi's work in [6] is based on the invariant theory of the general linear group which in turn is based on the double centralizer theorem between the general linear group and the symmetric group. The progression of ideas is similar here. In [9] Sergeev proved a double centralizer theorem for the queer superalgebra; and in [2] we used Sergeev's theory to study the invariant theory of the queer superalgebra. We proved that the pure J -trace polynomials in d variables give all maps M n,n ( ) d → E invariant under the action of the queer superalgebra. In the language of invariant theory this theorem could be called the First Fundamental Theorem for the queer superalgebra and in the same spirit the current work, giving the relations between the J -trace polynomials would be the Second Fundamental Theorem.
In section 2 of the current work we show how to associate multilinear pure J -trace-polynomials with elements of the Sergeev algebra and we show that this association has properties similar to Razmyslov and Procesi's association of trace polynomials with the group algebra of the symmetric group.
↦
(For the reader familiar with polynomial identities: But not analogous to Regev's association of ordinary multilinear polynomials with elements of the group algebra of the symmetric group.) Then in section 3 we describe which elements of the Sergeev algebra correspond to J -trace-polynomial identities for M n,n ( ) and prove our main theorem, that in characteristic zero all such identities are consequences of the ones of degree ( n +2 2 ) . It then follows from the non-degeneracy of the J -trace that all mixed J -trace-identities are consequences of the ones of degree ( n +2 2 ) -1 = 1 2 n n ( +3).
## 1.4 Queer Traces
Readers familiar with Kemer's structure theory of p.i. algebras (see [4]) will know that it is based on the three families of verbally prime algebras, M F n ( ), M k,/lscript , and M E n ( ). Although the polynomial identities are not well understood for any of the families, the trace identities are known for the first two: The case of trace identities of matrices was done by Razmysolv and Procesi, see [7] and [6], and the case of M k,/lscript by Razmyslov and Berele, see [8] and [1]. The third family M E n ( ) is not graced with a natural trace, but it does have a queer trace. The queer trace of a matrix with Grassmann entries is defined to be the degree one part of the sum of the diagonal entries. This map is similar to a trace function in that it satisfies qtr ab ( ) = qtr ba ( ), but unlike ordinary traces, queer trace anticommute, qtr a qtr b ( ) ( ) = -qtr b qtr a . ( ) ( ) In this section we show how the theory of J -trace-identities of M n,n ( ) can be used to describe the queer trace idenities of M E n ( ).
In general there are two ways to construct algebras with queer trace from graded algebras with supertrace. If A is such an algebra and if str A ( 0 ) = 0 then it is not hard to see that str ab ( ) = str ba ( ) and str a str b ( ) ( ) = -str b str a ( ) ( ) and so the supertrace is a queer trace. So, given any superalgebra A with supertrace, we can simply define qtr a ( ) to be the degree one part of str a ( ). This is how the queer trace on M E n ( ) was defined. The second construction can be used if in addition A had an idempotent element j ∈ A 1 . In this case we could let B be the centralizer of j . B would be a graded superalgebra with str B ( 0 ) = 0 and so the supertrace on B would be a queer trace. For our purposes we will need the case of A = M n,n ( ). The centralizer of j is the set of matrices of the form ( A B B A ) . This algebra is denoted SM n,n ( ), and the degree zero part will be denoted SM n,n . ( S for symmetric.)
There are technical issues in defining mixed queer trace identies since queer traces are not central. Pure queer trace polynomials can easily be defined as the free algebra on the symbols qtr u ( ) where u is a monomial modulo
the obvious relations. As an example, the Grassmann algebra satisfies the mixed queer trace identity
$$[ x _ { 1 }, x _ { 2 } ] = 2 q t r ( x _ { 1 } ) q t r ( x _ { 2 } )$$
and the pure queer trace identities
$$\ q t r ( x _ { 1 } [ x _ { 2 }, x _ { 3 } ] ) = q t r ( x _ { 1 } ) q t r ( x _ { 2 } ) q t r ( x _ { 3 } ),$$
$$q t r ( x _ { 1 } ) q t r ( x _ { 2 } x _ { 3 } ) + q t r ( x _ { 2 } ) q t r ( x _ { 1 } x _ { 3 } ) + q t r ( x _ { 3 } ) q t r ( x _ { 1 } x _ { 2 } ) = 0. \quad ( 4 )$$
The reader may wish to show that (3) and (4) are each consequences of (2). The intrepid reader may wish to show that all queer trace identities (appropriately defined) of E are consequences of (2), and that all pure queer trace identities are consequences of (3) and (4).
There is an isomorphism from SM n,n to M E n ( ) given by
↦
$$\begin{pmatrix} & & \ddots, \ddots & \ddots & \end{pmatrix} \\ & \begin{pmatrix} A & B \\ B & A \end{pmatrix} \mapsto A + B. \\ \dots & & \end{pmatrix}$$
It is desirable to make this map queer trace preserving and to that end we define ˜ qtr on SM n,n as or, more naturally, as ˜ qtr a ( ) = 1 2 str ja ( ), Since we can describe all J -trace-identities of M n,n ( ) the following theorem will describe all queer trace identities of M E n ( ). If A is J -trace-algebra then we make the univeral algebra U A,J ( -trace) into an algebra with queer trace by modding out by the relation that J is central. Denote by Q the map
$$\widetilde { q t r } \begin{pmatrix} A & B \\ B & A \end{pmatrix} = t r ( B ), \\ \widetilde { l } t r ( \begin{matrix} a \\ b \end{matrix} ) = \frac { 1 } { 2 } s t r ( j a ),$$
$$Q \colon \overline { F } \langle X, J \text{-trace} \rangle \rightarrow \overline { F } \langle X, q t r \rangle,$$
from supertrace polynomials to queer trace polynomials gotten by making J central and identifying str ja ( ) with 2 qtr a ( ). Note that when we make J central we make supertraces of even terms 0.
Theorem 1.1. In characteristic zero, the ideal of queer trace identities of M E n ( ) , equals the ideal of J -trace-identities of M n,n ( ) together with the identity that J is central.
Proof. Since M E n ( ) is isomorphic to SM n,n as an algebra with queer trace, we consider queer trace identities of SM n,n . Also, since SM n,n is the centralizer of j in M n,n , all supertrace of identities of M n,n become queer trace identities of SM n,n under the map Q . Since all identities are consequences of multilinear ones, in order to prove the theorem it suffices to prove that if f ( x , . . . , x 1 k ) is a multilinear queer trace identity for SM n,n there exists a J -trace polynomial g x , . . . , x ( 1 k ) such that Q g ( ) = f . First, let f 1 ( x , . . . , x 1 k ) be any supertrace polynomial such that Q f ( 1 ) = f . Then, f 1 will be a J -trace-identity for SM n,n , and so we let
$$g ( x _ { 1 }, \dots, x _ { k } ) = f _ { 1 } ( x _ { 1 } + J x _ { 1 } J, \dots, x _ { k } + J x _ { k } J ).$$
For each degree zero element a of M n,n a + jaj commutes with j and so is a degree zero element of SM n,n . Hence, g is a J -trace-identity for M n,n and Q g ( ) = f (2 x , . . . , 1 2 x k ) = 2 k f ( x , . . . , x 1 k ) for all x , . . . , x 1 k ∈ SM n,n .
## 2 Properties of Multilinear J -Trace Identities
## 2.1 The Sergeev algebra
Given a positive integer d the Clifford algebra C d ( ) is the 2 -dimensional d algebra generated by the anticommuting idempotents c 1 , . . . , c d . The symmetric group S d acts on C d ( ), and the Segeev algebra W d ( ) is the semidirect product of FS d with C d ( ). More concretely, S d ( ) is spanned by all σc /epsilon1 1 1 · · · c /epsilon1 d d with relations
$$c _ { i } ^ { 2 } = 1, \, c _ { i } c _ { j } = - c _ { j } c _ { i }, \, \sigma c _ { i } = c _ { \sigma ( i ) } \sigma.$$
In keeping with [2] we will be using the graded opposite of W d ( ) instead of W d ( ). It is generated by permutations σ and by Clifford elements C i with relations
$$\cdots \cdot \cdot C _ { i } ^ { 2 } = 1, \ C _ { i } C _ { j } = - C _ { j } C _ { i }, C _ { i } \sigma = \sigma C _ { \sigma ( i ) }, \sigma. \tau = \tau \sigma$$
for i, j = 1 , . . . , d and σ, τ ∈ S d . In the last relation the dot is for the product in W d ( ) o and the juxtaposition τσ indicates the ordinary product in S d .
There is a well-known identification of elements of FS d with multilinear, degree d pure trace polynomials. If σ has cycle decomposition
$$\sigma = ( i _ { 1 }, \dots, i _ { a } ) \cdots ( j _ { 1 }, \dots, j _ { b } )$$
then tr σ ( x , . . . , x 1 d ) is defined to be
$$t r ( x _ { i _ { 1 } } \cdots x _ { i _ { a } } ) \cdots t r ( x _ { j _ { 1 } } \cdots x _ { j _ { b } } ).$$
Note that this is well-defined because cycles commute and traces commute, and elements can be permuted cyclically in both traces and cycles. Another nice property of this identification is that
$$t r _ { \sigma \tau \sigma ^ { - 1 } } ( x _ { 1 }, \dots, x _ { d } ) = t r _ { \tau } ( x _ { \sigma ( 1 ) }, \dots, x _ { \sigma ( d ) } )$$
Hypothetically, one would like to identify elements of the Sergeev algebra with multilinear J -trace polynomials via
$$\sigma C _ { 1 } ^ { \epsilon _ { 1 } } \cdots C _ { d } ^ { \epsilon _ { d } } \to t r _ { \sigma } ( J ^ { \epsilon _ { 1 } } x _ { 1 }, \dots, J ^ { \epsilon _ { d } } x _ { d } ).$$
Unfortunately, such an identification would not be well-defined. Two pure J -traces may commute or anticommute: They will anticommute if they are each of odd degree in J . Likewise, permuting the elements within a trace may result in a negative sign. On the other hand, for given σc ∈ W d ( ) there are only two different J -trace polynomials that could result from the above putative identification and they are negatives of each other, and so we need to choose one in a natural way.
To identify elements of the Sergeev algebra with J -trace polynomials, we first define str σ for permutations. Assume that some of the variables x , . . . , x 1 d are even and the remainder odd, and let e , . . . , e 1 d be supercentral variables with deg x i = deg e i for all i , and such that str e u ( i ) = e str i ( u ) for all u . Then str σ ( x , . . . , x 1 d ) is uniquely determined by the condition
$$t r _ { \sigma } ( e _ { 1 } x _ { 1 }, \dots, e _ { d } x _ { d } ) = e _ { d } \cdots e _ { 1 } s t r _ { \sigma } ( x _ { 1 }, \dots, x _ { d } ).$$
If w = σC /epsilon1 1 1 . . . C /epsilon1 d d ∈ W d ( ) o we define
$$j t r _ { w } ( x _ { 1 }, \dots, x _ { d } ) = s t r _ { \sigma } ( J ^ { \epsilon _ { 1 } } x _ { 1 }, \dots, J ^ { \epsilon _ { d } } x _ { d } ),$$
recalling that the x i are all even and J is odd. Combining the previous two equations we get
$$e _ { d } \cdots e _ { 1 } j t r _ { w } ( x _ { 1 }, \dots, x _ { d } ) = t r _ { \sigma } ( e _ { 1 } J ^ { \epsilon _ { 1 } } x _ { 1 }, \dots, e _ { d } J ^ { \epsilon _ { d } } x _ { d } ),$$
which can be taken as the definition of jtr w when w is a monomial. Finally, using linearity we can define jtr w for any w ∈ W d ( ) o .
More generally, if w = σC i 1 · · · C i k with the i a distinct, then jtr w is gotten from tr σ ( x , . . . , x 1 d ) by substituting Jx i a for each x i a and pulling out a factor of e i k · · · e i 1 on the left, whether the sequence of i a is increasing or not.
## 2.2 Conjugation and J -trace Polynomials
The goal of this section is to generalize (7) to J -trace polynomials. There are two generalizations, one deals with conjugation by a permutation and the other deals with conjugation by a generator of the Clifford algebra.
Lemma 2.1. Let w be an element of W d ( ) o and let σ ∈ S d ⊆ W d ( ) o be a permutation. Then jtr σwσ -1 ( x , . . . , x 1 d ) = jtr w ( x σ -1 (1) , . . . , x σ -1 ( d ) ) .
As the proof will show, there is a σ -1 instead of a σ in the formula unlike (7) because we are using the opposite of S d .
Proof. Without loss, we may take w = τC /epsilon1 1 1 · · · C /epsilon1 d d = τc and so by (5)
$$\sigma w \sigma ^ { - 1 } = \sigma. \tau. \sigma ^ { - 1 } C _ { \sigma ^ { - 1 } ( 1 ) } ^ { \epsilon _ { 1 } } \cdots C _ { \sigma ^ { - 1 } ( d ) } ^ { \epsilon _ { d } } = \sigma ^ { - 1 } \tau \sigma C _ { \sigma ^ { - 1 } ( 1 ) } ^ { \epsilon _ { 1 } } \cdots C _ { \sigma ^ { - 1 } ( d ) } ^ { \epsilon _ { d } }.$$
Then
$$C _ { \sigma ^ { - 1 } ( 1 ) } ^ { \epsilon _ { 1 } } \cdots C _ { \sigma ^ { - 1 } ( d ) } ^ { \epsilon _ { d } } = g C _ { 1 } ^ { \epsilon _ { \sigma ( 1 ) } } \cdots C _ { d } ^ { \epsilon _ { \sigma ( d ) } },$$
where g = ± 1. Let e i be supercentral with degrees equal to /epsilon1 σ i ( ) . Then
$$e _ { d } \dots e _ { 1 } j t r _ { \sigma w \sigma ^ { - 1 } } ( x _ { 1 }, \dots, x _ { d } ) = g \cdot t r _ { \sigma ^ { - 1 } \tau \sigma } ( e _ { 1 } J ^ { \epsilon _ { \sigma ( 1 ) } } x _ { 1 }, \dots, e _ { d } J ^ { \epsilon _ { \sigma ( d ) } } x _ { d } ).$$
By (7) this equals
$$g \cdot t r _ { \tau } ( e _ { \sigma ^ { - 1 } ( 1 ) } & J ^ { \epsilon _ { 1 } } x _ { \sigma ^ { - 1 } ( 1 ) }, \dots, e _ { \sigma ^ { - 1 } ( d ) } J ^ { \epsilon _ { d } } x _ { \sigma ^ { - 1 } ( d ) } ) \\ & = g \cdot e _ { \sigma ^ { - 1 } ( d ) } \dots e _ { \sigma ^ { - 1 } ( 1 ) } j t r _ { w c } ( x _ { \sigma ^ { - 1 } ( 1 ) }, \dots, x _ { \sigma ^ { - 1 } ( d ) } ) \\ & = \pm e _ { d } \dots e _ { 1 } j t r _ { w c } ( x _ { \sigma ^ { - 1 } ( 1 ) }, \dots, x _ { \sigma ^ { - 1 } ( d ) } ).$$
To complete the proof we need to check that the sign is positive. Since the degree of e i equals the degree of /epsilon1 σ i ( ) , the definition of g implies that
$$\ g e _ { 1 } \dots e _ { d } = e _ { \sigma ^ { - 1 } ( 1 ) } \dots e _ { \sigma ^ { - 1 } ( d ) }.$$
This is not exactly what we want because the indices are increasing rather than decreasing. We separate the last step of the proof into a sublemma.
Sublemma. Let e , . . . , e 1 d be homogeneous elements of the Grassmann algebra E and let σ ∈ S d be a permutation such that ge 1 . . . e d = e σ -1 (1) . . . e σ -1 ( d ) . Then ge d . . . e 1 = e σ -1 ( d ) . . . e σ -1 (1) .
Proof of Sublemma: Let E be the Grassmann algebra on the vector space V with basis v , v 1 2 , . . . and assume without loss of generality that each e i is a monomial in the v j . There is a antiisomorphism θ on E given by
↦
$$v _ { i _ { 1 } } \cdots v _ { i _ { t } } \mapsto v _ { i _ { t } } \cdots v _ { i _ { 1 } }.$$
For each monomial e i we have θ e ( i ) = g e i i where g i = ± 1. Now,
$$\theta ( e _ { d } \cdots e _ { 1 } ) & = \theta ( e _ { 1 } ) \cdots \theta ( e _ { d } ) \\ & = g _ { 1 } \cdots g _ { d } e _ { 1 } \cdots e _ { d } \\ & = g g _ { 1 } \cdots g _ { d } e _ { \sigma ^ { - 1 } ( 1 ) } \cdots e _ { \sigma ^ { - 1 } ( d ) } \\ & = g \theta ( e _ { \sigma ^ { - 1 } ( 1 ) } ) \dots \theta ( e _ { \sigma ^ { - 1 } ( d ) } ) \\ & = g \theta ( e _ { \sigma ^ { - 1 } ( d ) } \dots e _ { \sigma ^ { - 1 } ( 1 ) } ). \\. \quad.. \quad. \quad e \quad.. \quad.. \quad..$$
Since θ is a one-to-one linear transformation the sublemma, and hence the lemma, follow.
Lemma 2.2. Let w = σc = σC /epsilon1 1 1 · · · C /epsilon1 d d ∈ W d ( ) o be a monomial. Then
$$j t r _ { C _ { i } w C _ { i } } ( x _ { 1 }, \dots, x _ { d } ) = \pm j t r _ { w } ( x _ { 1 }, \dots, J x _ { i } J, \dots, x _ { d } ),$$
where the sign is ( -1) ∑ /epsilon1 j .
Proof. There are three slightly different cases depending on whether σ i ( ) is less than, equal to, or greater than i . Trusting the reader to do the other two cases, we assume that σ i ( ) < i . In this case
$$C _ { i } \sigma c C _ { i } = \sigma C _ { \sigma ( i ) } c C _ { i } = g _ { 1 } \sigma c ^ { \prime },$$
where
$$c ^ { \prime } = C _ { 1 } ^ { \epsilon _ { 1 } } \cdots C _ { \sigma ( i ) } ^ { \epsilon _ { 1 + \sigma ( i ) } } \cdots C _ { i } ^ { 1 + \epsilon _ { i } } \cdots C _ { d } ^ { \epsilon _ { d } }$$
and where g 1 = ( -1) γ 1 and
$$\gamma _ { 1 } = \epsilon _ { 1 } + \cdots + \epsilon _ { \sigma ( i ) - 1 } + \epsilon _ { i + 1 } + \cdots \epsilon _ { d }.$$
To compute jtr σc ′ we choose e j with degrees /epsilon1 j , unless j = i or σ i ( ) in which case the degree is one greater. Now
$$e _ { d } \cdots e _ { 1 } j t r _ { \sigma c ^ { \prime } } ( x _ { 1 }, \dots, x _ { d } ) = t r _ { \sigma } ( e _ { 1 } J ^ { \epsilon _ { 1 } } x _ { 1 }, \dots, e _ { d } J ^ { \epsilon _ { d } } x _ { d } ).$$
The i -th term is C J i 1+ /epsilon1 i x i and the σ i ( ) term is e σ i ( ) J 1+ /epsilon1 σ i ( ) x σ i ( ) . We can assume without loss of generality that e σ i ( ) factors as f f 1 2 , where f 1 is of degree 1 and f 2 is of degree /epsilon1 σ i ( ) . An important but elementary fact about trace monomials is that if y σ i ( ) = ab then tr σ ( y , . . . , y 1 d ) = tr σ ( y , . . . , y 1 i a, . . . , b, . . . y d ) , where in the right hand side of the equation y i is replaced by y a i and y σ i ( ) is replaced by b . In order to apply this to (9) we first note that y σ i ( ) = f f J 1 2 1+ /epsilon1 σ i ( ) x σ i ( ) which we re-write as ( -1) /epsilon1 σ i ( ) f Jf J 1 2 /epsilon1 σ i ( ) x σ i ( ) . For the sake of bookkeeping, we let g 2 = ( -1) γ 2 where γ 2 = /epsilon1 σ i ( ) . Then (9) equals
$$g _ { 2 } \cdot t r _ { \sigma } ( e _ { 1 } J ^ { \epsilon _ { 1 } } x _ { 1 }, \dots, f _ { 2 } J ^ { \epsilon _ { \sigma ( i ) } } x _ { \sigma ( i ) }, \dots, e _ { i } J ^ { 1 + \epsilon _ { i } } x _ { i } f _ { 1 } J, \dots ).$$
Since f 1 is degree one and supercentral, the i th argument equals f e J 1 i /epsilon1 i Jx J i , and since f e 1 i has degree /epsilon1 i this equals
$$g _ { 2 } \cdot e _ { d } \cdots e _ { i + 1 } f _ { 1 } e _ { i } e _ { i - 1 } \cdots e _ { \sigma ( i ) + 1 } f _ { 2 } e _ { \sigma ( i ) - 1 } \cdots e _ { 1 } j t r _ { w } ( x _ { 1 }, \dots, J x _ { i } J, \dots, x _ { n } )$$
and to complete the proof we need only pull the f 1 back to be in front of the f 2 and compute the sign. First,
$$e _ { d } \cdots f _ { 1 } \cdots f _ { 2 } e _ { i } \cdots e _ { 1 } = g _ { 3 } \cdot e _ { d } \cdots e _ { 1 }$$
where g 3 = ( -1) γ 3 and γ 3 = /epsilon1 σ i ( )+1 · · · + /epsilon1 i . Altogether the sign is g g g 1 2 3 = ( -1) γ 1 + γ 2 + γ 3 and γ 1 + γ 2 + γ 3 = /epsilon1 1 + · · · + /epsilon1 d ,which proves the lemma.
## 2.3 Multilinear Identities
Given a monomial c = C /epsilon1 1 1 · · · C /epsilon1 d d in C d ( ) o we define | c | to be ∑ /epsilon1 j in Z 2 , and we sometimes refer to c as even or odd depending on whether | c | is 0 or 1. Likewise, for σc ∈ W d ( ) o we define | σc | to be | c | .
Lemma 2.3. Let A be an algebra with a J -trace. Assume that
$$\upsilon \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \begin{array} { c } \upsilon \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \, \text{with} \end{array} \\ u = \sum _ { \sigma, \epsilon } \alpha ( \sigma, \epsilon ) \sigma c ^ { \epsilon } \in W ( d ) \\ \left. \left. \left. \left. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right. \right.$$
is such that jtr u identity of A , and write u as u = u 0 + u 1 , the even terms plus the odd terms. Then jtr u 0 and jtr u 1 are each identities for A .
Proof. Given any a , . . . , a 1 d ∈ A 0 ,
$$j t r _ { u } ( a _ { 1 }, \dots, a _ { d } ) = j t r _ { u _ { 0 } } ( a _ { 1 }, \dots, a _ { d } ) + j t r _ { u _ { 1 } } ( a _ { 1 }, \dots, a _ { d } ) = 0.$$
But since the first term is in A 0 and the second in A 1 each must be zero.
Combining this lemma with lemmas 2.1 and 2.2 of the previous section we get this theorem.
Theorem 2.4. Let A be a J -trace algebra and let jtr f be an identity for A for some f ∈ W d ( ) o . Let w = σc be a monomial in W d ( ) o . Then jtr wfw -1 is also an identity for A .
The correspondence w ↔ jtr w is a vector space isomorphism between W d ( ) o and the degree d , multilinear, pure J -trace polynomials in x , . . . , x 1 d , which we denote JTR d . Using this identification we may consider JTR d as a bimodule for W d ( ) o . Also note that the incusions W d ( ) o ⊆ W e ( ) o for d ≤ e correspond to the inclusions JTR d ⊆ JTR d +1 gotten by correspnding f ( x , . . . , x 1 d ) to f ( x , . . . , x 1 d ) str x ( d +1 ) · · · str x ( e ). For a fixed J -tracealgebra A we let I d ⊆ JTR d be the set of identities for A in JTR d .
Convention. Theorem 2.4 shows that I d is invariant under conjugation by monomials in W d ( ) o but in general it need not be invariant under right or left multiplication. For the remainder of this section we will consider the special case in which I d is both a left and right W d ( ) o module and we will prove that the elements of W e I W e ( ) d ( ) are all consequences of I d and hence are in I e , for all d ≤ e .
We remark that the corresponding statement is a key lemma in the derivations of the trace identities for matrices and for M k,/lscript by Razmyslov, Procesi and Berele mentioned previously.
First of all, by induction, it suffices to prove the statement for e = d +1. Next, it suffices to prove that if f ∈ I d then fw, wf ∈ I d +1 for each monomial w ∈ W d ( + 1) , and by Theorem 2.4 we need only consider the o case of wf .
Lemma 2.5. Let σ ∈ S d +1 fix d + 1 . Then tr σ i,d ( +1) ( x , . . . , x 1 d +1 ) = tr σ ( x , . . . , x 1 d +1 x , . . . , x i d ) .
Proof. The proof follows from the computation
$$( i, a, \dots, b ) ( i, d + 1 ) = ( i, d + 1, a, \dots, b )$$
together with equation (6).
Lemma 2.6. Let f = jtr w for some w ∈ I d . Then for each u ∈ W d ( +1) o , uf ≡ jtr uw is a consequence of I n .
Proof. By the previous lemma we may assume without loss of generality that either all terms of w are even or all terms are odd.
Let u = σc , σ ∈ S d +1 , c ∈ C d ( +1). We first consider the case of σ ∈ S d so that σ d ( +1) = d +1. In this case u can be re-written as C /epsilon1 d +1 d +1 u ′ where u ′ ∈ C d ( ) and so u f ′ ∈ I d . Hence, in this case, we may replace f by u f ′ and so assume without loss that u = C /epsilon1 d +1 d +1 . Also, by Lemma 2.3 we may assume that every term in w is of degree g mod 2. In this case
$$j t r _ { C ^ { \epsilon _ { d + 1 } } _ { d + 1 } w } ( x _ { 1 }, \dots, x _ { d + 1 } ) & = ( - 1 ) ^ { g \epsilon _ { d + 1 } } j t r _ { w C ^ { \epsilon _ { d + 1 } } _ { d + 1 } } ( x _ { 1 }, \dots, x _ { d + 1 } ) \\ & = ( - 1 ) ^ { g \epsilon _ { d + 1 } } j t r _ { w } ( x _ { 1 }, \dots, x _ { d } ) s t r ( J ^ { \epsilon _ { d + 1 } } x _ { d + 1 } ),$$
which is certainly a consequence of f .
If w = σc with σ / ∈ S d we may write σ as ( i, d + 1) σ ′ , where σ ′ ( d + 1) = d + 1; and we write c as c = C /epsilon1 d +1 d +1 c ′ where c ′ ∈ C d ( ). Then C d +1 commutes with σ ′ and so w = ( i, d +1) C /epsilon1 d +1 d +1 w ′ where w ′ ∈ W d ( ) o , and since W d ( ) o I n ⊆ I n it suffices to consider the case in which w = ( i, d +1) C /epsilon1 d +1 d +1 . We now compute the J -trace polynomial corresponding to wu where w = ( i, d +1) C /epsilon1 d +1 d +1 and u is a monomial u = σc = σC /epsilon1 1 1 . . . C /epsilon1 d d . First note that wu = ( -1) γ ( i, d + 1) .uc /epsilon1 d +1 d +1 , where γ equals /epsilon1 d +1 ( /epsilon1 1 + . . . + /epsilon1 d ); and since the multiplication of the permutations in W d ( +1) o is in the opposite their product in S d +1 , wu equals ( -1) γ σ i, d ( + 1) cC /epsilon1 d +1 d +1 . Let e i have degree /epsilon1 i . Then by (8)
$$e _ { d + 1 } \dots e _ { 1 } j t r _ { w u } = ( - 1 ) ^ { \gamma } \cdot t r _ { \sigma ( i, d + 1 ) } ( e _ { 1 } J ^ { \epsilon _ { 1 } } x _ { 1 }, \dots, e _ { d + 1 } J ^ { \epsilon _ { d + 1 } } x _ { d + 1 } ). \quad ( 1 0 )$$
Applying Lemma 2.5, (10) equals
$$( - 1 ) ^ { \gamma } \cdot t r _ { \sigma } ( e _ { 1 } J ^ { \epsilon _ { 1 } } x _ { 1 }, \dots, e _ { i } J ^ { \epsilon _ { i } } x _ { i } e _ { d + 1 } J ^ { \epsilon _ { d + 1 } } x _ { d + 1 }, \dots, e _ { d } J ^ { \epsilon _ { d } } ) x _ { d }.$$
Focusing on the i th term:
$$e _ { i } J ^ { \epsilon _ { i } } x _ { i } e _ { d + 1 } J ^ { \epsilon _ { d + 1 } } x _ { d + 1 } = e _ { d + 1 } e _ { i } J ^ { \epsilon _ { i } + \epsilon _ { d + 1 } } J ^ { \epsilon _ { d + 1 } } x _ { d + 1 } J ^ { \epsilon _ { d + 1 } } x _ { d + 1 }$$
and so (10) equals
$$( - 1 ) ^ { \gamma } \cdot e _ { d } \dots e _ { i + 1 } e _ { d + 1 } e _ { i } \dots e _ { 1 } j t r _ { \sigma c ^ { \prime } } ( x _ { 1 }, \dots, J ^ { \epsilon _ { d + 1 } } x _ { i } J ^ { \epsilon _ { d + 1 } } x _ { d + 1 }, \dots, x _ { d } ),$$
where c ′ = C /epsilon1 1 1 · · · C /epsilon1 i + /epsilon1 d +1 i · · · C /epsilon1 d d . The above equals
$$( - 1 ) ^ { \gamma } \cdot e _ { d + 1 } \dots e _ { 1 } j t r _ { u C _ { i } ^ { \epsilon _ { d + 1 } } } ( x _ { 1 }, \dots, J ^ { \epsilon _ { d + 1 } } x _ { i } J ^ { \epsilon _ { d + 1 } } x _ { d + 1 }, \dots, x _ { d } ).$$
and so
$$j t r _ { w u } = ( - 1 ) ^ { \gamma } \cdot j t r _ { u C _ { i } ^ { \epsilon _ { d + 1 } } } ( x _ { 1 }, \dots, J ^ { \epsilon _ { d + 1 } } x _ { i } J ^ { \epsilon _ { d + 1 } } x _ { d + 1 }, \dots, x _ { d } ).$$
↦
Since uC /epsilon1 d +1 i ∈ W d ( ) o and the above polynomial is gotten from jtr uC /epsilon1 d +1 i by the homogeneous substitution x i → J /epsilon1 d +1 x J i /epsilon1 d +1 x d +1 it is a consequence of it.
Theorem 2.7. Assume that W d ( ) o I d W d ( ) o = I d . Then for all e ≥ d the elements of W e ( ) o I d W e ( ) o are all consequences of I d . In particular, they are all identities of A .
Proof. Follows from the previous lemma by induction on e .
## 3 J -trace-Identies of M n,n ( )
## 3.1 Theorems of Sergeev and Berele
For the remainder of the paper we take F to be characteristic zero. Let U be a Z 2 -graded vector space with graded dimension ( n, n ) and assume that J is a degree 1 idempotent isomorphism of U , i.e., a linear transformation with J 2 = id , J U ( 0 ) = U 1 and J U ( 1 ) = U 0 . In[9] Sergeev defined an action of W d ( ) o on U ⊗ d such that if u , . . . , u 1 d ∈ U are homogeneous, then
$$\sigma ( u _ { 1 } \otimes \dots \otimes u _ { d } ) = \pm u _ { \sigma ( 1 ) } \otimes \dots \otimes u _ { \sigma ( d ) }$$
and
$$C _ { i } ( u _ { 1 } \otimes \cdots \otimes u _ { d } ) = \pm u _ { 1 } \otimes \cdots \otimes J u _ { i } \otimes \cdots \otimes u _ { d },$$
where the sign depends on how many degree one elements are switched. The exact details are not necessary for our purposes. Sergeev proved two basic properties of this action, one about the image of W d ( ) o and one about the kernel of the action. The former result is that the image of W d ( ) o in End U ( ⊗ d ) equals the centralizer of the action of q n ( ), the queer superalgebra. For the kernel, it is important to know that W d ( ) o is a direct sum of two sided ideals ∑ ⊕ I λ , each I λ is a simple graded algebra, and the subscripts are naturally indexed by partitions of d into distinct parts. If we write φ : W d ( ) o → End U ( ⊗ d ) then Sergeev proved that the kernel is the sum ∑ ⊕ I λ summed over all partitions of height greater than n . Note that the smallest such partition would be δ = ( n +1 , n, . . . , 1) and in general if λ is a partition with distinct parts, λ will have height greater than n if and only if δ ⊆ λ . For convenience, we record these results of Sergeev.
Theorem 3.1. The algebra W d ( ) o is semisimple as an algebra and as a graded algebra. As a graded algebra it decomposes as a direct sum of graded simple ideals ∑ ⊕ I λ indexed by partitions of d with all distinct parts. There is a natural map φ : W d ( ) o → End V ( ⊗ d ) whose kernel is the sum of the I λ where λ has height greater than d .
Definition 3.2. We let DP d ( ) be the set if partitions of d into distinct parts.
In [2] we let V be a 2 n dimensional space over the Grassmann algebra E . The algebra End E ( V ) is isomorphic to M n,n ( ) as a superalgebra with supertrace. Using the notion of extending an F -linear map to an E linear one, we embed End U ( ⊗ d ) into End V ( ⊗ d ). Composing this with φ creates a map
$$\Phi \, \colon W ( d ) ^ { o } \to E n d ( V ^ { \otimes d } )$$
which has the same kernel as φ . Next, adapting some standard techniques from invariant theory we constructed an isomorphism
$$E n d ( V ^ { \otimes d } ) \cong ( E n d ( V ) ^ { \otimes d } ) ^ { * }.$$
We denoted by T w the functional on End V ( ) ⊗ d ) corresponding to w ∈ W d ( ) o .
In section 5 of [2] we computed T w ( A 1 ⊗··· ⊗ A d ) for A , . . . , A 1 d degree zero elements of End V ( ). Let w = σC /epsilon1 1 1 · · · C /epsilon1 d d ∈ W d ( ) o . In lemma 5.8 we showed that
$$T _ { w } ( A _ { 1 } \otimes \cdots \otimes A _ { d } ) = T _ { \sigma } ( \tilde { P } ^ { \epsilon _ { 1 } } A _ { 1 } \otimes \cdots \otimes \tilde { P } ^ { \epsilon _ { d } } A _ { d } ),$$
where ˜ is a degree one element of P End V ( ) corresponding to J . Letting e i be Grassmann elements of degree /epsilon1 i we showed in the proof of lemma 5.7 that
$$e _ { d } \cdots e _ { 1 } T _ { \sigma } ( \tilde { P } ^ { \epsilon _ { 1 } } A _ { 1 } \otimes \cdots \otimes \tilde { P } ^ { \epsilon _ { d } } A _ { d } ) = T _ { \sigma } ( e _ { 1 } \tilde { P } ^ { \epsilon _ { 1 } } A _ { 1 } \otimes \cdots \otimes e _ { d } \tilde { P } ^ { \epsilon _ { d } } ).$$
Continuing to work backwards through the lemmas of [2], lemma 5.6 can be used to evaluate the right hand side of this equation, since the factors in the tensor product are now all degree zero. It implies that
$$T _ { \sigma } ( e _ { 1 } \tilde { P } ^ { \epsilon _ { 1 } } A _ { 1 } \otimes \dots \otimes e _ { d } \tilde { P } ^ { \epsilon _ { d } } A _ { d } ) = t r _ { \sigma } ( e _ { 1 } \tilde { P } ^ { \epsilon _ { 1 } } A _ { 1 }, \dots, e _ { d } \tilde { P } ^ { \epsilon _ { d } } A _ { d } ).$$
Combining all this we get
$$e _ { d } \dots e _ { 1 } T _ { w } ( A _ { 1 } \otimes \dots \otimes A _ { d } ) = t r _ { \sigma } ( e _ { 1 } \tilde { P } ^ { \epsilon _ { 1 } } A _ { 1 }, \dots, e _ { d } \tilde { P } ^ { \epsilon _ { d } } A _ { d } ),$$
and comparing with (8) gives this lemma:
Lemma 3.3. If A , . . . , A 1 d ∈ End V ( ) 0 and w ∈ W d ( ) o then T w ( A 1 ⊗··· ⊗ A d ) = jtr w ( A , . . . , A 1 d ) .
It follows that w ∈ W d ( ) o corresponds to a J -trace-identity for End V ( ) if and only if T w is zero. But T w = 0 precisely when w is in the kernel of Φ, which is the same as the kernal of φ and which was identified by Sergeev. Finally, End V ( ) 0 is isomorphic to M n,n ( ) as an algebra with J -trace. We now have the main theorem of this section:
Theorem 3.4. Let w ∈ W d ( ) o . Then jtr w is a J -trace-identity for M n,n ( ) if and only if w lies in the two sided ideal ∑ ⊕ I λ , summed over all λ of height greater than n .
## 3.2 A Combinatorial Lemma
Theorem 3.4 tells us that the multilinear, degree d identities of M n,n ( ), I d = I d ( M n,n ( )), form a two sided ideal in each W d ( ) o ; and Theorem 2.7 says that the two sided ideal induced from I d , considered as a subspace of W d ( ) o , to each W e ( ) o , e > d consists of algebraic consequences of I d . Fortunately, the question of how ideals in W d ( ) o induce up is easily handled by a theorem of Schur and Jozefiak, and a theorem of Stembridge, as we now relate. We first focus on the case of W d ( +1) o I λ W d ( +1) , where o λ is a partition of d in DP d ( ) so I λ is an ideal of W d ( ) o .
/negationslash
/negationslash
Lemma 3.5. If I /triangleleft W d ( ) o is a two sided ideal, and if W d ( +1) o I decomposes as a direct sum of left ideals as ∑ m M µ µ , where M µ ⊆ I µ is a minimal left ideal and m µ = 0 , then W d ( + 1) o IW d ( + 1) o = ∑ I µ summed over all µ with m µ = 0 .
Proof. Since M µ is a right ideal contained in the minimal two-sided ideal I µ ,
/negationslash
$$M _ { \mu } I _ { \nu } = \begin{cases} 0, & \mu \neq \mu \\ I _ { \mu }, & \mu = \nu \end{cases}$$
$$\begin{smallmatrix} \\ \\ \\ \end{smallmatrix}$$
The computation of W d ( +1) o I λ W d ( +1) o is accomplished in the next lemma.
Lemma 3.6. Let λ ∈ DP d ( ) . Then W d ( +1) o I λ W d ( +1) o = ∑ ⊕ I µ where µ runs over all partitions in DP d ( +1) containing λ.
Proof. As a right W d ( ) o module I λ decomposes as a direct sum of isomorphic simple modules M λ . Now
$$W ( d + 1 ) ^ { o } M _ { \lambda } \cong W ( d + 1 ) \otimes _ { W ( d ) ^ { o } } M _ { \lambda } \cong W ( d + 1 ) ^ { o } / W ( d ) ^ { o } \otimes _ { F } M _ { \lambda }.$$
We compare this to
$$M _ { [ 1 ] } \hat { \otimes } M _ { \lambda } = W ( d + 1 ) ^ { o } \otimes _ { W ( 1 ) ^ { o } \times W ( d ) ^ { o } } \left ( M _ { [ 1 ] } \otimes _ { F } M _ { \lambda } \right ). \\ \text{wever, } W ( 1 ) ^ { o } \text{ is graded simple of dimension } ( 1, 1 ) \text{ and } M _ { [ 1 ] } \text{ is isomorph}$$
However, W (1) o is graded simple of dimension (1 1) and , M [1] is isomorphic to W (1) o . It follows that the right hand side is isomorphic to
$$W ( d + 1 ) ^ { o } \otimes _ { W ( d ) ^ { o } } M _ { \lambda }$$
and so W d ( +1) o M λ is isomorphic to M [1] ˆ ⊗ M λ . The outer tensor product now follows from the theorems we alluded to earlier. First, just like the outer tensor product of modules for the symmetric group is reflected in the product of Schur functions, Schur and Jozefiak proved that the outer product of Sergeev modules is reflected in the product of Q-Schur functions. See [5] for an account of these functions.
↦
This means that M µ ˆ ⊗ M ν = ∑ m λ µ,nu M ( ; ) λ if and only if Q µ ( x Q ) ν ( x ) = ∑ m λ µ,ν Q ( ; ) λ ( x . ) The latter products were computed by Stembridge in [10], see also [5] (8.18). Stembridge proved an analogue of the Littlewood-Richardson rule for the functions P λ ( x ) which have the property that each Q λ ( x ) is a (non-zero) constant multiple of P λ ( x ) and so we can use Stembridge's rule to determine when the coefficients we are calling m λ µ,ν ( ; ) are non-zero. We will need only the following special case:
Theorem (Schur, Jozefiak) . The Grothendick ring of W d ( ) o modules is isomorphic to the ring of Q-Schur functions with the isomorphism given by M λ → Q λ ( x ) .
Theorem (Stembridge) . Let λ ∈ DP n ( ) and µ ∈ DP n ( + 1) . Then m µ λ, ( ; [1]) = 0 if and only if λ ⊆ µ .
/negationslash
$$O u r lemma now follows.$$
The following corollary follows by induction.
Corollary 3.7. Let δ be the partition ( n +1 , n, . . . , 1) ∈ DP ( ( n +2 2 ) ) and let e > ( n +2 2 ) . Then W e ( ) o I W e δ ( ) o equals the sum ∑ ⊕ I λ , where λ runs over partitions of e with distinct parts, of height greater than n .
## The Main Theorem
Theorem 2.7, Theorem 3.4 and Corollary 3.7 are the ingredients we need to prove our main theorem.
Theorem 3.8. All pure J -trace identities of M n,n ( ) are consequences of the identities of degree ( n +2 2 ) .
Proof. By theorem 3.4 there are no J -trace identities of degree less than d = ( n +2 2 ) , and in degree ( n +2 2 ) the multilinear identities correspond to the two sided ideal of W d ( ) o , I δ . By theorem 2.7 for each e > d all elements of W e ( ) o I W e δ ( ) o are consequences of I δ . By corollary 3.7 the product W e ( ) o I W e δ ( ) o equals the sum of the two sided ideals I λ where λ has height greater than n . Finally, by theorem 3.4 again, this sum is precisely the space of multilinear identities of degree e , and since we are in characteristic zero all identities are consequences of multilinear ones and the theorem follows.
Using the following lemma we can also describe the mixed J -traceidentties of M n,n ( ).
Lemma 3.9. Let f ( x , . . . , x 1 n +1 ) = ∑ α str x ( n +1 u α ) g α be a multilinear pure J -trace-identity for M n,n ( ) , where each u α is a polynomial (without trace) in { J, x 1 , . . . , x n } and each g α is a pure J -trace-polynomial. Then F x , . . . , x ( 1 n ) = ∑ α u g α α is a multilinear mixed J -trace-identity for M n,n ( ) of degree one smaller and f is a consequence of F .
Proof. Since the supertrace is E -linear we have f = str x ( n +1 ∑ u g α α ), and the lemma follows from the non-degeneracy of the supertrace.
The lemma and theorem combine to give this description of the J -traceidentities of M n,n ( ).
Theorem 3.10. All J -trace identities of M n,n ( ) are consequences of the identities of degree ( n +2 2 ) -1 .
## References
- [1] A. Berele, Trace identities and Z/ Z 2 -graded invariants, Transactions of the AMS 309 (1988), 581-589.
- [2] A. Berele, Invariant theory for matrices over the Grassmann algebra, Adv. Math. 237 (2013), 3361.
- [3] A. Berele and A. Regev. Hook Young diagrams with applications to combinatorics and to representations of Lie superalgebras, Adv. in Math. 64 (1987), 118-175.
- [4] A. R. Kemer, Ideals of identities of associative algebras , Translated from the Russian by C. W. Kohls. Translations of Mathematical Monographs, 87. American Mathematical Society, Providence, RI, 1991.
- [5] I. G. MacDonald, Symmetric Functions and Hall Polynomials, 2nd edition, Clarendon Press, Oxford, 1995.
- [6] C. Procesi, The invariant theory of n × n matrices, Advances in Math. 19 (1976), no. 3, 306-381.
- [7] Ju. P. Razmyslov, Identities with trace in full matrix algebras over a field of characteristic zero. (Russian) Izv. Akad. Nauk SSSR Ser. Mat. 38 (1974), 723-756. [English translation: Math. USSR-Izv. 8 (1974), no. 4, 727-760 (1975).]
- [8] Yu. P. Razmyslov, Trace identities and central polynomials in matrix superalgebras M n,k . (Russian) Mat. Sb. (N.S.) 128(170) (1985), no. 2, 194-215, 287. [English translation: Math. USSR-Sb. 56 (1987), no. 1, 187-206.]
- [9] A. N. Sergeev, The tensor algebra of the identity representation as a module over the Lie superalgebra Gl ( n, m ) and Q ( n ), Mat. Sb. 123 (1984), 422-430. [English translation: Math. USSR Sbornik 51 (1985), 419-427.]
- [10] J. R. Stembridge, Shifted tableaux in the projective representations of symmetric groups, Adv. in Math. 74 (1989), 87-134.
| null |
[
"Allan Berele"
] |
2014-08-04T23:30:08+00:00
|
2014-08-04T23:30:08+00:00
|
[
"math.RA",
"16r30"
] |
$J$-trace identities and invariant theory
|
We generalize the notion of trace identity to $J$-trace. Our main result is
that all $J$-traces of $M_{n,n}$ are consequence of those of degree $\frac12n(n
+ 3)$. This also gives an indirect description of the queer trace identities of
$M_n(E)$.
|
1408.0839v2
|
## Evaluation of Constant Potential Method in Simulating Electric Double-Layer Capacitors
Zhenxing Wang, 1 Yang Yang, 2 David L. Olmsted, 2 Mark Asta, 2 and Brian B. Laird 1, a)
1) Department of Chemistry, University of Kansas, Lawrence, KS 66045, USA 2) Department of Materials Science and Engineering, University of California, Berkeley, CA 94720,
- USA
(Dated: 25 September 2014)
A major challenge in the molecular simulation of electric double layer capacitors (EDLCs) is the choice of an appropriate model for the electrode. Typically, in such simulations the electrode surface is modeled using a uniform fixed charge on each of the electrode atoms, which ignores the electrode response to local charge fluctuations in the electrolyte solution. In this work, we evaluate and compare this Fixed Charge Method (FCM) with the more realistic Constant Potential Method (CPM), [Reed, et al., J. Chem. Phys., 126 , 084704 (2007)], in which the electrode charges fluctuate in order to maintain constant electric potential in each electrode. For this comparison, we utilize a simplified LiClO 4 -acetonitrile/graphite EDLC. At low potential difference (∆Ψ ≤ 2V), the two methods yield essentially identical results for ion and solvent density profiles; however, significant differences appear at higher ∆Ψ. At ∆Ψ ≥ 4V, the CPM ion density profiles show significant enhancement (over FCM) of 'inner-sphere adsorbed' Li + ions very close to the electrode surface. The ability of the CPM electrode to respond to local charge fluctuations in the electrolyte is seen to significantly lower the energy (and barrier) for the approach of Li + ions to the electrode surface.
Keywords: electric double-layer capacitor, molecular simulation
## I. INTRODUCTION
Electric double-layer capacitors (EDLCs) are nonFaradaic, high power-density devices that have wide application in energy storage. Together with pseudocapacitors, EDLCs make up a class of energy storage devices called super capacitors. The energy storage and release mechanism in EDLCs is rapid and possesses a long cycle life due to the physical nature of the charging/discharging process - in the charging process, ions in the electrolyte solution aggregate at the interface to form an electric double layer, which in turn induces charges on the electrode surfaces. Over the past decade, accompanying the recent bloom of novel electrode (i.e. nanoporous material) and electrolyte materials (i.e. ionic liquid), EDLCs have been the subject of numerous experimental studies. 1,2 From these studies, the performance of EDLCs is affected by a number of different factors including, but not limited to, electrolyte composition 3,4 , interface structure 5,6 and surface area. 7,8 As a result of these efforts, the performance of EDLCs have been significantly enhanced, extending their applicability.
are more accurate than continuum models but considerably less computationally demanding than atomistic simulation. Atomistic simulations, however, have an advantage over the continuum models and cDFT because they provide a molecular-level description of the structure, dynamics and thermodynamics of EDLCs. Molecular simulation tools, such as ab-initio molecular dynamics (AIMD) simulation 13 and classical molecular-dynamics (MD) simulation are widely used to investigate EDLC interfacial phenomena at the molecular level.
These experimental studies have been complemented by a number of theoretical/computational studies ranging from atomistic simulation to mesoscale continuum modeling. 9 Analytical continuum models have been developed to describe the electrode/electrolyte interface, for example, the Gouy-Chapman-Stern model. 10,11 Classical Density Functional Theories (cDFT) 12 have also been applied to estimate properties of EDLCs, which
a) Electronic mail: [email protected]; Author to whom correspondence should be addressed
In molecular simulations of EDLCs, the modeling of the electrode is a particular challenge because of the difficulty in defining a consistent classical atomistic model for a conductor. In many MD studies of EDLCs, the electrode atoms are assumed to carry a uniform fixed charge. This Fixed Charge Method (FCM), however, neglects charge fluctuations on the electrode induced by local density fluctuations in the electrolyte solution. 14-18 To explicitly take into account such fluctuations, the Constant Potential Method (CPM) was developed by Reed, et al. 19 This method is based on earlier work of Siepmann and Sprik 20 in which the constraint of constant electrode potential was enforced on average using an extended Hamiltonian approach (similar to the Nos´ method e 21 for constant temperature); however, in the CPM the constraint is applied instantaneously at every step. Additional corrections to the CPM were added later by Gingrich and Wilson. 22 In the CPM, the electric potential Ψ i on each electrode atom is constrained at each simulation step to be equal to a preset applied external potential V , which is constant over a given electrode. This constraint leads to the following equation for the charge, q i , on each electrode atom (where i indexes the atoms in the electrode):
$$V = \Psi _ { i } = \frac { \partial U } { \partial q _ { i } } \quad \quad \ \ ( 1 ) \quad d u$$
where U is the total Coulomb energy of the system.
The structure of the Coulomb energy expression is such that Eq. 1 is a system of linear equations for q i and can be solved with standard linear algebra techniques. To guarantee that the linear system corresponding to Eq. 1 has a solution, the electrode point charges are generally replaced with a narrow Gaussian charge distribution. For a detailed study of the optimal choice for Gaussian width, see Gingrich . 1
Several studies of EDLCs employing the CPM have been reported. Merlet, et al. 24-27 studied a nanoporous carbon electrode in contact with electrolyte consisting of an ionic liquid or an ionic liquid/acetonitrile mixture. Vatamanu and coworkers 28-31 investigated ionic liquid electrolytes with carbon or gold electrodes using the smooth particle mesh Ewald (SMPE) 32,33 method to simplify the calculation. The hydration of metalelectrode surfaces was examined by Limmer, et al. 34,35
In all of these studies, detailed comparisons of the CPMwith the fixed charge method have been lacking. In a recent paper, Merlet, et al. 36 examined the differences between CPM and FCM simulations as measured by the relaxation kinetics in EDLC with nanoporous carbidederived carbon electrode and the electrolyte structure at interface in EDLC with planar graphite electrode.It was showed that CPM predicts more reasonable relaxation time than FCM. For the electrolyte structure, this study showed that there are some quantitative differences between the results of the two methods, but the qualitative features were unchanged for these ionic-liquid based EDLCs.
In this work, we study an organic electrolyte/saltbased ELDC, namely an LiClO /acetonitrile electrolyte 4 at a graphite electrode. To compare the results for this system using the CPM and FCM, several structural aspects normally reported in EDLC simulations are studied for comparison, specifically, the particle and charge density profiles near the electrodes and the solvation structure of the cation (Li + ) both in bulk and near the surface.
## II. MODELS AND METHODOLOGY
In our simulations, the atoms of the electrolyte solution are placed between two carbon electrodes, each consisting of three graphite layers. The simulation geometry is shown in Fig. 1, which shows a snapshot from a simulation at 298K with a potential difference, ∆Ψ of 2 V. For the production runs, the distance between the two innermost electrode layers (labeled L1 and R1, respectively, in Fig. 1) is 6 365 nm, which is far beyond the Debye length . (0 2 nm). . The electrolyte between the electrodes consists of 588 acetonitrile molecules and 32 Li + /ClO -4 pairs, corresponding to a LiClO 4 concentration of 1 00 M. The to-.
tal dimensions of the cell are 2.951, 2.982 and 8 040 nm. . The positions of the electrode carbon atoms are held fixed during the simulation.
To model the molecular interactions we employ a variety of literature force fields. For acetonitrile, we use the united atom model of Edwards, et al. 37 For the ions, we use the forcefield of Eilmes and Kubisiak, 38 excluding the polarizability terms. The interaction parameters for the graphite electrode carbon atoms are taken from Ref. 39. Lorentz-Berthelot mixing rules are used to construct all cross interactions.
FIG. 1. Simulation snapshot at 298K with ∆Ψ = 2V: negative electrode is on left and positive is on right. The color of each electrode atom indicates its charge (refer to color scale bar, unit e ). The electrolyte solution is shown between the two electrodes. Orange spheres: Li + , Red spheres: O in ClO -4 , cyan spheres: Cl in ClO , transparent stick models: -4 acetonitrile. For clarity, in this figure, the distance between L1 and R1 (5 43 nm) is smaller than that used in the production runs. .
In this slab geometry, we define the z -axis as the direction normal to the electrodes and apply periodic boundary conditions only in the x-y plane (parallel to the graphite layers). Unlike the original CPM, which used 2d-periodic Ewald sums, 32,40 3d-periodic Ewald sums with shape corrections 41 were used in this work to improve the calculation speed, with a volume factor set to 3. The correction term to the usual 3d-periodic energy expression is given by (see Eq. A.21 in Ref. 1):
$$\frac { 2 \pi } { V } \left ( \sum _ { i } q _ { i } z _ { i } \right ) ^ { 2 } \quad \quad \ \ ( 2 ) \\. \quad \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{-} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \, \text{--} \,$$
In studies using FCM, either the uniform charge on each electrode atom is either arbitrarily chosen 15,16,18 or is estimated using a time-consuming trial and error procedure to yield the specified electric potential difference, 17 which is calculated by numerically integrating the Poisson equation using the charge density profile. The CPM is able to predict the explicit average charge on each electrode atom at a given potential difference; therefore, for consistency, in our FCM simulations we set the charge on each electrode atom to the average charge per atom obtained by the CPM calculations at the same potential difference. Otherwise, all other force-field pa-
TABLE I. Average total charge on each electrode layer a
| Layer | 0 V | 2 V | 4 V | 5 V |
|---------|------------|-------------|-------------|-------------|
| L3 | 0 . 00(2) | 0 . 03(2) | 0 . 06(2) | 0 . 06(2) |
| L2 | 0 . 02(5) | 0 . 33(5) | 0 . 64(4) | 0 . 75(5) |
| L1 | - 0 . 1(4) | - 3 . 4(4) | - 6 . 6(4) | - 8 . 4(5) |
| R1 | 0 . 0(4) | 3 . 3(4) | 6 . 5(4) | 8 . 4(5) |
| R2 | 0 . 01(5) | - 0 . 29(4) | - 0 . 58(5) | - 0 . 75(6) |
| R3 | 0 . 00(2) | - 0 . 02(2) | - 0 . 03(2) | - 0 . 06(2) |
| Total | 0 . 01(6) | 0 . 02(6) | 0 . 04(6) | 0 . 01(6) |
a The numbers in parentheses represent 95% confidence intervals in the last significant figure
rameters in the FCM simulations are identical to those used in the CPM simulations.
All simulations were performed using the moleculardynamics simulation code LAMMPS, 42 modified to implement the CPM, using a time step of 1 fs. Constant NVT conditions are enforced using a Nos´-Hoover there mostat with a relaxation time of 100 fs and a temperature of 298 K. The cutoffs for all non-bonded interactions are 1 4 nm. . For the Ewald sums, an accuracy (relative RMS error in per-atom forces) is set to 10 -8 . The parameter of the Gaussian electrode charge distribution (see Eq. S2 in supplementary material) is set to 19 79 nm . -1 , which is the same as in Ref. 19. All results reported here are statistical averages taken from runs of 25 to 30ns in length, each preceded by 2 ns of equilibration. Further details as to the CPM method and our implementation in LAMMPS can be found in the supplementary material.
## III. RESULTS AND DISCUSSION
## A. Electrode charge distribution
Unlike the FCM, the CPM allows the individual atom charges on the electrode to fluctuate in response to local charge rearrangements in the electrolyte. The total charges on each layer for several potential differences are shown in Table I. The net electrode charge for each simulation does not show values statistically significant from zero, as expected. In a perfect conductor, the charges on the electrode atoms would be concentrated entirely on the electrode surface layers (L1 and R1 in Fig. 1); however, in the CPM the charge distribution on the electrode is approximated by discrete point charges centered on the electrode itself, so a small amount of charge is found on the second layer and to a much lesser extent the third.
Figure 2 is a log-linear plot of the probability distribution of individual electrode atom charges, p q ( ), on the inner electrode layers for various potential differences. For the inner layer of the positively charged electrode (R1), the charge distribution is well described by a Gaussian distribution over the entire range of potential differences studied (∆Ψ = 0V to 5V). For the negatively charged electrode (L1), this is true at the lower poten- tial differences (up to ∆Ψ = 2 V), but the charge distribution takes on a bimodal structure at higher potential differences (∆Ψ = 4 V and 5 V). A similar non-Gaussian charge distribution has also been seen in simulations at the negative electrode of a H 2 O/Pt system; 34 however, that system differs somewhat from our simulated system in that no ions are present. The bimodal charge distribution at the higher potential differences in this system highlights an important difference between the CPM and the FCM. Unlike the FCM, the electrode charges in the CPM can adjust to respond to local fluctuations in the electrolyte/ion charge density. As we will see in the next section, this second peak in p q ( ) seen at high potential differences is due to the presence of Li + ions near the electrode surface, which induce higher than average local charges on the adjacent electrode.
FIG. 2. Distribution of electrode atom charges for the inner electrode layers at various potential differences: L1 (solid lines) and R1(dashed lines).
## B. Density profiles for ions and electrolytes
To better understand the origin of the bimodal charge distributions in the CPM at high potential difference, we examine the ion density profiles at the interface and compare the results obtained using CPM and FCM (Fig. 3).
At the lowest non-zero potential difference (∆Ψ = 2 V), the ion density profiles for the CPM and FCM methods show relatively minor quantitative differences in peak heights, but otherwise the density peak positions and overall structure are identical. At higher potential dif-
-
FIG. 3. Ion densities near each of the two electrodes (L1 and R1) for various applied potential differences (a) L1; ∆Ψ = 2 V, (b) R1; ∆Ψ = 2 V, (c) L1; ∆Ψ = 4 V (d) R1 ∆Ψ = 4 V (e) L1; ∆Ψ = 5V, (f) R1; ∆Ψ = 5V. The negatively and positively charged electrodes (L1 and R1) are at z = 0 and 6 365 nm, respectively. .
ferences (4 V and 5 V), however, qualitative differences emerge. At ∆Ψ = 4 V, a peak in the Li + density profile near the negatively charged electrode (L1) at 0 22 nm ap-. pears in the CPM calculation, but is absent in the FCM results. This peak represents inner-sphere adsorbed Li + ions that no longer possess a full solvation shell of acetonitrile, but are also 'partially solvated' by the electrode. The existence of these inner-sphere adsorbed ions is made possible because the CPM allows for more physically correct electrode charge distribution - one that can respond to the presence of the nearby Li + ion with locally larger-than-average negative electrode charges. The partial desolvation was reported to be highly related with the confinement of ions, which fundamentally affects the capacitance. 43
As the potential difference is increased beyond 4 V, the inner-sphere adsorbed ion peak grows substantially. For the FCM, we see this peak at 5 V, but it has an amplitude that is much smaller than that predicted by the CPM. In addition, Fig. 3 also shows that the height of the Li + density peak at 0 69 nm, representing the first fully solvated . outer-sphere adsorbed Li + layer in the ELDC, is overestimated at ∆Ψ = 4V and 5V in the FCM. This is due to the migration of Li + ion density from the inner-spher adsorbed peak at 0 21 nm, compared with the CPM. .
In contrast to Li + , the ClO -4 density profiles are nearly identical for both methods at all studied potential differ- ences. The charge density in ClO -4 is spread out over a larger radius and does not create as large a local charge concentration in the electrolyte solution to perturb the electrode charge distribution enough to affect the density profiles.
FIG. 4. Acetonitirle density and charge profiles at various applied potentials (a) ∆Ψ = 2 V, (b) ∆Ψ = 4 V, (c) ∆Ψ = 5 V. Vertical lines show the position of the electrode inner layers L1 (left) and R1 (right).
In Fig. 4 we plot the solvent (acetonitrile, center of mass) and charge density profiles. Unlike the ion density profiles, the density profiles for acetonitrile are identical for both methods up to a potential difference of 4V. However, at ∆Ψ = 5V the acetonitrile peak closest to the negative electrode (L1) splits into two peaks in the CPM, a feature not seen in the FCM calculation. This is because the large amount of electrode solvated Li + seen in the CPM pulls the acetonitrile molecules in its solvation shell closer to the surface.
The structure of the electric double layer in this system is best illustrated by the charge density profile (see Figure 4). For this quantity the results for both methods (CPM and FCM) are very similar with the only difference coming at high potential difference, where small charge peaks corresponding to the electrode-solvated Li + are present in the CPM result, but not in the FCM.
## C. Structure of inner-sphere adsorbed Li +
In this work, the principal difference between the results derived from the CPM and FCM is the appearance in the CPM of Li + ions very close to the nega-
+
NS
FIG. 5. Radial distribution functions (RDF,solid lines) and corresponding number of solvent molecules (NS, dashed lines) between Li + and N atom in acetonitrile for ∆Ψ = 4 V. In the legend the number indicates the distance (nm) from the negative electrode surface to the center of bin (with width ± 0 02 nm) containing the Li . + ion. The insets show typical snapshots of the Li + solvation structure for each distance studied.
tively charged electrode at high potentials. This innersphere adsorbed Li + also is seen in the FCM, but only at very low concentration at the highest potential difference studied. To examine this feature more closely, we analyze the solvation structure of Li + by plotting in Fig. 5 (for ∆Ψ = 4 V) the CPM radial distribution function (RDF) between the Li + ion and the acetonitrile N for three Li + distances from the electrode surface: 0.21, 0.69 and 4 50nm. . These correspond to the inner-sphere adsorbed Li + , the first layer of outer-sphere adsorbed Li + and the bulk conditions, respectively. Also plotted in Fig. 5 are the number of solvent molecules (NS) within a given radius for the bulk ions, the first plateau in this quantity corresponds to the equilibrium coordination number. The results of RDF and NS for the FCM or other potential differences are identical to that shown for the CPM at 4 V (Fig. 5) - except for the fact that no Li + ions at 0 21 nm for the CPM at ∆Ψ = 2 V and for . the FCM at 2V and 4V are detected and therefore no RDF (nor NS) data was obtained.
At z = 0 69nm and 4 50nm, the Li . . + ion is fully solvated by the acetonitrile solvent and the first solvation shell peaks for both distances are nearly identical with a peak distance of 0 22 nm. . In addition, the coordination number (given by the plateau value of NS) of each Li + ion is equal to about 5.0 for the 'bulk' case (4 50 nm), .
just slightly smaller than that for the first fully solvated peak with 5.4, presumably due to the enhanced acetonitrile density near the electrode region (see Fig. 4). The value for the 'bulk' is consistent with that seen in other simulation studies of bulk acetonitrile-lithium salt solutions. 44,45
For Li + ions at 0 21 nm from the electrode, however, . the solvent coordination is significantly reduced from the bulk value to 3.1 due to partial solvation by the electrode atoms themselves. The bottom inset in Fig. 5 shows a representative snapshot from the CPM simulation (∆Ψ = 4V) of an electrode-solvated ion and its nearby environment. In the CPM, because the charges on the electrode can fluctuate individually in response to the local electrolyte charge distribution, the presence of the Li + ion very close to the electrode induces larger than average negative charges on the nearby electrode ions, as seen in the inset. The ability of the electrode to respond to local fluctuations is necessary for the stabilization of the electrode-solvated Li + below ∆Ψ < ∼ 4V, as no such ions are seen in the FCM simulations in this range. At 5V, such electrode-solvated ions are seen in the FCM, but at significantly lower concentration than in the CPM, indicating a much higher energy for such configurations in the FCM, relative to the CPM. We have confirmed that sampling in the simulations is sufficient, as we see multiple crossings and recrossings of Li + ions between the positions at 0 69 nm and 0 21 nm. . .
## IV. SUMMARY
In this work, two methods, the Constant Potential Method (CPM) and the Fixed Charge Method (FCM), are compared in a simulation of a model for a LiClO 4 acetonitrile/graphite electric double-layer capacitor. The major difference between the two methods is that in the CPM the charges on the individual electrode atoms can fluctuate in response to local fluctuations in electrolyte charge density, whereas those charges for the FCM are static. For this system, there are no measurable differences between the results of these two methods at low potential differences (∆Ψ ≤ 2 V); however, at larger potential differences significant qualitative differences emerge in the EDLC ion spatial distribution.
At a potential difference of 4V, a new peak in the Li + density profile appears in the CPM calculation at 0.21 nm away from the electrode. This peak - absent in the FCM calculation - corresponds to a inner-sphere adsorbed Li + ion that is close enough to the electrode to have lost some of its acetonitrile solvation shell and is partially solvated by the electrode atoms. For this electrode-solvated Li + ion the acetonitrile coordination number is reduced from about 5.0 for a fully solvated ion to 3.1. This partial solvation is possible because in the CPM the electrode atom charges can respond to local fluctuations in the electrolyte charge density - in this case negative charges build up in the CPM electrode near the lithium ion. This
ability lowers the energy of the 'electrode-solvated' Li + ion relative to that of a fixed-charge electrode (as in the FCM). This close approach of Li + to the electrode is possible in the FCM, but only at higher electrode potential differences and a considerably lower concentrations than when CPM is used. The energetics of the approach of a Li + ion to the electrode is important in many pseudocapacitor applications, and, as our calculations show, the CPM would be preferable to the FCM in the calculation of the barrier energy of this process because it more accurately represents the fluctuating charges on the electrode.
## ACKNOWLEDGMENTS
The authors thank Dr. David Limmer for helpful discussions on the constant potential method. This work is supported as part of the Molecularly Engineered Energy Materials, an Energy Frontier Research Center funded by the US Department of Energy, Office of Science, Office of Basic Energy Sciences under Award No. DE-SC0001342.
- 1 N. S. Choi, Z. Chen, S. A. Freunberger, X. Ji, Y.-K. Sun, K. Amine, G. Yushin, L. F. Nazar, J. Cho, and P. G. Bruce, Angew. Chemie 51 , 9994 (2012).
- 2 V. Aravindan, J. Gnanaraj, Y. S. Lee, and S. Madhavi, Chem. Rev. (2014).
- 3 T. Sato, G. Masuda, and K. Takagi, Electrochim. Acta 49 , 3603 (2004).
- 4 K. Yuyama, G. Masuda, H. Yoshida, and T. Sato, J. Power Sources 162 , 1401 (2006).
- 5 D. Wei, M. R. Astley, N. Harris, R. White, T. Ryh¨ anen, and J. Kivioja, Nanoscale (2014).
- 6 H. Ji, X. Zhao, Z. Qiao, J. Jung, Y. Zhu, Y. Lu, L. L. Zhang, A. H. MacDonald, and R. S. Ruoff, Nat. Commun. 5 , 3317 (2014).
- 7 C. Largeot, C. Portet, J. Chmiola, P. L. Taberna, Y. Gogotsi, and P. Simon, J. Am. Chem. Soc. 130 , 2730 (2008).
- 8 H. Itoi, H. Nishihara, T. Kogure, and T. Kyotani, J. Am. Chem. Soc. 133 , 1165 (2011).
- 9 M. V. Fedorov and A. a. Kornyshev, Chem. Rev. 114 , 2978 (2014). 10 H. Wang and L. Pilon, Electrochim. Acta 64 , 130 (2012).
- 11 H. Wang, A. Thiele, and L. Pilon, J. Phys. Chem C 117 , 18286 (2013).
- 12 D. E. Jiang and J. Wu, J. Phys. Chem. Lett. 4 , 1260 (2013).
- 13 Y. Ando, Y. Gohda, and S. Tsuneyuki, Chem. Phys. Lett. 556 , 9 (2013).
- 14 L. Yang, B. H. Fishbine, A. Migliori, and L. R. Pratt, J. Am. Chem. Soc. 131 , 12373 (2009).
- 15 L. Yang, B. H. Fishbine, A. Migliori, and L. R. Pratt, J. Chem. Phys. 132 , 044701 (2010).
- 16 Y. Shim, H. J. Kim, and Y. Jung, Faraday Discuss. 154 , 249 (2012).
- 17 G. Feng and P. T. Cummings, J. Phys. Chem. Lett. 2 , 2859 (2011).
- 18 G. Feng, S. Li, J. S. Atchison, V. Presser, and P. T. Cummings, J. Phys. Chem. C 117 , 9178 (2013).
- 19 S. K. Reed, O. J. Lanning, and P. A. Madden, J. Chem. Phys. 126 , 084704 (2007).
- 20 J. I. Siepmann and M. Sprik, J. Chem. Phys. 102 , 511 (1995). 21 S. Nose, J. Chem. Phys. 81 , 511 (1984).
- 22 T. R. Gingrich and M. Wilson, Chem. Phys. Lett. 500 , 178 (2010).
- 1 T. R. Gingrich, Simulating Surface Charge Effects in Carbon Nanotube Templated Ionic Crystal Growth , Master thesis, University of Oxford (2010).
- 24 C. Merlet, M. Salanne, B. Rotenberg, and P. A. Madden, J. Phys. Chem. C 115 , 16613 (2011).
- 25 C. Merlet, B. Rotenberg, P. A. Madden, P. L. Taberna, P. Simon, Y. Gogotsi, and M. Salanne, Nat. Mater. 11 , 306 (2012).
- 26 C. Merlet, M. Salanne, and B. Rotenberg, J. Phys. Chem. C 116 , 7687 (2012).
- 27 C. Merlet, M. Salanne, B. Rotenberg, and P. A. Madden, Electrochim. Acta 101 , 262 (2013).
- 28 J. Vatamanu, O. Borodin, and G. D. Smith, J. Am. Chem. Soc. 132 , 14825 (2010).
- 29 J. Vatamanu, O. Borodin, D. Bedrov, and G. D. Smith, J. Phys. Chem. C 116 , 7940 (2012).
- 30 J. Vatamanu, O. Borodin, and G. D. Smith, J. Phys. Chem C 116 , 1114 (2012).
- 31 J. Vatamanu, Z. Hu, D. Bedrov, C. Perez, and Y. Gogotsi, J. Phys. Chem. Lett. 4 , 2829 (2013).
- 32 M. Kawata and M. Mikami, Chem. Phys. Lett. 340 , 157 (2001).
- 33 M. Kawata, M. Mikami, and U. Nagashima, J. Chem. Phys. 116 , 3430 (2002).
- 34 D. T. Limmer, C. Merlet, M. Salanne, D. Chandler, P. A. Madden, R. van Roij, and B. Rotenberg, Phys. Rev. Lett. 111 , 106102 (2013).
- 35 D. T. Limmer, A. P. Willard, P. Madden, and D. Chandler, Proc. Natl. Acad. Sci. 110 , 4200 (2013).
- 36 C. Merlet, C. Pe, B. Rotenberg, P. A. Madden, P. Simon, and M. Salanne, J. Phys. Chem. Lett. 4 , 264 (2013).
- 37 D. M. F. Edwards, P. A. Madden, and I. R. Mcdonald, Mol. Phys. 51 , 1141 (1984).
- 38 A. Eilmes and P. Kubisiak, J. Phys. Chem. B 115 , 14938 (2011). 39 M. W. Cole and J. R. Klein, Surf. Sci. 124 , 547 (1983).
- 40 S. W. De Leeuw and J. W. Perram, Mol. Phys. 37 , 1313 (1979). 41 I. C. Yeh and M. L. Berkowitz, J. Chem. Phys. 111 , 3155 (1999). 42 S. Plimpton, J. Comput. Phys. 117 , 1 (1995).
- 43 C. Merlet, C. P´ ean, B. Rotenberg, P. A. Madden, B. Daffos, P. L. Taberna, P. Simon, and M. Salanne, Nat. Commun. 4 , 2701 (2013).
- 44 S. Oliveira Costa and J. L´pez Cascales, J. Electroanal. Chem. o 644 , 13 (2010).
- 45 D. Sp˚ngberg and K. Hermansson, Chem. Phys. a 300 , 165 (2004).
## Supplemental Materials: Evaluation of Constant Potential Method in Simulating Electric Double-Layer Capacitors
Zhenxing Wang 1 Yang Yang , David L. Olmsted , Mark Asta 2 2 2 and Brian B. Laird ∗ 1
1 Department of Chemistry, University of Kansas, Lawrence, KS 66045, USA
2 Department of Materials Science and Engineering, University of California, Berkeley, CA 94720, USA
∗ Author to whom correspondence should be addressed
## Table of Contents
S1 Details of Constant Potential Method
- S2 Implementation of CPM in LAMMPS
## S1 Constant Potential Method
In what follows, the charge and position vector of an electrode (electrolyte) atom are designated as Q i ( q j ) and R i ( r j ), respectively. Starting from Eq. 1 in the main text, the potential at the position of a selected electrode atom ( ) l with the charge as Q l and the positon as R l can be calculated by:
$$\Psi _ { l } = \Psi _ { l } ^ { k s p a c } + \Psi _ { l } ^ { \text{real} } + \Psi _ { l } ^ { \text{self} } + \Psi _ { l } ^ { \text{slab} }$$
Here Ψ kspace l the k -space contribution from the 3-d Ewald sum; Ψ real l is the corresponding real-space contribution; Ψ self l is the self-correction and Ψ slab l is the correction for the slab geometry. To guarantee the solvability of the CPM linear equations, Gaussian charges are used for electrode atoms 1
$$Q _ { i } ( { \mathbf r } ) = Q _ { i } \left ( \frac { \eta ^ { 2 } } { \pi } \right ) ^ { 3 / 2 } e ^ { - \eta ^ { 2 } ( { \mathbf r } - { \mathbf R } _ { i } ) ^ { 2 } } \\ \dots \dots \dots \dots \dots$$
Each portion of the potential for the mixed electrode (Gaussian charge) and electrolyte (point charge) system are
Here η is the parameter defining the width of the distribution. obtained as follows.
Real-space contributions : The Ewald expression for the real-space contribution is
$$\Psi _ { l } ^ { \text{real} } = \frac { 1 } { 2 } \sum _ { i = 1 } ^ { n ^ { \prime } } Q _ { i } \frac { \text{erfc} \left ( \alpha r _ { l i } \right ) - \text{erfc} \left ( \frac { \eta } { \sqrt { 2 } } r _ { l i } \right ) } { r _ { l i } } + \sum _ { j = 1 } ^ { m } q _ { j } \frac { \text{erfc} \left ( \alpha r _ { l j } \right ) - \text{erfc} \left ( \eta r _ { l j } \right ) } { r _ { l j } } \\ \intertext { w a n ^ { \prime } \text{ and out to the $i=1$ to w a n d d } \ w. = \left | \mathfrak { D }. \quad \mathfrak { D } \right | \text{ and out } \ w. = \left | \mathfrak { D }. \quad \mathfrak { D } \right |.$$
Where n ′ denotes the i = l term is excluded. r li = | R l -R i | and r lj = | R l -r j | .
k -space contributions : The Ewald expression for the k -space contribution is
$$\Psi _ { l } ^ { k s p a c e } = \frac { 1 } { V } \sum _ { k > 0 } \Gamma ( \kappa ) \left [ e ^ { - i k \cdot R _ { l } } S ( k ) + e ^ { i k \cdot R _ { l } } S ( - k ) \right ] \\ \text{notation in Role 1 and 2 have to nonrencional latitude counter } \varepsilon = | k | \text{and } \Gamma ( \varepsilon ) \text{ to the Enum}$$
Following the notation in Refs. 1 and 2, here k is a reciprocal lattice vector, κ = | k | and Γ( κ ) is the Fourier coefficient of the Gaussian function used in the Ewald sum
$$\Gamma ( \kappa ) = \frac { 4 \pi e ^ { - \kappa ^ { 2 } / 4 \alpha ^ { 2 } } } { \kappa ^ { 2 } }$$
In Eq. S4, S ( k ) is the structure factor, given as
$$S ( { \mathbf k } ) = \sum _ { i = 1 } ^ { n } Q _ { i } e ^ { i { \mathbf k } \cdot { \mathbf R } _ { i } } + \sum _ { j = 1 } ^ { m } q _ { j } e ^ { i { \mathbf k } \cdot { \mathbf r } _ { j } }$$
Self correction :
$$\Psi _ { l } ^ { \text{self} } = \frac { \sqrt { 2 } \eta - 2 \alpha } { \sqrt { \pi } } Q _ { l }$$
Slab correction :
$$\Psi _ { l } ^ { \text{slab} } = \frac { 4 \pi Z _ { l } } { V } \left ( \sum _ { i = 1 } ^ { n } Q _ { i } Z _ { i } + \sum _ { j = 1 } ^ { m } q _ { j } z _ { j } \right ) \\ \intertext { t e t o d e o t o m } \text{option and } z _ { j } \text{ is the corresponding quantity for an electrolyte atom.}$$
where Z i is the z component electrode atom position and z j is the corresponding quantity for an electrolyte atom.
The potential expression can then be rewritten as
$$\Psi _ { l } & = \sum _ { i } ^ { n } a _ { l i } Q _ { i } + \sum _ { j } ^ { m } b _ { l j } q _ { j } \\ \dots \quad \ell \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon$$
where a li is a function of the position of selected electrode atom ( R l ) and that of other electrode atoms ( R i ) only, which are constant because all electrode atoms are fixed. In opposite, r j is varying during the simulation and thus b lj is updated every timestep. Because Ψ l is equal to the external potential V , above equation becomes
$$\sum _ { i } ^ { n } a _ { l i } Q _ { i } & = V - \sum _ { j } ^ { m } b _ { l j } q _ { j } \\. &. &. &. &. &. &. &. &. &.$$
Similarly we can obtain a linear equation for each electrode atom. With Q = [ Q , Q , 1 2 · · · , Q n ] as unknown, the linear system can be then written as
$$A Q = V - B$$
By solving this well-determined linear equations system the explicit charge for electrode atoms can be determined.
## S2 Implementation in LAMMPS
The implemented code includes three functions: setting up the linear system of equations, solving the linear system and updating atom charges, energy and per-atom forces. The code can be downloaded through https://code.google.com/p/lammps-conp/
Determining the real-space contribution to A and B is similar to the real-space Coulombic energy calculation in LAMMPS, but without using the table technique (see the table keyword in pair modify command in the LAMMPS manual for details). The self correction and slab correction portion are straightforward and the approach is similar to that of the corresponding components of the energy calculation in LAMMPS. However the Ewald sum contribution is obtained in a different way. Elements in A depend on the positions of electrode atoms only, which is constant during the simulation (assuming fixed electrode atoms). Therefore, at the initial stage, all MPI tasks collect the coordinates of all electrode atoms and calculate Ewald sum contribution to A individually. In contrast, B is only the summation of the contribution from only electrolyte atoms, so that Ewald sum portion in B is calculated in a similar way as the Ewald sum in group/group interaction calculation in LAMMPS, which each MPI task only calculates the portion in its atom list and then sum them together.
At the initial stage, the matrix inverse of A is obtained using LAPACK. In each step, at the second half step of the velocity-Verlet, the B is calculated and the charges of electrode atoms obtained, after which the additional energy and per-atom force are calculated (See reference 1 for details of the additional terms) and added to original energy and per-atom force.
1 T. Gingrich, Simulating Surface Charge Effects in Carbon Nanotube Templated Ionic Crystal Growth , Ph.D. thesis, University of Oxford (2010).
2 S. G. Moore, 'Calculating Per-Atom and Group/Group Quantities Using the Lattice Sum Method,' http://lammps.sandia.gov/doc/ PDF/kspace.pdf .
|
10.1063/1.4899176
|
[
"Zhenxing Wang",
"Yang Yang",
"David L. Olmsted",
"Mark Asta",
"Brian B. Laird"
] |
2014-08-04T23:38:59+00:00
|
2014-09-24T14:50:19+00:00
|
[
"physics.comp-ph",
"cond-mat.mtrl-sci",
"physics.chem-ph"
] |
Evaluation of Constant Potential Method in Simulating Electric Double-Layer Capacitors
|
A major challenge in the molecular simulation of electric double layer
capacitors (EDLCs) is the choice of an appropriate model for the electrode.
Typically, in such simulations the electrode surface is modeled using a uniform
fixed charge on each of the electrode atoms, which ignores the electrode
response to local charge fluctuations induced by charge fluctuations in the
electrolyte. In this work, we evaluate and compare this Fixed Charge Method
(FCM) with the more realistic Constant Potential Method (CPM), [Reed, et al.,
J. Chem. Phys., 126, 084704 (2007)], in which the electrode charges fluctuate
in order to maintain constant electric potential in each electrode. For this
comparison, we utilize a simplified LiClO$_4$-acetonitrile/graphite EDLC. At
low potential difference ($\Delta\Psi\le 2V$), the two methods yield
essentially identical results for ion and solvent density profiles; however,
significant differences appear at higher $\Delta\Psi$. At $\Delta\Psi\ge 4V$,
the CPM ion density profiles show significant enhancement (over FCM) of
"partially electrode solvated" Li$^+$ ions very close to the electrode surface.
The ability of the CPM electrode to respond to local charge fluctuations in the
electrolyte is seen to significantly lower the energy (and barrier) for the
approach of Li$^+$ ions to the electrode surface.
|
1408.0840v1
|
## Detecting compact binary coalescences with seedless clustering
M Coughlin, 1, a E Thrane, 2 and N Christensen 3
1 Department of Physics, Harvard University, Cambridge, MA 02138, USA
2 LIGO Laboratory, California Institute of Technology, MS 100-36, Pasadena, CA, 91125, USA 3 Physics and Astronomy, Carleton College, Northfield, Minnesota 55057
(Dated: August 6, 2014)
Compact binary coalescences are a promising source of gravitational waves for second-generation interferometric gravitational-wave detectors. Although matched filtering is the optimal search method for well-modeled systems, alternative detection strategies can be used to guard against theoretical errors (e.g., involving new physics and/or assumptions about spin/eccentricity) while providing a measure of redundancy. In previous work, we showed how 'seedless clustering' can be used to detect long-lived gravitational-wave transients in both targeted and all-sky searches. In this paper, we apply seedless clustering to the problem of low-mass ( M total ≤ 10 M /circledot ) compact binary coalescences for both spinning and eccentric systems. We show that seedless clustering provides a robust and computationally efficient method for detecting low-mass compact binaries.
PACS numbers: 95.75.-z,04.30.-w
## I. INTRODUCTION
Compact binary coalescences (CBCs) of black holes (BHs) and/or neutron stars (NSs) are a likely source of gravitational waves (GWs) [1-3]. CBC events include binary neutron stars (BNSs), neutron-star black holes (NSBHs), and binary black holes (BBHs). As a CBC passes through its inspiral and merger stage, it generates GWs which sweep upward in frequency and strain amplitude through the sensitive band of GW detectors. The detection of GWs from CBCs will provide information about the populations of compact objects in the universe, elucidate the properties of strong field gravity, and provide a means to test general relativity.
Here, we focus on relatively low-mass binaries ( M total ≤ 10 M /circledot ). There are two reasons for restricting our attention to this region of parameter space. First, the rate of low-mass CBCs is less subject to theoretical uncertainty than high stellar-mass binary BHs and intermediate-mass BBH. Second, we are interested in long-lived signals ( ≈ 54-270 s), which appear as curved tracks in spectrograms of GW strain power, and therefore provide an appealing target for seedless clustering [4, 5] (described in greater detail below).
Searches for CBCs often use matched filtering, which requires precise knowledge of astrophysical waveforms. (Excess power searches are also used, especially for high-mass systems associated with shorter signals; see, e.g., [6].) Since CBCs are, for the most part, wellmodeled systems, matched filtering provides an essentially optimal strategy for detecting compact binaries. However, there are several reasons why it is useful to consider alternative detection strategies.
Independent verification. Alternative methods can provide independent verification of detections by
a Electronic address: [email protected]
matched filter pipelines, thereby increasing confidence in the veracity of a result (of course, because seedless clustering will be less sensitive than matched filtering searches in most cases, non-detection by seedless clustering is not a concern either). Although there is some redundancy provided by the multiple implementations of matched filtering used in current searches, seedless clustering provides a very different approach to gravitationalwave detection and detection by both methods potentially indicates the robustness of the result.
Visualization of the GW signal. In general, advanced detector CBC events are expected to be buried in noise to the extent that it will be difficult to see by eye their signature in a time series or strain auto-power spectrogram. Here we show that, by coherently combining the output of multiple detectors, CBCs can be visualized as faint but visible arcs on 'radiometric spectrograms'especially when the eye is guided by the reconstructed track of a search algorithm. (The curious reader is encouraged to skip ahead to Fig. 1 for an example of a radiometric spectrogram.)
Visualizing the signal helps confirm that the detected signal looks like one expects. The coherent combination also allows for confirmation that the parameter estimation of the signals, including the direction, masses, and time of coalescence, are all consistent 'by eye' with the radiometric spectrogram. For example, an error in the reconstructed CBC direction creates characteristic stripes; see [5]. If, on the other hand, the masses are incorrect, the reconstructed track will have the wrong frequency evolution as a function of time.
Data processing corner cases. Real-world GW searches require design choices, which take into account the complicated nature of GW detectors. Detector performance is non-stationary, the noise contains non-Gaussian 'glitches,' and data-taking is sometimes interrupted by lock-loss, just to name a few relevant effects. As a result, workarounds are employed, e.g., to estimate background, to discard noisy data, and to handle gaps. Matched fil-
tering [7] and seedless clustering [8] have different ways of performing these tasks. In general, these technical details are (by design) not important factors in determining the average sensitivity of a search. However, by employing multiple search methods we can guard against individual events falling between the cracks.
An example of a possible data processing corner case is shown in Fig. 2. This event was identified correctly by both matched filtering and seedless clustering. The lefthand panel shows ρ t ( ; f ) obtained using 'engineeringrun' data[34], in which data from a LIGO sub-system, in this case the pre-stabilized laser, is recolored to match the Advanced LIGO noise curve. Such engineering run data does not contain astrophysically useful strain measurements, but is nonetheless useful for its non-Gaussian noise characteristics. The data contains five segments consistent with non-Gaussian noise and would be removed in a search. Nonetheless, despite these noise artifacts, it is still possible to detect a simulated binary neutron star signal. The right-hand panel shows the reconstructed signal, obtained with the seedless clustering algorithm we describe below.
Waveform uncertainties. Theoretical errors in matched filter waveforms can arise from the computational limitations and/or imperfect approximations. Due to computational limitations, most CBC searches so far use template banks composed of non-spinning, noneccentric waveforms, which are less computationally challenging than search with spin and eccentricity. High-spin systems take longer to simulate with numerical relativity and the addition of extra spin parameters creates larger, more unwieldy template banks. Searches that ignore spin can suffer significant losses in sensitivity [9]; the cases considered here have between a 23-36% match, which is maximized over time and phase, between the spinning and non-spinning waveforms; these numbers increase to at least 97% when maximized over mass as well. When spin is included, it is often assumed that the spins are aligned in order to make the calculation more tractable. Even so, the inclusion spin effects can lead to a factor of two increase in sensitive volume [10]; see also [11, 12]. Main sequence binaries circularize by the time they enter the sensitive frequency band of terrestrial detectors [13]. However, dynamical capture may produce gravitational waves from highly eccentric binaries [1317]. The cases considered here have a less than 1% match between the eccentric and circular waveforms of equivalent mass. These numbers increase to between 20-60% when maximized over mass. This highlights the difficulty of detecting them using a template bank composed of circular templates. Imperfect assumptions about eccentricity and spin may therefore create openings for ostensibly sub-optimal detection strategies.
New physics. One can also imagine significant waveform errors due to the existence of new or unforeseen physics. For example, Piro raised concerns about the effects of magnetic interactions in BNS [18]. While these magnetic interactions were subsequently shown to be ig- norable [19] for GW astronomy, one can imagine a comparable source of theoretical error. More speculatively, non-standard theories of gravitation can lead to modifications of the waveform [20, 21].
Thus, there are many reasons why it is worth considering alternatives to matched filtering. One common alternative technique for detection of GW transients is to search for excess power in spectrograms (also called frequency-time ft -maps) of GW detector data [22-24]. This method casts GW searches as pattern recognition problems.
Previous work has shown how 'seedless clustering' can be used to perform sensitive searches for long-lived transients [4, 5]. The idea of seedless clustering is to integrate the signal power along spectrogram tracks chosen to capture the salient features of a wide class of signal models. Seedless clustering calculations are embarrassingly parallel, and so the technique benefits from the recent proliferation of highly parallel computing processors including graphical processor units and multi-core central processing units. Previous papers [4, 5] have pointed out that seedless clustering algorithms might be useful for CBC detection/confirmation.
In this work, we apply the seedless clustering formalism to efficiently search for CBC signals. In section II, we review the basics of seedless clustering. We show how the formalism of [4, 5] can be tuned to more sensitively detect CBC signals. In section IV, we determine the sensitivity of seedless clustering algorithms (with different levels of tuning) to CBC waveforms. We conclude with a discussion of topics for further study in section V.
## II. SEEDLESS CLUSTERING FOR CHIRPS
Searches for unmodeled GW transients typically begin with spectrograms proportional to GW strain power. The pixels of these spectrograms are computed by dividing detector strain time series in segments and computing Fourier transform of the segments. The Fourier transform of the strain data from detector I for the segment with a mid-time of t is denoted ˜ ( ; s I t f ). For the results presented here, we use 50%-overlapping, Hann-windowed segments with duration of 1 s. The frequency resolution is 1 Hz.
Searches for long-duration GW transients in particular use the cross-correlation of two GW strain channels from spatially separated detectors to construct ft -maps of cross-power signal-to-noise ratio [24]:
$$\text{ric} ^ { - } _ { \text{ibly} } \quad \rho ( t ; f | \hat { \Omega } ) = \text{Re} \left [ \lambda ( t ; f ) e ^ { 2 \pi i f \Delta \vec { x } \cdot \hat { \Omega } / c } \tilde { s } ^ { * } _ { I } ( t ; f ) \tilde { s } _ { J } ( t ; f ) \right ]. \ ( 1 ) \\ \text{rm} \quad \hat { \lambda } \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cd$$
Here, Ω is the direction of the GW source, ∆ ˆ /vector x is a vector describing the relative displacement of the two detectors, c is the speed of light, and e 2 πif ∆ /vector x · ˆ Ω /c is a directiondependent phase factor, which takes into account the time delay between the two detectors. The λ t ( ; f ) term is
a normalization factor, which employs data from neighboring segments to estimate the background at time t :
$$\lambda ( t ; f ) = \frac { 1 } { \mathcal { N } } \sqrt { \frac { 2 } { P _ { I } ^ { \prime } ( t ; f ) P _ { J } ^ { \prime } ( t ; f ) } }. \quad \begin{pmatrix} \text{be 1} \\ \text{by} \\ \max \\ \text{par} \end{pmatrix}$$
P ′ I ( ; t f ) and P ′ J ( ; t f ) are the auto-power spectral densities for detectors I and J in the segments neighboring t . For additional details, see [4, 5, 24]
GWs appear as tracks or blobs in the ft -maps. The morphology of the the GWs are dependent on the signal. CBC signals appear as chirps of increasing frequency. Clustering algorithms are used to identify clusters of pixels Γ likely to be associated with a GW signal. The total signal-to-noise ratio for a cluster of pixels can be written as a sum of over ρ t ( ; f | ˆ ): Ω
$$S N R _ { t o t } \equiv \frac { 1 } { N ^ { 1 / 2 } } \sum _ { \{ t ; f \} \in \Gamma } \rho ( t ; f | \hat { \Omega } ), \quad \ \ ( 3 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\quad \quad \quad$$
where N is the number of pixels in Γ.
Different clustering algorithms employ different methods for choosing Γ. Seed-based algorithms connect statistically significant seed pixels to form clusters [4, 25]. In seedless clustering algorithms [4], Γ is chosen from a bank of parametrized frequency-time tracks. Each such track is referred to as a 'template.' Calculations for many templates can be carried in parallel, which facilitates rapid calculations on multi-core devices such as graphical processor units (GPUs).
It must be noted that our templates are different from matched-filtering templates. Seedless clustering templates describe the morphology of a power spectrogram track whereas matched filter templates describe the phase evolution of a signal appearing in just one detector. Matched filter templates contain all the available information about the signal, whereas seedless clustering templates are a lossy, binned representation of the signal. By throwing away information, the seedless clustering search is less sensitive than a matched filter search, but by throwing away information, it can simultaneously become more robust against waveform uncertainties and new physics.
A very general search with minimal assumptions may employ, e.g., a template bank of randomly generated B´ ezier curves [26], which have been shown to do a reasonably good job of mimicking long-lived narrowband gravitational-wave signals [5]. However, one may equally well carry out a more specialized search, targeting a specific class of signals. Given our present interest in CBC signals, we opt to work with a more specialized template bank consisting of parametrized chirps:
$$f ( t ) = \frac { 1 } { 2 \pi } \frac { c ^ { 3 } } { 4 G M _ { \text{total} } } \sum _ { k = 0 } ^ { 7 } p _ { k } \tau ^ { - ( 3 + k ) / 8 }, \quad ( 4 ) \quad \text{$\text{$EAF}$} \\ \text{here}$$
where
$$\tau = \frac { \eta c ^ { 3 } ( t _ { c } - t ) } { 5 G M }. \quad \ \ \ \ ( 5 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
Here, G is the gravitational constant and M total is the total mass of the binary. The expansion coefficients p k can be found in [27]. Each chirp template is parametrized by two numbers: the coalescence time and the chirp mass. (This is in contrast to B´ ezier curves, which are parametrized by six numbers.) While technically, the waveform depends on the individual component masses, the main features of the signal can be well-approximated by only the chirp mass. Therefore, to reduce the parameter space by one variable, we employ the approximation that the individual component masses are equal.
The space of arbitrary long-lived gravitational-wave signals is very large, and so general algorithms, employing B´ ezier templates, use randomly generated numbers to span as much of the signal space as possible. The space of CBC chirp signals is much smaller. A search for binary neutron star signals with component masses of 1 4-3 . M /circledot plateaus in sensitivity using just 50 chirp mass bins. The same search requires 825 time bins for 660 s of data-a typical on-source window (in which the signal is assumed to exist) for a GW search triggered by a gammaray burst [28]. Thus, the template density for a targeted CBC search is ≈ 6 3 . × 10 4 ks -1 whereas a B´ ezier bank might require ≈ 8 × 10 8 ks -1 [5]; (1 ks = 1000 s). Since there are so many fewer templates in a chirp-template search (about four orders of magnitude fewer), it is computationally feasible to employ every template.
Of course, the previous calculation was for a targeted search in which the source location is previously determined, e.g., by an electromagnetic trigger or by a different search algorithm. In [5], we showed how the introduction of a phase factor can be used to allow to carry out an efficient all-sky search with seedless clustering. This formalism is straightforwardly applied to our chirping templates. For the CBCs detected by the LIGO Hanford-Livingston detector pair, it is sufficient to consider 40 time delays, each corresponding to a ring on the sky. Thus, even an all-sky search using seedless clustering to detect CBC signals can employ a relatively modest template density: ≈ 2 5 . × 10 6 ks -1 .
To estimate the computational cost of an all-sky seedless clustering search (with chirp-like templates), we carried out a benchmark study using a Kepler GK104s GPU and an 8-core Intel Xeon E5-4650 CPU. Each job was allotted 8 g of memory. The GPU was able to analyze 660 s of data in 48 s, corresponding to a duty cycle of ≈ 7%. Using all eight cores, the CPU duty cycle was comparable; the job-by-job variability in run time is greater than the difference between GPUs and 8-core CPUs on average.
If we require background estimation at the level of FAP = 1%, it follows that a continuously running seedless clustering search with chirp-like templates can be carried out with just 8 GPUs (or 8-core CPUs). (Background estimation at the level of FAP = 0 1% would re-. quire 74 GPUs / 8-core CPUs.) In reality, the duty cycle from coincident GW detectors may be ≈ 50%, in which case these computing requirements are conservative by a
factor of two. Repeating the test for a targeted search (for which the source location is known), we obtained an 8-core CPU duty cycle of 2%, a factor of 3 speed-up. The targeted search run on GPUs does not run appreciably faster than the all-sky version.
In addition to improved computational efficiency, there is another important advantage to be gained through the use of CBC templates compared to B´ ezier templates. In [5], we showed that quadratic B´zier curves do a mediocre e job approximating CBC signals. By adopting the chirping templates described in Eq. 4, we expect to capture more signal-to-noise ratio, and thereby extend the sensitive distance of the search. An example of a weak BNS signal recovered with a seedless chirping template is shown in Fig. 1.
## III. SENSITIVITY STUDY
In order to determine the sensitivity of seedless clustering with chirp templates, we perform a sensitivity study with Monte Carlo noise. We assume Gaussian noise consistent with the design sensitivity of Advanced LIGO. Following [28], we assume that an external trigger has predicted that the signal exists in a 660 s on-source window. For each trial, we search for a chirp signal four different ways: using B´zier curves and a known sky locae tion (BK), using B´zier with unknown sky location (BU), e using chirp templates with a known sky location (CK), and using chirp templates with an unknown sky location (CU).
The first step of the sensitivity study is background estimation. We perform many trials to estimate the distribution of SNR tot for noise. We generate separate noise distributions for all four search variations (BK, BU, CK, CU). Using these noise distributions, we determine the value of SNR tot (for each search variation), which corresponds to a false alarm probability (FAP) of 0 1%. .
The next step is to determine the distance to which different signals can be detected with SNR tot sufficient for a detection with FAP < 0 1%. . We add GW signals to realizations of detector noise. Each injected signal is injected with an optimal sky location and an optimal source orientation. We define the 'sensitive distance' as the distance at which 50% of the singals are recovered with FAP < 0 1%. We consider 14 CBC waveforms . with component masses ranging from 1 4-3 . M /circledot . Of these waveforms, eight characterize eccentric systems and three characterize systems where one or more object has a large dimensionless spin:
$$a \equiv c J / G m ^ { 2 }. \quad \quad \ \ ( 6 ) \quad \text{of} \quad \ \ \.$$
Here J is the angular momentum and m is the component mass.
Non-eccentric waveforms are generated using a SpinTaylorT4 approximation. Eccentric waveforms are generated using CBWaves , which employs all the contributions that have been worked out for generic eccentric orbits up to 2PN order [29]. The parameters for each waveform are give in Table I.
## IV. RESULTS
The results of our sensitivity study are summarized in Table I. There are a number of interesting trends. First, while all of the (known direction) seedless clustering distances are astrophysically interesting, the chirping templates perform consistently better than the B´zier teme plates. The ratio of detection distance for chirping templates / B´zier templates ranges from 100-161% with a e mean of 131%. The average ratio of sensitive volumes is 240%. This is comparable to the gain in sensitivity for a matched filter search to be had through the inclusion of spin [10], indicating that while it is certainly advantageous to use chirping templates, the B´zier templates do e surprisingly well.
We find no significant difference in the chirpingtemplate detection distance between systems that do or do not contain spin. The similarity in the sensitivity distances between the non-spinning and spinning cases indicates that the spins do not effect the signal morphology in a significant enough way to deviate from the nonspinning track. The advantage of chirping templates appears to increase slightly for spinning systems. Finally, we observe no loss in sensitivity going from the targeted CK search to the all-sky CU search. Evidently, the increase in signal space from the additional parameter of sky location is not sufficient to meaningfully affect the background distribution.
Highly eccentric signals generally have a longer duration than than those with low or no eccentricity. The sensitivity (using both B´ ezier templates and chirp-like templates) decreases with eccentricity. There is a similar advantage in distance of the chirp-like templates over the B´ ezier templates at low eccentricity. This benefit decreases slightly as eccentricity increases. This is due to the breakdown of the circular binary approximation. The breakdown becomes more pronounced at higher eccentricities, as one would expect.
Using matched filtering, Advanced LIGO, operating at design sensitivity, is expected to reliably detect BNS (with optimal orientation and sky location) out to distances of 450 Mpc [30], 2 4 . × further than the seedless clustering detection distance quoted here. It follows that ≈ 8% of the events detected by matched filtering will produce a FAP < 0 1% signature when followed up with . seedless clustering. Given a realistic astrophysical rate of 40 yr -1 BNS detections by Advanced LIGO [31], this implies that we can expect to confirm ≈ 3 events per year of science data using seedless clustering. The use of expanded template banks that include waveforms with spin will allow matched filtering searches to observe to comparable distances as the non-spinning case [11]; therefore the follow-up numbers will be similar to the above. In the event that circular template banks are used to search
FIG. 1: The plot on the left shows ρ t ( ; f ) for a simulated eccentric ( /epsilon1 = 0 2) BNS signal injected on top of Monte Carlo . detector noise. The component masses are 1 4 . M /circledot . The chirping signal appears as a faintly-visible track of lighter-than-average pixels. The horizontal lines are frequency notches to remove instrumental artifacts. On the right is the recovery obtained with seedless clustering. The signal is recovered with a FAP < . 0 1%.
0
50
100
150
200
250
0.5
0
-0.5
-1
t (s)
FIG. 2: The plot on the left shows ρ t ( ; f ) during a recent LIGO engineering run, in which data from a LIGO subsystem-not sensitive to GW strain-is recolored to produce semi-realistic detector noise. On the right is the seedless clustering recovery, which is able to to detect the injected signal with high confidence FAP < . 0 1%, despite the relatively poor data quality. (Though it is not immediately apparent from these plots, five segments are identified as characteristic of non-stationary noise [8].) The signal is recovered with FAP < . 0 1%.
for eccentric signals, there will be a non-negligable loss in sensitivity for these searches. Huerta and Brown estimate signal-to-noise ratio loss factors of about 0.5 and 0.2 for BNS systems with eccentricities of 0.2 and 0.4 respectively [32]. This would bring the matched filtering sensitivity distances of these signals to 225 Mpc and 90 Mpc; therefore seedless clustering may provide further opportunities for observing these types of signals.
In addition to providing confirmation of these loudest
CBC events, seedless clustering will provide a safety net by potentially detecting events missed due to waveform error, data-processing subtleties, and/or new physics.
## V. CONCLUSIONS
Seedless clustering provides a computationally efficient tool for the follow-up and detection of compact binary
100
1
90
80
70
60
50
40
30
20
f (Hz)
SNR
TABLE I: Sensitive distances for different waveforms (assuming optimal sky location and source orientation) given the design sensitivity of Advanced LIGO [30]. Each row represents a different waveform: BNS='binary neutron star,' NSBH='neutron star black hole binary,' BBH='binary black hole. A waveform beginning with an 'E' is eccentric. The columns marked m 1 and m 2 give the component masses in units of M /circledot . The columns marked a 1 and a 2 give the component spins; see Eq. 6. The next columns list the ellipticity /epsilon1 and the waveform duration in seconds. The final four columns list the (FAP = 0 1%, . FDP = 50%) detection distance (in Mpc) for B´zier templates with known sky location (BK), B´zier templates with unknown e e sky location (BU), chirp-like templates with known sky location (CK), and chirp-like templates with unknown sky location (CU).
| Waveform | m 1 | m 2 | a 1 | a 2 | /epsilon1 | t dur (s) | D BK | D BU | D CK | D CU |
|------------|-------|-------|-------|-------|-------------|-------------|--------|--------|--------|--------|
| BNS 1 | 1.4 | 1.4 | 0 | 0 | 0 | 170 | 160 | 130 | 190 | 190 |
| NSBH 1 | 3 | 1.4 | 0 | 0 | 0 | 96 | 200 | 200 | 290 | 290 |
| NSBH 2 | 3 | 1.4 | 0.95 | 0 | 0 | 97 | 220 | 180 | 320 | 320 |
| BBH 1 | 3 | 3 | 0 | 0 | 0 | 54 | 330 | 330 | 470 | 470 |
| BBH 2 | 3 | 3 | 0.95 | 0 | 0 | 55 | 320 | 270 | 470 | 470 |
| BBH 3 | 3 | 3 | 0.95 | 0.95 | 0 | 55 | 320 | 260 | 470 | 470 |
| BBH 4 | 5 | 5 | 0 | 0 | 0 | 42 | 620 | 560 | 750 | 750 |
| BBH 5 | 5 | 5 | 0.95 | 0 | 0 | 42 | 680 | 680 | 750 | 750 |
| BBH 6 | 5 | 5 | 0.95 | 0.95 | 0 | 43 | 750 | 680 | 830 | 830 |
| EBNS 1 | 1.4 | 1.4 | 0 | 0 | 0.2 | 120 | 150 | 120 | 180 | 180 |
| EBNS 2 | 1.4 | 1.4 | 0 | 0 | 0.4 | 224 | 150 | 120 | 160 | 160 |
| ENSBH 1 | 3 | 1.4 | 0 | 0 | 0.2 | 69 | 180 | 180 | 290 | 290 |
| ENSBH 2 | 3 | 1.4 | 0 | 0 | 0.4 | 127 | 180 | 160 | 240 | 240 |
| ENSBH 3 | 3 | 1.4 | 0 | 0 | 0.6 | 237 | 180 | 160 | 240 | 240 |
| EBBH 1 | 3 | 3 | 0 | 0 | 0.2 | 40 | 270 | 220 | 320 | 320 |
| EBBH 2 | 3 | 3 | 0 | 0 | 0.4 | 70 | 220 | 200 | 240 | 240 |
| EBBH 3 | 3 | 3 | 0 | 0 | 0.6 | 128 | 220 | 180 | 220 | 220 |
coalescences. While seedless clustering is expected to be less sensitive than matched filtering, it provides a number of useful features including independent verification, visualization of the gravitational-wave signal, the ability to catch corner-case signals, and robustness to both waveform uncertainty and existence of new physics.
We compared a specially tuned implementation of seedless clustering, optimized for compact binary coalescences, to a more generic search using B´zier curves. e We find that the CBC-tuned search can expand the sensitive volume by as much as a factor of 4 2 . × depending on the waveform (a factor of two on average) compared to the generic B´zier search. e Perhaps more importantly, the tuned search requires 10 4 fewer templates per unit of time, allowing for a significantly faster search.
There are a number of potential improvements to the algorithm worth exploring. It may be possible to improve the implementation of seedless clustering described here by more optimally weighting different time-frequency bins based on the known waveform. It is also worth exploring the effect of only using equal mass templates to recover potentially non-equal mass signals. It is possible that for cases with larger mass ratios than those considered here, it may be necessary to relax this assumption in order to reconstruct a majority of the signal. Another possibility for improvement is the implementation of a better parametrization for the eccentric waveforms. Unlike for circular binaries, there is, at present, no closed expression for the phase evolution of a binary with arbitrary eccentricity. A more effective parametrization is likely to capture more signal-to-noise ratio, thereby extending the sensitive range. Finally, it will be useful to carry out a systematic comparison of seedless clustering with matched filtering pipelines and using non-Gaussian noise.
## VI. ACKNOWLEDGMENTS
MC is supported by National Science Foundation Graduate Research Fellowship Program, under NSF grant number DGE 1144152. ET is a member of the LIGO Laboratory, supported by funding from United States National Science Foundation. NC's work was supported by NSF grant PHY-1204371. LIGO was constructed by the California Institute of Technology and Massachusetts Institute of Technology with funding from the National Science Foundation and operates under cooperative agreement PHY-0757058. This paper has been assigned LIGO document number P1400110-v5.
FIG. 3: Data from a recent LIGO engineering run, in which data from a LIGO subsystem-not sensitive to GW strain-is recolored to produce semi-realistic detector noise. A simulated binary neutron star signal has been added to the data. The top plot shows a wavelet transform of single-detector auto-power [33], currently in wide use for diagnostics. The injected waveform, which ends at t = 0, is difficult to make out by eye. The middle plot shows a spectrogram of ρ t ( ; f ). The injection, though faint, is visible between 200-250 s. The bottom plot shows the reconstructed track using seedless clustering; FAP < . 0 1%.
- [1] J. Aasi et al. (LIGO Scientific Collaboration and Virgo Collaboration), Phys. Rev. D 87 , 022002 (2013).
- [2] J. Abadie et al. (LIGO Scientific Collaboration and Virgo Collaboration), Physical Review D 85 , 082002 (2012).
- [3] Abadie J. et al. (LIGO Scientific Collaboration and Virgo Collaboration), Classical and Quantum Gravity 27 , 173001 (2010).
- [4] E. Thrane and M. Coughlin, Phys. Rev. D 88 , 083010 (2013).
- [5] E. Thrane and M. Coughlin, Phys. Rev. D 89 , 063012 (2014).
- [6] J. Abadie et al., Phys. Rev. D 85 , 102004 (2012).
- [7] S. Babak, R. Biswas, P. R. Brady, D. A. Brown, K. Cannon, C. D. Capano, J. H. Clayton, T. Cokelaer, J. D. E. Creighton, T. Dent, et al., Phys. Rev. D 87 , 024033 (2013), URL http://link.aps.org/doi/10.1103/PhysRevD.87.024033 .
- [8] T. Prestegard, E. Thrane, et al., Class. Quant. Grav. 29 , 095018 (2012).
- [9] Canton T. et al, Implementing a search for alignedspin neutron star -black hole systems with advanced ground based gravitational wave detectors (2014), URL http://arxiv.org/abs/1405.6731 .
- [10] S. Privitera, The importance of spin for observing gravitational waves from coalescing compact binaries with LIGO and Virgo. (2014).
- [11] D. A. Brown, I. Harry, A. Lundgren, and A. H. Nitz, Phys. Rev. D 86 , 084017 (2012).
- [12] S. Privitera, S. R. P. Mohapatra, P. Ajith, K. Cannon, N. Fotopoulos, M. A. Frei, C. Hanna, A. J. Weinstein, and J. T. Whelan, Phys. Rev. D 89 , 024003 (2014).
- [13] D. A. Brown and P. J. Zimmerman, Phys. Rev. D 81 , 024007 (2010).
- [14] R. M. O'Leary, B. Kocsis, and A. Loeb, MNRAS 395 , 2127 (2009).
- [15] U. Sperhake, E. Berti, V. Cardoso, J. A. Gonz´ alez, B. Br¨gmann, and M. Ansorg, Phys. Rev. D u 78 , 064069 (2008).
- [16] I. Hinder, F. Herrmann, P. Laguna, and D. Shoemaker, Physical Review D. 82 (2010), ISSN 0556-2821.
- [17] R. Gold, S. Bernuzzi, M. Thierfelder, B. Brugmann, and F. Pretorius, Phys.Rev. D86 , 121501 (2012), 1109.5128.
- [18] T. Piro, Astrophys. J. 755 , 80 (2012).
- [19] D. Lai, 757 , L3 (2012).
- [20] M. Shibata, K. Taniguchi, H. Okawa, and A. Buonanno, Phys. Rev. D 89 , 084005 (2014), URL http://link.aps.org/doi/10.1103/PhysRevD.89.084005 .
- [21] E. Barausse, C. Palenzuela, M. Ponce, and L. Lehner, Phys. Rev. D 87 , 081506 (2013), URL http://link.aps.org/doi/10.1103/PhysRevD.87.081506 .
- [22] P. Sutton et al., New Journal of Physics 12 , 053034 (2010).
- [23] S. Klimenko, I. Yakushin, A. Mercer, and G. Mitselmakher, Class. Quant. Grav. 25 , 114029 (2008).
- [24] E. Thrane et al., Physical Review D 83 , 083004 (2011).
- [25] T. Prestegard and E. Thrane, Burstegard: a hierarchical clustering algorithm (2012), https://dcc.ligo.org/LIGO-L1200204-v1 .
- [26] G. Farin, Curves and Surfaces for CAGD, Fourth Edition: A Practical Guide (Academic Press, 1996).
- [27] Blanchet L., Living Reviews in Relativity 4 (2006), URL http://www.livingreviews.org/lrr-2006-4 .
- [28] Abadie J. et al. (LIGO Scientific Collaboration and Virgo Collaboration), The Astrophysical Journal 760 , 12 (2012).
- [29] Csizmadia P. et al., Classical and Quantum Gravity 29 , 245002 (2012).
- [30] J. Aasi et al., Prospects for Localization of Gravitational Wave Transients by the Advanced LIGO and Advanced Virgo Observatories (2013), URL http://arxiv.org/abs/1304.0670 .
- [31] J. Abadie et al., Class. Quant. Grav 27 , 173001 (2010).
- [32] E. A. Huerta and D. A. Brown, Phys. Rev. D 87 , 127501 (2013).
- [33] J. Rollins, Multimessenger Astronomy with Low-Latency Searches for Transient Gravitational Waves. (2011).
- [34] The data in question are from LIGO Engineering Run 5, collected during 2014.
|
10.1103/PhysRevD.90.083005
|
[
"Michael Coughlin",
"Eric Thrane",
"Nelson Christensen"
] |
2014-08-05T00:01:10+00:00
|
2014-08-05T00:01:10+00:00
|
[
"gr-qc",
"astro-ph.IM"
] |
Detecting compact binary coalescences with seedless clustering
|
Compact binary coalescences are a promising source of gravitational waves for
second-generation interferometric gravitational-wave detectors. Although
matched filtering is the optimal search method for well-modeled systems,
alternative detection strategies can be used to guard against theoretical
errors (e.g., involving new physics and/or assumptions about spin/eccentricity)
while providing a measure of redundancy. In previous work, we showed how
"seedless clustering" can be used to detect long-lived gravitational-wave
transients in both targeted and all-sky searches. In this paper, we apply
seedless clustering to the problem of low-mass ($M_\text{total}\leq10M_\odot$)
compact binary coalescences for both spinning and eccentric systems. We show
that seedless clustering provides a robust and computationally efficient method
for detecting low-mass compact binaries.
|
1408.0841v1
|
## Supercontinuum optimization for dual-soliton based light sources using genetic algorithms in a Grid platform
F. R. Arteaga-Sierra, 1 2 , , ∗ C. Mili´ an, 2 3 6 , , I. Torres-G´ omez, 1 M. Torres-Cisneros, 4 G. Molt´ o, 3 and A. Ferrando 2 5 ,
1 Centro de Investigaciones en Optica, A.C., Le´n Gto. 37150, M´ exico ´ o 2 Universidad Polit´ ecnica de Valencia, Grupo de Modelizaci ´n Multidisciplinar Intertech, o 46022, Valencia, Spain
3 Instituto de Instrumentaci´n para Imagen Molecular (I3M). Centro mixto CSIC-Universitat o Polit`cnica de Val` encia-CIEMAT, camino de Vera s/n, 46022 Valencia, Espa˜ na e 4 NanoBioPhotonics Group, DICIS, Universidad de Guanajuato, Mexico 5 Departament d' ` ptica, Universitat de Val` encia, Burjassot (Val`ncia). 46100, Spain O e 6 Current address: Centre de Physique Th´ eorique, Ecole Polytechnique, CNRS, F-91128 ´ Palaiseau, France
*[email protected]
6 [email protected]
Abstract: We present a numerical strategy to design fiber based dual pulse light sources exhibiting two predefined spectral peaks in the anomalous group velocity dispersion regime. The frequency conversion is based on the soliton fission and soliton self-frequency shift occurring during supercontinuum generation. The optimization process is carried out by a genetic algorithm that provides the optimum input pulse parameters: wavelength, temporal width and peak power. This algorithm is implemented in a Grid platform in order to take advantage of distributed computing. These results are useful for optical coherence tomography applications where bell-shaped pulses located in the second near-infrared window are needed.
OCIS codes: (190.4370) Nonlinear optics; (060.5530) Pulse propagation and temporal solitons; (230.6080) Sources; (110.2945) Illumination design; (170.4500) Optical coherence tomography.
## References and links
- 1. J. P. Gordon, 'Theory of the soliton self-frequency shift,' Opt. Lett. 11, 662-664 (1986).
- 2. F. M. Mitschke and L. F. Mollenauer, 'Discovery of the soliton self-frequency shift,' Opt. Lett. 11, 659-661 (1986).
- 3. J. M. Dudley, G. Genty and S. Coen, 'Supercontinuum generation in photonic cristal fibers,' Rev. Mod. Phys. 78, 135-1184 (2006).
- 4. D. V. Skryabin and A. V. Gorbach, 'Colloquium: looking at a soliton through the prism of optical supercontinuum,' Rev. Mod. Phys. 82, 1287-1299 (2010).
- 5. A. V. Gorbach and D. V. Skryabin, 'Light trapping in gravity-like potentials and expansion of supercontinuum spectra in photonic-crystal fibres,' Nat. Photon. 1, 1749-4885 (2007).
- 6. A. Hause, T. X. Tran, F. Biancalana, A. Podlipensky, P. St.J. Russell and F. Mitschke, 'Understanding Ramanshifting multipeak states in photonic crystal fibers: two convergent approaches,' Opt. Lett. 35, 2167-2169 (2010).
- 7. A. Hause and F. Mitschke, 'Soliton trains in motion,' Phys. Rev. A 82, 043838 (2010).
- 8. T. X. Tran, A. Podlipensky, P. St. J. Russell and F. Biancalana, 'Theory of Raman multipeak states in solid-core photonic crystal fibers,' J. Opt. Soc. Am. B 27, 1785-1791 (2010).
- 9. Andrey V. Gorbach and Dmitry V. Skryabin, 'Soliton self-frequency shift, non-solitonic radiation and selfinduced transparency in air-core fibers,' Opt. Express 16, 4858-4865 (2008).
- 10. C. Mili´ an, D. V . Skryabin and A. Ferrando, 'Continuum generation by dark solitons,' Opt. Lett. 34, 2096-2098 (2009).
- 11. C. Mili´ an, A. Ferrando, and D. V . Skryabin, 'Polychromatic Cherenkov radiation and supercontinuum in tapered optical fibers,' J. Opt. Soc. Am. B 29, 589-593 (2012).
- 12. F. R. Arteaga-Sierra, C. Mili´n, I. Torres-G´mez, M. Torres-Cisneros, A. Ferrando, and A. D´ avila, 'Multi-peaka o spectra generation with Cherenkov radiation in a non-uniform single mode fiber,' Opt. Express 22, 2451-2458 (2014).
- 13. S. A. Dekker, A. C. Judge, R. Pant, I. Gris-S´ anchez, J. C. Knight, C. M. De Sterke and B. J. Eggleton, 'Highlyefficient, octave spanning soliton self-frequency shift using a specialized photonic crystal fiber with low OH loss,' Opt. Express 18, 17766-17773 (2011).
- 14. J. Rothhardt, A. M. Heidt, S. H¨ adrich, S. Demmler, J. Lim pert and A. T¨ unnermann, 'High stability soliton frequency-shifting mechanisms for laser synchronization applications,' J. Opt. Soc. Am. B 29, 1257-1262 (2012).
- 15. A. M. Al-kadry and M. Rochette, 'Mid-infrared sources based on the soliton self-frequency shift,' J. Opt. Soc. Am. B 29, 1347-1355 (2012).
- 16. A. C. Judge, O. Bang, B. J. Eggleton, B. T. Kuhlmey, E. C. M¨gi, R. Pant and C. Martijn de Sterke, 'Optimization a of the soliton self-frequency shift in a tapered photonic crystal fiber,' J. Opt. Soc. Am. B 26, 2064-2071 (2009).
- 17. S. Pricking and H. Giessen, 'Tailoring the soliton and supercontinuum dynamics by engineering the profile of tapered fibers,' Opt. Express 18, 20151-20163 (2010).
- 18. R. Pant, A. C. Judge, E. C. Magi, B. T. Kuhlmey, M. De Sterke and B. J. Eggleton, 'Characterization and optimization of photonic crystal fibers for enhanced soliton self-frequency shift,' J. Opt. Soc. Am. B 27, 18941901 (2010).
- 19. G. Molt´ o, M. Arevalillo-Herr´ez, a C. Mili´ an, M. Zacar ´ es, V . Hern´ndez, and A. Ferrando, 'Optimization of sua percontinuum spectrum using genetic algorithms on service-oriented grids,' in Proceedings of the 3rd Iberian Grid Infrastructure Conference (IberGrid), pp. 137-147 (2009).
- 20. A. Ferrando, C. Mili´ an, N. Gonz´lez, G. Molt´, P. Loza , M. Arevalillo-Herr´ez, M. Zacar´s, I. Torres-G´mez an d a o a e o V. Hern´ andez, 'Designing supercontinuum spectra using Grid technology,' Proc. SPIE 7839, 78390W (2010).
- 21. N. Akhmediev and M. Karlsson, 'Cherenkov radiation emitted by solitons in optical fibers,' Phys. Rev. A 51, 2602-2607 (1995).
- 22. J. Wang, Y. J. Geng, B. Guo, T. Klima, B. N. Lal, J. T Willerson and W. Casscells, 'Near-infrared spectroscopic characterization of human advanced atherosclerotic plaques,' J. Am. Coll. Cardiol. 39, 1305-1313 (2002).
- 23. Y. M. Wang, J. S. Nelson, Z. P. Chen, B. J. Reiser, R. S Chuck and R. S. Windeler,'Optimal wavelength for ultrahigh-resolution optical coherence tomography,' Opt. Express 11, 1411-1417 (2003).
- 24. G, Humbert, W. J. Wadsworth, S. G. Leon-Saval, J. C. Knight, T. A. Birks, P. St. J. Russell, M. J. Lederer, D. Kopf K. Wiesauer, E. I. Breuer and D. Stifter, 'Supercontinuum generation system for optical coherence tomography based on tapered photonic crystal fibre,' Opt. Express 14, 1596-1603 (2006).
- 25. Y. Wang, Y. Zhao, J. S. Nelson, Z. Chen, R. S. Windeler, 'Ultrahigh-resolution optical coherence tomography by broadband continuum generation from a photonic crystal fiber,' Opt. Lett. 28, 182-184 (2003).
- 26. F. Spoeler, S. Kray, P. Grychtol, B. Hermes, J. Bornemann, M. Foerst and H. Kurz, 'Simultaneous dual-band ultra-high resolution optical coherence tomography,' Opt. Express 15, 10832-10841 (2007).
- 27. A.M. Smith, M.C. Mancini, and S. Nie, 'Bioimaging: Second window for in vivo imaging,' Nat. Nanotechnol. 4, 710-711 (2009).
- 28. J.M. Huntley, T. Widjanarko, and P.D. Ruiz, 'Hyperspectral interferometry for single-shot absolute measurement of two-dimensional optical path distributions,' Meas. Sci. Technol. 21, 075304 (2010).
- 29. Q. Cao, N.G. Zhegalova, S.T. Wang, W.J. Akers, and M.Y. Berezin, 'Multispectral imaging in the extended near-infrared window based on endogenous chromophores,' J. Biomed. Opt. 18, 101318-101318 (2013).
- 30. Y. Kodama and A. Hasegawa, 'Nonlinear Pulse Propagation in a Monomode Dielectric Guide,' IEEE J. Quantum Elect. 23, 510-524 (1987).
- 31. R. Driben, B. A. Malomed, A. V. Yulin and D. V. Skryabin, 'Newton's cradles in optics: From N -soliton fission to soliton chains,' Phys. Rev. A 87, 063808 (2013).
- 32. S. N. Sivanandam and S. N. Deepa, Introduction to Genetic Algorithms (Springer, 2010).
- 33. F. I. Feldchtein, G. V . Gelikonov, V . M. Gelikonov, R. R. Iksanov, R. V . Kuranov, A. M. Sergeev, N. D. Gladkova, M. N. Ourutina, J. A.Warren, and D. H. Reitze, 'In vivo OCT imaging of hard and soft tissue of the oral cavity,' Opt. Express 3, 239?250 (1998).
- 34. V. M. Gelikonov, G. V. Gelikonov, and F. I. Feldchtein, 'Two-wavelength optical coherence tomography,' Radiophys. Quantum Electron. 47, 848?859 (2004).
- 35. J.G. Fujimoto, C. Pitris, S. A. Boppart and M.E, Brezinski, 'Optical coherence tomography: An emerging technology for biomedical imaging and optical biopsy,' Neoplasia 2, 9-25 (2000).
- 36. J. G. Fujimoto, 'Optical coherence tomography for ultrahigh resolution in vivo imaging,' Nat. Biotechnol. 21, 1361-1367 (2003).
- 37. E. Kerrinckx, L. Bigot, M. Douay and Y. Quiquempois, 'Photonic crystal fiber design by means of a genetic algorithm,' Opt. Express, 12, pp. 1990-1995 (2004).
- 38. W. Q. Zhang, J. E. Sharping, R. T. White, T. M. Monro and S. Afshar V., 'Design and optimization of fiber optical parametric oscillators for femtosecond pulse generation,' Opt. Express 18, 17294-17305 (2010).
- 39. W. Q. Zhang, S. Afshar V. and T. M. Monro, 'A genetic algorithm based approach to fiber design for high coherence and large bandwidth supercontinuum generation,' Opt. Express 17, 19311-19327 (2009).
- 40. R. R. Musin and A. M. Zheltikov, 'Designing dispersion-compensating photonic-crystal fibers using a genetic algorithm,' Opt. Commun. 281, 567-572 (2008).
- 41. Y. Guo-Bing, L. Shu-Guang, L. Shuo and W. Xiao-Yan, ' The Optimization of Dispersion Properties of Photonic Crystal Fibers Using a Real-Coded Genetic Algorithm,' Chinese Phys. Lett. 28, 064215 (2011).
- 42. G. P. Agrawal, Nonlinear Fiber Optics, 4th ed. (Academic Press. 2007).
- 43. S. Afshar V. and T. M. Monro, 'A full vectorial model for pulse propagation in emerging waveguides with subwavelength structures part I: Kerr nonlinearity,' Opt. Express 17, 2298-2318 (2009).
- 44. R. H. Stolen, J. P. Gordon, W. J. Tomlinson and H. A. Haus, 'Raman response function of silica-core fibers,' J. Opt. Soc. Am. B 6, 1159-1166 (1989).
- 45. R. B. Agrawal and K. Deb, Simulated Binary Crossover for Continuous Search Space (Technical report, 1994).
- 46. K. Deb, Multi-Objective Optimization using Evolutionary Algorithms (Wiley & Sons, 2001).
- 47. A.N. Bashkatov1, E.A. Genina, V.I. Kochubey, and V.V. Tuchin, 'Optical properties of human skin, subcutaneous and mucous tissues in the wavelength range from 400 to 2000 nm,' J. Phys. D: Appl. Phys. 38, 25432555 (2005).
- 48. R. Tripathi, N. Nassif, J. S. Nelson, B. H. Park and J. F. de Boer, 'Spectral shaping for non-Gaussian source spectra in optical coherence tomography,' Opt. Lett. 27, 406-408 (2002).
- 49. J. A. Izatt and M. A. Choma, Optical Coherence Tomography Technology and Applications (Springer, 2008).
## 1. Introduction
The Soliton self-frequency shift (SSFS) [1, 2] plays a central role in many effects taking place during supercontinuum (SC) generation in optical fibers (see Refs. [3, 4] for a review on the topic). To mention only a few examples, light trapping [5], multi-peak soliton states [6-8], emission of Airy waves [9], intense dark-soliton SC [10] or broad and intense blue shifted polychromatic dispersive waves [11,12], would not be possible (or strong enough) without the SSFS. One of the most notorious feature of the Raman effect in SC generation with femtosecond pulses corresponds to the Raman soliton carrying the lowest frequency. Its large frequency shift from the laser pulse has motivated infra-red (IR) Raman soliton sources [13-15] and their optimization [16-20].
In the previous work of Ref. [12], we have shown that fs-pulses traveling in a dispersion engineered single mode fiber (SMF) can generate several pre-defined spectral peaks in the normal group velocity dispersion (GVD) region. These peaks correspond to the narrow band Cherenkov radiation emitted by the bright solitons [21] undergoing Raman red-shift and recoil. Such spectra were important for applications in optical coherence tomography (OCT) with wavelengths in the near infra-red (NIR) window [22-24]: λ /lessorsimilar 1 µ m.
In the present work, we are interested in OCT applications enabled by fiber based illumination [25,26] in the second near IR window (NIR II) [27-29] where λ ∈ [ 1 1 4 , . ] µ m, typically in the anomalous GVD regime of highly nonlinear photonic crystal fibers (PCFs). Because of the typical dispersion landscape, sources in the NIR II window may be based on the bright optical solitons arising during SC generation through the intricate soliton fission effect [30, 31]. With this picture in mind, we used an efficient and general computational optimization method based on genetic algorithms (GA) [32] that is capable of finding output spectra [see Fig. 1] exhibiting two peaks centered at pre-defined wavelengths. This type of multi-peak spectral illumination is very often required for OCT applications in the NIR II [29] and the simultaneous presence of the two operating spectral components is a prerequisite for real time imaging [33, 34]. The two peaks we obtain are the first and second Raman solitons presenting a clean bell-shaped spectral profile, essential for OCT [35], with widths providing a decent longitudinal resolution l c ≈ 10 µ m [36]. To demonstrate the usefulness of this method, we consider in this work the two spectral channels separated by 100 nm (see, e.g., Ref. [29]). The optimization method finds the optimal input pulse parameters, namely central wavelength, λ 0, temporal width, T 0, and peak power, P 0, yielding the desired spectra. The obtained peak powers are of up to 90 mW
for each spectral band, satisfying the needs for OCT imaging applications [35]. We find by this method the possibility to tune the wavelength of the target spectral channels, what represents an important feature of this strategy since the greatest potential of the given PCF is exploited, specially in situations where limited choice of PCF designs is available.
It is worth mentioning that the use of GAs in optics has indeed proofed useful previously in solving satisfactorily the inverse optimization problem of the one we present here, i.e., the design of PCFs to control SC dynamics, in a wide range of situations [37-41]. The use of GAs generally requires a large amount of simulations. For this reason we used a distributed computing (Grid) platform to reduce the time required to find the optimal solutions. The advantage of this infrastructure is that it enables the use of the same code in a platform of scalable resources which are adapted according to the needs of the particular problem.
## 2. Supercontinuum modeling and the Genetic Algorythm
We simulate the nonlinear propagation of the complex electric field envelope, A , along the fiber axis, z , by integrating numerically (with split-step fourier method) the generalized nonlinear Schr¨ odinger equation (GNLSE) [42],
$$- i \partial _ { z } A ( z, t ) = \sum _ { q \geq 2 } \frac { \beta _ { q } ( \omega _ { 0 } ) } { q! } [ i \partial _ { t } ] ^ { q } A ( z, t ) + \gamma A ( z, t ) \int _ { - \infty } ^ { + \infty } d t ^ { \prime } R ( t ^ { \prime } ) | A ( z, t - t ^ { \prime } ) | ^ { 2 }, \quad ( 1 ) \\. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad.$$
where the β q 's account for the linear fiber dispersion and
$$\gamma = \frac { \varepsilon \varepsilon _ { 0 } ^ { 2 } \omega _ { 0 } c } { 3 } \frac { \int \int d x d y n _ { 2 } ( x, y ) [ 2 | \vec { E } | ^ { 4 } + | \vec { E } ^ { 2 } | ^ { 2 } ] } { [ \int \int d x d y R e \{ \vec { E } \times \vec { H } ^ { * } \} \hat { u } _ { z } ] ^ { 2 } }, \\ \text{earity parameter [43]} \text{ which has been computed with a FEM solver (comsol)}$$
is the nonlinearity parameter [43] which has been computed with a FEM solver (comsol) by integrating the electromagnetic components of the modal field at λ = 800 nm along the transverse fiber cross section. ε = 2 09 and . n 2 = 2 6 . × 10 -20 m 2 /W are the relative permitivitty and nonlinear index of silica glass, respectively. The nonlinear response of the glass is R t ( ) = [ 1 -f R ] δ ( ) + t f RhR ( ) t [44], where δ ( ) t is the Dirac delta function and the Raman (delayed) contribution is weighted with f R = 0 18 and described by the damped oscillator .
$$h _ { R } ( t ) = \frac { \tau _ { 1 } ^ { 2 } + \tau _ { 2 } ^ { 2 } } { \tau _ { 1 } \tau _ { 2 } ^ { 2 } } \Theta ( t ) \mathrm e ^ { - \frac { t } { \tau _ { 2 } } } \sin \left ( \frac { t } { \tau _ { 1 } } \right ),$$
where τ 1 = 12 2 fs, . τ 2 = 32 fs, and Θ ( ) t is the Heaviside step function. The input pulses used in our simulations are of the form √ P 0 sech ( t / T 0 ) , where (for a fixed λ 0) the input peak power controls the soliton order N ≡ T 0 √ γ P 0 / | β 2 | .
We consider a length L = 25 cm of the NL-2.4-800 PCF, exhibiting the lowest zero GVD at 798 nm [see details in Fig. 1], convenient for broad band SC generation with Ti:Sapphire lasers. The frequency conversion performance is investigated within the attainable ranges of input pulse parameters: λ ∈ [ 750 850 , ] nm, P ∈ [ 5 15 , ] kW, and T ∈ [ 30 150 , ] fs. For each simulation along the PCF, the GA generates an individual with the genome | g 〉 ≡ [ g 1 , g 2 , g 3 ] T = [ T 0 , λ 0 , P 0 ] T and evaluates how suitable that individual is from the simulation output through the fitness function (to be minimized) defined as
$$\phi \equiv \psi _ { 1 } ^ { - 1 } \cdot \psi _ { 2 } ^ { - 1 }, \quad \psi _ { j } ( \omega _ { c _ { j } } ; \Delta \omega ) \equiv \int _ { \omega _ { c _ { j } } - \Delta \omega } ^ { \omega _ { c _ { j } } + \Delta \omega } d \omega ^ { \prime } | \tilde { A } ( L, \omega ^ { \prime } ) | ^ { 2 }, \quad j = 1, 2, \quad ( 4 )$$
where 2 ∆ω is the chosen spectral channel widths and ω c j = 2 π c / λ c j the central frequency. Note that the definition of φ as a product tends to favor output spectra in the form ψ 1 ≈ ψ 2
Fig. 1. (Bottom) Dispersion and cross section of the PCF used in our modeling. Shaded area marks the range of the input pump, λ 0 . (Top) Typical spectral output and target channels centered at λ c 1 2 , (delimited by dashed lines) in which ψ 1 2 , are evaluated [See Eq. (4)]. The vertical black line marks the zero GVD and the red one the λ 0 used in this example.
Fig. 2. (a) Sketch of the three GA stages: pth = 50, pmax = 100. Probability distributions for (b) simulated binary crossover (SBX) and (c) polynomial mutation.
amongst all solutions with ψ 1 + ψ 2 = const . [see Fig. (3) for optimum results]. The optimization process is depicted in Fig. 2(a). After an initial set of pth randomly (uniformly distributed) generated individuals (stage 1), the genetic operators (GOs) crossover, ˆ , X and mutation, ˆ M , are responsible for generating the new offsprings , which are added to the population until the maximum size p = pmax > pth is reached (stage 2). During stage 3 the population size is kept constant, p = pmax , and a replace the worst strategy is used, i.e., the offspring is added to the population if φ of f spring < φ max or disregarded otherwise. This steady state GA keeps the Grid constantly computing new individuals in parallel and fully exploits the processing power of the Grid. We briefly describe the GOs below.
Cross-over, ˆ , is the first GO applied to the current population and generates two X childs | gc 1 2 , 〉 by combining two randomly chosen parents , | gp 1 2 , 〉 , i.e., ˆ X [ | g p 1 〉 T , | g p 2 〉 T , | / 0 〉 T , | / 0 〉 T T ] = [ | g p 1 〉 T , | g p 2 〉 T , | g c 1 〉 T , | g c 2 〉 T ] T (being | / 0 〉 the empty set). We have used simulated binary crossover
(SBX) [45] and the [ 12 × 12 operator ]
$$\hat { x } \equiv \left [ \begin{array} { c c c c } \hat { 1 } & \hat { 0 } & \hat { 0 } & \hat { 0 } \\ \hat { 0 } & \hat { 1 } & \hat { 0 } & \hat { 0 } & \hat { 0 } \\ \hat { \alpha } _ { + } & \hat { \alpha } _ { - } & \hat { 0 } & \hat { 0 } \\ \hat { \alpha } _ { - } & \hat { \alpha } _ { + } & \hat { 0 } & \hat { 0 } \end{array} \right ] ; \, ( \hat { \alpha } _ { + } ) _ { j k } \equiv x _ { k } \frac { 1 \pm \bar { \sigma } _ { k } } { 2 } \zeta _ { j k } ; \, x _ { k } \equiv \Theta ( u _ { k } - 0. 0 5 ), \quad \ \ ( 5 ) \\ \dots \, \hat { \alpha } _ { - } \, \hat { \alpha } _ { + } \, \hat { \alpha } _ { + } \, \hat { \alpha } _ { + } \, \hat { \alpha } _ { + } \, \dots \, \dots \, \dots \, \pi \, \pi \, \dots \, \dots \, \pi \, \pi \, \dots \, \dots \, \pi \, \pi \, \dots \, \dots \, \pi \, \pi \, \dots \, \dots \, \pi \, \pi \, \dots \, \dots \, \pi \, \pi \, \dots \, \dots \, \pi \, \dots \, \pi \, \dots \, \pi \, \dots \, \pi \, \dots \, \pi \, \dots \, \pi \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \delta \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \pi \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \delta \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dts \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dDS \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dors \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \dots \, \dots \, \dots \dots \, \dots \dots \dots \, \dots \dots \dots \, \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \ dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots. \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \Dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots$$
where ˆ 0 ≡ × 0 ˆ (being ˆ the 2 I I × 2 identity matrix). The crossover activators , xk , set a probability for cross over of 95% per gene. The stochastic variables in this case, ¯ σ k [see Fig. 2(b)], are chosen from a uniform random number uk ∈ [ 0 1 , ] according to,
$$\bar { \sigma } _ { k } = \sigma \, / \int _ { 0 } ^ { \sigma } \mathcal { P } _ { x } ( \sigma ) = u _ { k } ; \, \mathcal { P } _ { x } ( \sigma ) = \begin{cases} 0. 5 [ n + 1 ] \sigma ^ { n }, \, \sigma \leq 1 \\ 0. 5 [ n + 1 ] \sigma ^ { - [ n + 2 ] }, \, \sigma > 1 \end{cases} \,.$$
The polynomial mutation, ˆ M , [46] suitable for real coded problems (continuous valued variables), is applied after ˆ X and generates new genes as ˆ M : | g 〉 →| g ′ 〉 ,
$$g _ { k } ^ { \prime } = g _ { k } + m _ { k } \Delta _ { k } \bar { \zeta } _ { k } ; \quad m _ { k } \equiv \Theta \left ( u _ { k } - \frac { 2 } { 3 } \right ),$$
where 2 ∆ k is the interval size of each variable ( ∆τ = 60 fs, ∆λ = 50 nm, ∆ P = 5 kW). Hence, in average only one gene is mutated per individual when mutation is applied: Θ ( uk -2 / 3 ) ⇒ P mk ( = ) = 1 1 / 3. ¯ ζ k ∈ -[ 1 1 , ] satisfies the normalized probability distribution [see Fig. 2(c)]
$$\mathcal { P } _ { m } ( \zeta ) = 0. 5 \left \{ n + 1 [ 1 - | \zeta | ] ^ { n } \right \},$$
with n = 20, peaked around ζ = 0 and clearly different from the random generation. For each gene the stochastic variable ¯ is chosen from a random ζ u ∈ [ 0 1 : , ]
$$\bar { \zeta } _ { k } = \zeta / \int _ { - 1 } ^ { \zeta } \mathcal { P } _ { m } ( \zeta ) = u _ { k } ; \quad u _ { k } \in [ 0, 1 ].$$
Statistically, ˆ M provides diversity to the population and ˆ X explores the parameter space in the vicinity of the parents. In this particular optimization problem, each individual evaluation typically required ∼ 90 s what amounted for about 7 5 h of CPU time to perform a single run of . the GA with 300 individuals [see Fig. 4]. We used a cluster of 12 cores within a Grid infrastructure, which reduced the computation time to ∼ 40 minutes. The Grid protocols supporting the GA execution make infrastructure resizable according to the needs of the problem: number of executions, dimensionality of the search space, etc. (see Ref. [19] for details on the Grid).
## 3. Dual-pulse solitonic source optimization
As mentioned above, the desired output spectral channels in this work intend to cover OCT applications in the NIR II window, where transparency of the biological tissues increases and scattering decreases [47]. Because spectral bell-shaped pulses avoid spurious structures in OCT images [48], the bright optical solitons are very good candidates for OCT applications. Figure 3 shows the spectral evolutions (bottom) and output spectrograms (top) corresponding to the best individuals obtained by the GA strategy and fitness function, Eq. 4, described in the previous section. All output spectra shown in Fig. 3 present the two reddest solitonic pulses, ejected from the soliton fission, accurately centered in the predefined channels ( λ c 1 2 , ) delimited by the dashed lines (see Table I for parameter values associated to results in Fig. 3). In Figs. 3(a) and 3(d), the target spectral channels where chosen from Ref. [29] in order to illustrate the solution for a
Fig. 3. (a-c) Spectrograms of the output spectra, z = 25 cm, corresponding to the optimization results given by the GA algorithm after m = 300 evaluations. S 1 2 , label the two solitonic pulses. (d-f) Spectral evolutions along the fiber associated to (a-c), respectively, retrieved from the optimal input pulse parameters corresponding to three different pairs of channels, λ c 1 2 , . Input pulse parameters are given in Table I. Vertical solid lines mark the zero GVD wavelength.
dual-pulse source required in a realistic application. The other two cases, Figs. 3(b,e) and Figs. 3(c,f), demonstrate the tunability of such source, keeping λ c 1 -λ c 2 fixed to 100 nm without replacing the PCF but merely adjusting the input pulse parameters. We checked by benchmarks that several runs of the GA with fixed λ c 1 2 , provided systematically very similar optimal results and therefore only one is shown here for each different case.
Regarding OCT applications, another important aspect of the source presented here is that the fs-SC dynamics typically exhibits a very high coherence and negligible shot-to-shot fluctuations [3], known to be detrimental for OCT [35]. Moreover, the two output solitonic pulses ( S 1 2 , in Fig. 3) constituting the proposed OCT light source, provide a decent resolution l c ≡ 2ln2 λ s / [ π∆λ s FWHM , ] [49] of ∼ 10 µ mfor the two solitons, S 1 2 , .
Table 1. Parameters associated to the best individuals found by the GA, shown in Fig. 3.
| Optimal pulse parameters | Optimal pulse parameters | Optimal pulse parameters | Optimal pulse parameters | Spectral bands | Resolution | Fitness | Shown in |
|----------------------------|----------------------------|----------------------------|----------------------------|--------------------|----------------------|-----------------|--------------|
| T 0 (fs) | λ 0 (nm) | P 0 (kW) | N | λ c 1 , λ c 2 (nm) | l c 1 , l c 2 ( µ m) | φ (10 - 4 / W ) | |
| 90.01 | 834.98 | 5.012 | 20.14 | 1075, 975 | 9.8, 9.5 | 1.138 | Figs. 3(a,d) |
| 70.80 | 817.27 | 12.3501 | 36.24 | 1150, 1050 | 10.6, 9.2 | 1.107 | Figs. 3(b,e) |
| 101.13 | 849.24 | 9.6617 | 26.43 | 1225, 1125 | 10.1, 10.1 | 1.018 | Figs. 3(c,f) |
Figure 4(a) shows the 3 D chart in the parameter space containing all 300 individuals involved in the optimization process. Data points distributed all over the volume are typically generated by the random stage 1 ( m < pth = 50) and GOs tend to accumulate solutions around small volumes where fitness is typically small, with the overall effect of monotonically decreasing the average fitness value, observed when fitness is represented in order of execution [black
Fig. 4. (a) Parameter space cloud of the 300 individuals (and fitness) generated by the GA in the optimization yielding to the solution in Figs. 3 (b),(e). The best individual is marked in red and dashed mark its input parameters. (b) Fitness evolution versus generated individuals in chronological order. Dashed vertical line marks the threshold population pth = 50 corresponding to the end of stage 1 (random generation). Best (at m ≈ 260), Instantaneous minimum, and average fitness are also plotted (see legend).
curve in Fig. 4(b)]. However, the scattering ability of GOs often results in finding slightly better individuals in nearby regions presenting smaller agglomeration. An important reason for the convergence of our GA towards the optimal solutions is the fact that the operator ˆ M is given a lower probability of action than ˆ X (probabilities are 1 / 3 and 0 95 respectively, see . previous section). This combination gives both a good diversity and probability to conserve the properties of the best individuals during the execution of the GA.
## 4. Conclusions
We presented an efficient optimization procedure based on GAs deployed in the Grid platform, providing faster results and potential scalability of the computational resources. The optimization provides the optimum input pulse parameters required to control the SC dynamics in a way that the first two ejected Raman solitons are centered at two pre-defined wavelengths. The results are shown to be of interest for practical OCT applications in the NIR II region where dual frequency, pulsed sources enable in vivo imaging, and avoid spurious results.
|
10.1364/OE.22.023686
|
[
"F. R. Arteaga-Sierra",
"C. Milián",
"I. Torres-Gómez",
"M. Torres-Cisneros",
"G. Moltó",
"A. Ferrando"
] |
2014-08-05T00:33:47+00:00
|
2014-08-05T00:33:47+00:00
|
[
"physics.optics"
] |
Supercontinuum optimization for dual-soliton based light sources using genetic algorithms in a Grid platform
|
We present a numerical strategy to design fiber based dual pulse light
sources exhibiting two predefined spectral peaks in the anomalous group
velocity dispersion regime. The frequency conversion is based on the soliton
fission and soliton self-frequency shift occurring during supercontinuum
generation. The optimization process is carried out by a genetic algorithm that
provides the optimum input pulse parameters: wavelength, temporal width and
peak power. This algorithm is implemented in a Grid platform in order to take
advantage of distributed computing. These results are useful for optical
coherence tomography applications where bell-shaped pulses located in the
second near-infrared window are needed.
|
1408.0842v1
|
## Two spatially separated phases in semiconducting Rb 0 8 . Fe 1 5 . S 2
Meng Wang, 1, ∗ Wei Tian, 2 P. Valdivia, 1 Songxue Chi, 2 E. Bourret-Courchesne, 3 Pengcheng Dai, 4, 5 and R. J. Birgeneau 1, 3, 6
1
Department of Physics, University of California, Berkeley, California 94720, USA 2 Quantum Condensed Matter Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831-6393, USA 3 Materials Science Division, Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA 4 Department of Physics and Astronomy, Rice University, Houston, Texas 77005, USA 5 Institute of Physics, Chinese Academy of Sciences, P. O. Box 603, Beijing 100190, China 6 Department of Materials Science and Engineering, University of California, Berkeley, California 94720, USA
We report neutron scattering and transport measurements on semiconducting Rb 0 8 . Fe 1 5 . S , 2 a compound isostructural and isoelectronic to the well-studied A 0 8 . Fe Se ( y 2 A = K, Rb, Cs, Tl/K) superconducting systems. Both resistivity and DC susceptibility measurements reveal a magnetic phase transition at T = 275 K. Neutron diffraction studies show that the 275 K transition originates from a phase with rhombic iron vacancy order which exhibits an in-plane stripe antiferromagnetic ordering below 275 K. In addition, interdigitated mesoscopically with the rhombic phase is an ubiquitous phase with √ 5 × √ 5 iron vacancy order. This phase has a magnetic transition at T N = 425 K and an iron vacancy order-disorder transition at T S = 600 K. These two different structural phases are closely similar to those observed in the isomorphous Se materials. Based on the close similarities of the in-plane antiferromagnetic structures, moments sizes, and ordering temperatures in semiconducting Rb 0 8 . Fe 1 5 . S 2 and K 0 81 . Fe 1 58 . Se , 2 we argue that the in-plane antiferromagnetic order arises from strong coupling between local moments. Superconductivity, previously observed in the A 0 8 . Fe Se y 2 -z S z system, is absent in Rb 0 8 . Fe 1 5 . S , which has a semiconducting ground state. 2 The implied relationship between stripe/block antiferromagnetism and superconductivity in these materials as well as a strategy for further investigation is discussed in this paper.
PACS numbers: 74.25.Ha, 74.70.-b, 78.70.Nx
## I. INTRODUCTION
The A 0 8 . Fe 1 6 . ± δ Se 2 ( A =K, Rb, Cs, Tl/K) materials, the so-called '245' systems were discovered at the end of 2010, and have since generated a great deal of interest, in large part because of their unique properties: iron vacancy order, block antiferromagnetism (AF) with large 3.3 µ B moments aligned along the c -axis, and the existence of superconductivity for appropriate chemical compositions 1-6 . In the Fe pnictide systems, the parent compounds of the superconductors exhibit a collinear antiferromagnetic structure with small ordered moments, typically less than 1 µ B 7-10 . Superconductivity arises upon electron or hole doping of the parent compounds, which concomitantly suppresses the AF order. Spin fluctuations associated with the AF order, which exist throughout the superconducting (SC) dome, are thought to play a crucial role in the mechanism of superconductivity 11-13 . In the standard interpretation, nesting between the hole and electron Fermi surfaces gives rise to spin density wave (SDW) order. In addition, the ubiquitous occurrence of a neutron 'spin-resonance' at the SDW wave vector in superconducting iron pnictide compounds has been suggested to correlate with ' s ± ' pairing symmetry 14-16 .
A spin resonance mode was also found in the '245' system, but at a wave vector different from those of both the block and stripe AF orders 17,18 . Importantly, unlike the Fe pnictides, a weak electron-like Fermi pocket and hole-like bands below the Fermi surface are found in place of hole Fermi surfaces around the Γ point in the A 0 8 . Fe 1 6 . Se 2 system 19-21 . The occurrence of superconductivity with T c up to 32 K in the '245' system in the absence of electron-hole nesting presents a significant challenge to current theories of these phenomena 22 .
There is extensive empirical evidence that the SC phase occurs mesoscopically separated from the block AF insulator 23-29 . The block AF phase exists throughout the two dimensional phase diagrams of A x Fe Se y 2 over wide variations in the alkali metal (0 77 . ≤ x ≤ 0 98) . and iron contents (1 48 . ≤ y ≤ 1 65), . with little change of T N 30 . We emphasize that ( x, y ) are for the sample as a whole, not the two separate constituent phases in most studies. The reports focused on the composition of the superconducting phase remain conflicting 27,31-33 . Thus, the nature of the real superconducting phase and its parent compound are still under debate 25-27,31-34 . Both theory and photoemission experiment proposed an insulating or semiconducting phase as a candidate for the parent compound of the superconducting phase in (K, Tl) x Fe Se y 2 25,35 . Importantly, the same stripe AF structure with in-plane ordered moments that occurs in the parent compounds of pnictide superconductors was observed in semiconducting K 0 81 . Fe 1 58 . Se 2 by neutron diffraction 33 . If the stripe phase with in-plane AF order is, in fact, the parent compound of the superconducting phase in the '245' system, then the SC in this system may have the same underlying mechanism as that in the other iron based superconductors, in spite of the absence
of electron-hole nesting and different neutron spin resonance wave vectors 14-21 . Therefore, determining the origin of the in-plane AF order in the semiconducting phase and its relationship with superconductivity is crucial to understanding the mechanism of superconductivity in the A x Fe Se y 2 system.
The low-temperature electrical resistivity of the '245' system can be changed from insulating to semiconducting or superconducting by controlling the iron content as in A 0 8 . Fe Se , generally in concert with the alkali conceny 2 tration A , or by substitution of sulfur on the selenium sites as in A 0 8 . Fe Se y 2 -z S z 33,36-40 . In studies to-date, changing the iron content of the pure Se two-phase material results in the sudden disappearance of the superconductivity, while sulfur substitution for selenium appears to suppress superconductivity gradually, resulting in a semiconducting ground state 39 . Accordingly, semiconducting A 0 8 . Fe S y 2 may also be viewed as the parent compound of the A 0 8 . Fe Se y 2 -z S z superconductors, although the magnetic phase diagram has not yet been determined for high sulfur substitution. Both high temperature transport and Raman scattering measurements indicate that the block AF phase also exists in the A x Fe S y 2 system 41,42 . Thus, it is important to investigate whether or not the in-plane AF order occurs in A 0 8 . Fe S y 2 and, if so, to determine its relationships with superconductivity in the S-substituted A 0 8 . Fe Se y 2 -z S . z
In this paper, we present transport and elastic neutron scattering measurements on single crystals of semiconducting Rb 0 8 . Fe 1 5 . S . 2 Two magnetic phases are found in this material with the next nearest (NN) Fe neighbor bond distances at 180 K 3.765 A and 3.889 ˚ for ˚ A the two phases respectively. The first phase, the '245' phase, which has the more compact in-plane lattice constants, has the √ 5 × √ 5 iron vacancy order and block AF order as in the A 0 8 . Fe 1 6 . ± δ Se 2 system 30 . The N´ eel temperature of the block AF order is 425 K; this is reduced significantly compared with ∼ 560 K in A 0 8 . Fe Se , and y 2 also well separated from the √ 5 × √ 5 iron vacancy ordering temperature of 600 K in Rb 0 8 . Fe 1 5 . S 2 5 . Schematics of the three-dimensional structure together with that of the iron plane with ordered moments and iron vacancies are shown in Fig. 1 (a, b). The second phase has rhombic iron vacancy order with in-plane stripe AF order below 275 K (Fig. 1 (c, d)) 43 . We named it as the '234' phase (this assumes an ideal stoichiometry RbFe 1 5 . S ) in spite of the 2 possible deviation of Rb in the discussion below. The estimated in-plane magnetic moment size of 2 8 . ± 0 5 . µ B and the N´ eel temperature of 275 K for the stripe AF order in semiconducting Rb 0 8 . Fe 1 5 . S 2 are surprisingly close to the 2.88 µ B moments and T N = 280 K of the stripe AF order in semiconducting K 0 81 . Fe 1 58 . Se 2 33 . These results suggest that strong coupling of local moments plays the dominant role in the formation of in-plane AF order in semiconducting A 0 8 . Fe y X 2 ( X =Se, S).
FIG. 1: (color online). (a) Three dimensional structure of the block AF phase in Rb 0 8 . Fe 1 5 . S . (b) The Fe plane of block 2 AF phase with √ 5 × √ 5 iron order, where a magnetic unit cell with lattice parameter a s = √ 5 × 3 765 ˚ has been marked as . A green. The tetragonal lattice cell used throughout this paper is shaded light orange. The red and blue arrows represent the out-of-plane spin directions up and down. The orange, lime, turquoise and light gray balls are Rb, Fe, S atoms and Fe vacancies, respectively. (c) A three-dimensional magnetic unit cell of the in-plane AF order and (d) Fe-plane with the rhombic iron vacancy order. A magnetic unit cell is shaded green. The diagonal Fe bonds are 3.889 ˚ at 180 K. A
## II. EXPERIMENTAL DETAILS
Our experiments were carried out on the HB-1A tripleaxis spectrometer and HB-2C wide-angle neutron diffractometer (WAND) at the High-Flux Isotope Reactor, Oak Ridge National laboratory. The triple-axis experiment employed two pyrolytic graphite (PG) filters before the sample to reduce λ/ 2 contamination and horizontal collimation 40 -40 -′ ′ S -40 -80 ′ ′ with a fixed incident beam energy of E i = 14 64 meV. A single piece of crystal weighing . 220 mg with a mosaic of 1.5 ◦ was loaded into a closed cycle refrigerator (CCR) which covers the temperature range from 30 K to 750 K. The sample was aligned in the [ H,H,L ] zone and the [ H, 3 H,L ] zone in tetragonal notation with lattice parameters a = b = 3 889 ˚ , . A c = 13 889 . ˚ for the '234' phase and A a = b = 3 765 ˚ , . A c = 13 889 ˚ . A for the '245' phase optimized at 180 K. The momentum transfer ( q x , q y , q z ) is defined as (2 πH/a, 2 πK/b, 2 πL/c ) in ˚ A -1 . ( H,K,L ) are the Miller indices in reciprocal lattice units (r.l.u). We labeled the wave vectors as Q = [ H,K,L ] r.l.u throughout this paper. By employing the degrees of freedom of the upper and lower goniometers of HB-1A, we were able to probe wave vectors in the [ H,K,L ] plane near [ H,H,L ]. A Ge(115) monochromator was used to produce a neutron beam with λ = 1 482 .
˚ in the experiment at WAND. The one-dimensional A 3 He detectors with 624 anodes can cover a wide range in reciprocal space by rotating the sample. A standard CCR was used to cover the temperature range between 4 K and 300 K.
The Rb 0 8 . Fe 1 5 . S 2 single crystals were grown by the Bridgman method with a one-step reacting. Stoichiometric amounts of high purity of a Rb ingot, Fe powder, and pieces of S were loaded in an alumina crucible in an argon gas filled glovebox; then the alumina crucible was sealed in a quartz tube under vacuum. The quartz tube was loaded into a box furnace and kept at 200 ◦ C for 24 hours; then warmed up to 500 ◦ C and held for 20 hours; heated slowly to 1050 ◦ C for melting 5 hours; cooled down to 750 ◦ C at a rate of 4 ◦ C/hour; and finally cooled to room temperature. We obtained single crystals with dimensions up to 5 × 5 × 4 mm . 3
## III. RESULTS
We characterized the transport properties of several Rb 0 8 . Fe 1 5 . S 2 single crystals with a Quantum Design Physical Property Measurement System (PPMS). The results were very consistent among the different samples measured and indicated consistent phases. The in-plane resistivity shown in Fig. 2 (a) on a logarithm scale represents clear semiconducting behavior. This semiconducting characteristic is quite similar to that of the potassium compound with equivalent composition, K 0 8 . Fe 1 5 . S 2 40 . These results reveal, as expected, that Rb 0 8 . Fe S y 2 and K 0 8 . Fe S y 2 have similar transport characteristics. The enlarged resistivity from 240 K to 300 K in the inset of Fig. 2 (a) implies a phase transition at 275 K. This transition temperature corresponds to the onset of the in-plane stripe AF order observed by neutron diffraction, which is discussed in more detail below. The kink at 275 K corresponding to the stripe AF transition can also be seen in the susceptibility measurement. The difference between the zero field cooled (ZFC) and field cooled (FC) susceptibilities in Fig. 2 (b) suggests the possibility of a spin-glass phase coincident with the long range AF order in Rb 0 8 . Fe 1 5 . S , similar to that which has been proposed 2 for K 0 88 . Fe 1 63 . S 2 44 .
The lattice constants can be optimized by carrying out longitudinal scans: scanning the angle of the incident beam and exit beam S 2 (2 θ ), and rotating the sample angle ( θ ) by half of the step. The two well separated peaks of the longitudinal scan at Q = (1 1 0) are strong , , evidence for two structural phases existing in this sample[Fig. 2 (c)]. As estimated from the integrated peak intensities at 180 K, the '245' phase with a = b = 3 765 . ˚ (peak centered at A -52.70) has ∼ 65% volume fraction, and the '234' phase with a = b = 3 889 ˚ (peak cen-. A tered at -50.90) has ∼ 35% volume fraction. We observed that the volume fractions of the two phases varied among our different samples 45 . The transition temperatures of each phase are consistent. The two phases have
FIG. 2: (color online). (a) In-plane resistivity measurement of semiconducting Rb 0 8 . Fe 1 5 . S . 2 (b) 1 Tesla Zero field cooled (ZFC) and field cooled (FC) dc magnetic susceptibility measurements with H ‖ ab -plane. A spin-glass like behavior appears below 55 K. A kink corresponding to the stripe AF order transition is observed in both resistivity and susceptibility measurements at T =275 K. (c) The well separated peaks of S 2 scans at nuclear reflection wave vector Q = (1 1 0) , , demonstrate two different sets of in-plane lattice constants at 180 K and 590 K. (d) θ -2 θ scans at Q = (0 0 2) show a sin-, , gle lattice constant c within the instrument resolution. The error bars are one standard deviation and solid lines are fits to Gaussian function throughout this paper.
the same lattice constant c = 13 889 ˚ based on the . A θ -2 θ scans at Q = (0 0 2) at 180 K as shown in Fig. 2 (d). , , The peaks shift slightly due to the change of lattice constants at 590 K, but the peaks at Q = (1 1 0) are still , , clearly distinguishable at 590 K 45 . This is in marked contrast with the behavior in phase-separated superconducting K 0 8 . Fe 1 6 . Se , where the non-magnetic phase with the 2 more compact in-plane lattice constant merges together with the block AF phase at temperatures above the iron vacancy order-disorder transition at 520 K 23,32 .
We first discuss the '245' phase, which has the block AF order and √ 5 × √ 5 iron vacancy order in Fig. 1 (a, b). The block AF order generates magnetic peaks at the wave vectors shown as the solid circles in the scattering planes of [ H,K,L = odd ] in the inset figure of Fig. 3 (b). The √ 5 × √ 5 iron vacancy order produces nuclear peaks at the positions of the blank squares in the [ H,K,L = even ] planes. The wave vectors connected by red and blue lines in the inset of Fig. 3 (b) originate from the left and right chiralities, respectively. The details of the diffraction have been described elsewhere 29 . By comparing the peak centers under the two sets of lattice constants, the set with a = b = 3 765 ˚ was de-. A termined to correspond to the block AF phase. Fig. 3 (a-c) represent scans at the magnetic wave vectors of Q = (0 2 0 4 1), . , . , Q = (0 2 0 4 3) and . , . , Q = (1 2 1 4 1) . , . , at 212 K. The dramatic decrease of the magnetic peak intensity at L = 3 compared with that at L = 1 is consistent with c -axis-aligned moments together with the
FIG. 3: (color online). Magnetic and nuclear reflection peaks associated with the block AF phase and √ 5 × √ 5 iron vacancy order. Scans through magnetic peaks at (a) Q = (0 2 0 4 1), . , . , (b) Q = (0 2 0 4 3) and (c) . , . , Q = (1 2 1 4 1) along [ . , . , H,H +0 2] . direction at T = 212 K. The red solid points in (a) through Q = (0 4 0 2 1) . , . , are from the other chirality. (d) Rocking curve scans through magnetic peak Q = (0 2 0 4 1) at se-. , . , lected temperatures of T = 300, 385, 415 and 430 K. (e) Scan through the nuclear peak at Q = (0 2 0 6 2) . , . , along [ H,H +0 4] direction associated with the . √ 5 × √ 5 iron vacancy order at T = 300 K. (f) θ -2 θ scans through nuclear peak at Q = (0 4 0 8 2) at . , . , T = 300, 585, 597 and 605 K. The inset in (b) is the expected magnetic Bragg peaks (solid circles, L = odd) and nuclear Bragg peaks (empty squares, L = even) in tetragonal reciprocal space. The red and blue lines indicate two different chiralities. The insets in (d) and (f) are color maps of temperature dependence of reflection peaks corresponding to (d) and (f).
Fe 2+ magnetic form factor. Fig. 3 (a) also shows a scan at the equivalent wave vector Q = (0 4 0 2 1) from . , . , the other chirality. The temperature dependence of the rocking curve scans demonstrates that the N´el temperae ture is approximately 425 K, which is significantly lower than that in the A 0 8 . Fe Se y 2 system . 5 The fingerprint reflection peaks of the √ 5 × √ 5 iron vacancy order at Q = (0 2 0 6 2) and . , . , Q = (0 4 0 8 2) were also investi-. , . , gated and are represented in Fig. 3 (e, f). The orderdisorder transition temperature of the iron vacancies occurs at 600 K. Here we have carried out θ -2 θ scans in order to track the temperature dependence of the iron vacancy order, while accounting for thermal expansion.
Figure 4 summarizes the Bragg peaks of the '234' phase associated with the in-plane stripe AF order and rhombic iron vacancy order. The magnetic peaks are accurately centered at the wave vectors Q = (0 5 0 5 . , . , L =
FIG. 4: (color online). Diffraction studies on the in-plane AF phase with rhombic iron vacancy order. Scans of magnetic peaks at wave vector of (a) Q = (0 5 0 5 1), . , . , (b) Q = (0 5 0 5 3), . , . , (d) Q = (0 5 0 5 5) . , . , along the [ H,H ] direction, and (c) Q = (0 5 0 5 . , . , 1) along L direction at selected temperatures. The inset of (a) is a color map of detailed temperature dependence. (e, f) show the nuclear peaks at the wave vectors of H = 0 25 0 5 0 75; . , . , . L = even and magnetic peaks at H = 0 5 . , L = odd from the in-plane AF phase in the [ H, 3 H,L ] plane at 6 K and 280 K, respectively. The nuclear peaks were not changed, but the magnetic peaks disappeared at 280 K. The peaks at H = 0 4 . , L = even originate from the √ 5 × √ 5 iron vacancy order in block AF phase. (g) θ -2 θ scans at peaks of the rhombic iron vacancy order at Q = (0 25 0 75 0) at 180 K, 500 K and 718 K. The evolution . , . , of the center of S 2 versus temperature indicates an in-plane lattice constant change across the in-plane magnetic structure transition in the inset of (g).
1 3 5) at 180 K with lattice constants , , a = b = 3 889 . ˚, A c = 13 889 ˚ . The magnetic peaks disappear com-. A pletely by 280 K. The rhombic iron vacancy order together with the stripe AF order will induce magnetic peaks at Q = (0 25 0 25 . , . , L = odd ), Q = (0 75 0 75 . , . , L = odd ); and nuclear peaks at Q = (0 25 0 75 . , . , L = even ), Q = (0 75 0 25 . , . , L = even ), Q = (0 5 0 5 . , . , L = even ) as demonstrated in the inset of Fig. 4 (d). We show reflection peaks in the [ H, 3 H,L ] plane in Fig. 4 (e) at 6 K and (f) at 280 K. The peaks centered at Q = (0 25 0 75 . , . , L ), Q = (0 5 0 5 . , . , L ) and Q = (0 75 2 25 . , . , L ) , L = 0 , -2 , -4 are consistent with the rhombic iron vacancy order. The magnetic peak at Q = (0 5 1 5 3) at 6 K in Fig. 4 (e) . , . , disappears at a temperature above T N = 275 K. The
FIG. 5: (color online). Temperature dependence of order parameters for the in-plane stripe AF, the rhombic iron vacancy, the block AF and the √ 5 × √ 5 iron vacancy orders. The height of peaks at Q = (0 5 0 5 1) fitted from [ . , . , H,H ] scans shown as the black squares represent the stripe AF transition at T N 1 = 275 K. The temperature dependence of peak height at Q = (0 2 0 4 1) (green circle) . , . , fitted from rocking curve scans shows the block AF transition at T N 2 = 425 K. The red diamonds obtained from θ -2 θ scans through Q = (0 4 0 8 2) . , . , indicate a first order-like transition of the √ 5 × √ 5 iron vacancies at T S = 600 K. The rhombic iron vacancy order parameter integrated from the θ -2 θ scans at Q = (0 25 0 75 0) was . , . , collected from another piece of single crystal with the same composition aligned in the [ H, 3 H,L ] zone.
peaks at Q = (0 4 1 2 . , . , L = 0 , -2 , -4) originate from the √ 5 × √ 5 iron vacancy order of the '245' phase. The temperature dependence of the θ -2 θ scans in Fig. 4 (g) shows the existence of the rhombic iron vacancy order at temperatures as high as 718 K; this is the reason that the two phases did not merge together at the temperature above the iron vacancy order-disorder transition at T S = 600 K of the '245' phase in Rb 0 8 . Fe 1 5 . S 2 23,32 . From the inset, one can see a clear anomaly in the temperature dependence of the in-plane lattice constant at the AF transition indicating strong coupling between the structure and the antiferromagnetism. Residual peaks with temperature-independent intensities were observed at the magnetic peak positions above T N in semiconducting K 0 81 . Fe 1 58 . Se 2 33 . However, we did not observe residual intensity at these positions above T N in our semiconducting Rb 0 8 . Fe 1 5 . S 2 single crystals. This significant difference in these two systems, which otherwise behave quite similarly, remains to be understood.
In order to determine the transition temperatures of the '234' and '245' phases in Rb 0 8 . Fe 1 5 . S , 2 we carefully measured the intensities of the fingerprint reflection peaks versus temperature; the results are shown in Fig. 5. The N´ eel temperature of the in-plane stripe AF order of T N 1 = 275 K in the '234' phase of semiconducting Rb 0 8 . Fe 1 5 . S 2 is very close to T N = 280 K of the stripe AF order in semiconducting K 0 81 . Fe 1 58 . Se 2 33 . The block AF order of the '245' phase has a N´el temperature at e T N 2 = 425 K and an iron vacancy ordering temperature of T S = 600 K. These have a much larger separation than
## IV. DISCUSSIONS AND CONCLUSIONS
The similarity of the N´el temperatures suggests the e crucial role of local moment superexchange interactions between the iron spins. Thus, we propose that strong correlation effects are essential to the formation of the stripe AF phase, in contrast with the spin-density-wave mechanism, which has been proposed as the origin of the magnetic order in the parent compounds of the iron pnictide superconductors 14,15 . The strong coupling scenario can also be reconciled with the absence of hole Fermi surfaces in A 0 8 . Fe Se y 2 19-21 . Similar to the iron pnictides, the spin resonance modes associated with superconductivity in iron chalcogenide (FeTe 1 -x Se x and A 0 8 . Fe Se ) y 2 systems are compatible with nesting between the holeelectron or electron-electron Fermi surfaces 18 . In contrast with the pnictides, the in-plane magnetic orders in the iron chalcogenides are not compatible with nesting. The iron chalcogenides also have much larger local moments than the pnictide systems. The moments in the former are strongly suggestive of a localized rather than itinerant model for the magnetism.
The data in Fig. 5 show a surprising feature which indicates that the two different structural phases are in communication with each other. Specifically, there is a small increase ( ∼ 10%) with increasing temperature in the intensity of the superlattice reflection associated with the rhombic vacancy order at the temperature at which the vacancies in the √ 5 × √ 5 phase become disordered. The intensity change suggests that the iron, partially occupied on the rhombic vacancy sites of the '234' phase below T S = 600 K, moves to the iron vacancy disordered '245' phase. The movement of iron vacancies between the two phases in Rb 0 8 . Fe 1 5 . S 2 suggests a possible way to understanding the complex relationship between the AF structures and the superconductivity in the A 0 8 . Fe Se y 2 system. In the compounds with net composition A 0 8 . Fe Se y 2 (1 5 . < y < 1 6), the material stabilized is a combination . of the semiconducting phase A x Fe 1 5 . ± δ Se 2 ('234' phase) with in-plane stripe AF order and rhombic iron vacancy order together with the insulating phase, A 0 8 . Fe 1 6 . ± δ Se 2 ('245' phase) with the block AF order and √ 5 × √ 5 iron vacancy order. By adding more iron, only the volume fraction of the two phases is changed, that is, one traverses a first order two phase coexistence region between the '234' and '245' phases. This explains naturally why the N´ eel temperature of the in-plane stripe AF order is so stable. The block AF phase with √ 5 × √ 5 iron vacancy order, A 0 8 . Fe 1 6 . ± δ Se , 2 ('245' phase) with δ = 0, represents an end point of the two-phase coexistence region. In this picture, by further increasing the iron content beyond y = 1 6, the material then separates into a new iron . rich superconducting, non-magnetic phase, and the block AF phase with √ 5 × √ 5 iron vacancy order. We speculate that the '245' phase is a stable stoichiometric phase
and that the √ 5 × √ 5 ordered iron vacancies cannot be readily occupied. This means that increasing the Fe content above 1.6 causes the formation of a new iron-rich phase which exhibits superconductivity. Concomitantly, the iron-rich SC phase is always accompanied by the '245' phase but the '245' phase is not the parent compound of the superconducting phase.
The results reported in this paper suggest a new strategy for probing the onset of superconductivity in the A x Fe Se y 2 type systems. In the pnictide systems, important insights have been gained by continuously tuning variables, such as the electron concentration by substitution (e.g., replacing Fe by Co or Ni) and thereby studying the evolution of the magnetism from the AF parent material to the superconducting material 46 . This is especially important at the onset of superconductivity where rich magnetic and superconducting behavior are observed. This approach does not seem to be possible in the A x Fe Se y 2 systems since the superconductivity seems to appear discontinuously. Yet it is clear from the results reported here that systematic variation of the S content in the A 0 8 . Fe Se y 2 -z S z system should enable one to study the continuous evolution from the 'parent' stripe AF sulphide to the superconducting mixed sulphide-selenide thus emulating studies in pnictide materials like those in BaFe As 2 2 -x P x 47 . The A 0 8 . Fe Se y 2 -z S z system may be closely analogous to the BaFe 2 As 2 -x P x system.
In summary, we have studied the magnetic and nuclear structures of semiconducting Rb 0 8 . Fe 1 5 . S 2 single crystals.
- ∗ Electronic address: [email protected]
- 1 J. G. Guo, S. F. Jin, G. Wang, S. C. Wang, K. X. Zhu, T. T. Zhou, M. He, and X. L. Chen, Phys. Rev. B 82 , 180520(R) (2010).
- 2 A. Krzton-Maziopa, Z. Shermadini, E. Pomjakushina, V. Pomjakushin, M. Bendele, A. Amato, R. Khasanov, H. Luetkens, K. Conder, J. Phys. Condens. Matter 23 , 052203 (2011).
- 3 M. H. Fang, H. D. Wang, C. H. Dong, Z. J. Li, C. M. Feng, J. Chen, H. Q. Yuan, Europhys. Lett. 93 , 47004 (2011).
- 4 A. F. Wang, J. J. Ying, Y. J. Yan, R. H. Liu, X. G. Luo, Z. Y. Li, X. F. Wang, M. Zhang, G. J. Ye, P. Cheng, Z. J. Xiang, X. H. Chen, Phys. Rev. B 83 , 060512 (2011).
- 5 W. Bao, Q. Huang, G. F. Chen, M. A. Green, D. M. Wang, J. B. He, X. Q. Wang, Y. Qiu, Chin. Phys. Lett. 28 , 086104 (2011).
- 6 F. Ye, S. Chi, W. Bao, X. F. Wang, J. J. Ying, X. H. Chen, H. D. Wang, C. H. Dong, and M. H. Fang, Phys. Rev. Lett. 107 , 137003 (2011).
- 7 C. de la Cruz, Q. Huang, J. W. Lynn, J. Li, W. Ratcliff II, J. L. Zarestky, H. A. Mook, G. F. Chen, J. L. Luo, N. L. Wang, P. C. Dai, Nature (London) 453 , 899 (2008).
- 8 Q. Huang, Y. Qiu, W. Bao, M. A. Green, J. W. Lynn, Y. C. Gasparovic, T. Wu, G. Wu, and X. H. Chen, Phys. Rev. Lett. 101 , 257003 (2008).
- 9 S. L. Li, C. de la Cruz, Q. Huang, G. F. Chen, T. L. Xia, J. L. Luo, N. L. Wang, and P. C. Dai, Phys. Rev. B 80 , 020504(R) (2009).
Similar to semiconducting K 0 81 . Fe 1 58 . Se , there is an in2 plane stripe AF phase with rhombic iron vacancy order, in addition to the block AF phase with √ 5 × √ 5 iron vacancy order. The robust 2 8 . ± 0 5 . µ B in-plane ordered moments and ∼ 280 K N´ eel temperature of the stripe AF phase in semiconducting Fe-Se and Fe-S based systems suggest that strong electronic correlations play a dominant role in determining the nature of the magnetic state. The relationships among the block AF phase, the superconducting phase, and the in-plane stripe AF phase have been discussed in this paper. The A 0 8 . Fe Se y 2 -z S z system opens a new window to study the relationship between the antiferromagnetism and the superconductivity.
## V. ACKNOWLEDGEMENTS
This work is supported by the Director, Office of Science, Office of Basic Energy Sciences, U.S. Department of Energy, under Contract No. DE-AC02-05CH11231 and the Office of Basic Energy Sciences U.S. DOE Grant No. DE-AC03-76SF008. The research at Oak Ridge National Laboratory's High-Flux Isotope Reactor is sponsored by the Scientific User Facilities Division, Office of Basic Energy Sciences, U.S. Department of Energy. We also acknowledge support from the US NSF DMR-1362219 and the Robert A. Welch Foundation Grant No. C-1839 at Rice University.
- 10 J. W. Lynn and Pengcheng Dai, Physica C 469 , 469-476 (2009).
- 11 D. C. Johnston, Adv. in Phys. 59 , 803 (2010).
- 12 P. C. Dai, J. P. Hu, and E. Dagotto, Nat. Phys. 8 , 709-718 (2012).
- 13 M. Wang et al. , Nat. Commun. 4 , 2874 (2013).
- 14 J. Dong et al. , Europhys. Lett. 83 , 27006 (2008).
- 15 I. I. Mazin, D. J. Singh, M. D. Johannes, and M. H. Du, Phys. Rev. Lett. 101 , 057003 (2008).
- 16 T. A. Maier and D. J. Scalapino, Phys. Rev. B 78 , 020514 (2008).
- 17 J. T. Park, G. Friemel, Yuan Li, J.-H. Kim, V. Tsurkan, J. Deisenhofer, H.-A. Krug von Nidda, A. Loidl, A. Ivanov, B. Keimer, and D. S. Inosov, Phys. Rev. Lett. 107 , 177005 (2011).
- 18 G. Friemel, J. T. Park, T. A. Maier, V. Tsurkan, Yuan Li, J. Deisenhofer, H. A. Krug von Nidda, A. Loidl, A. Ivanov, B. Keimer, and D. S. Inosov, Phys. Rev. B. 85 , 140511(R) (2012).
- 19 Y. Zhang, L. X. Yang, M. Xu, Z. R. Ye, F. Chen, C. He, J. Jiang, B. P. Xie, J. J. Ying, X. F. Wang, X. H. Chen, J. P. Hu, D. L. Feng, Nat. Mater. 10 , 273 (2011).
- 20 T. Qian, X.-P. Wang, W.-C. Jin, P. Zhang, P. Richard, G. Xu, X. Dai, Z. Fang, J.-G. Guo, X.-L. Chen, and H. Ding, Phys. Rev. Lett. 106 , 187001 (2011).
- 21 D. X. Mou et al. , Phys. Rev. Lett. 106 , 107001 (2011).
- 22 E. Dagotto, Rev. Mod. Phys. 85 , 849 (2013).
- 23 A. Ricci, N. Poccia, B. Joseph, G. Arrighetti, L. Barba, J.
- Plaisier, G. Campi, Y. Mizuguchi, H. Takeya, Y. Takano, N. L. Saini, A. Bianconi, Supercond. Sci. and Technol. 24 , 082002 (2011).
- 24 A. Ricci et al. , Phys. Rev. B 84 , 060511(R) (2011).
- 25 F. Chen et al. , Phys. Rev. X 1 , 021020 (2011).
- 26 W. Li, H. Ding, Z. Li, P. Deng, K. Chang, K. He, S. Ji, L. Wang, X. Ma, J. P. Hu, X. Chen, and Q. K. Xue, Phys. Rev. Lett. 109 , 057003 (2012).
- 27 W. Li, H. Ding, P. Deng, K. Chang, C Song, K. He, L. Wang, X. M, J. Hu, X. Chen, and Q. Xue, Nat. Phys. 8 , 126-130 (2012).
- 28 Z. Wang, Y. J. Song, H. L. Shi, Z.W. Wang, Z.Chen, H. F. Tian, G. F. Chen, J. G. Guo, H. X. Yang, J. Q. Li, Phys. Rev. B 83 , 140505(R) (2011).
- 29 M. Wang et al. , Phys. Rev. B 84 , 094504(R) (2011).
- 30 W. Bao, G. N. Li, Q. Huang, G. F. Chen, J. B. He, D. M. Wang, M. A. Green, Y. M. Qiu, J. L. Luo, M. M. Wu, Chin. Phys. Lett. 30 , 027402 (2013).
- 31 N. Lazarevi´, M. Abeykoon, P. W. Stephens, H. Lei, E. S. c Bozin, C. Petrovic, and Z. V. Popovi´, Phys. Rev. B c 86 , 054503 (2012).
- 32 S. V. Carr, D. Louca, J. Siewenie, Q. Huang, A. F. Wang, X. H. Chen, and P. C. Dai, Phys. Rev. B 89 , 134509 (2014).
- 33 J. Zhao, H. Cao, E. Bourret-Courchesne, D. H. Lee, and R. J. Birgeneau, Phys. Rev. Lett. 109 , 267003 (2012).
- 34 X. X. Ding, D. L. Fang, Z. Y. Wang, H. Yang, J. Z. Liu, Q. Deng, G. B. Ma, C. Meng, Y. H. Hu, and Hai-Hu Wen, Nat. Commun. 4 , 1897 (2013).
- 35 R. Yu, J. X. Zhu, and Q. Si, Phys. Rev. Lett. 106 , 186401 (2011).
- 36 D. M. Wang, J. B. He, T. L. Xia, and G. F. Chen, Phys.
- Rev. B. 83 , 132502 (2011).
- 37 J. G. Guo, X. L. Chen, G. Wang, S. F. Jin, T. T. Zhou, and X. F. Lai, Phys. Rev. B 85 , 054507 (2012).
- 38 K. F. Wang, H. C. Lei, and C. Petrovic, Phys. Rev. B. 84 , 054526 (2011).
- 39 H. C. Lei, M. Abeykoon, E. S. Bozin, K. F. Wang, J. B. Warren, and C. Petrovic, Phys. Rev. Lett. 107 , 137002 (2011).
- 40 Y. Cai, Z. Wang, Z. W. Wang, Z. A. Sun, H. X. Yang, H. F. Tian, C. Ma, B. Zhang, and J. Q. Li, Europhys. Lett. 103 , 37010 (2013).
- 41 J. J. Ying, Z. J. Xiang, Z. Y. Li, Y. J. Yan, M. Zhang, A. F. Wang, X. G. Luo, and X. H. Chen, Phys. Rev. B 85 , 054506 (2012).
- 42 N. Lazarevi´, H. Lei, C. Petrovic, and Z. V. Popovi´, Phys. c c Rev. B 84 , 214305 (2011).
- 43 H. Sabrowsky, M. Rosenberg, D. Welz, P. Deppe, Sch¨fer, a J. Magn. Magn. Mater. 54-57, 1497 (1986).
- 44 H. C. Lei, M. Abeykoon, E. S. Bozin, and C. Petrovic, Phys. Rev. B 83 , 180503(R) (2011).
- 45 The longitudinal scan at Q = (1 1 0) on another piece of , , single crystal indicated 45 ± 5% '245' phase and 55 ± 5% '234' phase. The (1 1 0) peaks from the two phases are , , separated as high as 718 K. Our neutron powder diffraction refinement on 2 grams sample ground by single crystals revealed that only 2% '245' phase and 98% '234' phase in the sample at 500 K.
- 46 D. C. Johnston, Adv. in Phys. 59 , 803-1061 (2010).
- 47 S. Kasahara, T. Shibauchi, K. Hashimoto et al. , Phys. Rev. B 81 , 184519 (2010).
|
10.1103/PhysRevB.90.125148
|
[
"Meng Wang",
"Wei Tian",
"P. Valdivia",
"Songxue Chi",
"E. Bourret-Courchesne",
"Pengcheng Dai",
"R. J. Birgeneau"
] |
2014-08-05T00:38:59+00:00
|
2014-08-05T00:38:59+00:00
|
[
"cond-mat.supr-con"
] |
Two spatially separated phases in semiconducting Rb$_{0.8}$Fe$_{1.5}$S$_2$
|
We report neutron scattering and transport measurements on semiconducting
Rb$_{0.8}$Fe$_{1.5}$S$_2$, a compound isostructural and isoelectronic to the
well-studied $A_{0.8}$Fe$_{y}$Se$_2 (A=$ K, Rb, Cs, Tl/K) superconducting
systems. Both resistivity and DC susceptibility measurements reveal a magnetic
phase transition at $T=275$ K. Neutron diffraction studies show that the 275 K
transition originates from a phase with rhombic iron vacancy order which
exhibits an in-plane stripe antiferromagnetic ordering below 275 K. In
addition, interdigitated mesoscopically with the rhombic phase is an ubiquitous
phase with $\sqrt{5}\times\sqrt{5}$ iron vacancy order. This phase has a
magnetic transition at $T_N=425$ K and an iron vacancy order-disorder
transition at $T_{S}=600$ K. These two different structural phases are closely
similar to those observed in the isomorphous Se materials. Based on the close
similarities of the in-plane antiferromagnetic structures, moments sizes, and
ordering temperatures in semiconducting Rb$_{0.8}$Fe$_{1.5}$S$_2$ and
K$_{0.81}$Fe$_{1.58}$Se$_2$, we argue that the in-plane antiferromagnetic order
arises from strong coupling between local moments. Superconductivity,
previously observed in the $A_{0.8}$Fe$_{y}$Se$_{2-z}$S$_z$ system, is absent
in Rb$_{0.8}$Fe$_{1.5}$S$_2$, which has a semiconducting ground state. The
implied relationship between stripe/block antiferromagnetism and
superconductivity in these materials as well as a strategy for further
investigation is discussed in this paper.
|
1408.0843v2
|
## Geometric stability of topological lattice phases
T. S. Jackson, 1, ∗ Gunnar Möller, 2 and Rahul Roy 1
1
Department of Physics and Astronomy, University of California at Los Angeles, 475 Portola Plaza, Los Angeles, California 90095, USA 2 TCM Group, Cavendish Laboratory, J. J. Thomson Avenue, Cambridge CB3 0HE, UK (Dated: November 17, 2015)
The fractional quantum Hall (FQH) effect illustrates the range of novel phenomena which can arise in a topologically ordered state in the presence of strong interactions. The possibility of realizing FQH-like phases in models with strong lattice effects has attracted intense interest as a more experimentally accessible venue for FQH phenomena which calls for more theoretical attention. Here we investigate the physical relevance of previously derived geometric conditions which quantify deviations from the Landau level physics of the FQHE. We conduct extensive numerical many-body simulations on several lattice models, obtaining new theoretical results in the process, and find remarkable correlation between these conditions and the many-body gap. These results indicate which physical factors are most relevant for the stability of FQH-like phases, a paradigm we refer to as the geometric stability hypothesis, and provide easily implementable guidelines for obtaining robust FQH-like phases in numerical or real-world experiments.
The fractional quantum Hall effect (FQHE) provides a spectacular manifestation of the breakdown of the spinstatistics relation in two dimensions: one obtains quantum number fractionalization 1-5 and, potentially, non-Abelian statistics 6 which can form the substrate for topologically robust quantum computing. 7 Progress has been hampered by the considerable experimental difficulties involved in realizing the FQHE in the usual setting of a semiconductor heterostructure, but a flurry of interest in the field was set off by the recent insight 8-10 that these exotic phases of matter may also arise in topologically nontrivial insulators with partially filled flat bands, or fractional Chern insulators 10 (FCIs). The attractiveness of FCIs stems from the fact that the bandgap ∆ may be set without the use of a large external magnetic field, the strength of which is one of the limiting factors in the semiconductor FQHE. There are currently a range of experimental proposals for realizing FCIs in cold atom systems, 12-16 transition metal oxides 17-19 and elsewhere; a successful experimental implementation in ultracold fermions was recently announced in ref. 20.
projection, but GMP found that the set of operators remains a closed algebra under commutation. Intuitively, one expects that the form of this algebra plays a crucial role in the stability of the FQHE phases by limiting the set of possible destabilizing interactions. In a generic FCI, however, the analogous set of band-projected density operators is not a closed algebra, nor do the projected densities span the space of single-particle operators: 25 there is no canonical mapping between a general lattice FCI and the continuum FQHE.
Moreover, FCIs raise theoretical questions independent of their experimental interest. The majority of theoretical work on the FQHE over its thirty-year history has focused on the influence of interactions on the Landau level Hamiltonian, which occupies a unique, highly symmetric point in the space of single-particle Hamiltonians. FCI phenomena constitute a non-trivial and poorly-understood generalization of the FQHE in which lattice effects are non-negligible; a generic FCI does not have the FQHE as a continuum limit, and examples of lattice effects without any continuum analog have already been noted. 21,22 A generalization of our theoretical understanding of the FQHE to cover the case of FCIs is hence both nontrivial and experimentally relevant.
One possible approach to the stability of FCIs is via the single-mode approximation used by Girvin, MacDonald and Platzman 23,24 (GMP), who made the ansatz that the most relevant excitations which destabilize an FQH ground state are neutral magnetoroton modes generated by the action of electron density operators projected to the lowest Landau level. These operators do not commute with each other, due to the
In previous work, one of us 9 derived sufficient conditions for the band-projected density operators in an FCI to satisfy a closed algebra isomorphic to that present in the FQHE, which justified and elaborated upon a heuristic criterion used in previous FCI literature. Quantities describing the geometry of the Chern band (its embedding in Hilbert space) enter this analysis in a natural way as coefficients of terms which must necessarily vanish in order to obtain a closed algebra; remarkably, only three conditions need to be placed on the band's geometry for the isomorphism to be present to all orders in a longwavelength expansion. Heuristically, one might expect that reproducing the density operator algebra would then suffice to reproduce the full physics of the FQHE, but this argument has not been fully tested in the literature.
In the present work, we report the results of extensive numerical simulations which demonstrate that quantitative measures based on the band-geometric conditions of ref. 9 are robustly correlated with the many-body gap in realizations of FQH-type phases in different FCI lattice models. In addition to numerical data, we obtain several theoretical results, such as a scaling relation between the gap of an FQH-like state and the number of bands in an FCI model, which is essential for comparing different models. We find that the Berry curvature was computed incorrectly in a number of prior references; in Supplementary Note 1, we discuss why this quantity is defined unambiguously. The remarkably high degree of correlation we find between band geometry and the many-body gap leads us to propose a geometric stability hypothesis: that the algebra of band-projected density operators governs FQH-type phenomena in FCIs, even when the isomorphism doesn't hold exactly, and that the single-particle conditions investigated here
are accurate qualitative estimators of the stability of an FQHlike state. This frames the theoretical problem of generalizing results on the FQHE to cover FCI physics by distilling the effects of the lattice into a small number of quantities measuring the relevant deviations of an FCI from lowest Landau level behavior. Our results are also of use in experimental design, as they provide a computationally inexpensive means to estimate which choices of FCI model parameters are most likely to yield a FQH-like state with the largest possible gap; a naive analysis of the scales involved has suggested this may be on the order of room temperature. 8 From the opposite point of view, our results also indicate which areas of parameter space should be searched to find possible FCI states which do not correspond to FQH universality classes.
## RESULTS
## Geometry of Chern bands
We begin by introducing the quantities studied below. A necessary ingredient in engineering a fractional Chern insulator is a flat, topologically non-trivial band, defined as follows. Let | R , b 〉 be a tight-binding orbital localized at position R + d b ; the Fourier transform of the b th basis orbital (where b ranges from 1 to N ) is
$$| { \mathbf k }, b \rangle = \frac { 1 } { \sqrt { N _ { c } } } \sum _ { \mathbf R } e ^ { i { \mathbf k } \cdot ( { \mathbf R } + { \mathbf d } _ { b } ) } | { \mathbf R }, b \rangle, \quad \ \ ( 1 ) \quad g _ { \mu \nu } ^ { \alpha } ($$
where k is a crystal momentum restricted to the first Brillouin zone (BZ) and N c is the number of unit cells in the system, which are indexed by lattice vectors R . Eigenstates of the tight-binding Hamiltonian are Bloch functions
$$| { \mathbf k }, \alpha \rangle = \sum _ { b = 1 } ^ { \mathcal { N } } u _ { b } ^ { \alpha } ( { \mathbf k } ) | { \mathbf k }, b \rangle, \text{ \quad \ \ } ( 2 )$$
where α indexes the bands. At a fixed k , the tight-binding Hamiltonian is an N × N matrix with entries
$$H _ { b c } ( { \mathbf k } ) = \sum _ { \alpha = 1 } ^ { \mathcal { N } } E _ { \alpha } ( { \mathbf k } ) u _ { b } ^ { \alpha * } ( { \mathbf k } ) u _ { c } ^ { \alpha } ( { \mathbf k } ) \quad \quad ( 3 )$$
and band energies E α ( k ) . In general, for N > 1 , neither H bc ( k ) nor u α b ( k ) will have the periodicity of the reciprocal lattice, and this property is not needed to define the Berry curvature via eq. (5) below. Imposing this periodicity by hand has led to demonstrably incorrect calculations in previous literature. We clarify this point with a discussion in Supplementary Note 1 and illustrate the consequences of incorrect computations in Supplementary Figs. 1, 2.
Nontrivial topological order in a band α is indicated by a non-vanishing value of the (first) Chern number
$$c _ { 1 } = \frac { A _ { B Z } } { 2 \pi } \langle B _ { \alpha } \rangle, \text{ \quad \ \ } ( 4 )$$
where A BZ is the area of the momentum-space Brillouin zone, 〈· · · 〉 denotes the average over the BZ, normalized so 〈 1 〉 = 1 , and the Berry curvature 13,27 of the band α is defined as
$$B _ { \alpha } ( { \mathbf k } ) = - i \sum _ { b = 1 } ^ { \mathcal { N } } \left ( \frac { \partial u _ { b } ^ { \alpha * } } { \partial k _ { x } } \frac { \partial u _ { b } ^ { \alpha } } { \partial k _ { y } } - \frac { \partial u _ { b } ^ { \alpha * } } { \partial k _ { y } } \frac { \partial u _ { b } ^ { \alpha } } { \partial k _ { x } } \right ). \quad ( 5 )$$
The leading-order condition (in terms of a long-wavelength expansion) for the existence of an isomorphism between the band-projected density operators and the GMP algebra found in ref. 9 is that the Berry curvature should be constant as a function of k . In the results we report here, we quantify fluctuations of Berry curvature over the BZ by their root-meansquare (RMS) value,
$$\sigma _ { B } \equiv \sqrt { \frac { A _ { B Z } ^ { 2 } } { 4 \pi ^ { 2 } } \langle B ^ { 2 } \rangle - c _ { 1 } ^ { 2 } }.$$
We normalize σ B in the same way as the Chern number, so that (6) is dimensionless as well as insensitive to the scales over which deviations from the mean curvature occur.
The higher-order conditions obtained ref. 9 involve the pullback of the Fubini-Study metric on Hilbert space, 29 which we refer to below as the quantum metric. In terms of Bloch functions, it is given by
$$( 1 ) \quad g _ { \mu \nu } ^ { \alpha } ( \mathbf k ) & = \frac { 1 } { 2 } \sum _ { b = 1 } ^ { \mathcal { N } } \left [ \left ( \frac { \partial u _ { b } ^ { \alpha * } } { \partial k _ { \mu } } \frac { \partial u _ { b } ^ { \alpha } } { \partial k _ { \nu } } + \frac { \partial u _ { b } ^ { \alpha * } } { \partial k _ { \nu } } \frac { \partial u _ { b } ^ { \alpha } } { \partial k _ { \mu } } \right ) \\ \text{illouin} \quad & \quad - \sum _ { c = 1 } ^ { \mathcal { N } } \left ( \frac { \partial u _ { b } ^ { \alpha * } } { \partial k _ { \mu } } u _ { b } ^ { \alpha * } u _ { c } ^ { \alpha * } \frac { \partial u _ { c } ^ { \alpha } } { \partial k _ { \nu } } + \frac { \partial u _ { b } ^ { \alpha * } } { \partial k _ { \nu } } u _ { b } ^ { \alpha } u _ { c } ^ { \alpha * } \frac { \partial u _ { c } ^ { \alpha } } { \partial k _ { \mu } } \right ) \right ]. \\ & \quad \text{--} \quad. \. \. \. \. \. \. \. \. \.$$
The next-to-leading order condition of ref. 9 is that the quantum metric also be constant over the BZ. We adopt
$$\sigma _ { g } \equiv \sqrt { \frac { 1 } { 2 } \sum _ { \mu, \nu } \langle g _ { \mu \nu } g _ { \nu \mu } \rangle - \langle g _ { \mu \nu } \rangle \langle g _ { \nu \mu } \rangle }$$
as the appropriate generalization of RMS fluctuation to tensor quantities. The final constraint on the band geometry is that
$$D ( { \mathbf k } ) \equiv \det g ^ { \alpha } ( { \mathbf k } ) - \frac { B _ { \alpha } ( { \mathbf k } ) ^ { 2 } } { 4 } = 0. \quad \quad ( 9 )$$
It was shown in ref. 9 that the left-hand side of (9) is always nonnegative; the condition that it vanishes is equivalent to the condition that g α and F α = ( 0 B α -B α 0 ) are the real and imaginary components of a Kähler metric h α = g α + iF α / 2 . This means that, unlike the first two conditions, the metric determinant inequality D ( k ) ≥ 0 measures deviations from lowest Landau level physics, specifically. Analogous conditions may be derived for higher Landau levels.
A stronger condition can be obtained by considering the trace of the quantum metric instead. It was additionally shown in ref. 9 that
$$T ( { \mathbf k } ) \equiv \mathbf t r \, g ^ { \alpha } ( { \mathbf k } ) - | B _ { \alpha } ( { \mathbf k } ) | \geq 0. \quad \quad ( 1 0 )$$
In Supplementary Note 2 we show that if this inequality is saturated, the quantum metric is isotropic and D ( k ) must vanish. Hence the condition T ( k ) = 0 is equivalent to requiring that the algebra of band-projected density operators be identical to the GMP algebra, while D ( k ) = 0 merely requires that they be isomorphic.
## Band geometry hypothesis
The purpose of the present work is to investigate the degree to which the above criteria are satisfied in several FCI models known to exhibit FQH-like phases. 1,30,31 In this section, we outline procedures common to all models studied.
The stability of an FCI phase is trivially influenced by the dispersion of the occupied band. Fortunately, the dispersion of a band is independent of its Berry curvature and quantum metric: the former only depends on the Hamiltonian's spectrum while the latter depend only on its eigenvectors. This allows us to eliminate any dispersion-related confounding effects by energetically flattening the bands of each lattice model, which is equivalent to smearing nearest-neighbor hopping terms over an exponentially-localized area. 9,10
Differences between Chern bands and Landau levels also enter in the form of the Hamiltonian's interaction term. Unlike the energetic considerations, this dependence is still poorly understood, so we have limited the scope of the present paper to on-site repulsive interactions only (which necessitates bosonic statistics), since this is the lattice interaction which most closely matches the isotropy present in the continuum. For each lattice model considered, we therefore investigate the bosonic Laughlin state at filling fraction ν = 1 2 / (stabilized by a two-body delta-function interaction) and the bosonic Moore-Read state at ν = 1 (stabilized by a three-body deltafunction interaction). These states have completely different topological orders; furthermore the Laughlin state is known to be more robust in general than the Moore-Read state, so examining both provides a useful probe of the sensitivity of band-geometric arguments.
The band geometry hypothesis predicts that the most important factor will be Berry curvature fluctuations. Low curvature fluctuations were heuristically identified as a desirable criterion in the earliest FCI literature, 8,10 which has been well established by subsequent work (see in particular refs. 33 and 34). In the present work we therefore focus on the sub-leading conditions, namely the influence the quantum metric has on the gap.
Fluctuations of the quantum metric are predicted to be the next most relevant quantity, but in the models examined this was found to have a high degree of linear correlation with the Berry curvature fluctuations (see Supplementary Fig. 3). There is no a priori reason this should be the case: other metric-derived quantities were found to be largely independent of Berry curvature. In addition, we found that the trace inequality (10) was far more correlated with the gap than the determinant inequality (9) for all models examined, despite corresponding to a stronger condition on the algebra of density operators. These findings go beyond the scheme laid out in ref. 9. Due to space constraints, we present data on the dependence of the gap on the determinant condition for low values of σ B in Supplementary Fig. 4 for the kagomé lattice model and Supplementary Fig. 5 for the ruby lattice model.
## Haldane model
The first Chern insulator model was introduced by Haldane, 35 who considered a tight-binding model on the honeycomb lattice with nearest- and next-nearest-neighbor hoppings (Fig. 1a) and a Peierls phase due to non-uniform threading of magnetic flux through each hexagon. The singleparticle Hamiltonian for the Haldane model is
$$\text{par} \text{mann} \text{on} \text{or} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{l} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{rl} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{rm} \text{rl} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{rl} \text{rl} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{rm} \text{r} \text{rl} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{rl} \text{r} \text{rl} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{l} \text{r} \text{rl} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{rm} \text{rl} \text{rl} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{rl} \text{rl} \text{rl} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{l} \text{rl} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{l} \text{rl} \text{rl} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{rm} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{sl} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{ r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{\text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \text{r} \\ \text{where} \text{the} \text{Pauli} \text{sigma} \text{matrices} \text{act} \text{on} \text{the} \text{band} \text{index} \text{. } k _ { i } \equiv$$
where the Pauli sigma matrices act on the band index, k i ≡ k · a i , a 3 = -a 1 -a 2 and i = 1 2 3 , , is interpreted cyclically mod 3. The lower band has a nonzero Chern number when | M/t 2 | ≤ 3 √ 3 sin | φ | . This model has been extensively studied numerically, and the addition of short-ranged repulsive interactions has been shown to yield both the bosonic 30,31 and fermionic 1 Laughlin states at appropriate filling fractions.
For purposes of comparison with refs. 1 and 33, we consider the model at t 1 = t 2 = 1 . The energy spectrum has band crossings for these parameters, meaning that the bands cannot be flattened by local operators and the model analyzed in those references is not adiabatically connected to (11). In practice, however, one is most interested in the M = 0 subspace; the Hamiltonian then depends only on the combination t 2 sin φ/t 1 , and an increase in t 2 /t 1 which removes the crossing may then be compensated by a shift in φ which leaves the Hamiltonian (11) unchanged up to a scale.
The momentum dependence of the Berry curvature and quantum metric is shown in Fig. 1b for parameters which minimize σ B and maximize the gaps for the Laughlin state of bosons and fermions; these values are listed in Supplementary Table 1. We see that the distribution of Berry curvature minimizing σ B interpolates between that which maximizes the gap for the bosonic and fermionic Laughlin states, and likewise the value of φ minimizing σ B lies between the values minimizing the gaps (Fig. 1e, f). The band geometry argument doesn't distinguish the statistics of the underlying particles, due to the fact that the projected density operators are bilinear in particle operators and bosonic in either case. The ( φ, M ) parameter space may be sampled exhaustively; manybody gaps for the bosonic and fermionic Laughlin states are shown over the full topologically non-trivial region of parameter space in Figs. 1c, 1d. The largest gaps and most uniform band geometry both occur for M = 0 .
Figure 1. Band geometry and gap data for the Haldane model. ( a ) Honeycomb lattice used to define the Haldane model. Basis vectors a 1 , a 2 are shown in red; basis elements are shown by differently colored/numbered sites. Hopping elements are shown with black and blue edges; arrowheads indicate the chirality convention for complex hoppings. ( b ) Band geometry over the reciprocal lattice unit cell spanned by b 1 , b 2 , for parameter values maximizing the gap and minimizing σ B . Axes of ellipses are proportional to the eigenvectors of the quantum metric g α ( k ) , and ellipse color is given by the relative deviation of Berry curvature B α ( k ) from its Brillouin zone-averaged value. ( c ) Gap ∆ as a function of ( φ, M ) for N = 8 bosons at ν = 1 2 / with an on-site repulsion. ( d ) Gap as a function of ( φ, M ) for N = 8 fermions at ν = 1 3 / with nearest-neighbor repulsion. Note that this plot differs from Fig. 8 of ref. 1 because we exclude (light gray) parameters failing to meet energetic and entanglement-based criteria for Laughlin-type order. ( e ) Berry curvature fluctuations σ B (left scale) and metric fluctuations σ g and average trace inequality 〈 T 〉 (right scale) as functions of φ at M = 0 . ( f ) Reproduction of the gap data from panels ( c ), ( d ) along M = 0 .
The other band-geometric criteria are highly correlated with the curvature fluctuation σ B and yield little new information for this model (Fig. 1e), beyond being close to the location of the maximum gaps (Fig. 1f). We prove in Supplementary Note 3 that the remaining band-geometric criterion, the determinant condition (9), is necessarily saturated for any two-band model, but the trace condition (10) remains nontrivial here.
saw that a sub-lattice chemical potential M always reduces the gap of an FCI phase, so we set M = 0 below. The remaining couplings in the new Hamiltonian may be parameterized by t 2 sin φ/t 1 and t 3 /t 1 , so the interesting region of parameter space is still two-dimensional.
## Augmented Haldane model
Although the Haldane model at fractional filling exhibits a robust Laughlin state, its Berry curvature remains highly nonuniform even in the best case (Fig. 1b). We would like to be able to compare this model with the kagomé and ruby lattice models, in which more uniform curvature may be achieved. In addition, since we're interested in the sub-leading effects the quantum metric has on the gap, we want to examine band configurations for which the metric is more independent of the Berry curvature than in Fig. 1e.
This may be accomplished by adding a third-nearestneighbor hopping term, with independent coupling t 3 , to the Haldane model Hamiltonian (11). In the previous section, we
The momentum dependence of the Berry curvature and quantum metric is shown in Fig. 2a for parameters which minimize σ B and maximize the gaps for the Laughlin and MooreRead states; these values are given in Supplementary Table 2. Comparison with Fig. 1b shows that curvature fluctuations have been reduced; furthermore, the minimum of σ B (shown in Fig. 2b) occurs at different parameter values than the minimum of the trace condition (shown in Fig. 2c). Gaps for the Laughlin and Moore-Read states are shown in Figs. 2d and 2e, respectively; the maximum gaps in both cases occur at lower values of t 2 than the minimum value of σ B , which one may attribute to the influence of 〈 T 〉 .
Figure 2. Band geometry and gap data for the augmented
Haldane model. ( a ) Band geometry over the reciprocal lattice unit cell spanned by b 1 , b 2 , for parameter values maximizing the gaps and minimizing σ B . Axes of ellipses are proportional to the eigenvectors of the quantum metric g α ( k ) . Ellipse color is given by the relative deviation of Berry curvature B α ( k ) from its Brillouin zoneaveraged value. ( b ) Berry curvature fluctuations σ B and ( c ) Brillouin zone-averaged trace inequality 〈 T 〉 as a function of couplings t 2 , t 3 , where we set t 1 = 1 and φ = π/ 2 in this and remaining panels without loss of generality. ( d ) Gap ∆ as a function of couplings for the bosonic Laughlin state of N = 8 bosons at ν = 1 2 / . ( e ) Gap as a function of couplings for the bosonic Moore-Read state of N = 10 bosons at ν = 1 .
## Kagomé lattice model
A Chern insulator defined on the kagomé lattice was introduced by Tang, Mei and Wen 8 (Fig. 3a). This model is attractive for our purposes since it has three bands, while remaining structurally similar to the Haldane model.
Defining a complex hopping matrix element for the relative embedding of the sublattices in the unit cell as
$$h _ { b c } ( { \mathbf k } ) \equiv e ^ { i { \mathbf k } \cdot ( { \mathbf d } _ { b } - { \mathbf d } _ { c } ) } \widehat { e } _ { b, c }, \quad \ ( 1 2 ) \\ \simeq \ i c t h a \min i t m o t i v \ u b o c a \, / h \, \cap \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ v b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b c \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b \ u b.$$
where ̂ e b,c is the unit matrix whose ( b, c ) th entry is equal to 1 , the momentum space Hamiltonian for the kagomé lattice model is
$$H _ { \mathbf K } ( \mathbf k ) = & - \sum _ { j = 1 } ^ { 3 } \left [ ( t _ { 1 } - i \lambda _ { 1 } ) ( 1 + e ^ { i \mathbf k \cdot \mathbf a _ { j } } ) h _ { j, j + 1 } ( \mathbf k ) \\ & + ( t _ { 2 } - i \lambda _ { 2 } ) ( e ^ { i \mathbf k \cdot \mathbf a _ { j } } + e ^ { i \mathbf k \cdot \mathbf a _ { j + 1 } } ) h _ { j, j + 2 } ( \mathbf k ) \right ] \\ & + \mathbf h. c.$$
where h.c. is an abbreviation for the hermitian conjugate, a 3 = -a 1 -a 2 , and j is interpreted cyclically mod 3. The relative offsets d b are as depicted by the numbered sites in Fig. 3a. The momentum dependence of the Berry curvature and quantum metric is shown in Fig. 3b for parameters minimizing σ B and maximizing the gaps for the Laughlin and Moore-Read states; these values are listed in Supplementary Table 3.
Figure 3. Definition of couplings and band geometry for the
kagomé lattice model. ( a ) Lattice and chirality conventions for hopping terms in the kagomé lattice model Hamiltonian. Basis vectors a 1 , a 2 are shown in red; basis elements are shown by differently colored/numbered sites. Hopping elements are shown with black and blue edges; arrowheads indicate the chirality convention for complex hoppings. ( b ) Band geometry over the reciprocal lattice unit cell spanned by b 1 , b 2 , for parameter values minimizing σ B and maximizing gaps, respectively. Axes of ellipses are proportional to the eigenvectors of the quantum metric g α ( k ) . Ellipse color is given by the relative deviation of Berry curvature B α ( k ) from its Brillouin zone-averaged value.
Figure 4. Gap vs. band geometry for the kagomé lattice model. ( a ). Gap and geometry data for the kagomé lattice model with NN-only hoppings, as a function of the only coupling ratio λ /t 1 1 . ( b ) The same data as a one-dimensional submanifold in band geometry space, which we parameterize in terms of Berry curvature fluctuations σ B and the averaged trace condition 〈 T 〉 . The upper set (larger gap ∆ ) of points are gaps for the N = 8 bosonic Laughlin state at ν = 1 2 / , while the lower are for the bosonic Moore-Read state at ν = 1 . ( c ) Gap for the kagomé lattice model with both nearest-neighbor and next-nearest-neighbor couplings, as a function of band-geometric parameters ( σ B , 〈 T 〉 ) for the Laughlin state of N = 8 bosons at ν = 1 2 / . ( d ) Gap as a function of ( σ B , 〈 T 〉 ) for the Moore-Read state of N = 10 bosons at ν = 1 , for the same coupling values.
Figure 5. Gap vs. trace condition for the kagomé lattice model subject to constraints on σ B . ( a ) Gaps for the bosonic Laughlin state of N = 8 bosons at ν = 1 2 / , as a function of the Brillouin zone average of the trace condition 〈 T 〉 . We only plot gaps for parameter values which have Berry curvature fluctuations σ B less than twice its minimum value. ( b ) Gap of the bosonic Laughlin state vs. 〈 T 〉 for parameter values randomly chosen on isosurfaces of constant σ B in the space of couplings. The parameter space sampling procedure used to obtain these sets of points is described in the Methods. ( c ), ( d ) The same, for the bosonic Moore-Read state of N = 10 bosons at ν = 1 . Note that the same sets of model parameters are used in each column.
Wefirst consider the model with nearest-neighbor hoppings only (Figs. 4a, b). Because band geometry is determined by single-particle quantities, the values of ( σ B , 〈 T 〉 ) for a given value of the effective nearest-neighbor (NN) coupling λ /t 1 1 are identical for the Laughlin and Moore-Read states. Despite having different topological orders, gaps for both states decline monotonically as one proceeds from the region of minimum σ B and 〈 T 〉 , and this trend holds over the entire phase (i.e., up to the closure of the gap.)
To establish that the above trends were not coincidental, we studied the entire c 1 = -1 phase containing the NN-only point t 1 = λ 1 ; t 2 = λ 2 = 0 for the bosonic Laughlin (Fig. 4c) and Moore-Read (Fig. 4d) states. The gap's sensitivity to band geometry is most apparent for the more fragile MooreRead state: the state is only stable in a small region, with the largest gaps (white points) attained at parameters with the lowest values of σ B and 〈 T 〉 found. The fact that the region of stability is an arc, rather than a vertical line, demonstrates that 〈 T 〉 describes independent, non-negligible factors influencing the stability of this state. These phenomena are less evident in the Laughlin state (Fig. 4c), which remains stable over a wide range of parameter values.
Because quantum metric-dependent quantities enter at a higher order than Berry curvature fluctuations in the band geometry analysis, they should have a subdominant effect on the gap. In Fig. 5 we control for the effects of large curvature fluctuations by restricting attention to parameter values giving near-minimal values of σ B . Including all such parameters yields a one-way relationship for the Laughlin and Moore-Read states (Figs. 5a, c), in the sense that large gaps are obtained only at low values of 〈 T 〉 , but small gaps can be obtained at any value of 〈 T 〉 . Further detail is evident if we take parameter values chosen to give the same value of σ B (see the Methods for a description of the sampling procedure used.) Results for sets of points chosen to have four different values of σ B are shown in Figs. 5b, d for the Laughlin and Moore-Read states. Removing the variation in σ B reveals a full-fledged negative correlation between 〈 T 〉 and the gaps: the trend is approximately linear for models having curvature fluctuations near the minimum and becomes less so as curvature fluctuations are allowed to increase.
## Ruby lattice model
Hu, Kargarian and Fiete 36 described a Chern insulator model on the ruby lattice (Fig. 6a). In the limit of total spin polarization, they showed that hopping parameters could be chosen such that the lowest band had c 1 = 1 and a bandgap to bandwidth ratio of ∼ 70 . The Hamiltonian for this model is
$$\text{even} \quad \text{bandwidth ratio of } \sim 7 0. \, The Hamiltonian for this model is \\ 1 / t _ { 1 } \\ \text{pite} \quad \text{-} \\ \text{de-} \\ \text{min-} \\ \text{has} \\ \text{ne} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \text{n-} \\ \end{cases}$$
where h b,c ( k ) is defined in eq. (12). Here a 3 = -a 1 -a 2 and the index on a is interpreted cyclically mod 3, but the indices on h b,c ( k ) are interpreted cyclically mod 6. The relative offsets d b are as depicted by the numbered sites in Fig. 6a. We considered the c 1 = 1 phase containing the flat-band point found in ref. 36, with t = 1 0 + 1 2 . . i , t 1 = -1 2 + 2 6 . . i , and t 4 = -1 2 . . Momentum dependence of the Berry curvature and quantum metric is shown in Fig. 6b for parameters which minimize σ B and maximize the gaps for the Laughlin and Moore-Read states; these values are listed in Supplementary Table 4. Remarkably, the complexity of this Hamiltonian works in our favor: one can find parameter values which greatly reduce the fluctuations in band geometry relative to the kagomé lattice model, which means that the ruby lattice model may be tuned to produce a much closer approximation to lowest Landau level physics.
As a consequence, trends identified in the kagomé lattice model are manifest here with a much higher degree of correlation. Figs. 6c, d show that the gaps of the Laughlin and Moore-Read states are strongly correlated with band geometry as measured by both σ B and 〈 T 〉 . In both cases the gap can be seen to decrease with increasing 〈 T 〉 , even for the same values of σ B ; in particular, the Moore-Read state is only stable in the lower right half of the plot area. Restricting our attention to parameters yielding small fluctuations in Berry curvature, in Fig. 7 we display data analogous to that presented for the kagomé lattice model in Fig. 5. Again, if we only impose an upper bound on σ B we see that it's not possible to have a large gap for large values of 〈 T 〉 (Figs. 7a, c), while upon restricting to specific values of σ B (Figs. 7b, d) we see that the negative correlation between 〈 T 〉 and the gaps is even tighter, becoming nearly linear for low σ B and gaining more scatter as σ B is increased.
Figure 6. Definition of couplings, band geometry and gaps for the ruby lattice model. ( a ) Lattice and chirality conventions for hopping terms in the ruby lattice model Hamiltonian. All nearest-neighbor (arrowed) bonds are the same length. Basis vectors a 1 , a 2 are shown in red; basis elements are shown by differently colored/numbered sites. Hopping elements are shown with black ( ), blue ( t t 1 ) and brown ( t 4 ) edges; arrowheads indicate the chirality convention for complex hoppings. ( b ) Band geometry over the reciprocal lattice unit cell spanned by b 1 , b 2 , for parameter values minimizing σ B and maximizing gaps, respectively. Axes of ellipses are proportional to the eigenvectors of the quantum metric g α ( k ) . Ellipse color is given by the relative deviation of Berry curvature B α ( k ) from its Brillouin zone-averaged value. ( c ) Gap ∆ as a function of the average Berry curvature fluctuation σ B and averaged trace condition 〈 T 〉 for the Laughlin state of N = 8 bosons at ν = 1 2 / . ( d ) Gap as a function of ( σ B , 〈 T 〉 ) for the Moore-Read state of N = 10 bosons at ν = 1 .
Figure 7. Gap vs. trace condition for the ruby lattice model subject to constraints on σ B . ( a ) Gaps for the bosonic Laughlin state of N = 8 bosons at ν = 1 2 / , as a function of the Brillouin zone average of the trace condition 〈 T 〉 . Weonly plot gaps for parameter values which have Berry curvature fluctuations σ B less than twice its minimum value. ( b ) Gap of the bosonic Laughlin state vs. 〈 T 〉 for parameter values randomly chosen on isosurfaces of constant σ B in the space of couplings. ( c ), ( d ) The same, for the bosonic Moore-Read state of N = 10 bosons at ν = 1 . Note that the same sets of model parameters are used in each column.
## Significance of correlations
In this section we describe two approaches to quantifying the degree of correlation between the band geometry and the many-body gaps described above. In particular, the fact that FQH-type states are destabilized by fluctuations in Berry curvature is readily apparent and was anticipated in the first work on FCIs; in the present work we are interested in possible additional dependence on conditions derived from the quantum metric, so we seek to measure correlation between the gap and the trace condition 〈 T 〉 . This is not fully straightforward, due to correlation of 〈 T 〉 and σ B with each other, as evidenced by the fact that parameter values are not uniformly distributed in Figs. 4c and 6c.
One approach is to compare data for parameters yielding the same value of σ B , some of which is shown in Figs. 5b, d and 7b, d (see the Methods for the procedure used to sample from isosurfaces of constant σ B ). We do not have quantitative predictions for the functional dependence of the gap on any band-geometric quantity, so to avoid introducing assumptions we use Spearman's ρ as a nonparametric measure of correlation. This is defined as the linear (Pearson) correlation coefficient between the rankings of the data points when rank-ordered by ∆ and by 〈 T 〉 , and takes values ranging from +1 for any monotonically increasing function to -1 for any monotonically decreasing function.
Figure 8. Correlation between gaps and the trace condition
when the magnitude of Berry curvature fluctuations is held constant. Each data point shows (the negative of) Spearman's ρ for the correlation between gap and averaged trace condition 〈 T 〉 on separate samples of n = 200 points sampled from isosurfaces of constant root-mean-square Berry curvature fluctuation σ B . A value of ρ = -1 corresponds to perfectly monotonic anti-correlation. We show results for Laughlin (dark blue) and Moore-Read (light blue) states in ( a ) the kagomé lattice model and ( b ) the ruby lattice model. Error bars show the bootstrapped standard error.
In Fig. 8 we plot the results of this test for the Laughlin and Moore-Read data in the kagomé and ruby lattice models. This score was found to be within ∼ 10% of the linear correlation between 〈 T 〉 and ∆ for each isosurface, which indicates the robustness of our conclusion and implies that all values of σ B considered lie in a weak-fluctuation regime. The general trend evident in Fig. 8 is that ∆ is highly negatively correlated with 〈 T 〉 when curvature fluctuations are constrained to be near their minimum value. As curvature fluctuations are allowed to increase, this relationship becomes less exact, but converges to an asymptotic value well above 0 for both the Laughlin and Moore-Read states. This confirms the qualitative picture evident in Figs. 5b, d and 7b, d.
Alternatively, we can analyze the larger set of data having unrestricted values of σ B , at the cost of assuming linear relationships between all variables; the similarity between the results for Spearman's and Pearson's ρ mentioned above suggest that this is justified. One can then compute the partial correlation, denoted here by ρ B , as the degree of linear correlation between the residuals of ∆ and 〈 T 〉 , after first subtracting the best-fit linear dependence of each on σ B ; similarly, this score ranges from +1 to -1 . For the augmented Haldane, kagomé and ruby lattice models, respectively, we find ρ B = -0 91 . , -0 42 . and -0 62 . for the Laughlin state and ρ B = -0 77 . , -0 44 . and -0 46 . for the Moore-Read state. Sample sizes were, respectively, n = 1900 2100 , and 1800 . The fact that these values are lower than those obtained by the isosurface method describes the non-negligible correlation between σ B and 〈 T 〉 .
All correlation scores quoted above and shown in Fig. 8 are statistically significant at the 1% level (at most). This is the main result of this section: gaps for both states are independently sensitive to variations in the trace condition, beyond the correlations induced in the latter via curvature fluctuations. This confirms that band geometry plays a significant role in realistic FCI models.
## Cross-model comparisons
The band geometry hypothesis claims that the most stable FQH-like phases are obtained when the band geometry is tuned to be as close to that of a Landau level as possible. We have shown above that this holds for several different lattice models as their couplings are varied, but comparison of the models shows an apparent contradiction: the gaps reported above are smallest for the ruby lattice model, despite the fact that this model can be made to approximate Landau level physics more closely than the other models studied here.
The resolution of this apparent contradiction lies in the fact that the models considered have different numbers of tightbinding sites per unit cell. This factor enters into the interaction term of the Hamiltonian, and hence the gap: because we have flattened the dispersion of the kinetic term, the interaction strength is the only energy scale in the problem. In Supplementary Note 4 we give a scaling argument that the strength of a delta-function interaction in the continuum should be multiplied by a factor of N when discretized to an on-site repulsion in a model with N bands. This means that, when comparing the gaps of bosonic Laughlin states in different models, the appropriate quantity to compare is N ∆ . For the three-body delta-function interaction which stabilizes the bosonic Moore-Read state, the same considerations yield a scaling factor of N 2 .
In Fig. 9 we compare the gaps for the Laughlin state for the augmented Haldane, kagomé and ruby lattice models when scaled by this factor, as a function of the band-geometric pa-
rameters ( σ B , 〈 T 〉 ) . The domain of each plot is restricted to be the common overlap of the three models in ( σ B , 〈 T 〉 ) space. Because of this restriction, we were unable to perform a meaningful analysis for the Moore-Read state, as it is unstable in most of this region (compare Figs. 4d, 6d). Despite the fact that the scaling argument is exact only in the largeN limit and the models we compare have N = 2 3 , and 6 , we find roughly similar behavior across all three models, both in terms of the magnitude of the scaled gap and of its dependence on σ B and 〈 T 〉 .
Figure 9. Cross-model comparison of scaled gaps as a function of band geometry. Panels show interpolated energy gaps of the bosonic Laughlin state, scaled by the number of bands of each model, for the ( a ) augmented Haldane, ( b ) kagomé lattice and ( c ) ruby lattice models described above. Scaled gaps are plotted as a function of band geometry as measured by Berry curvature fluctuations σ B and the average value of the trace condition 〈 T 〉 . Data are only plotted in the region of common overlap of the three models in ( σ B , 〈 T 〉 ) space (polygonal outline). The color scale and contours used are identical for all three panels.
## DISCUSSION
In this work we have presented numerical results which systematically map out the robustness of topologically ordered FQH-like phases in FCI systems with realistic, short-ranged Hamiltonians and band geometry which is less than perfectly uniform. We presented quantitative evidence that the bandgeometric quantities identified in ref. 9 remain strongly correlated with the size of the gap even when its conditions on band geometry are not met exactly. This leads us to propose a geometric stability hypothesis for FQH-like phenomena in FCIs: in spite of the fact that the GMP algebra is not perfectly reproduced, we conjecture that an approximate version of the single-mode approximation correctly describes the lowenergy physics of these FCI models. As a practical corollary, we predict that single-particle Hamiltonians with more uniform band geometry - specifically, as measured by the hierarchy of three criteria - will produce more stable FQH-like states.
posed by Qi, 42 but the image of FQH pseudopotential interactions under this mapping is not well localized and strongly anisotropic, 8,42-44,46 making the relationship to physical FCI Hamiltonians unclear.
The validity of the single-mode approximation in FCIs has been investigated by a number of other authors using approaches complementary to that taken here. 9,25,33,37 Within the context of their Hamiltonian approach to the FQHE, Murthy and Shankar showed that composite fermion degrees of freedom could be chosen which reproduce an exact version of the GMP algebra. 38,39 More directly, in ref. 40, the lowest-lying neutral excitation of the kagomé and ruby lattice models was found to be well described by the magnetoroton mode of the corresponding FQH state on a torus, using a phenomenological mapping between the FQH and FCI Hilbert spaces described in ref. 41. A related mapping was originally pro-
Comparing the results from the kagomé and ruby lattice models, it appears easier to engineer uniform geometry in more complicated Hamiltonians, both in the sense of having more tunable couplings and in the sense of having more bands. The latter property is expected to hold on general grounds, as noted in refs. 13 and 47. Increasing the size of the unit cell reduces the effectiveness of a fixed-strength repulsion, however, so an optimal choice would balance these two factors. This has immediate relevance to experimental design: laboratory Hamiltonians are necessarily more complicated than those in idealized theoretical models (e.g., the proposal in ref. 16 involves an eight-dimensional parameterization of the applied electric field used to obtain a synthetic gauge potential.) Performing many-body simulations on a representative set of parameters in such a large space is prohibitively timeconsuming; the geometric stability hypothesis can be used to reduce this to a manageable subspace. In addition, band geometry may, by definition, be tuned independently of energetic considerations such as the bandwidth.
A pressing direction for future work is to further develop the band geometry hypothesis by investigating its validity in less straightforward scenarios: stable FCI states where the Berry curvature is not particularly uniform have also been proposed; furthermore, the stability of the state may also depend on the filling fraction and the particular state sought to be stabilized. 48,49 Among other aspects which would be interesting to clarify are the role of bosonic versus fermionic
particle statistics, NN- and longer-ranged inter-particle interactions, anisotropic interactions induced by the lattice structure, etc. In particular, the distribution of geometric quantities differs from band to band; this can be selected by fully filling a number of bands in a fermionic system, which could lead to new phenomena. One could also consider the wide range of more elaborate FCI models in the literature, possessing, e.g., Chern numbers | c 1 | > 1 , non-Abelian Berry curvature arising from multiple degenerate bands, etc. We direct the reader to the recent reviews 50,51 for a more extensive discussion and bibliography.
## METHODS
## Parameter space sampling
We study the dependence of the many-body gap on bandgeometric quantities in the Haldane model 35 and models proposed for the kagomé 8 and ruby 36 lattices. The parameter space for the Haldane model Hamiltonian is small enough that the vector of parameters t ∆ which maximize the gap ∆ may be found from exhaustive sampling. The parameter spaces for the other two models are higher-dimensional and non-compact, so this strategy will not work.
According to the geometric stability hypothesis, the Berry curvature fluctuations σ B should be inversely correlated with the gap, meaning that it may be employed as a proxy for the latter which requires far less effort to compute. In Supplementary Note 5 we derive an expression for the derivative of σ B with respect to any parameter appearing in the singleparticle Hamiltonian, allowing steepest-descent methods to be employed. Of course, we do not expect the vector of parameters t 0 minimizing σ B to precisely coincide with those maximizing the gap, but if the single mode approximation is applicable, t 0 will be a viable initial guess for t ∆ , to be refined as described below. Indeed, a state with a robust topological gap in the presence of large curvature fluctuations would be of immediate interest as an FCI phase not describable in terms of an FQH universality class.
In the neighborhood of t 0 , a surface of constant σ B in parameter space will be approximated by an ellipsoid given by the Hessian of σ B ( t ) at t 0 , which may be calculated by the method in Supplementary Note 5. We sample points uniformly from the surface of the ellipsoid defined by the Hessian by the well-known method of projecting vectors of parameters t sampled from the corresponding multinormal distribution. Wethen perturb this ellipsoid by shifting the vectors t radially along the rays connecting each with t 0 until we find parameters t ′ such that σ B ( t ′ ) is equal to the target value. These isosurfaces of different parameters with the same value of σ B permit us to study the sub-leading effects of the quantum metric on the gap predicted by the geometric stability hypothesis: in Fig. 10 we depict the isosurfaces found for the kagomé lattice model which were used to generate the data shown in Figs. 5b, d.
Finally, we locate the parameters giving the maximum gap by fitting a quadratic form
$$\Delta ( \mathfrak { t } ) = \Delta ^ { * } - \frac { 1 } { 2 } \sum _ { i, j } \Sigma _ { i j } ^ { - 1 } ( t _ { i } - t _ { i } ^ { * } ) ( t _ { j } - t _ { j } ^ { * } ) \quad ( 1 5 )$$
to the gaps from the isosurface data and sampling new parameters from a Gaussian distribution centered on t ∗ with covariance matrix Σ . We have verified that deviations of the actual t ∆ found from this data from the fitted value t ∗ are negligible compared to the scales set by Σ , hence this search does not need to be iterated further.
Figure 10. Isosurfaces of constant Berry curvature fluctuations in the parameter space of the kagomé lattice model. Parameter values which minimize the root-mean-square Berry curvature fluctuation σ B of the kagomé lattice model are marked by the large, central black dot. The large red dot identifies parameter values which yielded the maximum gap for the Laughlin state of N = 8 bosons at ν = 1 2 / . Concentric shaded surfaces are isosurfaces upon which σ B takes a constant value equal to 1 05 1 1 . , . , 1 25 . and 1 5 . times its minimum value (respectively). Blue contours on the walls of the box are sections through these surfaces at the minimum σ B point. Small grey points identify random samples taken from the σ B = 1 25 min . σ B isosurface, to illustrate uniformity.
## Numerical exact diagonalization
Unless explicitly identified otherwise, all data were obtained from exact diagonalization of the many-body Hamiltonian for N = 8 bosons interacting with a two-body on-site repulsion at a filling fraction (ratio of the number of particles to the number of unit cells) of ν = 1 2 / on a periodic lattice of 4 × 4 unit cells (for the Laughlin-like FCI state) or N = 10 bosons at a filling fraction of ν = 1 on a lattice of 5 × 2 unit cells, interacting with a three-body on-site repulsion only (for the Moore-Read-like FCI state). In the main text
we also presented additional data obtained for the fermionic Laughlin-like state in the Haldane model with N = 8 particles at ν = 1 3 / on a 6 × 4 lattice with two-body nearestneighbor repulsion. The fact that the bosonic Moore-Read state is realized in a filled Landau level means that, relative to the Laughlin state, we are able to simulate more particles with a many-body Hilbert space of roughly the same size. Flattening the spectrum of the single-particle Hamiltonian removes the only other energy scale from the problem, so the strength of the repulsive interaction sets the units of the many-body gap ∆ .
Because we do simulations for thousands of parameter values, a detailed finite-size scaling analysis of each is beyond the scope of this paper. We have, however, made spot checks by doing simulations of N = 6 to N = 12 particles at select parameter values. Representative data is shown in Supplementary Figure 6. From this analysis, we have concluded that the system sizes described above strike an acceptable balance between accurately approximating the thermodynamic ( N → ∞ ) limit and making the volume of simulations manageable on the computational resources available to us.
We employ several criteria to identify topological order in the results of numerical simulations, employing information from both the many-body energy spectrum and the many-body wavefunctions themselves. We first require that the spectrum has the correct number of degenerate ground states in the proper momentum sectors (two states at ( k , k x y ) = (0 0) , for bosonic Laughlin-type order, and three states at ( k , k x y ) =
## ∗ [email protected]
- 1 Laughlin, R. B. Anomalous quantum Hall effect: an incompressible quantum fluid with fractionally charged excitations. Phys. Rev. Lett. 50 , 1395-1398 (1983).
- 2 de Picciotto, R. et al. Direct observation of a fractional charge. Nature 389 , 162-164 (1997).
- 3 Saminadayar, L., Glattli, D., Jin, Y. & Etienne, B. Observation of the e/ 3 fractionally charged Laughlin quasiparticle. Phys. Rev. Lett. 79 , 2526-2529 (1997).
- 4 Camino, F., Zhou, W. & Goldman, V. Realization of a Laughlin quasiparticle interferometer: Observation of fractional statistics. Phys. Rev. B 72 , 075342 (2005).
- 5 Camino, F., Zhou, W. & Goldman, V. e/ 3 Laughlin quasiparticle primary-filling ν = 1 3 / interferometer. Phys. Rev. Lett. 98 , 076805 (2007).
- 6 Moore, G. W. & Read, N. Nonabelions in the fractional quantum Hall effect. Nucl. Phys. B 360 , 362-396 (1991).
- 7 Nayak, C., Stern, A., Freedman, M. H. & Das Sarma, S. NonAbelian anyons and topological quantum computation. Rev. Mod. Phys. 80 , 1083-1159 (2008).
- 8 Tang, E., Mei, J.-W. & Wen, X.-G. High-temperature fractional quantum Hall states. Phys. Rev. Lett. 106 , 236802 (2011).
- 9 Neupert, T., Santos, L., Chamon, C. & Mudry, C. Fractional quantum Hall states at zero magnetic field. Phys. Rev. Lett. 106 , 236804 (2011).
- 10 Sun, K., Gu, Z., Katsura, H. & Das Sarma, S. Nearly flatbands with nontrivial topology. Phys. Rev. Lett. 106 , 236803 (2011).
- 11 Regnault, N. & Bernevig, B. A. Fractional Chern insulator. Phys. Rev. X 1 , 021014 (2011).
(0 , 0) (0 , , 0) and (0 1) , for bosonic Moore-Read-type order). In a finite-size simulation these ground states will only be approximately degenerate; we require that the gap to the lowestenergy excited state (out of all momentum sectors; i.e. the indirect gap) be at least as large as the spread in ground state energies.
We ascertain whether a ground state wavefunction has the topological order of the Laughlin or Moore-Read state through properties of its entanglement spectrum: 52 we first verify that the spectrum of the reduced density matrix obtained by tracing out four bosons is gapped. There is no current quantitative theoretical interpretation of the magnitude of the entanglement gap; instead, the discriminative power of this criterion comes from requiring that the number of eigenvalues below the gap in each momentum sector obeys counting rules dictated by the topological order of continuum FQH state. 10,41 For the lattice geometry used in our studies, Laughlin-type order in a system of N = 8 bosons at ν = 1 2 / is identified by the (1 2) , counting rule (in the terminology of the previous references), which requires 48 states below the gap at ( k , k x y ) = (0 0) , , 44 states below the gap in sectors ( k , k x y ) = (0 2) (2 0) , , , , and (2 2) , , and 40 states below the gap in all other momentum sectors. Similarly, the counting rule for Moore-Read-type order in N = 10 bosons at ν = 1 requires 76 states below the gap at ( k , k x y ) = (0 0) , and (0 , 1) , and 75 states below the gap in all other momentum sectors. Simulations that fail any of the above tests are assigned an energy gap of zero.
- 12 Hafezi, M., Sørensen, A., Demler, E. & Lukin, M. Fractional quantum Hall effect in optical lattices. Phys. Rev. A 76 , 023613 (2007).
- 13 Cooper, N. R. & Moessner, R. Designing topological bands in reciprocal space. Phys. Rev. Lett. 109 , 215302 (2012).
- 14 Cooper, N. R. & Dalibard, J. Reaching fractional quantum Hall states with optical flux lattices. Phys. Rev. Lett. 110 , 185301 (2013).
- 15 Yao, N. Y. et al. Topological flat bands from dipolar spin systems. Phys. Rev. Lett. 109 , 266804 (2012).
- 16 Yao, N. Y. et al. Realizing fractional Chern insulators in dipolar spin systems. Phys. Rev. Lett. 110 , 185302 (2013).
- 17 Xiao, D., Zhu, W., Ran, Y., Nagaosa, N. & Okamoto, S. Interface engineering of quantum Hall effects in digital transition metal oxide heterostructures. Nature Communications 2 , 596 (2011).
- 18 Venderbos, J. W. F., Kourtis, S., van den Brink, J. & Daghofer, M. Fractional quantum-Hall liquid spontaneously generated by strongly correlated t 2 g electrons. Phys. Rev. Lett. 108 , 126405 (2012).
- 19 Kourtis, S., Venderbos, J. W. F. & Daghofer, M. Fractional Chern insulator on a triangular lattice of strongly correlated t 2 g electrons. Phys. Rev. B 86 , 235118 (2012).
- 20 Jotzu, G. et al. Experimental realization of the topological Haldane model with ultracold fermions. Nature 515 , 237-240 (2014).
- 21 Kol, A. & Read, N. Fractional quantum Hall effect in a periodic potential. Phys. Rev. B 48 , 8890-8898 (1993).
- 22 Möller, G. & Cooper, N. Composite fermion theory for bosonic quantum Hall states on lattices. Phys. Rev. Lett. 103 , 105303 (2009).
- 23 Girvin, S., MacDonald, A. H. & Platzman, P. Collectiveexcitation gap in the fractional quantum Hall effect. Phys. Rev. Lett. 54 , 581-583 (1985).
- 24 Girvin, S., MacDonald, A. H. & Platzman, P. Magneto-roton theory of collective excitations in the fractional quantum Hall effect. Phys. Rev. B 33 , 2481-2494 (1986).
- 25 Parameswaran, S. A., Roy, R. & Sondhi, S. L. Fractional Chern insulators and the W ∞ algebra. Phys. Rev. B 85 , 241308 (2012).
- 26 Roy, R. Band geometry of fractional topological insulators. Phys. Rev. B 90 , 165139 (2014).
- 27 Simon, B. Holonomy, the quantum adiabatic theorem, and Berry's phase. Phys. Rev. Lett. 51 , 2167-2170 (1983).
- 28 Berry, M. V. Quantal phase factors accompanying adiabatic changes. Proc. R. Soc. A 392 , 45-57 (1984).
- 29 Provost, J. P. & Vallee, G. Riemannian structure on manifolds of quantum states. Commun. Math. Phys. 76 , 289-301 (1980).
- 30 Wang, Y.-F., Gu, Z.-C., Gong, C.-D. & Sheng, D. N. Fractional quantum Hall effect of hard-core bosons in topological flat bands. Phys. Rev. Lett. 107 , 146803 (2011).
- 31 Scaffidi, T. & Möller, G. Adiabatic continuation of fractional Chern insulators to fractional quantum Hall states. Phys. Rev. Lett. 109 , 246805 (2012).
- 32 Wu, Y.-L., Bernevig, B. A. & Regnault, N. Zoology of fractional Chern insulators. Phys. Rev. B 85 , 075116 (2012).
- 33 Dobardzic, E., Milovanovic, M. V. & Regnault, N. Geometrical description of fractional Chern insulators based on static structure factor calculations. Phys. Rev. B 88 , 115117 (2013).
- 34 Dobardzic, E., Dimitrijevi´ c, M. & Milovanovic, M. V. Effective description of Chern insulators. Phys. Rev. B 89 , 235424 (2014).
- 35 Haldane, F. D. M. Model for a quantum Hall effect without Landau levels: condensed-matter realization of the 'parity anomaly'. Phys. Rev. Lett. 61 , 2015-2018 (1988).
- 36 Hu, X., Kargarian, M. & Fiete, G. A. Topological insulators and fractional quantum Hall effect on the ruby lattice. Phys. Rev. B 84 , 155116 (2011).
- 37 Goerbig, M. O. From fractional Chern insulators to a fractional quantum spin Hall effect. Eur. Phys. J. B 85 , 15 (2012).
- 38 Murthy, G. & Shankar, R. Composite fermions for fractionally filled Chern bands. Preprint at http://arxiv.org/abs/1108.5501 (2011).
- 39 Murthy, G. & Shankar, R. Hamiltonian theory of fractionally filled Chern bands. Phys. Rev. B 86 , 195146 (2012).
- 40 Repellin, C., Neupert, T., Papi´ c, Z. & Regnault, N. Single-mode approximation for fractional Chern insulators and the fractional quantum Hall effect on the torus. Phys. Rev. B 90 , 045114 (2014).
- 41 Bernevig, B. A. & Regnault, N. Emergent many-body translational symmetries of Abelian and non-Abelian fractionally filled topological insulators. Phys. Rev. B 85 , 075128 (2012).
- 42 Qi, X.-L. Generic wave-function description of fractional quantum anomalous Hall states and fractional topological insulators.
- Phys. Rev. Lett. 107 , 126803 (2011).
- 43 Wu, Y.-L., Regnault, N. & Bernevig, B. A. Gauge-fixed Wannier wave functions for fractional topological insulators. Phys. Rev. B 86 , 085129 (2012).
- 44 Wu, Y.-L., Regnault, N. & Bernevig, B. A. Bloch model wave functions and pseudopotentials for all fractional Chern insulators. Phys. Rev. Lett. 110 , 106802 (2013).
- 45 Lee, C. H., Thomale, R. & Qi, X.-L. Pseudopotential formalism for fractional Chern insulators. Phys. Rev. B 88 , 035101 (2013).
- 46 Lee, C. H. & Qi, X.-L. Lattice construction of pseudopotential Hamiltonians for fractional Chern insulators. Phys. Rev. B 90 , 085103 (2014).
- 47 Kovrizhin, D. L., Douçot, B. & Moessner, R. Multicomponent skyrmion lattices and their excitations. Phys. Rev. Lett. 110 , 186802 (2013).
- 48 Kapit, E. & Mueller, E. Exact parent Hamiltonian for the quantum Hall states in a lattice. Phys. Rev. Lett. 105 , 215303 (2010).
- 49 Liu, Z., Bergholtz, E. J. & Kapit, E. Non-Abelian fractional Chern insulators from long-range interactions. Phys. Rev. B 88 , 205101 (2013).
- 50 Parameswaran, S. A., Roy, R. & Sondhi, S. L. Fractional quantum Hall physics in topological flat bands. Comptes Rendus Physique 14 , 816-839 (2013).
- 51 Bergholtz, E. J. & Liu, Z. Topological flat band models and fractional Chern insulators. Int. J. Mod. Phys. B 27 , 1330017 (2013).
- 52 Li, H. & Haldane, F. D. M. Entanglement spectrum as a generalization of entanglement entropy: Identification of topological order in non-Abelian fractional quantum Hall effect states. Phys. Rev. Lett. 101 , 010504 (2008).
## ACKNOWLEDGMENTS
The authors thank Abhishek Roy for preliminary code written in the initial stage of the project and useful discussions and Emil Bergholtz for useful comments on a draft of this manuscript. R. R. acknowledges support from the Sloan Foundation. G. M. acknowledges support from the Leverhulme Trust under grant no. ECF-2011-565, from the Newton Trust of the University of Cambridge, and from the Royal Society under grant UF120157. This work used computational and storage services associated with the Hoffman2 Shared Cluster provided by UCLA Institute for Digital Research and Education's Research Technology Group. Part of our numerical work was performed using the Darwin Supercomputer of the University of Cambridge High Performance Computing Service funded by Strategic Research Infrastructure Funding from the Higher Education Funding Council for England and funding from the Science and Technology Facilities Council.
## SUPPLEMENTARY FIGURES
Supplementary Figure 1. Effect of Hamiltonian transformation on momentum space distribution of Berry curvature. We plot the deviations from uniform Berry curvature (neutral grey color) over a unit cell of the reciprocal lattice spanned by b 1 , b 2 , for models treated in the main text. Plots in each column use the same color scale and set of contours, as indicated. ( a ) - ( c ) Berry curvature distribution for the Haldane, kagomé lattice and ruby lattice models, respectively, at parameter values which minimize the root-mean-square curvature fluctuation σ B over the unit cell. Curvature fluctuations of the ruby lattice model at its min σ B point ( c ) are too small to be visible on the common scale. ( d ) - ( ) Berry curvature distribution for the transformed versions of these models used in Supplementary Ref. 1, evaluated the parameter f values minimizing the values of σ B as computed with the transformed Hamiltonians.
Supplementary Figure 2. Effect of Hamiltonian transformation on Berry curvature-gap correlations. Representative plot of the correlation between the many-body gap ∆ and root-mean-square Berry curvature fluctuation σ B . Data shown are for the Laughlin state of N = 8 fermions in the kagomé lattice model with nearest-neighbor repulsion on a 6 × 4 lattice. The single-particle parameters used are t 2 = λ 2 = 0 and t 1 = 1 , with the color scale corresponding to λ 1 . This model and parameters are chosen to match those used in Supplementary Ref. 1 (compare Fig. 26 of that reference); the thick black line corresponds to values of σ B computed using the transformed Hamiltonian of that reference, while colored circles denote values of σ B computed using the Hamiltonian in the main text. The two sets of data have the same gaps; only the computed values of σ B differ.
Supplementary Figure 3. Correlation between fluctuations of Berry curvature and of the quantum metric. The root-mean-square (RMS) fluctuation of the quantum metric over the Brillouin zone ( σ g ) is plotted against the RMS fluctuation of the Berry curvature σ B , for all parameter values used in the text for the ( a ) the kagomé lattice model and ( b ) the ruby lattice model. Despite sampling a large volume of the parameter space of both models, the fluctuations in both quantities show a large degree of linear correlation.
Supplementary Figure 4. Gap vs. determinant condition for the kagomé lattice model subject to constraints on σ B . ( a ) Gaps for the bosonic Laughlin state of N = 8 bosons at ν = 1 2 / , as a function of the Brillouin zone average of the determinant condition 〈 D 〉 . We only plot gaps for parameter values which have Berry curvature fluctuations σ B less than twice its minimum value. ( b ) Gap of the bosonic Laughlin state vs. 〈 D 〉 for coupling values randomly chosen on isosurfaces of constant σ B in the space of couplings. The parameter space sampling procedure used to obtain these sets of points is described in the main text (see Methods). ( c ), ( d ) The same, for the bosonic Moore-Read state of N = 10 bosons at ν = 1 . Note that the same sets of couplings are used in each column.
Supplementary Figure 5. Gap vs. determinant condition for the ruby lattice model subject to constraints on σ B . ( a ) Gaps for the bosonic Laughlin state of N = 8 bosons at ν = 1 2 / , as a function of the Brillouin zone average of the determinant condition 〈 D 〉 . We only plot gaps for parameter values which have Berry curvature fluctuations σ B less than twice its minimum value. ( b ) Gap of the bosonic Laughlin state vs. 〈 D 〉 for coupling values randomly chosen on isosurfaces of constant σ B in the space of couplings. The parameter space sampling procedure used to obtain these sets of points is described in the main text (see Methods). ( c ), ( d ) The same, for the bosonic Moore-Read state of N = 10 bosons at ν = 1 . Note that the same sets of couplings are used in each column.
Supplementary Figure 6. Finite-size scaling of the many-body gap in the kagomé lattice model with nearest-neighbor hoppings. The many-body gap data presented in the main text were obtained for systems of N = 8 particles, which represents a compromise between the need to accurately estimate the gap in thermodynamic limit and the need to exhaustively sample the parameter space of single-particle Hamiltonians using the computational resources available to us. Here we plot the many-body gap ∆ for the bosonic Laughlin state in the kagomé lattice model with nearest-neighbor hopping t 1 = 1 , variable λ 1 , and next-nearest-neighbor hopping t 2 = λ 2 = 0 . Red, black, green and blue points denote data for systems with N = 6 8 10 , , and 12 particles, respectively: the behavior changes significantly between system sizes N = 6 and N = 8 , but the gap data for N = 10 and N = 12 closely mirrors that of N = 8 , especially near the maximum of the gap. Inset: finite size scaling of the gap as a function of 1 /N for selected values of λ 1 . Near the maximal gap, the thermodynamic extrapolation ( N →∞ ) for the gap differs from the finite size value at N = 8 by about 10 %.
## SUPPLEMENTARY TABLES
Supplementary Table 1. Parameter values, band-geometric quantities and gaps for the Haldane model.
| | min σ B | max∆ L , Fermions | max∆ L , Bosons |
|---------|-----------|---------------------|-------------------|
| φ | 0 . 225 | 0 . 131 | 0 . 354 |
| M | 0 . 0 | 0 . 0 | 0 . 0 |
| σ B | 0 . 99326 | 1 . 24755 | 1 . 12760 |
| σ g | 2 . 85734 | 3 . 97585 | 3 . 25586 |
| 〈 D 〉 | 0 | 0 | 0 |
| 〈 T 〉 | 0 . 46705 | 0 . 84597 | 0 . 99358 |
| ∆ | | 0 . 02554 | 0 . 17243 |
As in the main text, we take the hopping amplitudes to be t 1 = t 2 = 1 , with the remaining couplings φ M , as free variables. Columns list information for the following values of ( φ, M ) of interest: those which minimize Berry curvature fluctuations σ B (obtained using the steepestdescent procedure described in Supplementary Note 5) and those maximizing the many-body gap ∆ for the Laughlin state for N = 8 fermions at ν = 1 3 / and N = 8 bosons at ν = 1 2 / . The latter two parameter values were found using our parameter space sampling procedure (see Methods). For these parameters, we compute the root-mean-square average over the Brillouin zone of fluctuations in the Berry curvature ( σ B ) and quantum metric ( σ g ), as well as the Brillouin zone averages of the determinant ( 〈 D 〉 ) and trace ( 〈 T 〉 ) conditions.
Supplementary Table 2. Parameter values, band-geometric quantities and gaps for the augmented Haldane model with third-nearest neighbor hopping.
| | min σ B | max∆ L | max∆ MR |
|---------|-----------|-----------|-----------|
| t 2 | 1 . 0 | 0 . 55 | 0 . 75 |
| t 3 | - 1 . 0 | - 0 . 65 | - 0 . 65 |
| σ B | 0 . 14699 | 0 . 43579 | 0 . 34909 |
| σ g | 3 . 00134 | 1 . 70829 | 2 . 29776 |
| 〈 D 〉 | 0 | 0 | 0 |
| 〈 T 〉 | 3 . 44154 | 0 . 75821 | 1 . 46273 |
| ∆ | | 0 . 19153 | 0 . 12614 |
As in the main text, we fix t 1 = 1 , φ = π/ 2 , and M = 0 , allowing t 2 , t 3 to vary. Columns list information for the following values of ( t 2 , t 3 ) of interest: those which minimize Berry curvature fluctuations σ B (obtained using the steepest-descent procedure described in Supplementary Note 5) and those maximizing the many-body gap ∆ for the Laughlin state of N = 8 bosons at ν = 1 2 / and for the Moore-Read state of of N = 10 bosons at ν = 1 . The latter two parameter values were found using our parameter space sampling procedure (see Methods). For these parameters, we compute the root-mean-square average over the Brillouin zone of fluctuations in the Berry curvature ( σ B ) and quantum metric ( σ g ), as well as the Brillouin zone averages of the determinant ( 〈 D 〉 ) and trace ( 〈 T 〉 ) conditions.
Supplementary Table 3. Parameter values, band-geometric quantities and gaps for the kagomé lattice model.
| | min σ B | max∆ L | max∆ MR |
|---------|-----------|-----------|-----------|
| λ 1 | 0 . 745 | 0 . 725 | 0 . 934 |
| t 2 | - 0 . 361 | - 0 . 361 | - 0 . 168 |
| λ 2 | 0 . 078 | 0 . 055 | - 0 . 129 |
| σ B | 0 . 22178 | 0 . 23231 | 0 . 34967 |
| σ g | 1 . 31746 | 1 . 25972 | 1 . 25434 |
| 〈 D 〉 | 0 . 09608 | 0 . 09857 | 0 . 41665 |
| 〈 T 〉 | 0 . 79612 | 0 . 70181 | 0 . 42712 |
| ∆ | | 0 . 17614 | 0 . 08246 |
We fix an energy scale by setting t 1 = 1 and letting the remaining model parameters ( λ 1 , t 2 , λ 2 ) vary. Columns list information for the following parameter values of interest: those which minimize Berry curvature fluctuations σ B (obtained using the steepest-descent procedure described in Supplementary Note 5) and those maximizing the many-body gap ∆ for the Laughlin state of N = 8 bosons at ν = 1 2 / and for the Moore-Read state of of N = 10 bosons at ν = 1 . The latter two parameter values were found using our parameter space sampling procedure (see Methods). For these parameters, we compute the root-mean-square average over the Brillouin zone of fluctuations in the Berry curvature ( σ B ) and quantum metric ( σ g ), as well as the Brillouin zone averages of the determinant ( 〈 D 〉 ) and trace ( 〈 T 〉 ) conditions.
Supplementary Table 4. Parameter values, band-geometric quantities and gaps for the ruby lattice model.
| | min σ B | max∆ L | max∆ MR |
|---------|-----------|-----------|-----------|
| Im t | 1 . 857 | 1 . 886 | 1 . 672 |
| Re t 1 | 0 . 685 | - 0 . 569 | - 1 . 261 |
| Im t 1 | 2 . 079 | 2 . 536 | 2 . 789 |
| t 4 | - 2 . 118 | - 2 . 022 | - 2 . 007 |
| σ B | 0 . 13398 | 0 . 14347 | 0 . 17980 |
| σ g | 0 . 47669 | 0 . 21574 | 0 . 38617 |
| 〈 D 〉 | 0 . 17740 | 0 . 25569 | 0 . 30456 |
| 〈 T 〉 | 0 . 17168 | 0 . 08262 | 0 . 11139 |
| ∆ | | 0 . 10173 | 0 . 03132 |
We fix an energy scale by setting Re t = 1 and letting the remaining parameters (Im t , t 1 , t 4 ) vary. Columns list information for the following parameter values of interest: those which minimize Berry curvature fluctuations σ B (obtained using the steepest-descent procedure described in Supplementary Note 5) and those maximizing the many-body gap ∆ for the Laughlin state of N = 8 bosons at ν = 1 2 / and for the Moore-Read state of of N = 10 bosons at ν = 1 . The latter two parameter values were found using our parameter space sampling procedure (see Methods). For these parameters, we compute the root-mean-square average over the Brillouin zone of fluctuations in the Berry curvature ( σ B ) and quantum metric ( σ g ), as well as the Brillouin zone averages of the determinant ( 〈 D 〉 ) and trace ( 〈 T 〉 ) conditions.
are only defined up to an overall phase
$$| u _ { \mathbf k } ^ { \alpha } \rangle \rightarrow e ^ { i \phi _ { \alpha } ( \mathbf k ) } | u _ { \mathbf k } ^ { \alpha } \rangle,$$
where φ α ( k ) is any smooth function satisfying φ α ( k + G ) = φ α ( k ) . This is the gauge symmetry of the band Hamiltonian, and it is the only such symmetry in the absence of energy degeneracies (assumed for simplicity throughout this section). As with the U (1) gauge symmetry of electromagnetism, gauge transformations of the form (5) alter the Berry connection but leave the Berry curvature (analogue of the magnetic field) and quantum metric unchanged, as can be seen from their explicitly gauge-invariant forms (28), (29).
The reader will note that the band Hamiltonians used in this article do not have the periodicity of the reciprocal lattice; nor do their eigenfunctions | u α k 〉 , and so neither can immediately be viewed as functions on the BZ torus defined by identifying the points k and k + G for any reciprocal lattice vector G . We first explain that the formalism of band geometry is unchanged in this situation, and then argue that this choice of basis is the canonically correct one, in the sense of corresponding to the observable quantities mentioned above.
Because the full Hamiltonian is periodic in real space, Bloch's theorem implies
$$u _ { k } ^ { \alpha } ( { \mathbf r } = { \mathbf d } _ { b } ) \equiv \langle 0 b | u _ { k } ^ { \alpha } \rangle = \sum _ { G } c _ { k - G } e ^ { i G \cdot { \mathbf d } _ { b } }$$
## SUPPLEMENTARY NOTES
## Supplementary Note 1: Gauge freedom and geometric phases in band Hamiltonians
Here we discuss issues arising in the definition of Berry curvature for Bloch bands which are absent from the general formalism of geometric phases. The issues identified here are not new; 2,3 we discuss them here in order to demonstrate that the values the Berry curvature and quantum metric take at a specific crystal momentum k are unambiguously defined and, in principle, measurable.
Berry curvature arises in physical applications at the as the commutator of band-projected position operators; as examples, we cite the semiclassical approximation to the orbital magnetization 4 which contributes to anomalous thermoelectric transport, and the intrinsic contribution to the anomalous Hall conductance 5
$$\sigma _ { x y } ^ { \text{int} } & = \frac { e ^ { 2 } } { \hbar } \int \frac { d ^ { d } k } { ( 2 \pi ) ^ { d } } \, f ( E _ { \mathbf k } ) B ( \mathbf k ),$$
Other applications are reviewed in Supplementary Ref. 6. We mention these observables here because they depend on the Berry curvature through forms other than its Brillouin zone (BZ) average (the Chern number), meaning that the distribution of curvature within the BZ is an experimentally meaningful quantity. There is also a recent experimental proposal to measure the Berry curvature directly. 7
Textbook discussions of Berry's phase are usually framed in the context of adiabatic evolution of a quantum state tracing out a closed cycle in some parameter manifold. In Chern insulator applications, the parameter manifold is the Brillouin zone, and the instantaneous eigenfunctions at a parameter k are the spatially periodic part of the Bloch functions u α b ( k ) . The Berry connection is A α ( k ) = - 〈 i u α k |∇ | k u α k 〉 , which has curvature given by
$$B _ { \alpha } ( { \mathbf k } ) = - i \sum _ { b = 1 } ^ { \mathcal { N } } \left ( \frac { \partial u _ { b } ^ { \alpha * } } { \partial k _ { x } } \frac { \partial u _ { b } ^ { \alpha } } { \partial k _ { y } } - \frac { \partial u _ { b } ^ { \alpha * } } { \partial k _ { y } } \frac { \partial u _ { b } ^ { \alpha } } { \partial k _ { x } } \right ).$$
The first Chern number is defined as the surface integral of the Berry curvature
$$c _ { 1 } = \frac { 1 } { 2 \pi } \int _ { B Z } d ^ { 2 } k \, B _ { \alpha } ( { \mathbf k } )$$
and is topologically quantized to integer values due to the fact that the Brillouin zone is a compact manifold (a torus).
Eigenstates of the band Hamiltonian
$$H _ { b c } ( \mathbf k ) = \sum _ { \alpha = 1 } ^ { \mathcal { N } } E _ { \alpha } ( \mathbf k ) u _ { b } ^ { \alpha * } ( \mathbf k ) u _ { c } ^ { \alpha } ( \mathbf k )$$
is unchanged under d b → d b + R ; in other words, there exists a unitary matrix U G = e i G r · ̂ such that
$$u _ { b } ^ { \alpha } ( { \mathbf k } + { \mathbf G } ) = \sum _ { c = 1 } ^ { \mathcal { N } } ( U _ { \mathbf G } ) _ { b c } u _ { c } ^ { \alpha } ( { \mathbf k } ),$$
for all k , α . Because U G is independent of k , it drops out of the expressions for the Berry curvature and quantum metric, which are therefore periodic in k .
One could also obtain manifestly periodic Bloch functions by performing momentum-dependent phase shifts c † b, k → e -i r b · k c † b, k , with different offsets r b for each sublattice b , so that the transformed Bloch functions are invariant under k → k + G : the transformed curvature and metric are then periodic as well. The resulting Hamiltonian is, of course, gauge-inequivalent to the original one, which can be seen from the fact that the curvature itself changes: under transformations
$$u _ { a } ^ { \alpha } ( { \mathbf k } ) \rightarrow \widetilde { u } _ { b } ^ { \alpha } ( { \mathbf k } ) = e ^ { i { \mathbf r } _ { b } \cdot { \mathbf k } } u _ { b } ^ { \alpha } ( { \mathbf k } )$$
for b = 1 , . . . , N , the Berry curvature at k changes by
$$\widetilde { B } _ { \alpha } ( { \mathbf k } ) - B _ { \alpha } ( { \mathbf k } ) = \sum _ { b = 1 } ^ { \mathcal { N } } r _ { b, y } \frac { \partial } { \partial k _ { x } } | u _ { b } ^ { \alpha } ( { \mathbf k } ) | ^ { 2 } - r _ { b, x } \frac { \partial } { \partial k _ { y } } | u _ { b } ^ { \alpha } ( { \mathbf k } ) | ^ { 2 }.$$
Because this is a sum of total derivatives, the surface integrals of ˜ B α ( k ) and B α ( k ) yield the same Chern number. The difference itself, however, only vanishes when r b is the same for all b , which is the gauge transformation (5).
Phase shifts of the form (8) were employed in recent publications 1,8 to obtain band Hamiltonians that were periodic in k . This was described as a 'gauge transformation' in these references, but as we've noted, the only gauge symmetry of the Hamiltonian is with respect to bands (i.e. U (1) rotations in the eigenbasis γ α † k → e -iφ α ( k ) γ α † k ). For the transformations made in Supplementary Ref. 1, the difference in curvature fluctuations is substantial, as shown in Supplementary Fig. 1 for the Haldane, kagomé lattice and ruby lattice models. Note that each panel of this figure shows the curvature distribution for parameters which minimize σ B as computed with each panel's respective Hamiltonian; i.e. they depict the closest one may get to uniform band geometry in the parameter space of the Hamiltonian considered.
For completeness, we note that unlike these single-particle properties, the many-body gap is invariant under the generalized phase shift (8), because single-particle density operators ρ k are left unchanged by the transformation. As an example, in Supplementary Fig. 2 we reproduce the results shown in Fig. 26 of Supplementary Ref. 1, along with the fluctuations in the canonically defined Berry curvature for the same system. The latter are lower than in the transformed Hamiltonian used in that reference, meaning that the negative correlation between σ B and the gap is stronger than depicted there: introducing phase shifts by hand distorts the Berry curvature distribution to a degree which significantly affects the conclusions one may draw from that data.
We have shown that band geometry may be defined for non-periodic band Hamiltonians; we now argue that the non-periodic basis used in this paper is in fact the one measured by any (direct or indirect) experimental probe, and hence is the only one which should be regarded as physical. The feature possessed by Bloch bands which is absent from the general theory of Berry phases is the fact that the parameter space in question is defined via the Fourier transform of the kinematic setting of the physical system. Because this is a global transform, in doing the Fourier sum
$$| { \mathbf k }, b \rangle = \frac { 1 } { \sqrt { N _ { c } } } \sum _ { \mathbf R } e ^ { i { \mathbf k } \cdot ( { \mathbf R } + { \mathbf d } _ { b } ) } | { \mathbf R }, b \rangle,$$
we have already implicitly chosen a basis for the band Hilbert space at each k . For example, the the transformations made in Supplementary Refs. 1 and 8 correspond to defining modified tight-binding states on the reciprocal lattice via
$$\widetilde { \chi } _ { b } ( \mathbf k ) = \frac { 1 } { \sqrt { N _ { c } } } \sum _ { \mathbf R } e ^ { i \mathbf k \cdot \mathbf R } \chi _ { b } ( \mathbf r - \mathbf R - \mathbf d _ { b } )$$
where χ b ( r -( R + d b )) = 〈 r R | b 〉 ; this is manifestly invariant under k → k + G , but in doing so the position operator is no longer consistently defined for orbitals with different offsets d b : the transformation is equivalent to shifting the orbitals of the crystal basis to the origins of various lattice cells. As an obviously apparent consequence, the crystal symmetry of the transformed Berry curvatures in the second row of Supplementary Fig. 1 are unphysically broken. Because experiments probe position-space quantities, we conclude that the basis defined by (10) and used in this paper is the canonical, physically relevant quantity.
The inequalities
## Supplementary Note 2: Relation between the trace and determinant inequalities
$$\text{tr} \, g ^ { \alpha } ( \mathbf k ) \geq | B _ { \alpha } ( \mathbf k ) | ;$$
$$\det g ^ { \alpha } ( \mathbf k ) & \geq B _ { \alpha } ( \mathbf k ) ^ { 2 } / 4,$$
were proved in Supplementary Ref. 9. Assuming the quantum metric is nondegenerate, it may be factored as
$$g ^ { \alpha } = \sqrt { g ^ { \alpha } } \begin{pmatrix} \tilde { g } _ { 1 1 } & \tilde { g } _ { 1 2 } \\ \tilde { g } _ { 1 2 } & \tilde { g } _ { 2 2 } \end{pmatrix}$$
where the scalar √ g α = (det g α ) 1 / 2 and the second factor is a unimodular matrix. Using the standard inequality between arithmetic and geometric means,
$$\text{tr} \, g ^ { \alpha } & = \sqrt { g ^ { \alpha } } ( \tilde { g } _ { 1 1 } + \tilde { g } _ { 2 2 } ) \geq 2 \sqrt { g ^ { \alpha } } \sqrt { \tilde { g } _ { 1 1 } \tilde { g } _ { 2 2 } }.$$
By unimodularity √ ˜ g 11 ˜ g 22 = √ 1 + ˜ g 2 12 ≥ 1 , so
$$( \text{tr} \, g ^ { \alpha } ( \text{k} ) ) ^ { 2 } \geq 4 \det g ^ { \alpha } ( \text{k} ),$$
with equality if and only if g α is proportional to the identity matrix. Because the Berry curvature is algebraically independent of the components of the quantum metric, we can conclude that saturation of trace inequality (12) implies saturation of the determinant inequality (13). In physical terms, a constant curvature and metric that saturate the determinant inequality imply the closure of a modified Girvin-MacDonald-Platzman (GMP) algebra; saturation of the trace inequality then corresponds to the stronger condition that the algebra of band projected density operators is not only isomorphic to the GMP algebra, but identical to it.
## Supplementary Note 3: Two-band models
By adding suitably chosen exponentially-localized couplings, any two-band Hamiltonian may be brought to a band-flattened form parameterized by a unit 3-vector H flat ( k ) = -̂ n k ( ) · σ , where σ = ( σ , σ 1 2 , σ 3 ) are the three Pauli matrices. The eigenvectors of the original and flattened Hamiltonians are identical (by construction) and can be obtained analytically:
$$u ^ { \pm } ( { \mathbf k } ) = \frac { 1 } { \sqrt { 2 ( 1 \mp \widehat { n } _ { 3 } ) } } \begin{pmatrix} \widehat { n } _ { 3 } \mp 1 \\ \widehat { n } _ { 1 } + i \widehat { n } _ { 2 } \end{pmatrix}.$$
Using this expression gives the Berry curvature and quantum metric of the lower band as
B
=
1
2
̂
n
·
∂
x
̂
n
×
∂
y
̂
n
;
(18)
$$g _ { \mu \nu } = \frac { 1 } { 4 } \partial _ { \mu } \widehat { n } \cdot \partial _ { \nu } \widehat { n },$$
since ̂ n · ∂ µ ̂ n = 0 . Using this and standard identities relating multiple dot and cross products, it follows that
$$g _ { x x } ( { \mathbf k } ) g _ { y y } ( { \mathbf k } ) - g _ { x y } ( { \mathbf k } ) ^ { 2 } - \frac { 1 } { 4 } B ( { \mathbf k } ) ^ { 2 } = 0 \, \colon$$
the determinant condition det g α ( k ) -B α ( k ) 2 / 4 = 0 is necessarily satisfied for any two-band model. As our results with the kagomé and ruby lattice models show, this ceases to be the case for models having more than two bands.
## Supplementary Note 4: Scaling of gaps with the number of bands
Fractional Chern insulators (FCIs) exhibit states which may be thought of as discretizations of continuum fractional quantum Hall (FQH) states, in that they have identical topological and long-wavelength properties (see, e.g., Supplementary Ref. 10). To that end, consider a continuum position-space wavefunction ψ ( r ) which is discretized to set of values ψ b in a tight-binding
model with N sites per unit cell. The normalization conditions on the corresponding Bloch functions are
$$\int _ { \text{UC} } d \mathbf r \, | u _ { \mathbf k } ( \mathbf r ) | ^ { 2 } = \sum _ { b = 1 } ^ { \mathcal { N } } | u _ { \mathbf k, b } | ^ { 2 } = 1,$$
for any k , where the integral is taken over a unit cell of the lattice. In the limit of large N , we may assume that u k ( r ) at r = d b is proportional to the discretized value u k ,b . Approximating the continuum normalization integral by a sum introduces a factor of 1 / N from the integration measure, which is compensated by the scaling
$$u _ { \mathbf k } ( \mathbf d _ { b } ) & \sim \sqrt { \mathcal { N } } u _ { \mathbf k, b }.$$
Now consider the matrix elements of the delta-function interaction employed in the main text to stabilize the bosonic Laughlin state. In the continuum, these are
$$\langle k _ { 3 } k _ { 4 } | \widehat { V } | k _ { 1 } k _ { 2 } \rangle & = V \int d { \mathbf r } \, u _ { k _ { 3 } } ^ { * } ( { \mathbf r } ) u _ { k _ { 4 } } ^ { * } ( { \mathbf r } ) u _ { k _ { 1 } } ( { \mathbf r } ) u _ { k _ { 2 } } ( { \mathbf r } ) \\ & \sim \frac { V } { \mathcal { N } } \sum _ { b } u _ { k _ { 3 } } ^ { * } ( { \mathbf d } _ { b } ) u _ { k _ { 4 } } ^ { * } ( { \mathbf d } _ { b } ) u _ { k _ { 1 } } ( { \mathbf d } _ { b } ) u _ { k _ { 2 } } ( { \mathbf d } _ { b } ),$$
times a momentum-conserving δ -function. Comparing this with an on-site interaction in the discretized model
$$V _ { \text{disc} } \sum _ { b } u _ { \text{k} _ { 3 }, b } ^ { * } u _ { \text{k} _ { 4 }, b } ^ { * } u _ { \text{k} _ { 1 }, b } u _ { \text{k} _ { 2 }, b }$$
and using (22) shows that the discretized interaction strength should be scaled as
$$V _ { \text{disc} } \sim V \mathcal { N } \quad ( \delta \text{-function} ).$$
This means that, given gaps which have been obtained for the FCI Laughlin state in two lattice models A,B , the quantities which should be compared are N A ∆ A and N B ∆ B (assuming that the single-particle dispersion has been flattened; i.e. that the gap is only set by the interaction term.) Similar considerations show that for the three-body delta-function interaction used to stabilize the Moore-Read state, the leading scaling should be V disc ∼ V N 2 .
We note that the above argument is not as simple as it may appear: we're implicitly assuming ψ ( r ) has the character of a low-lying eigenstate - more specifically, that it has support over almost all tight-binding orbitals. This is the case when u k ,b is a randomly chosen vector from the N -dimensional band Hilbert space, but one can easily construct counterexamples violating this assumption: for example, let lattice model B (with N B > N A ) have a block-diagonal kinetic term, the first block of which is identical to the kinetic term of model A and the second block of which describes additional, trivial 'spectator' orbitals with very high occupation energies. The low-lying bands u B k ,b will have vanishing amplitude on these spectator orbitals, and will numerically be identical to the corresponding eigenfunctions u A k ,b of model A , with zeros appended. In this scenario, the eigenfunction scaling assumption (22) is violated, and the interaction strength (and gap size) would not scale with the number of spectator orbitals.
The scaling argument can be placed on a more rigorous footing in the context of the Hofstadter model 11 in the limit of small flux per plaquette φ = 1 /N , which offers a sequence of lattice models (with N bands) which converges to the continuum FQHE in the N →∞ limit. In this case, the scalings (22), (23) are numerically observed to be obeyed to high accuracy. 12 For the case of the FCI models studied in this work, we know the ground states of different models lie in the same universality class as the continuum FQHE state (as can be determined by topological order, done here by entanglement spectrum counting), but we do not expect, e.g., the ruby lattice model ground state to be an interpolation of the kagomé lattice model ground state. The fact that we observe the scaled gaps to be so close to each other provides an a posteriori confirmation of this argument's validity even for the relatively small values of N = 2 3 6 , , .
## Supplementary Note 5: Expansion of band-geometric quantities in parameter space
The Berry curvature and quantum metric are defined in terms of the derivatives of a wavefunction over some parameter manifold. For both numerical and theoretical purposes, it becomes convenient to recast these expressions in terms of derivatives of the Hamiltonian itself, rather than its eigenfunctions, since ordinarily only the former is known analytically. This, of course, is not a new observation. 13
In order to avoid overall phase ambiguities and the necessity of using multiple charts to cover the Brillouin zone (BZ) in
topologically non-trivial situations, it's preferable to work with the occupied band projector P α = | k , α 〉〈 k , α | , instead of the eigenfunction | k , α 〉 itself. In these terms, the Feynman-Hellman theorems are
$$\partial _ { \lambda } E _ { \alpha } = \mathbf r r \, P _ { \alpha } \, \partial _ { \lambda } H ;$$
$$\partial _ { \lambda } P _ { \alpha } = R _ { \alpha } \, \partial _ { \lambda } H \, P _ { \alpha } + P _ { \alpha } \, \partial _ { \lambda } H \, R _ { \alpha },$$
where the projected resolvent operator R α is glyph[negationslash]
$$R _ { \alpha } = \sum _ { \beta \neq \alpha } \frac { | \beta \rangle \langle \beta | } { E _ { \alpha } - E _ { \beta } } = \frac { 1 - P _ { \alpha } } { E _ { \alpha } - H }.$$
Equations (24) and (25) hold for any parameter λ upon which the Hamiltonian smoothly depends, and they are valid for arbitrary values of λ and hence may be further differentiated.
Assuming no additional degeneracies, ( E α -H ) may be inverted in the subspace orthogonal to | k , α 〉 , and
$$\partial _ { \lambda } R _ { \alpha } & = - R _ { \alpha } ^ { 2 } \, \partial _ { \lambda } H \, P _ { \alpha } - P _ { \alpha } \, \partial _ { \lambda } H \, R _ { \alpha } ^ { 2 } - R _ { \alpha } \partial _ { \lambda } ( E _ { \alpha } - H ) \, R _ { \alpha }.$$
The relations (24), (25) and (27) then form a closed system and may be iterated to any order to develop the Taylor expansion of band-geometric quantities.
For example, the Berry curvature and quantum metric for a single occupied Chern band are 14,15
$$B _ { \alpha } = 2 \, \text{Im} \, \text{tr} \, P _ { \alpha } \, \partial _ { y } H \, R _ { \alpha } ^ { 2 } \, \partial _ { x } H,$$
g
α
2
µν
=
Re tr
P
α
∂
µ
HR ∂ H.
α
ν
(29)
The derivative of the Berry curvature with respect to a coupling λ follows from a straightforward computation as
$$\partial _ { \lambda } B _ { \alpha } & = 2 \, \text{Im} \, \text{tr} \, & P _ { \alpha } \left \{ - 2 ( \partial _ { \lambda } E _ { \alpha } ) \partial _ { y } H \, R _ { \alpha } ^ { 3 } \, \partial _ { x } H + \partial _ { y } H \, R _ { \alpha } ^ { 2 } \, \partial _ { x, \lambda } H + \partial _ { y, \lambda } H \, R _ { \alpha } ^ { 2 } \, \partial _ { x } H \\ & \quad + \partial _ { y } H \, R _ { \alpha } ( \partial _ { \lambda } H \, R _ { \alpha } + R _ { \alpha } \, \partial _ { \lambda } H ) \, R _ { \alpha } \, \partial _ { x } H \\ & \quad + \left [ ( \partial _ { \lambda } H \, R _ { \alpha } \, \partial _ { y } H \, R _ { \alpha } ^ { 2 } \, \partial _ { x } H + ( \partial _ { x } E _ { \alpha } ) \partial _ { \lambda } H \, R _ { \alpha } ^ { 3 } \, \partial _ { y } H \, \right ) - ( x \leftrightarrow y ) \right ] \right \}.$$
Other quantities such as the Hessian ∂ ∂ λ λ ′ B α may be calculated in a similar manner, although the algebra rapidly becomes tedious. These may, in turn, be used to find the variation in Brillouin zone-averaged quantities with respect to couplings in the Hamiltonian; for example, denoting by 〈· · · 〉 the Brillouin zone average,
$$c _ { 1 } & = \frac { A _ { B Z } } { 2 \pi } \langle B \rangle,$$
$$\sigma _ { B } = \frac { A _ { B Z } } { 2 \pi } \underset { \mu \nu } { \sqrt { \langle B ^ { 2 } \rangle - \langle B \rangle ^ { 2 } } },$$
$$\text{and so} \, \partial _ { \lambda } \sigma _ { B } = \frac { \stackrel { \mu \cdot \cdot } { ( A _ { B Z } / 2 \pi ) } ^ { 2 } } { \sigma _ { B } } \langle B \partial _ { \lambda } B \rangle.$$
Here we've made use of the fact that ∂ λ 〈 B 〉 = 0 while we remain in the same topological phase, which can easily be checked during numerical computations. Using (30) in (33) then permits us to rapidly perform a steepest-descent minimization of Berry curvature fluctuations, even in a high-dimensional parameter space.
## SUPPLEMENTARY REFERENCES
- ∗ [email protected]
- 1 Wu, Y.-L., Bernevig, B. A. & Regnault, N. Zoology of fractional Chern insulators. Phys. Rev. B 85 , 075116 (2012).
- 2 Zak, J. Berry's phase for energy bands in solids. Phys. Rev. Lett. 62 , 2747-2750 (1989).
- 3 Zak, J. Berry's geometrical phase for noncyclic Hamiltonians. Europhys. Lett. 9 , 615-620 (1989).
- 4 Sundaram, G. & Niu, Q. Wave-packet dynamics in slowly perturbed crystals: Gradient corrections and Berry-phase effects. Phys. Rev. B 59 , 14915-14925 (1999).
- 5 Karplus, R. & Luttinger, J. Hall effect in ferromagnetics. Phys. Rev. 95 , 1154-1160 (1954).
- 6 Xiao, D., Chang, M.-C. & Niu, Q. Berry phase effects on electronic properties. Rev. Mod. Phys. 82 , 1959-2007 (2010).
- 7 Duca, L. et al. An Aharonov-Bohm interferometer for determining Bloch band topology. Science 347 , 288-292 (2015).
- 8 Lee, C. H., Thomale, R. & Qi, X.-L. Pseudopotential formalism for fractional Chern insulators. Phys. Rev. B 88 , 035101 (2013).
- 9 Roy, R. Band geometry of fractional topological insulators. Phys. Rev. B 90 , 165139 (2014).
- 10 Regnault, N. & Bernevig, B. A. Fractional Chern insulator. Phys. Rev. X 1 , 021014 (2011).
- 11 Hofstadter, D. Energy levels and wave functions of Bloch electrons in rational and irrational magnetic fields. Phys. Rev. B 14 , 2239-2249 (1976).
- 12 Bauer, D., Jackson, T. & Roy, R. Quantum geometry and stability of the fractional quantum Hall effect in the Hofstadter model. Preprint at http://arxiv.org/abs/1504.07185 (2015).
- 13 Berry, M. V. Quantal phase factors accompanying adiabatic changes. Proc. R. Soc. A 392 , 45-57 (1984).
- 14 Avron, J. & Seiler, R. Quantization of the Hall conductance for general, multiparticle Schrödinger Hamiltonians. Phys. Rev. Lett. 54 , 259-262 (1985).
- 15 Campos Venuti, L. & Zanardi, P. Quantum critical scaling of the geometric tensors. Phys. Rev. Lett. 99 , 095701 (2007).
|
10.1038/ncomms9629
|
[
"T. S. Jackson",
"G. Möller",
"R. Roy"
] |
2014-08-05T00:49:14+00:00
|
2015-11-14T22:19:34+00:00
|
[
"cond-mat.str-el",
"cond-mat.mes-hall",
"cond-mat.quant-gas"
] |
Geometric stability of topological lattice phases
|
The fractional quantum Hall (FQH) effect illustrates the range of novel
phenomena which can arise in a topologically ordered state in the presence of
strong interactions. The possibility of realizing FQH-like phases in models
with strong lattice effects has attracted intense interest as a more
experimentally accessible venue for FQH phenomena which calls for more
theoretical attention. Here we investigate the physical relevance of previously
derived geometric conditions which quantify deviations from the Landau level
physics of the FQHE. We conduct extensive numerical many-body simulations on
several lattice models, obtaining new theoretical results in the process, and
find remarkable correlation between these conditions and the many-body gap.
These results indicate which physical factors are most relevant for the
stability of FQH-like phases, a paradigm we refer to as the geometric stability
hypothesis, and provide easily implementable guidelines for obtaining robust
FQH-like phases in numerical or real-world experiments.
|
1408.0844v2
|
## On transcendental analytic functions mapping an uncountable class of U -numbers into Liouville numbers
Diego Marques 1, ∗ , Josimar Ramirez 2
Departamento de Matem´ atica, Universidade de Bras´ ılia, Bras´ ılia, 70910-900, Brazil
## Abstract
In this paper, we shall prove, for any m ≥ 1, the existence of an uncountable subset of U -numbers of type ≤ m (which we called the set of m -ultra numbers ) for which there exists uncountably many transcendental analytic functions mapping it into Liouville numbers.
Keywords: m -ultra numbers, transcendental functions 2010 MSC: 11Jxx
## 1. Introduction
A transcendental function is a function f ( x ) such that the only complex polynomial satisfying P x, f ( ( x )) = 0, for all x in its domain, is the null polynomial. For instance, the trigonometric functions, the exponential function, and their inverses.
The study of the arithmetic behavior of transcendental functions at complex points has attracted the attention of many mathematicians for decades. The first result concerning this subject goes back to 1884, when Lindemann proved that the transcendental function e z assumes transcendental values at all nonzero algebraic point. In 1886, Weierstrass gave an example of a transcendental entire function which takes rational values at all rational points. Later, St¨ckel a [6] proved that for each countable subset Σ ⊆ C and each
∗ Corresponding author
Email addresses: [email protected] (Diego Marques), [email protected]
(Josimar Ramirez)
1 Supported by FAP-DF and CNPq-Brazil
2 Scholarship holder of CAPES-Brazil
dense subset T ⊆ C , there is a transcendental entire function f such that f (Σ) ⊆ T (F. Gramain showed that St¨ckel's theorem is valid if Σ and a T are subsets of R ). Another construction due to St¨ckel [7] produces a transcena dental entire function f whose derivatives f ( ) t , for t = 0 1 2 , , , . . . , all map Q into Q . Two years later, G. Faber refined this result by showing the existence of a transcendental entire function such that f ( ) t ( Q ) ⊆ Q ( ), for all i t ≥ 0. In 1968, van der Poorten [5] proved the existence of a transcendental function f , such that f ( s ) ( α ) ∈ Q ( α ), for all α ∈ Q . In 2011, Marques [4] proved some of these results in the hypertranscendental context (for more on this subject, we refer the reader to [8] and the references therein). Let den( z ) be the denominator of the rational number z . Very recently, Marques and Moreira [3] proved the existence of uncountable many transcendental entire functions f , such that f ( Q ) ⊆ Q and den( f ( p/q )) < q 8 q 2 , for all p/q ∈ Q , with q > 1.
A real number ξ is called a Liouville number , if there exists a rational sequence ( p /q k k ) k ≥ 1 , with q k > 1, such that
$$0 < \left | \xi - \frac { p _ { k } } { q _ { k } } \right | < q _ { k } ^ { - k }, \, \text{for $k=1,2,\dots$} \\ \text{e Liouville numbers is denoted by $\mathbb{L}$ and it is a}$$
∣ ∣ The set of the Liouville numbers is denoted by L and it is a dense G δ set (and therefore uncountable).
All the previously mentioned results deal with the arithmetic behavior of a countable set by a transcendental function. However, in 1984, in one of his last papers, K. Mahler [2] raised the following question:
Question. Are there transcendental entire functions f ( z ) such that if ξ is any Liouville number, then so is f ( ξ )?
He also said that: 'The difficulty of this problem lies of course in the fact that the set of all Liouville numbers is non-enumerable'. In this direction, Marques and Moreira [3] showed the existence of an uncountable subset of Liouville numbers for which there exists uncountably many transcendental entire functions mapping it into itself.
As usual, H ( α ) will denote the height of the algebraic number α (that is, the maximum of the absolute values of its primitive minimal polynomial over Z ) and exp [3] ( x ) = e e e x . Now, let us define the following class of numbers:
Definition 1. A real number ξ is called an m -ultra number if there exist infinitely many m -degree real algebraic numbers ( α n n ) such that
$$0 < | \xi - \alpha _ { n } | < ( \exp ^ { [ 3 ] } ( \mathcal { H } ( \alpha _ { n } ) ) ) ^ { - n }, \, f o r \, n = 1, 2, \dots.$$
The set of the m -ultra numbers will be denoted by U m -ultra .
It follows from the definition that U m -ultra is a subset of U -numbers of type at most m and it is also a dense G δ set (in particular it is uncountable) -it means that U m -ultra is a large set in a topological sense. In particular, every real number can be written as the sum (and the product, provided it is non zero) of two m -ultra numbers, as in [1].
The aim of this paper is to prove the following result:
Theorem 1. There exist uncountable many analytic transcendental functions φ : R → R such that φ ( U m -ultra ) ⊆ L .
Recall that Q m denotes the set of all m -degree real algebraic numbers. Since Q m and Q are dense countable sets of R , there exist uncountable many transcendental analytic functions φ with φ ( Q m ) ⊆ Q . In order to prove Theorem 1 we shall find a class of such functions with an upper bound for den( φ α ( )) in terms of m and H ( α ). More precisely, we have
Theorem 2. For any given positive integer m ≥ 1 , there exist uncountably many transcendental analytic functions φ : R → R with | φ ′ ( x ) | < 0 0001 . , φ ( Q m ) ⊆ Q and such that for all α ∈ Q m , it holds that
$$\text{den} ( \phi ( \alpha ) ) \leq ( 2 q ) ^ { 4 5 0 m ^ { 5 } 2 ^ { 1 8 m ^ { 2 } } q ^ { 6 m } },$$
where q = H ( α ) .
## 2. The proofs
2.1. Proof that Theorem 2 implies Theorem 1
Given an m -ultra number ξ , there exist infinitely many α n ∈ Q m , with height at least max { m, 8 } , and such that
$$0 < | \xi - \alpha _ { n } | < \frac { 1 } { ( \exp ^ { [ 3 ] } ( \mathcal { H } ( \alpha _ { n } ) ) ^ { n } }, \, \text{for } n = 1, 2, \dots.$$
Let φ be a function as in Theorem 2. By the Mean Value Theorem, we obtain
$$\i \tan \pi \tan \pi | \phi ( \xi ) - \phi ( \alpha _ { n } ) | \leq 0. 0 0 0 1 \cdot | \xi - \alpha _ { n } | < \frac { 1 } { ( \exp ^ { [ 3 ] } ( H ( \alpha _ { n } ) ) ) ^ { n } }.$$
We know that φ α ( n ) = p /q n n , with q n ≤ (2 t n ) 450 m 5 2 18 m 2 t 6 m n , where t n = H ( α n ). Since t n ≥ max { m, 8 } , then a straightforward calculation gives q n ≤ exp [3] ( t n ) and hence
$$\cdot \cdot \cdot \cdot \left | \phi ( \xi ) - \frac { p _ { n } } { q _ { n } } \right | = | \phi ( \xi ) - \phi ( \alpha _ { n } ) | < \frac { 1 } { q _ { n } ^ { n } }, \, \text{for } n = 1, 2, \dots. \right | \\ \text{is implies that } \phi ( \xi ) \text{ is a Liouville number as desired}.$$
∣ ∣ This implies that φ ξ (
) is a Liouville number as desired. /square
## 2.2. Proof of Theorem 2
Before starting the proof, we shall state some useful facts
- · For any distinct y, b ∈ -[ 1 1], it holds that , | sin( y -b ) | > y | - | b / 3. (Indeed, the function sin( x /x ) is decreasing for x ∈ (0 , π ], and sin(2) / 2 > 1 / 3.)
- · For any x, y ∈ Q m we have H ( y -x ) ≤ 2 4 m 2 H ( x ) m H ( y ) m (Indeed, let W x ( ) be the absolute multiplicative Weil height, we know that
$$\frac { 1 } { 2 ^ { \partial ( x ) } } W ( x ) ^ { \partial ( x ) } \leqslant \mathcal { H } ( x ) \leqslant 2 ^ { \partial ( x ) } W ( x ) ^ { \partial ( x ) }, \quad \quad ( 2 )$$
where ∂ x ( ) is the degree of the algebraic number x . Using this together with the inequality W x ( + ) y ≤ 2 W x W y ( ) ( ) the result follows.)
- · For any distinct x, y ∈ Q m ∩ [0 , 1 / 2], with H ( x , ) H ( y ) ≤ n , we have
$$| \cos ( \pi x ) - \cos ( \pi y ) | \geq \frac { \pi } { 2 ^ { 4 m ^ { 2 } + 1 } n ^ { 2 m + 1 } }.$$
(Indeed, we can assume x < y . Then by the mean value theorem, one has | cos( πx ) -cos( πy ) | ≥ sin( πx )( πy -πx ) ≥ 2 πx y ( -x . ) There is a simple lower bound for the modulus of a nonzero complex algebraic numbers α in terms of the height H ( α ), namely | α | ≥ ( H ( α ) + 1) -1 ≥ (2 H ( α )) -1 (see [9, Page 82]). Thus,
$$\int _ { 0 \infty } \lfloor \sigma, \, \Gamma \underset { \left [ \sigma, 1 \infty \infty, 1 \infty \infty, 1 \infty \infty \infty, 1 \infty \infty \infty \infty \infty } { 1 } \left | \cos ( \pi x ) - \cos ( \pi y ) \right | \ > \ \frac { \pi } { 2 H ( x ) H ( y - x ) } \\ \geq \ \frac { \pi } { 2 ^ { 4 m ^ { 2 } + 1 } n ^ { 2 m + 1 } }. \right )$$
- · For every /epsilon1 > 0, any interval of length > /epsilon1 contains at least two rational numbers with denominator ≤ /ceilingleft 2 //epsilon1 /ceilingright . /ceilingleft /ceilingright
- (Indeed, if m = 2 //epsilon1 and ( a, b ) is the interior of the interval, we have b -a > /epsilon1 ≥ 2 /m , and so, for k = /floorleft ma /floorright +1, we have ma < k ≤ ma +1, and so ma < k < k +1 ≤ ma +2 < ma + m b ( -a ) = mb , which implies a < k/m < ( k +1) /m < b .)
- · Let f : R → R be a periodic function which assumes infinitely many
- values. Then f is transcendental. (Indeed, suppose that f is algebraic, and let P x, y ( ) = ∑ n i =0 a i ( x y ) i be the polynomial with minimal degree n in the variable y , such that a n ( x ) has minimal degree and P x, f ( ( x )) = 0, ∀ x ∈ R . We may suppose that a n ( x ) is monic. Since f assumes infinitely many values, then a 0 ( x , . . . , a ) n ( x ) cannot be all constants. Set /lscript as the largest index with a /lscript ( x ) non constant. If t is the period of f , then Q k ( x, f ( x )) := P x ( + tk, f ( x + tk )) -P x, f ( ( x )) = ∑ /lscript i =0 ( a i ( x + tk ) -a i ( x ))( f ( x )) i = 0, for all ( x, k ) ∈ R × Z . Note that for some integer k 0 , a /lscript ( x + tk 0 ) -a /lscript ( x ) is nonzero. If /lscript = n , then a n ( x + tk ) -a n ( x ) is nonzero having degree smaller than the degree of a n ( x ) which contradicts the minimality of the degree of a n ( x ). In the case of /lscript < n , then Q k 0 ( x, f ( x )) = 0 and Q k 0 has degree /lscript < n , in y , which contradicts our assumption on the minimality of n .)
- · Let f : R → R be a transcendental function and let g : R → R be a non constant algebraic function. Then f ◦ g is transcendental. (Indeed, by assumption, C ( y, f ( y )) is transcendental over C ( y ). Setting y = g x ( ) is merely making an algebraic extension of each, so C ( g x , f ( ) ( g x ( ))) is transcendental over C ( g x ( )). Thus C ( g x , f ( ) ( g x ( ))) is transcendental over C ( g x ( )). Hence the tower C ( g x , f ( ) ( g x ( ))) ⊃ C ( g x ( )) ⊃ C ( x ) is transcendental, so f ( g x ( )) is transcendental over C ( x ).)
Now, we are ready to deal with the proof of the theorem. Consider the following enumeration of A := Q m ∩ [0 , 1 / 2]:
$$A = \{ \alpha _ { 1 }, \alpha _ { 2 }, \alpha _ { 3 }, \dots \},$$
constructed as follows. Let S k be the set of all irreducible and primitive polynomials in Z [ x ] with degree m and height k . Denote by t k := |S | k <
/negationslash
( m + 1)(2 k + 1) m . Let R k be the set all distinct roots of polynomial in S k belonging to the interval [0 , 1 / 2], (Note that R ∩ R k t = ∅ , for k = ) and t l k = |R | k , then R k = { γ ( k ) 1 , . . . , γ ( k ) l k } with γ ( k ) i < γ ( k ) i +1 ∀ k ≥ 1. So the desired enumeration is given by
$$A = \{ \alpha _ { 1 }, \alpha _ { 2 }, \alpha _ { 3 }, \dots \} = \{ \mathcal { R } _ { 1 }, \mathcal { R } _ { 2 }, \mathcal { R } _ { 3 }, \dots \}.$$
Now, we will give estimates for the height of the algebraic numbers in A as a function of the position in the enumeration. Although estimates are not the best, they will be sufficient for our purposes.
If α n ∈ R k +1 , then H ( α n ) = k + 1. We have n ≤ l 1 + · · · + l k +1 ≤ t 1 + · · · + t k +1 ≤ ( m +1)(2 k +3) m +1 , therefore
$$\mathcal { H } ( \alpha _ { n } ) \geqslant \frac { 1 } { 2 } \, ^ { m + 1 } \sqrt { \frac { n } { m + 1 } } - 2.$$
On another hand, n ≥ l 1 + · · · + l k . Let j be an odd number with 4 < j ≤ k , then l j ≥ 1 (because (2 /j ) 1 /m ∈ R j ). Thus, if k ≥ 5, we have n ≥ /floorleft k -4 2 /floorright > k -6 2 , therefore H ( α n ) < 2 n + 7 (the cases k = 1 2 3 4 are trivial). , , , Define B n = { y , y 1 2 , . . . , y n } with y k := cos( πα k ).
Set h : C C
→ given by
$$h ( x ) = g ( \cos ( \pi x ) ),$$
where g y ( ) = ∑ ∞ n =1 c n g n ( y ), with g n ( y ) = ∏ b ∈ B n sin( y -b ).
Suppose that c n = 0 for 1 ≤ n ≤ 5 and | c n | < 1 /n n for every positive integer n . We claim that h is an entire function. In fact, for all y belonging to the open ball B (0 , R ) one has that
$$| g _ { n } ( y ) | < \prod _ { b \in B _ { n } } e ^ { | y - b | } \leq e ^ { n ( R + 1 ) },$$
where we used that b ∈ -[ 1 1]. , Thus, since | c n | < 1 /n n , we get | c n g n ( y ) | ≤ ( e R +1 /n ) n yielding that g (and so h ) is an entire function, since the series g y ( ) = ∑ ∞ n =1 c n g n ( y ), which defines g , converges uniformly in any of these balls. Let f : R → R be the restriction of h to R . In particular, f is analytic and | f ′ ( x ) | ≤ ∞ n =6 1 /n n -1 < 0 0002, for all . x ∈ R .
∑ Now, we shall choose inductively c n 's conveniently such that f satisfies f ( α k ) ∈ Q , for all k , and den( f ( α k )) < (72 m 2 (6 q ) 4 m ) 10 m 3 (6 q ) 2 m , where q = H ( α k ).
Suppose that c , . . . , c 1 n -1 were chosen such that f ( α 1 ) , . . . , f ( α n ) have the desired properties (notice that the choice of c , . . . , c 1 n -1 determines the values of f ( α 1 ) , . . . , f ( α n ), independently of the values of c k , k ≥ n ; in particular, since c k = 0 for 1 ≤ k ≤ 5, we have f ( α n ) = 0 for 1 ≤ n ≤ 6). Now, we shall choose c n for which f ( α n +1 ) satisfies the requirements.
/negationslash
Let t ≤ n be positive integers with n ≥ 5. Then H ( α n +1 ) , H ( α t ) ≤ 2 n +9. Since cos( πα n +1 ) = cos( πα t ), then
$$| y _ { n + 1 } - y _ { t } | \geq \frac { \pi } { 2 ^ { 4 m ^ { 2 } + 1 } ( 2 n + 9 ) ^ { 2 m + 1 } }.$$
## Therefore
$$| \sin ( y _ { n + 1 } - y _ { t } ) | > \frac { | y _ { n + 1 } - y _ { t } | } { 3 } > \frac { \pi / 3 } { 2 ^ { 4 m ^ { 2 } + 1 } ( 2 n + 9 ) ^ { 2 m + 1 } }$$
/negationslash yielding | g n ( y n +1 ) | > ( π/ 3 2 4 m 2 +1 (2 n +9) 2 m +1 ) n . Thus c n g n ( y n +1 ) runs an interval of length larger than 2 π n (3 n ) -n 2 -n (4 m 2 +1) (2 n +9) -3 mn . Now, we may choose (in at least two ways) c n = 0 such that g y ( n +1 ) is a rational number with denominator at most n n 2 n (4 m 2 +1) (2 n +9) 3 mn . Thus den( f ( α k )) = den( g (cos( πα k )) = den( g y ( k )) ≤ ( k -1) ( k -1) 2 ( k -1)(4 m 2 +1) (2 k +7) 3 m k ( -1) < k k 2 k (4 m 2 +1) (2 k +7) 3 mk . Since q := H ( α k ) ≥ 1 2 m +1 √ k/ m ( +1) -2, we get k ≤ (2 q +4) ( m +1) ( m +1). Then
$$\text{den} ( f ( \alpha _ { k } ) ) \ & \geq \ \left ( ( 2 q + 4 ) ^ { ( m + 1 ) } ( m + 1 ) \right ) ^ { ( 2 q + 4 ) ^ { ( m + 1 ) } ( m + 1 ) } 2 ^ { ( 2 q + 4 ) ^ { ( m + 1 ) } ( m + 1 ) ( 4 m ^ { 2 } + 1 ) } \\ & \quad \times ( 2 ( 2 q + 4 ) ^ { ( m + 1 ) } ( m + 1 ) + 7 ) ^ { 3 m ( 2 q + 4 ) ^ { ( m + 1 ) } ( m + 1 ) } \\ & < \ ( 7 2 m ^ { 2 } ( 6 q ) ^ { 4 m } ) ^ { 1 0 m ^ { 3 } ( 6 q ) ^ { 2 m } }.$$
Now, consider the function ψ : R → R , given by ψ x ( ) := x 2(1+ x 2 ) . Note that ψ ( Q m ) ⊆ Q m ∩ [0 , 1 / 2]. Therefore, our desired function is φ := f ◦ ψ . In fact, φ ( Q m ) ⊆ Q , | φ ′ ( x ) | = | f ′ ( ψ x ( )) || ψ ′ ( x ) | < 0 0001, for all . x ∈ R , and by our previous argument, if α ∈ Q m , then
$$\text{den} ( \phi ( \alpha ) ) = \text{den} ( f ( \psi ( \alpha ) ) ) \leq ( 7 2 m ^ { 2 } ( 6 t ) ^ { 4 m } ) ^ { 1 0 m ^ { 3 } ( 6 t ) ^ { 2 m } },$$
where t = H ( ψ α ( )). On the other hand, we can use (2) together with the fact that W x/y ( ) ≤ W x W y ( ) ( ) to obtain t = H ( α 2(1+ α 2 ) ) ≤ 2 6 m q 3 , where q = H ( α ). Thus
$$\text{den} ( \phi ( \alpha ) ) < ( 2 q ) ^ { 4 5 0 m ^ { 5 } 2 ^ { 1 8 m ^ { 2 } } q ^ { 6 m } },$$
## as desired.
Note that since there is a binary tree of different possibilities for f (if we have choosen c , c 1 2 , . . . , c n -1 , different choices of c n give different values of f ( y n +1 ), which does not depend on the values of c k for k > n , and so different functions f ), we constructed uncountably many possible functions f . So there exist uncountable many functions φ (since ψ is non constant).
Now, it remains to prove that all functions constructed above are transcendental: in fact, since f assumes infinitely many values (because it is continue and non constant) and it is periodic (with period 2), then f is transcendental. Therefore f ◦ ψ is transcendental, because ψ is a non constant rational function. /square
## acknowledgements
The authors would like to thank Joseph Silverman and Carlos Gustavo Moreira for helpful suggestions which improved the quality of this paper. The first author is supported by FAP-DF and CNPq-Brazil and the second author is scholarship holder of CAPES-Brazil.
## References
- [1] P. Erd˝s, Representations of real numbers as sums and products of o Liouville numbers, Michigan Math. J. 9 , 59-60 (1962).
- [2] K. Mahler, Some suggestions for further research, Bull. Austral. Math. Soc. 29 (1984), 101-108.
- [3] D. Marques, C. G. Moreira, A variant of a question proposed by K. Mahler concerning Liouville numbers, to appear in Bull. Australian Math. Soc.
- [4] D. Marques, On the arithmetic nature of hypertranscendental functions at complex points. Expositiones Mathematicae 29 (2011), 361370.
- [5] A. J. van der Poorten, Transcendental entire functions mapping every algebraic number field into itself, J. Austral. Math. Soc. 8 (1968), 192-193.
- [6] P. St¨ckel, Ueber arithmetische Eingenschaften analytischer Functioa nen, Math. Ann. 46 (1895), no. 4, 513-520.
- [7] P. St¨ckel, a Arithmetische eingenschaften analytischer functionen, Acta Math. 25 (1902) 371-383.
- [8] M. Waldschmidt, Algebraic values of analytic functions, Proceedings of the International Conference on Special Functions and their Applications (Chennai, 2002). J. Comput. Appl. Math. 160 (2003), no. 1-2, 323-333.
- [9] M. Waldschmidt, Diophantine Approximation on Linear Algebraic Groups . Springer-Verlag, Berlin Heidelberg, New York (2000).
| null |
[
"Diego Marques",
"Josimar Ramirez"
] |
2014-08-05T00:56:57+00:00
|
2014-08-28T23:07:40+00:00
|
[
"math.NT",
"11Jxx"
] |
On transcendental analytic functions mapping an uncountable class of $U$-numbers into Liouville numbers
|
In this paper, we shall prove, for any $m\geq 1$, the existence of an
uncountable subset of $U$-numbers of type $\leq m$ (which we called the set of
{\it $m$-ultra numbers}) for which there exists uncountably many transcendental
analytic functions mapping it into Liouville numbers.
|
1408.0845v1
|
## Predicting missing links and their weights via reliable-route-based method
Jing Zhao (a)1 , Lili Miao , Haiyang Fang , Qian-Ming Zhang 2 1 2 3 , , Min Nie 2 and Tao Zhou (b)2 3 ,
- 1 Department of Mathematics, Logistical Engineering University, Chongqing 404100, People's Republic of China
- 2 Web Sciences Center, School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu 610054, People's Republic of China
- 3 Big Data Research Center, University of Electronic Science and Technology of China, Chengdu 611731, People's Republic of China
PACS 89.20.Ff -
Computer science and technology
PACS 89.75.Hc -
Networks and genealogical trees
PACS 89.65.-s - Social and economic systems
Abstract -Link prediction aims to uncover missing links or predict the emergence of future relationships according to the current networks structure. Plenty of algorithms have been developed for link prediction in unweighted networks, with only a very few of them having been extended to weighted networks. Thus far, how to predict weights of links is important but rarely studied. In this Letter, we present a reliable-route-based method to extend unweighted local similarity indices to weighted indices and propose a method to predict both the link existence and link weights accordingly. Experiments on different real networks suggest that the weighted resource allocation index has the best performance to predict the existence of links, while the reliableroute-based weighted resource allocation index performs noticeably better on weight prediction. Further analysis shows a strong correlation for both link prediction and weight prediction: the larger the clustering coefficient, the higher the prediction accuracy.
Introduction. -Link prediction aims to predict whether there exists a missing link between a pair of nonconnected nodes or whether a link may appear in the future [1]. Link prediction can not only help to find missing data in real networks, but also complement our understanding on the evolution processes of networks [2, 3]. Thus it has attracted increasing attentions in recent years. Ideally, topological features of the network and profiles of the considered nodes are combined to assess the possibility of the existence of links between nodes. For example, when predicting literature citations based on the citation network, Popescul and Ungar considered not only topological characteristics of the network, but also the node attributes, such as authors, journal names and contents of the papers [4]. However, in most cases, it is very difficult to get attribute information of nodes or the information is not reliable. For example, in online social networks, the information of many users is confidential or false. Therefore, many algorithms only make use of topological information.
(a) Corresponding author: [email protected]
(b) Corresponding author: [email protected]
A few variant problems make use of functional outputs [5] or predict link direction instead of link existence [6]. Up to now, the mainstream of topological based methods is the similarity-based algorithm, including algorithms that use local, global or quasi-local information respectively to compute the similarity indices [1,7].
Most previous studies in link prediction focus on unweighted networks. Recently, a few works tried to extend similarity indices from unweighted networks to weighted networks [8-13]. Some strategies have been proposed to enhance the precision of weighted indices, for example, to emphasize the contributions of weak links [8], or to integrate multiple indices, including local and global ones [9]. However, these studies did not consider the prediction of weights, which is also very significant in addition to the prediction of the existence of links.
In this Letter, we try to predict missing links and their weights using local similarity measures. Inspired by the solution of the most reliable route problem in communication networks [14], we propose a reliable-route-based method to generalize unweighted similarity indices to weighted ones.
Considering that the similarity index between two unconnected nodes essentially reflects interaction strength between the two nodes, we set weights of missing links proportional to the similarity indices. We apply an optimization algorithm to identify the optimum weight prediction function and measure the accuracy of weight prediction by the Pearson correlation coefficient. Six real networks are used in experiments to evaluate our methods, indicating that the weighted resource allocation index and the reliable-route-based weighted resource allocation index outperform other indices in predicting the link existence and link weights, respectively.
Problem and Metrics. Given a weighted undirected network G V,E,W ( ), where V , E and W are sets of nodes, links and link weights, respectively, we want to find out its missing links (or links that may appear in the future) and predict their weights as well. To do this, for each pair of unconnected nodes x, y ∈ V , we assign a likelihood score S xy to quantify the probability that ( x , y ) is a missing link. Then all unconnected pairs are ranked decreasingly according to their scores, and the links on the top are considered to be of the highest existence likelihoods. S xy is usually called similarity index in literatures.
To test the algorithm's accuracy, we randomly divide the link set E into a training set E T and a test set E V , so that, E = E T ∪ E V and E T ∩ E V = φ . In this study, E T and E V contain 90% and 10% links of E , respectively. We use precision to quantify the accuracy, which is the ratio of real missing links to those taken as predicted links, i.e., if the top L links are considered as predicted links while L r of which appear in the test set, the precision is L /L r .
In most weighted networks, the link weights actually measure pairwise affinities between nodes. That is, the larger the weight, the higher extent of relevance between corresponding nodes. When conducting link prediction, the likelihood score S xy is defined to measure the possibility of the existence of a link between nodes x and y , which in essence reflects the affinity extent of nodes x and y . Therefore, we can adjust the likelihood scores to predict link weights. Specifically, same as in link prediction, the link set E is randomly divided into a training set E T and a test set E V , thus the weighted adjacency matrices of the training and test sets are W T and W V , respectively. Then the likelihood score for link prediction is in fact a function S W ( T ) of the input matrix W T . We need to define a weight prediction function F W ( T ) so that the difference between F W ( T ) and W V can be as small as possible.
We set F W ( T ) = λ · S W ( T ) where λ is a scaling coefficient which can be identified by solving the following optimization problem:
$$\min _ { \lambda } \| \lambda \cdot S ( W _ { T } ) = W _ { V } \| _ { F } \quad \quad ( 1 ) \quad \ \ n \pi _ { \dots }$$
$$s. t. \ 0 < \lambda \leq M _ { V } / M _ { S T }$$
where M V and M ST are maximum values of elements in the matrices W V and S W ( T ), respectively, and ‖ · ‖ denotes the Frobenius norm, defined as the square root of the sum of the squares of the matrix's elements [15]. We measure the accuracy of weight prediction by the Pearson correlation coefficient between the vectors of predicted and known link weights in the test set.
Methods. Systematical comparison suggested that the Common Neighbors (CN), Adamic-Adar (AA) and Resource Allocation (RA) indices perform best among lots of local similarity indices in unweighted networks [1,7,16,17]. Thus we focus on these three measures whose definitions are as follows.
- (i) CN index. The CN index simply counts the number of common neighbors between nodes x and y as:
$$S _ { x y } = | \Gamma ( x ) \cap \Gamma ( y ) | ;$$
where Γ( x ) is the set of neighbors of node x and |·| denotes cardinality of the set.
- (ii) AA index [18]. This index depresses the contribution of the high-degree common neighbors by assigning larger contribution to less-connected neighbors:
$$S _ { x y } = \sum _ { z \in \Gamma ( x ) \cap \Gamma ( y ) } \frac { 1 } { \log k _ { z } } ;$$
where k z is the degree of node z .
(iii) RA index [16]. Similar to AA index, RA index also punishes the high-degree common neighbors, but to a higher extent, as
$$S _ { x y } = \sum _ { z \in \Gamma ( x ) \cap \Gamma ( y ) } \frac { 1 } { k _ { z } }. \quad \quad ( 4 )$$
Previous studies extended similarity indices from unweighted networks to weighted networks by introducing the sum of weights of the two links ( z, x ) and ( z, y ), where z runs over all common neighbors of nodes x and y , as [8, 12]:
- (i) Weighted CN index (WCN):
$$S _ { x y } = \sum _ { z \in \Gamma ( x ) \cap \Gamma ( y ) } ( W _ { x z } + W _ { z y } ) ; \quad \quad ( 5 )$$
(ii) Weighted AA index (WAA):
$$S _ { x y } = \sum _ { z \in \Gamma ( x ) \cap \Gamma ( y ) } \frac { W _ { x z } + W _ { y z } } { \log \left ( 1 + S _ { z } \right ) }, \quad \quad ( 6 )$$
(iii) Weighted RA index (WRA):
$$S _ { x y } = \sum _ { z \in \Gamma ( x ) \cap \Gamma ( y ) } \frac { W _ { x z } + W _ { y z } } { S _ { z } }. \quad \quad ( 7 )$$
Here, S z denotes the strength of node z , namely the sum of weights of associated links, as
$$S _ { z } = \sum _ { z ^ { \prime } \in \Gamma ( z ) } W _ { z z ^ { \prime } }.$$
Table 1: Basic topological features of the networks under study. | V | and | E | are the number of nodes and links, 〈 k 〉 is the average degree, and C and C w are the unweighted and weighted versions (see Eq. (13) and Eq. (14) for definitions) of clustering coefficient, respectively.
| Networks | hsaPPi | Cel | Geom | String | Corum | String Corum |
|------------|-----------|----------|-----------|-----------------|-----------|----------------|
| | V | | 2821 | 281 | 7343 | 16886 | 2314 | 2270 |
| | E | | 13880 | 2402 | 11898 | 1.52093e+06 | 34148 | 153788 |
| 〈 k 〉 | 9.84 | 17.1 | 3.24 | 180.14 | 29.51 | 135.49 |
| C | 0.169 | 0.346 | 0.486 | 0.301 | 0.747 | 0.371 |
| C w | 0.167 | 0.291 | 0.535 | 0.231 | 0.795 | 0.278 |
In operations research, the most reliable route problem on a communication network requires to select the most reliable route to transmit data packages from a source node to a destination node. That is, one needs to find a route that links these two nodes and maximizes the probability that a package can reach the destination but not be corrupted in a non repairable fashion on the route. In this case, the communication network is represented as a weighted network, in which the weight of a link is its reliability, i.e., the probability that this link is safe for data transmit. Thus the reliability of a route is the product of weights of links along this route [14]. As we have mentioned, weight in most weighted networks actually measures pairwise affinity between nodes, which is usually strongly related to reliability. For example, weight of a collaboration network is the number of co-authorized publications between two scientists, which is statistically correlated to the possibility that these two scientists will collaborate in the future [19]. In a protein-protein interaction network, weight is the confidence score of the interaction, which represents the probability that the interaction occurs [20]. In a collaborative network of e-commerce users, weight characterizes the co-purchases between two users [21]. Thus we think it is reasonable to measure the affinity of a pair of unconnected nodes by the product of weights of paths connecting them. Therefore, we define the so-called reliable-route-based weighted similarity indices as follows:
(i) Reliable-route-based weighted CN index (rWCN):
$$S _ { x y } = \sum _ { z \in \Gamma ( x ) \cap \Gamma ( y ) } ( W _ { x z } \cdot W _ { z y } ) \quad \quad ( 9 ) \ \underset { \substack { \text{STRI} \\ \dots \ e } } { \text{genre} }.$$
(ii) Reliable-route-based weighted AA index (rWAA):
$$S _ { x y } = \sum _ { z \in \Gamma ( x ) \cap \Gamma ( y ) } \frac { W _ { x z } \cdot W _ { y z } } { \log \left ( 1 + S _ { z } \right ) } \quad \ \ ( 1 0 ) \ \text{done} \\ \text{and } u \\ \text{toract}$$
(iii) Reliable-route-based weighted RA index (rWRA):
$$S _ { x y } = \sum _ { z \in \Gamma ( x ) \cap \Gamma ( y ) } \frac { W _ { x z } \cdot W _ { y z } } { S _ { z } } \quad \quad ( 1 1 ) \stackrel { n e n s r } { \, \text{We d} } \\ \text{whose} \quad \text{non} \quad \text{.}$$
Notice that, the reliable-route-based indices are borrowed from the definition of route reliability in the most reliable route problem in communication networks, where the edge weight of the communication network is the probability that this link is in good condition, which is no more than 1. Therefore, for networks whose weights do not lie in the range [0,1], before calculating reliable-route-based weighted similarity indices, we first map their weights to (0,1) through
$$w ^ { \prime } = e ^ { - \frac { 1 } { w } },$$
where w and w ′ denote the original and regulated weights, respectively. Since the function (12) is a one-to-one mapping, it is easy to extract the original weight w from the weight w ′ .
Results. Six real weighted networks are used to test our method. (i) hsaPPI: a high-confidence protein-protein interaction network of human beings constructed from the experimental biochemical co-fractionation data with overlap information derived from curated public databases and literature searches, in which the weight denotes the interaction confidence level [22]. (ii) Cel: the updated version of the neural network of C.elegans, in which nodes are neurons, links are synaptic contacts between neurons, and the weight represents numbers of synapses between the neuron pair [23]. (iii) Geom: the collaboration network covering scientific collaborations between scientists in computational geometry till February 2002, in which the link weight corresponds to the number of coauthorized publications between two scientists (can be downloaded from the Pajek Datasets or the Computational Geometry Database ). (iv) String: weighted human gene-association network constructed from the database STRING [24] (Search Tool for the Retrieval of Interacting Genes/Proteins). STRING integrates both physical interactions and functional associations from numerous sources and provides each link with a probabilistic confidence score. The version 9.05 of STRING was downloaded and used in this study. (v) Corum: a protein-protein interaction network of component proteins in human protein complexes collected by the database CORUM (Comprehensive Resource of Mammalian protein complexes) [25]. We downloaded the database CORUM in June of 2013, whose core data includes 1343 complexes and 2314 component proteins. In this network, two proteins are linked if they appear in the same complex while the weight of each link is given by the number of shared complexes. (vi) String Corum: a sub-network of String constructed by ex-
Table 2: Link prediction accuracy measured by precision on the networks under study.
| Networks | hsaPPI | Cel | Geom | String | Corum | String Corum |
|------------|----------|-------|--------|----------|---------|----------------|
| CN | 0.207 | 0.123 | 0.196 | 0.171 | 0.626 | 0.249 |
| WCN | 0.207 | 0.131 | 0.177 | 0.186 | 0.631 | 0.274 |
| rWCN | 0.206 | 0.13 | 0.163 | 0.187 | 0.67 | 0.27 |
| AA | 0.217 | 0.127 | 0.391 | 0.18 | 0.63 | 0.263 |
| WAA | 0.217 | 0.134 | 0.391 | 0.197 | 0.674 | 0.285 |
| rWAA | 0.216 | 0.137 | 0.352 | 0.198 | 0.703 | 0.28 |
| RA | 0.221 | 0.124 | 0.485 | 0.221 | 0.845 | 0.304 |
| WRA | 0.223 | 0.13 | 0.493 | 0.24 | 0.895 | 0.323 |
| rWRA | 0.23 | 0.13 | 0.479 | 0.244 | 0.891 | 0.315 |
Table 3: Weight prediction accuracy measured by Pearson correlation coefficient on the networks under study.
| Networks | hsaPPI | Cel | Geom | String | Corum | String Corum |
|------------|----------|-------|--------|----------|---------|----------------|
| CN | 0.283 | 0.178 | 0.246 | 0.189 | 0.668 | 0.242 |
| WCN | 0.292 | 0.183 | 0.239 | 0.251 | 0.71 | 0.299 |
| rWCN | 0.3 | 0.181 | 0.235 | 0.348 | 0.75 | 0.386 |
| AA | 0.286 | 0.194 | 0.347 | 0.201 | 0.699 | 0.251 |
| WAA | 0.294 | 0.205 | 0.227 | 0.078 | 0.749 | 0.309 |
| rWAA | 0.305 | 0.206 | 0.249 | 0.108 | 0.786 | 0.4 |
| RA | 0.236 | 0.209 | 0.431 | 0.261 | 0.795 | 0.295 |
| WRA | 0.24 | 0.214 | 0.446 | 0.292 | 0.832 | 0.339 |
| rWRA | 0.256 | 0.216 | 0.454 | 0.391 | 0.858 | 0.438 |
tracting the proteins in CORUM and their links from the network String. See Table 1 for the basic topological measures of these networks. In Cel, Geom and Corum, weights stand for the numbers of synapses, co-authors and shared complexes, respectively. As mentioned above, we will first transform the weight w in these three networks to area (0,1) by Eq. (12).
For each of the six networks, we randomly split its links into training and test set, which contain 90% and 10% of the links, respectively. Accuracy of link prediction is measured by precision, while accuracy of weight prediction is assessed by the Pearson correlation coefficient between the vectors of predicted and known link weights in the test set. When calculating precision metric, we set L as the size of the test set. Repeating this process for 100 times, we obtained the average accuracy measures for link and weight prediction as listed in Table 2 and Table 3. For each network, the bold number in the row highlights the best prediction result.
Tables 2 and Table 3 show that the algorithms of RA series perform overall best for both link and weight predictions, which is in consistent with earlier studies for link prediction [1]. For link prediction, the only exception is Cel, whose highest precision is obtained by rWAA, while for weight prediction, the only exception is hsaPPI, whose best correlation coefficient comes from rWAA. In particular, rWRA significantly outperforms others in weight prediction.
It can be seen that the network Corum gets remarkably much higher prediction accuracy than the other net- works, which may be because of that the network contains a plenty of cliques, as indicated by its large clustering coefficients for both unweighted and weighted versions (see Table 1), respectively defined as [26,27]:
$$C ( i ) = \frac { \sum _ { j k } a _ { i j } a _ { j k } a _ { k i } } { \sum _ { j k } a _ { i j } a _ { k i } }, \quad \quad ( 1 3 )$$
where a ij equals to 1 when there is a link between node i and node j , else a ij is zero. And
$$C _ { w } ( i ) = \frac { \sum _ { j k } w _ { i j } w _ { j k } w _ { k i } } { \max _ { i j } w _ { i j } \sum _ { j k } w _ { i j } w _ { k i } }, \quad \ \ ( 1 4 )$$
where w ij represents the weight of link ( i, j ). The clustering coefficient of a network is the average clustering coefficient over all nodes whose degree is bigger than 1. Since the current local indices only takes into account common neighborhoods of two nodes, it is very straightforward to infer that the larger the clustering coefficient, the better the prediction. Indeed, the highest precision accuracies, corresponding to the bold numbers in Table 2 and Table 3, exhibit a strongly positive correlation with the clustering coefficients, as indicated by the high Pearson correlation coefficients reported in Table 4. Furthermore, one can also infer (not solid yet) that the weighted clustering coefficient gives a better characterization than the unweighted version, as indicated by the larger correlation coefficients.
Conclusion and Discussion. This work aims to predict missing links and their weights based on local information. A reliable-route-based method is proposed to
Table 4: Pearson correlations between clustering coefficients and the highest prediction accuracies corresponding to bold numbers in Table 2 and Table 3.
| | Link Prediction | Weight Prediction |
|-----|-------------------|---------------------|
| C | 0.866 | 0.773 |
| C w | 0.94 | 0.863 |
generalize similarity indices defined from unweighted to weighted networks. Extensive experiments on real networks show that the series of resource allocation indices performs overall best in both link prediction and weight prediction. In addition, we show that the larger the clustering coefficient is, the higher the prediction accuracy is.
The prediction accuracy is also affected by the network background. For example, there are four networks whose nodes are human proteins. The network hsaPPI only includes high-confidence physical interactions between proteins, which is thus the most sparse network among the four. The network String comprises functionally associated pairs, including physical interactions, co-expression, co-localization, forming complex, and participating same biological process. In fact, the database STRING come from experiments, curated databases, literature mining, and computational prediction. Therefore, String is the most noisy one and its links are built up by different organizing principles, which is usually not easy for link prediction algorithms. That may be the reason why prediction accuracies for hsaPPI and String are relatively poor. In comparison, the network Corum was constructed to represent theoretical links between component proteins of complexes, who is the most regularly organized and thus is most easily to be explained and predicted.
## ∗ ∗ ∗
This work was partially supported by the National Natural Science Foundation of China under Grant Nos. 61372194, 81260672 and 11222543. T.Z. acknowledge the program for New Century Excellent Talents in University under Grant No. NCET-11-0070.
## REFERENCES
- [1] L¨ u. L. and Zhou T. , Physica A , 390 (2011) 1150.
- [2] Wang W. Q., Zhang Q. M. and Zhou T. , EPL , 98 (2012) 28004.
- [3] Zhang Q. M., L¨. L.,Wang W. Q.,Zhu Y. X. u and Zhou T. , PLOS ONE , 8 (2013) e55437.
- [4] Popescul A. and Ungar L. H. , Workshop on learning statistical models from relational data (JCA Press) 2003, pp. 81.
- [5] Barzel B. and Barab´ asi A.-L. , Nat. Biotech. , 31 (2013) 720.
- [6] Guo F., Yang Z. and Zhou T. , Physica A , 392 (2013) 3402.
- [7] Liben-Nowell D. and Kleinberg J. M. , J. Am. Soc. Inform. Sci. &. Technol. , 58 (2007) 1019.
- [8] L¨ u L. and Zhou T , Europhys. Lett. , 89 (2010) 18001.
- [9] Yang Z., Fu D., Tang Y., Zhang Y., Hao Y., Gui C., Ji X. and Yue X. , Communications in Computer and Information Science 2012 (Springer Berlin Heidelberg Press) 2012, pp. 119.
- [10] De S´ a H. R. and Prudencio R. B. C. , The 2011 International Joint Conference on Neural Networks (IEEE Press) 2011, pp. 2281.
- [11] Kunegis J. and Lommatzsch A. , ICML '09 Proceedings of the 26th Annual International Conference on Machine Learning (ACM Press) 2009, pp. 561.
- [12] Murata T. and Moriyasu S. , Web Intelligence, IEEE/WIC/ACM International Conference on. IEEE (IEEE Press) 2007, pp. 85.
- [13] Wind D. K and Morup M. , The 2012 IEEE International Workshop on Machine Learning for Signal Processing (IEEE Press) 2012, pp. 1.
- [14] Taha H. A. , Operations Research: An Introduction (Prentice-Hall, Inc.) 2006.
- [15] Golub G. H. and Van Loan C. F. , Matrix Computations (Johns Hopkins University Press) 2013, p. 756.
- [16] Zhou T., L¨, L. u and Zhang Y.-C , Eur. Phys. J. B. , 71 (2009) 623.
- [17] L¨ u, L., Jin C. H., and Zhou T. , Phys. Rev. E , 80 (2009) 046122.
- [18] Adamic L. A. and Adar E. , Soc. Networks , 25 (2003) 211.
- [19] Newman M. E. J. , Phys. Rev. E , 64 (2001) 016132.
- [20] Chua H. N., Sung W. K., and Wong L. , Bioinformatics , 22 (2006) 1623.
- [21] Zhou T., Ren J., Medo M., and Zhang Y. C. , Phys. Rev. E , 76 (2007) 046115.
- [22] Havugimana P. C., Hart G. T., Nepusz T., Yang H., Turinsky A. L., Li Z., Wang P. I., Boutz D. R., Fong V., Phanse S., Babu M.,Craig S. A., Hu P., Wan C., Vlasblom J., Dar V., Bezginov A.,Clark G. W.,Wu G. C., Wodak S. J., Tillier E. R. M., Paccanaroemail A. and Emili A. , Cell , 150 (2012) 1068.
- [23] Varshney L. R., Chen B. L., Paniagua E., Hall D. H. and Chklovskii D. B. , PLOS Comput. Biol. , 7 (2011) e1001066.
- [24] Jensen L. J., Kuhn M., Stark M., Chaffron S., Creevey C., Muller J., Doerks T., Julien P., Roth A., Simonovic M., Bork P. and von Mering C. , Nucleic Acids Res. , 37 (2009) D412.
- [25] Ruepp A., Brauner B., Dunger-Kaltenbach I., Frishman G., Montrone C., Stransky M., Waegele B., Schmidt T., Doudieu O. N., Stmpflen V. and Mewes W. , Nucleic Acids Res. , 36 (2008) D646.
- [26] Watts D. J. and Strogatz S. H. , Nature , 393 (1998) 440.
- [27] Holme P., Park S. M., Kim B. J. and Edling C. R , Physica A , 373 (2007) 821.
|
10.1038/srep12261
|
[
"Jing Zhao",
"Lili Miao",
"Haiyang Fang",
"Qian-Ming Zhang",
"Min Nie",
"Tao Zhou"
] |
2014-08-05T01:07:41+00:00
|
2014-08-05T01:07:41+00:00
|
[
"cs.SI",
"cs.IR",
"physics.soc-ph"
] |
Predicting missing links and their weights via reliable-route-based method
|
Link prediction aims to uncover missing links or predict the emergence of
future relationships according to the current networks structure. Plenty of
algorithms have been developed for link prediction in unweighted networks, with
only a very few of them having been extended to weighted networks. Thus far,
how to predict weights of links is important but rarely studied. In this
Letter, we present a reliable-route-based method to extend unweighted local
similarity indices to weighted indices and propose a method to predict both the
link existence and link weights accordingly. Experiments on different real
networks suggest that the weighted resource allocation index has the best
performance to predict the existence of links, while the reliable-route-based
weighted resource allocation index performs noticeably better on weight
prediction. Further analysis shows a strong correlation for both link
prediction and weight prediction: the larger the clustering coefficient, the
higher the prediction accuracy.
|
1408.0846v3
|
## A Brooks-type result for sparse critical graphs
Alexandr Kostochka ∗
Matthew Yancey †
April 5, 2017
## Abstract
A graph G is k -critical if it has chromatic number k , but every proper subgraph of G is ( k -1)-colorable. Let f k ( n ) denote the minimum number of edges in an n -vertex k -critical graph. Recently the authors gave a lower bound, f k ( n ) ≥ ⌈ ( k +1)( k -2) | V ( G ) |-k k ( -3) 2( k -1) ⌉ , that solves a conjecture by Gallai from 1963 and is sharp for every n ≡ 1 (mod k -1). It is also sharp for k = 4 and every n ≥ 6. In this paper we refine the result by describing all n -vertex k -critical graphs G with | E G ( ) | = ( k +1)( k -2) | V ( G ) |-k k ( -3) 2( k -1) . In particular, this result implies exact values of f 5 ( n ) for n ≥ 7.
Mathematics Subject Classification : 05C15, 05C35
Key words and phrases : graph coloring, k -critical graphs, Brooks' Theorem.
## 1 Introduction
glyph[negationslash]
A proper k -coloring , or simply k -coloring , of a graph G = ( V, E ) is a function f : V →{ 1 2 , , . . . , k } such that for each uv ∈ E f u , ( ) = f ( v ). A graph G is k -colorable if there exists a k -coloring of G . The chromatic number , χ G ( ), of a graph G is the smallest k such that G is k -colorable. A graph G is k -chromatic if χ G ( ) = k .
A graph G is k -critical if G is k -chromatic, but every proper subgraph of G is ( k -1)-colorable. Critical graphs were first defined and used by Dirac [7, 8, 9] in 1951-52. A reason to study k -critical graphs is that every k -chromatic graph contains a k -critical subgraph and k -critical graphs have more restricted structure. For example, k -critical graphs are 2-connected and ( k -1)-edgeconnected.
One of the basic questions on k -critical graphs is: What is the minimum number f k ( n ) of edges in a k -critical graph with n vertices? This question was first asked by Dirac [12] in 1957 and then was reiterated by Gallai [17] in 1963, Ore [29] in 1967 and others [21, 22, 34]. Gallai [17] has found the values of f k ( n ) for n ≤ 2 k -1.
∗ University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA and Sobolev Institute of Mathematics, Novosibirsk 630090, Russia. Email: [email protected]. Research of this author is supported in part by NSF grants DMS-1266016 and DMS-1600592 and by grants 15-01-05867 and 16-01-00499 of the Russian Foundation for Basic Research.
† Department of Mathematics, University of Illinois, Urbana, IL 61801, USA. E-mail: [email protected]. Research of this author is partially supported by the Arnold O. Beckman Research Award of the University of Illinois at Urbana-Champaign.
Theorem 1 (Gallai [17]) If k ≥ 4 and k +2 ≤ n ≤ 2 k -1 , then
$$f _ { k } ( n ) = \frac { 1 } { 2 } \left ( ( k - 1 ) n + ( n - k ) ( 2 k - n ) \right ) - 1.$$
Kostochka and Stiebitz [24] found the value f k (2 k ) = k 2 -3. Gallai [16] also conjectured the exact value for f k ( n ) for n ≡ 1 ( mod k -1).
Conjecture 2 (Gallai [16]) If k ≥ 4 and n ≡ 1 ( mod k -1) , then
$$f _ { k } ( n ) = \frac { ( k + 1 ) ( k - 2 ) n - k ( k - 3 ) } { 2 ( k - 1 ) }.$$
The upper bound on f k ( n ) follows from Gallai's construction of k -critical graphs with only one vertex of degree at least k . So the main difficulty of the conjecture is in proving the lower bound on f k .
For a graph G and a vertex u ∈ V ( G ), a split of u is a construction of a new graph G ′ such that V ( G ′ ) = V ( G ) -u + { u , u ′ ′′ } , where G -u ∼ = G ′ - { u , u ′ ′′ } , N u ( ′ ) ∪ N u ( ′′ ) = N u ( ), and N u ( ′ ) ∩ N u ( ′′ ) = ∅ . A DHGO-composition O G ,G ( 1 2 ) of graphs G 1 and G 2 is a graph obtained as follows: delete some edge xy from G 1 , split some vertex z of G 2 into two vertices z 1 and z 2 of positive degree, and identify x with z 1 and y with z 2 . Note that DHGO-composition could be found in Dirac's paper [13] and has roots in [10]. It was also used by Gallai [16] and Haj´s [19]. o Ore [29] used it for a composition of complete graphs.
The mentioned authors observed that if G 1 and G 2 are k -critical and G 2 is not k -critical after z has been split, then O G ,G ( 1 2 ) also is k -critical. This observation implies
$$f _ { k } ( n + k - 1 ) \leq f _ { k } ( n ) + \frac { ( k + 1 ) ( k - 2 ) } { 2 } = f _ { k } ( n ) + ( k - 1 ) \frac { ( k + 1 ) ( k - 2 ) } { 2 ( k - 1 ) }. \quad \ \ ( 1 )$$
Ore believed that this construction starting from an extremal graph on at most 2 k vertices with G 2 = K k at each iteration yields sparsest k -critical n -vertex graphs for all n ≥ 2 k .
glyph[negationslash]
Conjecture 3 (Ore [29]) If k ≥ 4 , n ≥ k and n = k +1 , then f k ( n + k -1) = f k ( n ) +( k -2)( k + 1) / . 2
Note that Conjecture 2 is equivalent to the case n ≡ 1 ( mod k -1) of Conjecture 3.
Some lower bounds on f k ( n ) were obtained in [12, 28, 16, 24, 25, 15]. Recently, the authors [26] proved Conjecture 2 valid.
glyph[negationslash]
Theorem 4 ([26]) If k ≥ 4 and G is k -critical, then | E G ( ) | ≥ ⌈ ( k +1)( k -2) | V ( G ) |-k k ( -3) 2( k -1) ⌉ . In other words, if k ≥ 4 and n ≥ k, n = k +1 , then
$$f _ { k } ( n ) \geq F ( k, n ) \coloneqq \left \lceil \frac { ( k + 1 ) ( k - 2 ) n - k ( k - 3 ) } { 2 ( k - 1 ) } \right \rceil.$$
The result also confirms Conjecture 3 in several cases.
Corollary 5 ([26]) Conjecture 3 is true if (i) k = 4 , (ii) k = 5 and n ≡ 2 ( mod 4) , or (iii) n ≡ 1 ( mod k -1) .
Some applications of Theorem 4 are given in [26] and [5]. In [27], the authors derive from a partial case of Theorem 4 a half-page proof of the well-known Gr¨tzsch Theorem [18] that every o planar triangle-free graph is 3-colorable. Conjecture 3 is still open in general. By examining known values of f k ( n ) when n ≤ 2 k , it follows that f k ( n ) -F k, n ( ) ≤ k / 2 8.
The goal of this paper is to describe the k -extremal graphs, i.e. the k -critical graphs G such that | E G ( ) | = ( k +1)( k -2) | V ( G ) |-k k ( -3) 2( k -1) . This is a refinement of Conjecture 2: For n ≡ 1 ( mod k -1), we describe all n -vertex k -critical graphs G with | E G ( ) | = f k ( n ). This is also the next step towards the full solution of Conjecture 3.
By definition, if G is k -extremal, then ( k +1)( k -2) | V ( G ) |-k k ( -3) 2( k -1) is an integer, and so | V ( G ) | ≡ 1 ( mod k -1). For example, K k is k -extremal.
Suppose that G 1 and G 2 are k -extremal and G = O G ,G ( 1 2 ). Then
$$| E ( G ) | = | E ( G _ { 1 } ) | + | E ( G _ { 2 } ) | - 1 = \frac { ( k + 1 ) ( k - 2 ) ( | V ( G _ { 1 } ) | + | V ( G _ { 2 } ) | ) - 2 k ( k - 3 ) } { 2 ( k - 1 ) } - 1$$
$$| E ( G ) | = | E ( G _ { 1 } ) | + | E ( G _ { 2 } ) | - 1 = \frac { ( k + 1 ) ( k - 2 ) ( | V ( G _ { 1 } ) | + | V ( G _ { 2 } ) | ) - 2 i } { 2 ( k - 1 ) } \\ = \frac { ( k + 1 ) ( k - 2 ) | V ( G ) | - k ( k - 3 ) } { 2 ( k - 1 ) }. \\ \cdot \widehat { \ } \quad. \ \dots \ \sim \ \dots \ \sim \ \dots \ \sim \ \dots \ \dots \.$$
After z is split, G 2 will still have F k, V ( | ( G 2 ) ) | < F k, V ( | ( G 2 ) | +1) edges, and therefore will not be k -critical. Thus the DHGO-composition of any two k -extremal graphs is again k -extremal.
A graph is a k -Ore graph if it is obtained from a set of copies of K k by a sequence of DHGOcompositions. By the above, every k -Ore graph is k -extremal. This yields an explicit construction of infinitely many k -extremal graphs.
The main result of the present paper is the following.
Theorem 6 Let k ≥ 4 and G be a k -critical graph. Then G is k -extremal if and only if it is a k -Ore graph. Moreover, if G is not a k -Ore graph, then | E G ( ) | ≥ ( k +1)( k -2) | V ( G ) |-y k 2( k -1) , where y k = max 2 { k -6 , k 2 -5 k +2 } . Thus y 4 = 2 , y 5 = 4 , and y k = k 2 -5 k +2 for k ≥ 6 .
The message of Theorem 6 is that although for every k ≥ 4 there are infinitely many k -extremal graphs, they all have a simple structure. In particular, every k -extremal graph distinct from K k has a separating set of size 2. The theorem gives a slightly better approximation for f k ( n ) and adds new cases for which we now know the exact values of f k ( n ):
Corollary 7 Conjecture 3 holds and the value of f k ( n ) is known if (i) k ∈ { 4 5 , } , (ii) k = 6 and n ≡ 0 ( mod 5) , (iii) k = 6 and n ≡ 2 ( mod 5) , (iv) k = 7 and n ≡ 2 ( mod 6) , or (v) k ≥ 4 and n ≡ 1 ( mod k -1) .
This value of y k in Theorem 6 is best possible in the sense that for every k ≥ 4, there exist infinitely many 3-connected graphs G with | E G ( ) | = ( k +1)( k -2) | V ( G ) |-y k 2( k -1) . The idea of this construction (Construction 58) and the examples for k = 4 5 are due to Toft ([33], based on [32]). , Construction 60 produces the examples for k ≥ 6.
Theorem 6 has already found interesting applications. In [3], it was used to describe the 4-critical planar graphs with exactly 4 triangles. This problem was studied by Axenov [1] in the seventies, and then mentioned by Steinberg [31] (quoting Erd˝s from 1990), and Borodin [2]. o It was proved in [3] that the 4-critical planar graphs with exactly 4 triangles and no 4-faces are exactly the 4-Ore graphs with exactly 4 triangles. Also, Kierstead and Rabern [23] and independently Postle [30]
have used Theorem 6 to describe the infinite family of 4-critical graphs G with the property that for each edge xy ∈ E G ( ), d x ( ) + d y ( ) ≤ 7. It turned out that such graphs form a subfamily of the family of 4-Ore graphs.
Our proofs will use the language of potentials .
Definition 8 Let G be a graph. For R ⊆ V ( G ) , define the k -potential of R to be
$$\rho _ { k, G } ( R ) = ( k + 1 ) ( k - 2 ) | R | - 2 ( k - 1 ) | E ( G [ R ] ) |.$$
When there is no chance for confusion, we will use ρ k ( R ) . Let P k ( G ) = min ∅glyph[negationslash] = R ⊆ V ( G ) ρ k ( R ) .
Informally, ρ k,G ( R ) measures how many edges are needed to be added to G R [ ] (or removed, if the potential is negative) in order to obtain a graph with average degree ( k +1)( k -2) k -1 . Our proofs below will involve adding and deleting edges and vertices, so using the language of potentials helps keep track of whether or not the resulting graph maintains the assumptions of the theorem.
Translated into the language of potentials, Theorem 4 sounds as follows.
Corollary 9 ([26]) If G is k -critical then ρ k ( V ( G )) ≤ k k ( -3) . In particular, if a graph H satisfies ρ k,H ( S ) > k k ( -3) for all nonempty S ⊆ V ( H ) , then H is ( k -1) -colorable.
Similarly, our main result, Theorem 6, is:
Theorem 10 If G is k -critical and not a k -Ore graph, then
$$\rho _ { k } ( V ( G ) ) \leq y _ { k },$$
where y k = max 2 { k -6 , k 2 -5 k +2 } . In particular, if a graph H does not contain a k -Ore graph as a subgraph and ρ k,H ( S ) > y k for all nonempty S ⊆ V ( H ) , then H is ( k -1) -colorable.
Our strategy of the proof (similar to those in [4, 6, 26, 27]) is to consider a minimum counterexample J to Theorem 10 and derive a set of its properties leading to a contradiction. Quite useful claims will be that all nontrivial proper subsets of V ( J ) have 'high' potentials. Important examples of such claims are Claim 27 and Lemma 38 below. This will help us to provide ( k -1)-colorings of subgraphs of J with additional properties. For example, Claim 27 will imply Claim 28 that adding any edge to a subgraph H of J with 1 < V H | ( ) | < V J | ( ) | leaves the subgraph ( k -1)-colorable. Important new ingredient of the proof is the study in the next section of the properties of k -Ore graphs and their colorings. In Section 3 we prove basic properties of our minimum counter-example J , including Claim 27 mentioned above. Then in Section 4 we introduce and study properties of clusters - sets of vertices of degree k -1 in J with the same closed neighborhood. This will allow us to prove Lemma 38. Based on this lemma and its corollaries, we prove Theorem 10 in Section 5 using some variations of discharging; the cases of small k will need separate considerations. In Section 6 we discuss the sharpness of our result and in Section 7 - some algorithmic aspects of it.
## 2 Potentials and Ore graphs
The fact below summarizes useful properties of ρ k and y k following directly from the definitions or Corollary 9.
## Fact 11 For the k -potential defined by (2), we have
- 1. Potential is submodular:
$$\rho _ { k } ( X \cap Y ) + \rho _ { k } ( X \cup Y ) = \rho _ { k } ( X ) + \rho _ { k } ( Y ) - 2 ( k - 1 ) | E _ { G } [ X - Y, Y - X ] |. \quad \ \ ( 3 )$$
- 2. ρ k ( V ( K 1 )) = ( k +1)( k -2) .
- 3. ρ k ( V ( K 2 )) = 2( k 2 -2 k -1) .
- 4. ρ k ( V ( K k -1 )) = 2( k -2)( k -1) .
- 5. ρ k ( V ( K k )) = k k ( -3) .
- 6. If k ≥ 4 , then ρ k ( V ( K k )) ≤ ρ k ( V ( K 1 )) ≤ ρ k ( V ( K k -1 )) ≤ ρ k ( V ( K 2 )) ≤ ρ k ( V ( K i )) for all 3 ≤ i ≤ k -2 .
- 7. For any vertex set S , ρ k ( S ) ≥ ρ k ( K | S | ) . In particular, if 1 ≤ | S | ≤ k -1 , then ρ k ( S ) ≥ ( k + 1)( k -2) . If 2 ≤ | S | ≤ k -1 , then ρ k,G ( S ) ≥ 2( k -2)( k -1) with equality only if | S | = k -1 and G S [ ] = K k -1 .
- 8. k k ( -3) ≤ y k +2 k -2 < k ( +1)( k -2) .
- 9. ρ k ( A ) is even for each k and A .
- 10. If G is a graph with a spanning subgraph H that is a k -Ore graph, then ρ k,G ( V ( G )) ≤ k k ( -3) . If H = G , then we have equality. If H is a proper subgraph of G , then ρ k,G ( V ( G )) ≤ y k .
A common technique in constructing critical graphs (see [21, 31]) is to use quasi-edges and quasi-vertices . For k ≥ 3, a graph G , and x, y ∈ V ( G ), a k -quasixy -edge Q k ( x, y ) is a subset Q of V ( G ) such that x, y ∈ Q and
- (Q1) G Q [ ] has a ( k -1)-coloring,
glyph[negationslash]
- (Q2) φ x ( ) = φ y ( ) for every proper ( k -1)-coloring of G Q [ ], and
- (Q3) for any edge e ∈ G Q [ ], G Q [ ] -e has a ( k -1)-coloring φ such that φ x ( ) = φ y ( ).
Symmetrically, a k -quasixy -vertex Q ′ k ( x, y ) is a subset Q ′ of V ( G ) such that x, y ∈ Q ′ and
- (Q 1) ′ G Q [ ′ ] has a ( k -1)-coloring,
- (Q 2) ′ φ x ( ) = φ y ( ) for every proper ( k -1)-coloring of G Q [ ′ ], and
glyph[negationslash]
- (Q 3) for any edge ′ e ∈ G Q [ ′ ], G Q [ ′ ] -e has a ( k -1)-coloring φ such that φ x ( ) = φ y ( ).
If G is a k -critical graph, then for each e = xy ∈ E G ( ), graph G -e is a k -quasixy -vertex. On the other hand, given some k -quasi-vertices and k -quasi-edges, one can construct from copies of them infinitely many k -critical graphs. In particular, the DHGO-composition can be viewed in this way. The next observation is well known and almost trivial, but we state it, because we use it often.
Fact 12 Let k ≥ 4 . If a k -critical graph G has a separating set { x, y } , then
- (1) G -{ x, y } has exactly two components, say with vertex sets A ′ and B ′ ;
- (2) xy / E G ∈ ( ) ;
- (3) one of A ′ ∪ { x, y } and B ′ ∪ { x, y } is a k -quasixy -edge and the other is a k -quasixy -vertex.
A quasi-edge and a quasi-vertex are very related structures, as seen by the following construction.
Fact 13 If Q k ( x, z ) is a k -quasixz -vertex and Q x,y ′ ( ) is obtained from Q k ( x, z ) by appending a leaf y that is adjacent only to z , then Q x,y ′ ( ) is a k -quasixy -edge. If Q ′ k ( x, y ) is a k -quasixy -edge and N y ( ) ∩ Q ′ k ( x, y ) = { z } , then the vertex set Q k ( x, z ) = Q ′ k ( x, y ) -y is a k -quasixz -vertex.
Fact 12 together with the definition of k -Ore graphs, implies the following.
glyph[negationslash]
Fact 14 Every k -Ore graph G = K k has a separating set { x, y } and two vertex subsets A = A G,x,y ( ) and B = B G,x,y ( ) such that
- (i) A ∩ B = { x, y } , A ∪ B = V ( G ) and no edge of G connects A -x -y with B -x -y ,
- (ii) the graph ˜ G x, y ( ) obtained from G A [ ] by adding edge xy is a k -Ore graph,
- (iii) the graph ˇ ( G x,y ) obtained from G B [ ] by gluing x with y into a new vertex x ∗ y is a k -Ore graph, and
- (iv) xy / E G ∈ ( ) .
In terms of Fact 14, G is the DHGO-composition of ˜ G x,y ( ) and ˇ ( G x,y ). We will repeatedly use the notation in this fact. The next fact directly follows from the definitions.
## Fact 15 Using the notation in Fact 14, we have
- 1. A is a k -quasixy -vertex;
- 2. B is a k -quasixy -edge;
- 3. ρ k,G ( A ) = ρ k,K 1 ( V ( K 1 )) = ( k +1)( k -2) ;
- 4. ρ k,G ( B ) = ρ k,K 2 ( V ( K 2 )) = 2( k 2 -2 k -1) ;
- 5. N x ( ) ∩ B ∩ N y ( ) = ∅ ;
- 6. N ˜ G ( v ) = N G ( v ) for each v ∈ A -x -y ;
- 7. d ˇ G ( v ) = d G ( v ) for each v ∈ B -x -y ;
- 8. · If R ⊆ A -x or R ⊆ A -y , then ρ k,G ( R ) = ρ k,G ˜ ( R ) .
- · If R ⊆ B -{ x, y } , then ρ k,G ( R ) = ρ k,G ˇ ( R ) .
- · If R ⊆ B -x or R ⊆ B -y and ˇ = R R -{ x, y } + x ∗ y , then ρ k,G ( R ) ≥ ρ k,G ˇ ( ˇ ) R .
- · If { x, y } ⊆ R ⊆ B and ˇ = R R -{ x, y } + x ∗ y , then ρ k,G ( R ) = ρ k,G ˇ ( ˇ ) + ( R k +1)( k -2) .
- 9. ρ k,G ( V ( G )) = ρ k,V ( ˜ G ) ( V ( ˜ G )) = ρ k,V ( ˇ ) G ( V ( ˇ )) = G k k ( -3) .
Claim 16 For every k -Ore graph G and each ∅ glyph[negationslash] = R glyph[subsetnoteql] V ( G ) , we have ρ k,G ( R ) ≥ ( k +1)( k -2) .
Proof. Let G be a smallest counter-example to the claim and let
$$R \subsetneq V ( G ) \text{ be a smallest nonempty proper subset of } V ( G ) \text{ with } \rho _ { k, G } ( R ) < ( k + 1 ) ( k - 2 ). \quad ( 4 )$$
glyph[negationslash]
By Fact 11.7, G = K k . So, by Fact 14 there is a separating set { x, y } and two vertex subsets A = A G,x,y ( ) and B = B G,x,y ( ) as in Fact 14. By the minimality of G , every proper subset of V ( ˜ G x, y ( )) and of V ( ˇ ( G x,y )) has potential at least ( k +1)( k -2). If G R [ ] were disconnected, then the vertex set of some component of G R [ ] would also have potential less than ( k +1)( k -2), contradicting the minimality of R . So, G R [ ] is connected. Since ρ k,G ( R < k ) ( +1)( k -2) by (4) and R is non-empty, by Fact 11.7, | R | ≥ k .
Case 1: { x, y } ∩ R = ∅ . Since G R [ ] is connected, R is a non-empty proper subset either of A or B . This contradicts Fact 15 and the minimality of G .
Case 2: { x, y } ∩ R = { x } . The set R ∩ A induces a non-empty connected subgraph of G , and so by the minimality of | R | , ρ k,G ( R ∩ A ) ≥ ( k +1)( k -2). Similarly, ρ k,G ( R ∩ B ) ≥ ( k +1)( k -2). By Fact 11.1,
$$\rho _ { k, G } ( R ) = \rho _ { k, G } ( R \cap A ) + \rho _ { k, G } ( R \cap B ) - \rho _ { k, G } ( \{ x \} ) \geq ( k + 1 ) ( k - 2 ),$$
a contradiction to (4).
Case 3: { x, y } ⊆ R . If A ⊆ R , then by Facts 11.1 and 15.3,
$$\rho _ { k, \check { G } ( x, y ) } ( ( R - A ) + x * y ) = \rho _ { k, G } ( R ) - \rho _ { k, G } ( A ) + \rho _ { k, \check { G } ( x, y ) } ( \{ x * y \} ) = \rho _ { k, G } ( R ).$$
But by the minimality of G , this is at least ( k + 1)( k -2), a contradiction to (4). Similarly, if B ⊆ R , then
$$\rho _ { k, \tilde { G } ( x, y ) } ( R \cap A ) = \rho _ { k, G } ( R ) - \rho _ { k, G } ( B ) + \rho _ { k, \tilde { G } ( x, y ) } ( \{ x, y \} ) = \rho _ { k, G } ( R ),$$
glyph[negationslash]
a contradiction to (4) again. So, suppose A -R = ∅ and B -R = ∅ . By the minimality of G , we have ρ k,G x,y ˜ ( ) ( R ∩ A ) ≥ ( k + 1)( k -2). Since xy is an edge in ˜ G x,y ( ) but not in G , this yields ρ k,G ( R ∩ A ) ≥ ( k +1)( k -2) + 2( k -1). Similarly, ρ k,G x,y ˇ ( ) (( R -A ) + x ∗ y ) ≥ ( k +1)( k -2) and thus ρ k,G ( R ∩ B ) ≥ 2( k +1)( k -2). Then glyph[negationslash]
$$\rho _ { k, G } ( R ) = \rho _ { k, G } ( R \cap A ) + \rho _ { k, G } ( R \cap B ) - 2 \rho _ { k, G } ( K _ { 1 } ) \geq ( k + 1 ) ( k - 2 ) + 2 ( k - 1 ),$$
a contradiction to (4). glyph[square]
glyph[negationslash]
Definition 17 For a set U of vertices in a graph G , the border of U is U ∗ = { w ∈ U : N w ( ) ⊆ U } , i.e. the set of vertices in U that have neighbors outside of U .
glyph[negationslash]
If U ∗ is the border of U ⊂ V ( G ) and U ∗ = U , then U ∗ is a separating set in G .
A set S of vertices in a graph G is standard , if
- (a) ρ k,G ( S ) = ( k +1)( k -2) and
- (b) the border of S is a 2-element set { x, y } such that G S [ -{ x, y } ] is connected, and
- (c) S is a k -quasi-{ x, y }
- -vertex.
Note that a standard set is a k -quasi-vertex whose k -potential is the same as that of a vertex.
Lemma 18 Let G be a k -Ore graph. Let W ⊂ V ( G ) with | W | ≥ 2 and ρ k ( W ) ≤ ( k +1)( k -2) . Then G W [ ] is connected and contains a standard set.
Proof. Suppose ρ k ( W ) ≤ ( k +1)( k -2). Then by Claim 16, ρ k ( W ) = ( k +1)( k -2). If G W [ ] is disconnected, say W = W 1 ∪ W 2 with W 1 ∩ W 2 = ∅ and no edges between W 1 and W 2 , then ρ k,G ( W 1 ) + ρ k,G ( W 2 ) = ρ k,G ( W ). This implies min { ρ k,G ( W ,ρ 1 ) k,G ( W 2 ) } ≤ ( k +1)( k -2) / 2, which contradicts Claim 16. So G W [ ] is connected, i.e., the first part of the lemma holds.
To prove the second part, choose a counter-example G with the fewest vertices and let W ⊆ V ( G ) be a smallest subset of V ( G ) such that
$$| W | \geq 2, \, \rho _ { k, G } ( W ) = ( k + 1 ) ( k - 2 ), \, \text{and} \, W \, \text{does not contain a standard set}.$$
glyph[negationslash]
By Fact 11.7, the graph K k simply does not have sets W satisfying (5). So G = K k and thus by Fact 14 has a separating set { x, y } . Let A B , , ˜ G x,y ( ), and ˇ ( G x, y ) be defined as in Fact 14. First we show that
$$G [ W ] \text{ is 2-connected.}$$
Indeed, suppose not. Then by the first part of the lemma, G W [ ] has a cut vertex, say z . Let W 1 and W 2 be two subsets of W such that W 1 ∩ W 2 = { z } , W 1 ∪ W 2 = W and there are no edges between W 1 -z and W 2 -z . Then by Fact 11.1 and Fact 11.2,
$$\rho _ { k, G } ( W _ { 1 } ) + \rho _ { k, G } ( W _ { 2 } ) = \rho _ { k, G } ( W ) + \rho _ { k, G } ( \{ z \} ) = 2 ( k + 1 ) ( k - 2 ).$$
So by Claim 16, ρ k,G ( W 1 ) = ρ k,G ( W 2 ) = ( k +1)( k -2). Thus by the minimality of W , each of W 1 and W 2 contains a standard subset, a contradiction to (5). This proves (6).
glyph[negationslash]
Let W A = A ∩ W and W B = B ∩ W . Suppose S ⊆ W A , ρ k,G ( S ) = ( k +1)( k -2) (which implies S = ∅ ), and y / S ∈ . Because (a) by Fact 15.8, S has the same potential in ˜ G as in G , (b) by Fact 14.ii, ˜ G is also k -Ore, and (c) G is a minimal counterexample,
$$S \text{ contains a standard set } W ^ { \prime } \text{ in } \widetilde { G } ( x, y ).$$
We will use (7) in Cases 1 and 3 below.
Case 1: W ⊆ A . If { x, y } ⊆ W , then ρ k,G x,y ˜ ( ) ( W ) = ρ k,G ( W ) -2( k -1) = k k ( -3), which by Claim 16 means that W = A . But A is a standard set, a contradiction to the choice of W . So by symmetry, we may assume that y / W ∈ . By (7) with S = W , the set W contains a standard set W ′ in ˜ G x,y ( ). Since y / W ∈ ′ , by Fact 15.8, W ′ has the same potential in G as in ˜ G . So by the minimality of W W , ′ = W . Furthermore, if x ∈ W ′ , then it is in the border of W ′ in ˜ G because y / W ∈ = W ′ and xy ∈ E G ( ˜ ). By this and Fact 15.6, we conclude that the border of W ′ in ˜ G coincides with the border of W ′ = W in G . So W ′ = W is also a standard set in G with the same border, a contradiction to (5).
Case 2: W ⊆ B . Let W ′ B = W -{ x, y } + x ∗ y . If { x, y } ⊆ W , then by Fact 15.8 ρ k,G x,y ˇ ( ) ( W ′ B ) = ρ k,G ( W ) -( k +1)( k -2) = 0, which contradicts Claim 16.
Suppose now that { x, y }∩ W = . Then ∅ W ⊂ V ( ˇ ( G x,y )). By the minimality of G W , contains a standard set W ′ in ˇ ( G x,y ), and W ′ does not contain x ∗ y . As in Case 1, by Fact 15.8, W ′ has the same potential in G as in ˇ , G and by Fact 15.7, W ′ has the same border in G as in ˇ . G So W ′ is also a standard set in G with the same border, a contradiction to (5). Thus by the symmetry between x and y we may assume { x, y } ∩ W = { x } . Let i denote the number of neighbors of y in W .
Case 2A: i = 0. Then ρ k,G x,y ˇ ( ) ( W ′ B ) = ρ k,G ( W ) = ( k +1)( k -2), and by the minimality of G W , ′ B contains a standard set W ′ in ˇ ( G x,y ). If x ∗ y / W ∈ ′ , then W ′ is standard in G exactly as in the previous paragraph. So assume x ∗ y ∈ W ′ . By the case, y has no neighbors in W ′ , but it does have a neighbor in B . This means, x ∗ y is a border vertex of W ′ in ˇ ( G x,y ). But then the set W ′′ := W ′ -x ∗ y + x is a standard set in G with the border W ′′ ∗ = W ′ ∗ -x ∗ y + , a contradiction x to (5).
Case 2B: i ≥ 1. By Fact 15.5, we can calculate the potential of W ′ B in ˇ ( G x, y ) exactly:
$$\rho _ { k, \zeta ( x, y ) } ( W _ { B } ^ { \prime } ) = \rho _ { k, G } ( W ) - 2 i ( k - 1 ) = k ( k - 3 ) - ( i - 1 ) 2 ( k - 1 ).$$
By Claim 16 and the definition of Ore-graphs, this yields that W ′ B = V ( ˇ ( G x, y )) and i = 1. It follows that W = B -y , and y has exactly one neighbor, say z in W . But now we are exactly in the situation described by Fact 13 (as B is a k -quasixy -edge by Fact 15.2), and so W is a k -quasixz -vertex, and therefore a standard set, a contradiction to (5).
glyph[negationslash]
Case 3: W -A = ∅ and W -B = ∅ . Since { x, y } separates A -{ x, y } from B -{ x, y } in G , the set { x, y } ∩ W is a separating set in G W [ ]. Thus by (6), { x, y } ⊆ W .
glyph[negationslash]
Case 3A: W A = A (equivalently, A ⊆ W ). We are done, since A is standard by Fact 15.1, Fact 15.3, and the definition of a standard set.
Case 3B: W B = B (equivalently, B ⊆ W ). Since ρ k,G ( B ) = 2( k 2 -2 k -1) by Fact 15.4 and x, y have no common neighbors in B by 15.5, we have
$$\rho _ { k, \tilde { G } ( x, y ) } ( W _ { A } ) = \rho _ { k, G } ( W ) - \rho _ { k, G } ( B ) + \rho _ { k, \tilde { G } ( x, y ) } ( \{ x, y \} )$$
$$= ( k + 1 ) ( k - 2 ) - 2 ( k ^ { 2 } - 2 k - 1 ) + 2 ( k + 1 ) ( k - 2 ) - 2 ( k - 1 ) = ( k + 1 ) ( k - 2 ).$$
So for S = W A the three conditions (a), (b), (c) that imply (7) hold, since (8) implies (a), and statements (b) and (c) are about G (and hence still hold). Thus by (7), W A contains a standard set W ′ in ˜ G x,y ( ). If | W ′ ∩{ x, y }| ≤ 1, then W ′ is a standard set in G with the same border using the same argument as in Case 1. If { x, y } ⊂ W ′ , then replacing in (8) W A and W with W ′ and W ′ ∪ B respectively, we get ρ k,G ( W ′ ∪ B ) = ρ k,G x,y ˜ ( ) ( W ′ ) = ( k +1)( k -2).
We claim that W ′ ∪ B has the same border in G as W ′ does in ˜ G . By Fact 15.6, the only possible new elements of the border of W ′ in W ′ ∩ A are x or y . But the only new neighbors of x or y are in B , and B ⊂ W ′ ∪ B so neither x nor y can be a new element of the border. And the neighbors of each v ∈ B -x -y are in B , so none of the vertices in B -x -y is in the border. This means that W ′ ∪ B ⊆ W is a standard set in G .
glyph[negationslash]
Case 3C: W A = A and W B = B . Similarly to Case 2, let W ′ B = W B -{ x, y } + x ∗ y . By Claim 16 and since each of ˇ and G ˜ G is k -Ore, ρ k,G ˜ ( W A ) ≥ ( k +1)( k -2) and ρ k,G ˇ ( W ′ B ) ≥ ( k +1)( k -2). Recall that { x, y } ⊂ W by (6), so { x, y } = W A ∩ W B , and so ρ k,G ˜ ( W A ) = ρ k,G ( W A ) -2( k -1) and by Fact 15.8, ρ k,G ˇ ( W ′ B ) = ρ k,G ( W B ) -( k +1)( k -2). Therefore glyph[negationslash]
ρ k,G ( W A )+ ρ k,G ( W B ) ≥ ( k +1)( k -2)+2( k -1)+( k +1)( k -2)+( k +1)( k -2) = 3( k +1)( k -2)+2( k -1) .
But since W A ∩ W B = { x, y } and xy / E G ∈ ( ), by Fact 11.1,
$$\rho _ { k, G } ( W _ { A } ) + \rho _ { k, G } ( W _ { B } ) = \rho _ { k, G } ( W ) + \rho _ { k, G } ( \{ x, y \} ) = 3 ( k + 1 ) ( k - 2 ),$$
a contradiction. glyph[square]
Now we will prove two statements on colorings and structure of subgraphs not containing standard sets of k -Ore graphs.
Lemma 19 Let G be a k -Ore graph. Let uv be an edge in G such that
$$\rho _ { k, G - u v } ( W ) > ( k + 1 ) ( k - 2 ) \text{ for every } W \subseteq V ( G - u v ) \text{ with } 2 \leq | W | \leq | V ( G ) | - 1. \quad ( 9 )$$
glyph[negationslash]
Then for each w ∈ V ( G ) -u -v , there is a ( k -1) -coloring φ w of G -uv such that φ w ( w ) = φ w ( u ) = φ w ( v ) .
Proof. We use induction on | V ( G ) . | For G = K k , the statement is evident. Otherwise, let x, y, A, B, G x, y ˜ ( ) and ˇ ( G x,y ) be as in Fact 14. By Fact 15, ρ k,G ( A ) = ( k + 1)( k -2), and thus by (9), uv ∈ E G A ( [ ]).
glyph[negationslash]
Case A: w ∈ A . By the induction assumption, there exists a ( k -1)-coloring φ ′ w of ˜ G x,y ( ) -uv such that φ ′ w ( w ) = φ ′ w ( u ) = φ ′ w ( v ). Since φ ′ w ( x ) = φ ′ w ( y ) and B is a quasixy -edge, this coloring extends to a ( k -1)-coloring of the whole G -uv .
glyph[negationslash]
glyph[negationslash]
Case B: w ∈ B -x -y . Let φ ′ be any ( k -1)-coloring of ˜ G x, y ( ) -uv . By Fact 14(iv), xy / E G ∈ ( ), so uv = xy . By the definition of ˜ G xy , ∈ E G ( ˜ ), so φ ′ ( x ) = φ ′ ( y ). Since ˜ G x, y ( ) is k -critical, φ ′ ( u ) = φ ′ ( v ).
glyph[negationslash]
Case B1: φ ′ ( u ) = φ ′ ( x ). Let G 0 = G B [ ] + xw if xw / E G ∈ ( ) and G 0 = G B [ ] otherwise. Then for each W ⊆ V ( G 0 ),
$$\rho _ { k, G _ { 0 } } ( W ) = \rho _ { k, G } ( W ) \text{ if } \{ x, w \} \subseteq W \text{ and } \rho _ { k, G _ { 0 } } ( W ) \geq \rho _ { k, G } ( W ) - 2 ( k - 1 ) \text{ if } \{ x, w \} \subseteq W. \quad ( 1 0 )$$
Since { u, v } glyph[negationslash]⊆ V ( G 0 ), by (9), ρ k,G ( W > k ) ( +1)( k -2) for each W ⊆ V ( G 0 ) with | W > | 1. This together with (10) imply that
$$\rho _ { k, G _ { 0 } } ( W ) \geq \rho _ { k, G } ( W ) - 2 ( k - 1 ) > ( k + 1 ) ( k - 2 ) - 2 ( k - 1 ) = k ( k - 3 )$$
glyph[negationslash]
glyph[negationslash]
for every W ⊆ V ( G 0 ) with | W > | 1. If | W | = 1, then ρ k,G 0 ( W ) = ( k +1)( k -2) > k k ( -3), and so ρ k,G 0 ( W > k k ) ( -3) for all nonempty W ⊆ V ( G 0 ). By the second part of Corollary 9, this implies that G 0 has a ( k -1)-coloring φ ′′ . Since G B [ ] ⊆ G B 0 [ ], coloring φ ′′ is also a coloring of quasixy -edge G B [ ], which yields φ ′′ ( x ) = φ ′′ ( y ). By Fact 14(i) and because φ ′′ ( x ) = φ ′′ ( y ) and φ ′ ( x ) = φ ′ ( y ), we can rename the colors in φ ′′ so that φ ′′ ( x ) = φ ′ ( x ) and φ ′′ ( y ) = φ ′ ( y ), and obtain a ( k -1)-coloring φ = φ ′ | A ∪ φ ′′ | B . By construction, φ u ( ) = φ x ( ) = φ w ( ).
glyph[negationslash]
glyph[negationslash]
glyph[negationslash]
Case B2: φ ′ ( u ) / ∈ { φ ′ ( x , φ ) ′ ( y ) } and k ≥ 5. Since B is a quasi-edge, G B [ ] has a ( k -1)-coloring φ ′′ of G B [ ] such that φ ′′ ( x ) = φ ′ ( x ) and φ ′′ ( y ) = φ ′ ( y ). If φ ′′ ( w ) ∈ { φ ′′ ( x , φ ) ′′ ( y ) } , then by the assumption of the case, φ = φ ′ | A ∪ φ ′′ | B is a ( k -1)-coloring we are looking for. Otherwise, since k -1 ≥ 4, we can rename the colors of φ ′′ distinct from the colors of x and y so that φ ′′ ( w ) = φ ′ ( u ) and again take φ = φ ′ | A ∪ φ ′′ | B .
glyph[negationslash]
Case B3: φ ′ ( u ) / ∈ { φ ′ ( x , φ ) ′ ( y ) } and k = 4. Let G 0 be obtained from G B [ ] by adding a new vertex z adjacent to x, y and w . Suppose first that G 0 has a 3-coloring φ ′′ . Since G B [ ] ⊂ G 0 and B is a quasixy -edge, φ ′′ ( x ) = φ ′′ ( y ). So the color of z is distinct from φ ′′ ( x ) and φ ′′ ( y ), and thus because there are only 3 colors, φ ′′ ( w ) ∈ { φ ′′ ( x , φ ) ′′ ( y ) } . In this case by renaming the colors in φ ′′ so that φ ′′ ( x ) = φ ′ ( x ) and φ ′′ ( y ) = φ ′ ( y ), we get a required coloring of G . Now suppose that G 0 has no 3-coloring. Then G 0 contains a 4-critical subgraph G 1 . Since G 1 is not a subgraph of G , it follows that z ∈ V ( G 1 ). Since G 1 is 4-critical, δ G ( 1 ) ≥ 4 -1 = 3, and so { x, y, w } ⊂ V ( G 1 ). Let
W = V ( G 1 ). Since ρ 4 ,G 0 ( W ) ≤ 4 by Corollary 9, we have ρ 4 ,G ( W -z ) = ρ 4 ,G ( W ) -10+3(6) ≤ 12. So Fact 11.1 (because G A [ ∩ W ] = G x,y [ { } ] ∼ = 2 K 1 ) implies that ,
$$\rho _ { 4, G } ( A \cup W - z ) \leq \rho _ { 4, G } ( A ) + \rho _ { 4, G } ( W - z ) - 2 \rho _ { 4 } ( K _ { 1 } ) \leq 1 0 + 1 2 - 2 0 = 2.$$
glyph[negationslash]
By Claim 16, this yields that A ∪ W -z either is empty or is V ( G ). But | A ∪ W - | ≥ z 3 because { x, y, w } ⊂ W -z . Also A ∪ W -z = V ( G ), since G is 4-Ore, and the vertex set of each 4-Ore graph has potential k k ( -3) = 4, since it is 4-extremal. glyph[square]
Claim 20 Let G be a k -Ore graph. Let u be a vertex in G such that
$$\rho _ { k, G } ( W ) > ( k + 1 ) ( k - 2 ) \text{ for every } W \subseteq V ( G ) - u \text{ with } | W | \geq 2.$$
Then there exists a ( k -1) -clique S ⊆ V ( G ) -u such that d G ( v ) = k -1 for all v ∈ S and N S ( ) -S is an independent set.
Proof. We use induction on | V ( G ) . | For G = K k , the statement is evident. Otherwise, let x, y, A, B, G x, y ˜ ( ) and ˇ ( G x,y ) be as in Fact 14. Then ρ k,G ( A ) = ( k +1)( k -2), and so u ∈ A .
If there exists a W ⊆ V ( ˇ ( G x, y )) such that | W | ≥ 2 and ρ k,G x,y ˇ ( ) ( W ) ≤ ( k + 1)( k -2), then by (11), x ∗ y ∈ W . So by induction, ˇ has a set G S ⊆ V ( ˇ ( G x,y )) - ∗ x y such that ˇ ( G x, y )[ S ] ∼ = K k -1 , d ˇ ( G x,y ) ( v ) = k -1 for all v ∈ S , and N S ( ) -S is an independent set in ˇ ( G x,y ). Recall that ˇ G -x ∗ y is a subgraph of G . So since u ∈ A , we have S ⊆ V ( G ) -u , G S [ ] ∼ = K k -1 , and N G ( S ) -S is an independent set in G . By Fact 15.7, d G ( v ) = k -1 for all v ∈ S . glyph[square]
## 3 Basic properties of minimal counter-examples
The closed neighborhood of a vertex u in a graph H is N H [ u ] = N H ( u ) ∪ { u } . We will use the following partial order on the set of graphs. A graph H is smaller than a graph G , if either
(S1) | V ( G ) | > V H | ( ) , | or
$$( \mathbf S 2 ) \ | V ( G ) | = | V ( H ) | \text{ and } | E ( G ) | > | E ( H ) |, \text{ or }$$
(S3) | V ( G ) | = | V ( H ) , | | E G ( ) | = | E H ( ) | and G has fewer pairs of adjacent vertices with the same closed neighborhood.
Note that if H is a subgraph of G , then H is smaller than G . Let k ≥ 4 and J be a minimal with respect to relation 'smaller' counter-example to Theorem 10: J is a k -critical graph with ρ k ( V ( J )) > y k that is not k -Ore. Since y k and the values of ρ k are always even, the restriction ρ k ( V ( J )) > y k is equivalent to
$$\rho _ { k } ( V ( J ) ) \geq y _ { k } + 2.$$
Let n := | V ( J ) . | In this section, we derive basic properties of J and its colorings.
Claim 21 J is 3 -connected.
Proof. Suppose that J has a separating set { x, y } and sets A ⊂ V ( J ) and B ⊂ V ( J ) such that A ∩ B = { x, y } , A ∪ B = V ( J ), and no edge of J connects A -x -y with B -x -y . By Fact 12 and the symmetry between A and B , we may assume that A is a k -quasixy -vertex and B is a
k -quasixy -edge. It follows that the graph ˜ J obtained from J A [ ] by inserting edge xy and the graph ˇ obtained from J J B [ ] by gluing x with y are k -critical. Then
$$\rho _ { k } ( V ( J ) ) \leq ( \rho _ { k } ( V ( \widetilde { J } ) ) + 2 & ( k - 1 ) ) + ( \rho _ { k } ( V ( \check { J } ) ) + ( k + 1 ) ( k - 2 ) ) - 2 \cdot ( k + 1 ) ( k - 2 ) \\ & = \rho _ { k } ( V ( \widetilde { J } ) ) + \rho _ { k } ( V ( \check { J } ) ) - k ( k - 3 ).$$
$$( 1 3 )$$
By assumption, y k < ρ k ( V ( J )). By Corollary 9, ρ k,J ˜ ( V ( ˜ J ) ≤ k k ( -3) and ρ k ( V ( ˇ )) J ≤ k k ( -3). Moreover, if ˜ J (respectively, ˇ ) J is not a k -Ore graph, then by the minimality of J , the potential of its vertex set is at most y k . If at least one of ˜ J or ˇ is not J k -Ore, then we get a contradiction with (13). If both are k -Ore, then J is k -Ore, which contradicts the definition of J . glyph[square]
Fact 22 By the definition of ρ k and the assumption ρ k ( V ( J )) > y k , for each v ∈ V ( J ) ,
$$\rho _ { k } ( V ( J ) - v ) = \rho _ { k } ( V ( J ) ) - ( k + 1 ) ( k - 2 ) + 2 ( k - 1 ) d ( v ) >$$
- · y k + k 2 -3 k +4 , if d v ( ) = k -1 ,
- · y k + k 2 -k +2 , if d v ( ) = k ,
- · y k + k 2 + k , if d v ( ) ≥ k +1 .
Because y k ≥ k 2 -5 k +2 , we see that ρ k ( V ( J ) -v ) is also more than
- · 2 k 2 -8 k +6 = 2( k -3)( k -1) , if d v ( ) = k -1 ,
- · 2 k 2 -6 k +4 = 2( k -2)( k -1) , if d v ( ) = k ,
- · 2 k 2 -4 k +2 = 2( k -1) 2 , if d v ( ) ≥ k +1 .
Now we define graph Y ( J, R, φ, X ). The idea of Y ( J, R, φ, X ) is that it is often smaller than J , and every ( k -1)-coloring of it extends to a ( k -1)-coloring of J .
glyph[negationslash]
Definition 23 For a graph G , a set R ⊂ V ( G ) and a ( k -1) -coloring φ : R → [ k -1] of G R [ ] , the graph Y ( G,R,φ,X ) is constructed as follows. Let R ∗ be the border of R , i.e., R ∗ = { v ∈ R : N v ( ) -R = ∅} . Let t be the number of colors used by φ on R ∗ . We may renumber the colors so that the colors used by φ on R ∗ are 1 , . . . , t . First, for i = 1 , . . . , t , let R ′ i denote the set of vertices in V ( G ) -R adjacent in G to at least one vertex v ∈ R with φ v ( ) = i . Now, let Y ( G,R,φ,X ) be obtained from G -R by adding a set X = { x , . . . , x 1 t } of new vertices such that N x ( i ) = R ′ i ∪ { ( x , . . . , x 1 t } -x i ) for i = 1 , . . . , t .
Informally, the definition can be rephrased as follows: For a given R ⊂ V ( G ) and a ( k -1)coloring φ of G R [ ], we glue each color class of φ G R ( [ ]) into a single vertex, then add all possible edges between the new vertices (corresponding to the color classes) and then delete those that have no neighbors outside of R . Graph Y ( G,R,φ,X ) will be a helpful gadget for deriving properties of G , since it inherits some structure from G .
First we will prove some properties of Y ( J, R, φ, X ).
Claim 24 Suppose R ⊂ V ( J ) and φ is a ( k -1) -coloring of J R [ ] . Then χ Y ( ( J, R, φ, X )) ≥ k .
glyph[negationslash]
Proof. Let Y = Y ( J, R, φ, X ). Suppose Y has a ( k -1)-coloring φ ′ . By the construction of Y , the colors of all x i in φ ′ are distinct. We can change the names of the colors so that φ ′ ( x i ) = i for 1 ≤ i ≤ t , where t is given in Definition 23. Again by the construction of Y , φ ′ ( u ) = i for each vertex u ∈ R ′ i . Therefore φ | R ∪ φ ′ | V ( J ) -R is a proper coloring of J , a contradiction. glyph[square]
The next claim is a submodularity-type equation that is a direct extension of Fact 11.1.
Claim 25 Let R ⊂ V ( J ) , φ be a ( k -1) -coloring of J R [ ] and Y = Y ( J, R, φ, X ) . Let W ⊆ V ( Y ) . If W ∩ X = { x i 1 , . . . , x i q } , then let R | W denote the set of vertices v ∈ R ∗ such that φ v ( ) ∈ { i 1 , . . . , i q } . Then
$$\rho _ { k, J } ( W - X + R ) = \rho _ { k, Y } ( W ) - \rho _ { k, Y } ( W \cap X ) + \rho _ { k, J } ( R ) - 2 ( k - 1 ) | E _ { J } ( W - X, R - R | _ { W } ) |. \ \ ( 1 4 )$$
Proof. Since ρ k,J ( U ) is a linear combination of the numbers of vertices and edges in J U [ ], it is enough to check that the weight of every vertex and edge of J W [ -X + R ] is accounted exactly once in the RHS of (14) and the weight of every other vertex or edge either does not appear at all or appears once with plus and once with minus. In particular, the weight of every vertex and edge of Y W [ ∩ X ] appears once with plus and once with minus. glyph[square]
By Corollary 9 and Claim 24, Y ( J, R, φ, X ) contains a vertex set with potential at most k k ( -3). In some instances this will not be enough for our purposes, and we will want Y ( J, R, φ, X ) to contain a vertex set with potential at most y k . The next claim helps us with this.
Claim 26 Let R ⊂ V ( J ) , φ be a ( k -1) -coloring of J R [ ] and Y = Y ( J, R, φ, X ) . Then Y contains a k -critical subgraph Y ′ , and so ρ k,Y ( V ( Y ′ )) ≤ k k ( -3) . Furthermore, if | R | ≥ k , then Y is smaller than J and
- (a) either Y ′ is an induced k -Ore subgraph, or
(b) ρ k,Y ( V ( Y ′ )) ≤ y k .
glyph[negationslash]
Moreover, V ( Y ′ ) ∩ X = ∅ .
Proof. By Claim 24, Y has a k -critical subgraph Y ′ . The bound on the potential of V ( Y ′ ) follows from Corollary 9. In order to prove the 'Furthermore' part, observe that if | R | ≥ k , then Y is smaller than J by Rule (S1) in the definition of 'smaller', since φ uses at most k -1 < R | | colors on R . Because Y ′ is a subgraph of Y , Y ′ is smaller or equal to Y , and so by transitivity is smaller than J . Thus, by the minimality of J , either Y ′ is k -Ore or V ( Y ′ ) has potential at most y k . If Y ′ is an induced subgraph and k -Ore, then (a) holds. If Y ′ is not induced, then by Fact 11.10, ′
(b) holds. If Y is not k -Ore, then (b) holds by the minimality of J .
glyph[negationslash]
Since χ Y ( ′ ) > χ J R ( [ ]), Y ′ is not a subgraph of J . So, V ( Y ′ ) ∩ X = ∅ . glyph[square]
Now we will use Y ( J, R, φ, X ) to prove lower bounds on potentials of nontrivial sets.
$$\text{Claim 27} \, \text{If $\emptyset\neq R\subseteq V(J)$, then} \quad \rho _ { k, J } ( R ) \geq \rho _ { k } ( V ( J ) ) + 2 ( k - 1 ) > y _ { k } + 2 ( k - 1 ).$$
glyph[negationslash]
Proof. Let R be a nonempty proper subset of V ( J ) with the smallest potential. Since J is k -critical, J R [ ] has a proper coloring φ : R → [ k -1]. Let Y = Y ( J, R, φ, X ). By Claim 26, Y contains a subset S with potential at most k k ( -3) and S ∩ X = . Let ∅ Z = S -X + R . Because
| X | ≤ k -1, by Fact 11.7, each non-empty subset of X has potential at least ( k +1)( k -2). So by (14),
$$\rho _ { k, J } ( Z ) & \leq \rho _ { k, Y } ( S ) - \rho _ { k, Y } ( S \cap X ) + \rho _ { k, J } ( R ) \\ & \leq k ( k - 3 ) - ( k + 1 ) ( k - 2 ) + \rho _ { k, J } ( R ) = \rho _ { k, J } ( R ) - 2 ( k - 1 ).$$
Since Z ⊃ R , it is nonempty. So, by the minimality of the potential of R , we have Z = V ( J ). The final statement comes from our assumption that ρ k ( V ( J )) > y k . glyph[square]
The following fact implies that J contains no quasi-vertex.
Claim 28 For each R glyph[subsetnoteql] V ( J ) with | R | ≥ 2 and any distinct x, y ∈ R , the graph J R [ ] + xy is ( k -1) -colorable.
Proof. Let R be a smallest subset of vertices of J such that 2 ≤ | R < n | and for some distinct x, y ∈ R , the graph H = J R [ ] + xy is not ( k -1)-colorable. Since J is k -critical, xy / E J ∈ ( ).
Let H ′ be a k -critical subgraph of H . By the minimality of R V H , ( ′ ) = R . By Claim 27, ρ k,H ′ ( R ) = -(2 k -2) + ρ k,J ( R > y . ) k Because | R < n | , by Rule (S1), H ′ is smaller than J . So by the minimality of J and because ρ k,H ′ ( R > y ) k , H ′ is k -Ore and ρ k,H ′ ( R ) = k k ( -3). If there is an edge e ∈ E H ( ) -E H ( ′ ), then
$$\rho _ { k, H } ( R ) \leq \rho _ { k, H ^ { \prime } } ( R ) - 2 ( k - 1 ) = k ( k - 3 ) - 2 ( k - 1 ) \leq y _ { k },$$
contradicting Claim 27. Hence H = H ′ . Thus H is k -Ore and
$$\rho _ { k, J } ( R ) = \rho _ { k, H } ( R ) + 2 ( k - 1 ) = k ( k - 3 ) + 2 ( k - 1 ) = ( k + 1 ) ( k - 2 ). \text{ \quad \ } ( 1 6 )$$
By Claim 21, | R ∗ | ≥ 3. We want to prove that
$$J [ R ] \, \text{has a $(k-1)$-coloring $\psi$ such that $R_{*}$ is not monochromatic.}$$
glyph[negationslash]
Case 1: { x, y } ⊂ R ∗ . Since | R ∗ | ≥ 3, we may choose w ∈ R ∗ -x -y . If there exists a subset R ′ glyph[subsetnoteql] R with | R ′ | ≥ 2 such that { x, y } glyph[negationslash]⊂ R ′ and ρ k ( R ′ ) = ( k +1)( k -2), then by Lemma 18, H contains a standard set A ⊆ R ′ . But then there exists a pair of vertices { a, b } ⊂ A ⊆ R ′ glyph[subsetnoteql] R such that J A [ ] + ab is not ( k -1)-colorable, which contradicts the minimality of R . Otherwise, by Lemma 19, there is a ( k -1)-coloring φ w of H -xy such that φ w ( w ) = φ w ( x ) = φ w ( y ). Then for ψ = φ w , (17) holds.
Case 2: { x, y } glyph[negationslash]⊂ R ∗ . Let u, v be any vertices in R ∗ . If uv ∈ E J ( ), then (17) is immediately true. Otherwise, let H 0 = J R [ ] + uv . If H 0 has a ( k -1)-coloring, then (17) holds. If not, then by the minimality of R , exactly as above, H 0 is a k -Ore graph. So, we have Case 1. This proves (17).
glyph[negationslash]
Let ψ satisfy (17). Let Y = Y ( J, R, ψ, X ). By Claim 26, Y contains a vertex set W such that ρ k,Y ( W ) ≤ k k ( -3) and W ∩ X = ∅ . Recall that X induces a complete graph and that by Fact 11.7 the subgraph of K k -1 with the smallest potential is K 1 with ρ k ( V ( K 1 )) = ( k +1)( k -2) and the subgraph with the second smallest potential is K k -1 with ρ k ( V ( K k -1 )) = 2( k -2)( k -1). This together with (16) and the choice of W yields
$$\rho _ { k, J } ( W - X + R ) \leq \rho _ { k, Y } ( W ) - \rho _ { k, Y } ( X \cap W ) + \rho _ { k, J } ( R ) \leq k ( k - 3 ) - ( k + 1 ) ( k - 2 ) + ( k + 1 ) ( k - 2 ) = k ( k - 3 ).$$
Since W -X + R ⊃ R , we have | W -X + R | ≥ 2. By Fact 11.8, y k +(2 k -2) ≥ k k ( -3). This together with Claim 27 yields W -X + R = V ( J ). If | W ∩ X | ≥ 2, then we get the stronger bound ρ k ( X ∩ W ) ≥ 2( k -1)( k -2), and so the inequality in (18) strengthens to
$$\rho _ { k, J } ( W - X + R ) \leq k ( k - 3 ) - 2 ( k - 1 ) ( k - 2 ) + ( k - 2 ) ( k + 1 ) = 2 k - 6 \leq y _ { k },$$
a contradiction. Thus | X ∩ W | = 1. Because R ∗ is not monochromatic and | X ∩ W | = 1, there is a vertex z ∈ R ∗ -W . Then by (14), instead of (18) we have
$$\rho _ { k, J } ( W - X + R ) \leq k ( k - 3 ) - ( k + 1 ) ( k - 2 ) + ( k + 1 ) ( k - 2 ) - 2 k + 2 = k ^ { 2 } - 5 k + 2 \leq y _ { k },$$
a contradiction. glyph[square]
Claim 29 Let X be a ( k -1) -clique in J , u, v ∈ X N u , ( ) -X = { a } , and N v ( ) -X = { } b . Then a = b .
glyph[negationslash]
glyph[negationslash]
Proof. Assume a = b . Let J ′ = J -u -v + ab if ab / ∈ E J ( ) and J ′ = J -u -v otherwise. By Claim 28, J ′ has a ( k -1)-coloring φ . Because d u ( ) = d v ( ) = k -1, each of the sets C a = 1 { , . . . , k - }-∪ 1 w N u ∈ ( ) -v φ w ( ) and C b = { 1 , . . . , k - }-∪ 1 w N v ∈ ( ) -u φ w ( ) is nonempty. Since φ a ( ) = φ b ( ) and ( N u ( ) -a ) = ( N v ( ) -b ), the sets C a and C b are different. Therefore φ can be extended to u and v . But then we get a ( k -1)-coloring of J , a contradiction. glyph[square]
Claim 30 J does not contain K k -e .
glyph[negationslash]
Proof. Suppose J R [ ] = K k -e . The only k -critical graph on k vertices is K k , which is k -Ore. By assumption J is not k -Ore, so R = V ( J ), but adding the missing edge to J R [ ] creates a k -chromatic graph on R , a contradiction to Claim 28. glyph[square]
## 4 Clusters and sets with small potential
Recall that in Section 3 we defined a relation 'smaller,' and J is a 'smallest' counterexample to Theorem 10: it is a k -critical graph with ρ k ( V ( J )) > y k and is not k -Ore.
Definition 31 For S ⊆ V ( J ) , an S -cluster is an inclusion maximal set R ⊆ S such that for every x ∈ R d x , ( ) = k -1 and for every x, y ∈ R N x , [ ] = N y [ ] . A cluster is a V ( J ) -cluster.
Each cluster is a clique of vertices with degree k -1. In this section, results on clusters will help us to derive the main lower bound on the potentials of nontrivial vertex sets, Lemma 38, which in turn will help us to prove stronger results on the structure of clusters in J .
Having the same closed neighborhood is an equivalence relation, and so the set of clusters is a partition of the set of the vertices with degree k -1. Thus the following fact holds.
Fact 32 Every vertex with degree k -1 is in a unique cluster.
Furthermore, if the only S -cluster is the empty set, then every vertex in S has degree at least k . By definition, if a cluster T is contained in a vertex set S , then T is also an S -cluster.
Claim 33 Every cluster T satisfies | T | ≤ k -3 . Furthermore, for every ( k -1) -clique X in J , (i) there is a unique X -cluster T (possibly T = ∅ ), and (ii) every non-empty X -cluster is a cluster (in other words, every cluster is either contained in X or disjoint from X ). Each ( k -1) -clique in J contains at least 2 vertices of degree at least k .
Proof. If T is a cluster with | T | ≥ k -2, then T ∪ N T ( ) ⊇ K k -e , a contradiction to Claim 30.
Let X be a ( k -1)-clique in J . Two distinct X -clusters would contradict Claim 29. If T is a non-empty X -cluster contained in a larger cluster T ′ , then each v ∈ T ′ -X has to be adjacent to each vertex in X , and so J contains clique X ∪ T ′ of size at least k , a contradiction.
The final statement is proven as follows: by Fact 32 only vertices in clusters have degree k -1, by (i) X contains at most one cluster which, if exists, has at most k -3 of the k -1 vertices in X by the first part of this claim. glyph[square]
Claim 34 Let R ⊂ V ( J ) , let φ be a ( k -1) -coloring of J R [ ] , and let Y = Y ( J, R, φ, X ) be as in Definition 23. If Y is k -Ore, then Y ∼ = K k .
Proof. Suppose that Y is a k -Ore graph distinct from K k . Let a separating set { x, y } , vertex subsets A = A Y,x,y ( ) and B = B Y,x,y ( ), and graphs ˜ Y ( x, y ) and ˇ ( Y x, y ) be as in Fact 14. Since Y X [ ] is a clique and E Y ( A -x -y, B -x -y ) = ∅ , either X ⊆ A or X ⊆ B . Since xy / E Y ∈ ( ) we may assume that either X ⊂ A -y or X ⊂ B -y . By construction, Y -J = X , so this implies that B -x or A -x , respectively, is a subgraph of J . We will show that this is impossible.
Suppose first that X ⊆ A -y . The graph ˇ Y -x ∗ y is a subgraph of J , namely, it is J B [ -x -y ], and by Fact 15.7
$$d _ { \tilde { Y } } ( v ) = d _ { J } ( v ) \text{ for every } v \in B - x - y.$$
If ˇ Y -x ∗ y has a vertex subset S with | S | ≥ 2 of potential at most ( k + 1)( k -2), then by Lemma 18, S contains a standard set S ′ . But each standard set S ′ has two vertices u and w such that Y S [ ′ ]+ uw is not ( k -1)-colorable. This contradicts Claim 28. Thus ρ k,Y ˇ ( S ) > k ( +1)( k -2) for every S ⊆ V ( ˇ ) Y - ∗ x y with | S | ≥ 2. Then by Claim 20, there exists an S ⊆ V ( ˇ ) Y - ∗ x y = B - -x y such that ˇ [ Y S ] ∼ = K k -1 , and d ˇ Y ( v ) = k -1 for all v ∈ S . By (19), this contradicts the last part of Claim 33.
Now suppose that X ⊆ B -y . The graph ˜ Y -x is a subgraph of J , namely, it is J A [ -x ], and similarly to (19), by Fact 15.6
$$d _ { \widetilde { Y } } ( v ) = d _ { J } ( v ) \text{ for every } v \in A - x - y.$$
As above, ρ k,Y ˜ ( S ) > k ( +1)( k -2) for every S ⊆ V ( ˜ Y ) -x with | S | ≥ 2. So again by Claim 20, there exists an S ′ ⊆ V ( ˜ Y ) -x = A -x such that ˜ Y S [ ′ ] ∼ = K k -1 , and d ˜ Y ( v ) = k -1 for all v ∈ S ′ . But | S ′ - | ≥ y k -2, which together with (20) contradicts Claim 33. glyph[square]
Claim 35 For every partition ( A,B ) of V ( J ) with 2 ≤ | A | ≤ n -2 , | E J ( A,B ) | ≥ k .
Proof. Let A ∗ (respectively, B ∗ ) be the set of vertices in A (respectively, B ) that have neighbors in B (respectively, A ). Since J is 3-connected, | A ∗ | ≥ 3 and | B ∗ | ≥ 3. So by Claim 28, J A [ ] has a ( k -1)-coloring φ A such that A ∗ is not monochromatic, and J B [ ] has a ( k -1)-coloring φ B such that B ∗ is not monochromatic. But Gallai and Toft (see [32, p. 157]) independently proved that if | E J ( A,B ) | ≤ k -1, then either A ∗ is monochromatic in every ( k -1)-coloring of J A [ ] or B ∗ is monochromatic in every ( k -1)-coloring of J B [ ]. So, | E J ( A,B ) | ≥ k . glyph[square]
Sometimes below, our goal will be to extend to J a coloring φ of J R [ ] for some R and φ . Recall that Y ( J, R, φ, X ) is obtained from J by replacing the vertices of R with a clique whose vertices are the color classes of φ with at least one element in the border of R (which we called R ∗ ). One of the ways we will control φ is to add edge(s) to R before we generate a ( k -1)-coloring φ using Claim 28 and a lemma below. Our next lemma describes how edges can be placed in R so that no color class of φ is too large. The proof of this lemma will use the following old result of Hakimi.
Theorem 36 (Hakimi [20]) Let ( w , . . . , w 1 s ) be a list of nonnegative integers with w 1 ≥ . . . ≥ w s . Then there is a loopless multigraph F with vertex set { u , . . . , u 1 s } such that d F ( u j ) = w j for all j = 1 , . . . , s if and only if z = w 1 + . . . + w s is even and w 1 ≤ w 2 + . . . + w s .
For technical reasons, in one specific case of the lemma below we will allow for a hyperedge of size 3. Recall that an independent set in a hypergraph is a set that contains no edge.
Lemma 37 Let i ′ ≥ 1 and s ≥ 2 be integers. Let R ∗ = { u , . . . , u 1 s } be a vertex set. Then for each z ≥ 2 i ′ and any integral positive weight function w : R ∗ →{ 1 2 , , . . . } such that w u ( 1 )+ . . . + w u ( s ) = z and w u ( 1 ) ≥ w u ( 2 ) ≥ . . . ≥ w u ( s ) , there exists a graph H with V ( H ) = R ∗ and | E H ( ) | ≤ i ′ such that for each 1 ≤ j ≤ s , d H ( u j ) ≤ w u ( j ) , and for every independent set M in H with | M | ≥ 2 ,
$$\sum _ { u \in R _ { * } - M } w ( u ) \geq i ^ { \prime }.$$
Moreover, if s ≥ 3 , i ′ ≥ 1 , and z > 2 i ′ , then at least one of the three stronger statements below holds:
- (i) such H with Property (21) could be chosen as a graph with at most i ′ -1 edges, or
- (ii) such H with Property (21) could be chosen as a hypergraph instead of a graph with at most i ′ -1 graph edges and one edge of size 3 , or
(iii) the weight arrangement is i ′ -special , which means that s = i ′ +1 and w u ( 2 ) = . . . = w u ( s ) = 1 .
Proof. It is easy to check that the main part of the statement, and (ii) of the 'moreover' part hold for i ′ = 1. So assume i ′ ≥ 2.
Case 1: w u ( 2 ) + . . . + w u ( s ) ≤ i ′ -1. We make E H ( ) = { u u 1 j : 2 ≤ j ≤ s } . If M is any independent set with | M | ≥ 2, then u 1 / M ∈ and w u ( 1 ) ≥ 2 i ′ -( i ′ -1) yielding (21). To prove the 'Moreover' part in this case, observe that our H has at most i ′ -1 edges.
Case 2: s ≥ 2 i ′ + 1. Let the edge set of H consist of the matching { u u , . . . , u 1 2 2 i ′ -1 u 2 i ′ } . Every independent set misses at least one end of each edge in H , which implies (21). Moreover, if z ≥ 2 i ′ +1, then we extend edge u u 1 2 to the hyperedge { u , u 1 2 , u 2 i ′ +1 } . Then Claim (ii) of the 'Moreover' part of the lemma holds.
Case 3: w u ( 2 ) + . . . + w u ( s ) ≥ i ′ and s ≤ 2 i ′ . Since s ≤ 2 i ′ , there exists an auxiliary integral weight function w ′ : R ∗ →{ 1 2 , , . . . } such that
$$w ^ { \prime } ( u _ { 2 } ) + \dots + w ^ { \prime } ( u _ { s } ) \geq i ^ { \prime }, \, w ^ { \prime } ( u _ { 1 } ) + \dots + w ^ { \prime } ( u _ { s } ) = 2 i ^ { \prime } \text{ and } w ^ { \prime } ( u _ { j } ) \leq w ( u _ { j } ) \ \forall \, j = 1, \dots, s.$$
By (22) and Theorem 36, there exists a loopless multigraph H ′ with vertex set { u , . . . , u 1 s } such that d H ′ ( u j ) = w u ′ ( j ) for all j . We obtain a graph H from the multigraph H ′ by replacing each set of multiple edges with a single edge. Every independent set in H is also independent in H ′ . For every independent set M in H ′ , each of its i ′ edges has an end outside of M , so
$$\sum _ { u \in R _ { * } - M } w ( u ) \geq \sum _ { u \in R _ { * } - M } w ^ { \prime } ( u ) = \sum _ { u \in R _ { * } - M } d _ { H ^ { \prime } } ( u ) \geq | E ( H ^ { \prime } ) | = i ^ { \prime }.$$
This yields (21). Note that in this case, (21) holds for every independent set M , even if | M | = 1. Now we prove the 'Moreover' part of the statement. If H ′ had any multiple edge, then we satisfy (i) and are done. Suppose, H ′ is simple. Since z > 2 i ′ , w u ′ ( glyph[lscript] ) < w u ( glyph[lscript] ) for some 1 ≤ glyph[lscript] ≤ s . If H -u glyph[lscript] has an edge e , then after enlarging e to e + u l we still keep (21). This instance satisfies (ii), and we are done. Otherwise u glyph[lscript] is incident to every edge of H = H ′ , and so H is a star with center u glyph[lscript] and i ′ ≥ 2 edges. Each such star has only one central vertex, so every other vertex u j satisfies w u ( j ) = w u ′ ( j ) = d H ( u j ) = 1. By definition, this means that the weight arrangement is i ′ -special. So we satisfy (iii) and are done. glyph[square]
Recall that ρ k,K k -1 ( V ( K k -1 )) = 2( k -1)( k -2). Importantly, this is larger than the potential of a standard set. Our main lower bound on the potentials of nontrivial vertex sets is the following.
Lemma 38 If R glyph[subsetnoteql] V ( J ) and 2 ≤ | R | ≤ n -2 , then ρ k ( R ) ≥ 2( k -1)( k -2) . Moreover, if ρ k ( R ) = 2( k -1)( k -2) , then J R [ ] = K k -1 .
Proof. Assume that the lemma does not hold. Let i be the smallest integer such that there exists R glyph[subsetnoteql] V ( J ) with glyph[negationslash]
$$2 \leq | R | \leq n - 2, \, J [ R ] \neq K _ { k - 1 }, \, \rho _ { k } ( R ) \leq 2 ( k - 1 ) ( k - 2 ),$$
and
$$y _ { k } + 2 i ( k - 1 ) < \rho _ { k } ( R ) \leq y _ { k } + 2 ( i + 1 ) ( k - 1 ).$$
It is important that we are only minimizing i , and not necessarily minimizing ρ k ( R ). By Claim 27, i ≥ 1. In order for both (23) and (24) to hold, since
$$y _ { k } + ( k + 1 ) ( k - 1 ) \geq k ^ { 2 } - 5 k + 2 + ( k + 1 ) ( k - 1 ) > 2 ( k - 1 ) ( k - 2 ),$$
i ≤ k 2 . By the integrality, i ≤ ⌊ k 2 ⌋ . Moreover, if k = 4 then y k = max 2 { · 4 -6 4 , 2 -5 · 4 + 2 } = 2 and so y 4 +4(4 -1) = 14 > 12 = 2(4 -1)(4 -2). Thus
$$i \leq \left \lfloor \frac { k } { 2 } \right \rfloor, \text{ moreover, if } k = 4 \text{ then } i = 1.$$
Let R be a smallest subset of V ( J ) for which (23) and (24) hold. By Fact 11.7, each vertex set S with 2 ≤ | S | ≤ k -1 has potential at least 2( k -1)( k -2), with equality only when S induces a K k -1 . So (23) yields | R | ≥ k . Thus by Claim 26, for any proper ( k -1)-coloring φ of J R [ ], graph Y ( J, R, φ, X ) is smaller than J .
Let Q = V ( J ) -R , and for u ∈ R , let w u ( ) = | N u ( ) ∩ Q | . By Definition 17, R ∗ = { u ∈ R : w u ( ) ≥ } 1 . Let R ∗ = { u , . . . , u 1 s } and w u ( 1 ) ≥ . . . ≥ w u ( s ). By Claim 35,
$$z \coloneqq \sum _ { i = 1 } ^ { s } w ( u _ { i } ) = | E _ { J } ( R, V ( J ) - R ) | \geq k.$$
By Claim 21, s ≥ 3.
We will consider four cases, and the first is the main one.
Case 1: There is a ( k -1)-coloring φ of J R [ ] such that for every color class C of φ with | C ∩ R ∗ | ≥ 2 either
$$\sum _ { u \in R _ { * } - C } w ( u ) \geq i$$
or
$$\sum _ { u \in R _ { * } - C } w ( u ) = i - 1 \text{ and } \sum _ { u \in C } w ( u ) \leq k - 2.$$
Let Y = Y ( J, R, φ, X ) be as in Definition 23.
glyph[negationslash]
glyph[negationslash]
Let Y ′ be the k -critical subgraph of Y described in Claim 26. Let W = V ( Y ′ ) and X ′ = X ∩ W . Since | R | ≥ k , by Claim 26, X ′ = ∅ and one of the following two statements is true: (a) Y W [ ] contains a k -Ore graph, or (b) ρ k,Y ( W ) ≤ y k . Because X ′ = ∅ , by Fact 11.7, ρ k,Y ( X ′ ) ≥ ( k + 1)( k -2). By Corollary 9, ρ k,Y ( W ) ≤ ρ k,Y ′ ( W ) ≤ k k ( -3). By (14) and our bounds on ρ k,Y ( X ′ ) and ρ k,Y ( W ),
$$\rho _ { k, J } ( W - X + R ) \leq \rho _ { k, Y } ( W ) - \rho _ { k, Y } ( X ^ { \prime } ) + \rho _ { k, J } ( R ) \leq \rho _ { k, J } ( R ) - 2 ( k - 1 ).$$
glyph[negationslash]
Because | W -X + R | ≥ | R | ≥ k (which implies J W [ -X + R ] = K k -1 ) and the choice of i , | W -X + R | ≥ n -1. Suppose first that | W -X + R | = n -1. By Fact 22, ρ k,J ( W -X + R > ) y k + k 2 -3 k +4. By (29), ρ k,J ( R > y ) k + k 2 -k +2 ≥ 2( k -1)( k -2), contradicting the choice of R . So
$$W - X + R = V ( J ).$$
We claim that Y is a k -clique and will prove this in three steps. We will show, in order, that
$$( A ) \ | X ^ { \prime } | \geq 2, \, ( B ) \ Y ^ { \prime } \text{ is an induced $k$-Ore graph, and } ( C ) \ Y ^ { \prime } = Y.$$
These three steps prove that Y is k -Ore; Claim 34 yields that if Y is k -Ore, then Y is a k -clique.
Suppose X ′ = { x j } . Then W = V ( Y ′ ) -X + x j . Let R j = { u ∈ R ∗ : φ u ( ) = c j } , where c j is the j th color class in coloring φ . If | R j | = 1, then Y ′ ∼ = J W [ -x j ∪ R j ], which is a subgraph of J . Because | R | ≥ k > 1 = | R j | , Y ′ is a proper subgraph of J , but k -critical graphs do not have k -chromatic proper subgraphs. Thus | R j | ≥ 2. If (28) holds for R j , then d Y ′ ( x j ) ≤ ∑ u ∈ R j w u ( ) ≤ k -2, but the k -critical graph Y ′ cannot have vertices of degree less than k -1. Otherwise, by (27), at least i edges connect the vertices in R ∗ -R j with Q . Those edges are a part of E J ( W -X,R -R | W ) in (14), an edge set we conservatively estimated to be empty in the calculation of (29). Adjusting that calculation to account for our new bound on E J ( W -X,R -R | W ), instead of (29) we get
ρ k,J ( W -{ x j } + R ) ≤ k k ( -3) -( k -2)( k +1) + ρ k,J ( R ) -2 ( i k -1) = ρ k,J ( R ) -2( i +1)( k -1) .
By (24), we have ρ k,J ( R ) -2( i +1)( k -1) ≤ y k , which together with i ≥ 0, (29), and (30) imply ρ k,J ( V ( J )) ≤ y k , contradicting our assumption that ρ k,J ( V ( J )) > y k . This proves (A).
If Y ′ is not an induced k -Ore graph, then by Claim 26, ρ k,Y ( W ) ≤ y k . We adjust the calculation of (29) from (14) again: we use the inequality ρ k,Y ( W ) ≤ y k and that part (A) implies 2 ≤ | X ′ | ≤ k -1, so X ′ has potential at least 2( k -1)( k -2) by Fact 11.7. The result is
$$\rho _ { k, J } ( W - X + R ) \leq y _ { k } - 2 ( k - 1 ) ( k - 2 ) + \rho _ { k, J } ( R ).$$
We apply (30) to the left hand side and (23) to the right hand side to see that ρ k,J ( V ( J )) ≤ y k , which contradicts our choice of J . This proves (B).
glyph[negationslash]
By (30) and (B), if Y ′ = Y , then X ′ = X . If X ′ = X , then we again get a stronger bound on E J ( W -X,R -R | W ) in (14) (this time it contains at least one element), and so we may again adjust the bound in (29) by subtracting 2( k -1). We augment the application of (14) to (29) in two ways: the extra -2( k -1) described in the previous sentence and that | X ′ | ≥ 2 implies ρ k,Y ( X ′ ) ≥ 2( k -1)( k -2) by Fact 11.7; those augmentations and (23) produce glyph[negationslash]
glyph[negationslash]
$$\rho _ { k, J } ( W - X + R ) \leq - 2 ( k - 1 ) + k ( k - 3 ) - \rho _ { k, Y } ( X ^ { \prime } ) + \rho _ { k, J } ( R ) \leq k ^ { 2 } - 5 k + 2 \leq y _ { k },$$
which contradicts our choice of J for the same reason as the proof to part (B). This proves (C); and therefore Y = Y ( J, R, φ, X ) is k -Ore. By Claim 34, Y = K k .
Recall that { Q,R } is a partition of V ( J ). Then by (30) and (31C), { Q,X } is a partition of V ( Y ). Let t = | X | (and t = | X ′ | also by (31C)). By (31A), t ≥ 2. Because | R | ≤ n -2, | Q | ≥ 2. So since V ( Y ) = X ∪ Q , we have t ≤ k -2. Then J is obtained from J R [ ] by adding k -t vertices and at least ( ) k 2 -( ) t 2 edges (a vertex in Q may be adjacent to more than one vertex in a color class of φ ). So
$$\tau _ { i } ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ \nu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^{\nu } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i } \mu ^ { i }$$
We know that 2 ≤ t ≤ k -2. We will show that t = 2. This is clear when k = 4. Let k ≥ 5. Denote the RHS of (32) by µ k, t, R ( ). For fixed k and R µ k,t, R , ( ) is quadratic in t with a positive coefficient at t 2 . So, if 3 ≤ t ≤ k -2, then µ k, t, R ( ) ≤ max { µ k, ( 3 , R , µ ) ( k, k -2 , R ) } . Furthermore,
$$\mu ( k, k - 2, R ) = \rho _ { k } ( R ) + 2 ( k + 1 ) ( k - 2 ) - k ( k - 1 ) ^ { 2 } + ( k - 1 ) ( k - 2 ) ( k - 3 )$$
$$\leq 2 ( k - 1 ) ( k - 2 ) - 2 k ^ { 2 } + 8 k - 1 0 = 2 k - 6 \leq y _ { k },$$
$$& + 1 ) ( k - 2 ) - k ( k - 1 ) ^ { 2 } + ( k - 1 ) ( k - 2 ) ( k - 3 ) \\ & 2 ) - 2 k ^ { 2 } + 8 k - 1 0 = 2 k - 6 \leq y _ { k },$$
and when k ≥ 5,
$$\mu ( k, 3, R ) = \rho _ { k } ( R ) + ( k - 3 ) ( k + 1 ) ( k - 2 ) - k ( k - 1 ) ^ { 2 } + 6 ( k - 1 ) \leq 2 ( k - 1 ) ( k - 2 ) - 2 k ^ { 2 } + 6 k = 4 \leq y _ { k }.$$
Since ρ k ( V ( J )) > y k by the choice of J , we conclude that t = 2 and J Q [ ] = K k -2 . Moreover, µ (4 , 2 , R ) ≤ 2(4 -1)(4 -2) + 20 -(6 -1)6 = 2 = y 4 , so k ≥ 5. Thus by (32),
$$\rho _ { k } ( V ( J ) ) \leq \mu ( k, 2, R ) = \rho _ { k } ( R ) + ( k - 2 ) ^ { 2 } ( k + 1 ) - k ( k - 1 ) ^ { 2 } + 2 ( k - 1 ) = \rho _ { k } ( R ) - k ^ { 2 } + k + 2. \ ( 3 3 )$$
Hence by (33) and (12),
$$\rho _ { k } ( R ) \geq \rho _ { k } ( V ( J ) ) + k ^ { 2 } - k - 2 \geq y _ { k } + k ^ { 2 } - k.$$
Plugging in the values of y k , we get
$$\rho _ { k } ( R ) \geq 2 ( k - 1 ) ( k - 2 ) - 2 \text{ for } k \geq 6 \text{ and } \rho _ { 5 } ( R ) \geq 2 ( 5 - 1 ) ( 5 - 2 ) = 2 4.$$
Since for k ≥ 5, 2( k -1)( k -2) -2 > y k +4( k -1), we have
$$i \geq 2.$$
Also we conclude that each v ∈ Q has exactly two neighbors in R , since otherwise the upper bound on ρ k ( V ( J )) in (32) and (33) would be stronger by 2( k -1), and this combined with (23) would lead to
$$\rho _ { k } ( V ( J ) ) \leq - 2 ( k - 1 ) + 2 ( k - 1 ) ( k - 2 ) - k ^ { 2 } + k + 2 = k ^ { 2 } - 7 k + 8 \leq y _ { k }.$$
glyph[negationslash]
Let Q = { v , . . . , v 1 k -2 } and let N v ( j ) ∩ R = { u j, 1 , u j, 2 } for j = 1 , . . . , k -2. If there exists a proper ( k -1)-coloring φ ′ of J R [ ] such that φ ′ ( u j, 1 ) = φ ′ ( u j, 2 ) for some j , then φ ′ may be extended to all of J greedily by first coloring Q -v j and at the end coloring v j (at each step at most k -2 colors must be avoided). Similarly, if { φ ′ ( u j, 1 ) , φ ′ ( u j, 2 ) } glyph[negationslash] = { φ ′ ( u j , ′ 1 ) , φ ′ ( u j , ′ 2 ) } for some j = j ′ , then φ ′ may be extended to all of J greedily by first coloring X -v j -v j ′ and at the end coloring v j and v j ′ . Thus for any proper ( k -1)-coloring φ ′ of J R [ ], glyph[negationslash]
$$\text{for all $1\leq j,j^{\prime}\leq k-2$,} \quad \phi ^ { \prime } ( u _ { j, 1 } ) \neq \phi ^ { \prime } ( u _ { j, 2 } ) \text{ and } \{ \phi ^ { \prime } ( u _ { j, 1 } ), \phi ^ { \prime } ( u _ { j, 2 } ) \} = \{ \phi ^ { \prime } ( u _ { j ^ { \prime }, 1 } ), \phi ^ { \prime } ( u _ { j ^ { \prime }, 2 } ) \}. \quad ( 3 6 )$$
glyph[negationslash]
Because 3 ≤ s = | R ∗ | by Claim 21, there exist distinct vertices v , v ′ ′′ ∈ Q such that N v ( ′ ) ∩ R = N v ( ′′ ) ∩ R . By symmetry, we may assume u 1 1 , / N v ∈ ( 2 ). Let J ∗ be obtained from J R [ ] by adding edges e 1 = u 1 1 , u 2 1 , and e 2 = u 1 1 , u 2 2 , when they do not exist. By (36), χ J ( ∗ ) ≥ k , which implies that e 1 or e 2 was not in J . Thus J ∗ contains a k -critical subgraph J ◦ . By the minimality of J and the fact that | J ◦ | ≤ | R | ≤ | J | -2, graph J ◦ is k -Ore or ρ k,J ∗ ( V ( J ◦ )) ≤ y k . Because J ◦ = J V [ ( J ◦ )] + S for some nonempty S ⊆ { e , e 1 2 } , we know that J V [ ( J ◦ )] is not a clique. If V ( J ◦ ) satisfies (23), then the minimality of i applied to (24) states that ρ k,J ( V ( J ◦ )) > y k +2 ( i k -1); if V ( J ◦ ) does not satisfy (23), then (as χ J ( ◦ ) = k implies | V ( J ◦ ) | ≥ k , and the other conditions are proved above)
$$\rho _ { k, J } ( V ( J ^ { \circ } ) ) > 2 ( k - 1 ) ( k - 2 ) \geq y _ { k } + 2 i ( k - 1 ).$$
Since i ≥ 2 by (35) and because we have added at most two edges, by (24), ρ k,J ∗ ( V ( J ◦ )) ≥ ρ k,J ( V ( J ◦ )) -4( k -1) > y k , and so J ◦ is k -Ore. Recall by Fact 15.9, the vertex sets of k -Ore graphs have potential k k ( -3). So, ρ k,J ∗ ( V ( J ◦ )) ≤ k k ( -3) and our above inequality becomes by Fact 11.8
$$\rho _ { k, J } ( V ( J ^ { \circ } ) ) \leq \rho _ { k, J ^ { * } } ( V ( J ^ { \circ } ) ) + 4 ( k - 1 ) \leq k ( k - 3 ) + 4 ( k - 1 ) \leq y _ { k } + 3 ( 2 ( k - 1 ) ).$$
Hence V ( J ◦ ) satisfies (24) for some i ≤ 2. By (35), we have i = 2. By the minimality of i and of | R | , this gives
$$i = 2, \, V ( J ^ { \circ } ) = R \text{ and } J [ R ] = J ^ { \circ } - e _ { 1 } - e _ { 2 }.$$
Also, since i = 2 and ( k +1)( k -2) ≤ y k +2(2( k -1)), the minimality of the potential of R in (24), and Fact 11.(2,3,7) imply
$$J [ R ] \, \text{contains no set with potential at most } ( k + 1 ) ( k - 2 ).$$
For all S ⊆ V ( J ◦ ) -u 1 1 , , we have J ∗ [ S ] ∼ = J S [ ]. Thus, by (37) and (38), Claim 20 applies to J ◦ = J ∗ and u 1 1 , . By this claim, J ◦ -u 1 1 , contains a clique S of order k -1 such that each vertex in S has degree k -1. Since u 1 1 , / S ∈ , S is also a clique in J . Since e , e 1 2 ⊂ N Q ( ) = R ∗ , if u ∈ S -N Q ( ) then d J ( u ) = k -1. Because N Q ( ) is 2-colorable (by (36)), this implies that there is an S ′ ⊂ S with | S ′ | ≥ k -3 such that d J ( u ) = k -1 for all u ∈ S ′ . Each vertex of S ′ is in a cluster by Fact 32, and Claim 33 says that all of S ′ is one cluster and that | S ′ | = k -3. Let { u ′ } = N S ( ′ ) -S . Then by Fact 11 (parts 1, 2, and 4)
$$\rho _ { k, J } ( S \cup u ^ { \prime } ) \leq \rho _ { k, J } ( S ) + \rho _ { k, J } ( \{ u ^ { \prime } \} ) - 2 ( k - 1 ) ( k - 3 ) = k ^ { 2 } + k - 4 < y _ { k } + 3 ( 2 ( k - 1 ) ).$$
By the minimality of R , we have R = S ∪ { u ′ } . So u 1 1 , = u ′ and J R [ ] is a k -clique minus the edges e 1 = u 1 1 , u 2 1 , and e 2 = u 1 1 , u 2 2 , . But then for any possible choice of u 1 2 , , there exists a
( k -1)-coloring φ of J R [ ] such that { φ ′ ( u 1 1 , ) , φ ′ ( u 1 2 , ) } glyph[negationslash] = { φ ′ ( u 2 1 , ) , φ ′ ( u 2 2 , ) } . This contradiction to (36) finishes Case 1.
In all subsequent cases, we will use Lemma 37 in order to construct either a ( k -1)-coloring of J or a ( k -1)-coloring of J R [ ] fitting into Case 1. Lemma 37 uses variables i , z, s, w ′ . We will use several different values for i ′ , but z, s, w are always defined by the discussion leading to (26), and by Claim 21, s ≥ 3.
Case 2: 2 i ≥ z = | E R,Q ( ) . | By (25) and (26), in order to have 2 i ≥ | E R,Q ( ) , | we need i = k 2 , k ≥ 6, and | E R,Q ( ) | = k . For k ≥ 6, we know that y k = k 2 -5 k +2. By Lemma 37 with i ′ = i -1, we can add to J R [ ∗ ] a set E 1 of at most i -1 edges such that (21) holds with i ′ = i -1. Let H 1 = J R [ ] ∪ E 1 . By (24),
$$\rho _ { k, H _ { 1 } } ( R ^ { \prime } ) > y _ { k } + 2 i ( k - 1 ) - ( i - 1 ) 2 ( k - 1 ) = k ( k - 3 )$$
for every R ′ ⊆ R with | R ′ | ≥ 2 (as this inequality is also true for K k -1 ). So, by Corollary 9, H 1 has a proper ( k -1)-coloring φ . Moreover, φ is also a proper ( k -1)-coloring of J R [ ]. Since Case 1 does not hold, φ has a color class C that satisfies neither (27) nor (28). This means that ∑ u ∈ R ∗ -C w u ( ) = i -1 (the lower bound is from (21)) and ∑ u ∈ C w u ( ) ≥ k -1 . But then
$$| E ( R, Q ) | \geq k - 1 + i - 1 \geq k - 1 + \frac { k } { 2 } - 1 = \frac { 3 k } { 2 } - 2.$$
Since k ≥ 6, this contradicts | E R,Q ( ) | = k . This concludes Case 2.
If Case 2 does not hold, then z > 2 i and, since s = | R ∗ | ≥ 3 by Claim 21, the 'moreover' part of Lemma 37 holds when applied with i ′ = i (recall the calculation right after (24) showing that i ≥ 1). The cases below analyze the possibilities in Lemma 37.
Case 3: The set { w u ( 1 ) , . . . , w ( u s ) } is i -special: s = i +1 and w u ( 2 ) = . . . = w u ( s ) = 1. This means that many (exactly z -i ≥ i ) edges connect u 1 with Q and each of the vertices u , . . . , u 2 i +1 is connected to Q by exactly one edge. For j = 2 , . . . , i + 1, let q j be the vertex in Q such that u q j j ∈ E J ( ). Let E 0 = { u u 1 j : 2 ≤ j ≤ } i and H 0 = J R [ ] ∪ E 0 . Since | R < n | , H 0 is smaller than J . Since | E 0 | = i -1, by (24),
$$\rho _ { k, H _ { 0 } } ( R ^ { \prime } ) > y _ { k } + 2 i ( k - 1 ) - ( i - 1 ) 2 ( k - 1 ) \geq k ( k - 3 )$$
glyph[negationslash]
for every R ′ ⊆ R with | R ′ | ≥ 2 (as such an inequality is also true for K k -1 ). So, by the second part of Corollary 9, H 0 has a proper ( k -1)-coloring φ . By construction, φ is a proper ( k -1)coloring of J R [ ] that satisfies φ u ( j ) = φ u ( 1 ) for each 2 ≤ j ≤ i . If φ u ( i +1 ) = φ u ( 1 ), then for every monochromatic subset M of R ∗ in J ∪ E 0 with | M | ≥ 2, (21) holds. But this coloring satisfies (27), which contradicts that Case 1 does not apply, so suppose φ u ( i +1 ) = φ u ( 1 ).
glyph[negationslash]
Let J 0 be obtained from J V [ ( J ) -( R -u 1 )] by adding edge u q 1 i +1 . By Claim 28, J 0 has a ( k -1)coloring φ ′ . Since i ≤ k 2 by (25), we can rename the colors in φ ′ so that φ ′ ( u 1 ) = φ u ( 1 ) = φ u ( i +1 ) and φ ( { u , . . . , u 2 i } ) ∩ φ ′ ( { q , . . . , q 2 i } ) = ∅ . Then φ ∪ φ ′ is a proper ( k -1)-coloring of J , a contradiction.
Case 4: The set of weights { w u ( 1 ) , . . . , w ( u s ) } is not i -special and 2 i < z , so that Part (i) or (ii) of the 'moreover' part of Lemma 37 holds when applied with i ′ = i . If Part (i) holds, then we take this set E 0 of at most i -1 edges and let H 0 = J R [ ] ∪ E 0 . In this case by (24), ρ k,H 0 ( R ′ ) > y k +2 k -2 ≥ k k ( -3) for every R ′ ⊆ R with | R ′ | ≥ 2 (as such an inequality also holds
for K k -1 ). So, by the second part of Corollary 9, H 0 has a ( k -1)-coloring φ , satisfying (27) of Case 1.
Suppose now that Part (ii) holds: there is a hypergraph H with at most i -1 graph edges and a 3 -edge e 0 = { u, v, w } such that d H ( u j ) ≤ w u ( j ) for all j = 1 , . . . , s and (21) holds. Let H 1 be obtained from J R [ ] by adding the set of edges E H ( ) -e 0 and edge uv . Since | R < n H | , 1 is smaller than J . A proper ( k -1)-coloring of H 1 would satisfy (27) of Case 1, so χ H ( 1 ) ≥ k . Then H 1 has a k -critical subgraph H ′ 1 .
Let R ′ = V ( H ′ 1 ). Because Part (i) of the 'moreover' does not hold, { u, v } ⊆ R ′ and uv / J R ∈ [ ′ ]. So J R [ ′ ] is not a clique. If H ′ 1 is not a k -Ore graph, then by the minimality of J , ρ k,H 1 ( R ′ ) ≤ y k and so ρ k,J ( R ′ ) ≤ y k +2 ( i k -1), contradicting in (24) the minimality of i (as y k +2 ( i k -1) ≤ 2( k -1)( k -2) and J R [ ′ ] is not a clique). Thus, H ′ 1 is a k -Ore graph and ρ k,H 1 ( R ′ ) = k k ( -3) ≤ y k +2 k -2. Then
$$\rho _ { k, J } ( R ^ { \prime } ) \leq \rho _ { k, H _ { 1 } } ( R ^ { \prime } ) + 2 i ( k - 1 ) \leq y _ { k } + 2 ( i + 1 ) ( k - 1 ),$$
glyph[negationslash]
and by the minimality of R , R ′ = R . Furthermore, if H ′ 1 = H 1 , then it has the same vertex set as H 1 and at least one fewer edge, in which case,
$$\rho _ { k, J } ( R ^ { \prime } ) \leq \rho _ { k, H ^ { \prime } _ { 1 } } ( R ^ { \prime } ) + 2 i ( k - 1 ) \leq \rho _ { k, H _ { 1 } } ( R ^ { \prime } ) + 2 ( i - 1 ) ( k - 1 ) \leq k ( k - 3 ) + 2 ( i - 1 ) ( k - 1 ) \leq y _ { k } + 2 i ( k - 1 ),$$
a contradiction to (24). So, H ′ 1 = H 1 , H 1 is a k -Ore graph and so ρ k,J ( R ) = k k ( -3) + 2 ( i k -1). Recall that ρ k,J ( R ) ≤ 2( k -1)( k -2) by the last inequality of (23), so
$$i \leq \frac { 2 ( k - 1 ) ( k - 2 ) - k ( k - 3 ) } { 2 ( k - 1 ) } = \frac { 1 } { 2 } ( k - 2 + \frac { 2 } { k - 1 } ).$$
glyph[negationslash]
Because i is an integer, this inequality is strict, and so ρ k,J ( R ) < 2( k -1)( k -2). Therefore ρ k,J ( W > ρ ) k,J ( R ) if J W [ ] = K k -1 . By the minimality of i and R , any W ⊂ R such that | W | ≥ 2 and J W [ ] = K k -1 satisfies ρ k,J ( W > ρ ) k,J ( R ). Graph H 1 -uv is J R [ ] plus i -1 edges, so for any W ⊂ V ( H ′ 1 ) with | W | ≥ 2 we have
$$\rho _ { k, H _ { 1 } - u v } ( W ) = \rho _ { k, J } ( W ) - 2 ( i - 1 ) ( k - 1 ) > k ( k - 3 ) + 2 ( k - 1 ) = ( k + 1 ) ( k - 2 ).$$
glyph[negationslash]
Thus by Lemma 19, H 1 -uv has a ( k -1)-coloring φ with φ w ( ) = φ u ( ). This is a ( k -1)-coloring of J R [ ] that satisfies (27) of Case 1. glyph[square]
Recall that a standard set has potential ( k +1)( k -2). Because ( k +1)( k -2) < 2( k -1)( k -2) when k ≥ 4, Lemma 38 implies:
Corollary 39 For each R glyph[subsetnoteql] V ( J ) with 2 ≤ | R | ≤ n -2 , ρ k ( R > k ) ( +1)( k -2) . In particular, J does not contain a standard set of size at most n -2 .
Claim 40 If v is not in a ( k -1) -clique X , then | N v ( ) ∩ X | ≤ k -1 2 . Furthermore, if T is a cluster in a ( k -1) -clique X , then | T | ≤ k -1 2 .
Proof. If | N v ( ) ∩ X | ≥ glyph[ceilingleft] k/ 2 , then glyph[ceilingright] ρ k ( X + ) v ≤ 2( k -2)( k -1) -2. Since n ≥ k +2, this contradicts Lemma 38. This proves the first part.
Suppose now that T is a cluster in a ( k -1)-clique X . Since | X | = k -1 and d w ( ) = k -1 for every w ∈ T , each such w has the unique neighbor v w ( ) outside of X . But by the definition of a cluster, v w ( ) is the same, say v , for all w ∈ T . This means that T ⊆ X ∩ N v ( ), so | N v ( ) ∩ X | ≥ | T | . Thus the second part follows from the first. glyph[square]
Claim 41 Suppose T is a cluster in J , t = | T | ≥ 2 , and N T ( ) ∪ T contains a ( k -1) -clique X . Then d J ( v ) ≥ k -1 + t for every v ∈ X -T .
## Proof.
glyph[negationslash]
By the definition of a cluster, | T ∪ N T ( ) | = k . So, since | T | ≥ 2 and | X | = k -1, T ∩ X = ∅ . So by Claim 33(ii), T ⊆ X , and by Claim 33(i), X contains only one nonempty cluster, namely, T .
Suppose v ∈ X -T and d v ( ) ≤ k -2 + t . By the previous paragraph, v is not in a cluster and thus by Fact 32, d v ( ) ≥ k . By Claim 30, T is contained in at most one ( k -1)-clique (which is X ), and so
$$N ( T ) \cup T - v \, \text{ does not contain } K _ { k - 1 }.$$
Because T and v are parts of the same clique, | N v ( ) -T | = d v ( ) -| T | , and this is at most k -2, since d v ( ) ≤ k -2+ . Let t u ∈ T and J ′ = J -v + u ′ , where u ′ is a new vertex that satisfies N u [ ′ ] = N u [ ]. Suppose J ′ has a ( k -1)-coloring φ ′ : V ( J ′ ) → C = { c 1 , . . . c k -1 } . Then we find a ( k -1)-coloring φ of J as follows: set φ | V ( J ) - -T v = φ ′ | V ( J ′ ) - -T u ′ , φ v ( ) ∈ C -φ ′ ( N v ( ) -T ), and then color T using colors in φ ′ ( T ∪ u ′ ) -φ v ( ). This is a contradiction, so there is no ( k -1)-coloring of J ′ . Thus J ′ contains a k -critical subgraph J ′′ . Let W = V ( J ′′ ). By Corollary 9, ρ k,J ′ ( W ) ≤ k k ( -3).
By the criticality of J , graph J ′′ is not a subgraph of J . So u ′ ∈ W and u ′ has at least k -1 neighbors in W . By symmetry, we have T ⊂ W . But then
$$\rho _ { k, J } ( W - u ^ { \prime } ) \leq k ( k - 3 ) - ( k - 2 ) ( k + 1 ) + 2 ( k - 1 ) ( k - 1 ) = 2 ( k - 2 ) ( k - 1 ).$$
By Lemma 38, this implies that either J W [ -u ′ ] is a K k -1 or W -u ′ = V ( J ) -v . If the former holds, then because J W [ -u ′ ] is a complete graph and T ⊂ W -u ′ we have N T ( ) ∪ T ⊃ J W [ -u ′ ] ∼ = K k -1 , and because v / W ∈ this is a contradiction to (39). If the latter holds, then we have a contradiction to Fact 22, since d v ( ) ≥ k . glyph[square]
glyph[negationslash]
Claim 42 Suppose xy ∈ E J ( ) , N x [ ] = N y [ ] , x is in a cluster of size s , y is in a cluster of size t , and s ≥ t . Then x is in a ( k -1) -clique. Furthermore, t = 1 .
Proof. Assume that x is not in a ( k -1)-clique. Let J ′ = J -y + x ′ , where x ′ is a new vertex with N J ′ ( x ′ ) = N J [ x ] -y . By the definition of a cluster, d x ( ) = d y ( ) = k -1. Both J ′ and J have the same number of vertices and the same number of edges (because xy ∈ E J ( ), vertex x lost an edge to y and gained an edge to x ′ ). If distinct v, w ∈ V ( J ) -y , then they have the same closed neighborhood in J ′ if and only if they have the same closed neighborhood in J . By the definition of a cluster, there are exactly t -1 vertices in V ( J ) -y that have the same closed neighborhood as y . By construction, there are exactly s vertices in V ( J ′ ) -x ′ that have the same closed neighborhood in J ′ as x ′ . So since s ≥ t , by Rule (S3), J ′ is smaller than J .
If J ′ has a ( k -1)-coloring φ ′ : V ( J ′ ) → C = { c 1 , c 2 , . . . c k -1 } , then we extend it to a proper ( k -1)-coloring φ of J as follows: define φ | V ( J ) - -x y = φ ′ | V ( J ′ ) - -x x ′ , then choose φ y ( ) ∈ C -( φ ′ ( N y ( ) -x )), and φ x ( ) ∈ { φ ′ ( x , φ ) ′ ( x ′ ) } - { φ y ( ) } .
glyph[negationslash]
So, χ J ( ′ ) ≥ k and J ′ contains a k -critical subgraph J ′′ . Let W = V ( J ′′ ). By the criticality of J and because y / J ∈ ′′ , we have J ′′ = J , and J ′′ is not a subgraph of J . This yields x ′ ∈ W . By symmetry, x ∈ W . By Corollary 9, ρ k,J ′ ( W ) ≤ k k ( -3). Because of this and the fact that d x ( ′ ) = k -1,
$$\rho _ { k, J } ( W - x ^ { \prime } ) \leq k ( k - 3 ) - \rho _ { k, J ^ { \prime } } ( \{ x ^ { \prime } \} ) + 2 ( k - 1 ) d ( x ^ { \prime } ) = 2 ( k - 2 ) ( k - 1 ). \quad \ ( 4 0 )$$
glyph[negationslash]
By assumption, x is not in a ( k -1)-clique, so Lemma 38 implies that | W -x ′ | > n -2. Since y / W ∈ , this yields | W -x ′ | = n -1, which implies V ( J ′ ) = V ( J ′′ ) and W -x ′ = V ( J ) -y . By Corollary 9, ρ k,J ′′ ( W ) ≤ k k ( -3). Moreover, because J ′′ is smaller than or equal to J ′ which is smaller than J , the minimality of J implies that if J ′′ is not k -Ore then ρ k,J ′′ ( W ) ≤ y k . In this case we can replace the term k k ( -3) in (40) with y k (as J ′′ is a subgraph of J ′ and so ρ k,J ′ ( W ) ≤ ρ k,J ′′ ( W )) and get ρ k,J ( W -x ′ ) ≤ y k + k 2 -3 k +4 contradicting Fact 22. If J ′′ = J ′ then ρ k,J ′′ ( W ) -2( k -1) ≥ ρ k,J ′ ( W ), and instead of (40) we obtain ρ k,J ( W -x ′ ) ≤ 2( k -2)( k -1) -2( k -1), again contradicting Fact 22. So J ′′ = J ′ and J ′′ is k -Ore, thus J ′ is a k -Ore graph.
glyph[negationslash]
Since n > k , J ′ = K k . Let the separating set { u, v } , vertex subsets A = A J , u, v ( ′ ) and B = B J , u, v ( ′ ), and graphs ˜ J ′ ( u, v ) and ˇ ( J ′ u, v ) be as in Fact 14. By Corollary 39, the standard set A is not contained in V ( J ). Hence x ′ ∈ A . Therefore x ′ / V ∈ ( ˇ ( J ′ u, v )) -u ∗ v . Thus for every W ⊆ V ( ˇ ) J ′ -u ∗ v with | W | ≥ 2, by Corollary 39, we have ρ k,J ˇ ( ′ u,v ) ( W ) = ρ k,J ( W > k ) ( +1)( k -2). Then by Claim 20, there exists an
$$S \subseteq V ( \check { J } ^ { \prime } ( u, v ) ) - u * v \ s u c h \ t h a t \, \check { J } ^ { \prime } ( u, v ) [ S ] \cong K _ { k - 1 }, \ a n d \, d _ { \check { J } ^ { \prime } ( u, v ) } ( w ) = k - 1 \text{ for all } w \in S. \ ( 4 1 )$$
By Claim 30, vertex y in J is adjacent to at most k -3 vertices in S . Because V ( J ) -y ⊂ V ( J ′ ), (41) yields that vertices in S -N y ( ) have degree k -1 in J . So by Fact 32, S contains a V ( J )-cluster T , which by Claim 33 contains all vertices in S of degree k -1 and | T | ≥ | S -N y ( ) | ≥ 2. Then by Claim 41, the degree of each vertex in S -T in J is at least k +1. This is impossible, since by Fact 15.5 each of them has in J at most one extra neighbor (and it is y ) in comparison with ˇ ( J ′ u, v ), where their degree was only k -1. This proves the first part: x is in a ( k -1)-clique, say X .
glyph[negationslash]
Let T y be the cluster containing y . By the definition of a cluster, every vertex in T y has the same neighbors as y , and so T y ⊆ N x ( ). The ( k -1)-clique X containing x is a part of N x [ ]. By Claim 33(ii) if T y ∩ X = , then ∅ T y ⊆ X , and by Claim 33(i) y ∈ T y -X ; so T y ∩ X = ∅ . Therefore | T y | ≤ | N x ( ) -X | = d x ( ) -( k -2) = 1. This proves the second part. glyph[square]
Claim 43 Suppose T is a cluster in J , t = | T | ≥ 2 , and N T ( ) ∪ T does not contain K k -1 . Then d J ( v ) ≥ k -1 + t for every v ∈ N T ( ) -T .
Proof. By Claim 42, d v ( ) ≥ k . Now the proof follows exactly as the proof to Claim 41. glyph[square]
## 5 Proof of Theorem 10
Now we are ready to prove the theorem. Recall that in Section 3 we defined a relation 'smaller,' and J is a 'smallest' counterexample to Theorem 10: it is a k -critical graph with ρ k ( V ( J )) > y k and is not k -Ore.
We will use the following result on k -critical graphs which is Corollary 9 in [26].
Lemma 44 ([26]) Let J be a k -critical graph. Let disjoint vertex subsets A and B be such that (a) at least one of A an B is independent;
- (b) d a ( ) = k -1 for every a ∈ A ;
- (c) d b ( ) = k for every b ∈ B ;
(d) | A | + | B | ≥ 3 .
Then (i) e J ( ( A,B )) ≤ 2( | A | + | B | ) -4 and (ii) e J ( ( A,B )) ≤ | A | +3 | B | -3 .
## 5.1 Case k = 4
In this subsection we prove the theorem for k = 4. Specifically, we will prove that | E J ( ) | ≥ 5 3 | V ( J ) , | which will imply that ρ 4 ,J ( V ( J )) ≤ 0 ≤ y 4 = 2.
Claim 45 Each vertex with degree 3 has at most one neighbor with degree 3 .
Proof. Let x be such that N x ( ) = { a, b, c } and d a ( ) = 3. Then x and a are each in a cluster. Because no cluster is larger than k -3 = 1 by Claim 33, a and x are in different clusters. Then by Claim 42, J [ { x, b, c } ] is a K 3 . So by Claim 33, d b , d ( ) ( c ) ≥ 4. glyph[square]
We now use discharging to show that | E J ( ) | ≥ 5 3 n . Each vertex begins with charge equal to its degree. If d v ( ) ≥ 4, then v gives charge 1 6 to each neighbor. Note that v will be left with charge at least 5 6 d v ( ) ≥ 10 3 . By Claim 45, each vertex of degree 3 will end with charge at least 3 + 2 6 = 10 3 . Therefore the total charge is at least 10 3 n , and thus so is the sum of the vertex degrees. Hence the number of edges is at least 5 3 n . glyph[square]
## 5.2 Case k = 5
In this subsection we prove the theorem for k = 5. Specifically, we will prove that | E J ( ) | ≥ 9 4 | V ( J ) , | which will imply that ρ 5 ,J ( V ( J )) ≤ 0 < y 5 = 4.
Claim 46 If k = 5 , then each cluster has only one vertex.
glyph[negationslash]
Proof. Suppose the claim does not hold. By Claim 33, every cluster has size at most k -3 = 2, so assume that { x, y } is a cluster: N x [ ] = N y [ ] and d x ( ) = d y ( ) = 4. Let N x ( ) = { y, a, b, c } . By assumption J is not 5-Ore and therefore J = K 5 (and since J is critical, it does not contain a K 5 ). By Claim 28, J does not contain a subgraph isomorphic to K 5 -e . Therefore any five vertices in J induce at most ( ) 5 2 -2 edges, and thus | E J ( [ { a, b, c } ]) | ≤ 1. By Claims 41 and 43, we can rename the vertices in { a, b, c } so that ab, ac / E ∈ ( J ) and d c ( ) ≥ 6.
We obtain J ′ from J by deleting x and y and gluing a with b . If J ′ is 4-colorable, then so is J . This is because any 4-coloring of J ′ will have at most 2 colors on N x [ ] -{ x, y } and therefore could be extended greedily to x and y .
So J ′ contains a k -critical subgraph J ′′ . Let W ′ = V ( J ′′ ). Then by Corollary 9, ρ 5 ,J ′ ( W ′ ) ≤ ρ 5 ,J ′′ ( W ′ ) ≤ 10. Furthermore, because J ′′ is smaller than J ,
$$\text{if $J^{\prime\prime}$ is not $k$-Ore, then $\rho_{5,J^{\prime\prime}(W^{\prime}) \leq 4$}.$$
Because J is critical and x, y / J ∈ ′′ ⊆ J ′ , graph J ′′ is not a subgraph of J . This implies that a ∗ b ∈ V ( J ′′ ). Let W = W ′ -a ∗ b + a + b + x + . By the definition of potential, y
$$\rho _ { 5, J } ( W ) = \rho _ { 5, J ^ { \prime \prime } } ( W ^ { \prime } ) + 1 8 \left ( | W | - | W ^ { \prime } | \right ) - 8 \left ( | E ( J [ W ] ) | - | E ( J ^ { \prime \prime } [ W ^ { \prime } ] ) | \right ).$$
If c / W ∈ ′ , then by construction, W has 3 more vertices and induces at least 5 more edges than W ′ . If c ∈ W ′ , then W has 3 more vertices and induces at least 7 more edges than W ′ . Suppose first that c / W ∈ ′ , so that (43) becomes ρ 5 ,J ( W ) ≤ 10 + 54 -40 = 24. Because ab / E J ∈ ( ), J W [ ]
is not a K 4 . So by Lemma 38, | W | ≥ n -1. Therefore W = V ( J ) -c and ρ 5 ,J ( W ) ≤ 24, but this contradicts Fact 22 because d c ( ) ≥ 6 = k +1.
So now we assume that c ∈ W ′ , which means that (43) becomes ρ 5 ,J ( W ) ≤ 10+54 -56 ≤ 8. By Lemma 27, W = V ( J ), which then implies V ( J ′′ ) = V ( J ′ ). Furthermore, if J ′′ is not k -Ore, then by (42), ρ 5 ,J ′′ ( W ′ ) ≤ 4 and the right hand side of (43) has an extra -6. If J ′′ is a proper subgraph of J ′ , then | E J W ( [ ]) | -| E J ( ′′ [ W ′ ]) | is at least 8 instead of the previously applied bound of 7. In either case our bound becomes stronger by at least 6, and becomes ρ 5 ,J ( V ( J )) = ρ 5 ,J ( W ) ≤ 8 -6 = 2 < y k , a contradiction to the choice of J . So J ′′ is k -Ore and J ′′ = J ′ .
This also implies that N a ( ) ∩ N b ( ) = { x, y } , because otherwise, since J ′′ = J ′ , we would have gained an extra edge when we undo the merge of a and b into a ∗ b , which would increase | E J W ( [ ]) | - | E J ( ′′ [ W ′ ]) | by 1 in (43) yielding the same contradiction.
Since d c ( ) ≥ 6, J ′ cannot be K 5 . Let the separating set { u, v } , vertex subsets A = A J , u, v ( ′ ) and B = B J , u, v ( ′ ), and graphs ˜ J ′ ( u, v ) and ˇ ( J ′ u, v ) be as in Fact 14. By Fact 15.3, ρ k,J ′ ( A ) = ( k +1)( k -2). By Corollary 39, proper subgraphs of J have strictly larger potential than ( k +1)( k -2) and so a ∗ b ∈ A . Therefore J ′ [ B -u -v ] ⊂ J and so V ( ˇ ) J ′ -V ( J ) = u ∗ v . Hence by Corollary 39, ρ k,J ˇ ( ′ u,v ) ( U ) = ρ k,J ( U ) > k ( +1)( k -2) for every U ⊆ V ( ˇ ) J ′ -u ∗ v with | U | ≥ 2. Then by Claim 20, there exists an S ⊆ V ( ˇ ( J ′ u, v )) -u ∗ v such that ˇ ( J ′ u, v )[ S ] ∼ = K k -1 , and d ˇ ( J ′ u,v ) ( w ) = k -1 = 4 for all w ∈ S .
We claim that
$$\ e a c h { \ v e r t e x \ i n \ S - c } \ h a s \ d e g r e e \ k - 1 \ i n \ J.$$
Indeed, how it is possible that a vertex w ∈ S has a larger degree in J than in ˇ ? J ′ By Fact 15.7, d ˇ J ′ ( w ) = d J ′ ( w ). Because N a ( ) ∩ N b ( ) = { x, y } , the only vertices whose degree in J could be greater than in J ′ are a, b, c . But we already showed that a ∗ b / V ∈ ( ˇ ) J ′ -u ∗ v and S ⊆ V ( ˇ ( J ′ u, v )) -u ∗ v , so S ∩ { a, b, c } ⊆ { } c . This proves (44).
By Fact 32 and (44), every vertex in S -c is in a cluster. Because S is a ( k -1)-clique, by Claim 33 there is only one cluster in S , so the claim implies that T = S -c is a cluster and | T | = 3, which contradicts Claim 33 that each cluster in J has size at most k -3 = 2. glyph[square]
Claim 47 Each K 4 -subgraph of J contains at most one vertex with degree 4 . Furthermore, if d x ( ) = d y ( ) = 4 and xy ∈ E J ( ) , then each of x and y is in a K 4 .
Proof. Each vertex of degree 4 is in a cluster by definition, and by Claim 33, each K 4 contains only one cluster. The first statement of our claim then follows from Claim 46 and the second from Claim 42. glyph[square]
Definition 48 Let H ⊆ V ( J ) be the set of vertices of degree 5 not in a K 4 , and L ⊆ V ( J ) be the set of vertices of degree 4 not in a K 4 . Set glyph[lscript] = | L | , h = | H | and e 0 = | E L,H ( ) | .
Claim 49 e 0 ≤ 3 h + glyph[lscript] .
Proof. This is trivial if h + glyph[lscript] ≤ 2. By Claim 47, L is independent. So the claim follows by Lemma 44(ii) with A = L and B = H . glyph[square]
We will now use discharging to show that | E J ( ) | ≥ 9 4 n , which will finish the proof to the case k = 5. Let every vertex v ∈ V ( J ) have initial charge d v ( ). The discharging has one rule:
Rule R1: Each vertex in V ( J ) -H with degree at least 5 gives charge 1 / 6 to each neighbor.
We will show that the charge of each vertex in V ( J ) -H -L is at least 4 5, and then show that . the average charge of the vertices in H ∪ L is at least 4 5. .
Claim 50 After discharging, each vertex in V ( J ) -H -L has charge at least 4 5 . .
Proof. Let v ∈ V ( J ) -H -L . If d v ( ) = 4 and v / L ∈ , then v is in a K 4 and by Claim 47 v receives charge 1 / 6 from at least 3 neighbors and gives no charge. If d v ( ) = 5 and v / H ∈ , then v is in a K 4 and by Claim 47 N v ( ) contains at least 2 vertices with degree at least 5. Therefore v gives charge 1 6 to 5 neighbors, but receives charge 1 / / 6 from at least 2 neighbors. If d v ( ) ≥ 6, then v is left with charge at least 5 d v / ( ) 6 ≥ 4 5. . glyph[square]
Claim 51 After discharging, the sum of the charges on the vertices in H ∪ L is at least 4 5 . | H ∪ L | .
Proof. By Claim 47, if v ∈ L then every vertex in N v ( ) has degree at least 5. By Rule R1, the vertices in L receive from outside of H ∪ L the charge at least 1 6 (4 glyph[lscript] - | E H,L ( ) ). | By Claim 49, | E H,L ( ) | ≤ 3 h + . So, the total charge on glyph[lscript] H ∪ L is at least
$$5 h + 4 \ell + \frac { 1 } { 6 } ( 4 \ell - ( 3 h + \ell ) ) = 4. 5 ( h + \ell ),$$
as claimed. glyph[square]
Combining Claims 50 and 51, the total charge is at least 9 2 n . Thus the degree sum of V ( J ) is at least 9 2 n , and so | E J ( ) | ≥ 9 4 | V ( J ) . | glyph[square]
## 5.3 Case k ≥ 6
In this subsection we prove Theorem 10 for k ≥ 6. We will prove that | E J ( ) | ≥ ( k +1)( k -2) 2( k -1) | V ( J ) , | which will imply that ρ k,J ( V ( J )) ≤ 0 ≤ y k = k 2 -5 k +2. This proof will involve several claims.
Claim 52 Suppose k ≥ 6 , X is a ( k -1) -clique, and v ∈ X has degree k -1 . Then X contains at least ( k -1) / 2 vertices with degree at least k +1 .
Proof. Let { u } = N v ( ) -X . Assume that X contains at least k/ 2 vertices with degree at most k . By Claim 40, | N u ( ) ∩ X < k/ | 2, so there exists a w ∈ X such that uw / E J ∈ ( ) and d w ( ) ≤ k . By Claim 29, d w ( ) = k , so assume N w ( ) -X = { a, b } . Let J ′ be obtained from J -v by adding edges ua and ub if they do not already exist.
Suppose J ′ has a ( k -1)-coloring f . If f ( u ) is not used on X -w -v , then we recolor w with f ( u ). So, v will have at least two neighbors of color f ( u ), and we can extend the ( k -1)-coloring to v .
Thus J ′ is not ( k -1)-colorable and so contains a k -critical subgraph J ′′ . Let W = V ( J ′′ ). By Corollary 9, ρ k,J ′ ( W ) ≤ k k ( -3) and so
$$\rho _ { k, J } ( W ) \leq k ( k - 3 ) + 2 ( k - 1 ) ( 2 ) = k ^ { 2 } + k - 4 < 2 ( k - 2 ) ( k - 1 ).$$
glyph[negationslash]
If W = V ( J ′ ) then this contradicts Lemma 38, since in this case | W | ≤ | V ( J ′ ) | -1 ≤ n -2. So, W = V ( J ′ ).
If J ′′ is not a k -Ore graph, then by the minimality of J , ρ k,J ′′ ( W ) ≤ y k , and since adding edges only reduces potential, we have ρ k,J ′ ( W ) ≤ ρ k,J ′′ ( W ), and so
$$\rho _ { k, J } ( V ( J ) ) \leq \rho _ { k, J ^ { \prime } } ( W ) + ( k - 2 ) ( k + 1 ) ( 1 ) - 2 ( k - 1 ) ( k - 3 ) < y _ { k }$$
when k ≥ 6. This contradicts our choice of J , and so J ′′ is k -Ore.
glyph[negationslash]
Suppose J ′′ = J ′ . Since W = V ( J ′ ) = V ( J ′′ ), it follows that ρ k,J ′ ( W ) ≤ ρ k,J ′′ ( W ) -2( k -1), which leads to the same contradiction because by Fact 11.8, ρ k,J ′′ ( W ) -2( k -1) ≤ k k ( -3) -2( k -1) ≤ y k . So, our case is that J ′′ is a k -Ore graph and J ′′ = J ′ , i.e., J ′ is a k -Ore graph. Therefore ρ k,J ′′ ( W ) = k k ( -3) by Fact 11.10.
Since J ′ -ua -ub is a subgraph of J , by Corollary 39, ρ k,J ′ ( U ) = ρ k,J ( U ) > k ( +1)( k -2) for every U ⊆ V ( J ′ ) -u with | U | ≥ 2. Then by Claim 20, there exists an
$$S \subseteq V ( J ^ { \prime } ) - u \ s u c h \ t h a t \ J ^ { \prime } [ S ] \cong K _ { k - 1 }, \ a n d \ d _ { J ^ { \prime } } ( v ) = k - 1 \ f o r \ a l l \ v \in S. \quad \quad ( 4 5 )$$
Note that for every z ∈ S -a -b -N J ( v ), we have d J ( z ) = d J ′ ( z ) = k -1.
Suppose z ∈ S ∩ N J ( v ). Then since N J ( v ) = { u }∪ X and u / S ∈ , we have z ∈ X ∩ S . So z / ∈ { a, b } and therefore N J ′ ( z ) = N J ( z ) ∪{ } v , and thus d J ( z ) ≥ | X ∪ S | -1. Because | S | = k -1, there exists a z ′ ∈ S -X . By construction of J ′ and because S is a clique, we have that X ∩ S ⊆ N J ( z ′ ) ∩ X . By Claim 40, | X ∩ S | ≤ | N J ( z ′ ) ∩ X | ≤ ( k -1) / 2. So for k ≥ 6,
$$d _ { J ^ { \prime } } ( z ) = d _ { J } ( z ) - 1 \geq | X \cup S | - 2 \geq \lceil \frac { k - 1 } { 2 } \rceil + ( k - 1 ) - 2 \geq k,$$
a contradiction to (45). Thus, S ∩ N v ( ) = ∅ .
By Fact 32 and Claim 33, the vertices in S -a -b are part of one cluster in J with size at least k -1 -2 in the ( k -1)-clique S , which contradicts Claim 40 since k -3 > k -1 2 for k ≥ 6. glyph[square]
Claim 53 If k = 6 and a cluster T is contained in a 5 -clique X , then | T | = 1 .
Proof. By Claim 40, assume that T = { v , v 1 2 } . Let N v ( 1 ) -X = { y } and T -X = { u, u , u ′ ′′ } . By Claim 52, d u , d u ( ) ( ′ ) , d ( u ′′ ) ≥ 7. By Claim 40, | N y ( ) ∩ X < k/ | 2, and by the definition of a cluster, { v , v 1 2 } ⊂ N y ( ). Thus, N y ( ) ∩ { u, u , u ′ ′′ } = . Obtain ∅ J ′ from J -T by gluing u to y .
Suppose that J ′ has a 5-coloring. Then we can extend this coloring to a 5-coloring of J by greedily assigning colors to T , because only 3 different colors appear on the set { u, u , u ′ ′′ , y } . So we may assume that χ J ( ′ ) ≥ 6. Then J ′ contains a 6-critical subgraph J ′′ . Let W = V ( J ′′ ). Then by Corollary 9, ρ 6 ,J ′ ( W ) ≤ 6(6 -3) = 18. Since J ′′ is not a subgraph of J because J itself is critical, u ∗ y ∈ W . Let t = |{ u , u ′ ′′ } ∩ W | .
Case 1: t = 0. Then
$$\rho _ { 6, J } ( W - u * y + y + X ) \leq 1 8 + 2 8 ( 5 ) - 1 0 ( 1 2 ) = 3 8,$$
By Lemma 38, | W -u ∗ y + y + X | ≥ n -1. The bound (46) did not account for the edges in E ( { u , u ′ ′′ } , V ( J ) -X ), but each of u , u ′ ′′ has at least 3 neighbors outside of X . Thus we strengthen (46) to ρ 6 ,J ( W -u ∗ y + y + X ) ≤ 38 -10 · 4 < 0.
Denote R = W -u ∗ y + y + u + T .
Case 2: t = 1. Then ρ 6 ,J ( R ) ≤ 18+28(3) -10(7) = 32. By Lemma 38, this yields | R | ≥ n -1, so R is either V ( J ) -u ′ or V ( J ) -u ′′ . But because d u ( ′ ) , d ( u ′′ ) ≥ 7 = k + 1, Fact 22 says that ρ 6 ,J ( V ( J ) -u ′ ) , ρ 6 ,J ( V ( J ) -u ′′ ) > y k + k 2 + k = 50, a contradiction.
Case 3: t = 2. Then | R | ≥ 4 and
$$\rho _ { 6, J } ( R ) = \rho _ { k, J ^ { \prime \prime } } ( W ) + 2 8 \left ( | R | - | W | \right ) - 1 0 \left ( | E ( J [ R ] ) | - | E ( J ^ { \prime \prime } [ W ] ) | \right ) \leq 1 8 + 2 8 ( 3 ) - 1 0 ( 9 ) = 1 2. \ ( 4 7 )$$
Since | R | ≥ 4, by Lemma 38, | R | ≥ n -1. By Fact 22, every ( n -1)-element set has potential more than y k + k 2 -3 k +4 = 26. So | R | = n and R = W -u ∗ y + y + u + T = V ( J ).
If J ′′ is not k -Ore, then by the minimality of J , ρ k,J ′′ ( W ) ≤ y k = 8. In this case, we can replace in (47) the upper bound of 18 on ρ k,J ′′ ( W ) with 8, which yields ρ 6 ,J ( R ) ≤ 2 < y 6 , a contradiction. Thus J ′′ is k -Ore.
glyph[negationslash]
Similarly, if J ′′ = J ′ , then because V ( J ′ ) = V ( J ′′ ) we can replace in (47) the term -10(9) with -10(10) as there is an extra edge in | E J R ( [ ]) | - | E J ( ′′ [ W ]) , | which again yields ρ 6 ,J ( R ) ≤ 2 < y 6 . So J ′ = J ′′ .
Since J ′′ -u ∗ y is a subgraph of J , by Corollary 39, ρ k,J ′ ( U ) = ρ k,J ( U ) > k ( +1)( k -2) for every U ⊆ V ( J ′ ) -u ∗ y with | U | ≥ 2. Then by Claim 20, there exists an S ⊆ V ( J ′ ) -u ∗ y such that J ′ [ S ] ∼ = K 5 , and d J ′ ( v ) = 5 for all v ∈ S . By construction, S ⊆ V ( J ), so J S [ ] ∼ = K 5 . By Fact 32 each vertex with degree k -1 in J is in a cluster, by Claim 33 the S -cluster is unique, and by Claim 40, S has at most 2 vertices of degree k -1 in J . So there are at least three vertices z , z 1 2 , z 3 ∈ S such that d J ( z i ) > d J ′ ( z i ) for 1 ≤ i ≤ 3. But the vertices with larger degrees in J than in J ′ could be of only two types: (a) those in N J [ v 1 ] = N J [ v 2 ] = { y, v 1 , v 2 , u, u , u ′ ′′ } and (b) those in N J ( y ) ∩ N J ( u ). Because J ′′ -u ∗ y does not contain T = { v , v 1 2 } (which was deleted), and u ∗ y / S ∈ , we have at most two vertices of type (a). Then we have a vertex of type (b), but in this case we have an extra edge whose potential was not accounted in (47). So in this case we again instead of (47) get ρ 6 ,J ( R ) ≤ 12 -10 < y 6 , a contradiction. glyph[square]
Definition 54 We partition V ( J ) into four classes: L 0 , L 1 , H 0 , and H 1 . Let H 0 be the set of vertices with degree k , H 1 be the set of vertices with degree at least k +1 , and H = H 0 ∪ H 1 . Let
$$L = \{ u \in V ( J ) \, \colon d ( u ) = k - 1 \},$$
$$L _ { 0 } = \{ u \in L \, \colon N ( u ) \subseteq H \},$$
and
$$L _ { 1 } = L - L _ { 0 }.$$
Set glyph[lscript] = | L 0 | , h = | H 0 | and e 0 = | E L ,H ( 0 0 ) | .
Claim 55 e 0 ≤ 2( glyph[lscript] + h ) .
Proof. This is trivial if h + glyph[lscript] ≤ 2. By definition, L 0 is independent. The claim follows by applying Lemma 44(i) for A = L and B = H for h + glyph[lscript] ≥ 3. glyph[square]
Let every vertex v ∈ V ( J ) have initial charge d v ( ). Our discharging has two rules:
Rule R1: If a vertex in H 1 has neighbors of degree k -1, then it keeps for itself charge k -2 ( / k -1) and distributes the rest equally among its neighbors of degree k -1. Otherwise, it keeps all its charge for itself.
Rule R2: If a K k -1 -subgraph X contains exactly s ( k -1)-vertices adjacent to a ( k -1)-vertex x outside of X and not in a K k -1 , then each of these s vertices gives charge k -3 s k ( -1) to x .
Claim 56 Each vertex in H 1 gives to each neighbor of degree k -1 charge at least 1 k -1 .
Proof. If v ∈ H 1 , then v gives to each neighbor charge at least ψ d v ( ( )) := d v ( ) -k +2 ( / k -1) d v ( ) . Since ψ x ( ) is monotonically increasing for x ≥ k , ψ d v ( ( )) is minimized when d v ( ) = k +1. Then each neighbor of v of degree k -1 gets charge at least (1 + 2 / k ( -1)) / k ( +1) = 1 ( / k -1). glyph[square]
Claim 57 Each vertex in L 1 has charge at least k -2 ( / k -1) .
Proof. Let v ∈ L 1 . By Fact 32 every vertex in L ⊇ L 1 is in exactly one cluster. Let v be in a cluster C of size t .
By definition, if v gives anything out to some vertex x by Rule R2, then v is in a ( k -1)-clique X . In this case, since v has k -2 neighbors in X x , is the unique neighbor of v outside of X . Also, by Claim 33, C ⊂ X and by the definition of a cluster, each w ∈ C is adjacent to x . So
$$\text{if $v$ gives anything out by Rule $2$, then it gives it at most once and for $s=t$.}$$
Case 1: v is in a ( k -1)-clique X and t ≥ 2. Since v / H ∈ 1 , it can give charge only by Rule R2. Since v is in a ( k -1)-clique, it may receive charge only by Rule R1.
Recall that C ⊂ X . By Claim 41, each vertex in X -C has degree at least k -1 + t . Since t ≥ 2, this means X -C ⊆ H 1 . Furthermore, each vertex in X -C has at least k -2 -t neighbors with degree at least k (the other vertices of X -C ). Therefore each vertex u ∈ X -C gives by Rule R1 charge at least d u ( ) -k +2 ( / k -1) d u ( ) -k +2+ t to each neighbor of degree k -1. Note that this function increases as d u ( ) increases, so the charge is minimized when d u ( ) = k -1 + t . It follows that u gives to v charge at least t -1+2 ( / k -1) 2 +1 t .
So, v receives from the vertices in X -C by Rule R1 at least ( k -1 -t )( t -1+2 ( / k -1) 2 +1 t ) and in view of (48) gives out by Rule R2 to at most one vertex and with s = . Thus t v has charge at least k -1 + ( k -1 -t )( t -1+2 ( / k -1) 2 +1 t ) -k -3 t ( k -1) , which we claim is at least k -2 ( / k -1). Let
$$g _ { 1 } ( t ) = ( k - 1 - t ) ( ( t - 1 ) ( k - 1 ) + 2 ) - ( 2 t + 1 ) ( k - 3 ) ( 1 + \frac { 1 } { t } ).$$
We claim that g 1 ( ) t ≥ 0, which is equivalent to v having charge at least k -2 / k ( -1). Let
$$\widetilde { g } _ { 1 } ( t ) = ( k - 1 - t ) ( ( t - 1 ) ( k - 1 ) + 2 ) - ( 2 t + 1 ) ( k - 3 ) ( 3 / 2 ).$$
Note that ˜ g 1 ( ) t ≤ g 1 ( ) t when t ≥ 2, so it is enough to show that ˜ g 1 ( ) t ≥ 0 on the appropriate domain. Function ˜ g 1 ( ) t is quadratic with a negative coefficient at t 2 , so it suffices to check its values at the boundaries. They are
$$\widetilde { g } _ { 1 } ( 2 ) = ( k - 3 ) ( k - 6. 5 )$$
and
$$4 \widetilde { g } _ { 1 } ( \frac { k - 1 } { 2 } ) \ & = \ \ ( k - 1 ) \left ( ( k - 3 ) ( k - 1 ) + 4 \right ) - 6 k ( k - 3 ) \\ & = \ k ^ { 3 } - 1 1 k ^ { 2 } + 2 9 k - 7 \\ & = \ \ ( k - 7 ) ( k ^ { 2 } - 4 k + 1 ).$$
Each of these values is non-negative when k ≥ 7, and if k = 6, then the case does not apply by Claim 53.
Case 2: t ≥ 2 and v is not in a ( k -1)-clique. Then v cannot give charge by Rule R2 and may receive charge by Rule R1. By Claim 43, each neighbor of v outside of C has degree at least k -1 + t ≥ k +1 and is in H 1 . Therefore v has charge at least k -1 + ( k -t )( t -1+2 ( / k -1) k -1+ t ). We define
$$g _ { 2 } ( t ) \ & = \ ( k - t ) ( t - 1 + \frac { 2 } { k - 1 } ) - \frac { k - 3 } { k - 1 } ( k - 1 + t ) \\ & = \ t ( k - t ) - 2 ( 1 - \frac { 2 } { k - 1 } ) ( k - 1 ) \\ & = \ t ( k - t ) - 2 ( k - 3 ).$$
Note that g 2 ( ) t ≥ 0 is equivalent to v having charge at least k -2 ( / k -1). The function g 2 ( ) is t quadratic with a negative coefficient at t 2 , so it suffices to check its values at the boundaries. They are
$$g _ { 2 } ( 2 ) = 2 ( k - 2 ) - 2 ( k - 3 ) = 2$$
and
$$g _ { 2 } ( k - 3 ) = ( k - 3 ) ( 3 ) - 2 ( k - 3 ) = k - 3.$$
Each of these values is positive.
glyph[negationslash]
Case 3: t = 1. By the definition of L 1 , v is adjacent to at least one vertex w with degree k -1. Since | C | = t = 1 and so C = { v } , w / C ∈ . So by Fact 32, v and w are in different clusters. This means N w [ ] = N v [ ]. If v is not in a ( k -1)-clique X , then by Claim 42, w is in a ( k -1)-clique and a cluster of size at least 2. In this case v will receive charge ( k -3) / k ( -1) in total from the cluster containing w using Rule R2, and will not give away any charge. Therefore we may assume that v is in a ( k -1)-clique X .
By Claim 52, there exists a Y ⊂ X such that | Y | ≥ k -1 2 and every vertex in Y has degree at least k +1. By Claim 33, none of the vertices in X -C = X -{ } v is in a cluster, which by Fact 32 means that every vertex in X -{ } v has degree at least k . So each vertex in Y has at least k -3 neighbors with degree at least k (the vertices of X besides v and itself). Therefore by Rule R1 each vertex u ∈ Y donates charge at least g 3 ( d u ( )) = d u ( ) -k +2 ( / k -1) d u ( ) -k +3 to each neighbor of degree k -1. Since d 3 ( d u ( )) increases as d u ( ) increases, the charge is minimized when d u ( ) = k +1. It follows that u gives to v charge at least d 3 ( k +1) = 1+2 ( / k -1) 4 , and v has charge at least
$$k - 1 + \frac { k - 1 } { 2 } \left ( \frac { 1 + 2 / ( k - 1 ) } { 4 } \right ) = k + \frac { k - 7 } { 8 },$$
which is at least k -2 ( / k -1) when k ≥ 6. glyph[square]
By the discharging rules and Claims 56 and 57, after discharging,
- a) the charge of each vertex in H 1 ∪ L 1 is at least k -2 ( / k -1);
- b) the charges of vertices in H 0 did not decrease;
- c) along every edge from H to L the charge at least 1 / k ( -
- 1 0 1) is sent.
Thus by Claim 55, the total charge F of the vertices in H 0 ∪ L 0 is at least
$$k h + ( k - 1 ) \ell + \frac { 1 } { k - 1 } \left ( \ell ( k - 1 ) - e _ { 0 } \right ) \geq k ( h + \ell ) - \frac { 1 } { k - 1 } 2 ( h + \ell ) = ( h + \ell ) \left ( k - \frac { 2 } { k - 1 } \right ),$$
and so by a), the total charge of all the vertices of J is at least n ( k -2 k -1 ) . Therefore the degree sum of J is at least n ( k -2 k -1 ) = ( ( k +1)( k -2) k -1 ) n , i. e., | E J ( ) | ≥ ( ( k +1)( k -2) 2( k -1) ) n . glyph[square]
## 6 Sharpness
First we prove Corollary 7, and then we will construct sparse 3-connected k -critical graphs. As it was pointed out in the introduction, Construction 58 and infinite series of 3-connected sparse 4- and 5-critical graphs are due to Toft [33] (based on [32]).
Proof of Corollary 7. By (1), if we construct an n 0 -vertex k -critical graph for which our lower bound on f k ( n 0 ) is exact, then the bound on f k ( n ) is exact for every n of the form n 0 + ( s k -1). So, by Corollary 5, we only need to construct
- · a 5-critical 7-vertex graph with ⌈ 15 1 2 ⌉ = 16 edges,
- · a 5-critical 8-vertex graph with ⌈ 17 3 4 ⌉ = 18 edges,
- · a 6-critical 10-vertex graph with ⌈ 27 1 5 ⌉ = 28 edges,
- · a 6-critical 12-vertex graph with ⌈ 32 4 5 ⌉ = 33 edges, and
- · a 7-critical 14-vertex graph with ⌈ 45 1 3 ⌉ = 46 edges.
These graphs are presented in Figure 1. glyph[square]
Figure 1: Minimal k -critical graphs.
Construction 58 (Toft [33]) Let G be a k -critical graph, e = uv ∈ E G ( ) , and w ∈ V ( G ) -{ u, v } be such that for all ( k -1) -colorings φ of G -e , φ w ( ) = φ u ( ) = φ v ( ) . Let S 1 ∪ S 2 ∪ S 3 be a partition of the vertex set X of a copy of K k -1 such that each S i is non-empty. We define G ′ as follows: V ( G ′ ) = V ( G ) ∪ V ( X ) and E G ( ′ ) = ( E G ( ) -e ) ∪ E X ( ) ∪ E ′ , where
$$E ^ { \prime } = \{ u a \, \colon a \in S _ { 1 } \} \cup \{ v b \, \colon b \in S _ { 2 } \} \cup \{ w c \, \colon c \in S _ { 3 } \}.$$
Claim 59 If k ≥ 4 , G is a 3 -connected k -critical graph and G ′ is created from G via Construction 58, then G ′ is a 3 -connected k -critical graph.
Proof. We will use the names and definitions from Construction 58.
glyph[negationslash]
If there exists a ( k -1)-coloring φ of G ′ , then all k -1 colors must appear on X . Then φ u ( ) appears on a vertex in S 2 or S 3 . But then either φ v ( ) = φ u ( ) or φ w ( ) = φ u ( ), which contradicts the assumptions of Construction 58. So χ G ( ′ ) ≥ k .
glyph[negationslash]
glyph[negationslash]
Suppose there exists an f ∈ E G ( ′ ) such that χ G ( ′ -f ) ≥ k . If f ∈ E G ( ), then let φ 1 be a ( k -1)-coloring of G -f . Because e ∈ E G ( ) -f , φ 1 ( u ) = φ 1 ( v ), and so φ 1 extends easily to G ′ -f . If f ⊂ X , then a ( k -1)-coloring of G -e can be extended to G ′ -f , because X can be colored with k -2 colors, while N X ( ) = { u, v, w } is colored with 1 color. If f ∈ E ′ , then a ( k -1)-coloring of G -e extends to G ′ -f , because the unique color on { u, v, w } can be given to f ∩ X . Therefore G ′ is k -critical.
glyph[negationslash]
Suppose now that there exists a set S such that | S | < 3 and there are nonempty A and B such that E A,B ( ) = ∅ and A ∪ B ∪ S = V ( G ′ ). Because k -critical graphs are 2-connected, | S | = 2. By Fact 12, S is independent. Because X is a clique, without loss of generality X glyph[subsetnoteql] A ∪ S . By construction, A -X = ∅ , so S also separates G -e . Since κ G ( ) ≥ 3, S ∩ X = ∅ and e has an endpoint in each component of G -S -e . So we may assume that u ∈ A and v ∈ B . Since v has a neighbor in S 2 ⊂ X , this contradicts to S ∩ X = ∅ and X glyph[subsetnoteql] A ∪ S . glyph[square]
The assumptions in Construction 58 are strong. Most edges e in k -critical graphs do not have such a vertex w , and some k -critical graphs do not have any edge-vertex pairs ( e, w ) that satisfy the assumptions. We will construct an infinite family G k of sparse 3-connected graphs that do satisfy the assumptions.
The family is generated for each k by finding a small 3-connected k -critical graph G ′ k such that ρ k ( G ′ k ) = y k . We will describe a subgraph H ′ k ≤ G ′ k with two vertices, u and w , such that in any ( k -1)-coloring φ ′ of H ′ k , φ ′ ( u ) = φ ′ ( w ). Construction 58 can then be applied to G ′ k , using any edge e incident to u that is not in H ′ k and not incident to w . Because Construction 58 does not decrease the degree of u , this process can be iterated indefinitely to populate G k .
Note that Construction 58 adds the same number of vertices and edges as DHGO-composition with G 2 = K k . Therefore every graph G ∈ G k has ρ k ( G ) = y k . Furthermore, G is also k -critical and 3-connected, and therefore not k -Ore. This implies the sharpness of Theorem 6.
All that is left is to find suitable graphs for G ′ k and H ′ k . Figure 2 illustrates G ′ 4 and G ′ 5 . We will need a second construction for larger k .
## Construction 60 Fix a t such that 1 ≤ t < k/ 2 . Let
$$V ( H _ { k, t } ) = \{ u _ { 1 }, u _ { 2 }, \dots, u _ { k - 1 }, v _ { 1 }, v _ { 2 }, \dots, v _ { k - 1 }, w \}$$
and
$$E ( H _ { k, t } ) = \{ u _ { i } u _ { j } \, \colon 1 \leq i < j \leq k - 1 \} \cup \{ v _ { i } v _ { j } \, \colon 1 \leq i < j \leq k - 1 \} \cup \{ u _ { i } v _ { j } \, \colon i, j \leq t \}$$
v
v
Figure 2: Graphs G ′ 4 and G ′ 5 , with substructures labeled for constructing G 4 and G 5 .
$$\cup \{ w u _ { i } \colon i > t \} \cup \{ w v _ { i } \colon i > t \}.$$
By construction, H k, 1 is a k -Ore graph, H k,t is k -critical, κ H ( k,t ) = t +1, | V ( H k,t ) | = 2 k -1, and | E H ( k,t ) | = k k ( -1) -2 t + t 2 . Moreover, ρ k ( H k, 2 ) = y k . For k ≥ 6, we choose G ′ k = H k, 2 . We will next find H ′ k for k ≥ 6, which will complete the argument.
Claim 61 Let H ′ k = H k, 2 -u v 1 1 . Then in every ( k -1) -coloring φ ′ of H ′ k , φ ′ ( u 1 ) = φ ′ ( v 1 ) = φ ′ ( w ) .
Proof. Let φ ′ be a ( k -1)-coloring of H ′ k . Note that each of the ( k -1) colors appears both, on { u , u 1 2 , . . . , u k -1 } and on { v , v 1 2 , . . . , v k -1 } . Then φ ′ ( w ) appears on a vertex a ∈ { u , u 1 2 } and again on a vertex b ∈ { v , v 1 2 } . So ab / E G ∈ ( ), which implies that a = u 1 and b = v 1 . glyph[square]
u
Figure 3: An example of a graph in G 4 .
## 7 Algorithm
The proof of Theorem 4 was constructive, and provided an algorithm for ( k -1)-coloring of sparse graphs. Let P k ( G ) be the minimum of ρ k,G ( W ) over all W ⊆ V ( G ) with 2 ≤ | W | .
Theorem 62 ([26]) If k ≥ 4 , then every n -vertex graph G with P k ( G > k k ) ( -3) can be ( k -1) -colored in O k ( 3 5 . n 6 5 . log( n )) time.
We present below a polynomial-time algorithm for checking whether a given graph is a k -Ore graph. Together with an analog of the algorithm in Theorem 62 that uses the proof of Theorem 6 instead of Theorem 4, it would yield a polynomial-time algorithm that for every n -vertex graph G with P k ( G > y ) k either finds a ( k -1)-coloring of G or finds a subgraph of G that is a k -Ore graph.
Our algorithm to determine whether an n -vertex graph G is k -Ore is simple:
- 0. If G is K k , return 'yes.'
- 1. Check whether n ≡ 1 ( mod k -1) and | E G ( ) | = ( k +1)( k -2) | V ( G ) |-k k ( -3) 2( k -1) . If not, then return 'no.'
- 2. Check whether the connectivity of G is exactly 2. If not, then return 'no.' Otherwise, choose a separating set { x, y } .
- 3. If G - -x y has more than two components or xy ∈ E G ( ), then return 'no.' Otherwise, let A and B be the vertex sets of the two components of G - -x y . If {| A | ( mod k -1) , | B | ( mod k -1) } glyph[negationslash] = { k -2 0 , } , then return 'no'. Otherwise, rename A and B so that | A | ( mod k -1) = k -2 and | B | ( mod k -1) = 0.
- 4. Create graphs ˜ G x,y ( ) and ˇ ( G x,y ) as defined in Fact 14. Recurse on each of ˜ G x,y ( ) and ˇ ( G x,y ). If at least one recursion call returns 'no,' then return 'no.' Otherwise, return 'yes.'
The longest procedure in this algorithm is checking whether the connectivity of G is exactly 2 at Step 2, which has complexity O kn ( 3 ) because | E G ( ) | ≤ kn/ 2. And it will be called fewer than 2 n/ k ( -2) times. So the overall complexity is at most O n ( 4 ).
Acknowledgment. We thank Michael Stiebitz and Bjarne Toft for helpful discussions. We also thank the referees for many helpful comments significantly improving the presentation of the paper.
## References
- [1] V. Aksenov, Private communication (1976).
- [2] O. V. Borodin, Colorings of plane graphs: A survey, Discrete Math. , 313 (2013), 517-533.
- [3] O. V. Borodin, Z. Dvoˇ r´k, A. V. Kostochka, B. Lidick´, and M. Yancey, Planar 4-critical graphs with a y four triangles, European J. of Combinatorics 41 (2014), 138-151.
- [4] O. V. Borodin and A. V. Kostochka, Defective 2-colorings of sparse graphs, J. Combin. Theory (B) 104 (2014), 72-80.
- [5] O. V. Borodin, A. V. Kostochka, B. Lidick´ and M. Yancey, Short proofs of coloring theorems on planar y graphs, European J. of Combinatorics 36 (2014), 314-321.
- [6] O. V. Borodin, A. V. Kostochka and M. Yancey, On 1-improper 2-coloring of sparse graphs, Discrete Mathematics , 313 (2013), no. 22, 2638-2649.
- [7] G. A. Dirac, Note on the colouring of graphs, Math. Z. 54 (1951), 347-353.
- [8] G. A. Dirac, A property of 4-chromatic graphs and some remarks on critical graphs, J. London Math. Soc. 27 (1952), 85-92.
- [9] G. A. Dirac, Some theorems on abstract graphs, Proc. London Math. Soc. (3) 2 (1952), 69-81.
- [10] G.A. Dirac, The structure of k-chromatic graphs, Fund. Math. 40 (1953) 42-55.
- [11] G. A. Dirac, Map colour theorems related to the Heawood colour formula, J. London Math. Soc. 31 (1956), 460-471.
- [12] G. A. Dirac, A theorem of R. L. Brooks and a conjecture of H. Hadwiger, Proc. London Math. Soc. (3) 7 (1957), 161-195.
- [13] G.A. Dirac, On the structure of 5- and 6-chromatic abstract graphs, J. Reine Angew. Math. 214-215 (1964) 43-52.
- [14] G. A. Dirac, The number of edges in critical graphs, J. Reine Angew. Math. 268/269 (1974), 150-164.
- [15] B. Farzad and M. Molloy, On the edge-density of 4-critical graphs, Combinatorica 29 (2009), 665-689.
- [16] T. Gallai, Kritische Graphen I, Publ. Math. Inst. Hungar. Acad. Sci. 8 (1963), 165-192.
- [17] T. Gallai, Kritische Graphen II, Publ. Math. Inst. Hungar. Acad. Sci. 8 (1963), 373-395.
- [18] H. Gr¨ otzsch, Zur Theorie der diskreten Gebilde. VII. Ein Dreifarbensatz f¨r dreikreisfreie Netze auf u der Kugel. Wiss. Z. Martin-Luther-Univ. Halle-Wittenberg. Math.-Nat. Reihe 8 (1958/1959), 109-120 (in German).
- [19] G. Haj´ os, ¨ ber U eine Konstruktion nichtn -f¨rbbarer a Graphen, Wiss. Z. Martin-Luther-Unive. HalleWittenberg Math.-Natur. Reihe 10 (1961), 116-117.
- [20] S. L. Hakimi, On realizability of a set of integers as degrees of the vertices of a linear graph. I., J. Soc. Indust. Appl. Math. 10 (1962), 496-506.
- [21] T. R. Jensen and B. Toft, Graph Coloring Problems, Wiley-Interscience Series in Discrete Mathematics and Optimization, John Wiley & Sons, New York, 1995.
- [22] T. R. Jensen and B. Toft, 25 pretty graph colouring problems, Discrete Math. 229 (2001), 167-169.
- [23] H. Kierstead and L. Rabern, Personal communication.
- [24] A. V. Kostochka and M. Stiebitz, Excess in colour-critical graphs, in: Graph Theory and Combinatorial Biology, Balatonlelle (Hungary), 1996, Bolyai Society, Mathematical Studies 7 , Budapest, 1999, 87-99.
- [25] A. V. Kostochka and M. Stiebitz, A new lower bound on the number of edges in colour-critical graphs and hypergraphs, J. Comb. Theory, Series B. 87 (2003), 374-402.
- [26] A. V. Kostochka and M. Yancey, Ore's Conjecture on color-critical graphs is almost true, J. Comb. Theory, Series B. 109 (2014), 73-101.
- [27] A. V. Kostochka and M. Yancey, Ore's Conjecture for k = 4 and Gr¨tzsch Theorem, o Combinatorica 34 (2014), 323-329.
- [28] M. Krivelevich, On the minimal number of edges in color-critical graphs, Combinatorica 17 (1997), 401-426.
- [29] O. Ore, The Four Color Problem, Academic Press, New York, 1967.
- [30] L. Postle, Personal communication.
- [31] R. Steinberg, The state of the three color problem, Quo Vadis, Graph Theory?, Ann. Discrete Math. , 55 (1993), 211-248.
- [32] B. Toft, Color-critical graphs and hypergraphs, J. Combin. Theory 16 (1974), 145-161.
- [33] B. Toft, Personal communication.
- [34] Zs. Tuza, Graph coloring, in: Handbook of graph theory. J. L. Gross and J. Yellen Eds., CRC Press, Boca Raton, FL, 2004. xiv+1167 pp.
| null |
[
"Alexandr Kostochka",
"Matthew Yancey"
] |
2014-08-05T01:12:33+00:00
|
2017-04-04T13:23:43+00:00
|
[
"math.CO",
"05C15, 05C35"
] |
A Brooks-type result for sparse critical graphs
|
A graph $G$ is $k$-{\em critical} if it has chromatic number $k$, but every
proper subgraph of $G$ is $(k-1)$--colorable. Let $f_k(n)$ denote the minimum
number of edges in an $n$-vertex $k$-critical graph. Recently the authors gave
a lower bound, $f_k(n) \geq \left\lceil
\frac{(k+1)(k-2)|V(G)|-k(k-3)}{2(k-1)}\right\rceil$, that solves a conjecture
by Gallai from 1963 and is sharp for every $n\equiv 1\,({\rm mod }\, k-1)$. It
is also sharp for $k=4$ and every $n\geq 6$. In this paper we refine the result
by describing all $n$-vertex $k$-critical graphs $G$ with $|E(G)|=
\frac{(k+1)(k-2)|V(G)|-k(k-3)}{2(k-1)}$. In particular, this result implies
exact values of $f_5(n)$ when $n\geq 7$.
|
1408.0847v2
|
## OPTIMALITY OF DOUBLY REFLECTED L ´ EVY PROCESSES IN SINGULAR CONTROL
## ERIK J. BAURDOUX ∗ AND KAZUTOSHI YAMAZAKI †
ABSTRACT. We consider a class of two-sided singular control problems. A controller either increases or decreases a given spectrally negative L´ evy process so as to minimize the total costs comprising of the running and controlling costs where the latter is proportional to the size of control. We provide a sufficient condition for the optimality of a double barrier strategy, and in particular show that it holds when the running cost function is convex. Using the fluctuation theory of doubly reflected L´ evy processes, we express concisely the optimal strategy as well as the value function using the scale function. Numerical examples are provided to confirm the analytical results.
AMS 2010 Subject Classifications: 60G51, 93E20, 49J40
Key words: singular control; doubly reflected L´ evy processes; fluctuation theory; scale functions
## 1. INTRODUCTION
We consider the problem of optimally modifying a stochastic process by means of singular control. An admissible strategy is two-sided and the process can be increased or decreased. The objective is to minimize the expected total costs comprising of the running and controlling costs; the former is modeled as some given function f of the controlled process that is accumulated over time, and the latter is proportional to the size of control. The problem of singular control arises in various contexts. For its applications, we refer the reader to, e.g., [14, 15] for inventory management, [19] for cash balance management, [5, 24] for monotone follower problems and [1, 6, 20, 30, 32, 36] for finance and insurance.
This paper studies a spectrally negative L´ evy model where the underlying process, in the absence of control, follows a general L´ evy process with only negative jumps. We pursue a sufficient condition on the running cost function f such that a strategy of double barrier type is optimal and the value function is obtained semi-explicitly. This generalizes the classical Brownian motion model [21] and complements the results on the continuous diffusion model as in [31].
Motivated by the recent research of spectrally negative L´ evy processes and their applications, we take advantage of their fluctuation theory as in [10, 28]. These techniques are used extensively in stochastic
This version: February 6, 2015.
∗ Department of Statistics, London School of Economics, Houghton Street, London, WC2A 2AE, UK. Email: [email protected].
† (corresponding author) Department of Mathematics, Faculty of Engineering Science, Kansai University, 3-3-35 Yamatecho, Suita-shi, Osaka 564-8680, Japan. Email: [email protected] . Tel: +81-6-6368-1527.
control problems in the last decade. Exemplifying examples include de Finetti's dividend problem as in [1, 6, 30], where a single barrier strategy is shown to be optimal under certain conditions. In these papers, the so-called scale function is commonly used to express the net present value of the barrier strategy. Thanks to its analytical properties such as continuity/smoothness (see, e.g., [23, 28]), the selection of the candidate barrier level and the verification of optimality can be carried out efficiently. While a part of the verification is still problem-dependent and is often a difficult task, these methods allow one to solve for this wide class of L´vy processes without specializing on a particular type, whether or not the process is e of infinite activity/variation.
This paper considers a variant of the above mentioned papers where the control is allowed to be twosided. Our objective is to show the optimality of a double barrier strategy where the resulting controlled process becomes a doubly reflected L´ evy process of [1, 34]. Existing research on the optimality of doubly reflected L´ evy processes includes the dividend problem with capital injection as in [1, 6]. Other related problems where two threshold levels characterize the optimal strategy include stochastic games [4, 3, 16, 22] and impulse control [7, 38].
In this paper, we take the following steps to achieve our goal:
- (1) We first write via the scale function the expected net present value corresponding to the double barrier strategy; this is a direct application of the results in [1, 34].
- (2) This is followed by the selection of the two barriers. The upper barrier is chosen so that the resulting candidate value function becomes twice differentiable at the barrier; the lower barrier is chosen so that it is continously (resp. twice) differentiable when the process is of bounded (resp. unbounded variation).
- (3) We then analyze the existence of such a pair that satisfy the two conditions simultaneously. We show that either such a pair exist, or otherwise a single barrier strategy (with the upper barrier set to infinity) is optimal.
- (4) In order to verify the optimality of the strategy defined in the previous steps, we study the verification lemma and identify some additional conditions that are sufficient for the optimality. Moreover, we show that it is satisfied whenever the running cost function f is convex.
As in the above mentioned papers, we use the special known properties of the scale function to solve the problem. In particular, the steps taken here are similar to those used in [22], where two parameters are shown to characterize the optimal strategies in the two-person game they considered. The main novelty and challenge here are that we solve the problem without specifying the form of the running cost function f and derive a most general condition on f that is sufficient for the optimality of a doubly reflected L´vy e process.
In addition to the above, we give examples with (piecewise) quadratic and linear cases for f , which have been used in, e.g., [2, 13, 37]. We shall see in particular that in the linear case the upper boundary can become infinity (or equivalently a single barrier strategy is optimal), whereas it does not occur in the
quadratic case. In order to confirm the obtained analytical results, we give numerical examples where the underlying process is a spectrally negative L´ evy process in the β -family of Kuznetsov [25].
The rest of the paper is organized as follows. Section 2 gives a mathematical model of the problem and a brief review on the spectrally negative L´vy process and the scale function. Section 3 expresses via e the scale function the expected net present value under the double barrier strategy. The candidate barrier levels are then selected by using the smoothness conditions at the barriers, and the existence of such a pair is shown. In Section 4, we study the verification lemma for this problem and analyze what additional conditions are required for the candidate value function to be optimal. Section 5 obtains a more concrete sufficient condition and in particular shows that it is satisfied when f is convex. In Section 6, we give examples with piecewise quadratic and linear cases. We conclude the paper with numerical examples in Section 7.
Throughout the paper, x + and x -are used to indicate the right and left hand limits, respectively. The superscripts x + := max( x, 0) , f + ( x ) := max( f ( x , ) 0) , x -:= max( -x, 0) and f -( x ) := max( -f ( x , ) 0) are used to indicate positive and negative parts. Monotonicity is understood in the strict sense; for the weak sense 'nondecreasing' and 'nonincreasing' are used. The convexity (unless otherwise stated) is in the weak sense.
## 2. MATHEMATICAL FORMULATION
Let (Ω , F , P ) be a probability space hosting a spectrally negative L´ evy process X = { X t t ; ≥ } 0 whose Laplace exponent is given by
$$\psi ( s ) \coloneqq \log \mathbb { E } \left [ e ^ { s X _ { 1 } } \right ] = c s + \frac { 1 } { 2 } \sigma ^ { 2 } s ^ { 2 } + \int _ { ( - \infty, 0 ) } ( e ^ { s z } - 1 - s z 1 _ { \{ - 1 < z < 0 \} } ) \nu ( \mathrm d z ), \quad s \geq 0,$$
where ν is a L´vy measure with the support e ( -∞ , 0) that satisfies the integrability condition ∫ ( -∞ , 0) (1 ∧ z 2 ) ν (d z ) < ∞ . It has paths of bounded variation if and only if σ = 0 and ∫ ( -1 0) , | z | ν (d z ) < ∞ ; in this case, we write (2.1) as
$$\psi ( s ) = \delta s + \int _ { ( - \infty, 0 ) } ( e ^ { s z } - 1 ) \nu ( \mathrm d z ), \ \ s \geq 0,$$
with δ := c -∫ ( -1 0) , z ν (d z ) . We exclude the case in which X is the negative of a subordinator (i.e., X has monotone paths a.s.). This assumption implies that δ > 0 when X is of bounded variation. Let P x be the conditional probability under which X 0 = x (also let P ≡ P 0 ), and let F := {F t , t ≥ } 0 be the filtration generated by X .
An admissible strategy π := { ( U , D π t π t ); t ≥ } 0 is given by a pair of nondecreasing, right-continuous and F -adapted processes such that U π 0 -= D π 0 -= 0 and, as is assumed in [21],
$$\mathbb { E } _ { x } \left [ \int _ { [ 0, \infty ) } e ^ { - q t } ( \mathrm d U _ { t } ^ { \pi } + \mathrm d D _ { t } ^ { \pi } ) \right ] < \infty, \quad x \in \mathbb { R }.$$
Let Π be the set of all admissible strategies and the discount q is assumed to be a strictly positive constant.
With the controlled process Y π t := X t + U π t -D π t , t ≥ 0 , the problem is to compute the total expected costs:
$$v ^ { \pi } ( x ) \coloneqq \mathbb { E } _ { x } \left [ \int _ { 0 } ^ { \infty } e ^ { - q t } f ( Y _ { t } ^ { \pi } ) \mathrm d t + \int _ { [ 0, \infty ) } e ^ { - q t } \left ( C _ { U } \mathrm d U _ { t } ^ { \pi } + C _ { D } \mathrm d D _ { t } ^ { \pi } \right ) \right ], \ \ x \in \mathbb { R },$$
for some running cost function f satisfying the conditions specified below and fixed constants C ,C U D ∈ R such that
$$C _ { U } + C _ { D } > 0,$$
$$\dots$$
and to obtain an admissible strategy over Π that minimizes it, if such a strategy exists. The inequality (2.3) is commonly assumed in the literature (see, e.g., [21, 31]); this implies that it is suboptimal to activate U π t and D π t simultaneously. Hence, we can safely assume that the supports of the Stieltjes measures d U π t ( ω ) and d D π t ( ω ) do not overlap for a.e. ω ∈ Ω .
Regarding the running cost function f , we assume the same assumptions as in [8, 9, 38]; this is a crucial condition when dealing with a process with negative jumps.
Assumption 2.1. We assume that f : R → R satisfies the following.
- (1) f is continuous and is a piecewise continuously differentiable function and grows (or decreases) at most polynomially (in the sense defined by Beyer et al. [11] ).
- (2) There exists a number a ∈ R such that the function
$$\tilde { f } ( x ) \coloneqq f ( x ) + C _ { U } q x, \ \ x \in \mathbb { R },$$
is increasing on ( a, ∞ ) and is decreasing and convex on ( -∞ , a ) .
- (3) There exist a c 0 > 0 and an x 0 ≥ a such that ˜ ( f ′ x ) ≥ c 0 for x ≥ x 0 .
For the problem to make sense, we assume that the L´ evy process X has a finite moment.
$$\text{Assumption 2.2.} \, \text{We assume that } \mathbb { E } [ X _ { 1 } ] = \psi ^ { \prime } ( 0 + ) \in ( - \infty, \infty ).$$
- 2.1. Scale functions. Fix q > 0 . For any spectrally negative L´ evy process, there exists a function called the q -scale function
$$W ^ { ( q ) } \, \colon \mathbb { R } \rightarrow [ 0, \infty ),$$
which is zero on ( -∞ , 0) , continuous and increasing on [0 , ∞ ) , and is characterized by the Laplace transform:
$$\int _ { 0 } ^ { \infty } e ^ { - s x } W ^ { ( q ) } ( x ) \text{d} x = \frac { 1 } { \psi ( s ) - q }, \quad s > \Phi ( q ),$$
where
$$\Phi ( q ) \coloneqq \sup \{ \lambda \geq 0 \, \colon \psi ( \lambda ) = q \}.$$
Here, the Laplace exponent ψ in (2.1) is known to be zero at the origin and strictly convex on [0 , ∞ ) ; therefore Φ( ) q is well defined and is strictly positive as q > 0 . We also define, for x ∈ R ,
$$q ) & \text{ is well defined and is strictly positive as } q > 0. \text{ We also define, for } x \in \mathbb { R }, \\ \overline { W } ^ { ( q ) } ( x ) & \coloneqq \int _ { 0 } ^ { x } W ^ { ( q ) } ( y ) d y, \\ Z ^ { ( q ) } ( x ) & \coloneqq 1 + q \overline { W } ^ { ( q ) } ( x ), \\ \overline { Z } ^ { ( q ) } ( x ) & \coloneqq \int _ { 0 } ^ { x } Z ^ { ( q ) } ( z ) d z = x + q \int _ { 0 } ^ { x } \int _ { 0 } ^ { z } W ^ { ( q ) } ( w ) d w d z. \\ q ) & \text{ is uniformly zero on the negative half line. we have }$$
Because W ( q ) is uniformly zero on the negative half line, we have
$$Z ^ { ( q ) } ( x ) = 1 \quad \text{and} \quad \overline { Z } ^ { ( q ) } ( x ) = x, \quad x \leq 0.$$
Let us define the first down- and up-crossing times, respectively, of X by
$$\tau _ { b } ^ { - } \coloneqq \inf \left \{ t \geq 0 \, \colon X _ { t } < b \right \} \quad \text{and} \quad \tau _ { b } ^ { + } \coloneqq \inf \left \{ t \geq 0 \, \colon X _ { t } > b \right \}, \quad b \in \mathbb { R }.$$
Then, for any b > 0 and x ≤ b ,
$$( 2. 6 ) \quad \mathbb { E } _ { x } \left [ e ^ { - q \tau _ { b } ^ { + } } 1 _ { \{ \tau _ { b } ^ { + } < \tau _ { 0 } ^ { - } \} } \right ] = \frac { W ^ { ( q ) } ( x ) } { W ^ { ( q ) } ( b ) } \quad \text{and} \quad \mathbb { E } _ { x } \left [ e ^ { - q \tau _ { 0 } ^ { - } } 1 _ { \{ \tau _ { b } ^ { + } > \tau _ { 0 } ^ { - } \} } \right ] = Z ^ { ( q ) } ( x ) - Z ^ { ( q ) } ( b ) \frac { W ^ { ( q ) } ( x ) } { W ^ { ( q ) } ( b ) }.$$
By taking limits on the latter,
$$\mathbb { E } _ { x } \left [ e ^ { - q \tau _ { 0 } ^ { - } } \right ] = Z ^ { ( q ) } ( x ) - \frac { q } { \Phi ( q ) } W ^ { ( q ) } ( x ), \ \ x \in \mathbb { R }.$$
Fix λ ≥ 0 and define ψ λ ( ) · as the Laplace exponent of X under P λ with the change of measure
$$\frac { d \mathbb { P } ^ { \lambda } } { d \mathbb { P } } \Big | _ { \mathcal { F } _ { t } } = \exp ( \lambda X _ { t } - \psi ( \lambda ) t ), \ \ t \geq 0 ;$$
see page 213 of [28]. Suppose W ( q ) λ and Z ( q ) λ are the scale functions associated with X under P λ (or equivalently with ψ λ ( ) · ). Then, by Lemma 8.4 of [28], W ( q -ψ λ ( )) λ ( x ) = e -λx W ( q ) ( x ) , x ∈ R , which is well defined even for q ≤ ψ λ ( ) by Lemmas 8.3 and 8.5 of [28]. In particular, we define
$$W _ { \Phi ( q ) } ( x ) \coloneqq W ^ { ( 0 ) } _ { \Phi ( q ) } ( x ) = e ^ { - \Phi ( q ) x } W ^ { ( q ) } ( x ), \quad x \in \mathbb { R },$$
which is known to be an increasing function and, as in Lemma 3.3 of [27],
$$W _ { \Phi ( q ) } ( x ) \nearrow \psi ^ { \prime } ( \Phi ( q ) ) ^ { - 1 } \quad \text{as} \ x \to \infty.$$
Remark 2.1. (1) If X is of unbounded variation or the L´ evy measure is atomless, it is known that W ( q ) is C 1 ( R \{ } 0 ) ; see, e.g., [12] . Hence,
- (a) Z ( q ) is C 1 ( R \{ } 0 ) and C 0 ( R ) for the bounded variation case, while it is C 2 ( R \{ } 0 ) and C 1 ( R ) for the unbounded variation case, and
- (b) Z ( q ) is C 2 ( R \{ } 0 ) and C 1 ( R ) for the bounded variation case, while it is C 3 ( R \{ } 0 ) and C 2 ( R ) for the unbounded variation case.
- (2) Regarding the asymptotic behavior near zero, as in Lemmas 4.3 and 4.4 of [29] ,
$$W ^ { ( q ) } ( 0 ) = \begin{cases} \begin{array} { c c } 0, & \text{if $X$ is of unbounded variation}, \\ \frac { 1 } { \delta }, & \text{if $X$ is of bounded variation}, \end{array}. \end{cases}$$
(2.9)
$$W ^ { ( q ) ^ { \prime } } ( 0 + ) \coloneqq \lim _ { x \downarrow 0 } W ^ { ( q ) ^ { \prime } } ( x ) = \begin{cases} \begin{array} { c c c } \frac { 2 } { \sigma ^ { 2 } }, & \text{if} \, \sigma > 0, \\ \infty, & \text{if} \, \sigma = 0 \text{ and} \, \nu ( - \infty, 0 ) = \infty, \\ \frac { q + \nu ( - \infty, 0 ) } { \delta ^ { 2 } }, & \text{if} \, \sigma = 0 \text{ and} \, \nu ( - \infty, 0 ) < \infty. \end{array} \end{cases}. \quad. \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon$$
(3) As in (8.18) and Lemma 8.2 of [28] ,
$$\frac { W ^ { ( q ) ^ { \prime } } ( y + ) } { W ^ { ( q ) } ( y ) } \leq \frac { W ^ { ( q ) ^ { \prime } } ( x + ) } { W ^ { ( q ) } ( x ) }, \ \ y > x > 0.$$
In all cases, W ( q ) ′ ( x - ≥ ) W ( q ) ′ ( x +) for all x > 0 .
The problem in this paper is a generalization of Section 6 of [38], where D π t is restricted to be zero. Its optimal solution is a (single) barrier strategy, which is described immediately below. Define, for any measurable function h and s ∈ R ,
$$\infty \text{ union and so} \, \cos \, \lambda, \\ \Psi ( s ; h ) & \coloneqq \int _ { 0 } ^ { \infty } e ^ { - \Phi ( q ) y } h ( y + s ) \text{d} y = \int _ { s } ^ { \infty } e ^ { - \Phi ( q ) ( y - s ) } h ( y ) \text{d} y, \\ \varphi _ { s } ( x ; h ) & \coloneqq \int _ { s } ^ { x } W ^ { ( q ) } ( x - y ) h ( y ) \text{d} y, \quad x \in \mathbb { R }. \\ \cdot \quad \sim \, \not \, \not \, \, \, \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots \, \quad \dots$$
Here ϕ s ( x h ; ) = 0 for any x ≤ s because W ( q ) is uniformly zero on ( -∞ , 0) .
The following, which holds directly from Assumption 2.1, is due to [9]. Here note that Ψ( ; · ˜ ) f ′ is equivalent to (4.23) of [9] (times a positive constant). While in [9], they focus on a special class of spectrally negative L´ evy processes, the results still hold for a general spectrally negative L´vy process. e
Lemma 2.1 (Proposition 5.1 of [9]) . (1) There exists a unique number a < a such that Ψ( ; ˜ ) = 0 a f ′ ,
$$\Psi ( x ; \tilde { f } ^ { \prime } ) < 0 \, i f \, x < \underline { a } \, a n d \, \Psi ( x ; \tilde { f } ^ { \prime } ) > 0 \, i f \, x > \underline { a }.$$
- (2) Ψ( ; ˜ ) ′ x f ′ > 0 for x ≤ a .
- (3) Ψ( ; ˜ ) x f ′ ≥ c / 0 Φ( ) q for x ≥ x 0 .
Namely, while a is the unique zero of ˜ f ′ , a is the unique zero of Ψ( ; · ˜ ) f ′ . We are now ready to state the results of the auxiliary problem.
## Theorem 2.1 (Theorem 6.1 of [38]) . Consider a version of the problem that minimizes
$$\tilde { v } ^ { \pi } ( x ) \coloneqq \mathbb { E } _ { x } \left [ \int _ { 0 } ^ { \infty } e ^ { - q t } f ( \widetilde { Y } _ { t } ^ { \pi } ) \mathrm d t + \int _ { [ 0, \infty ) } e ^ { - q t } C _ { U } \mathrm d U _ { t } ^ { \pi } \right ], \ \ x \in \mathbb { R },$$
with the controlled process ˜ Y π t := X t + U π t , t ≥ 0 for some fixed constant C U ∈ R , that satisfies Assumption 2.1.
Then the barrier strategy U a, ∞ defined by
$$U _ { t } ^ { \underline { \alpha }, \infty } \coloneqq \sup _ { 0 \leq t ^ { \prime } \leq t } \left ( \underline { a } - X _ { t ^ { \prime } } \right ) \vee 0, \quad t \geq 0,$$
is optimal and the value function is
$$( 2. 1 1 ) \quad \inf _ { \pi } \tilde { v } ^ { \pi } ( x ) = \tilde { v } _ { \underline { a } } ( x ) \coloneqq - C _ { U } \left ( \overline { Z } ^ { ( q ) } ( x - \underline { a } ) + \frac { \psi ^ { \prime } ( 0 + ) } { q } \right ) + \frac { f ( \underline { a } ) } { q } Z ^ { ( q ) } ( x - \underline { a } ) - \varphi _ { \underline { a } } ( x ; f ).$$
## 3. THE DOUBLE BARRIER STRATEGIES
Following Avram et al. [1] and Pistorius [33], we define a doubly reflected L´ evy process given by
$$Y _ { t } ^ { a, b } \coloneqq X _ { t } + U _ { t } ^ { a, b } - D _ { t } ^ { a, b }, \ \ t \geq 0, \ a < b,$$
which is reflected at two barriers a and b so as to stay on the interval [ a, b ] ; see page 165 of [1] for the construction of this process. We let π a,b be the corresponding strategy and v a,b the corresponding expected total cost. Our aim is to show that by choosing the values of ( a, b ) appropriately, the minimization is attained by the strategy π a,b .
For b ≥ a , let
$$( 3. 1 ) & \quad \Gamma ( a, b ) \coloneqq C _ { D } + C _ { U } Z ^ { ( q ) } ( b - a ) + f ( b ) W ^ { ( q ) } ( 0 ) + \int _ { a } ^ { b } f ( y ) W ^ { ( q ) ^ { \prime } } ( b - y ) \mathrm d y - W ^ { ( q ) } ( b - a ) f ( a ) \\ & \quad = C _ { D } + C _ { U } + \varphi _ { a } ( b ; \tilde { f } ^ { \prime } ).$$
Also let
$$R ^ { ( q ) } ( y ) \coloneqq \overline { Z } ^ { ( q ) } ( y ) + \frac { \psi ^ { \prime } ( 0 + ) } { q }, \ \ y \in \mathbb { R }.$$
Lemma 3.1. Fix any a < b . We have π a,b ∈ Π . Moreover, for x ≤ b ,
$$v _ { a, b } ( x ) & = \frac { \Gamma ( a, b ) } { q W ^ { ( q ) } ( b - a ) } Z ^ { ( q ) } ( x - a ) - C _ { U } R ^ { ( q ) } ( x - a ) + \frac { f ( a ) } { q } Z ^ { ( q ) } ( x - a ) - \varphi _ { a } ( x ; f ) \\ & = \frac { \Gamma ( a, b ) } { q W ^ { ( q ) } ( b - a ) } Z ^ { ( q ) } ( x - a ) - C _ { U } R ^ { ( q ) } ( x - a ) + \frac { f ( a ) } { q } - \int _ { a } ^ { x } \overline { W } ^ { ( q ) } ( x - y ) f ^ { \prime } ( y ) d y. \\ - \quad \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \end{array}$$
For x ≥ b , we have v a,b ( x ) = v a,b ( b ) + C D ( x -b ) .
Proof. As in Theorem 1 of [1], for all a ≤ x ≤ b ,
$$\hat { \Phi } _ { x } \left [ \int _ { ( 0, \infty ) } e ^ { - q t } \text{d} D _ { t } ^ { a, b } \right ] & = \frac { Z ^ { ( q ) } ( x - a ) } { q W ^ { ( q ) } ( b - a ) }, \\ \mathbb { E } _ { x } \left [ \int _ { [ 0, \infty ) } e ^ { - q t } \text{d} U _ { t } ^ { a, b } \right ] & = - R ^ { ( q ) } ( x - a ) + \frac { Z ^ { ( q ) } ( b - a ) } { q W ^ { ( q ) } ( b - a ) } Z ^ { ( q ) } ( x - a ),$$
which are finite under Assumption 2.2 and hence π a,b ∈ Π . The q -resolvent density of Y a,b is, by Theorem 1 of [33], for y ∈ [ a, b ] ,
$$\mathbb { E } _ { x } \left [ \int _ { [ 0, \infty ) } e ^ { - q t } 1 _ { \{ Y _ { t } ^ { a, b } \in \mathrm d y \} } \mathrm d t \right ] & = \left [ \frac { Z ^ { ( q ) } ( x - a ) W ^ { ( q ) ^ { \prime } } ( b - y ) } { q W ^ { ( q ) } ( b - a ) } - W ^ { ( q ) } ( x - y ) \right ] \mathrm d y \\ & + \left [ Z ^ { ( q ) } ( x - a ) \frac { W ^ { ( q ) } ( 0 ) } { q W ^ { ( q ) } ( b - a ) } \right ] \delta _ { b } ( \mathrm d y ),$$
where δ b is the Dirac measure at b . Summing up these,
$$\text{where} \, \partial _ { b } \, \text{is the distance measure at} \, b. \, \text{Summing up these}, \\ v _ { a, b } ( x ) & = C _ { D } \frac { Z ^ { ( q ) } ( x - a ) } { q W ^ { ( q ) } ( b - a ) } + C _ { U } \left [ - R ^ { ( q ) } ( x - a ) + \frac { Z ^ { ( q ) } ( b - a ) } { q W ^ { ( q ) } ( b - a ) } Z ^ { ( q ) } ( x - a ) \right ] \\ & + \int _ { a } ^ { b } f ( y ) \left [ \frac { Z ^ { ( q ) } ( x - a ) W ^ { ( q ) ^ { \prime } } ( b - y ) } { q W ^ { ( q ) } ( b - a ) } - W ^ { ( q ) } ( x - y ) \right ] d y + f ( b ) Z ^ { ( q ) } ( x - a ) \frac { W ^ { ( q ) } ( 0 ) } { q W ^ { ( q ) } ( b - a ) } \\ & = \frac { Z ^ { ( q ) } ( x - a ) } { q W ^ { ( q ) } ( b - a ) } \left [ C _ { D } + C _ { U } Z ^ { ( q ) } ( b - a ) + f ( b ) W ^ { ( q ) } ( 0 ) + \int _ { a } ^ { b } f ( y ) W ^ { ( q ) ^ { \prime } } ( b - y ) d y \right ] \\ & - C _ { U } R ^ { ( q ) } ( x - a ) - \varphi _ { a } ( x ; f ), \\ \text{which example the fact accumidity of $2$} \,. \, \text{$The second accumidity hold because variance information by more otherwise}$$
which equals the first equality of (3.2). The second equality holds because integration by parts gives
$$\varphi _ { a } ( x ; f ) = \overline { W } ^ { ( q ) } ( x - a ) f ( a ) + \int _ { a } ^ { x } \overline { W } ^ { ( q ) } ( x - y ) f ^ { \prime } ( y ) \text{d} y, \ \ x \geq a.$$
The case x < a holds because v a,b ( x ) = v a,b ( a ) -C U ( x -a ) and Z ( q ) ( x -a ) = Z ( q ) (0) and R ( q ) ( x -a ) = ( x -a ) + R ( q ) (0) . The case x > b similarly holds. glyph[square]
## 3.1. Smoothness conditions. Taking a derivative in (3.2),
$$v _ { a, b } ^ { \prime } ( x ) = \frac { \Gamma ( a, b ) } { W ^ { ( q ) } ( b - a ) } W ^ { ( q ) } ( x - a ) - C _ { U } - \varphi _ { a } ( x ; \tilde { f } ^ { \prime } ), \quad a < x < b,$$
and hence by (3.1)
$$v ^ { \prime } _ { a, b } ( b - ) = C _ { D } \quad \text{and} \quad v ^ { \prime } _ { a, b } ( a + ) = \frac { \Gamma ( a, b ) } { W ^ { ( q ) } ( b - a ) } W ^ { ( q ) } ( 0 ) - C _ { U }.$$
This implies, in view of Remark 2.1(2), that the differentiability of v a,b at b holds for all cases while it holds at a when Γ( a, b ) /W ( q ) ( b -a ) = 0 for the case of bounded variation and it holds automatically for the case of unbounded variation.
Taking another derivative, we have, for a.e. x ∈ ( a, b ) ,
$$v _ { a, b } ^ { \prime \prime } ( x ) = \frac { W ^ { ( q ) ^ { \prime } } ( x - a ) } { W ^ { ( q ) } ( b - a ) } \Gamma ( a, b ) - \int _ { a } ^ { x } W ^ { ( q ) ^ { \prime } } ( x - y ) \tilde { f } ^ { \prime } ( y ) \text{d} y - \tilde { f } ^ { \prime } ( x ) W ^ { ( q ) } ( 0 ),$$
and hence
$$v _ { a, b } ^ { \prime \prime } ( b - ) = \frac { \Gamma ( a, b ) } { W ^ { ( q ) } ( b - a ) } W ^ { ( q ) ^ { \prime } } ( ( b - a ) - ) - \gamma ( a, b ),$$
$$v _ { a, b } ^ { \prime \prime } ( a + ) = \frac { \Gamma ( a, b ) } { W ^ { ( q ) } ( b - a ) } W ^ { ( q ) ^ { \prime } } ( 0 + ) - \tilde { f } ^ { \prime } ( a + ) W ^ { ( q ) } ( 0 ),$$
where
$$\gamma ( a, b ) \coloneqq \int _ { a } ^ { b } W ^ { ( q ) ^ { \prime } } ( b - y ) \tilde { f } ^ { \prime } ( y ) \text{d} y + \tilde { f } ^ { \prime } ( b - ) W ^ { ( a ) } ( 0 ) = \frac { \partial } { \partial b } \Gamma ( a, b - ), \quad b > a.$$
Hence, our candidate levels ( a, b ) are such that
$$\frac { \Gamma ( a, b ) } { W ^ { ( q ) } ( b - a ) } = 0,$$
$$\frac { 1 ( a, b ) } { W ^ { ( q ) } ( b - a ) } & = 0, \\ \gamma ( a, b ) & = 0.$$
Here we understand for the case b = ∞ that lim b →∞ Γ( a, b ) /W ( q ) ( a, b ) = 0 . In such case, in view of (3.2), lim b →∞ v a,b ( x ) = v a, ∞ ( x ) := -C R U ( q ) ( x -a ) + f ( a Z ) ( q ) ( x -a /q ) -ϕ a ( x f ; ) and hence lim b →∞ v ′ a,b ( x ) = v ′ a, ∞ ( x ) and lim b →∞ v ′′ a,b ( x ) = v ′′ a, ∞ ( x ) also hold.
We summarize the results as follows.
Lemma 3.2. (1) If (3.6) holds for some a < b ≤ ∞ , then v a,b is differentiable (resp. twice-differentiable) at a when X is of bounded (resp. unbounded) variation.
- (2) If in addition b < ∞ , it is continuously differentiable at b . In particular, when (3.7) further holds, then it is twice-differentiable at b .
↦
- 3.2. Existence of ( a , b ∗ ∗ ) . Here we show the existence of a pair ( a , b ∗ ∗ ) where (3.6) and (3.7) hold simultaneously. Equivalently, we pursue ( a , b ∗ ∗ ) such that the function b → Γ( a , b ∗ ) attains a minimum 0 at b ∗ (if b ∗ < ∞ ).
First, by (2.3), (3.1) and (3.5),
$$\Gamma ( a, a ) = C _ { D } + C _ { U } > 0 \quad \text{and} \quad \gamma ( a, a + ) = \tilde { f } ^ { \prime } ( a + ) W ^ { ( q ) } ( 0 ), \quad a \in \mathbb { R }.$$
↦
Recall the definition of the level a as in Assumption 2.1. Fix any a ≥ a . Because γ a, b ( ) ≥ 0 for b > a in view of (3.5), the function b → Γ( a, b ) starts at a positive value Γ( a, a ) and increases in b . Therefore, it never crosses nor touches the x-axis.
↦
We now start at a and decrease its value until we arrive at the desired pair ( a , b ∗ ∗ ) such that b → Γ( a , b ∗ ) first touches the x-axis at b ∗ .
We shall first show that a as in Lemma 2.1(1) becomes a lower bound of such a ∗ . For any fixed a ∈ R ,
$$\frac { \varphi _ { a } ( b ; \tilde { f } ^ { \prime } ) } { W ^ { ( q ) } ( b - a ) } = \int _ { a } ^ { b } \tilde { f } ^ { \prime } ( y ) \frac { W ^ { ( q ) } ( b - y ) } { W ^ { ( q ) } ( b - a ) } \mathrm d y = e ^ { \Phi ( q ) a } \int _ { a } ^ { b } \tilde { f } ^ { \prime } ( y ) e ^ { - \Phi ( q ) y } \frac { W _ { \Phi ( q ) } ( b - y ) } { W _ { \Phi ( q ) } ( b - a ) } \mathrm d y.$$
Here, by (2.8), 1 { y ≤ } b | ˜ ( f ′ y ) | e -Φ( ) q y W Φ( ) q ( b -y /W ) Φ( ) q ( b -a ) ≤ | ˜ ( f ′ y ) | e -Φ( ) q y for a.e. y ≥ a , which is integrable over ( a, ∞ ) by Assumption 2.1(1). Hence, by (3.1), dominated convergence gives
$$\lim _ { b \to \infty } \frac { \Gamma ( a, b ) } { W ^ { ( q ) } ( b - a ) } = \Psi ( a ; \tilde { f } ^ { \prime } ).$$
By Lemma 2.1(1), this also implies that lim b →∞ Γ( a, b ) = ∞ if a > a and lim b →∞ Γ( a, b ) = -∞ if a < a . Therefore, for fixed a ∈ ( a, a ) , the infimum Γ( a ) := inf b ≥ a Γ( a, b ) exists and is increasing in a because the (right-)derivative with respect to a becomes
$$\frac { \partial } { \partial a } \Gamma ( a +, b ) = - \tilde { f } ^ { \prime } ( a + ) W ^ { ( a ) } ( b - a ), \quad a < b,$$
which is positive for a < a , and for any a < a < a ′ (such that ˜ f ′ < 0 on ( a , a ′ ) ),
$$\text{positive for $a<\overline{a}$, and for any $a^{\prime}<a<\overline{a}$ (such that $\tilde{f}^{\prime}<0$ on $(a^{\prime},a))$}, \\ \Gamma ( a ^ { \prime } ) \leq \inf _ { b \geq a } \Gamma ( a ^ { \prime }, b ) = \inf _ { b \geq a } \left ( \Gamma ( a, b ) + \int _ { a ^ { \prime } } ^ { a } \tilde { f } ^ { \prime } ( y ) W ^ { ( q ) } ( b - y ) d y \right ) \\ \leq \inf _ { b \geq a } \left ( \Gamma ( a, b ) + \int _ { a ^ { \prime } } ^ { ( a + a ^ { \prime } ) / 2 } \tilde { f } ^ { \prime } ( y ) W ^ { ( q ) } ( b - y ) d y \right ) \\ \leq \inf _ { b \geq a } \left ( \Gamma ( a, b ) + W ^ { ( q ) } \left ( b - \frac { a + a ^ { \prime } } { 2 } \right ) \int _ { a ^ { \prime } } ^ { ( a + a ^ { \prime } ) / 2 } \tilde { f } ^ { \prime } ( y ) d y \right ) \\ \leq \frac { \Gamma ( a ) + W ^ { ( q ) } \left ( \frac { a - a ^ { \prime } } { 2 } \right ) \int _ { a ^ { \prime } } ^ { ( a + a ^ { \prime } ) / 2 } \tilde { f } ^ { \prime } ( y ) d y < \frac { \Gamma ( a ) } { }. \\ \text{easy to see that the function $\Gamma ( a )$ is continuous on $(\underline{\omega},\overline{a}$)}.$$
It is also easy to see that the function Γ( a ) is continuous on ( a, a ) .
In view of these arguments, as we decrease the value of a from a to a , there are two scenarios:
↦
- (1) The curve b → Γ( a, b ) downcrosses the x-axis for a finite b for some a ∈ ( a, a ) ; i.e., there exists a ′ ∈ ( a, a ) such that Γ( a ′ ) < 0 .
↦
- (2) The curve b → Γ( a, b ) is uniformly positive for any choice of a ∈ ( a, a ) ; i.e., Γ( a ) ≥ 0 for all a ∈ ( a, a ) .
For the first scenario, due to the continuity and increasingness of Γ on ( a, a ) and because Γ( a ) = C D + C U > 0 , there must exist a unique a ∗ ∈ ( a, a ) such that Γ( a ∗ ) = 0 . By calling b ∗ the largest value of the minimizers of Γ( a , ∗ · ) , we must have that Γ( a , b ∗ ∗ ) = 0 . In addition, if the function γ a , ( ∗ · ) is continuous at b ∗ then γ a , b ( ∗ ∗ ) = 0 due to the property of the local minimum. Notice, in view of the definition of γ ( · , · ) as in (3.5), that γ a, b ( ) < 0 for any a < b ≤ a , and hence such b ∗ > a .
For the second scenario, we have Γ( a, b ) ≥ 0 for any a ∈ ( a, a ) and b ≥ a . Taking a ↓ a , we have Γ( a, b ) ≥ 0 for any b ≥ a . By (3.9), we see that v a, ∞ ( x ) := lim b →∞ v a,b ( x ) equals ˜ ( v a x ) as in (2.11), that is attained by the strategy π a, ∞ comprising of the single barrier strategy U a, ∞ as in (2.10) and D π = D a, ∞ ≡ 0 .
We summarize the results in the lemma below.
Lemma 3.3. There exist a unique a ∗ such that Γ( a , x ∗ ) ≥ 0 for all x ∈ [ a , ∗ ∞ ) , and b ∗ (defined as the largest minimizer of Γ( a , ∗ · ) ) such that either Case 1 or Case 2 defined below holds.
$$\text{Case } 1 \colon \underline { a } < a ^ { * } < \overline { a } < b ^ { * } \, a n d$$
$$\Gamma ( a ^ { * }, b ^ { * } ) = 0.$$
Moreover, if γ a , b ( ∗ ∗ ) is continuous at b ∗ then we also have that γ a , b ( ∗ ∗ ) = 0 . Case 2: a ∗ = a and b ∗ = ∞ and
$$\lim _ { b \to \infty } \frac { \Gamma ( \underline { a }, b ) } { W ^ { ( q ) } ( b - \underline { a } ) } = 0.$$
↦
Remark 3.1. In view of Case 1 of Lemma 3.3, the function b → γ a , b ( ∗ ) is continuous at b ∗ if f is differentiable at b ∗ or X is of unbounded variation (i.e. W ( q ) (0) = 0 as in (2.9) ).
Remark 3.2. (1) Because a ∗ < a and a →-∞ as C U →∞ , we must have a ∗ →-∞ as C U →∞ . On the other hand, a ∗ must be finite when C U is finite because a ∗ ≥ a .
- (2) By (3.9) and Lemma 2.1(1), for any a > a , Γ( a ) -C D -C U = inf b ≥ a ϕ a ( b ; ˜ ) f ′ > -∞ . Because this does not depend on the value of C D , we have Γ( a ) →∞ as C D →∞ . Equivalently, for any a > a , we can choose a sufficiently large C D so that Γ( a ) > 0 . Hence, a ∗ → a and b ∗ →∞ as C D →∞ .
## 4. VERIFICATION LEMMA
With ( a , b ∗ ∗ ) whose existence is proved in Lemma 3.3, our candidate value function becomes, by (3.2), for all x ≤ b ∗ ,
$$v _ { a ^ { * }, b ^ { * } } ( x ) & = - C _ { U } R ^ { ( q ) } ( x - a ^ { * } ) + \frac { f ( a ^ { * } ) } { q } Z ^ { ( q ) } ( x - a ^ { * } ) - \varphi _ { a ^ { * } } ( x ; f ) \\ & = - C _ { U } R ^ { ( q ) } ( x - a ^ { * } ) + \frac { f ( a ^ { * } ) } { q } - \int _ { a ^ { * } } ^ { x } \overline { W } ^ { ( q ) } ( x - y ) f ^ { \prime } ( y ) \text{d} y. \\ \intertext { a n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t jest i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a n t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i, a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n n D t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a c t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d e n t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n t d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i N d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i n d a t i [ A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { j } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + A _ { i } ] + a _ {$$
Integration by parts gives (for more details, see Lemma 4.1 of [38])
$$\varphi _ { a ^ { * } } ( x ; f ) = \varphi _ { a ^ { * } } ( x ; \tilde { f } ) - C _ { U } \left [ a ^ { * } Z ^ { ( q ) } ( x - a ^ { * } ) + \overline { Z } ^ { ( q ) } ( x - a ^ { * } ) - x \right ], \ \ x \in \mathbb { R }.$$
Hence we can also write
$$v _ { a ^ { * }, b ^ { * } } ( x ) = - C _ { U } \left ( \frac { \psi ^ { \prime } ( 0 + ) } { q } + x \right ) + \frac { \tilde { f } ( a ^ { * } ) } { q } Z ^ { ( q ) } ( x - a ^ { * } ) - \varphi _ { a ^ { * } } ( x ; \tilde { f } ), \ \ x \leq b ^ { * }.$$
Let L be the infinitesimal generator associated with the process X applied to a sufficiently smooth function h
$$\mathcal { L } h ( x ) \coloneqq c h ^ { \prime } ( x ) + \frac { 1 } { 2 } \sigma ^ { 2 } h ^ { \prime \prime } ( x ) + \int _ { ( - \infty, 0 ) } \left [ h ( x + z ) - h ( x ) - h ^ { \prime } ( x ) z 1 _ { \{ - 1 < z < 0 \} } \right ] \nu ( \mathrm d z ), \quad x \in \mathbb { R }.$$
By Lemma 3.2 and Remarks 2.1(1) and 3.1, the function v a ,b ∗ ∗ is C 1 ( R ) (resp. C 2 ( R ) ) when X is of bounded (resp. unbounded) variation. Moreover, the integral part is well defined and finite by Assumption 2.2 and because v a ,b ∗ ∗ is linear below a ∗ . Hence, L v a ,b ∗ ∗ ( ) · makes sense anywhere on R .
The following theorem addresses some additional conditions that are sufficient for the optimality of v a ,b ∗ ∗ .
Theorem 4.1 (Verification lemma) . Suppose
- (1) -C U ≤ v ′ a ,b ∗ ∗ ( x ) for all x ∈ ( a , b ∗ ∗ ) ,
- (2) ( Lq v ) a ,b ∗ ∗ ( x ) + f ( x ) ≥ 0 for all x > b ∗ .
Then, we have
$$v _ { a ^ { * }, b ^ { * } } ( x ) = \inf _ { \pi \in \Pi } v ^ { \pi } ( x ), \ \ x \in \mathbb { R },$$
and π a ,b ∗ ∗ is the optimal strategy.
We shall later show that the conditions (1) and (2) of the above theorem are satisfied if the function f is convex or more generally Assumption 5.1 below holds.
In order to show Theorem 4.1 above, we first show Lemmas 4.1 and 4.2 below.
$$\text{Lemma} \, 4. 1. \quad ( 1 ) \, \ W e h a v e \, ( \mathcal { L } - q ) v _ { a ^ { * }, b ^ { * } } ( x ) + f ( x ) = 0 \, & \text{for} \, a ^ { * } < x < b ^ { * }. \\ ( 2 ) \, \ W e h a v e \, ( \mathcal { L } - q ) v _ { a ^ { * }, b ^ { * } } ( x ) + f ( x ) \geq 0 \, & \text{for} \, x \leq a ^ { * }.$$
Proof. As in the proof of Theorem 2.1 in [6], ( L q Z ) ( q ) ( y -a ∗ ) = ( L q R ) ( q ) ( y -a ∗ ) = 0 for any a ∗ < y < b ∗ . On the other hand, as in the proof of Lemma 4.5 of [17], ( Lq ϕ ) a ∗ ( x f ; ) = f ( x ) . Hence in view of (4.1), (1) is proved.
For (2), by (4.2), v a ,b ∗ ∗ ( x ) = [ -C ψ U ′ (0+)+ ˜ ( f a ∗ )] /q -C x U , for x < a ∗ , and hence ( Lq v ) a ,b ∗ ∗ ( x )+ f ( x ) = ˜ ( f x ) -˜ ( f a ∗ ) . This is positive by x ≤ a ∗ < a and Assumption 2.1(2), as desired. glyph[square]
By (3.1) and (3.3),
$$v _ { a ^ { * }, b ^ { * } } ^ { \prime } ( x ) & = - \Gamma ( a ^ { * }, x ) + C _ { D }, \quad a ^ { * } \leq x \leq b ^ { * }.$$
Lemma 4.2. For all x ∈ R , we have v ′ a ,b ∗ ∗ ( x ) ≤ C D .
Proof. By Lemma 3.3, we must have Γ( a , x ∗ ) ≥ 0 on [ a , b ∗ ∗ ] and hence in view of (4.3), this inequality holds for x ∈ [ a , b ∗ ∗ ] . For x ∈ ( -∞ , a ∗ ) , we have v ′ a ,b ∗ ∗ ( x ) = -C U , which is smaller than C D by (2.3). Finally, for x ∈ ( b , ∗ ∞ ) , we have v ′ a ,b ∗ ∗ ( x ) = C D . glyph[square]
We are now ready to give a proof for Theorem 4.1.
Proof of Theorem 4.1. As a short-hand notation, let v ≡ v a ,b ∗ ∗ in this proof. By (2.3), Lemma 4.2 and the assumption (1),
$$- C _ { U } \leq v ^ { \prime } ( x ) \leq C _ { D }, \ \ x \in \mathbb { R }.$$
As discussed in the introduction, we can focus on the strategy π ∈ Π such that U π t and D π t are not increased simultaneously. Fix any such admissible strategy π ∈ Π . Thanks to the smoothness of v described above, Itˆ's formula (see, e.g., page 78 of [35]) gives o
$$v ( Y _ { t } ^ { \pi } ) - v ( Y _ { 0 - } ^ { \pi } ) = \int _ { [ 0, t ] } v ^ { \prime } ( Y _ { s - } ^ { \pi } ) d Y _ { s } ^ { \pi } + \frac { \sigma ^ { 2 } } { 2 } \int _ { 0 } ^ { t } v ^ { \prime \prime } ( Y _ { s } ^ { \pi } ) d s + \sum _ { 0 \leq s \leq t } [ v ( Y _ { s } ^ { \pi } ) - v ( Y _ { s - } ^ { \pi } ) - v ^ { \prime } ( Y _ { s - } ^ { \pi } ) \Delta Y _ { s } ^ { \pi } ].$$
Define the difference of the control processes ξ π t := U π t -D , t π t ≥ 0 . Then Y π t = X t + ξ π t and
$$\Delta _ { i }$$
$$\Delta \xi _ { t } ^ { \pi } = \begin{cases} \Delta U _ { t } ^ { \pi }, & \text{if $\Delta\xi_{t}^{\pi} \geq 0$,} \\ - \Delta D _ { t } ^ { \pi }, & \text{if $\Delta\xi_{t}^{\pi} < 0$,} \end{cases} \quad \text{and} \quad d \xi _ { t } ^ { \pi, c } = \begin{cases} \ d U _ { t } ^ { \pi, c }, & \text{if $\delta\xi_{t}^{\pi, c} \geq 0$,} \\ - \delta D _ { t } ^ { \pi, c }, & \text{if $\delta\xi_{t}^{\pi, c} < 0$,} \end{cases}$$
where we denote ∆ ζ t := ζ t -ζ t -and ζ c as the continuous part of a process ζ . We have
$$\int _ { [ 0, t ] } v ^ { \prime } ( Y _ { s - } ^ { \pi } ) \text{d} Y _ { s } ^ { \pi } = \int _ { [ 0, t ] } v ^ { \prime } ( Y _ { s - } ^ { \pi } ) \text{d} X _ { s } + \int _ { 0 } ^ { t } v ^ { \prime } ( Y _ { s } ^ { \pi } ) \text{d} \xi _ { s } ^ { \pi, c } + \sum _ { 0 \leq s \leq t } v ^ { \prime } ( Y _ { s - } ^ { \pi } ) \Delta \xi _ { s } ^ { \pi }.$$
From the L´ evy-Itˆ decomposition theorem (e.g., Theorem 2.1 of [28]), we know that o
$$X _ { t } = ( \sigma B _ { t } + c t ) + \left ( \int _ { [ 0, t ] } \int _ { ( - \infty, - 1 ] } y N ( \mathrm d s \times \mathrm d y ) \right ) + \left ( \lim _ { \varepsilon \downarrow 0 } \int _ { [ 0, t ] } \int _ { ( - 1, - \varepsilon ) } y ( N ( \mathrm d s \times \mathrm d y ) - \nu ( \mathrm d y ) \mathrm d s ) \right ),$$
where B t is a standard Brownian motion and N is a Poisson random measure in the measurable space ([0 , ∞ × -∞ ) ( , 0) , B [0 , ∞ ×B -∞ ) ( , 0) , d t × ν (d x )) . The last term is a square integrable martingale, to which the limit converges uniformly on any compact [0 , T ] .
Using this decomposition and defining A π t := Y π t -+∆ X t t , ≥ 0 (so that A π t +∆ ξ π t = Y π t ), integration by parts gives (see, e.g., the proof of Theorem 3.1 of [22] for details),
$$e ^ { - q t } v ( Y _ { t } ^ { \pi } ) - v ( Y _ { 0 - } ^ { \pi } ) = \int _ { 0 } ^ { t } e ^ { - q s } ( \mathcal { L } - q ) v ( Y _ { s } ^ { \pi } ) \mathrm d s + J _ { t } + M _ { t },$$
with
$$J _ { t } \coloneqq \int _ { 0 } ^ { t } e ^ { - q s } v ^ { \prime } ( Y _ { s } ^ { \pi } ) d U _ { s } ^ { \pi, c } + \sum _ { 0 \leq s \leq t } e ^ { - q s } [ v ( A _ { s } ^ { \pi } + \Delta U _ { s } ^ { \pi } ) - v ( A _ { s } ^ { \pi } ) ] 1 _ { \{ \Delta U _ { s } ^ { \pi } > 0 \} }$$
$$\}$$
$$- \int _ { 0 } ^ { t } e ^ { - q s } v ^ { \prime } ( Y _ { s } ^ { \pi } ) \mathrm d D _ { s } ^ { \pi, c } + \sum _ { 0 \leq s \leq t } ^ { - - } e ^ { - q s } [ v ( A _ { s } ^ { \pi } - \Delta D _ { s } ^ { \pi } ) - v ( A _ { s } ^ { \pi } ) ] 1 _ { \{ \Delta D _ { s } ^ { \pi } > 0 \} },$$
$$\cdot$$
$$& \stackrel { \kappa \, \vee } { \int } _ { 0 } ^ { t } \sigma e ^ { - q s } v ^ { \prime } ( Y _ { s } ^ { \pi } ) d B _ { s } + \lim _ { \varepsilon \downarrow 0 } \int _ { [ 0, t ] } \int _ { ( - 1, - \varepsilon ) } e ^ { - q s } v ^ { \prime } ( Y _ { s - } ^ { \pi } ) y ( N ( \mathrm d s \times \mathrm d y ) - \nu ( \mathrm d y ) \mathrm d s ) \\ & + \int _ { [ 0, t ] } \int _ { ( - \infty, 0 ) } e ^ { - q s } ( v ( Y _ { s - } ^ { \pi } + y ) - v ( Y _ { s - } ^ { \pi } ) - v ^ { \prime } ( Y _ { s - } ^ { \pi } ) y 1 _ { \{ y \in ( - 1, 0 ) \} } ) ( N ( \mathrm d s \times \mathrm d y ) - \nu ( \mathrm d y ) \mathrm d s ). \\ & \stackrel { \kappa } { \int } _ { \kappa }$$
By (4.4), we have the inequality: J t ≥ -C U ∫ [0 ,t ] e -qs d U π s -C D ∫ [0 ,t ] e -qs d D π s . Moreover, by Lemma 4.1(2) and the assumption (2) of this theorem, ( Lq v x ) ( ) ≥ -f ( x ) for all x ∈ R .
Let τ π n := inf { t ≥ 0 : | Y π t | > n } , n > 0 . Optional sampling gives
$$v ( x ) \leq \mathbb { E } _ { x } \left [ \int _ { 0 } ^ { t \wedge \tau _ { n } ^ { \pi } } e ^ { - q s } f ( Y _ { s } ^ { \pi } ) d s + C _ { U } \int _ { [ 0, t \wedge \tau _ { n } ^ { \pi } ] } e ^ { - q s } d U _ { s } ^ { \pi } + C _ { D } \int _ { [ 0, t \wedge \tau _ { n } ^ { \pi } ] } e ^ { - q s } d D _ { s } ^ { \pi } + e ^ { - q ( t \wedge \tau _ { n } ^ { \pi } ) } v ( Y _ { t \wedge \tau _ { n } ^ { \pi } } ^ { \pi } ) \right ].$$
By (2.4), E x [ ∫ t ∧ τ π n 0 e -qs f ( Y π s )d ] = s E x [ ∫ t ∧ τ π n 0 e -qs ˜ ( f Y π s )d ] s -C U E x [ ∫ t ∧ τ π n 0 qe -qs Y π s d ] s . Because ˜ f admits a global minimum at a by Assumption 2.1(2) (and hence is bounded from below), dominated convergence applied to the negative part of the integrand and monotone convergence for the other part give
$$\lim _ { t, n \uparrow \infty } \mathbb { E } _ { x } \Big { [ } \int _ { 0 } ^ { t \wedge \tau _ { n } ^ { \pi } } e ^ { - q s } \tilde { f } ( Y _ { s } ^ { \pi } ) \mathrm d s \Big { ] } = \mathbb { E } _ { x } \Big { [ } \int _ { 0 } ^ { \infty } e ^ { - q s } \tilde { f } ( Y _ { s } ^ { \pi } ) \mathrm d s \Big { ] }.$$
On the other hand, because ( Y π s ) -≤ -X s + | x | + D π s (where we define X t := inf 0 ≤ ≤ t ′ t X t ′ , t ≥ 0 , as the running infimum process),
$$\int _ { 0 } ^ { t \wedge \tau _ { n } ^ { \pi } } q e ^ { - q s } ( Y _ { s } ^ { \pi } ) ^ { - } d s & \leq - \int _ { 0 } ^ { t \wedge \tau _ { n } ^ { \pi } } q e ^ { - q s } ( \underline { X } _ { s } - | x | ) d s + \int _ { 0 } ^ { t \wedge \tau _ { n } ^ { \pi } } q e ^ { - q s } D _ { s } ^ { \pi } d s \\ & \leq - \int _ { 0 } ^ { \infty } q e ^ { - q s } \underline { X } _ { s } d s + | x | + \int _ { 0 } ^ { \infty } q e ^ { - q s } D _ { s } ^ { \pi } d s.$$
Notice that
$$- \mathbb { E } \Big [ \int _ { 0 } ^ { \infty } q e ^ { - q s } \underline { X } _ { s } \mathrm d s \Big ] = \Phi ( q ) ^ { - 1 } - \frac { \psi ^ { \prime } ( 0 + ) } { q }$$
by the duality and the Wiener-Hopf factorization (see, e.g., the proof of Lemma 4.4 of [22]), which is finite by Assumption 2.2.
Integration by parts gives E x [ ∫ [0 ,t ] e -qs d D π s ] = E x [ e -qt D π t ] + E x [ ∫ t 0 qe -qs D π s d ] s ≥ E x [ ∫ t 0 qe -qs D π s d ] s . Hence, by (2.2),
$$\infty > \mathbb { E } _ { x } \left [ \int _ { [ 0, \infty ) } e ^ { - q s } \mathrm d D _ { s } ^ { \pi } \right ] \geq \mathbb { E } _ { x } \left [ \int _ { 0 } ^ { \infty } q e ^ { - q s } D _ { s } ^ { \pi } \mathrm d s \right ].$$
This together with (4.5) and (4.6) gives lim t,n ↑∞ E x [ ∫ t ∧ τ π n 0 qe -qs ( Y π s ) -d s ] = E x [ ∫ ∞ 0 qe -qs ( Y π s ) -d s ] < ∞ . Similar arguments show that lim t,n ↑∞ E x [ ∫ t ∧ τ π n 0 qe -qs ( Y π s ) + d s ] = E x [ ∫ ∞ 0 qe -qs ( Y π s ) + d s ] < ∞ .
By these and the monotonicity of D π t and U π t in t , monotone convergence gives a bound:
$$\begin{array} { c c c c c c } \ell & \ell & \ell & \ell & \ell & \leq \\ & v ( x ) = v ( U _ { 0 - } ^ { \pi } ) & \leq \mathbb { E } _ { x } \left [ \int _ { 0 } ^ { \infty } e ^ { - q s } f ( Y _ { s } ^ { \pi } ) d s + C _ { U } \int _ { [ 0, \infty ) } e ^ { - q s } d U _ { s } ^ { \pi } + C _ { D } \int _ { [ 0, \infty ) } e ^ { - q s } d D _ { s } ^ { \pi } \right ] \\ & + \lim \sup _ { t, n \to \infty } \mathbb { E } _ { x } \left [ e ^ { - q ( t \wedge \tau _ { n } ^ { \pi } ) } v ^ { + } ( Y _ { t \wedge \tau _ { n } ^ { \pi } } ^ { \pi } ) \right ]. \end{array}$$
It remains to show that the last term of the right hand side vanishes. Indeed, we have X t -D π t ≤ Y π t ∧ τ π n ≤ X t + U π t , t ≥ 0 (where we define X t := sup 0 ≤ ≤ t ′ t X t ′ , t ≥ 0 , as the running supremum process). In view of this and (4.4), it is sufficient to show that E x [ e -qt ( X t + U π t )] and E x [ e -qt ( -X t + D π t )] vanish in the limit.
↦
First, as in Lemma 3.3 and Remark 3.2 of [22], E x [ e -qt X t ] and E x [ e -qt X t ] vanish in the limit as t →∞ . On the other hand, by (4.7) and the monotonicity of t → D π t ,
$$0 = \lim _ { t \to \infty } \mathbb { E } _ { x } \left [ \int _ { t } ^ { \infty } q e ^ { - q s } D _ { s } ^ { \pi } d s \right ] \geq \lim _ { t \to \infty } \sup \mathbb { E } _ { x } \left [ D _ { t } ^ { \pi } \int _ { t } ^ { \infty } q e ^ { - q s } d s \right ] = \lim _ { t \to \infty } \sup \mathbb { E } _ { x } \left [ e ^ { - q t } D _ { t } ^ { \pi } \right ] \geq 0.$$
Similarly, lim t →∞ E x [ e -qt U π t ] = 0 also holds.
These together with (4.8) show v x ( ) ≤ v π ( x ) for all π ∈ Π . We also have v x ( ) ≥ inf π ∈ Π v π ( x ) because v is attained by an admissible strategy π a ,b ∗ ∗ ∈ Π . This completes the proof. glyph[square]
Showing the conditions (1) and (2) of Theorem 4.1 above is the most challenging task of this problem. However, Case 2 (i.e. a ∗ = a and b ∗ = ∞ ) can be handled easily; we defer the discussion on Case 1 to the next section.
Theorem 4.2. In Case 2 , we have v a, ∞ ( x ) = inf π ∈ Π v π ( x ) , x ∈ R , and π a, ∞ is the optimal strategy.
Proof. In view of the conditions for Theorem 4.1, we only need to show the condition (1). By (4.3), v ′ a, ∞ ( x ) = -Γ( a, x ) + C D = -C U -ϕ a ( x f ; ˜ ) ′ . The proof is complete because ϕ a ( x f ; ˜ ) ′ is nonpositive by the proof of Proposition 7.4 of [38]. glyph[square]
## 5. SUFFICIENT CONDITION FOR OPTIMALITY FOR CASE 1
We shall now investigate a sufficient optimality condition for Case 1 so that the assumptions in Theorem 4.1 are satisfied. Throughout this section, we assume Case 1 and the following.
Assumption 5.1. We assume that, for every a < a , there exists ˜ ( b a ) ∈ ( a, ∞ ] such that
$$\gamma ( a, b ) \leq ( > ) 0 \Longleftrightarrow b < ( > ) b ( a ), \ \ b \geq a.$$
↦
Equivalently, this assumption says that the function b → Γ( a, b ) is first nonincreasing and then increasing (or nonincreasing monotonically), given a < a ; note from (3.5) that γ a, b ( ) < 0 for a < b < a and hence the function must first decrease.
As an important condition where Assumption 5.1 holds, we show the following. It is noted that majority of related control problems assume the convexity of f ; see, e.g., [14, 15, 21].
## Theorem 5.1. If f is convex, then Assumption 5.1 holds.
Proof. Fix a < a . Integration by parts applied to (3.5) gives, for all b > a ,
(5.1)
$$\ell _ { \lambda } ^ { \quad \cdot } \ell _ { \lambda } ^ { \quad \cdot } W ^ { ( q ) } ( b - y ) \tilde { f } ^ { \prime \prime } ( y ) \mathrm d y + \tilde { f } ^ { \prime } ( a + ) W ^ { ( q ) } ( b - a ) + \sum _ { a < y < b } W ^ { ( q ) } ( b - y ) [ \tilde { f } ^ { \prime } ( y + ) - \tilde { f } ^ { \prime } ( y - ) ],$$
where ˜ ( f ′′ y ) exists a.e. on ( a, b ) by the convexity of f . First, because ˜ ( f ′ x +) -˜ ( f ′ x -) ≥ 0 for any x ∈ R by the convexity of f , γ a, b ( +) -γ a, b ( - ≥ ) 0 .
Second, dividing both sides of (5.1) by W ( q ) ( b -a ) and taking a derivative with respect to b , we have for a.e. b > a ,
$$\Re \cdot \upsilon \succ \alpha, \\ \frac { \partial } { \partial b } \frac { \gamma ( a, b ) } { W ^ { ( q ) } ( b - a ) } & = \int _ { a } ^ { b } \frac { \partial } { \partial b } \frac { W ^ { ( q ) } ( b - y ) } { W ^ { ( q ) } ( b - a ) } \tilde { f } ^ { \prime \prime } ( y ) \text{d} y + \frac { W ^ { ( q ) } ( 0 ) } { W ^ { ( q ) } ( b - a ) } \tilde { f } ^ { \prime \prime } ( b - ) \\ & + \sum _ { a < y < b } \frac { \partial } { \partial b } \frac { W ^ { ( q ) } ( b - y ) } { W ^ { ( q ) } ( b - a ) } [ \tilde { f } ^ { \prime } ( y + ) - \tilde { f } ^ { \prime } ( y - ) ]. \\ \hat { \ } \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \,.$$
Here, for any a < y < b , the (right) derivative of the fraction W ( q ) ( b -y /W ) ( q ) ( b -a ) equals
$$\frac { \partial } { \partial b } \frac { W ^ { ( q ) } ( ( b - y ) + ) } { W ^ { ( q ) } ( ( b - a ) + ) } = \frac { W ^ { ( q ) } ( b - y ) } { W ^ { ( q ) } ( b - a ) } \left [ \frac { W ^ { ( q ) ^ { \prime } } ( ( b - y ) + ) } { W ^ { ( q ) } ( b - y ) } - \frac { W ^ { ( q ) ^ { \prime } } ( ( b - a ) + ) } { W ^ { ( q ) } ( b - a ) } \right ],$$
↦
which is positive by Remark 2.1(3). This together with the convexity of ˜ f shows that b → γ a, b /W ( ) ( q ) ( b -a ) is nondecreasing on ( a, ∞ ) . By (3.8) and Assumption 2.1(2), we have γ a, a ( +) = ˜ ( f ′ a +) W ( q ) (0) ≤ 0 . This means, by the positivity of W ( q ) ( b -a ) , that γ a, ( · ) is first negative and then positive (or uniformly negative). This completes the proof. glyph[square]
Lemma 5.1. Under Assumption 5.1, the function v a ,b ∗ ∗ is convex on R .
↦
Proof. Because a ∗ < a , Assumption 5.1 guarantees that ˜ ( b a ∗ ) = b ∗ . Because b → Γ( a , b ∗ ) is nonincreasing on ( a , b ∗ ∗ ) , v a ,b ∗ ∗ is convex on ( a , b ∗ ∗ ) in view of (4.3). The convexity can be extended to R by the differentiability at a ∗ and b ∗ of v a ,b ∗ ∗ (if b ∗ < ∞ ) by Lemma 3.2 and the linearity on ( -∞ , a ∗ ] and [ b , ∗ ∞ ) . glyph[square]
This lemma directly implies the following.
Proposition 5.1. Under Assumption 5.1, the condition (1) of Theorem 4.1 holds.
Fix any b ∈ R . Note that Γ( a, b ) > 0 for any a ∈ [ a ∧ b, b ] in view of (3.1) and Assumption 2.1(2). This together with (3.10) and Assumption 2.1(2) (which implies lim a ↓-∞ Γ( a, b ) = -∞ ) shows that there exists a unique a b ( ) ∈ -∞ ( , a ∧ b ) such that
$$\Gamma ( a ( b ), b ) = 0.$$
Lemma 5.2. Suppose Assumption 5.1 holds. (i) If b > b ′ > b ∗ , then a b ( ) < a b ( ′ ) < a ∗ and (ii) if b > b ∗ , γ a b , x ( ( ) - ≥ ) 0 for all x ≥ b .
Proof. We first suppose b > b ∗ and prove that a b ( ) < a ∗ . Assume for contradiction that b > b ∗ and a b ( ) ≥ a ∗ hold simultaneously. By a ∗ ≤ a b ( ) < a and (3.10), we have 0 = Γ( a b , b ( ) ) ≥ Γ( a , b ∗ ) , which is a contradiction because Γ( a , b ∗ ∗ ) = 0 and Γ( a , ∗ · ) is increasing on ( b , b ∗ ) (by Assumption 5.1 and because b ∗ = ˜ ( b a ∗ ) ). Hence whenever b > b ∗ we must have a ∗ > a b ( ) .
This also shows γ a b , b ( ( ) -) ≥ 0 , and hence (ii) by Assumption 5.1. Indeed, if γ a b , b ( ( ) -) < 0 , this means by Assumption 5.1 that γ a b , ( ( ) · ) < 0 on ( b , b ∗ ) and hence 0 = Γ( ( ) a b , b ) < Γ( a b , b ( ) ∗ ) .
However, this contradicts with 0 = Γ( a , b ∗ ∗ ) , which is larger than Γ( a b , b ( ) ∗ ) by a b ( ) < a ∗ < a and (3.10).
Now suppose b > b ′ > b ∗ and assume for contradiction that a b ( ) ≥ a b ( ′ ) to complete the proof for (i). By (3.10) we have 0 = Γ( a b , b ( ) ) ≥ Γ( a b ( ′ ) , b ) , which is a contradiction because Γ( a b ( ′ ) , · ) is increasing on ( b , b ′ ) (due to (ii)) and Γ( a b ( ′ ) , b ′ ) = 0 . glyph[square]
Lemma 5.3. Suppose Assumption 5.1 holds. For any b > b ∗ , we have Γ( a b , y ( ) ) ≤ 0 for b ∗ ≤ y ≤ b .
↦
Proof. By definition, Γ( a b , b ( ) ) = 0 . Because a → Γ( a, b ∗ ) is increasing on ( -∞ , a ) by (3.10) and a b ( ) < a ∗ < a by Lemma 5.2(i), we have Γ( a b , b ( ) ∗ ) < Γ( a , b ∗ ∗ ) = 0 .
↦
Because y → Γ( a b , y ( ) ) is nonincreasing and then increasing on ( b , b ∗ ) (or simply monotone on ( b , b ∗ ) ), we must have that Γ( a b , y ( ) ) ≤ max Γ( ( ) { a b , b ∗ ) , Γ( a b , b ( ) ) } = 0 for b ∗ ≤ y ≤ b . glyph[square]
Using Lemmas 5.2 and 5.3, we show the second condition of Theorem 4.1. Below, we use techniques similar to [22, 30].
Proposition 5.2. Under Assumption 5.1, the condition (2) of Theorem 4.1 holds.
Proof. Fix any x > b ∗ . It is sufficient to prove
$$( { \mathcal { L } } - q ) ( v _ { a ^ { * }, b ^ { * } } - v _ { a ( x ), x } ) ( x - ) \coloneqq \lim _ { y \uparrow x } ( { \mathcal { L } } - q ) ( v _ { a ^ { * }, b ^ { * } } - v _ { a ( x ), x } ) ( y ) \geq 0.$$
Indeed if both (5.3) and ( Lq v ) a ,b ∗ ∗ ( x ) + f ( x ) < 0 hold simultaneously, then
$$0 > ( \mathcal { L } - q ) v _ { a ^ { * }, b ^ { * } } ( x ) + f ( x ) \geq ( \mathcal { L } - q ) v _ { a ( x ), x } ( x - ) + f ( x ),$$
which leads to a contradiction because ( Lq v ) a x ,x ( ) ( y )+ f ( y ) = 0 for a x ( ) < y < x that holds similarly to Lemma 4.1(1). Notice that the function v a x ,x ( ) admits the same form as (4.1) (with a ∗ replaced with a x ( ) ) because Γ( a x , x ( ) ) = 0 .
Notice from (3.4) that both v a ,b ∗ ∗ and v a x ,x ( ) are differentiable on R . Similarly to (4.3),
$$( 5. 4 ) \quad \ v ^ { \prime } _ { a ( x ), x } ( y ) = - \Gamma ( a ( x ), y ) + C _ { D } \quad \text{and} \quad v ^ { \prime \prime } _ { a ( x ), x } ( y ) = - \gamma ( a ( x ), y ), \quad a ( x ) < y < x.$$
The dominated convergence theorem gives
$$( \cdot )$$
$$- \quad \cdot \\ & ( \mathcal { L } - q ) ( v _ { a ^ { * }, b ^ { * } } - v _ { a ( x ), x } ) ( x - ) = c ( v _ { a ^ { * }, b ^ { * } } - v _ { a ( x ), x } ^ { \prime } ) ( x ) + \frac { 1 } { 2 } \sigma ^ { 2 } ( v _ { a ^ { * }, b ^ { * } } ^ { \prime \prime } - v _ { a ( x ), x } ^ { \prime \prime } ) ( x - ) \\ & + \int _ { ( - \infty, 0 ) } \left [ ( v _ { a ^ { * }, b ^ { * } } - v _ { a ( x ), x } ) ( x + z ) - ( v _ { a ^ { * }, b ^ { * } } - v _ { a ( x ), x } ) ( x ) - ( v _ { a ^ { * }, b ^ { * } } ^ { \prime } - v _ { a ( x ), x } ^ { \prime } ) ( x ) z 1 _ { \{ - 1 < z < 0 \} } \right ] \nu ( \mathrm d z ) \\ & - q ( v _ { a ^ { * }, b ^ { * } } - v _ { a ( x ), x } ) ( x ).$$
By the differentiability of v a x ,x ( ) and v ′ a ,b ∗ ∗ ( x ) = C D and v ′′ a ,b ∗ ∗ ( x ) = 0 as x > b ∗ , this is simplified to
$$( 5. 5 ) \ \ ( \mathcal { L } - q ) ( v _ { a ^ { * }, b ^ { * } } - v _ { a ( x ), x } ) ( x - ) = - \frac { 1 } { 2 } \sigma ^ { 2 } v _ { a ( x ), x } ^ { \prime \prime } ( x - )$$
$$+ \int _ { ( - \infty, 0 ) } \left [ ( v _ { a ^ { * }, b ^ { * } } - v _ { a ( x ), x } ) ( x + z ) - ( v _ { a ^ { * }, b ^ { * } } - v _ { a ( x ), x } ) ( x ) \right ] \nu ( \mathrm d z ) - q ( v _ { a ^ { * }, b ^ { * } } - v _ { a ( x ), x } ) ( x ).$$
By taking limits in (5.4) and by Lemma 5.2(ii),
$$v _ { a ( x ), x } ^ { \prime \prime } ( x - ) = - \gamma ( a ( x ), x - ) \leq 0.$$
In order to prove the positivity of the integral part of (5.5), we shall prove that
$$v ^ { \prime } _ { a ( x ), x } ( y ) \geq v ^ { \prime } _ { a ^ { * }, b ^ { * } } ( y ), \ \ y \in ( - \infty, x ).$$
Recall also that a x ( ) < a ∗ by Lemma 5.2(i).
$$( i ) \, \text{For $b^{*} \leq y < x$, by Lemma $5.3$, $v_{a(x),x}^{\prime}(y)=-\Gamma(a(x),y)+C_{D} \geq C_{D}=v_{a^{*},b^{*}(y)$.}$$
$$( \text{ii} ) \, \text{For} \, a ^ { * } \leq y < b ^ { * },$$
$$v ^ { \prime } _ { a ( x ), x } ( y ) = - \Gamma ( a ( x ), y ) + C _ { D } \geq - \Gamma ( a ^ { * }, y ) + C _ { D } = v ^ { \prime } _ { a ^ { * }, b ^ { * } } ( y ).$$
Here the inequality holds because a x ( ) < a ∗ < a and Γ( · , y ) is increasing by (3.10).
(iii) For a x ( ) ≤ y < a ∗ , by Assumption 5.1, (3.8) and (3.10), Γ( a x , y ( ) ) ≤ Γ( a x , a x ( ) ( )) ∨ Γ( a x , a ( ) ∗ ) ≤ Γ( a x , a x ( ) ( )) ∨ Γ( a , a ∗ ∗ ) = C D + C U . Hence
$$v ^ { \prime } _ { a ( x ), x } ( y ) = - \Gamma ( a ( x ), y ) + C _ { D } \geq - C _ { U } = v ^ { \prime } _ { a ^ { * }, b ^ { * } } ( y ).$$
(vi) For y < a x ( ) , we have that v ′ a x ,x ( ) ( y ) = v ′ a ,b ∗ ∗ ( y ) = -C U . Hence (5.6) holds, and consequently the integral of (5.5) is positive.
Finally, we shall show that
$$v _ { a ( x ), x } ( x ) \geq v _ { a ^ { * }, b ^ { * } } ( x ).$$
By (4.2) (which also holds when a ∗ is replaced with a x ( ) ) and a x ( ) < a ∗ ,
$$v _ { a ( x ), x } ( a ( x ) ) & = - C _ { U } \frac { \psi ^ { \prime } ( 0 + ) } { q } + \frac { \tilde { f } ( a ( x ) ) } { q } - C _ { U } a ( x ), \\ v _ { a ^ { * }, b ^ { * } } ( a ( x ) ) & = - C _ { U } \frac { \psi ^ { \prime } ( 0 + ) } { q } + \frac { \tilde { f } ( a ^ { * } ) } { q } - C _ { U } a ( x ).$$
$$v _ { a ^ { * }, b ^ { * } } ( a ( x ) ) = - C _ { U } \frac { \psi ^ { \prime } ( 0 + ) } { q } + \frac { \tilde { f } ( a ^ { * } ) } { q } - C _ { U } a ( x ).$$
Because a x ( ) < a ∗ < a , we have v a x ,x ( ) ( a x ( )) ≥ v a ,b ∗ ∗ ( a x ( )) by Assumption 2.1(2). This together with (5.6) shows (5.7). Putting altogether, (5.3) indeed holds. This completes the proof. glyph[square]
Combining Propositions 5.1 and 5.2, we have the following.
Theorem 5.2. Under Assumption 5.1, we have v a ,b ∗ ∗ ( x ) = inf π ∈ Π v π ( x ) for x ∈ R , and π a ,b ∗ ∗ is the optimal strategy.
## 6. EXAMPLES
Recall from Theorems 5.1 and 5.2 that whenever the running cost function f is convex, the optimality of v a ,b ∗ ∗ holds. In this section we consider the following two special cases and study the criteria for Case 1 and Case 2 as in Lemma 3.3.
## 6.1. Quadratic case. Suppose the running cost function is f ≡ f Q where
$$f _ { Q } ( x ) \coloneqq \alpha ^ { - } x ^ { 2 } 1 _ { \{ x < 0 \} } + \alpha ^ { + } x ^ { 2 } 1 _ { \{ x \geq 0 \} }, \quad x \in \mathbb { R },$$
for some α , α -+ > 0 . The quadratic cost function of this form is used in, e.g., [2].
We then have f ′ Q ( x ) = 2[ α -1 { x< 0 } + α + 1 { x ≥ } 0 ] x and ˜ ( f ′ Q x ) = 2[ α -1 { x< 0 } + α + 1 { x ≥ } 0 ] x + qC U . Hence, Assumption 2.1(2) holds with a = -qC U / (2 α -) if C U ≥ 0 and a = -qC U / (2 α + ) if C U < 0 . Moreover, for a ≤ 0 ,
$$\Psi ( a ; \tilde { f } ^ { \prime } ) = \begin{cases} \ \frac { 2 \alpha ^ { - } a + q C _ { U } } { \Phi ( q ) } + \frac { 2 ( \alpha ^ { + } - \alpha ^ { - } ) e ^ { \Phi ( q ) a } + 2 \alpha ^ { - } } { \Phi ( q ) ^ { 2 } }, & a \leq 0, \\ \ \frac { 2 \alpha ^ { + } a + q C _ { U } } { \Phi ( q ) } + \frac { 2 \alpha ^ { + } } { \Phi ( q ) ^ { 2 } }, & a > 0, \end{cases}$$
and hence a is either a ≤ 0 satisfying
$$2 \alpha ^ { - } \underline { a } + q C _ { U } = - \frac { 2 } { \Phi ( q ) } \left ( ( \alpha ^ { + } - \alpha ^ { - } ) e ^ { \Phi ( q ) \underline { a } } + \alpha ^ { - } \right ),$$
or a > 0 satisfying
$$\cdot \alpha ^ { + }.$$
$$2 \alpha ^ { + } \underline { a } + q C _ { U } = - \frac { 2 } { \Phi ( q ) } \alpha ^ { + }.$$
In addition, direct computation gives for a ≤ 0
$$\Gamma ( a, b ) = C _ { U } + C _ { D } + ( 2 \alpha ^ { - } a + q C _ { U } ) \overline { W } ^ { ( q ) } ( b - a ) + 2 \alpha ^ { - } \int _ { a } ^ { 0 \wedge b } \overline { W } ^ { ( q ) } ( b - y ) d y + 2 \alpha ^ { + } \int _ { 0 \wedge b } ^ { b } \overline { W } ^ { ( q ) } ( b - y ) d y,$$
and for a > 0 ,
$$\Gamma ( a, b ) = C _ { U } + C _ { D } + ( 2 \alpha ^ { + } a + q C _ { U } ) \overline { W } ^ { ( q ) } ( b - a ) + 2 \alpha ^ { + } \int _ { a } ^ { b } \overline { W } ^ { ( q ) } ( b - y ) \text{d} y.$$
We first show the following.
Lemma 6.1. For any a ≤ 0 , we have lim b ↑∞ [ W ( q ) ( b ) -W ( q ) ( b -a e ) Φ( ) q a ] = ( e Φ( ) q a -1) /q .
Proof. The case a = 0 holds trivially and hence we assume a < 0 . By (2.6), E b [ e -qτ -0 1 { τ + b -a >τ -0 } ] = 1 + qW ( q ) ( b ) -(1 + qW ( q ) ( b -a )) W ( q ) ( b /W ) ( q ) ( b -a ) . Because this converges to 0 as b ↑ ∞ , we have the convergence:
$$- \frac { 1 } { q } = \lim _ { b \uparrow \infty } \left [ \overline { W } ^ { ( q ) } ( b ) - e ^ { \Phi ( q ) a } \overline { W } ^ { ( q ) } ( b - a ) - \frac { 1 } { q } \frac { W ^ { ( q ) } ( b ) } { W ^ { ( q ) } ( b - a ) } + \frac { \overline { W } ^ { ( q ) } ( b - a ) } { W ^ { ( q ) } ( b - a ) } \left ( e ^ { \Phi ( q ) a } W ^ { ( q ) } ( b - a ) - W ^ { ( q ) } ( b ) \right ) \right ].$$
By equation (95) of Kuznetsov et al. [27], we can decompose the scale function so that W ( q ) ( y ) = e Φ( ) q y /ψ ′ (Φ( )) q -ˆ u ( q ) ( y ) , y ≥ 0 , where ˆ u ( q ) ( z ) is uniformly bounded and vanishes as z ↑ ∞ . Hence, e Φ( ) q a W ( q ) ( b -a ) -W ( q ) ( b ) = -ˆ u ( q ) ( b -a e ) Φ( ) q a +ˆ u ( q ) ( b ) b ↑∞ - - → 0 . On the other hand, by (2.7) and (2.8), W ( q ) ( b /W ) ( q ) ( b -a ) b ↑∞ - - → e Φ( ) q a and W ( q ) ( b -a /W ) ( q ) ( b -a ) b ↑∞ - - → Φ( ) q -1 . This shows the claim. glyph[square]
Proposition 6.1. Suppose f ≡ f Q as in (6.1) . Then, Case 1 always holds.
↦
Proof. It is sufficient to show that Γ( a, b ) b ↑∞ - - → -∞ . Indeed, by the continuity of Γ( a, b ) in a , this means that there must exist a > a ′ such that b → Γ( a , b ′ ) downcrosses the x-axis, which means Case 1 .
- (i) We shall first consider the case a ≤ 0 . For b > 0 , by (6.2),
$$& ( i ) \, \text{We shall first consider the case } \underline { a } \leq 0 \,. \, \text{For } b > 0, \, \text{by } ( 6. 2 ), \\ & \Gamma ( \underline { a }, b ) = C _ { U } + C _ { D } + ( 2 \alpha ^ { - } \underline { a } + q C _ { U } ) \overline { W } ^ { ( q ) } ( b - \underline { a } ) + 2 \alpha ^ { - } \int _ { 0 } ^ { b - \underline { a } } \overline { W } ^ { ( q ) } ( y ) d y + 2 ( \underline { a } ^ { + } - \alpha ^ { - } ) \int _ { 0 } ^ { b } \overline { W } ^ { ( q ) } ( y ) d y \\ & \quad = C _ { U } + C _ { D } + ( 2 \alpha ^ { - } \underline { a } + q C _ { U } ) \left ( \overline { W } ^ { ( q ) } ( b - \underline { a } ) - \Phi ( q ) \int _ { 0 } ^ { b - \underline { a } } \overline { W } ^ { ( q ) } ( y ) d y \right ) \\ & \quad - 2 ( \alpha ^ { + } - \alpha ^ { - } ) \left ( e ^ { \Phi ( q ) \underline { a } } \int _ { 0 } ^ { b - \underline { a } } \overline { W } ^ { ( q ) } ( y ) d y - \int _ { 0 } ^ { b } \overline { W } ^ { ( q ) } ( y ) d y \right ). \\ & \quad \text{By Lemma } 7. 1 \, \text{of } [ 3 8 ], \, \text{for the process } U ^ { \underline { a } \times \infty } \, \text{as defined in } ( 2. 1 0 ), \, \text{we have}$$
By Lemma 7.1 of [38], for the process U (6.4)
a, ∞ as defined in (2.10), we have
$$\overset { \ell } { \mathbb { E } } _ { x } \left [ \overset { \ell } { \int } _ { ( 0, \infty ) } e ^ { - q t } \text d U _ { t } ^ { a, \infty } \right ] = - ( x - \underline { a } ) + \frac { q } { \Phi ( q ) } \left ( \overline { W } ^ { ( q ) } ( x - \underline { a } ) - \Phi ( q ) \int _ { 0 } ^ { x - \underline { a } } \overline { W } ^ { ( q ) } ( y ) \text d y \right ) + \frac { 1 } { \Phi ( q ) } - \frac { \psi ^ { \prime } ( 0 + ) } { q }.$$
For any x > 0( ≥ a ) , because 0 ≤ U a, ∞ t ≤ -X t ∨ 0 , we have 0 ≤ ∫ [0 , ∞ ) e -qt d U a, ∞ t ≤ ∫ ∞ 0 qe -qt U a, ∞ t d t ≤ ∫ ∞ 0 qe -qt [( -X t ) ∨ 0]d t . Moreover,
$$\lim _ { x \to \infty } \mathbb { E } _ { x } \Big { [ } \int _ { 0 } ^ { \infty } q e ^ { - q t } [ ( - \underline { X } _ { t } ) \vee 0 ] { \mathrm d } t \Big { ] } = \lim _ { x \to \infty } \mathbb { E } \Big { [ } \int _ { 0 } ^ { \infty } q e ^ { - q t } [ ( - ( \underline { X } _ { t } + x ) ) \vee 0 ] { \mathrm d } t \Big { ] } = 0,$$
where the last equality holds by dominated convergence by noting that, under P , 0 ≤ ∫ ∞ 0 qe -qt [( -( X t + x )) ∨ 0]d t ≤ ∫ ∞ 0 qe -qt ( -( X t ))d t , which is integrable by (4.6). Hence, (6.4) vanishes as x → ∞ . Therefore,
$$\overline { W } ^ { ( q ) } ( b - \underline { a } ) - \Phi ( q ) \int _ { 0 } ^ { b - \underline { a } } \overline { W } ^ { ( q ) } ( y ) \text{d} y \sim \frac { \Phi ( q ) } { q } ( b - \underline { a } ),$$
where x ∼ y means x/y → 1 as b →∞ . On the other hand, by l'Hˆpital's rule and Lemma 6.1, o
$$e ^ { \Phi ( q ) \underline { q } } \int _ { 0 } ^ { b - \underline { q } } \overline { W } ^ { ( q ) } ( y ) \mathrm d y - \int _ { 0 } ^ { b } \overline { W } ^ { ( q ) } ( y ) \mathrm d y \sim b \left ( e ^ { \Phi ( q ) \underline { q } } \overline { W } ^ { ( q ) } ( b - \underline { q } ) - \overline { W } ^ { ( q ) } ( b ) \right ) \sim b \frac { 1 - e ^ { \Phi ( q ) \underline { q } } } { q }.$$
Combining these and by (6.2),
$$\lim _ { b \uparrow \infty } \frac { \Gamma ( \underline { a }, b ) } { b } = ( 2 \alpha ^ { - } \underline { a } + q C _ { U } ) \frac { \Phi ( q ) } { q } - 2 ( \alpha ^ { + } - \alpha ^ { - } ) \frac { 1 - e ^ { \Phi ( q ) \underline { a } } } { q } = - \frac { 2 \alpha ^ { + } } { q } < 0.$$
This completes the proof for the case a ≤ 0 .
(ii) For the case a > 0 , by (6.3),
$$\Gamma ( a, b ) = C _ { U } + C _ { D } - \frac { 2 \alpha ^ { + } } { \Phi ( q ) } \left ( \overline { W } ^ { ( q ) } ( b - a ) - \Phi ( q ) \int _ { a } ^ { b } \overline { W } ^ { ( q ) } ( b - y ) \text{d} y \right ),$$
which goes to -∞ by (6.5). This completes the proof.
## 6.2. Linear case. Suppose the running cost function is f ≡ f L where
$$f _ { L } ( x ) \coloneqq \alpha ^ { - } | x | 1 _ { \{ x \leq 0 \} } + \alpha ^ { + } x 1 _ { \{ x > 0 \} }, \ \ x \in \mathbb { R },$$
glyph[negationslash]
for some α , α + -∈ R . This linear cost is specifically assumed in related control problems such as [13, 37]. For any x = 0 , f ′ L ( x ) = -α -1 { x< 0 } + α + 1 { x> 0 } and ˜ ( f ′ L x ) = -α -1 { x< 0 } + α + 1 { x> 0 } + qC U . Hence, in order for Assumption 2.1 to be satisfied, we need to require that
$$q C _ { U } + \alpha ^ { + } > 0 > q C _ { U } - \alpha ^ { - }.$$
Under (6.7), we have a = 0 , and a < 0 is the unique value such that
$$( \alpha ^ { - } + \alpha ^ { + } ) e ^ { \Phi ( q ) \underline { a } } = \alpha ^ { - } - q C _ { U },$$
which always exists because (6.7) guarantees that α -+ α + > 0 and α --qC U ∈ (0 , α -+ α + ) . In addition, direct computation gives, for any a ∈ [ a, a ] and b > a ,
$$\Gamma ( a, b ) & = C _ { U } + C _ { D } + ( q C _ { U } - \alpha ^ { - } ) \overline { W } ^ { ( q ) } ( b - a ) + ( \alpha ^ { - } + \alpha ^ { + } ) \overline { W } ^ { ( q ) } ( b ), \\ \gamma ( a, b ) & = ( q C _ { U } - \alpha ^ { - } ) W ^ { ( q ) } ( b - a ) + ( \alpha ^ { - } + \alpha ^ { + } ) W ^ { ( q ) } ( b ).$$
Proposition 6.2. Suppose f ≡ f L as in (6.1) such that (6.7) holds. Then, Case 1 holds if C D < α + /q and Case 2 holds otherwise.
Proof. Because a < a = 0 , the fraction W ( q ) ( b /W ) ( q ) ( b -a ) is increasing in b (see (5.2)). Hence, for any a < b , by (6.8) (together with (2.7) and (2.8)),
$$\frac { \gamma ( \underline { a }, b ) } { W ^ { ( q ) } ( b - \underline { a } ) } \nearrow ( \alpha ^ { - } + \alpha ^ { + } ) \lim _ { b \to \infty } \frac { W ^ { ( q ) } ( b ) } { W ^ { ( q ) } ( b - \underline { a } ) } + ( q C _ { U } - \alpha ^ { - } ) = ( \alpha ^ { - } + \alpha ^ { + } ) e ^ { \Phi ( q ) \underline { a } } + ( q C _ { U } - \alpha ^ { - } ) = 0.$$
↦
Hence, γ a, b ( ) ≤ 0 uniformly in b , or equivalently the function b → Γ( a, b ) is uniformly decreasing and inf b ≥ a Γ( a, b ) = lim b →∞ Γ( a, b ) . By this, Case 1 holds if lim b →∞ Γ( a, b ) < 0 while Case 2 holds if lim b →∞ Γ( a, b ) ≥ 0 . Now the proof is complete because, by Lemma 6.1 and (6.8),
$$\lim _ { b \to \infty } \Gamma ( a, b ) & = C _ { U } + C _ { D } + ( \alpha ^ { - } + \alpha ^ { + } ) \frac { e ^ { \Phi ( q ) \underline { a } } - 1 } { q } \\ & = C _ { U } + C _ { D } - ( \alpha ^ { - } + \alpha ^ { + } ) / q + \alpha ^ { - } / q - C _ { U } = C _ { D } - \alpha ^ { + } / q.$$
glyph[square]
glyph[square]
## 7. NUMERICAL RESULTS
In this section, we conduct numerical experiments using the spectrally negative L´ evy process in the β -family introduced by [25]. The following definition is due to Definition 4 of [25].
Definition 7.1. A spectrally negative L´ evy process is said to be in the β -family if (2.1) is written
$$\psi ( z ) = \hat { \delta } z + \frac { 1 } { 2 } \sigma ^ { 2 } z ^ { 2 } + \frac { \varpi } { \beta } \left \{ B ( \alpha + \frac { z } { \beta }, 1 - \lambda ) - B ( \alpha, 1 - \lambda ) \right \}$$
for some ˆ δ ∈ R , α > 0 , β > 0 , glyph[pi1] ≥ 0 , λ ∈ (0 , 3) \{ 1 2 , } and the beta function B x, y ( ) .
This is a subclass of the meromorphic L´ evy process [26] and hence the L´ evy measure can be written
$$\nu ( \mathrm d z ) = \sum _ { j = 1 } ^ { \infty } p _ { j } \eta _ { j } e ^ { - \eta _ { j } | z | } 1 _ { \{ z < 0 \} } \mathrm d z, \ \ z \in \mathbb { R },$$
for some { p , η k k ; k ≥ 1 } . The equation ψ ( ) = · q has countably many roots {-ξ k,q ; k ≥ 1 } that are all negative real numbers and satisfy the interlacing condition: · · · < -η k < -ξ k,q < · · · < -η 2 < -ξ 2 ,q < -η 1 < -ξ 1 ,q < 0 . By using similar arguments as in [18], the scale function can be written as
$$W ^ { ( q ) } ( x ) = \frac { e ^ { \Phi ( q ) x } } { \psi ^ { \prime } ( \Phi ( q ) ) } - \sum _ { i = 1 } ^ { \infty } B _ { i, q } e ^ { - \xi _ { i, q } x }, \ \ x \geq 0,$$
where glyph[negationslash]
$$B _ { i, q } \coloneqq \frac { \Phi ( q ) } { q } \frac { \xi _ { i, q } A _ { i, q } } { \Phi ( q ) + \xi _ { i, q } } \ \text{ and } \ A _ { i, q } \coloneqq \left ( 1 - \frac { \xi _ { i, q } } { \eta _ { i } } \right ) \prod _ { j \neq i } \frac { 1 - \frac { \xi _ { i, q } } { \eta _ { j } } } { 1 - \frac { \xi _ { i, q } } { \xi _ { j, q } } }, \ \ i \geq 1.$$
Throughout this section, we suppose ˆ = 0 1 δ . , λ = 1 5 . , α = 3 , β = 1 , glyph[pi1] = 0 1 . and σ = 0 2 . . This means that the process is of unbounded variation with jumps of infinite activity. We also let q = 0 03 . .
7.1. Quadratic case. Weshall first consider the quadratic case with f ≡ f Q as in (6.1) with α -= α + = 1 . As is discussed in Proposition 6.1, Case 1 necessarily holds and a ∗ must be in ( a, a ) . Due to the fact that Γ( ) · is monotone and γ a, ( · ) satisfies Assumption 5.1 by Theorem 5.1, the values of a ∗ and b ∗ can be easily computed by a (nested) bisection-type method.
↦
Figure 1 shows the functions b → Γ( a, b ) and b → γ a, b ( ) for a = a, ( a + a ∗ ) / 2 , a ∗ , ( a ∗ + ) 2 a / , a with the common values C U = C D = 10 . As has been studied in Subsection 3.2, Γ( a, · ) starts at a positive value and increases monotonically. On the other hand, as in the proof of Proposition 6.1, Γ( a, · ) goes to -∞ (and hence Case 1 always holds). The desired value of a ∗ is such that the function is tangent (at b ∗ ) to the x-axis. It can be confirmed by the graph on the right that Assumption 5.1 indeed holds for each a .
↦
In Figure 2, we show the value functions for C U ∈ E := { 35 30 25 20 15 10 5 0 , , , , , , , , - } 5 with the common value of C D = 6 (left) and also those for C D ∈ E with C U = 6 (right). With these selections of parameters, (2.3) is always satisfied. It is clear that the value function is monotone in both C U and C D .
↦
FIGURE 1. Quadratic case: the plots of b → Γ( a, b ) (left) and b → γ a, b ( ) (right) for a = a, ( a + a ∗ ) / 2 , a ∗ , ( a ∗ + a / ) 2 , a . The line in red corresponds to the one for a = a ∗ ; the point at which Γ( a , ∗ · ) is tangent to the x-axis (or γ a , ( ∗ · ) vanishes) becomes b ∗ .
↦
The distance between a ∗ and b ∗ tends to shrink as C U + C D decreases. In all cases, we can confirm that the optimal barrier levels ( a , b ∗ ∗ ) are indeed finite.
with respect to C U
with respect to C D
FIGURE 2. Quadratic case: the plots of the value function for C U ∈ E with the common value of C D = 6 (left) and also those for C D ∈ E with C U = 6 (right). The up-pointing and down-pointing triangles show the points at a ∗ and b ∗ , respectively.
↦
7.2. Linear case. We shall then consider the linear case with f ≡ f L as in (6.6) with α -= α + = 1 . Figure 3 shows the functions b → Γ( a, b ) and b → γ a, b ( ) for a = a, ( a + a ∗ ) / 2 , a ∗ , ( a ∗ + ) 2 a / , a with the common values C U = C D = 10 . A noticeable difference with the quadratic case in Figure 1 is that the function γ a, ( · ) is not differentiable at b = 0 . Moreover, Γ( a, · ) converges to a finite value. Recall that the limit is positive if and only if a ∗ = a and b ∗ = ∞ . We can confirm that Assumption 5.1 is indeed satisfied.
↦
Similarly to Figure 2, we show in Figure 4 the value functions using the same parameter set for C U and C D . Here, we exclude the case C U = 35 because it violates (6.7). Moreover, the case C D = 35 is an example of Case 2 because C D ≥ α /q + as in Proposition 6.2. We see that because the tail of the function f grows/decreases slower than the quadratic case, the levels ( a , b ∗ ∗ ) for this linear case are more sensitive to the values of C U and C D .
↦
FIGURE 3. Linear case: the plots of b → Γ( a, b ) (left) and b → γ a, b ( ) (right) for a = a, ( a + a ∗ ) / 2 , a ∗ , ( a ∗ + ) 2 a / , a .
↦
## ACKNOWLEDGEMENTS
We thank the anonymous referee for constructive comments and suggestions. E. J. Baurdoux was visiting CIMAT, Guanajuato when part of this work was carried out and he is grateful for their hospitality and support. K. Yamazaki is in part supported by MEXT KAKENHI grant numbers 22710143 and 26800092, JSPS KAKENHI grant number 23310103, the Inamori foundation research grant, and the Kansai University subsidy for supporting young scholars 2014.
with respect to C D
FIGURE 4. Linear case: the plots of the value function for C U ∈ E \{ 35 } with the common value of C D = 6 (left) and also those for C D ∈ E with C U = 6 (right). The largest function in the right plot shows the case with C D = 35 where the optimal barrier levels are given by a ∗ = a and b ∗ = ∞ .
## REFERENCES
- [1] F. Avram, Z. Palmowski, and M. R. Pistorius. On the optimal dividend problem for a spectrally negative L´ evy process. Ann. Appl. Probab. , 17(1):156-180, 2007.
- [2] S. Baccarin. Optimal impulse control for cash management with quadratic holding-penalty costs. Decisions in Economics and Finance , 25(1):19-32, 2002.
- [3] E. J. Baurdoux and A. E. Kyprianou. The McKean stochastic game driven by a spectrally negative L´ evy process. Electron. J. Probab. , 13:no. 8, 173-197, 2008.
- [4] E. J. Baurdoux, A. E. Kyprianou, and J. C. Pardo. The Gapeev-K¨ uhn stochastic game driven by a spectrally positive L´ evy process. Stochastic Process. Appl. , 121(6):1266-1289, 2008.
- [5] E. Bayraktar and M. Egami. An analysis of monotone follower problems for diffusion processes. Math. Oper. Res. , 33(2):336-350, 2008.
- [6] E. Bayraktar, A. E. Kyprianou, and K. Yamazaki. On optimal dividends in the dual model. Astin Bull. , 43(3):359-372, 2013.
- [7] E. Bayraktar, A. E. Kyprianou, and K. Yamazaki. Optimal dividends in the dual model under transaction costs. Insurance: Math. Econom. , 54:133-143, 2014.
- [8] L. Benkherouf and A. Bensoussan. Optimality of an ( s, S ) policy with compound Poisson and diffusion demands: a quasi-variational inequalities approach. SIAM J. Control Optim. , 48(2):756-762, 2009.
- [9] A. Bensoussan, R. H. Liu, and S. P. Sethi. Optimality of an ( s, S ) policy with compound Poisson and diffusion demands: a quasi-variational inequalities approach. SIAM J. Control Optim. , 44(5):1650-1676, 2005.
- [10] J. Bertoin. L´ evy processes , volume 121 of Cambridge Tracts in Mathematics . Cambridge University Press, Cambridge, 1996.
- [11] D. Beyer, S. P. Sethi, and M. Taksar. Inventory models with Markovian demands and cost functions of polynomial growth. J. Optim. Theory Appl. , 98(2):281-323, 1998.
- [12] T. Chan, A. E. Kyprianou, and M. Savov. Smoothness of scale functions for spectrally negative L´ evy processes. Probab. Theory Relat. Fields , 150:691-708, 2011.
- [13] G. M. Constantinides and S. F. Richard. Existence of optimal simple policies for discounted-cost inventory and cash management in continuous time. Oper. Res. , 26(4):620-636, 1978.
- [14] J. Dai, D. Yao, et al. Brownian inventory models with convex holding cost, part 1: Average-optimal controls. Stochastic Systems , 3(2):442-499, 2013.
- [15] J. Dai, D. Yao, et al. Brownian inventory models with convex holding cost, part 2: Discount-optimal controls. Stochastic Systems , 3(2):500-573, 2013.
- [16] M. Egami, T. Leung, and K. Yamazaki. Default swap games driven by spectrally negative L´ evy processes. Stochastic Process. Appl. , 123(2):347-384, 2013.
- [17] M. Egami and K. Yamazaki. Precautionary measures for credit risk management in jump models. Stochastics , 85(1):111-143, 2013.
- [18] M. Egami and K. Yamazaki. Phase-type fitting of scale functions for spectrally negative L´ evy processes. J. Comput. Appl. Math. , 264:1-22, 2014.
- [19] G. D. Eppen and E. F. Fama. Cash balance and simple dynamic portfolio problems with proportional costs. International Economic Review , 10(2):119-133, 1969.
- [20] X. Guo and H. Pham. Optimal partially reversible investment with entry decision and general production function. Stochastic Process. Appl. , 115(5):705-736, 2005.
- [21] J. M. Harrison and M. I. Taksar. Instantaneous control of Brownian motion. Math. Oper. Res. , 8(3):439-453, 1983.
- [22] D. Hernandez-Hernandez and K. Yamazaki. Games of singular control and stopping driven by spectrally one-sided L´ evy processes. Stochastic Process. Appl. , forthcoming.
- [23] F. Hubalek and A. E. Kyprianou. Old and new examples of scale functions for spectrally negative L´ evy processes. Sixth Seminar on Stochastic Analysis, Random Fields and Applications, eds R. Dalang, M. Dozzi, F. Russo. Progress in Probability, Birkh¨user a , 2010.
- [24] I. Karatzas. The monotone follower problem in stochastic decision theory. Appl. Math. Optim. , 7(1):175-189, 1981.
- [25] A. Kuznetsov. Wiener-Hopf factorization and distribution of extrema for a family of L´ evy processes. Ann. Appl. Probab. , 2009.
- [26] A. Kuznetsov, A. E. Kyprianou, and J. C. Pardo. Meromorphic L´ evy processes and their fluctuation identities. Ann. Appl. Probab. , 22(3):1101-1135, 2012.
- [27] A. Kuznetsov, A. E. Kyprianou, and V. Rivero. The theory of scale functions for spectrally negative levy processes. Springer Lecture Notes in Mathematics , 2061:97-186, 2013.
- [28] A. E. Kyprianou. Introductory lectures on fluctuations of L´ evy processes with applications . Universitext. SpringerVerlag, Berlin, 2006.
- [29] A. E. Kyprianou and B. A. Surya. Principles of smooth and continuous fit in the determination of endogenous bankruptcy levels. Finance Stoch. , 11(1):131-152, 2007.
- [30] R. L. Loeffen. On optimality of the barrier strategy in de Finetti's dividend problem for spectrally negative L´ evy processes. Ann. Appl. Probab. , 18(5):1669-1680, 2008.
- [31] P. Matom¨ aki. On solvability of a two-sided singular control problem. Math. Method Oper. Res. , 76(3):239-271, 2012.
- [32] A. Merhi and M. Zervos. A model for reversible investment capacity expansion. SIAM J. Control Optim. , 46(3):839-876, 2007.
- [33] M. R. Pistorius. On doubly reflected completely asymmetric L´ evy processes. Stochastic Process. Appl. , 107(1):131-143, 2003.
- [34] M. R. Pistorius. On exit and ergodicity of the spectrally one-sided L´ evy process reflected at its infimum. J. Theoret. Probab. , 17(1):183-220, 2004.
- [35] P. E. Protter. Stochastic integration and differential equations , volume 21 of Stochastic Modelling and Applied Probability . Springer-Verlag, Berlin, 2005. Second edition. Version 2.1, Corrected third printing.
- [36] S. P. Sethi and M. I. Taksar. Optimal financing of a corporation subject to random returns. Math. Finance , 12(2):155-172, 2002.
- [37] A. Sulem. A solvable one-dimensional model of a diffusion inventory system. Math. Oper. Res. , 11(1):125-133, 1986.
- [38] K. Yamazaki. Inventory control for spectrally positive L´ evy demand processes. arXiv:1303.5163 , 2013.
| null |
[
"Erik J. Baurdoux",
"Kazutoshi Yamazaki"
] |
2014-08-05T01:35:49+00:00
|
2015-02-05T08:06:22+00:00
|
[
"math.OC",
"math.PR",
"60G51, 93E20, 49J40"
] |
Optimality of doubly reflected Levy processes in singular control
|
We consider a class of two-sided singular control problems. A controller
either increases or decreases a given spectrally negative Levy process so as to
minimize the total costs comprising of the running and control costs where the
latter is proportional to the size of control. We provide a sufficient
condition for the optimality of a double barrier strategy, and in particular
show that it holds when the running cost function is convex. Using the
fluctuation theory of doubly reflected Levy processes, we express concisely the
optimal strategy as well as the value function using the scale function.
Numerical examples are provided to confirm the analytical results.
|
1408.0848v8
|
## Multilayer Bootstrap Networks
Xiao-Lei Zhang GLYPH<3>
Center for Intelligent Acoustics and Immersive Communications, School of Marine Science and Technology, Northwestern Polytechnical University, China
## Abstract
Multilayer bootstrap network builds a gradually narrowed multilayer nonlinear network from bottom up for unsupervised nonlinear dimensionality reduction. Each layer of the network is a nonparametric density estimator. It consists of a group of k-centroids clusterings. Each clustering randomly selects data points with randomly selected features as its centroids, and learns a one-hot encoder by one-nearest-neighbor optimization. Geometrically, the nonparametric density estimator at each layer projects the input data space to a uniformly-distributed discrete feature space, where the similarity of two data points in the discrete feature space is measured by the number of the nearest centroids they share in common. The multilayer network gradually reduces the nonlinear variations of data from bottom up by building a vast number of hierarchical trees implicitly on the original data space. Theoretically, the estimation error caused by the nonparametric density estimator is proportional to the correlation between the clusterings, both of which are reduced by the randomization steps.
Keywords: Resampling, ensemble, nearest neighbor, tree
## 1. Introduction
Principal component analysis (PCA) (Pearson, 1901) is a simple and widely used unsupervised dimensionality reduction method, which finds a coordinate system in the original Euclidean space that the linearly uncorrelated coordinate axes (called principal components) describe the most variances of data. Because PCA is insu GLYPH<14> cient to capture highly-nonlinear data distributions, many dimensionality reduction methods are explored.
Dimensionality reduction has two core steps. The first step finds a suitable feature space where the density of data with the new feature representation can be well discovered, i.e. a density estimation problem. The second step discards the noise components or small variations of the data with the new feature representation, i.e. a principal component reduction problem in the new feature space.
Dimensionality reduction methods are either linear (He & Niyogi, 2004) or nonlinear based on the connection between the data space and the feature space. This paper focuses on nonlinear methods, which can be categorized to three classes. The first class is kernel methods. It first projects data to a kernelinduced feature space, and then conducts PCA or its variants in the new space. Examples include kernel PCA (Schölkopf et al., 1998), Isomap (Tenenbaum et al., 2000), locally linear embedding (LLE) (Roweis & Saul, 2000), Laplacian eigenmaps (Shi & Malik, 2000; Ng et al., 2002; Belkin & Niyogi, 2003), t -distributed stochastic neighbor embedding (t-SNE) (Van der
GLYPH<3> Corresponding author
Email address:
[email protected],
[email protected] (Xiao-Lei Zhang)
URL: www.xiaolei-zhang.net (Xiao-Lei Zhang)
Maaten & Hinton, 2008), and their generalizations (Yan et al., 2007; Nie et al., 2011). The second class is probabilistic models. It assumes that data are generated from an underlying probability function, and takes the posterior parameters as the feature representation. Examples include Gaussian mixture model and latent Dirichlet allocation (Blei et al., 2003). The third class is autoassociative neural networks (Hinton & Salakhutdinov, 2006). It learns a piecewise-linear coordinate system explicitly by backpropagation, and uses the output of the bottleneck layer as the new representation.
However, the feature representations produced by the aforementioned methods are defined in continuous spaces. A fundamental weakness of using a continuous space is that it is hard to find a simple mathematical form that transforms the data space to an ideal continuous feature space, since a real-world data distribution may be non-uniform and irregular. To overcome this di GLYPH<14> culty, a large number of machine learning methods have been proposed, such as distance metric learning (Xing et al., 2002) and kernel learning (Lanckriet et al., 2004) for kernel methods, and Dirichlet process prior for Baysian probabilistic models (Teh et al., 2005), in which advanced optimization methods have to be applied. Recently, learning multiple layers of nonlinear transforms, named deep learning, is a trend (Hinton & Salakhutdinov, 2006). A deep network contains more than one nonlinear layers. Each layer consists of a group of nonlinear computational units in parallel. Due to the hierarchical structure and distributed representation at each layer, the representation learning ability of a deep network is exponentially more powerful than that of a shallow network when given the same number of nonlinear units. However, the development of deep learning was mostly supervised, e.g. (Schmidhuber, 2015; Hinton et al., 2012; Wang & Chen, 2017; He et al., 2016;
Zhou & Feng, 2017; Wang et al., 2017). To our knowledge, deep learning for unsupervised dimensionality reduction seems far from explored (Hinton & Salakhutdinov, 2006).
To overcome the aforementioned weakness in a simple way, we revisit the definition of frequentist probability for the density estimation subproblem of dimensionality reduction. Frequentist probability defines an event's probability as the limit of its relative frequency in a large number of trials (Wikipedia, 2017). In other words, the density of a local region of a probability distribution can be approximated by counting the events that fall into the local region. This paper focuses on exploring this idea. To generate the events, we resort to random resampling in statistics (Efron, 1979; Efron & Tibshirani, 1993). To count the events, we resort to one-nearest-neighbor optimization and binarize the feature space to a discrete space.
To further reduce the small variations and noise components of data, i.e. the second step of dimensionality reduction, we extend the density estimator to a gradually narrowed deep architecture, which essentially builds a vast number of hierarchical trees on the discrete feature space. The overall simple algorithm is named multilayer bootstrap networks (MBN).
To our knowledge, although ensemble learning (Breiman, 2001; Freund & Schapire, 1995; Friedman et al., 2000; Dietterich, 2000; Tao et al., 2006), which was triggered by random resampling, is a large family of machine learning, it is not very prevalent in unsupervised dimensionality reduction. Furthermore, we did not find methods that estimate the density of data in discrete spaces by random resampling, nor their extensions to deep learning.
This paper is organized as follows. In Section 2, we describe MBN.In Section 3, we give a geometric interpretation of MBN. In Section 4, we justify MBN theoretically. In Section 5, we study MBN empirically. In Section 6, we introduce some related work. In Section 7, we summarize our contributions.
## 2. Multilayer bootstrap networks
## 2.1. Network structure
MBN contains multiple hidden layers and an output layer (Fig. 1). Each hidden layer consists of a group of mutually independent k -centroids clusterings; each k -centroids clustering has k output units, each of which indicates one cluster; the output units of all k -centroids clusterings are concatenated as the input of their upper layer. The output layer is PCA.
The network is gradually narrowed from bottom up, which is implemented by setting parameter k as large as possible at the bottom layer and be smaller and smaller along with the increase of the number of layers until a predefined smallest k is reached.
## 2.2. Training method
MBNis trained layer-by-layer from bottom up.
For training each layer given a d -dimensional input data set X = f x 1 ; : : : ; x n g either from the lower layer or from the original data space, we simply need to focus on training each k -centroids clustering, which consists of the following steps:
W
1
W
2
W
(a) Figure 1: Network structure. The dimension of the input data for this demo network is 4. Each colored square represents a k -centroids clustering. Each layer contains 3 clusterings. Parameters k at layers 1, 2, and 3 are set to 6, 3, and 2 respectively. The outputs of all clusterings in a layer are concatenated as the input of their upper layer.
- GLYPH<15> Random sampling of features. The first step randomly selects ˆ d dimensions of X ( ˆ d GLYPH<20> d ) to form a subset of X , denoted as ˆ X = f ˆ x 1 ; : : : ; ˆ x n g .
- GLYPH<15> Random sampling of data. The second step randomly selects k data points from ˆ X as the k centroids of the clustering, denoted as f w 1 ; : : : ; w k g .
- GLYPH<15> One-nearest-neighbor learning. The new representation of an input ˆ produced by the current clustering is an inx dicator vector h which indicates the nearest centroid of ˆ. x For example, if the second centroid is the nearest one to ˆ, x then h = [0 ; 1 0 ; ; : : : ; 0] T . The similarity metric between the centroids and ˆ at the bottom layer is customized, e.g. x the squared Euclidean distance arg min k i = 1 k w i GLYPH<0> ˆ x k 2 , and set to arg max k i = 1 w x T i ˆ at all other hidden layers.
## 2.3. Novelty and advantages
Two novel components of MBN distinguish it from other dimensionality reduction methods.
The first component is that each layer is a nonparametric density estimator based on resampling, which has the following major merits:
- GLYPH<15> It estimates the density of data correctly without any predefined model assumptions. As a corollary, it is insensitive to outliers.
- GLYPH<15> The representation ability of a group of k -centroids clusterings is exponentially more powerful than that of a single k -centroids clustering.
- GLYPH<15> The estimation error introduced by binarizing the feature space can be controlled to a small value by simply increasing the number of the clusterings.
The second component is that MBN reduces the small variations and noise components of data by an unsupervised deep ensemble architecture, which has the following main merits:
- GLYPH<15> It reduces larger and larger local variations of data gradually from bottom up by building as many as O kL ( 2 V ) hierarchical trees on the data space (instead of on data points)
3
Cyclic-s
(b)
Figure 2: Principle of compressive MBN. The numbers are the dimensions of representations.
implicitly, where L is the total number of layers, kL is parameter k at the L -th layer, V is the number of the clusterings at the layer, and function O ( ) is the GLYPH<1> order in mathematics.
- GLYPH<15> It does not inherit the problems of deep neural networks such as local minima, overfitting to small-scale data, and gradient vanishing, since it is trained simply by random resampling and stacking. As a result, it can be trained with as many hidden layers as needed and with both small-scale and large-scale data.
See Sections 3 and 4 for the analysis of the above properties.
## 2.4. Weaknesses
The main weakness of MBN is that the size of the network is large. Although MBN supports parallel computing, its prediction process is ine GLYPH<14> cient, compared to a neural network.
To overcome this weakness, we propose compressive MBN (Fig. 2). Compressive MBN uses a neural network to learn a mapping function from the training data to the output of MBN at the training stage, and uses the neural network for prediction. More generally, if MBN is applied to some specific task, then we can use compressive MBN to learn a mapping function from the input data to the output of the task directly.
## 3. Geometric interpretation
In this section, we analyze the two components of MBN from the geometric point of view.
## 3.1. Feature learning by a nonparametric density estimator based on resampling
As presented in Section 1, a core problem of machine learning is to find a suitable similarity metric that maps the original data space to a uniformly-distributed feature space. Here we give an example on its importance. As shown in Fig. 3, the similarity between the data points that are far apart in a distribution with a large variance (e.g., x 1 and x 2 in a distribution P 1 ) might be the same as the similarity between the data points that are close to each other in a distribution with a small variance (e.g., x 3 and x 4 in a distribution P 2 ). If we use the Euclidean distance as the similarity metric, then the similarity between x 1 and x 2 is smaller than that between x 3 and x 4 , which is not true.
The proposed method provides a simple solution to the above similarity metric learning problem. It outputs a new feature representation of x 1 as follows. Each k -centroids clustering contributes a neighboring centroid of x 1 . The centroid partitions the local region of x 1 to two disconnected parts, one containing x 1 and the other not. The data point x 1 owns a local region supported by the centroids of all clusterings that are closest to the data point. 1 These centroids partition the local region to as many as O (2 V ) fractions. Each fraction is represented as a unique binary code at the output space of the estimator in a way illustrated in Fig. 4. 2 The new representation of x 1 in the output feature space is a binary code that represents the local fraction of the data space where x 1 is located.
With the new representation, the similarity between two data points is calculated by counting the number of the nearest centroids they share in common-a method in frequentist methodology. In Fig. 3, (i) if the local region in P 1 is partitioned in the same way as the local region in P 2 , and if the only di GLYPH<11> erence between them is that the local region in P 1 is an amplification of the local region in P 2 , then the surfaces of the two local regions are the same in the discrete feature space. (ii) x 1 and x 2 share 5 common nearest centroids, and x 3 and x 4 also share 5 common nearest centroids, so that the similarity between x 1 and x 2 in P 1 equals to the similarity between x 3 and x 4 in P 2 .
Note that because V k -centroids clusterings are able to partition the data space to O k ( 2 V ) fractions at the maximum, its representation ability is exponentially more powerful than a single k -centroids clustering.
## 3.2. Dimensionality reduction by a deep ensemble architecture
The nonparametric density estimator captures the variances of the input data, however, it is not responsible for reducing the small variances and noise components. To reduce the nonlinear variations, we build a gradually narrowed deep network. The network essentially reduces the nonlinear variations of data by building a vast number of hierarchical trees on the data space (instead of on data points) implicitly. We present its geometric principle as follows.
Suppose MBN has L layers, parameters k at layers 1 to L are k 1 ; k 2 ; : : : ; kL respectively, and k 1 > k 2 > : : : > kL . From Fig. 4, we know that the l -th layer partitions the input data space to O kl ( 2 V ) disconnected small fractions. Each fraction is encoded as a single point in the output feature space. The points in the output feature space are the nodes of the trees. Hence, the l -th layer has O kl ( 2 V ) nodes. The bottom layer has O k ( 1 2 V ) leaf nodes. The top layer has O kL ( 2 V ) root nodes, which is the number of the trees that MBN builds.
1 It is important to keep parameter k of the k -centroids clusterings in a layer the same. Otherwise, the density of the centroids between the clusterings are di GLYPH<11> erent, that is to say, if we regard the centroids as the coordinates axes of the local region, then the coordinate axes are built in di GLYPH<11> erent data spaces.
2 A small fraction should not have to contain data points as shown in Fig. 4.
(a)
(b)
Figure 3: Illustration of the similarity metric problem at the data space. (a) The similarity problem of two data points x 1 (in red color) and x 2 (in blue color) in a distribution P 1. The local region of x 1 (or x 2) is the area in a colored circle that is centered at x 1 (or x 2). The small hollow points that lie in the cross area of the two local regions are the shared centroids by x 1 and x 2. (b) The similarity problem of two data points x 3 and x 4 in a distribution P 2.
Figure 4: Encoding the local region of a data point x 1 (in blue color) by (a) one k -centroids clustering, (b) two k -centroids clusterings, and (c) three k -centroids clusterings. The centroids of the three k -centroids clusterings are the points colored in red, black, and green respectively. The local region of a centroid is the area in the circle around the centroid. The centroid itself and the edge of its local region are drawn in the same color.
For two adjacent layers, because kl GLYPH<0> 1 < kl , it is easy to see that O kl ( GLYPH<0> 1 = kl ) neighboring child nodes at the ( l GLYPH<0> 1)-th layer will be merged to a single father node at the l -th layer on average. To generalize the above property to the entire MBN, a single root node at the L -th layer is a merging of O k ( 1 = kL ) leaf nodes at the bottom layer, which means that the nonlinear variations among the leaf nodes that are to be merged to the same root node will be reduced completely. Because kL GLYPH<28> k 1 , we can conclude that MBNis highly invariant to the nonlinear variations of data.
One special case is that, if kL = 1, the output feature spaces of all k -centroids clusterings at the top layer are just a single point. However, in practice, we never set kL = 1; instead, we usually set the termination condition of MBN as kL GLYPH<21> 1 5 : c where c is the ground-truth number of classes. The termination condition makes each k -centroids clustering stronger than random guess (Schapire, 1990).
Note that because O k ( 2 V ) GLYPH<29> n , the data points are distributed sparsely in the output feature space. Hence, MBN seldom merges data points. In other words, MBN learns data representations instead of doing agglomerative data clustering.
## 4. Theoretical analysis
Although MBN estimates the density of data in the discrete feature space, its estimation error is small and controllable. Specifically, as shown in Fig. 4c, V k -centroids clusterings can partition a local region to O (2 V ) disconnected small fractions at the maximum. It is easy to imagine that, for any local region, when V is increasing and the diversity between the centroids is still reserved, an ensemble of k -centroids clusterings approximates the true data distribution. Because the diversity is important, we have used two randomized steps to enlarge it.
In the following, we analyze the estimation error of the proposed method formally:
Theorem 1. The estimation error of the building block of MBN E ensemble and the estimation error of a single k-centroids clustering E single have the following relationship:
$$\mathcal { E } _ { e n s e n b l e } = \left ( \frac { 1 } { V } + \left ( 1 - \frac { 1 } { V } \right ) \rho \right ) \mathcal { E } _ { s i n g l e }$$
where GLYPH<26> is the pairwise positive correlation coe GLYPH<14> cient between the k-centroids clusterings, 0 GLYPH<20> GLYPH<26> GLYPH<20> 1 .
Proof. We prove Theorem 1 by first transferring MBN to a supervised regression problem and then using the bias-variance
decomposition of the mean squared error to get the bias and variance components of MBN. The detail is as follows.
Suppose the random samples for training the k -centroids clusterings are identically distributed but not necessarily independent, and the pairwise positive correlation coe GLYPH<14> cient between two random samples (i.e., f w v 1 ; i g k i = 1 and f w v 2 ; j g k j = 1 , 8 v 1 ; v 2 = 1 ; : : : ; V and v 1 , v 2 ) is GLYPH<26> , 0 GLYPH<20> GLYPH<26> GLYPH<20> 1.
We focus on analyzing a given point x , and assume that the true local coordinate of x is s which is an invariant point around x and usually found when the density of the nearest centroids around x goes to infinity (i.e. n ! 1 ; V ! 1 f , w v g V v = 1 are identically and independently distributed, and parameter k is unchanged). We also suppose that x is projected to ˆ when given s a finite number of nearest centroids f w v g V v = 1 . The correlation coe GLYPH<14> cient between the centroids f w v g V v = 1 is GLYPH<26> . Note that s is used as an invariant reference point for studying ˆ. s
The e GLYPH<11> ectiveness of MBN can be evaluated by the estimation error of the estimate ˆ to the truth s s . The estimation error is defined by E ( k s GLYPH<0> ˆ s k 2 ) where E ( ) is the expectation of a random GLYPH<1> variable.
From the geometric interpretation, we know that each w v owns a local space S v , and moreover, both s and ˆ are in s S v . When parameter k ! n S v , is small enough to be locally linear. Under the above fact and an assumption that the features of w v , i.e. wv ; 1 ; : : : ; wv d ; ; : : : ; wv D ; , are uncorrelated, we are able to assume that w v follows a multivariate normal distribution around s : w v GLYPH<24> MN ( s ; GLYPH<27> 2 I ) where I is the identity matrix, GLYPH<27> describes the variance of w v and GLYPH<27> 2 I is the covariance matrix. We can further derive E ( k s GLYPH<0> ˆ s k 2 ) = P D d = 1 E (( sd GLYPH<0> ˆ sd ) 2 ) where we denote s = [ s 1 ; : : : ; sD ] T and ˆ s = [ ˆ s 1 ; : : : ; ˆ sD ] T .
In the following, we focus on analyzing the estimation error at a single dimension E (( sd GLYPH<0> ˆ sd ) 2 ) and will omit the index d for clarity. It is known that the mean squared error of a regression problem can be decomposed to the summation of a squared bias component and a variance component (Hastie et al., 2009):
$$E ( ( s - \hat { s } ) ^ { 2 } ) & = ( s - E ( \hat { s } ) ) ^ { 2 } + E \left ( ( \hat { s } - E ( \hat { s } ) ) ^ { 2 } \right ) & 5. 1. \ F \\ & = \text{Bias} ^ { 2 } ( \hat { s } ) + \text{Var} ( \hat { s } ) & ( 2 ) & \text{Wh}$$
Because wv follows a univariate normal distribution, it is easy to obtain:
$$E ( w _ { v } ) \, = \, s & & ( 3 ) & & \stackrel { \dots } { \text{M} }$$
$$E ( w _ { v } ^ { 2 } ) & = \sigma ^ { 2 } + s ^ { 2 } \sum _ { \cdot } ^ { 2 } & ( 4 ) & \quad a \in [ \begin{matrix} \dots \\ \end{matrix} }$$
$$E ( w _ { v _ { 1 } } w _ { v _ { 2 } } ) & = \rho \sigma ^ { 2 } + s ^ { 2 } & ( 5 ) \quad \text{the de}$$
For a single estimate wv , we have ˆ sv = wv . For a set of estimates f wv v g V = 1 , we have ˆ s GLYPH<6> = 1 V P V v = 1 wv . Based on Eqs. (3) to (5), we can derive:
$$\ B i a s ^ { 2 } ( \hat { s } _ { \nu } ) & = 0 & ( 6 ) & \quad \text{ine nc} \\ \text{defaul}$$
$$\sigma _ { \text{single} } ^ { 2 } = \text{Var} ( \hat { s } _ { v } ) = \sigma ^ { 2 } & & ( 7 ) \quad \text{the ou}$$
$$\ B i a s ^ { 2 } ( \hat { s } _ { \Sigma } ) & = 0 & ( 8 ) & \quad t i o n a l k$$
and
$$\sigma _ { \text{ensemble} } ^ { 2 } \, = \, \text{Var} \, ( \hat { s } _ { \Sigma } ) = \frac { \sigma ^ { 2 } } { V } + \left ( 1 - \frac { 1 } { V } \right ) \rho \sigma ^ { 2 }.$$
Substituting Eqs. (6) and (7) to Eq. (2) gets:
$$t \quad \mathcal { E } _ { \text{single} } = \sigma ^ { 2 }$$
and substituting Eqs. (8) and (9) to Eq. (2) gets:
$$\mathbb { E } _ { \substack { 1 \\ \lambda } } \quad \mathcal { E } _ { \text{ensemble} } \, = \, \frac { \sigma ^ { 2 } } { V } + \left ( 1 - \frac { 1 } { V } \right ) \rho \sigma ^ { 2 }.$$
Theorem 1 is proved.
From Theorem 1, we can get the following corollaries easily:
Corollary 1. When GLYPH<26> is reduced from 1 to 0 , E ensemble is reduced from E single to E single = V accordingly.
Corollary 2. When V ! 1 , E ensemble reaches a lower bound GLYPH<26> E single .
From Corollary 1, we know that increasing parameter V and reducing GLYPH<26> can help MBN reduce the estimation error.
From Corollary 2, we know that it is important to reduce GLYPH<26> . We have adopted two randomization steps to reduce GLYPH<26> . However, although decreasing parameters a and k can help MBN reduce GLYPH<26> , it will also cause E single rise. In other words, reducing E single and reducing GLYPH<26> is a pair of contradictory factors, which needs a balance through a proper parameter setting.
## 5. Empirical evaluation
In this section, we first introduce a typical parameter setting of MBN, then demonstrate the density estimation ability of MBN on synthetic data sets, and finally apply the low dimensional output of MBN to the tasks of visualization, clustering, and document retrieval.
## 5.1. Parameter setting
Whenadata set is small-scale, we use the linear-kernel-based kernel PCA (Schölkopf et al., 1998; Canu et al., 2005) as the PCA toolbox of MBN. When a data set is middle- or largescale, we use the expectation-maximization PCA (EM-PCA) (Roweis, 1998). 3
MBN is insensitive to parameters V and a as if V > 100 and a 2 [0 5 : ; 1]. If not specified, we used V = 400 and a = 0 5 as : the default values.
We denote parameter k at layer l as kl . To control kl , we introduce a parameter GLYPH<14> defined as kl + 1 = GLYPH<14> kl , GLYPH<14> 2 (0 ; 1). MBN is relatively sensitive to parameter GLYPH<14> : if data are highly-nonlinear, then set GLYPH<14> to a large value, otherwise, set GLYPH<14> to a small value; if the nonlinearity of data is unknown, set GLYPH<14> = 0 5 which is our : default.
Finally, if kL + 1 < 1 5 : c , then we stop MBN training and use the output of the L -th nonlinear layer for PCA. This terminating
3 The word 'large-scale' means that the data cannot be handled by traditional kernel methods on a common personal computer.
Table 1: Hyperparameters of MBN.
| Parameter | Description |
|-------------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| GLYPH<14> | A parameter that controls the network structure by k l + 1 = GLYPH<14> k l , 8 l = 1 ; : : : ; L |
| a | Fraction of randomly selected dimensions (i.e., ˆ d ) over all dimensions (i.e., d ) of input data. |
| V | Number of k -centroids clusterings per layer. |
| k 1 | A parameter that controls the time and storage complexities of the network. For small-scale problems, k 1 = 0 : 5 n . For large-scale problems, k 1 should be tuned smaller to fit MBNto the computing power. |
Figure 5: Density estimation by MBN on synthetic data sets. The data points in di GLYPH<11> erent classes are drawn in di GLYPH<11> erent colors.
condition guarantees the validness of data resampling which requires each random sample of data to be stronger than random guess (Schapire, 1990). The hyperparameters of MBN are summarized in Table 1.
In the following experiments, we adopted the above default parameter setting of MBN, unless otherwise specified.
## 5.2. Density estimation
## 5.2.1. Density estimation for nonlinear data distributions
Four synthetic data sets with non-uniform densities and nonlinear variations are used for evaluation. They are Gaussian data, Jain data (Jain & Law, 2005), pathbased data (Chang & Yeung, 2008), and compound data (Zahn, 1971), respectively. Parameter a was set to 1.
The visualization result in Fig. 5 shows that the synthetic data in the new feature spaces produced by MBN not only are distributed uniformly but also do not contain many nonlinear variations.
## 5.2.2. Density estimation in the presence of outliers
A data set of two-class Gaussian data with a randomly generated outlier is used for evaluation. The results of PCA and
MBN are show in Fig. 6. From the figure, we observe that MBNis robust to the presence of the outlier.
## 5.3. Data visualization
## 5.3.1. Visualizing AML-ALL biomedical data
The acute myeloid leukemia and acute lymphoblastic leukemia (AML-ALL) biomedical data set (Golub et al., 1999) is a two-class problem that consists of 38 training examples (27 ALL, 11 AML) and 34 test examples (20 ALL, 14 AML). Each example has 7,129 dimensions produced from 6,817 human genes.
We compared MBN with PCA and 2 nonlinear dimensionality reduction methods which are Isomap (Tenenbaum et al., 2000) and LLE (Roweis & Saul, 2000). We tuned the hyperparameters of Isomap and LLE for their best performance. 4 The visualization result is shown in Fig. 7.
## 5.3.2. Visualizing MNIST digits
The data set of the MNIST digits (Lecun et al., 2004) contains 10 handwritten integer digits ranging from 0 to 9. It con-
4 It is di GLYPH<14> cult and sometimes unable to tune the hyperparameters in practice.
Figure 6: Density estimation by MBN on a Gaussian data with the presence of an outlier. The red cross denotes the outlier. Because the outlier is far from the Gaussian data, the resolution of the sub-figure in the top-right corner is not high enough to di GLYPH<11> erentiate the two classes of the Gaussian data.
sists of 60,000 training images and 10,000 test images. Each image has 784 dimensions.
We compared MBN with PCA, Isomap (Tenenbaum et al., 2000), LLE (Roweis & Saul, 2000), Spectral (Ng et al., 2002), deep belief networks (Hinton & Salakhutdinov, 2006) (DBN), and t-SNE (Van der Maaten & Hinton, 2008). We tuned the hyperparameters of the 5 nonlinear comparison methods for their best performance.
Because the comparison nonlinear methods are too costly to run with the full MNIST data set except DBN, we randomly sampled 5,000 images with 500 images per digit for evaluation. The visualization result in Fig. 8 shows that the lowdimensional feature produced by MBN has a small within-class variance and a large between-class distance.
To demonstrate the scalability of MBN on larger data sets and its generalization ability on unseen data, we further trained MBNonall 60,000 training images, and evaluated its e GLYPH<11> ectiveness on the 10,000 test images, where k 1 was set to 8000 to reduce the training cost. The visualization result in Fig. 9 shows that MBN on the full MNIST provides a clearer visualization than that on the small subset of MNIST.
## 5.4. Clustering
Ten benchmark data sets were used for evaluation. They cover topics in speech processing, chemistry, biomedicine, image processing, and text processing. The details of the data sets are given in Table 2. The data set '20-Newsgroups', which originally has 20,000 documents, was post-processed to a corpus of 18,846 documents, each belonging to a single topic.
Wecompared MBN with PCA and Spectral (Ng et al., 2002). We applied k -means clustering to the low-dimensional outputs of all comparison methods as well as the original highdimensional features. To prevent the local minima problem of k -means clustering, we ran k -means clustering 50 times and picked the clustering result that corresponded to the optimal objective value of the k -means clustering among the 50 candidate objective values as the final result. We ran each comparison method 10 times and reported the average performance.
The parameter settings on the data sets with IDs from 1 to 9 are as follows. For PCA, we preserved the top 98% largest eigenvalues and their corresponding eigenvectors. For Spectral, we set the output dimension to the ground-truth number of classes and adopted the RBF kernel. We reported the results with a fixed kernel width 2 GLYPH<0> 4 A which behaves averagely the best over all data sets, as well as the best result by searching the kernel width from f 2 GLYPH<0> 4 A ; 2 GLYPH<0> 3 A ; : : : ; 2 4 A g on each data set, where A is the average pairwise Euclidean distance between data. The two parameter selection methods of Spectral are denoted as Spectral no tuning \_ and Spectral optimal respectively. For MBN, we set the output dimension to the ground-truth number of classes. We reported the results with GLYPH<14> = 0 5, as well as the : best results by searching GLYPH<14> from f 0 1 : ; 0 2 : ; : : : ; 0 9 : g on each data set. The two parameter selection methods of MBN are denoted as MBN no tuning \_ and MBN optimal respectively.
The parameter settings on the 20-Newsgroups are as follows. For PCA, we set the output dimension to 100. For all comparison methods, we used the cosine similarity as the similarity metric.
We evaluated the clustering result in terms of normalized mutual information (NMI) and clustering accuracy. The clustering results in Tables 3 and 4 show that (i) MBN no \_ tuning achieves better performance than Spectral no tuning \_ and PCA, and (ii) MBN optimal achieves better performance than Spectral optimal .
Besides the data sets in Table 2, we have also conducted experiments on the following data sets: Lung-Cancer biomedical data, COIL20 images, USPS images, Extended-YaleB images, Reuters-21578 text corpus, and TDT2 text corpus. The experimental conclusions are consistent with the results in Tables 3 and 4.
## 5.4.1. E ect of hyperparameter GLYPH<11> GLYPH<14>
We showed the e GLYPH<11> ect of parameter GLYPH<14> on each data set in Fig. 10. From the figure, we do not observe a stable interval of GLYPH<14> where MBN is supposed to achieve the optimal performance across the data sets; if the data are highly variant, then setting GLYPH<14> to a large value yields good performance, and vise versa. Generally, if the nonlinearity of data is unknown, then setting GLYPH<14> = 0 5 is safe. :
## 5.4.2. E ects of hyperparameters a and V GLYPH<11>
Theorem 1 has guaranteed that the estimation error can be reduced by enlarging parameter V , and may also be reduced by
Figure 8: Visualizations of a subset (5,000 images) of MNIST produced by MBN and 6 comparison methods. For clarity, only 250 images per digit are drawn.
decreasing parameter a . In this subsection, we focus on investigating the stable working intervals of V and a .
Because the experimental phenomena on all data sets are similar, we reported the phenomena on the small subset of MNIST as a representative in Fig. 11. From the figure, we know that (i) we should prevent setting a to a very small value. Empirically, we set a = 0 5. : (ii) We should set V GLYPH<21> 100. Empirically, we set V = 400.
structure. We only preserved the 82 leaf topics. As a result, there were 107,132 unlabeled documents. We preprocessed each document as a vector of 2,000 commonest word stems by the rainbow software (McCallum, 1998) where each entry of a vector was set to the word count.
## 5.5. Document retrieval
3-dimensional feature 0.7 0.8 0.9 1 2-dimensional feature 5-dimensional feature B We applied MBN to document retrieval and compared it with latent semantic analysis (LSA) (Deerwester et al., 1990), a document retrieval method based on PCA, on a larger data setReuters newswire stories (Lewis et al., 2004) which consist of 804,414 documents. The data set of the Reuters newswire stories are divided into 103 topics which are grouped into a tree
We randomly selected half of the data set for training and the other half for test. We recorded the average accuracy over all 402,207 queries in the test set at the document retrieval setting, where a query and its retrieved documents were di GLYPH<11> erent documents in the test set. If an unlabeled document was retrieved, it was considered as a mistake. If an unlabeled document was used as a query, no relevant documents would be retrieved, which means the precisions of the unlabeled query at all levels were zero.
Because the data set is relatively large, we did not adopt the default setting of MBN where k 1 = 201103. Instead, we set k 1
0.6
0.5
0.4
Accuracy
0.3
0.2
0 1
2 3 4 5 6 7 8 9
0 1
2 3
4 5
6 7 8 9
0 1
2 3 4 5 6
10-dimensional feature
20-dimensional feature
30-dimensional feature
7 8 9
Table 2: Description of data sets.
| ID | Name | # data points | # dimensions | # classes | Attribute |
|------|---------------|-----------------|----------------|-------------|-----------------------------|
| 1 | Isolet1 | 1560 | 617 | 26 | Speech data |
| 2 | Wine | 178 | 13 | 3 | Chemical data |
| 3 | New-Thyroid | 215 | 5 | 3 | Biomedical data |
| 4 | Dermathology | 366 | 34 | 6 | Biomedical data |
| 5 | COIL100 | 7200 | 1024 | 100 | Images |
| 6 | MNIST(small) | 5000 | 768 | 10 | Images (handwritten digits) |
| 7 | MNIST(full) | 70000 | 768 | 10 | Images (handwritten digits) |
| 8 | UMIST | 575 | 1024 | 20 | Images (faces) |
| 9 | ORL | 400 | 1024 | 40 | Images (faces) |
| 10 | 20-Newsgroups | 18846 | 26214 | 20 | Text corpus |
Table 3: NMI on 10 data sets. The numbers after GLYPH<6> are the standard deviations.
| | | Raw feature | PCA | Spectral no _ tuning | MBN no _ tuning | Spectral optimal | MBN optimal |
|----|---------------|-----------------------|-----------------------|------------------------|-----------------------|-----------------------|-----------------------|
| 1 | Isolet1 | 77.21% GLYPH<6> 0.92% | 56.74% GLYPH<6> 0.75% | 75.51% GLYPH<6> 0.58% | 77.89% GLYPH<6> 0.81% | 79.66% GLYPH<6> 0.58% | 77.97% GLYPH<6> 0.64% |
| 2 | Wine | 42.88% GLYPH<6> 0.00% | 40.92% GLYPH<6> 0.00% | 41.58% GLYPH<6> 0.00% | 55.49% GLYPH<6> 4.07% | 43.02% GLYPH<6> 0.00% | 76.96% GLYPH<6> 3.24% |
| 3 | New-Thyroid | 49.46% GLYPH<6> 0.00% | 49.46% GLYPH<6> 0.00% | 39.63% GLYPH<6> 0.00% | 68.80% GLYPH<6> 5.03% | 43.20% GLYPH<6> 0.00% | 75.78% GLYPH<6> 2.75% |
| 4 | Dermathology | 9.11% GLYPH<6> 0.11% | 59.50% GLYPH<6> 0.10% | 51.04% GLYPH<6> 0.15% | 82.40% GLYPH<6> 2.24% | 51.04% GLYPH<6> 0.15% | 92.79% GLYPH<6> 1.09% |
| 5 | COIL100 | 76.98% GLYPH<6> 0.27% | 69.64% GLYPH<6> 0.45% | 89.28% GLYPH<6> 0.59% | 81.66% GLYPH<6> 0.59% | 89.28% GLYPH<6> 0.59% | 90.37% GLYPH<6> 0.65% |
| 6 | MNIST(small) | 49.69% GLYPH<6> 0.14% | 27.86% GLYPH<6> 0.08% | 62.06% GLYPH<6> 0.04% | 77.12% GLYPH<6> 0.35% | 62.06% GLYPH<6> 0.04% | 78.50% GLYPH<6> 0.84% |
| 7 | MNIST(full) | 50.32% GLYPH<6> 0.12% | 28.05% GLYPH<6> 0.07% | Timeout | 91.36% GLYPH<6> 0.07% | Timeout | Timeout |
| 8 | UMIST | 65.36% GLYPH<6> 1.21% | 66.25% GLYPH<6> 1.10% | 81.83% GLYPH<6> 0.58% | 74.48% GLYPH<6> 1.91% | 84.66% GLYPH<6> 2.38% | 84.84% GLYPH<6> 1.98% |
| 9 | ORL | 75.55% GLYPH<6> 1.36% | 75.81% GLYPH<6> 1.17% | 80.21% GLYPH<6> 1.13% | 79.22% GLYPH<6> 0.84% | 83.62% GLYPH<6> 1.13% | 81.73% GLYPH<6> 1.19% |
| 10 | 20-Newsgroups | Timeout | 22.49% GLYPH<6> 1.41% | 27.03% GLYPH<6> 0.05% | 41.61% GLYPH<6> 0.05% | 27.03% GLYPH<6> 0.05% | 42.23% GLYPH<6> 0.46% |
manually to 500 and 2000 respectively, V = 200, and kept other parameters unchanged, i.e. a = 0 5 and : GLYPH<14> = 0 5. :
Experimental results in Fig. 12 show that the MBN with the small network reaches an accuracy curve of over 8% higher than LSA; the MBN with the large network reaches an accuracy curve of over 13% higher than LSA. The results indicate that enlarging the network size of MBN improves its generalization ability.
2-dimensional feature with MBN on the training set and generalizes well on the test set. The prediction time of MBN and the compressive MBN is 4857.45 and 1.10 seconds, respectively.
## 5.6. Empirical study on compressive MBN
This subsection studies the e GLYPH<11> ects of the compressive MBN on accelerating the prediction process of MBN.
We first studied the generalization ability of the compressive MBN on the 10,000 test images of MNIST, where the models of MBN and the compressive MBN are trained with the 60,000 training images of MNIST. The neural network in the compressive MBN contains 2 hidden layers with 2048 rectified linear units per layer. It projects the data space to the 2-dimensional feature space produced by MBN. The visualization results in Fig. 13 show that the compressive MBN produces an identical
We then studied the generalization ability of the compressive MBN in retrieving the Reuters newsware stories. The neural network here has the same structure with that on MNIST. It projects the data space to the 5-dimensional representation produced by MBN. The result in Fig. 14 shows that the compressive MBN produces an almost identical accuracy curve with MBN. The prediction time of MBN and the compressive MBN is 190,574.74 and 88.80 seconds, respectively.
## 6. Discussions
MBN is related to many methods in statistics and machine learning. Here we introduce its connection to histogram-based density estimators, bootstrap methods, clustering ensemble, vector quantization, product of experts, sparse coding, and unsupervised deep learning.
Table 4: Clustering accuracy on 10 data sets.
| | | Raw feature | PCA | Spectral no _ tuning | MBN no _ tuning | Spectral optimal | MBN optimal |
|----|---------------|-----------------------|-----------------------|------------------------|-----------------------|-----------------------|-----------------------|
| 1 | Isolet1 | 61.47% GLYPH<6> 1.93% | 38.62% GLYPH<6> 0.99% | 49.63% GLYPH<6> 1.88 | 61.13% GLYPH<6> 2.52% | 72.29% GLYPH<6> 1.92% | 62.50% GLYPH<6> 1.14% |
| 2 | Wine | 70.22% GLYPH<6> 0.00% | 78.09% GLYPH<6> 0.00% | 61.24% GLYPH<6> 0.00 | 81.91% GLYPH<6> 2.61% | 70.79% GLYPH<6> 0.00% | 93.20% GLYPH<6> 1.77% |
| 3 | New-Thyroid | 86.05% GLYPH<6> 0.00% | 86.05% GLYPH<6> 0.00% | 79.49% GLYPH<6> 0.94 | 93.02% GLYPH<6> 1.60% | 82.79% GLYPH<6> 0.00% | 94.51% GLYPH<6> 0.93% |
| 4 | Dermathology | 26.17% GLYPH<6> 0.28% | 61.67% GLYPH<6> 0.26% | 50.38% GLYPH<6> 0.35 | 82.81% GLYPH<6> 7.67% | 50.38% GLYPH<6> 0.35% | 96.07% GLYPH<6> 0.79% |
| 5 | COIL100 | 49.75% GLYPH<6> 1.31% | 43.42% GLYPH<6> 1.21% | 61.83% GLYPH<6> 2.27 | 57.38% GLYPH<6> 1.85% | 61.83% GLYPH<6> 2.27% | 68.29% GLYPH<6> 1.58% |
| 6 | MNIST(small) | 52.64% GLYPH<6> 0.14% | 34.49% GLYPH<6> 0.10% | 53.30% GLYPH<6> 0.01 | 82.36% GLYPH<6> 0.46% | 56.55% GLYPH<6> 0.20% | 87.06% GLYPH<6> 1.75% |
| 7 | MNIST(full) | 53.48% GLYPH<6> 0.11% | 35.14% GLYPH<6> 0.11% | Timeout | 96.64% GLYPH<6> 0.04% | Timeout | Timeout |
| 8 | UMIST | 43.20% GLYPH<6> 1.66% | 43.44% GLYPH<6> 1.92% | 70.99% GLYPH<6> 2.19 | 56.96% GLYPH<6> 3.73% | 70.99% GLYPH<6> 2.19% | 73.22% GLYPH<6> 4.02% |
| 9 | ORL | 54.37% GLYPH<6> 2.41% | 54.55% GLYPH<6> 2.81% | 57.33% GLYPH<6> 2.37 | 59.68% GLYPH<6> 1.58% | 68.85% GLYPH<6> 2.48% | 64.25% GLYPH<6> 3.50% |
| 10 | 20-Newsgroups | Timeout | 22.61% GLYPH<6> 1.29% | 28.52% GLYPH<6> 0.03 | 46.57% GLYPH<6> 1.10% | 28.52% GLYPH<6> 0.03% | 47.69% GLYPH<6> 0.92% |
Figure 9: Visualization of the 10,000 test images of MNIST produced by the MBNatlayer 8. For clarity, only 250 images per digit are drawn. See Appendix D for the visualizations produced by other layers.
## 6.1. Histogram-based density estimators
Histogram-based density estimation is a fundamental density estimation method, which estimates a probability function by accumulating the events that fall into the intervals of the function (Freedman et al., 2007) where the intervals may have an equivalent length or not. Our proposed density estimator is essentially such an approach. An interval in our approach is defined as the data space between two data points. The accumulated events in the interval are the shared centroids by the two data points.
## 6.2. Bootstrap methods
Bootstrap resampling (Efron, 1979; Efron & Tibshirani, 1993) has been applied successfully to machine learning. It resamples data with replacement . MBN does not adopt the standard bootstrap resampling. In fact, MBN uses random subsampling without replacement , also known as delete-d jackknife resampling in statistics (Efron & Tibshirani, 1993). The reason why we do not adopt the standard bootstrap resampling is that the resampling in MBN is used to build local coordinate systems, hence, if a data point is sampled multiple times, the duplicated data points are still viewed as a single coordinate axis. Moreover, it will cause the k -centroids clusterings built in different data spaces. However, MBN was motivated from and shares many common properties with the bootstrap methods, such as building each base clustering from a random sample of the input and de-correlating the base clusterings by the random sampling of features (Breiman, 2001) at each base clustering, hence, we adopted the phrase 'bootstrap' in MBN and clarify its usage here for preventing confusion.
## 6.3. Clustering ensemble
Clustering ensemble (Strehl & Ghosh, 2003; Dudoit & Fridlyand, 2003; Fern & Brodley, 2003; Fred & Jain, 2005; Zheng et al., 2010; Vega-Pons & Ruiz-Shulcloper, 2011) is a clustering technique that uses a consensus function to aggregate the clustering results of a set of mutually-independent base clusterings. Each base clustering is usually used to classify data to the ground-truth number of classes. An exceptional clustering ensemble method is (Fred & Jain, 2005), in which each base k -means clustering produces a subclass partition by assigning the parameter k to a random value that is slightly larger than the ground-truth number of classes.
Each layer of MBN can be regarded as a clustering ensemble. However, its purpose is to estimate the density of data instead of producing an aggregated clustering result. Moreover, MBN has clear theoretical and geometric explanations. We have also found that setting k to a random value does not work for MBN, particularly when k is also small.
## 6.4. Vector quantization
Each layer of MBN can be regarded as a vector quantizer to the input data space. The codebook produced by MBN is exponentially smaller than that produced by a traditional k -means clustering, when they have the same level of quantization errors. A similar idea, named product quantization , has been explored in (Jegou et al., 2011). If product quantization uses random
Figure 10: E GLYPH<11> ect of decay factor GLYPH<14> on 9 benchmark data sets.
Figure 11: E GLYPH<11> ect of parameters a and V on the subset of MNIST.
sampling of features instead of a fixed non-overlapping partition of features in (Jegou et al., 2011), and uses random sampling of data points to train each k -means clustering instead of the expectation-maximization optimization, then product quantization equals to a single layer of MBN. We are also aware of the hierarchical product quantizer (Wichert, 2012), which builds multiple sets of sub-quantizers (where the name 'hierarchy' comes) on the original feature. Our MBN builds each layer of sub-quantizers on the output sparse feature of its lower layer which is fundamentally di GLYPH<11> erent from the hierarchical product quantization method.
Animportant di GLYPH<11> erence between existing vector quantization methods (e.g. binarized neural networks and binary hashing)
and MBN is that the former are designed for reducing the computational and storage complexities, while MBN is not developed for this purpose. MBN aims at providing a simple way to overcome the di GLYPH<14> culty of density estimation in a continuous space. Hence, it generates larger codebooks than common vector quantization methods and does not transform the sparse codes to compact binary codes.
## 6.5. Product of experts
PoE aims to combine multiple individual models by multiplying them, where the individual models have to be a bit more complicated and each contains one or more hidden variables (Hinton, 2002). Its general probability framework is:
$$p ( { \mathbf x } ) = \frac { \prod _ { v = 1 } ^ { V } g _ { v } ( { \mathbf x } ) } { \sum _ { { \mathbf x } ^ { \prime } } \prod _ { v = 1 } ^ { V } g _ { v } ( { \mathbf x } ^ { \prime } ) }$$
where x 0 indexes all possible vectors in the data space, and gv is called an expert . A major merit of PoE is that, a function that can be fully expressed by a mixture of experts with N experts (i.e. mixtures), such as Gaussian mixture model or k -means clustering, can be expressed compactly by a PoE with only log 2 N experts at the minimum, with the expense of the optimization di GLYPH<14> culty of the partition function P Q x 0 V v = 1 gv ( x 0 ) which consists of an exponentially large number of components.
The connection between MBN and PoE is as follows:
Theorem 2. Each layer of MBN is a PoE that does not need to optimize the partition function.
Accuracy
0.4
0.35
0.3
0.25
0.2
0.15
LSA
0.1
DNN approximation of LSA
0.4
0.35
0.3
0.25
0.2
0.15
MBN at layer 1
0.1
Compressive MBN at layer 1
511 1023
0.05
0.05
Figure 12: Average accuracy curves of retrieved documents on the test set (82 topics) of the Reuters newswire stories. 0 0
1
3
7
15
31
63
127
255
511 1023
1
3
7
15
31
63
127
255
Number of retrieved documents
511 1023
Number of retrieved documents
Compressive MBN at layer 4
1
3
7
15
31
63
127
255
511 1023
Number of retrieved documents
Visualizing training data (With random reconstruction) Generalization on test data (With random reconstruction) C D Figure 13: Visualizations of the training data (left) and test data (right) of MNIST produced by the compressive MBN.
Proof. See Appendix B for the proof.
## 6.6. Sparse coding
Given a learned dictionary W , sparse coding typically aims to solve min h i P n i = 1 k x i GLYPH<0> Wh i k 2 2 + GLYPH<21> k h i k 1 1 , where k GLYPH<1> k q represents ' q -norm, h i is the sparse code of the data point x i , and GLYPH<21> is a hyperparameter controlling the sparsity of h i . Each column of W is called a basis vector.
To understand the connection between MBN and sparse coding, we may view GLYPH<21> as a hyperparameter that controls the number of clusterings. Specifically, if we set GLYPH<21> = 0, it is likely that h i contains only one nonzero element. Intuitively, we can understand it as that we use only one clustering to learn a sparse code. A good value of GLYPH<21> can make a small part of the elements of h i nonzero. This choice approximates to the method of partitioning the dictionary to several (probably overlapped) subsets and then grouping the basis vectors in each subset to a base clustering. Motivated by the above intuitive analysis, we introduce the connection between sparse coding and MBN formally as follows:
Theorem 3. The ' 1 -norm sparse coding is a convex relaxation of the building block of MBN when given the same dictionary.
Figure 14: Comparison of the generalization ability of MBN and the compressive MBN on retrieving the Reuters newsware stories.
## 6.7. Unsupervised deep learning
Learning abstract representations by deep networks is a recent trend. From the geometric point of view, the abstract representations are produced by reducing larger and larger local variations of data from bottom up in the framework of trees that are built on data spaces. 5 For example, convolutional neural network merges child nodes by pooling. Hierarchical Dirichlet process (Teh et al., 2005) builds trees whose father nodes generate child nodes according to a prior distribution. DBN (Hinton & Salakhutdinov, 2006) merges child nodes by reducing the number of the nonlinear units gradually from bottom up. Subspace tree (Wichert & Moreira, 2015) merges nodes by reducing the dimensions of the subspaces gradually. PCANet (Chan et al., 2015) merges nodes by reducing the output dimensions of the local PCA associated with its patches gradually. Our MBN merges nodes by reducing the number of the randomly sampled centroids gradually.
A fundamental di GLYPH<11> erence between the methods is how to build e GLYPH<11> ective local coordinate systems in each layer. To our knowledge, MBN is a simple method and needs little assumption and prior knowledge. It is only more complicated than
5 Some recent unsupervised deep models for data augmentation are out of the discussion, hence we omit them here.
Accuracy
random projection which is, to our knowledge, not an e GLYPH<11> ective method for unsupervised deep learning to date. Moreover, MBN is the only method working in discrete feature spaces, and it works well.
## 7. Conclusions
In this paper, we have proposed multilayer bootstrap network for nonlinear dimensionality reduction. MBN has a novel network structure that each expert is a k -centroids clustering whose centroids are randomly sampled data points with randomly sampled features; the network is gradually narrowed from bottom up.
MBN is composed of two novel components: (i) each layer of MBN is a nonparametric density estimator by random resampling. It estimates the density of data correctly without any model assumption. It is exponentially more powerful than a single k -centroids clustering. Its estimation error is proven to be small and controllable. (ii) The network is a deep ensemble model. It essentially reduces the nonlinear variations of data by building a vast number of hierarchical trees on the data space. It can be trained as many layers as needed with both large-scale and small-scale data.
MBN performs robustly with a wide range of parameter settings. Its time and storage complexities scale linearly with the size of training data. It supports parallel computing naturally. Empirical results demonstrate its e GLYPH<14> ciency at the training stage and its e GLYPH<11> ectiveness in density estimation, data visualization, clustering, and document retrieval. We have also demonstrated that the high computational complexity of MBN at the test stage can be eliminated by the compressive MBN-a framework of unsupervised model compression based on neural networks.
A problem left is on the selection of parameter GLYPH<14> which controls the network structure. Although the performance of MBN with GLYPH<14> = 0 5 is : good, there is still a large performance gap between GLYPH<14> = 0 5 and the best : GLYPH<14> . Hence, how to select GLYPH<14> automatically is an important problem.
## Acknowledgement
The author thanks Prof. DeLiang Wang for providing his computing resources. The author also thanks the action editor and reviewers for their comments which improved the quality of the paper. This work was supported by the National Natural Science Foundation of China under Grant 61671381.
## References
## References
Belkin, M., & Niyogi, P. (2003). Laplacian eigenmaps for dimensionality reduction and data representation. Neural Comput. , 15 , 1373-1396.
- Bishop, C. M. et al. (2006). Pattern Recognition and Machine Learning . Springer New York:.
- Blei, D. M., Ng, A. Y., & Jordan, M. I. (2003). Latent dirichlet allocation. J. Mach. Learn. Res. , 3 , 993-1022.
- Boyd, S. P., & Vandenberghe, L. (2004). Convex Optimization . Cambridge University Press.
Breiman, L. (2001). Random forests. Machine Learning , 45 , 5-32.
- Canu, S., Grandvalet, Y., Guigue, V., & Rakotomamonjy, A. (2005). SVM and kernel methods Matlab toolbox. http://asi.insa-rouen.fr/ enseignants/~arakoto/toolbox/index.html .
- Chan, T.-H., Jia, K., Gao, S., Lu, J., Zeng, Z., & Ma, Y. (2015). Pcanet: A simple deep learning baseline for image classification? IEEE Transactions on Image Processing , 24 , 5017-5032.
- Chang, H., & Yeung, D.-Y. (2008). Robust path-based spectral clustering. Pattern Recognition , 41 , 191-203.
- Deerwester, S. C., Dumais, S. T., Landauer, T. K., Furnas, G. W., & Harshman, R. A. (1990). Indexing by latent semantic analysis. Journal of the American Society for Information Science , 41 , 391-407.
- Dietterich, T. G. (2000). Ensemble methods in machine learning. In Multiple Classifier Systems (pp. 1-15). Cagliari, Italy: Springer.
- Dudoit, S., & Fridlyand, J. (2003). Bagging to improve the accuracy of a clustering procedure. Bioinformatics , 19 , 1090-1099.
- Efron, B. (1979). Bootstrap methods: another look at the jackknife. Annals of Statistics , 7 , 1-26.
- Efron, B., & Tibshirani, R. (1993). An Introduction to the Bootstrap . CRC press.
- Fern, X. Z., & Brodley, C. E. (2003). Random projection for high dimensional data clustering: A cluster ensemble approach. In Proceedings of the 20th International Conference on Machine Learning (pp. 186-193). volume 3.
- Fred, A. L., & Jain, A. K. (2005). Combining multiple clusterings using evidence accumulation. IEEE Transactions on Pattern Analysis and Machine Intelligence , 27 , 835-850.
- Freedman, D., Pisani, R., & Purves, R. (2007). Statistics (4th edition).
- Freund, Y., & Schapire, R. E. (1995). A decision-theoretic generalization of on-line learning and an application to boosting. In Proceedings of the 2nd European Conference on Computational Learning Theory (pp. 23-37). Barcelona, Spain.
- Friedman, J., Hastie, T., Tibshirani, R. et al. (2000). Additive logistic regression: A statistical view of boosting (with discussion and a rejoinder by the authors). Annals of Statistics , 28 , 337-407.
- Golub, T. R., Slonim, D. K., Tamayo, P., Huard, C., Gaasenbeek, M., Mesirov, J. P., Coller, H., Loh, M. L., Downing, J. R., Caligiuri, M. A. et al. (1999). Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science , 286 , 531-537.
- Hastie, T., Tibshirani, R., & Friedman, J. (2009). The Elements of Statistical Learning . Springer.
- He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of 2016 IEEE International Conference on Computer Vision and Pattern Recognition (pp. 770-778). Las Vegas, NV.
- He, X., & Niyogi, X. (2004). Locality preserving projections. In Advances in Neural Information Processing Systems 17 (pp. 153-160). Vancouver, British Columbia, Canada volume 16.
- Hinton, G., Deng, L., Yu, D., Dahl, G. E., Mohamed, A.-r., Jaitly, N., Senior, A., Vanhoucke, V., Nguyen, P., Sainath, T. N. et al. (2012). Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups. IEEE Signal Processing Magazine , 29 , 82-97.
- Hinton, G. E. (2002). Training products of experts by minimizing contrastive divergence. Neural Computation , 14 , 1771-1800.
- Hinton, G. E., & Salakhutdinov, R. R. (2006). Reducing the dimensionality of data with neural networks. Science , 313 , 504-507.
- Jain, A. K., & Law, M. H. (2005). Data clustering: A user's dilemma. Lecture Notes in Computer Science , 3776 , 1-10.
- Jegou, H., Douze, M., & Schmid, C. (2011). Product quantization for nearest neighbor search. IEEE transactions on pattern analysis and machine intelligence , 33 , 117-128.
- Lanckriet, G. R. G., Cristianini, N., Bartlett, P., Ghaoui, L. E., & Jordan, M. I. (2004). Learning the kernel matrix with semidefinite programming. J. Mach. Learn. Res. , 5 , 27-72.
- Lecun, Y., Cortes, C., & Burges, C. J. C. (2004). THE MNIST DATABASE of handwritten digits. http://yann.lecun.com/exdb/mnist/index. html .
- Lewis, D. D., Yang, Y., Rose, T. G., & Li, F. (2004). RCV1: a new benchmark collection for text categorization research. Journal of Machine Learning Research , 5 , 361-397.
- Van der Maaten, L., & Hinton, G. E. (2008). Visualizing data using t-sne. Journal of Machine Learning Research , 9 , 2579-2605.
- McCallum, A. (1998). Rainbow. http://www.cs.cmu.edu/~mccallum/
## bow/rainbow .
Ng, A. Y., Jordan, M. I., & Weiss, Y. (2002). On spectral clustering: Analysis and an algorithm. In Advances in Neural Information Processing Systems 14 (pp. 849-856). Vancouver, British Columbia, Canada.
Nie, F., Zeng, Z., Tsang, I. W., Xu, D., & Zhang, C. (2011). Spectral embedded clustering: A framework for in-sample and out-of-sample spectral clustering. IEEE Transactions on Neural Networks , 22 , 1796-1808.
Pearson, K. (1901). On lines and planes of closest fit to systems of points in space. Philosophical Magazine , 2 , 559-572.
Roweis, S. T. (1998). EM algorithms for PCA and SPCA. In Advances in Neural Information Processing Systems 10 (pp. 626-632). Denver, CO.
Roweis, S. T., & Saul, L. K. (2000). Nonlinear dimensionality reduction by locally linear embedding. Science , 290 , 2323-2326.
- Schapire, R. E. (1990). The strength of weak learnability. Machine Learning , 5 , 197-227.
- Schmidhuber, J. (2015). Deep learning in neural networks: An overview. Neural Networks , 61 , 85-117.
- Schölkopf, B., Smola, A., & Müller, K.-R. (1998). Nonlinear component analysis as a kernel eigenvalue problem. Neural Computation , 10 , 1299-1319.
Shi, J., & Malik, J. (2000). Normalized cuts and image segmentation. IEEE Trans. Pattern Anal. Machine Intell. , 22 , 888-905.
Strehl, A., & Ghosh, J. (2003). Cluster ensembles-a knowledge reuse framework for combining multiple partitions. Journal of Machine Learning Research , 3 , 583-617.
Tao, D., Tang, X., Li, X., & Wu, X. (2006). Asymmetric bagging and random subspace for support vector machines-based relevance feedback in image retrieval. IEEE Transactions on Pattern Analysis and Machine Intelligence , 28 , 1088-1099.
Teh, Y. W., Jordan, M. I., Beal, M. J., & Blei, D. M. (2005). Sharing clusters among related groups: Hierarchical dirichlet processes. In Proc. Adv. Neural Inform. Process. Syst. (pp. 1385-1392).
Tenenbaum, J. B., De Silva, V., & Langford, J. C. (2000). A global geometric framework for nonlinear dimensionality reduction. Science , 290 , 23192323.
Vega-Pons, S., & Ruiz-Shulcloper, J. (2011). A survey of clustering ensemble algorithms. International Journal of Pattern Recognition and Artificial Intelligence , 25 , 337-372.
Wang, D. L., & Chen, J. (2017). Supervised speech separation based on deep learning: An overview. arXiv preprint arXiv:1708.07524 , .
Wang, Q., Qin, Z., Nie, F., & Yuan, Y. (2017). Convolutional 2d lda for nonlinear dimensionality reduction. In Proceedings of the 26th International Joint Conference on Artificial Intelligence (pp. 2929-2935).
Wichert, A. (2012). Product quantization for vector retrieval with no error. In ICEIS (1) (pp. 87-92).
Wichert, A., & Moreira, C. (2015). On projection based operators in lp space for exact similarity search. Fundamenta Informaticae , 136 , 461-474.
Wikipedia (2017). Frequentist probability. https://en.wikipedia.org/ wiki/Frequentist\_probability .
Xing, E. P., Jordan, M. I., Russell, S., & Ng, A. (2002). Distance metric learning with application to clustering with side-information. In Advances in Nueral Information Processing Systems 15 (pp. 505-512). Vancouver, British Columbia, Canada.
Yan, S., Xu, D., Zhang, B., Zhang, H.-J., Yang, Q., & Lin, S. (2007). Graph embedding and extensions: a general framework for dimensionality reduction. IEEE Transactions on Pattern Analysis and Machine Intelligence , 29 , 40-51.
Zahn, C. T. (1971). Graph-theoretical methods for detecting and describing gestalt clusters. IEEE Transactions on Computers , 100 , 68-86.
Zheng, L., Li, T., & Ding, C. (2010). Hierarchical ensemble clustering. In Prof. IEEE Int. Conf. Data Min. (pp. 1199-1204).
Zhou, Z.-H., & Feng, J. (2017). Deep forest: Towards an alternative to deep neural networks. arXiv preprint arXiv:1702.08835 , (pp. 1-10).
## Appendix A. Complexity analysis
Theorem 4. The computational and storage complexities of MBN are:
$$O _ { t i m e } & = O \left ( d s k V n \right ) & \text{(A.1)}$$
$$O _ { s t o r a g e } = O ( ( d s + V ) n + k V + k d s )$$
respectively at the bottom layer, and are:
$$\stackrel { s } { s } & & & O _ { t i m e } & = O \left ( k V ^ { 2 } n \right ) & & & ( \text{A.3} ) \\ \vdash & & O _ { s t o r a g e } & = O ( 2 V ( n + k ) ) & & & ( \text{A.4} )$$
respectively at other layers, where d is the dimension of the original feature, s is the sparsity of the data (i.e., the ratio of the non-zero elements over all elements).
If MBN is not used for prediction, then MBN needs not to be saved, which further reduces the storage complexity to O (( ds + V n ) ) at the bottom layer and O (2 Vn at other layers. )
Proof. Due to the length limitation of the paper, we omit the simple proof.
Fortunately, as shown in Fig A.15, the empirical time complexity grows with O V ( ) instead of O V ( 2 ). The only explanation is that the input data is sparse. Specifically, the multiplication of two sparse matrices only considers the element-wise multiplication of two elements that are both nonzero, as a result, when the input data is sparse, one factor V is o GLYPH<11> set by the sparsity factor s . The empirical time and storage complexities with other parameters are consistent with our theoretical analysis. We omit the results here.
Note that, our default parameter k 1 = 0 5 : n makes the time complexity scale squarely with the size of the data set n . To reduce the computational cost, we usually set k 1 manually to a small value irrelevant to n as what we have done for visualizing the full MNSIT and retrieving the RCV1 documents in Section 5.
## Appendix B. Proof of Theorem 2
The optimization objective of k -means clustering can be written as:
$$& \max _ { \ } g ( x | W, h ) \\ c & \quad = \max _ { \ } \exp \left ( - \frac { 1 } { 2 \sigma ^ { 2 } } \| \mathbf x - W ^ { T } \mathbf h \| _ { 2 } ^ { 2 } \right ) \\ h & \quad \text{subject to } h \, \text{is a one-hot code } \quad \text{(B.1)}$$
where W = [ w 1 ; : : : ; w k ] is the weight matrix whose columns are centroids, h is a vector of hidden variables, and the covariance matrix of the clustering is GLYPH<27> 2 I with I being the identity matrix and GLYPH<27> ! 0 (Bishop et al., 2006).
Substituting g ( x W h j ; ) = exp GLYPH<16> GLYPH<0>k x GLYPH<0> W h T k 2 2 = 2 GLYPH<27> 2 GLYPH<17> to Eq. (12) gets:
$$\vee \vdash ^ { \prime } \varepsilon \times \begin{matrix} p ( x | \{ W _ { v }, \mathbf h _ { v } \} _ { v = 1 } ^ { V } ) \\ & = \frac { \prod _ { v = 1 } ^ { V } \exp \left ( - \| \mathbf x - W _ { v } ^ { T } \mathbf h _ { v } \| _ { 2 } ^ { 2 } / 2 \sigma ^ { 2 } \right ) } { \sum _ { x ^ { \prime } } \prod _ { v = 1 } ^ { V } \exp \left ( - \| \mathbf x ^ { \prime } - W _ { v } ^ { T } \mathbf h _ { v } \| _ { 2 } ^ { 2 } / 2 \sigma ^ { 2 } \right ) } \\ & = \frac { \exp \left ( - \sum _ { v = 1 } ^ { V } \| \mathbf x - W _ { v } ^ { T } \mathbf h _ { v } \| _ { 2 } ^ { 2 } / 2 \sigma ^ { 2 } \right ) } { \sum _ { x ^ { \prime } } \exp \left ( - \sum _ { v = 1 } ^ { V } \| \mathbf x ^ { \prime } - W _ { v } ^ { T } \mathbf h _ { v } \| _ { 2 } ^ { 2 } / 2 \sigma ^ { 2 } \right ) } \end{matrix} \quad ( B. 2 )$$
Running time of training each layer
Figure A.15: Time complexity of MBN with respect to parameter V .
where we assume that all experts have the same covariance matrix GLYPH<27> 2 I . Maximizing the likelihood of Eq. (B.2) is equivalent to the following problem:
Because GLYPH<27> ! 0, problem (B.3) can be rewritten as:
$$\mathcal { J } _ { \text{PoE} } = \sum _ { \{ W _ { \nu }, h _ { \nu } \} _ { V = 1 } ^ { V } } - \log p ( x | \{ W _ { \nu }, h _ { \nu } \} _ { V = 1 } ^ { V } ) \\ = \sum _ { \{ W _ { \nu }, h _ { \nu } \} _ { V = 1 } ^ { V } } \frac { 1 } { 2 \sigma ^ { 2 } } \sum _ { \nu = 1 } ^ { V } \| x - W _ { \nu } ^ { T } h _ { \nu } \| _ { 2 } ^ { 2 } \Big { \text{ Becau} } \\ + \log \sum _ { x ^ { \prime } } \exp \left ( - \frac { 1 } { 2 \sigma ^ { 2 } } \sum _ { \nu = 1 } ^ { V } \| x ^ { \prime } - W _ { \nu } ^ { T } h _ { \nu } \| _ { 2 } ^ { 2 } \Big { \text{ Becau} } \\ \infty \sum _ { \{ W _ { \nu }, h _ { \nu } \} _ { V = 1 } ^ { V } \| _ { V = 1 } ^ { 2 } } \\ + \log \left ( \sum _ { x ^ { \prime } } \exp \left ( - \frac { 1 } { 2 \sigma ^ { 2 } } \sum _ { \nu = 1 } ^ { V } \| x ^ { \prime } - W _ { \nu } ^ { T } h _ { \nu } \| _ { 2 } ^ { 2 } \right ) \right ) ^ { 2 \sigma ^ { 2 } } \quad \text{which} \quad \text{It is} \\ \text{subject to} \quad h _ { \nu } \text{ is a one-hot code} \quad \text{(B.3)} \quad \text{an} \text{en}.$$
$$- \quad \text{Because $\sigma\to 0$, problem (B.3) can be rewritten as:} \\ \Gamma _ { \text{E} } \in \in \min _ { \{ W, h _ { v } \} _ { v = 1 } ^ { V } } \sum _ { v = 1 } ^ { V } \| \mathbf x - W _ { v } ^ { T } \mathbf h _ { v } \| _ { 2 } ^ { 2 } \\ \quad \quad + \log \left ( \max \left ( \exp \left ( - \frac { 1 } { 2 \sigma ^ { 2 } } \sum _ { v = 1 } ^ { V } \| \mathbf x ^ { \prime } - W _ { v } ^ { T } \mathbf h _ { v } \| _ { 2 } ^ { 2 } \right ) \right ) ^ { 2 \sigma ^ { 2 } } \\ \quad = \min _ { \{ W, h _ { v } \} _ { v = 1 } ^ { V } \| \mathbf x - W _ { v } ^ { T } \mathbf h _ { v } \| _ { 2 } ^ { 2 } - \min _ { v = 1 } ^ { V } \| \mathbf x ^ { \prime } \\ \quad \quad - W _ { v } ^ { T } \mathbf h _ { v } \| _ { 2 } ^ { 2 } \\ \quad \quad \text{subject to} \ \ \mathbf h _ { v } \text{ is a one-hot code} \quad \quad ( B. 4 ) \\ \quad \text{Because $\mathbf x^{\prime}$ can be any possible vector in the data space, we} \,. \quad \,. \quad \, - v \,. \,. \,. \,. \,. \,. \,. \,. \,. \,.$$
Because x 0 can be any possible vector in the data space, we have min x 0 P V v = 1 k x 0 GLYPH<0> W h T v v k 2 2 = 0. Eventually, the optimization objective of PoE is:
$$\mathcal { J } _ { \text{PoE} } & \, \in \min _ { \{ W _ { v }, h _ { v } \} _ { v = 1 } ^ { V } } \sum _ { v = 1 } ^ { V } \| \mathbf x - W _ { v } ^ { T } \mathbf h _ { v } \| _ { 2 } ^ { 2 } \\ \text{subject to} \quad & \, \mathbf h _ { v } \, \text{is a one-hot code} \quad & \, ( B. 5 )$$
which is irrelevant to the partition function.
It is clear that problem (B.5) is the optimization objective of an ensemble of k -means clusterings. When we assign Wv by
random sampling, then problem (B.5) becomes:
$$\mathcal { J } _ { \text{PoE} } & \in \min _ { \{ h _ { v } \} _ { v = 1 } ^ { V } } \sum _ { v = 1 } ^ { V } \| x - W _ { v } ^ { T } h _ { v } \| _ { 2 } ^ { 2 } \\ \text{subject to} \quad & h _ { v } \text{ is a one-hot code} \quad & ( B. 6 )$$
which is the building block of MBN. Theorem 2 is proved.
## Appendix C. Proof of Theorem 3
Each layer of MBN maximizes the likelihood of the following equation:
$$p ( { \mathbf x } ) & = \prod _ { v = 1 } ^ { V } g _ { v } ( { \mathbf x } ) & ( C. 1 ) & \quad \text{mon} \, \ell \\ & \quad \max _ { v \cdots }.$$
where gv ( x ) is a k -means clustering with the squared error as the similarity metric:
$$g _ { v } ( \mathbf x ) & = \mathcal { M N } \left ( \mathbf x ; \mathbf W _ { v } \mathbf h _ { v }, \sigma ^ { 2 } \mathbf l \right ) & ( C. 2 ) \quad \text{regula} \\ \text{subject to} \quad \mathbf h _ { v } \text{ is a one-hot code}$$
where MN denotes the multivariate normal distribution, W v = [ w v ; 1 ; : : : ; w v k ; ] is the weight matrix whose columns are centroids, I is the identity matrix, and GLYPH<27> ! 0.
Given a data set f x i g n i = 1 . We assume that f W v g V v = 1 are fixed, and take the negative logarithm of Eq. (C.1):
$$\min _ { \{ \hbar { v }, i \} _ { i = 1 } ^ { n } / \ V _ { v = 1 } } \sum _ { v = 1 } ^ { V } \sum _ { i = 1 } ^ { n } \| \mathbf x _ { i } - W _ { v } \mathbf h _ { v, i } \| _ { 2 } ^ { 2 }, & & ( C. 3 ) \\ \text{subject to } \mathbf h _ { v, i } \text{ is a one-hot code.}$$
If we denote W = [ W 1 ; : : : ; W V ] and further complement the head and tail of h v i ; with multiple zeros, denoted as h 0 v i ; , such that Wh 0 v i ; = W h v v i ; , we can rewrite Eq. (C.3) to the following equivalent problem:
$$\min _ { \{ \mathbf h _ { v, i } \} _ { i = 1 } ^ { n } \} _ { v = 1 } ^ { V } } \sum _ { v = 1 } ^ { n } \| \mathbf x _ { i } - \mathbf W \mathbf h _ { v, i } ^ { \prime } \| _ { 2 } ^ { 2 }, & & ( C. 4 ) \\ \text{subject to } \mathbf h _ { v, i } \text{ is a one-hot code.}$$
It is an integer optimization problem that has an integer matrix variable H 0 v = [ h 0 v ; 1 ; : : : ; h 0 v n ; ]. Suppose there are totally j H 0 v j possible solutions of H 0 v , denoted as H 0 v ; 1 ; : : : ; H 0 v ; j H 0 v j , we first relax Eq. (C.4) to a convex optimization problem by constructing a convex hull (Boyd & Vandenberghe, 2004) on H 0 v :
$$\bigoplus _ { \substack { \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \float } \, \\ \min _ { \left \{ \langle \mu _ { v, k } | ^ { \mathcal { H } ^ { \prime } _ { v } | } _ { k = 1 } \rangle _ { v = 1 } ^ { V } \sum _ { i = 1 } ^ { n } \left \| \mathbf x _ { i } - W \left ( \sum _ { k = 1 } ^ { | \mathcal { H } ^ { \prime } _ { v } | } \mu _ { v, k } | ^ { \prime } _ { v, k, i } \right ) \right \| _ { 2 } ^ { 2 } \quad ( C. 5 ) \\ \text{subject to } 0 \leq \mu _ { v, k } \leq 1, \sum _ { k = 1 } ^ { | \mathcal { H } ^ { \prime } _ { v } | } \mu _ { v, k } = 1, \quad \forall v = 1, \dots, V. \\ \text{$\int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int\int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \int \end{array}$$
Because Eq. (C.5) is a convex optimization problem, according to Jensen's inequality, the following problem learns a lower bound of Eq. (C.5):
$$\min _ { \left \{ \{ \mu _ { v, k } \} _ { k = 1 } ^ { i \mathcal { H } ^ { \prime } _ { i } \} _ { v = 1 } ^ { V } } } \ V \sum _ { i = 1 } ^ { n } \left \| { \mathbf x } _ { i } - { \mathbf W } { \mathbf y } ^ { \prime \prime } _ { i } \right \| _ { 2 } ^ { 2 } & & ( C. 6 )$$
$$\text{subject to } \ 0 \leq \mu _ { v, k } \leq 1, \sum _ { k = 1 } ^ { | \mathcal { H } ^ { \prime } _ { v | } } \mu _ { v, k } = 1, \quad \forall v = 1, \dots, V$$
where h 00 i = 1 V P V v = 1 P j H 0 v j k = 1 GLYPH<22> v k ; h 0 v k i ; ; with GLYPH<22> v k ; as a variable.
Recalling the definition of sparse coding given a fixed dictionary W , we observe that Eq. (C.6) is a special form of sparse coding with more strict constraints on the format of sparsity.
Therefore, given the same dictionary W , each layer of MBN is a distributed sparse coding that is lower bounded by the common ' 1 -norm sparse coding. When we discard the expectationmaximization optimization of each k -means clustering (i.e., dictionary learning) but only preserve the default initialization method - random sampling, Eq. (C.1) becomes the building block of MBN. Given the same dictionary, the ' 1 -normregularized sparse coding is a convex relaxation of the building block of MBN. Theorem 3 is proved.
## Appendix D. Visualizations produced by intermediate layers
Figure D.16: Visualizations of MNIST produced by MBN at di GLYPH<11> erent layers. This figure is a supplement to Fig. 9.
| null |
[
"Xiao-Lei Zhang"
] |
2014-08-05T02:13:50+00:00
|
2018-03-06T15:59:10+00:00
|
[
"cs.LG",
"cs.NE",
"stat.ML"
] |
Multilayer bootstrap networks
|
Multilayer bootstrap network builds a gradually narrowed multilayer nonlinear
network from bottom up for unsupervised nonlinear dimensionality reduction.
Each layer of the network is a nonparametric density estimator. It consists of
a group of k-centroids clusterings. Each clustering randomly selects data
points with randomly selected features as its centroids, and learns a one-hot
encoder by one-nearest-neighbor optimization. Geometrically, the nonparametric
density estimator at each layer projects the input data space to a
uniformly-distributed discrete feature space, where the similarity of two data
points in the discrete feature space is measured by the number of the nearest
centroids they share in common. The multilayer network gradually reduces the
nonlinear variations of data from bottom up by building a vast number of
hierarchical trees implicitly on the original data space. Theoretically, the
estimation error caused by the nonparametric density estimator is proportional
to the correlation between the clusterings, both of which are reduced by the
randomization steps.
|
1408.0849v1
|
## Convergence of loop erased random walks on a planar graph to a chordal SLE(2) curve
## Hiroyuki Suzuki
## Abstract
In this paper we consider the 'natural' random walk on a planar graph and scale it by a small positive number δ . Given a simply connected domain D and its two boundary points a and b , we start the scaled walk at a vertex of the graph nearby a and condition it on its exiting D through a vertex nearby b , and prove that the loop erasure of the conditioned walk converges, as δ ↓ 0, to the chordal SLE 2 that connects a and b in D , provided that an invariance principle is valid for both the random walk and the dual walk of it. Our result is an extension of one due to Dapeng Zhan [12] where the problem is considered on the square lattice. A convergence to the radial SLE 2 has been obtained by Lawler, Schramm and Werner [3] for the square and triangular lattices and by Yadin and Yehudayoff [10] for a wide class of planar graphs. Our proof, though an adaptation of that of [3] and [10], involves some new ingredients that arise from two sources: one for dealing with a martingale observable that is different from that used in [3] and [10] and the other for estimating the harmonic measures of the random walk started at a boundary point of a domain.
## 1 Introduction
The Schramm-Loewner evolutions driven by Brownian motion √ κB t ( ) of variance κ , abbreviated as SLE , introduced by Oded Schramm [6], have been studied from various κ points of view and are now recognized to well describe the scaling limits of certain lattice models of both physical and mathematical interest. Lawler, Schramm and Werner [3] have proved that the scaling limit of a loop erased random walk (or loop erasure (for the definition, see p.9) of random walk, abbreviated as LERW) on either of the square and triangular lattices is the radial SLE . 2 Dapeng Zhan [12] have studied LERW's on the square lattice but in a multiply connected domain and derived the convergence of them. In the case of a simply connected domain in particular, he has proved the convergence to the chordal SLE . Yadin and Yehudayoff [10] extend the result of [3], the convergence of 2 LERW to a radial SLE to that for the natural random walks on planar graphs under a natural setting of the problem. In this paper we consider the LERW in a similar setting to [10] and show that LERW conditioned to connecting two boundary points in a simply connected domain converges to a chordal SLE 2 curve.
Key words and phrases, Loop erased random walk, Schramm-Loewner evolution.
Here we state our result in an informal way by using the terminology familiar in the theory of SLE of which we shall give a brief exposition in the next section. Let V be the set of vertices of a planar graph on which a random walk (of discrete time) is defined and supposed to satisfy invariance principle in the sense that the linear interpolation of its space-scaled trajectory converges to that of Brownian motion (in a topology where two curves are identified if they agree by some change of time parameter). For each δ > 0 we make the scale change of the space by δ : V δ = { δv : v ∈ V } , the set of scaled lattice points and accordingly we make the δ -scaling of our random walk so that it moves on V δ . Given a simply connected bounded domain D and two distinct boundary points a and b of it, let γ δ denote the loop erasure of the random walk scaled by δ that starts a vertex a δ of V δ nearby a and is conditioned to exit D ∩ V δ through a vertex b δ nearby b so that γ δ is a random self-avoiding path on D ∩ V δ connecting a δ and b δ , which may be regarded as a 'path' in the planar graph. We prove that the polygonal curve given by linearly interpolating γ δ converges to the chordal SLE 2 curve connecting a and b in D under a certain natural assumption on D , the pair a, b , the planar graph and the random walk (Theorem 5.6).
For obtaining the result as stated above we first prove the convergence of the driving function of the loop erasure (Theorem 5.1). The proof is made in a way similar to [3], [10] and [7]. In [7] the harmonic explorer, an evolution of a self avoiding random curve, is introduced and proved to converge to a chordal SLE 4 curve. For the proof a suitably chosen martingale associated with the evolving random curve, called martingale observable, plays a dominant role. Not as in [7] we take the martingale observable given by the ratio of harmonic measures of a (random) point relative to two points, the starting site of the walk and a suitably chosen site in a random domain defined by the loop erasure. This martingale is suggested in [3] as a suitable candidate of a martingale observable but we need to normalize it in an appropriate way; moreover we must change the normalization as the loop erasure grows. We apply the approximation result on the harmonic measure (Poisson kernel) proved in [10]. To this end we need a delicate probability estimate, since our random walk starts at a boundary point and we must deal with the conditional law given that it exits D ∩ V δ through another boundary point.
We deduce the convergence of the loop erasure in a uniform topology from that of the driving function under the hypothesis that not only the random walk but also the dual walk of it satisfy the invariance principle (Theorem 5.6). For the deduction we prove Proposition 4.1 asserting that the law of the time reversal of loop erasure of a walk agrees with the law of loop erasure of the time reversal of the same walk.
By the way, Proposition 4.1 provides an improvement of the convergence to a radial SLE . 2 In [10] the loop erasure is unti-chronological (loops are discarded in the reverse order). The reason is that one wants to consider the loop erasure determined from the boundary. Because the radial SLE 2 starts at a boundary point and stops at an inner point, and one wants to use a domain Markov property of the loop erasure. In [3], they used the reversibility property of the loop erasure of a simple random walk proved by Lawler [2]. Proposition 4.1 implies that the convergence to SLE 2 in the result of Yadin and Yehudayoff is valid also for LERW with the loops discarded in the chronological order instead of unti-chronological order.
The rest of the paper is organized as follows. In Sections 2 and 3 we give brief expositions of the Loewner evolution and SLE, respectively, and the fundamental results
relevant to the present issue or used in the proof of our results. In Section 4, consisting of three subsections, we first give the framework of our problem, the planar graph as well as the random walk on it, and bring in the LERW together with results associated with it (Subsection 4.1); we then present a martingale associated with the LERW (Subsection 4.2); we also present the result of [10] which asserts an approximation of the harmonic measure of our random walk by the classical Poisson kernel and a trivial lemma of the planar graph (Subsection 4.3). The statement and proof of the main result of the present paper are given in Section 5. The convergence of the loop erasure to SLE 2 curve with respect to the driving function is given in Subsection 5.1, where a certain probability estimate proved in Section 6 is taken for granted. The convergence of the loop erasure to SLE 2 curve in a uniform topology is given in Subsection 5.2, where we prove the invariance of law of LERW in (a double) time reversion. In Section 6 we verify the aforementioned probability estimate which plays an crucial role in the proof of our result, a probability estimate of the scaled random walk on D ∩ V δ starting at a boundary vertex under the conditional law given that it exists the domain through another boundary vertex that is specified in advance.
## 2 Loewner chain
In this section, consisting of four subsections, we give a brief exposition of the Loewner evolution and some results relevant to the present issue. The standard results in the theory as given in Lawler's book [1] are stated under the heading as P 2.k ( k = 1 2 , , . . . ).
## 2.1 Conformal map and half-plane capacity
Let H := { z ∈ C : Im z > 0 } be the upper half plane. A bounded subset A ⊂ H is called a compact H -hull if A = A ∩ H and H \ A is a simply connected domain. Let Q denote the set of compact H -hulls. For any A ∈ Q , there exists a unique conformal map g A : H \ A → H satisfying | g A ( z ) - | → z 0 as z →∞ . The half-plane capacity hcap( A ) is defined by
$$\ h { \text{cap} } ( A ) \coloneqq \lim _ { z \rightarrow \infty } z ( g _ { A } ( z ) - z ).$$
Then, g A has the expansion
$$\ v a p m o w a n d \\ g _ { A } ( z ) = z + \frac { \text{hcap} ( A ) } { z } + O \left ( \frac { 1 } { | z | ^ { 2 } } \right ), \quad z \to \infty. \\ \int _ { \dots } \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \end{array}$$
The half-plane capacity has some nice properties, of which we need the following.
$$\mathbf P \ 2. 1 \cdot \left ( p 6 9 \text{-} 7 1 \right ) I f \, r > 0, x \in \mathbb { R }, A \in \mathcal { Q }, \text{ then}$$
$$\ h c { \text{ap} } ( r A ) = r ^ { 2 } \text{hcap} ( A ), \text{ \ \ h c { \text{ap} } ( A + x ) = \text{hcap} ( A ).$$
$$I f \, A, B \in \mathcal { Q }, A \subset B, \ t h e n$$
$$\ h \text{cap} ( B ) = \text{hcap} ( A ) + \text{hcap} ( g _ { A } ( B \, \Lambda ) ).$$
## 2.2 Chordal Loewner Chain in H
A chordal Loewner chain is the solution of a type of Loewner equation that describes the evolution of a curve growing from the boundary to the boundary of a domain in C . In this section we consider the special case when the domain is H := { z ∈ C : Im z > 0 } , the upper half plane and the curve grows from the origin to the infinity in H . Suppose that γ : [0 , ∞ → ) H is a simple curve with γ (0) = 0 , γ (0 , ∞ ⊂ ) H . Then, for each t ≥ 0, there exists a unique conformal map g t : H \ γ (0 , t ] → H satisfying | g t ( z ) - | → z 0 as z →∞ . It is noted that g t can be continuously extended to the (two sided) boundary of H \ γ (0 , t ] along γ (0 , t ]. If γ is parametrized by half plane capacity (i.e., if lim z →∞ z g ( t ( z ) -z ) = 2 ), t g t satisfies the following differential equation
$$\frac { \partial } { \partial t } g _ { t } ( z ) & = \frac { 2 } { g _ { t } ( z ) - U ( t ) }, \quad g _ { 0 } ( z ) = z, \\ \int _ { t = 1 } \infty \cdots \int _ { t \pi \nu } \cdots \int _ { t \pi \nu } \cdots \int _ { t \pi \nu } \cdots \int _ { t \pi \nu \nu } \cdots \int _ { t \pi \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \varepsilon }$$
where U t ( ) = g t ( γ t ( )) and U ( ) · is a R -valued continuous function (see [1]). We call the equation (1) the chordal Loewner equation and U ( ) the driving function. ·
P 2.2. (p108) Suppose for some r < √ 2 and all s < t ,
Conversely, suppose that U ( ) : [0 · , ∞ → ) R , a continuous function, is given in advance, for z ∈ H , solve the ordinary differential equation (1) to obtain the solution g t ( z ) up to the time T z := sup { t > 0 : | g t ( z ) -U t ( ) | > 0 } and put K t := { z ∈ H : T z ≤ } t . Then for t > 0, g t ( z ) is a conformal map from H \ K t to H . The family ( g t ) t ≥ 0 describes the evolution of hulls ( K t ) t ≥ 0 corresponding to U ( ) · and growing from the boundary to ∞ . Therefore, we have a one-to-one correspondence between U ( ) · and ( K t ) t ≥ 0 . If U ( ) · is the driving function of a simple curve γ , we can recover γ from U ( ) · by the formula γ t ( ) = g -1 t ( U t ( )) and we can write K t = γ (0 , t ]. If U ( ) is sufficiently nice, then ( · K t ) t ≥ 0 is generated by a curve γ with γ (0) ∈ R , lim t →∞ γ t ( ) = ∞ (i.e., for any t ≥ 0, H \ K t is the unbounded component of H \ γ (0 , t ]). However, there exists a continuous function U ( ) · such that ( K t ) t ≥ 0 can not be generated by a curve. There is known a sufficient condition for U ( ) to drive a curve as given by ·
$$| U ( t ) - U ( s ) | \leq r \sqrt { t - s }.$$
Then ( K t ) t ≥ 0 is generated by a simple curve.
The family g , t t ≥ 0 is called the (chordal) Loewner chain generated by a curve γ or driven by a function U t ( ). In summary, a simple curve γ brings out a Loewner chain, whereby it determines the driving function U t ( ), and conversely a continuous function U t ( ) with appropriate regularity generates a curve through the Loewner chain driven by U t ( ).
Proposition 2.3. (Lemma 2.1. in [3]) There exists a constant C > 0 such that the following holds. Let K t be the corresponding hull for a Loewner chain driven by a continuous function U t ( ) . Set
$$\vec { \cdot } \cdot \\ k ( t ) \coloneqq \sqrt { t } + \sup \{ | U ( s ) - U ( 0 ) | \, \colon 0 \leq s \leq t \}.$$
Then, for any t > 0 ,
$$C ^ { - 1 } k ( t ) \leq \text{diam} ( K _ { t } ) \leq C k ( t ).$$
## 2.3 Chordal Loewner chains in simply connected domains
Let D /subsetnoteql C be a simply connected domain and ∂D be a set of prime ends. If D is a Jordan domain, then ∂D may be identified with the topological boundary of D . Let a, b be distinct points on ∂D . For p ∈ D , we define the inner radius of D with respect to p ,
$$\text{rad} _ { p } ( D ) \coloneqq \inf \{ | z - p | \, \colon z \not \in D \}.$$
/negationslash
Let φ : D → H be a conformal map with φ a ( ) = 0 , φ ( b ) = ∞ . Although φ is not unique, any other such map can be written as rφ for some r > 0. For a simple curve γ : (0 , T ) → D connecting a and b so that γ (0+) = a and γ T ( -) = b , let g t be the Loewner chain generated by the curve φ ◦ γ : (0 , T ) → H and put
$$\phi _ { t } = g _ { t } \circ \phi, \ \ t \in [ 0, \infty ).$$
We reparametrize the curve γ so that the curve φ ◦ γ in H is parametrized by half plane capacity. Denote by ( γ t ( )) the function representing the curve in this parametrization, so that 2 t = hcap( φ ◦ γ [0 , t ]). The driving function U t ( ) of the chain g t is then given by
$$U ( t ) = \phi _ { t } ( \gamma ( t ) ).$$
The family of conformal maps φ , t t ≥ 0 may also be called a chordal Loewner chain (in D ) with driving function U t ( ). For each s > 0, φ s conformally maps D s ( ) := D \ γ (0 , s ] onto H with φ s ( a s ) = U s ( ), φ s ( b ) = ∞ , where a s = γ s ( ) and the curve γ ( s ) ( ) := t γ s ( + ) t connects a s and b in D s ( ). On putting
$$g _ { t } ^ { ( s ) } = g _ { s + t } \circ g _ { s } ^ { - 1 } \quad \text{and} \quad \phi _ { t } ^ { ( s ) } = \phi _ { s + t },$$
substitution into U s ( + ) = t φ s + t ( γ s ( + )) yields t
$$U ( t + s ) = \phi _ { t } ^ { ( s ) } ( \gamma ^ { ( s ) } ( t ) ).$$
It follows from (2) that φ ( s ) t = g ( s ) t ◦ φ s and g ( s ) t (and φ ( s ) t ) is the Loewner chain generated by the curve γ ( s ) ; and also, from (3) that U ( s ) ( ) := t U s ( + ) is the driving function of the t chain φ ( s ) t in D s ( ).
Define p t ( ) ∈ D by
$$\phi _ { t } ( p ( t ) ) = U ( t ) + i.$$
p t ( ) serves as a reference point for the study of the conformal map φ t . (See Proposition 4.5 and the remark advanced before Lemma 5.3.)
Lemma 2.4. Let T > 1 and /epsilon1 > 0 , and, given a pair ( D,γ ) , put ˜ := sup T { t ∈ [0 , T ] : | U t ( ) | < 1 //epsilon1 } . Then there exists a constant c T, /epsilon1 ( ) > 0 , which may also depend on ( D,γ (0)) but does not on ( γ t , t > ( ) 0) , such that
$$\text{rad} _ { p ( t ) } ( D ( t ) ) \geq c ( T, \epsilon ) \text{rad} _ { p ( 0 ) } ( D ) \text{ \ for \ } t < \tilde { T }.$$
Proof. We claim that
$$| \phi ( p ( t ) ) - \phi ( \gamma ( t ^ { \prime } ) ) | \geq 2 ^ { - 1 } e ^ { - 4 \tilde { T } } \quad \text{if} \quad t ^ { \prime } \leq t < \tilde { T }.$$
Let t ′ ≤ t < T ˜ and z = φ γ t ( ( ′ )), and put
$$y ( s ) = g _ { s } ( \phi ( p ( t ) ) ) - g _ { s } ( z ), \quad 0 \leq s \leq t.$$
We prove | y (0) | = | φ p t ( ( )) - | ≥ z 2 -1 e -4 ˜ T . Recalling that Im g s ( w ) is decreasing in s for any w ∈ H , we see that
$$\text{Im} \, g _ { s } \circ \phi ( p ( t ) ) \geq \text{Im} \, g _ { t } \circ \phi ( p ( t ) ) = 1 \quad \text{if} \quad s \leq t.$$
Applying this with s = 0 we have | y (0) | ≥ 1 / 2 if Im z ≤ 1 / 2. Let Im z > 1 / 2 and define τ := inf { t ≥ 0 : Im g t ( z ) = 1 / 2 } . Then τ < t ′ ≤ t (since Im g t ′ ( z ) = 0) and the Loewner equation together with the inequality (5) shows
$$\text{equation together with the inequality } ( ) \text{ shows} \\ \left | \frac { d } { d s } y ( s ) \right | = \frac { 2 | y ( s ) | } { | g _ { s } \circ \phi ( p ( t ) ) - U ( s ) | \cdot | g _ { s } ( z ) - U ( s ) | } \leq 4 | y ( s ) | \quad \text{for} \quad 0 \leq s \leq \tau. \\ \text{ Hence } | y ( s ) | \text{ is absolutely continuous and satisfies } \frac { d } { d s } | y ( s ) | \leq 4 | y ( s ) |, \text{ so that} \\.$$
∣ Hence | y s ( ) | is absolutely continuous and satisfies d ds | y s ( ) | ≤ 4 | y s ( ) , so that |
$$| y ( \tau ) | \leq | y ( 0 ) | e ^ { 4 \tau }.$$
Using (5) again we have 1 / 2 ≤ Im y τ ( ) so that 1 / 2 ≤ | y (0) | e 4 ˜ T , which is the same as what we need to prove. Thus the claim (4) is verified.
It is proved in [8] (the proof of Corollary 4.3) that the set { φ p t ( ( )) : t < T ˜ } is included in a compact set of H depending only on T and ε , whence according to the Koebe distortion theorem rad p t ( ) ( D ) ≥ c 0 ( T, /epsilon1 ) rad p (0) ( D ) for some constant c 0 ( T, /epsilon1 ) > 0. For the proof of the lemma it therefore suffices to show that
$$| p ( t ) - \gamma ( t ^ { \prime } ) | \geq c _ { 1 } ( T, \epsilon ) \, \text{dist} ( p ( t ), \partial D ) \ \text{ for } \ t ^ { \prime } \leq t < \tilde { T }.$$
To this end we may suppose | p t ( ) -γ t ( ′ ) | < 2 -1 dist( p t , ∂D ( ) ). Applying (4) and the distortion theorem in turn yields
$$2 ^ { - 1 } e ^ { - 4 \tilde { T } } \leq | \phi ( p ( t ) ) - \phi ( \gamma ( t ^ { \prime } ) ) | \leq 1 6 | p ( t ) - \gamma ( t ^ { \prime } ) | \cdot \frac { \text{dist} ( \phi ( p ( t ) ), \mathbb { R } ) } { \text{dist} ( p ( t ), \partial D ) }.$$
We know that dist( φ p t ( ( )) , R ) ≤ M for some constant M = M T,/epsilon1 ( ) > 0 from the result of [8] mentioned above. Hence | p t ( ) -γ t ( ′ ) | ≥ [ e -4 T / 32 M ] dist( p t , ∂D ( ) ) as desired.
## 2.4 Metrics on curves
Let γ, γ j ( j = 1 2 , , . . . ) be curves which generate the Loewner chains. Let U t ( ) and U j ( ) t be driving functions corresponding to γ and γ j , respectively. If U j ( ) converges uniformly t to U t ( ) on any bounded interval, then we will say that γ j converges to γ with respect to the driving function.
Next, we consider the metric on the space of unparametrized curves in C . Let f , f 1 2 : [0 , 1] → C be a continuous, non-locally constant functions. If there exists a continuously increasing bijection α : [0 , 1] → [0 , 1] such that f 2 = f 1 ◦ α , then we will say f 1 and f 2 are the same up to reparametrization, denoted by f 1 ∼ f 2 . A unparametrized curve γ is
defined to be an equivalence class modulo ∼ . Let d ∗ be the spherical metric on ̂ C . We define the metric on the space of unparametrized curves by
$$\dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \delta$$
where f i any function in the equivalence class γ i , and the infimum is taken over all reparametrizations α which are continuously increasing bijections of [0 , 1]. We often denote by the same notation γ a parametrized curve as well as an unparametrized curve. Let us denote by γ -the time reversal of γ .
The convergence with respect to the driving function is weaker than the convergence with respect to the metric d U . We will consider a sufficient condition for the convergence with respect to the metric d U when we have the convergence with respect to the driving function. Let D /subsetnoteql C be a simply connected domain and ∂D be the set of prime ends of D . Let a, b ∈ ∂D be distinct points. Let φ : D → H be a conformal map with φ a ( ) = 0 , φ ( b ) = ∞ . Let φ -: D → H be a conformal map with φ -( b ) = 0 , φ -( a ) = ∞ .
Theorem 2.5. (Theorem 1.2 in [8]) Let { γ j } be a sequence of simple curves travelling from a to b in D . Suppose that there exists simple curves γ and η such that φ ◦ γ j converges to φ ◦ γ with respect to the driving function and φ -◦ γ j -converges to φ -◦ η with respect to the driving function. Then γ -= η and γ j converges to γ with respect to the metric d U .
## 3 Schramm-Loewner evolutions
## 3.1 SLE in the upper half plane
Let B t be a one-dimensional standard Brownian motion with B 0 = 0. A chordal SchrammLoewner evolution with parameter κ > 0 (abbreviated as chordal SLE κ ) is the random family of conformal map g t obtained from the chordal Loewner equation
$$\frac { \partial } { \partial t } g _ { t } ( z ) & = \frac { 2 } { g _ { t } ( z ) - \sqrt { \kappa } B _ { t } }, \quad g _ { 0 } ( z ) = z \quad ( z \in \mathbb { H } ).$$
Let K t be an evolving (random) hull corresponding to SLE . κ Because B t is not (1/2)H¨ older continuous, we can not use P2.2 and it is not easy to see whether K t is generated by a curve. However, according to the following results K t is actually generated by a curve with full probability.
P 3.1. (p148) With probability 1, the limit γ t ( ) := lim z → 0 g -1 t ( z + √ κB t ) exists for any t ≥ 0 and K t is generated by the curve γ .
This curve γ is called a chordal SLE κ curve in H from 0 to ∞ . The following properties of SLE κ curves are easily verified.
P 3.2. (p148) Suppose that γ is a chordal SLE κ curve in H and r > 0 . Let ̂ γ t ( ) := r -1 γ r t ( 2 ) . Then, γ has the same distribution as γ .
̂ P 3.3. (p147) Suppose that γ is a chordal SLE κ curve in H . Let τ be a stopping time. Let ̂ γ t ( ) := g τ ( γ t ( + )) τ - √ κB τ . Then, ̂ γ has the same distribution as γ . 7
The behaviour of a chordal SLE κ curve depends on the value of the parameter κ . There is three phases in the behaviour of a chordal SLE κ curve. The two phases transitions take place at the values κ = 4 and κ = 8.
P 3.4. (p150-151) Suppose that γ be a chordal SLE κ curve in H .
- · If 0 < κ ≤ 4 , then w.p.1, γ is a simple curve with γ (0 , ∞ ⊂ ) H .
- · If 4 < κ < 8 , then w.p.1, γ (0 , ∞ ∩ ) H = H and ∪ t> 0 K t = H .
/negationslash
- · If κ ≥ 8 , then w.p.1, γ is a space-filling curve, i.e., γ [0 , ∞ ) = H .
## 3.2 SLE in simply connected domains
Let γ be a chordal SLE κ curve in H from 0 to ∞ . As in the subsection 2.3 let D /subsetnoteql C be a simply connected domain, ∂D a set of prime ends, a, b two distinct points on ∂D and φ : D → H a conformal map with φ a ( ) = 0 , φ ( b ) = ∞ . Although φ is not unique, any other such map ˜ can be written as φ rφ for some r > 0. By P 3.2 , φ -1 ( γ ) is independent of the choice of the map up to a time change and we consider SLE κ curves in D as unparametrized curves. A chordal SLE κ curve in D from a to b is defined by φ -1 ( γ ).
The two properties stated in the next proposition, called the domain Markov property and conformal invariance, respectively, immediately follow from the definition of SLE.
P 3.5. Let γ be a chordal SLE κ curve in D from a to b and µ a,b D ; be a law of γ . Let f : D → D ′ be a conformal map. Then,
$$\mu _ { a, b ; D } ( \cdot | \gamma ( 0, t ] ) = \mu _ { \gamma ( t ), b ; D \nmid \gamma ( 0, t ] } ( \cdot ),$$
and
$$f \circ \mu _ { a, b ; D } ( \cdot ) = \mu _ { f ( a ), f ( b ) ; D ^ { \prime } } ( \cdot ).$$
In the theory of SLE, it is easier to prove the convergence with respect to the driving function than in the metric d U . Theorem 2.5 implies the following result, which we shall apply the following result to derive the convergence with respect to d U of LERW from that of the driving function. Let φ -: D → H be a conformal map with φ -( b ) = 0 , φ -( a ) = ∞ .
Theorem 3.6. ([8]) Let { γ j } be a sequence of simple random curves travelling from a to b in D . Let κ ≤ 4 , and γ a, b ( ) be the chordal SLE κ curve in D from a to b . φ ◦ γ j and φ -◦ γ j -converge weakly to a chordal SLE κ curve in H with respect to the driving function. Then γ j converges weakly to γ a, b ( ) with respect to d U .
The reversibility of SLE holds at least for κ ≤ 4.
Theorem 3.7. (Theorem 2.1 in [11]) Let κ ≤ 4 . The time-reversal of a chordal SLE κ curve in D from a to b has the same distribution as chordal SLE κ curve in D from b to a .
If κ > 8, then SLE curve is not reversible.
## 4 Loop erased random walks
## 4.1 Some property of LERW
For any u, v ∈ C , we write [ u, v ] = { (1 -t ) u + tv : 0 ≤ t ≤ 1 } for the line segment whose end points are u and v . Let V ⊂ C be a countable subset with 0 ∈ V . Let E : V × V → [0 , ∞ ) and E = { ( u, v ) : E u, v ( ) > 0 } . We call G = ( V, E ) a directed weighted graph. We assume that ∑ v ∈ V E u, v ( ) < ∞ for every u ∈ V , and put
We call G that satisfies the following conditions a planar irreducible graph.
$$p ( u, v ) \coloneqq \frac { E ( u, v ) } { \sum _ { w \in V } E ( u, w ) }. \\ \colon & \text{following conditions a planar irre}.$$
- 1. G
$$u ^ { \prime }, v ^ { \prime } ) & \in E, \, [ u, v ] \cap [ u ^ { \prime }, v ^ { \prime } ] \in \{ \emptyset, \{ u \}, \{ v \} \}. ) \\ & \quad \quad \quad$$
- is a planar graph. (i.e. for every distinct edges ( u, v ) , ( u , v ′ ′ ) E , [ u, v ] [ u , v ′ ′ ] , u , v .)
- 2. For any compact set K ⊂ C , the number of vertices v ∈ K is finite.
- 3. The Markov chain S ( ) on · V with transition probability p u, v ( ) is irreducible. (i.e. for every u, v ∈ V , there exists n ∈ N such that P ( S n ( ) = v | S (0) = u ) > 0.)
We call S ( ) the natural random walk on · G . For the reminder of this paper we think that G is a planar irreducible graph.
For any simply connected domain D /subsetnoteql C , let V ( D ) := V ∩ D . Define
$$\partial _ { o u t } V ( D ) \coloneqq \{ ( u, v ) \in E \, \colon [ u, v ] \cap \partial D \neq \emptyset, u \in V ( D ) \}$$
/negationslash and
$$\partial _ { i n } V ( D ) \coloneqq \{ ( u, v ) \in E \colon [ u, v ] \cap \partial D \neq \emptyset, v \in V ( D ) \}. \\ \text{sub-} \colon \text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text
$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text#$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text}$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text`$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text'$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text.$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text51$\text$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{#\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{ $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text1$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text<$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$$
/negationslash
$$\begin{array} { l l l } \text{The first exit time from $D$ is defined by} \\ \tau _ { D } \coloneqq \begin{cases} \inf \{ n \geq 1 \colon ( S ( n - 1 ), S ( n ) ) \in \partial V _ { o u t } ( D ) \} & \text{if} \ \ S ( 0 ) \in V ( D ) \\ \inf \{ n \geq 2 \colon ( S ( n - 1 ), S ( n ) ) \in \partial V _ { o u t } ( D ) \} & \text{if} \ \ ( S ( 0 ), S ( 1 ) ) \in \partial _ { i n } V ( D ) \ \ \. \\ 0 & \text{otherwise} \end{cases} \\ \text{We sometimes consider the edge $(u,v)\in \partial_{o u t } V ( D )$ as the vertex $v$, and the edge $(u,v)\in \partial_{u t } V ( D )$ and $S(0)\in \partial_{i n } V ( D )$ and for a} \\ \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _{ - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ | \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { -. } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { + } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { ( } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { [ } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - ] \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { -} \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - + } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in. } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _{ + } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _{ \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - = } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { 0 } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ { - } \hat { \in } _ {$$
The first exit time from D is defined by
.
We sometimes consider the edge ( u, v ) ∈ ∂ out V ( D ) as the vertex v , and the edge ( u, v ) ∈ ∂ in V ( D ) as the vertex u ; e.g., we write S τ ( D ) ∈ ∂ out V ( D ) and S (0) ∈ ∂ in V ( D ) and for a set J ⊂ ∂D , we write S τ ( D ) ∈ J instead of writing [ S τ ( D -1) , S ( τ D )] ∩ J = ∅ .
/negationslash
Loop erasure. Let ω = ( ω , ω , . . . , ω 0 1 n ) be a finite sequence of points. Let s 0 = max { k ≥ 0 : ω 0 = ω k } . Inductively, we define s m = max { k ≥ 0 : ω s m -1 +1 = ω k } . If l = min { m ≥ 0 : ω s m = ω n } , then the loop erasure of ω is defined by
$$L [ \omega ] = ( \omega _ { s _ { 0 } }, \omega _ { s _ { 1 } }, \dots, \omega _ { s _ { l } } ).$$
The time-reversal of ω is defined by
$$\omega ^ { - } = ( \omega _ { n }, \omega _ { n - 1 }, \dots, \omega _ { 0 } ).$$
/negationslash
It is readily recognized that the operations L and -are not commutable, namely, L ω [ -] = L ω [ ] -in general. If the transition probability p u, v ( ) is symmetric, then the following result has been proved by Lawler in [2]. For our purpose, we prove the following result without assuming that p u, v ( ) is symmetric.
Proposition 4.1. Let S ( ) · be a natural random walk on G .
$$\mathbf P ( L [ ( S ( 0 ), S ( 1 ), \dots, S ( \tau _ { D } ) ) ^ { - } ] = \omega ) = \mathbf P ( L [ ( S ( 0 ), S ( 1 ), \dots, S ( \tau _ { D } ) ) ] ^ { - } = \omega ).$$
Remark. Theorem 4.1 implies that the convergence to the radial SLE 2 in the result of Yadin and Yehudayoff (Theorem1.1 in [10]) is valid also for LERW with the loops discarded in the chronological order instead of unti-chronological order.
Proof. Let ω = ( ω , . . . , ω 0 n ) and ω , . . . , ω 1 n -1 ∈ V ( D ) be distinct and ( ω n -1 , ω n ) ∈ ∂ out V ( D ). Our task is to show the identity
$$\mathbf P ( L [ ( S ( 0 ), \dots, S ( \tau _ { D } ) ) ] = \omega ) = \mathbf P ( L [ ( S ( 0 ), \dots, S ( \tau _ { D } ) ) ^ { - } ] = \omega ^ { - } ).$$
Let q : V × V → [0 , 1]. Set
$$G _ { q } ( x ; D ) = 1 + \sum _ { k = 0 } ^ { \infty } \sum _ { \omega ^ { \prime } \subset D \colon \omega _ { 0 } ^ { \prime } = x, \omega _ { k } ^ { \prime } = x } q ( \omega _ { 0 } ^ { \prime }, \omega _ { 1 } ^ { \prime } ) \cdots q ( \omega _ { k - 1 } ^ { \prime }, \omega _ { k } ^ { \prime } ), \\. \. \. \. \. \. \. \. \. \. \. \. \. \. \. \. \. \.$$
where the inner summation is taken over all paths ω ′ = ( ω , . . . , ω ′ 0 ′ k ) in D such that ω ′ 0 = x, ω ′ k = x .
The probability of LERW is described by the following (See [2]).
$$\mathbf P ( L [ ( S ( 0 ), \dots, S ( \tau _ { D } ) ] = \omega ) = \prod _ { j = 0 } ^ { n - 1 } p ( \omega _ { j }, \omega _ { j + 1 } ) G _ { p } ( \omega _ { j } ; D \, \left \{ \omega _ { 0 }, \dots, \omega _ { j - 1 } \right \} ) \\ \intertext { m a l l a s d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d }$$
By the exchange lemma (the equation (12.2.3) in [2]), we get
$$\mathbf P ( L [ ( S ( 0 ), \dots, S ( \tau _ { D } ) ] = \omega ) & = \prod _ { j = 0 } ^ { n - 1 } p ( \omega _ { j }, \omega _ { j + 1 } ) G _ { p } ( \omega _ { j } ; D \, \left \{ \omega _ { j + 1 }, \dots, \omega _ { n - 1 } \right \} ) \quad \quad ( 9 ) \\ \Omega _ { m } + \mathbf h a \sim \mathbf h a n d$$
On the other hand,
$$\text{$m$ was $m$,} \\ \text{$P(L[(S(0),\dots,S(\tau_{D}))}^{-}] = \omega^{-} ) } & = \sum _ { \omega ^ { \prime } \subset D \colon L [ \omega ^ { \prime } ) ^ { - } ] = \omega ^ { - } } \prod _ { i = 0 } ^ { | \omega ^ { \prime } | - 1 } p ( \omega ^ { \prime } _ { i }, \omega ^ { \prime } _ { i + 1 } ) \\ & = \sum _ { \omega ^ { \prime } \subset D \colon L [ \omega ^ { \prime } ] = \omega ^ { - } } \prod _ { i = 0 } ^ { | \omega ^ { \prime } | - 1 } p ^ { * } ( \omega ^ { \prime } _ { i }, \omega ^ { \prime } _ { i + 1 } ), \\ \text{where} \, | \omega ^ { \prime } | \text{ is the length of $\omega^{\prime}$ and $p^{*}(x,y) \colon=p(y,x).\text{ This equation and decomposing $\omega$} }$$
where | ω ′ | is the length of ω ′ and p ∗ ( x, y ) := p y, x ( ). This equation and decomposing ω ′ between its last visit to ω n -1 , . . . , ω 0 imply that
$$\text{SVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVV} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \, \text{sVVB} \\ \text{PH} ( L [ ( S ( 0 ), \dots, S ( \tau _ { D } ) ) ^ { - } ] = \omega ^ { - } ) & = \prod _ { j = 0 } ^ { n - 1 } p ^ { * } ( \omega _ { j + 1 }, \omega _ { j } ) G _ { p ^ { * } } ( \omega _ { j } ; D \, \left \{ \omega _ { j + 1 }, \dots, \omega _ { n - 1 } \right \} ). \quad ( 1 0 ) \\ \text{Finally observe that } G _ { p } ( x ; D ^ { \prime } ) = G _ { p ^ { * } } ( x ; D ^ { \prime } ). \text{ Thus, } ( 9 ) \text{ and } ( 1 0 ) \text{ imply } ( 8 ). \quad \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{} \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{$$
Finally observe that G p ( x D ; ′ ) = G p ∗ ( x D ; ′ ). Thus, (9) and (10) imply (8).
Let γ = ( γ , γ 0 1 , . . . , γ l ) be the loop erasure of the time-reversal of the natural random walk stopped on exiting D . By Proposition 4.1, we may think that γ is the time-reversal of the loop erasure. (In Section 5, we treat γ as the time-reversal of the loop erasure. But in this section, we treat γ as the loop erasure of the time-reversal because it is more suitable to consider the following properties of γ .)
$$\text{Let } D _ { j } \coloneqq D \, \bigcup _ { i = 0 } ^ { j - 1 } [ \gamma _ { i }, \gamma _ { i + 1 } ]. \text{ For any } j \in \mathbb { N },$$
$$n _ { j } \coloneqq \min \{ n \geq 0 \, \colon S ( n ) = \gamma _ { j } \}.$$
Because the loop erasure γ is determined from the boundary, γ has the following Markov property.
Proposition 4.2. (Lemma 3.2. in [3]) Conditioned on γ [0 , j ] , the following holds.
- 1. S [0 , n j ] and S n , τ [ j D ] are independent.
- 2. γ j, l [ ] has the same distribution as the loop erasure of time-reversal of the natural random walk S [0 , τ D j ] conditioned to exit at γ j .
## 4.2 Martingale observable for LERW
Let D /subsetnoteql C be a simply connected domain. Let S x ( ) · be a natural random walk on G started at x ∈ V . Let v 0 ∈ V ( D ) ∪ ∂ in V ( D ) and γ be the loop erasure of time-reversal of the natural random walk S v 0 [0 , τ D ]. Let D j := D \ ∪ j -1 i =0 [ γ , γ i i +1 ]. The hitting probability H u,v j ( ) is defined by
$$H _ { j } ( u, v ) \coloneqq \mathbf P ( S ^ { u } ( \tau _ { D _ { j } } ) = v ).$$
Let F j be a filtration generated by γ [0 , j ].
Proposition 4.3. For any w ∈ V ( D ) , let
$$M _ { j } \coloneqq \frac { H _ { j } ( w, \gamma _ { j } ) } { H _ { j } ( v _ { 0 }, \gamma _ { j } ) }.$$
Then, M j is a martingale with respect to F j .
Lawler, Schramm and Werner [3] point out that the martingale M j given above should be a possible martingale observable, although they don't adopt it but a martingale formed by the Green functions of evolving domains. They provide a curtailed proof that M j is a martingale. Since M j plays the central role in this paper we give a detailed proof of this fact.
Proof. First, we consider another representation of M j . Let ̂ S x ( ) be a independent copy · of S x ( ) · and L x be the loop erasure of the time-reversal of ̂ S x [0 , τ D ]. We will denote by Q the law of S . Fix γ [0 , j ]. By proposition 4.2,
$$S ^ { x } ( \cdot ) \text{ and } L _ { x } & \text{ be the loop erasure of the time-reversal of } \widehat { S } ^ { x } [ 0, \tau _ { D } ]. \text{ We will denote} \\ \text{the law of } \widehat { S }. \text{ Fix } \gamma [ 0, j ]. \text{ By proposition } 4. 2, \\ \frac { \mathbf Q ( L _ { w } [ 0, j ] = \gamma [ 0, j ] ) } { \mathbf Q ( L _ { v _ { 0 } } [ 0, j ] = \gamma [ 0, j ] ) } & = \frac { \mathbf Q ( \widehat { S } ^ { w } ( \tau _ { D _ { j } } ) = \gamma _ { j } ) \mathbf Q ( L _ { \gamma _ { j } } [ 0, j ] = \gamma [ 0, j ] ) } { \mathbf Q ( \widehat { S } ^ { v _ { 0 } } ( \tau _ { D _ { j } } ) = \gamma _ { j } ) \mathbf Q ( L _ { \gamma _ { j } } [ 0, j ] = \gamma [ 0, j ] ) } \\ & = \frac { H _ { j } ( w, \gamma _ { j } ) } { H _ { j } ( v _ { 0 }, \gamma _ { j } ) }.$$
Therefore, we can write
$$M _ { j } = \frac { \mathbf Q ( L _ { w } [ 0, j ] = \gamma [ 0, j ] ) } { \mathbf Q ( L _ { v _ { 0 } } [ 0, j ] = \gamma [ 0, j ] ) }.$$
Hence,
$$\mathbf E [ M _ { j + 1 } | \gamma [ 0, j ] ] = \sum _ { v \in V ( D _ { j } ) } \mathbf P ( \gamma _ { j + 1 } = v | \gamma [ 0, j ] ) \cdot \frac { \mathbf Q ( L _ { w } [ 0, j ] = \gamma [ 0, j ], L _ { w } ( j + 1 ) = v ) } { \mathbf Q ( L _ { v _ { 0 } } [ 0, j ] = \gamma [ 0, j ], L _ { v _ { 0 } } ( j + 1 ) = v ) }, \\. \quad - \cdot \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad.$$
and, since P ( γ j +1 = v γ | [0 , j ]) = Q ( L v 0 ( j +1) = v L | v 0 [0 , j ] = γ [0 , j ]), the right-hand side reduces to
$$& \sum _ { v \in V ( D _ { j } ) } \frac { \mathbf Q ( L _ { w } [ 0, j ] = \gamma [ 0, j ], L _ { w } ( j + 1 ) = v ) } { \mathbf Q ( L _ { v _ { 0 } } [ 0, j ] = \gamma [ 0, j ] ) } = \frac { \mathbf Q ( L _ { w } [ 0, j ] = \gamma [ 0, j ] ) } { \mathbf Q ( L _ { v _ { 0 } } [ 0, j ] = \gamma [ 0, j ] ) } = M _ { j }. \\ & \dots \cdot \cdot \cdot \cdot \cdot$$
Thus, M j is a martingale.
## 4.3 Estimates of discrete harmonic measures
For δ > 0, the graph G δ = ( V , E δ δ ) defined by
$$V _ { \delta } = \{ \delta u \, \colon u \in V \}, \ \ E _ { \delta } = \{ ( \delta u, \delta v ) \, \colon E ( u, v ) > 0 \}.$$
Let the Markov chain S δ ( ) on · V δ be the scaling of S ( ) by a factor of · δ . We call S δ ( ) the · natural random walk on G δ . Let S x δ ( ) be a natural random walk on · G δ started at x ∈ V δ . Similarly, we can define H ( δ ) j ( u, v ), V δ ( D ), ∂ out V δ ( D ), ∂ in V δ ( D ).
Let D = { z ∈ C : | z | < 1 } be the unit disc.
Definition 4.4. If the family of the random walks S x δ satisfies the following condition, then we say that S x δ satisfies invariance principle:
For any compact set K ⊂ D and /epsilon1 > 0 , there is some δ 0 > 0 such that the following holds. Let Z x be a two-dimensional Brownian motion started at x stopped on exiting D . For any 0 < δ < δ 0 and x ∈ K ∩ V δ , there exists a coupling of S x δ and Z x satisfying
$$\mathbf P ( d _ { \mathcal { U } } ( S _ { \delta } ^ { x } [ 0, \tau _ { \mathbb { D } } ], Z ^ { x } ) > \epsilon ) < \epsilon.$$
In view of the Skorokhod representation theorem the above condition is equivalent to holding that S x δ weakly converges to Z x uniformly for all x ∈ K .
In [10] (Lemma 1.2) the following result is proved.
Proposition 4.5. Suppose that S x δ satisfies invariance principle. For any positive constants r , ε and η < 1 , there exists some δ 0 > 0 such that for all 0 < δ < δ 0 the following holds. Let D ⊂ D , let p ∈ V δ ( D ) be such that rad ( p D ) ≥ r , and let ψ : D → D be a conformal map with ψ p ( ) = 0 . Let y ∈ V δ ( D ) be such that | ψ y ( ) | < 1 -η and let a ∈ ∂ out V δ ( D ) . Then, where K D stands for the Poisson kernel of D .
$$\mathfrak { u } _ { \ } \psi ( \mathfrak { p } ) & \, - \, \infty. \, \mathfrak { u } _ { \ } \in v _ { \ } \in \mathfrak { v } _ { \delta } ( \mathfrak { v } _ { \ } ) \, \in v _ { \delta } \, \mathfrak { u } _ { \ } \in \mathfrak { u } _ { \ } \mathfrak { u } _ { \ } | \psi ^ { \prime } \\, \\ & \left | \frac { H _ { 0 } ^ { ( \delta ) } ( y, a ) } { H _ { 0 } ^ { ( \delta ) } ( p, a ) } - \frac { K _ { \mathbb { D } } ( \psi ( y ), \psi ( a ) ) } { K _ { \mathbb { D } } ( \psi ( p ), \psi ( a ) ) } \right | < \epsilon, \\ & \text{the Poisson kernel of $\mathbb{D}$}. \\ & \quad. \,. \,.$$
The Poisson kernel of H is given by
$$\text{mounds} & \circ \alpha \nu \nu \nu \nu \nu \\ K _ { \mathbb { H } } ( u, v ) \coloneqq - \frac { 1 } { \pi } & { \text{Im} } \left ( \frac { 1 } { u - v } \right ) = \frac { 1 } { \pi } \frac { \text{Im} \, u } { | u - v | ^ { 2 } }. \\ \circ \max \, \text{mounds} \, \text{from-touted in terms of $K_{-}$}_{-}$} \text{or any non-one}.$$
The result above may be translated in terms of K H . For our purpose we apply it in a rather trivial fashion. Let
Corollary 4.6. Suppose that S x δ satisfies invariance principle. For any constants r > 0 , /epsilon1 > 0 , η > 0 and λ > 1 , there exists some δ 0 > 0 such that for all 0 < δ < δ 0 the following holds. Let D ⊂ D , let p ∈ D be such that rad ( p D ) ≥ r , and let φ : D → H be a conformal map with φ p ( ) = i . Let y, w ∈ V δ ( D ) be such that Im φ y ( ) > η, Im φ w ( ) > η and | φ y ( ) | < λ, φ w | ( ) | < λ . Then, for all a ∈ ∂ out V δ ( D )
∣
∣
∣ ∣ ∣ ∣ ∣ ∣ Proof. Let p δ ∈ V δ ( D ) be a nearest point of p . Applying Proposition 4.5 with p = p δ ,
H
(
δ
)
0
(
w, a
)
K
H
(
φ w , φ a
(
)
(
))
H
(
δ
)
< /epsilon1.
0
(
y, a
)
-
K
H
(
φ y , φ a
(
)
(
))
∣
∣
$$\frac { H _ { 0 } ^ { ( \delta ) } ( w, a ) } { H _ { 0 } ^ { ( \delta ) } ( y, a ) } & = \frac { H _ { 0 } ^ { ( \delta ) } ( w, a ) / H _ { 0 } ^ { ( \delta ) } ( p _ { \delta }, a ) } { H _ { 0 } ^ { ( \delta ) } ( y, a ) / H _ { 0 } ^ { ( \delta ) } ( p _ { \delta }, a ) } \\ & = \frac { K _ { \mathbb { D } } ( \psi ( w ), \psi ( a ) ) } { K _ { \mathbb { D } } ( \psi ( y ), \psi ( a ) ) } + O ( \epsilon ). \\ \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \,.$$
Because the ratio of the Poisson kernel is conformal invariance, we find
$$\frac { K _ { \mathbb { D } } ( \psi ( w ), \psi ( a ) ) } { K _ { \mathbb { D } } ( \psi ( y ), \psi ( a ) ) } = \frac { K _ { \mathbb { H } } ( \phi ( w ), \phi ( a ) ) } { K _ { \mathbb { H } } ( \phi ( y ), \phi ( a ) ) }.$$
This completes the proof.
Here we present the following trivial lemma for convenience of a later citation.
Lemma 4.7. Suppose that S x δ satisfy invariance principle. For any /epsilon1 > 0 , there exists δ 0 such that for all 0 < δ < δ 0 , the length of edges of G δ in D is bounded above by /epsilon1 .
Proof. Suppose that this Lemma is not true. Then, there exists /epsilon1 > 0 such that for some sufficiently small δ , there exists an edge e of G δ such that the length of e is bounded below by /epsilon1 . Since G δ is planar graph, S x δ can not cross the edge e , so that it cannot behave as a Brownian path and the invariance principle fails to hold.
## 5 Scaling limit
## 5.1 Convergence with respect to the driving function
Let D /subsetnoteql C be a simply connected domain and a, b two distinct points on ∂D . We say that ∂D is locally analytic at z ∈ ∂D if there exists a one-to-one analytic function f : D → C with f (0) = z and f ( D ) ∩ D = f ( { w ∈ D : Im w > 0 } ). Let G = ( V, E ) be a planar irreducible graph and S x δ a natural random walk on G δ started at x (see Section 4 for
detailed description). Let Γ a,b δ be a natural random walk on G δ started at a δ and stopped on exiting D and conditioned to hit ∂D at b δ , where a δ is a point of ∂ in V δ ( D ) close to a and b δ is a point of ∂ out V δ ( D ) close to b such that there exists a path on G δ connecting a δ and b δ in D . If ∂D is locally analytic at a and b , we can choose such a δ and b δ . Let γ a,b δ be the loop erasure of Γ a,b δ .
Theorem 5.1. Suppose that S x δ satisfy invariance principle. Let D be a bounded simply connected domain and a, b be distinct points on ∂D . Suppose that ∂D is locally analytic at a and b . Let φ : D → H be a conformal map with φ a ( ) = 0 , φ ( b ) = ∞ . Then, φ ◦ ( γ b,a δ ) -converges weakly to the chordal SLE 2 curve in H as δ → 0 with respect to the driving function.
Remark. In order to assure the uniformity of invariance principle so imposed in Definition 4.4 it suffices to suppose it only for the walk starting at a point, e.g., the origin as is shown in [9].
We abbreviate ( γ b,a δ ) -= γ = ( γ , γ 0 1 , . . . , γ l ). By Proposition 4.1, γ has the same distribution as the loop erasure of the time-reversal of Γ b,a δ . Hence, it is possible for γ to use results in Section 4. Let F j be a filtration generated by γ [0 , j ]. We may also think that γ [0 , j ] is the simple curve that is a linear interpolation.
Let U t ( ) be a driving function of φ γ ( ) and g t be a Loewner chain driven by U t ( ). Let t j := 1 2 hcap φ γ ( [0 , j ]) and
$$U _ { j } \coloneqq U ( t _ { j } ), \ \phi _ { j } \coloneqq g _ { t _ { j } } \circ \phi \text{ and } D _ { j } \coloneqq D \, \sum \gamma [ 0, j ].$$
Let p j := φ -1 j ( i + U j ). p j plays the role of a reference point, an 'origin', of D j . In radial case, such a point is fixed at the origin. But in chordal case, p j must be moved with j , so that there remains sufficient space around p j in D j , a sequence of reducing domains formed by encroachment of γ into D . (Cf. [7]).
We use the martingale introduced in Proposition 4.3, as in [3] and [10]. But we need to normalize it appropriately. We denote by S b δ a natural random walk on G δ started at b δ . Let A := φ -1 ([ -1 1]) and the normalization is made by multiplying , P ( S b δ ( τ D ) ∈ A ), which we name M j :
$$M _ { j } \coloneqq \frac { H _ { j } ^ { ( \delta ) } ( w, \gamma _ { j } ) } { H _ { j } ^ { ( \delta ) } ( b, \gamma _ { j } ) } H _ { 0 } ^ { ( \delta ) } ( b ; A ),$$
(for any δ > 0 and w ∈ V δ ( D )), where we write H ( δ ) 0 ( b ; A ) := P ( S b δ ( τ D ) ∈ A ).
In the rest of this section let r, R and η be arbitrarily fixed positive constants and suppose the same hypothesis of Theorem 5.1 to be valid. We write D for D ( r, R, η ) and consider ( D,a, b, p ) ∈ D . For dealing with the martingale observable M j defined above the following lemma plays a significant role and D ( r, R, η ) is introduced as a class for which the estimates given there is valid uniformly.
Let D /subsetnoteql C be a simply connected domain, a, b two distinct points on ∂D and φ : D → H a conformal map with φ a ( ) = 0 , φ ( b ) = ∞ as before. Let p = φ -1 ( ). i Put Ψ( ) = ( z z -i ) / z ( + ). i Define ψ := Ψ ◦ φ : D → D , which is a conformal map with ψ b ( ) = 1 , ψ ( p ) = 0 , ψ ( a ) = -1. Let D = D ( r, R, η ) be the collection of all quadruplets ( D,a, b, p ) such that rad ( p D ) ≥ r and D ⊂ R D and ψ -1 has analytic extension in { z ∈ C : | z - | 1 < η } .
Lemma 5.2. There exists a number λ 0 = λ 0 ( η ) > 1 / 2 such that for any ε > 0 and λ > λ 0 , there exists numbers δ 0 > 0 and α ∈ (0 , 1 / 2) such that if ( D,a, b, p ) ∈ D ( r, R, η ) , 0 < δ < δ 0 and D ′ = D \ φ -1 ( { z : | z | < 2 λ } ) , then and, if diam( φ γ ( [0 , j ])) < 1 , then
$$\dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \\ \mathbf P \left ( \text{Im} \, \phi ( S _ { \delta } ^ { b } ( \tau _ { D ^ { \prime } } ) ) < \alpha \lambda \, | \ S _ { \delta } ^ { b } ( \tau _ { D } ) \in A \right ) < \epsilon, \\ \iota ( \phi ( \gamma [ 0, j ] ) ) < 1, \, \text{then}$$
$$\vec { \ } \times \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \, \vec { \ } \,
\text{P} \left ( \text{Im} \, \phi ( S _ { \delta } ^ { b } ( \tau _ { D ^ { \prime } } ) ) < \alpha \lambda \ | \ S _ { \delta } ^ { b } ( \tau _ { D _ { j } } ) = \gamma _ { j } \right ) < \epsilon. \quad & & ( 1 3 ) \\ \text{of of Lemma} \, 5. 2 \text{ is involved and postponed to the end of Section 6.}$$
The proof of Lemma 5.2 is involved and postponed to the end of Section 6. For any /epsilon1 > 0, let
$$m \coloneqq \min \{ j \geq 1 \, \colon t _ { j } \geq \epsilon ^ { 2 } \text{ or } | U _ { j } - U _ { 0 } | \geq \epsilon \}.$$
Lemma 5.3. There exists a constant C > 0 and a number /epsilon1 0 > 0 such that for each positive /epsilon1 < /epsilon1 0 , there exists δ 0 > 0 such that if ( D,a, b, p ) ∈ D ( r, R, η ) and 0 < δ < δ 0 , then
$$| \mathbf E [ U _ { m } - U _ { 0 } ] | \leq C \epsilon ^ { 3 },$$
and
$$| \mathbf E [ ( U _ { m } - U _ { 0 } ) ^ { 2 } - 2 t _ { m } ] | \leq C \epsilon ^ { 3 }. \\ \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon$$
(Although U 0 = 0 , we write U 0 in the formulae above to indicate how they show be when the starting position U 0 = γ 0 is not mapped to the origin by φ .)
Proof. This proof is broken into four steps. It consists of certain estimations of the harmonic functions that constitutes the martingale observable defined by (11).
Step 1 . In this step we derive an expression, given in (16) below, of the ratio
$$H _ { j } ^ { ( \delta ) } ( b, \gamma _ { j } ) / H _ { 0 } ^ { ( \delta ) } ( b ; A ).$$
We take sufficiently small /epsilon1 0 > 0, which we need in this proof. Given 0 < /epsilon1 < /epsilon1 0 we take a number λ = 1 /ε 3 that will be specified shortly. Let D ′ := D \ φ -1 ( B U , ( 0 2 λ ) ∩ H ) (note that B U , ( 0 2 λ ) = { z : | z | < 2 λ )). In the following we consider for j = 0 1 2 , , , . . . , although we apply the resulting relation only for j = 0 , m ,
$$\lambda \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon$$
We split the sum on the right-hand side into two parts according as y is close to the boundary of D or not. The part of those y which are close to the boundary must be negligible.
Proposition 2.3 and the definition of m imply that diam( φ γ ( [0 , m -1])) = O /epsilon1 ( ). By Lemma 4.7, the harmonic measure from p of γ m [ -1 , m ] in D m is O /epsilon1 ( ) for sufficiently small δ > 0. By conformal invariance of harmonic measure, the harmonic measure from φ m -1 ( p ) of φ m -1 ( γ m [ -1 , m ]) in H \ φ m -1 ( γ m [ -1 , m ]) is O /epsilon1 ( ). This implies that diam( φ m -1 ( γ m [ -1 , m ])) = O /epsilon1 ( ), and we have
$$\text{diam} ( \phi ( \gamma [ 0, m ] ) ) = O ( \epsilon ).$$
By (14) and Lemma 5.2, we can choose α = α ε ( ) < 1 / 2 so that for all sufficiently small δ > 0, for j = 0 , m ,
This implies
$$- \, \upsilon, m, \\ \, \text{P} \left ( \text{Im} \, \phi ( S _ { \delta } ^ { b } ( \tau _ { D ^ { \prime } } ) ) < \alpha \lambda \, | \ S _ { \delta } ^ { b } ( \tau _ { D _ { j } } ) = \gamma _ { j } \right ) = O ( \epsilon ^ { 3 } ).$$
$$\frac { \mathbf P ( \text{Im} \, \phi ( S _ { \delta } ^ { b } ( \tau _ { D ^ { \prime } } ) ) < \alpha \lambda, S _ { \delta } ^ { b } ( \tau _ { D _ { j } } ) = \gamma _ { j } ) } { \mathbf P ( \text{Im} \, \phi ( S _ { \delta } ^ { b } ( \tau _ { D ^ { \prime } } ) ) \geq \alpha \lambda, S _ { \delta } ^ { b } ( \tau _ { D _ { j } } ) = \gamma _ { j } ) } & = \frac { \mathbf P ( \text{Im} \, \phi ( S _ { \delta } ^ { b } ( \tau _ { D ^ { \prime } } ) ) < \alpha \lambda \ | \ S _ { \delta } ^ { b } ( \tau _ { D _ { j } } ) = \gamma _ { j } ) } { \mathbf P ( \text{Im} \, \phi ( S _ { \delta } ^ { b } ( \tau _ { D ^ { \prime } } ) ) \geq \alpha \lambda \ | \ S _ { \delta } ^ { b } ( \tau _ { D _ { j } } ) = \gamma _ { j } ) } \\ & = O ( \epsilon ^ { 3 } ).$$
Therefore,
$$\intertext { r e e r e, } H _ { j } ^ { ( \delta ) } ( b, \gamma _ { j } ) = ( 1 + O ( \epsilon ^ { 3 } ) ) \sum _ { \substack { y \in V _ { \delta } ( D ) \\ \text{Im} \, \phi ( y ) \geq \alpha \lambda } } \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { D ^ { \prime } } ) = y, S _ { \delta } ^ { b } ( \tau _ { D _ { j } } ) = \gamma _ { j } ).$$
By strong Markov property,
$$D _ { y } & \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-} \text{-}; \\ & \frac { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { D ^ { \prime } } ) = y, S _ { \delta } ^ { b } ( \tau _ { D _ { j } } ) = \gamma _ { j } ) } { H _ { 0 } ^ { ( \delta ) } ( b ; A ) } = \frac { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { D ^ { \prime } } ) = y ) \mathbf P ( S _ { \delta } ^ { y } ( \tau _ { D } ) \in A ) } { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { D } ) \in A ) } \cdot \frac { \mathbf P ( S _ { \delta } ^ { y } ( \tau _ { D _ { j } } ) = \gamma _ { j } ) } { \mathbf P ( S _ { \delta } ^ { y } ( \tau _ { D } ) \in A ) }. \\ \text{Therefore, } ( 15 ) & \text{ implies}$$
Therefore, (15) implies
$$\frac { H _ { j } ^ { ( \delta ) } ( b, \gamma _ { j } ) } { H _ { 0 } ^ { ( \delta ) } ( b ; A ) } = ( 1 + O ( \epsilon ^ { 3 } ) ) \sum _ { \substack { y \in V _ { \delta } ( D ) \\ \text{Im} \phi ( y ) \geq \alpha \lambda } } \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { D ^ { \prime } } ) = y \ | \ S _ { \delta } ^ { b } ( \tau _ { D } ) \in A ) \cdot \frac { H _ { j } ^ { ( \delta ) } ( y, \gamma _ { j } ) } { H _ { 0 } ^ { ( \delta ) } ( y ; A ) }. \quad ( 1 6 )$$
Step 2 . Let w ∈ V δ and y ∈ V δ ( D ) satisfy
$$\text{Im} \, \phi ( w ) \geq \frac { 1 } { 2 }, \ | \phi ( w ) - U _ { 0 } | \leq 3 ; \ \text{Im} \, \phi ( y ) \geq \alpha \lambda, \ \lambda \leq | \phi ( y ) - U _ { 0 } | \leq 2 \lambda. \quad \text{(17)}$$
Applying Corollary 4.6 to the domain D with a reference point p ,
$$\frac { H _ { 0 } ^ { ( \delta ) } ( w, \gamma _ { 0 } ) } { H _ { 0 } ^ { ( \delta ) } ( y, \gamma _ { 0 } ) } = \frac { \text{Im} \, \phi ( w ) / | \phi ( w ) - U _ { 0 } | ^ { 2 } } { \text{Im} \, \phi ( y ) / | \phi ( y ) - U _ { 0 } | ^ { 2 } } + O ( \epsilon ^ { 3 } ), \\ \dots \dots \dots \dots \dots \dots \dots \dots \dots$$
and the assumed invariance principle implies
$$\text{summu navaue, p m m p s m p s } H _ { 0 } ^ { ( \delta ) } ( y ; A ) & = \frac { 1 } { \pi } \int _ { - 1 } ^ { 1 } \frac { \text{Im} \, \phi ( y ) } { | \phi ( y ) - x | ^ { 2 } } d x + O ( \varepsilon ^ { 3 } \alpha / \lambda ) & ( 1 9 ) \\ \Lambda & = \text{T-1} ^ { 2 } / \text{Im} \, \phi ( \mu ) \, < \, \mathbb { m } / \varepsilon \, \partial / \varepsilon \, \text{Local} \, \varepsilon / \lambda \, \text{mnet at email to author with } \mathbb { m } \, \mathbb { m }$$
since | φ y ( ) -U 0 | 2 / Im φ y ( ) ≤ 2 λ/α (recall α/λ must get small together with ε ). The relations (17), (18) and (19) together imply
$$\text{relations (17), (18) and (19) together imply} \\ \frac { H _ { 0 } ^ { ( \delta ) } ( w, \gamma _ { 0 } ) } { H _ { 0 } ^ { ( \delta ) } ( y, \gamma _ { 0 } ) } H _ { 0 } ^ { ( \delta ) } ( y ; A ) & = \frac { \text{Im} \, \phi ( w ) } { \pi | \phi ( w ) - U _ { 0 } | ^ { 2 } } \int _ { - 1 } ^ { 1 } \frac { | \phi ( y ) - U _ { 0 } | ^ { 2 } } { | \phi ( y ) - x | ^ { 2 } } d x + O ( \epsilon ^ { 3 } ) \\ & = \frac { 2 } { \pi } \frac { \text{Im} \, \phi ( w ) } { | \phi ( w ) - U _ { 0 } | ^ { 2 } } + O ( \epsilon ^ { 3 } ). \\ & \quad \. \.$$
From (16) and (20) we infer that
$$\frac { 1 } { M _ { 0 } } = \frac { \theta _ { 0 } ^ { ( \delta ) } ( b, \gamma _ { 0 } ) } { H _ { 0 } ^ { ( \delta ) } ( b ; A ) H _ { 0 } ^ { ( \delta ) } ( w ; \gamma _ { 0 } ) } = ( 1 + O ( \varepsilon ^ { 3 } ) ) \sum _ { \substack { y \in V _ { \delta } ( D ) \\ \text{Im} \phi ( y ) \geq \alpha \lambda } } p ( y ) \Big / \left [ \frac { 2 } { \pi } \frac { \text{Im} \, \phi ( w ) } { | \phi ( w ) - U _ { 0 } | ^ { 2 } } + O ( \epsilon ^ { 3 } ) \right ],$$
where p y ( ) = P ( S b δ ( τ D ′ ) = y | S b δ ( τ D ) ∈ A ). In view of Lemma 5.2, we can suppose
$$\lambda ^ { \prime } & \quad \nu ^ { \prime } + \sigma ^ { \prime } + \nu ^ { \prime } - \nu ^ { \prime } \\ & \quad \sum _ { \substack { y \in V _ { \delta } ( D ) \\ \text{Im} \phi ( y ) \geq \alpha \lambda } } p ( y ) = 1 + O ( \epsilon ^ { 3 } ),$$
by replacing α by smaller one if necessary. Since Im φ w / φ w ( ) | ( ) -U 0 | 2 is bounded by a universal constant, we now conclude
$$M _ { 0 } & = \frac { 2 } { \pi } \frac { \text{Im} \, \phi ( w ) } { | \phi ( w ) - U _ { 0 } | ^ { 2 } } + O ( \epsilon ^ { 3 } ) \\ & = \frac { 2 } { \pi } \text{Im} \, \left ( \frac { - 1 } { \phi ( w ) - U _ { 0 } } \right ) + O ( \epsilon ^ { 3 } ). \\. & \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \ \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } \quad \hat { \ } }$$
Step 3 . We derive an analogous formula for M m . Lemma 2.3 and (14) imply
$$t _ { m } = O ( \epsilon ^ { 2 } ), \ \ | U ( s ) - U ( 0 ) | = O ( \epsilon ) \ \text{ for } \ \forall s \in [ 0, t _ { m } ].$$
The Loewner equation (1) shows that
$$| g _ { t } ( z ) - z | & \leq t \cdot \sup _ { 0 \leq s \leq t } \frac { 2 } { | g _ { s } ( z ) - U ( s ) | }, \\. \. \. \. \. \. \. \.$$
and, observing the imaginary part of the Loewner equation,
1
Im
g
t
(
z
)
2
Im
≥
z
≥
(
- ·
exp
t
sup
.
(25)
0
≤ ≤
s
t
|
g
s
(
z
)
-
U s
(
)
|
2
)
Wealso find d dt Im g t ( z ) ≥ -2 / Im g t ( z ), and this implies d dt (Im g t ( z )) 2 ≥ -4. By integrating this relation over [0 , t ], we get (Im g t ( z )) 2 ≥ (Im z ) 2 -4 . t Since t m = O /epsilon1 ( 2 ), we have Im g s ◦ φ w ( ) ≥ 1 / 4 for 0 ≤ s ≤ t m . Therefore, (24) gives
$$| g _ { s } \circ \phi ( w ) - \phi ( w ) | = O ( \epsilon ^ { 2 } ) \quad \text{for} \quad \forall s \in [ 0, t _ { m } ].$$
Let σ := inf { t ≥ 0 : | g t ( z ) -U t ( ) | ≤ λ/ 2 } . Using (24), we get | g σ ( z ) - | ≤ z 4 σ/λ and
$$| z - U ( 0 ) | \leq \frac { 4 \sigma } { \lambda } + \frac { \lambda } { 2 } + | U ( \sigma ) - U ( 0 ) |.$$
Thus, if | z -U (0) | > λ , then σ > t m . This implies | g s ◦ φ y ( ) -U s ( ) | ≥ λ/ 2 for 0 ≤ s ≤ t m . Therefore, (24) and (25) lead to
$$| \phi _ { m } ( y ) - \phi ( y ) | = O ( \epsilon ^ { 3 } ) \quad \text{and} \quad \frac { \text{Im} \, \phi _ { m } ( y ) } { \text{Im} \, \phi ( y ) } = 1 + O ( \varepsilon ^ { 3 } ).$$
(26) and (27) imply
$$\text{Im} \, \phi _ { m } ( w ) \geq \frac { 1 } { 3 }, \, | \phi _ { m } ( w ) - U _ { m } | \leq 4 ; \quad \text{Im} \, \phi _ { m } ( y ) \geq \frac { \alpha \lambda } { 2 }, \, \frac { \lambda } { 2 } \leq | \phi _ { m } ( y ) - U _ { m } | \leq 3 \lambda.$$
and it follows from Lemma 2.4 that rad p m ( D m ) ≥ r ′ for some r ′ > 0. Therefore, we can apply Corollary 4.6 to the domain D m with the reference point p m , and hence the relation (19) implies
$$\frac { H _ { m } ^ { ( \delta ) } ( w, \gamma _ { m } ) } { H _ { m } ^ { ( \delta ) } ( y, \gamma _ { m } ) } H _ { 0 } ^ { ( \delta ) } ( y ; A ) = & \frac { \text{Im} \, \phi _ { m } ( w ) } { \pi | \phi _ { m } ( w ) - U _ { m } | ^ { 2 } } \int _ { - 1 } ^ { 1 } \frac { \text{Im} \, \phi ( y ) } { \text{Im} \, \phi _ { m } ( y ) } \cdot \frac { | \phi _ { m } ( y ) - U _ { m } | ^ { 2 } } { | \phi ( y ) - x | ^ { 2 } } d x \\ & + O ( \epsilon ^ { 3 } ).$$
Thus, from (16), (21) and (27) we get
$$\Phi, \, \left ( \frac { 1 } { 2 } \right ) \, \text{and} \, \left ( \frac { 1 } { 2 } \right ) \, \text{we ge} \\ M _ { m } & = \frac { 2 } { \pi } \text{$$
Step 4 . Proposition 4.3 implies that M j is a martingale. Because m is a bounded stopping time,
Thus, (22) and (28) lead to
$$\mathbf E [ M _ { m } - M _ { 0 } ] = 0.$$
$$& \xi, \, \left ( \frac { 1 } { \phi _ { m } ( w ) - U _ { m } } \right ) - \text{Im} \, \left ( \frac { 1 } { \phi ( w ) - U _ { 0 } } \right ) \right ] = O$$
(23) and (26) imply
$$\frac { 1 } { g _ { s } \circ \phi ( w ) - U ( s ) } = \frac { 1 } { \phi ( w ) - U _ { 0 } } + O ( \epsilon ) \quad \text{for} \quad \forall s \in [ 0, t _$$
By integrating this relation over [0 , t m ], Loewner equation and (23) show that
$$& \phi _ { m } ( w ) = \phi ( w ) + \frac { 2 } { \phi ( w ) - U _ { 0 } } \cdot t _ { m } + O ( \epsilon ^ { 3 } ). & & ( 3 0 ) \\ \quad \quad \quad$$
Let f ( u, v ) = 1 / u ( -v ). Using (23) and (30), we Taylor-expand f ( φ m ( w , U ) m ) -f ( φ w , U ( ) 0 ) with respect to φ m ( w ) -φ w ( ) and U m -U 0 , up to O /epsilon1 ( 3 ). Observing imaginary part of this Taylor expansion, from (29) and (30) we get
$$& \text{is layout expansion, from } \left ( \varepsilon \right ) \, \text{and } \left ( \varpi \right ) \, \text{we get} \\ & \text{Im} \, \left ( \frac { 1 } { ( \phi ( w ) - U _ { 0 } ) ^ { 2 } } \right ) \text{E} [ U _ { m } - U _ { 0 } ] + \text{Im} \, \left ( \frac { 1 } { ( \phi ( w ) - U _ { 0 } ) ^ { 3 } } \right ) \text{E} [ ( U _ { m } - U _ { 0 } ) ^ { 2 } - 2 t _ { m } ] = O ( \epsilon ^ { 3 } ). \ \ ( 3 1 ) \\ & \text{Now } \, \text{we consider two different choices of $w$ under the constraint $w$ $\, \in \ V$ such that}$$
Now, we consider two different choices of w under the constraint w ∈ V δ such that Im φ w ( ) ≥ 1 2 , | φ w ( ) | ≤ 3 . By the Koebe distortion theorem we can find w satisfying φ w ( ) -U 0 = i + O /epsilon1 ( 3 ). Then, (31) implies
$$\mathbf E [ ( U _ { m } - U _ { 0 } ) ^ { 2 } - 2 t _ { m } ] = O ( \epsilon ^ { 3 } ). \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
. \quad \quad. \quad. \quad. \quad. \quad. \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \pi \quad. \cdot \strut$$
Similarly, we can find w satisfying φ w ( ) -U 0 = e i π 3 + O /epsilon1 ( 3 ) and we get
$$\mathbf E [ U _ { m } - U _ { 0 } ] = O ( \epsilon ^ { 3 } ). \\ -$$
As in Subsection 2.3, let D t ( ) = D \ γ [0 , t ], φ t = g t ◦ φ and p t ( ) = φ -1 t ( i + U t ( )).
Lemma 5.4. Let T > 1 and ε > 0 , and, given a quadruplet ( D,a, b, p ) ∈ D , put ˜ T = sup { t ∈ [0 , T ] : | U t ( ) | < 1 //epsilon1 } . Then, there exists η 1 = η 1 ( T, /epsilon1 ) > 0 and r 1 = r 1 ( T, ε ) > 0 such that ( D t , γ ( ) ( ) t , b, p ( )) t ∈ D ( r , R, η 1 1 ) for all t < T ˜ .
Proof. Let g ∗ t ( z ) := g t ( z ) -U t ( ). Put Ψ( ) = ( z z -i ) / z ( + ). Define the conformal map i h t : D \ ψ γ ( [0 , t ]) → D by
$$h _ { t } ( z ) \coloneqq \Psi \circ g _ { t } ^ { * } \circ \Psi ^ { - 1 } ( z ). \\ \omega \times \omega ^ { + } \quad \varepsilon / \varepsilon \quad. \quad \mathbb { m } \cdot \quad. \quad. \quad.$$
Put ψ t ( z ) := h t ◦ ψ z ( ) so that ψ t : D t ( ) → D is a conformal map with ψ t ( γ t ( )) = -1 , ψ t ( b ) = 1 , ψ t ( p t ( )) = 0. Clearly ∂ ( D \ ψ γ ( [0 , t ])) is locally analytic at 1 and h t (1) = 1. On using the Loewner equation we infer that g ′ t ( z ) = 1 as z →∞ , which implies h ′ t (1) = 1. Now we can choose a positive η 1 < η/ 4 such that if t < T ˜ , then ψ γ ( [0 , t ]) does not intersect with B := { z ∈ C : | z - | 1 < 4 η 1 } . Thus, h t is analytically extended to B for t < T ˜ , so that in view of Koebe's 1/4 theorem h -1 t has an analytic extension in { z ∈ C : | z - | 1 < η 1 } for t < T ˜ . Since ψ -1 t = ψ -1 ◦ h -1 t and ψ -1 is analytic on B ψ , -1 t has an analytic extension in { z ∈ C : | z - | 1 < η 1 } for t < T ˜ . The existence of r 1 is deduced from Lemma 2.4. Thus the assertion of the lemma has been proved.
Proof of Theorem 5.1 . Having proved Lemma 5.3 it is easy to adapt the arguments given in [7]. Let D be as in the theorem and take R so that D ⊂ R D . Let r := rad ( p D ). From our hypothesis of local analyticity of ∂D at b , the function ψ has an analytic extension in a neighborhood of b . Thus, we can choose η > 0 such that ψ -1 is analytic in { z ∈ C : | z - | 1 < η } , hence ( D,a, b, p ) ∈ D ( r, R, η ).
Let T > 1 and /epsilon1 1 > 0 and put ˜ = sup T { t ∈ [0 , T ] : | U t ( ) | < 1 //epsilon1 1 } . Let /epsilon1 > 0 be small enough. Let m 0 = 0 and define m n inductively by
$$m _ { n } \coloneqq \min \{ j > m _ { n - 1 } \colon t _ { j } - t _ { m _ { n - 1 } } \geq \epsilon ^ { 2 } \text{ or } | U _ { j } - U _ { m _ { n - 1 } } | \geq \epsilon \}.$$
Let N := max { n ∈ N : t m n < T ˜ } . By Lemma 5.4, we can take some positive constants r 1 and η 1 such that ( D m n , γ m n , b, p m n ) ∈ D ( r , R, η 1 1 ) for any n ≤ N .
By the Markov property stated in Proposition 4.2, we find that γ ( t mn ) ( ) = · γ t ( m n + ) · is the same distribution as the time-reversal of the loop erasure of a natural random walk on G δ started at b δ and stopped on exiting D m n and conditioned to hit ∂D m n at γ m n . We apply Lemma 5.3 with ( D m n , γ m n , b, p m n ) for any n ≤ N . Then, we deduce from the fact stated at (3) that there exists δ 0 = δ 0 ( /epsilon1, /epsilon1 1 , T ) > 0 such that if δ < δ 0 , then for any n ≤ N
and
∣ ∣ The rest of proof of Theorem 5.1 is the proof that U t ( ) weakly converges to √ 2 B t ( ) uniformly on [0 , T ] as δ → 0, where B t ( ) is a one-dimensional standard Brownian motion with B (0) = 0. This proof follows from the above estimate and the Skorokhod embedding theorem as in [3] and [7]. (See Subsection 3.3 in [3] and Corollary 4.3 in [7].)
$$& \text{and} \\ & \text{E} \left [ ( U _ { m _ { n + 1 } } - U _ { m _ { n } } ) ^ { 2 } \, \Big | \, \gamma [ 0, m _ { n } ] \right ] = \text{E} \left [ 2 ( t _ { m _ { n + 1 } } - t _ { m _ { n } } ) \, \Big | \, \gamma [ 0, m _ { n } ] \right ] + O ( \epsilon ^ { 3 } ). \\ & \text{The rest of proof of Theorem 5.1 is the proof that } U ( t ) \, \text{weakly converges to } \sqrt { 2 } B ( t ) \\ \text{inform, on } [ 0 \, T ] \, \circ \delta \rightharpoonup \, \text{where} \, R ( t ) \, \text{is an explanation at and} \, \text{random} \, \text{motion}$$
$$\text{e exos } \upsilon _ { 0 } - \upsilon _ { 0 } \text{(c,c_{1},1)} \succ \text{such as } \partial \omega \in \upsilon \succ \upsilon _ { 0 }, \\ \text{$\mathbf E\left[ U_{m_{n+1}} - U_{m_{n}} \, \Big | \, \gamma[0,m_{n}} \right ] = O(\epsilon^{3})$,} \\ \\ - U_{m_{n}})^{2} \, \Big | \, \gamma[0,m_{n}} \right ] \right ] = \mathbf E \Big [ 2 ( t _ { m _ { n + 1 } } - t _ { m _ { n } } ) \, \Big | \, \gamma[0,m_{n}^{3}$}.$$
## 5.2 Convergence with respect to the metric d U
Now, we assume that there exists an invariant measure π for a natural random walk S ( ) · on G such that 0 < π v ( ) < ∞ for any v ∈ V . Let p u, v ( ) be the transition probability for S ( ). · We consider the dual walk S ∗ ( ). · The transition probability of S ∗ ( ), · denoted by p ∗ ( u, v ), is given by
$$p ^ { * } ( u, v ) \coloneqq \frac { \pi ( v ) } { \pi ( u ) } p ( v, u ).$$
Then, the dual walk S ∗ ( ) · is a natural random walk on some other planar irreducible graph. As in the case of S ( ), we define ( · S ∗ ) x δ , (Γ ) ∗ a,b δ , ( γ ∗ ) a,b δ corresponding to S ∗ ( ). · The following lemma is a relation between the time-reversal and the dual walk.
Proposition 5.5. Suppose that there exists an invariant measure π for a natural random walk S ( ) · on G such that 0 < π v ( ) < ∞ for any v ∈ V . Then, the time-reversal of Γ a,b δ has the same distribution as (Γ ) ∗ b,a δ . Similarly, the time-reversal of γ a,b δ has the same distribution as ( γ ∗ ) b,a δ .
Proof. The first assertion immediately follows from the definition of the dual walk and the conditional probability. In addition to the first assertion, applying Proposition 4.1,
$$( \gamma _ { \delta } ^ { a, b } ) ^ { - } = L [ \Gamma _ { \delta } ^ { a, b } ] ^ { - } \stackrel { d } { = } L [ ( \Gamma _ { \delta } ^ { a, b } ) ^ { - } ] \stackrel { d } { = } L [ ( \Gamma ^ { * } ) _ { \delta } ^ { b, a } ] = ( \gamma ^ { * } ) _ { \delta } ^ { b, a },$$
where = means the same distribution. Hence, we get the second assertion. d
Let η a,b be a chordal SLE 2 curve in D from a to b . Recall the metric d U defined by (6) in Subsection 2.4.
Theorem 5.6. Suppose that there exists an invariant measure π for a natural random walk S ( ) · on G such that 0 < π v ( ) < ∞ for any v ∈ V and S x δ and ( S ∗ ) x δ satisfy invariance principle. Let D be a bounded simply connected domain and a, b ∈ ∂D be distinct points. Suppose that ∂D is locally analytic at a and b . Then, γ a,b δ converges weakly to η a,b as δ → 0 with respect to the metric d U .
Proof. Let φ : D → H be a conformal map with φ a ( ) = 0 , φ ( b ) = ∞ and Let φ -: D → H be a conformal map with φ b ( ) = 0 , φ ( a ) = ∞ . Theorem 5.1 implies that φ -◦ ( γ a,b δ ) -converges weakly to a chordal SLE 2 with respect to the driving function. Because we also assume that ( S ∗ ) x δ satisfy invariance principle, Theorem 5.1 implies that φ ◦ (( γ ∗ ) b,a δ ) -converges weakly to a chordal SLE 2 with respect to the driving function. By Proposition 5.5, γ a,b δ is the same distribution as (( γ ∗ ) b,a δ ) -. Hence, φ ◦ γ a,b δ converges weakly to a chordal SLE 2 with respect to the driving function. Therefore, Theorem 3.6 completes the proof.
## 6 Estimates of hitting probabilities of the random walk started at a boundary point
In this section we prove Lemma 5.2. To this end it is convenient to work in the disc D instead of H . Let D /subsetnoteql C be a simply connected domain and a, b be distinct points on
∂D . Let φ : D → H be a conformal map with φ a ( ) = 0 , φ ( b ) = ∞ . Let p := φ -1 ( ). i Put Ψ( ) = ( z z -1) / z ( + 1) and ψ = Ψ ◦ φ so that ψ is a conformal map of D onto D with ψ a ( ) = -1 , ψ ( b ) = 1 , ψ ( p ) = 0. Let S b δ be a natural random walk on G δ started at b δ , where b δ is a point of ∂ in V δ ( D ) close to b .
Recall the class D ( r, R, η 0 ), which is the collection of all quadruplets ( D,a, b, p ) such that rad ( p D ) ≥ r and D ⊂ R D and ψ -1 has analytic extension in { z ∈ C : | z - | 1 < η 0 } . Throughout this section we consider the constants r , R and η 0 to be fixed and write D for D ( r, R, η 0 ); also suppose that S x δ satisfies invariance principle.
For ( D,a, b, p ) ∈ D and η < η 0 ∧ 1 2 put
$$U = U _ { \eta } = \{ z \in D \, \colon | \psi ( z ) - 1 | < \eta \}$$
and for any number α from the open interval (0 , 1 / 2),
$$J _ { \alpha } = \{ z \in \partial U \, \colon \text{dist} ( \psi ( z ), \partial \mathbb { D } ) < \alpha \eta, z \in D \}.$$
Proposition 6.1. Let U = U η and J α be as described above. Then for any ε > 0 there exists δ 0 = ( δ ε, η ) > 0 such that for all positive δ < δ , α < δ 0 0 and for all ( D,a, b, p ) ∈ D ,
$$\mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \in J _ { \alpha } \ | \ S _ { \delta } ^ { b } ( \tau _ { U } ) \in D ) < \epsilon,$$
Here δ 0 may depend on the graph ( V, E ) .
Remark. It is only for this proposition that we need the condition of the analyticity about b . Without that condition the estimate of the proposition is obtained by Uchiyama [9].
Proof. This proof is an adaptation of a part of the arguments given in [9]. Put
$$C = \{ z \in \partial D \, \colon \text{Im} \, \psi ( z ) > 0, | \psi ( z ) - 1 | < \eta / 3 \},$$
and
$$B = \{ z \in \mathbb { C } \, \colon | \psi ( z ) - 1 | < \eta / 3 \} \, \sum \overline { U }, \ \Omega = B \cup C \cup U.$$
Let
$$C _ { \delta } = \{ v \in V _ { \delta } ( D ) \, \colon [ u, v ] \cap C \neq \emptyset \text{ for some } u \in V _ { \delta } ( B ) \}, \\ \cdot \quad \cdot \quad \omega \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdots$$
/negationslash and v ∗ be a vertex in C δ such that Im ψ v ( ∗ ) is closest to η/ 6 among vertexes of C δ .
Let L denote the last time when the walk S v ∗ δ in Ω killed when it crosses the boundary ∂ Ω exits B :
/negationslash
$$\intertext { s u e x s } \intercolon & L = \begin{cases} 1 + \max \{ 0 \leq n < \tau _ { \Omega } \, \colon S _ { \delta } ^ { v ^ { * } } ( n ) \in B \} & \text{if} \ \ S _ { \delta } ^ { v ^ { * } } ( \tau _ { \Omega } ) \not \in \partial B \\ \infty & \text{if} \ \ S _ { \delta } ^ { v ^ { * } } ( \tau _ { \Omega } ) \in \partial B \end{cases}.$$
.
We write T = τ U . Putting J + α = J α ∩ H we compute q = P ( S v ∗ δ ( τ Ω ) ∈ J + α ), the probability that the walk exits Ω through J + α , which we rewrite as
$$q = \mathbf P ( S _ { \delta } ^ { v ^ { * } } ( T ) \circ \theta _ { L } \in J _ { \alpha } ^ { + }, L < \tau _ { \Omega } ),$$
where the shift operator θ L acts on T as well as on S v ∗ δ . By employing the strong Markov property
$$q & = \sum _ { n = 0 } ^ { \infty } \sum _ { y \in C _ { \delta } } \mathbf P ( S _ { \delta } ^ { v ^ { * } } ( T ) \circ \theta _ { n } \in J _ { \alpha } ^ { + }, L = n, S _ { \delta } ^ { v ^ { * } } ( n ) = y ) \\ & = \sum _ { n = 0 } ^ { \infty } \sum _ { y \in C _ { \delta } } \mathbf P ( S _ { \delta } ^ { v ^ { * } } ( T ) \circ \theta _ { n } \in J _ { \alpha } ^ { + }, S _ { \delta } ^ { v ^ { * } } ( n ) = y ) \\ & = \sum _ { n = 0 } ^ { \infty } \sum _ { y \in C _ { \delta } } \mathbf P ( S _ { \delta } ^ { v ^ { * } } ( n ) = y ) \mathbf P ( S _ { \delta } ^ { y } ( T ) \in J _ { \alpha } ^ { + } ) \\ \text{occurrence of the event } S _ { \delta } ^ { y } ( T ) \in J _ { \alpha } ^ { + } \text{ for } y \in C _ { \delta } \text{ entails } S _ { \delta } ^ { y } ( T ) \in \ D$$
The occurrence of the event S y δ ( T ) ∈ J + α for y ∈ C δ entails S y δ ( T ) ∈ D , so that P ( S y δ ( T ) ∈ J + α ) = P ( S y δ ( T ) ∈ J + α , S y δ ( T ) ∈ D ). Hence, bringing in the conditional probability
$$p ( y ) = \mathbf P ( S _ { \delta } ^ { y } ( T ) \in J _ { \alpha } ^ { + } \ | \ S _ { \delta } ^ { y } ( T ) \in D ),$$
we infer that where G Ω stands for the Green function of the walk killed on exiting Ω. We have
$$q = \sum _ { y \in C _ { \delta } } G _ { \Omega } ( v ^ { * }, y ) \mathbf P ( S _ { \delta } ^ { y } ( T ) \in D ) p ( y ), \\ r \, & \text{the Green function of the walk killed on exitin} \sigma$$
$$p ( y ) \geq p ( b ), \ \ y \in C _ { \delta },$$
for, if γ b denote a path joining b δ with J + α in V δ ( U ), then the walk starting at y ∈ C δ and conditioned on the event S y δ ( T ) ∈ D must hit γ b ∪ J + α before existing U . Observing the identity we finally obtain
$$\begin{array} { c } \kappa \cdot \, \cdot & \cdot & \cdot & \cdot & \cdot & \cdot \\ \sum _ { y \in C _ { \delta } } G _ { \Omega } ( v ^ { * }, y ) \mathbf P ( S _ { \delta } ^ { y } ( T ) \in D ) = \mathbf P ( S _ { \delta } ^ { v ^ { * } } ( \tau _ { \Omega } ) \in D ), \\ \text{tain} \end{array}$$
$$q \geq p ( b ) \mathbf P ( S _ { \delta } ^ { v ^ { * } } ( \tau _ { \Omega } ) \in D ). \\ \cdot \dots \ \mathbf \mathbf \mathbf u / c w ^ { * } / \cdots \ \mathbf \mathbf \mathbf u \cdots \ \mathbf \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots \ \mathbf u \cdots$$
This concludes p b ( ) < /epsilon1/ 2 since P ( S v ∗ δ ( τ Ω ) ∈ D > ) 1 / 3 and q < /epsilon1/ 6 for all sufficiently small δ and α . Let J -α = J α \ J + α . On defining C with Im ψ z ( ) ≤ 0 in place of Im ψ z ( ) > 0 we repeat the same argument to show that P ( S b δ ( T ) ∈ J -α | S b δ ( T ) ∈ D < /epsilon1/ ) 2.
Lemma 6.2. Let A := φ -1 ([ -1 1]) , . For any /epsilon1 > 0 , there exists δ 0 = δ 0 ( /epsilon1, η ) > 0 such that the following holds. Let ( D,a, b, p ) ∈ D . Then, for all 0 < δ < δ 0 and 0 < α < δ 0 ,
$$\mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \in J _ { \alpha } \ | \ S _ { \delta } ^ { b } ( \tau _ { D } ) \in A ) < \epsilon.$$
Proof. By the definition of the conditional probability and the strong Markov property,
/negationslash
/negationslash
/negationslash
$$\cdots \cdots _ { J } \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \ddots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \\ \frac { \mathsf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \in J _ { \alpha } \ | \ S _ { \delta } ^ { b } ( \tau _ { D } ) \in A ) } { \mathsf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \not \in J _ { \alpha } \ | \ S _ { \delta } ^ { b } ( \tau _ { D } ) \in A ) } } = \frac { \mathsf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) = y ) \mathsf P ( S _ { \delta } ^ { y } ( \tau _ { D } ) \in A ) } { \sum _ { y \not \in J _ { \alpha } } \mathsf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) = y ) \mathsf P ( S _ { \delta } ^ { y } ( \tau _ { D } ) \in A ) } }.$$
Because we assume invariance principle, the hitting probability P ( S y δ ( τ D ) ∈ A ) can be approximated by the same probability for a Brownian motion. Because the hitting probability for a Brownian motion is conformal invariant, we can calculate the hitting probability on the upper half plane instead of D . Therefore, we find that there exists a universal constant C such that for sufficiently small δ ,
/negationslash
$$\frac { \sup _ { y \in J _ { \alpha } } \mathbf P ( S _ { \delta } ^ { y } ( \tau _ { D } ) \in A ) } { \inf _ { y \not \in J _ { \alpha } } \mathbf P ( S _ { \delta } ^ { y } ( \tau _ { D } ) \in A ) } \leq C.$$
Thus, we obtain
Because
$$\begin{array} { c } \text{we ovavant} \\ \frac { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \in J _ { \alpha } \ | \ S _ { \delta } ^ { b } ( \tau _ { D } ) \in A ) } { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \not \in J _ { \alpha } \ | \ S _ { \delta } ^ { b } ( \tau _ { D } ) \in A ) } \leq C \frac { \sum _ { y \in J _ { \alpha } } \mathbf P ( S ^ { b } ( \tau _ { U } ) = y ) } { \sum _ { y \not \in J _ { \alpha } } \mathbf P ( S ^ { b } ( \tau _ { U } ) = y ) }. \\ \text{se} \\ \sum _ { \dots \sim r } \mathbf P ( S ^ { b } ( \tau _ { U } ) = y ) \quad \mathbf P ( S _ { \varepsilon } ^ { b } ( \tau _ { U } ) \in J _ { \varepsilon } \ | \ S _ { \varepsilon } ^ { b } ( \tau _ { U } ) \in D ) \end{array}$$
/negationslash
Proposition 6.1 completes the proof.
/negationslash
/negationslash
$$\sum _ { \substack { y \in J _ { \alpha } \\ \overline { \sum } _ { y \not \in J _ { \alpha } } } } \mathbf P ( S ^ { b } ( \tau _ { U } ) = y ) & = \frac { \mathbf P ( S ^ { b } _ { \delta } ( \tau _ { U } ) \in J _ { \alpha } \ | \ S ^ { b } _ { \delta } ( \tau _ { U } ) \in D ) } { \mathbf P ( S ^ { b } _ { \delta } ( \tau _ { U } ) \not \in J _ { \alpha } \ | \ S ^ { b } _ { \delta } ( \tau _ { U } ) \in D ) }, \\ \intertext { \text{tion} } 6. 1 \text{ completes the proof.} } \left ( \begin{array} { c } \wedge \end{array} \right ) \right ) \right ) \end{array} \right )$$
/negationslash
Lemma 6.3. For any /epsilon1 > 0 , there exists δ 0 = δ 0 ( /epsilon1, η ) > 0 such that the following holds. Let ( D,a, b, p ) ∈ D . Then, for all 0 < δ < δ 0 and 0 < α < δ 0 ,
$$\mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \in J _ { \alpha } \ | \ S _ { \delta } ^ { b } ( \tau _ { D } ) = a _ { \delta } ) < \epsilon,$$
where a δ is a point of ∂ out V δ ( D ) close to a .
Proof. By the definition of the conditional probability,
/negationslash
/negationslash
/negationslash
$$\begin{array} { c c c c c } \cdot & \cdot & \cdot \\ & \frac { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \in J _ { \alpha } \, | \, S _ { \delta } ^ { b } ( \tau _ { D } ) = a _ { \delta } ) } { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \not \in J _ { \alpha } \, | \, S _ { \delta } ^ { b } ( \tau _ { D } ) = a _ { \delta } ) } & = \frac { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \in J _ { \alpha }, S _ { \delta } ^ { b } ( \tau _ { D } ) = a _ { \delta } ) } { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \not \in J _ { \alpha }, S _ { \delta } ^ { b } ( \tau _ { D } ) = a _ { \delta } ) } \\ & = \frac { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \in J _ { \alpha }, S _ { \delta } ^ { b } ( \tau _ { D } ) = a _ { \delta } \, | \, S _ { \delta } ^ { b } ( \tau _ { D } ) \in A ) } { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \not \in J _ { \alpha }, S _ { \delta } ^ { b } ( \tau _ { D } ) = a _ { \delta } \, | \, S _ { \delta } ^ { b } ( \tau _ { D } ) \in A ) }. \\ \end{array}$$
Since the random walk conditioned on exiting D through A is Markovian, the right-hand side above may be written as
/negationslash
By Lemma 5.8. in [10], there exists a universal constant C such that for sufficiently small δ ,
$$e \, \text{above may be written as} \\ \frac { \sum _ { y \in J _ { \alpha } } \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) = y | S _ { \delta } ^ { b } ( \tau _ { D } ) \in A ) \mathbf P ( S _ { \delta } ^ { y } ( \tau _ { D } ) = a _ { \delta } | S _ { \delta } ^ { y } ( \tau _ { D } ) \in A ) } { \sum _ { y \not \in J _ { \alpha } } \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) = y | S _ { \delta } ^ { b } ( \tau _ { D } ) \in A ) \mathbf P ( S _ { \delta } ^ { y } ( \tau _ { D } ) = a _ { \delta } | S _ { \delta } ^ { y } ( \tau _ { D } ) \in A ) }. \\ \text{Lemma} \, 5. 8. \text{ in } [10 ], \, \text{there exists a universal constant $C$ such that for sufficiently sm}$$
/negationslash
Hence, we obtain
$$\frac { \sup _ { y \in J _ { \alpha } } \mathbf P ( S _ { \delta } ^ { y } ( \tau _ { D } ) = a _ { \delta } \ | \ S _ { \delta } ^ { y } ( \tau _ { D } ) \in A ) } { \inf _ { y \not \in J _ { \alpha } } \mathbf P ( S _ { \delta } ^ { y } ( \tau _ { D } ) = a _ { \delta } \ | \ S _ { \delta } ^ { y } ( \tau _ { D } ) \in A ) } \leq C.$$
$$\frac { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \in J _ { \alpha } \ | \ S _ { \delta } ^ { b } ( \tau _ { D } ) = a _ { \delta } ) } { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \not \in J _ { \alpha } \ | \ S _ { \delta } ^ { b } ( \tau _ { D } ) = a _ { \delta } ) } \leq C \frac { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \in J _ { \alpha } \ | \ S _ { \delta } ^ { b } ( \tau _ { D } ) \in A ) } { \mathbf P ( S _ { \delta } ^ { b } ( \tau _ { U } ) \not \in J _ { \alpha } \ | \ S _ { \delta } ^ { b } ( \tau _ { D } ) \in A ) } \\ \nu \quad \ e \quad \tau \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ t \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ a \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \ \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad $\ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad \ e \quad { \ e$$
/negationslash
/negationslash
Therefore, Lemma 6.2 completes the proof.
Proof of Lemma 5.2. By the mapping Ψ( ) = ( z z -i ) / z ( + ), the half disc i B + (2 λ ) := B U , ( 0 2 λ ) ∩ H is mapped to a small disc of radius ∼ 1 / 2 λ and centered at 1. For 1 / 2 λ < η 0 , (12) follows from applying Lemma 6.2 with this small disc in place of U η , the little discrepancy between them making no harm. If diam( φ γ ( [0 , j ])) < 1, the difference between B + (2 λ ) and g t j ( B + (2 λ )) is insignificant for sufficiently large λ . Hence, we also have (13) by applying Lemma 6.3 with ( D , γ , b, p j j j ), which is legitimate because of Lemma 5.4.
Acknowledgments. I would like to express my sincere gratitude to Professor K. Uchiyama: apart from his many invaluable and helpful comments on this paper, I am especially grateful for his teaching me the essential idea for the proof of Lemma 2.4 as well as for showing his manuscript prior to its publication and letting me include a result therein as Proposition 6.1.
## References
- [1] G. Lawler, Conformally invariant processes in the plane, Mathematical Surveys and monographs 114 , American Mathematical Society, Providence, RI, 2005.
- [2] G. Lawler, Perplexing problems in probability, Loop-erased random walks, Birkh¨user Boston, Boston, 1999, 197-217. a
- [3] G. Lawler, O. Schramm and W. Werner, Conformal invariance of planar loop-erased random walks and uniform spanning trees, Ann. Probab. 32 , 1B (2004), 939-995.
- [4] Ch. Pommerenke, Boundary Behaviour of Conformal maps, Grundlehren der mathematischen Wissenschaften 299 , Springer-Verlag, Berlin, 1992.
- [5] S. Rohde and O. Schramm, Basic properties of SLE, Ann. Math. 161 (2005), 879-920.
- [6] O. Schramm, Scaling limits of loop erased random walks and uniform spanning trees, Israel J. Math. 118 (2000), 221-288.
- [7] O. Schramm and S. Sheffield, The harmonic explorer and its convergence to SLE(4), Ann. Probab. 33 (2005), 2127-2148.
- [8] S. Sheffield and N. Sun, Strong convergence from Loewner driving convergence, Ann. Probab. 40 , 2 (2012), 578-610.
- [9] K. Uchiyama, Uniformity of Invariance Principle and Excursions for RW's on Planar Graphs, preprint.
- [10] A. Yadin and A. Yehudayoff, Loop-erased random walk and Poisson kernel on planar graphs, Ann. Probab. 39 , 4 (2011), 1243-1285.
- [11] D. Zhan, Reversibility of chordal SLE, Ann. probab. 36 4 , (2008), 1472-1494.
- [12] D. Zhan, The scaling limits of planar LERW in finitely connected domains, Ann. Probab. 36 , 2 (2008), 467-529.
| null |
[
"Hiroyuki Suzuki"
] |
2014-08-05T02:15:44+00:00
|
2014-08-05T02:15:44+00:00
|
[
"math.PR",
"60F17, 60J67, 82B41"
] |
Convergence of loop erased random walks on a planar graph to a chordal SLE(2) curve
|
In this paper we consider the natural random walk on a planar graph and scale
it by a small positive number $\delta$. Given a simply connected domain $D$ and
its two boundary points $a$ and $b$, we start the scaled walk at a vertex of
the graph nearby $a$ and condition it on its exiting $D$ through a vertex
nearby $b$, and prove that the loop erasure of the conditioned walk converges,
as $\delta \downarrow 0$, to the chordal SLE$_{2}$ that connects $a$ and $b$ in
$D$, provided that an invariance principle is valid for both the random walk
and the dual walk of it.
|
1408.0851v1
|
## Photometric study of the pulsating, eclipsing binary OO Dra
X.B. Zhang , L.C. Deng , J.F. Tian , K. Wang , J.J. Sun , Q.L. Liu , H.Q. Xin , Q. 1 1 1 1 2 2 2 Zhou , Z.Z. Yan 2 1 3 , , Z.Q. Luo , C.Q. Luo 3 1
1, Key Laboratory of Optical Astronomy, National Astronomical Observatories, Chinese Academy of Sciences, Beijing, 100012, China
2, Purple Mountain Observatory, Chinese Academy of Sciences, Nanjing, 210008, China
3, Department of Physics, China West Normal University, Nanchong 637002, China
Received
;
accepted
## ABSTRACT
We present a comprehensive photometric study of the pulsating, eclipsing binary OO Dra. Simultaneous B- and V-band photometry of the star was carried out on 14 nights. Revised orbital period and a new ephemeris were derived from the data. The first photometric solution of the binary system and the physical parameters of the component stars are determined. It reveals that OO Dra could be a detached system with the less-massive secondary component nearly filling in its Roche lobe. By subtracting the eclipsing light changes from the data, we obtained the intrinsic pulsating light curves of the hotter and massive primary component. Frequency analysis of the residuals light yields two confident pulsation modes in both B- and V-band data with the dominant frequency detected at 41.865 c/d. A brief discussion concerning the evolutionary status and the pulsation nature of the binary system is finally given.
Subject headings: stars: variables: oscillation - stars: binary, eclipsing - stars: individual (OO Dra)
## 1. Introduction
Eclipsing binaries with pulsating components are important objects to understand stellar structure and evolution. Investigating the frequencies of oscillations, known as asteroseismology, helps us to identify the physical processes behind the pulsating nature and so the stellar interiors. On the other hand, modeling the light and radial velocity curves of a eclipsing binary can precisely determine the physical parameters of the components. This enables one to definitively identify the pulsation modes and compare the results with theoretical models in detail. The study of pulsating stars in eclipsing binaries offers new and strict constraints for stellar theories, helps people to understand the influences of tidal forces and mass transfer among the interacting binaries.
The present study concentrates on the newly discovered pulsating eclipsing binary system OO Dra. The star was first found to be an eclipsing binary with an orbital period of 1.23837 days by Biyalieva & Khruslov (2007). The pulsation nature of the star was later discovered by Dimitrov et al. (2008) through follow-up observations. With the out-of-eclipse photometric data, they had made a preliminary frequency analysis and detected a main periodicity about 37 c/d. A linear ephemeris of the binary system was given as HJD MInI ( ) = 2451403 832(4) + 1 2383832(8) . . × E . In addition, Dimitrov et al. (2008) had contributed a spectroscopy of the binary system. Twelve radial-velocity measurements were given. The spectral type of the primary star in the binary system was identified to be A3 V-IV, with an effective temperature of 8500 K and logg = 4 0. .
As a contribution to the on-going program of searching for and studying of pulsating eclipsing binaries (Zhang et al. 2009, 2013, Liu et al. 2012), we have performed a new photometry of OO Dra and collected sufficient photometric data. We present the results of the observations in this paper, as well as a comprehensive study of the binary system and its intrinsic pulsation.
## 2. Photometric observations
The eclipsing binary OO Dra was observed over 3 weeks in March-April 2014. Time-series CCD photometry were carried out with the 50cm binocular telescope (Deng et al. 2013, Zhang et al. 2014) at the Qinghai Station of the Purple Mountain Observatory, Chinese Academy of Sciences. Data were collected with two Andor 2k × 2k CCD, one with a standard Johnson V and the other with a B filter equipped, so that a simultaneous two-color photometry is achieved. Each camera provides a field of view of 20' × 20', with an image scale of about 0'.59/pixel. Useful data were obtained in 90 hours during 14 nights from March 19 to April 12, 2014. A total of 1522 frames in B and 3428 in V filter were collected. The journal of observations is presented in Table 1.
The preliminary processing (bias, dark subtraction and flat fielding correction) of the CCD frames was performed with the standard routines of CCDPROC in the IRAF package. Photometry was extracted by using the DAOPHOT II package. Following Dimitrov et al. (2008), we used the stars BD+75 453 (=SAO 7402) and BD+75 415 (=GSC4050-1520) as the comparison and check stars. The magnitude difference between the two stars was confirmed to be stable within 0.015 mag during the observation. The differential magnitudes of OO Dra were then extracted. The time of each measurement was converted to Heliocentric Julian date. As an example, we plot in the upper panels of Figure 1 a part of the time-series light curves . The left one represents the B- and V-band measurements obtained on March 24. Those shown in the right one were derived on April 6. Where the short-term pulsational light variations in addition to the eclipsing light changes can be clearly seen. It confirms the discovery of Dimitrov et al. (2008).
Table 1: Observation log.
| Date | Start (HJD2456700+) | length (hrs) | Frames (B) | Frames (V) |
|---------------|-----------------------|----------------|--------------|--------------|
| 2014 March 22 | 39.134 | 3.8 | 79 | 154 |
| 2014 March 23 | 40.049 | 7.8 | 139 | 386 |
| 2014 March 24 | 41.042 | 8 | 141 | 275 |
| 2014 March 25 | 42.024 | 8.1 | 144 | 256 |
| 2014 March 27 | 44.044 | 7.8 | 137 | 312 |
| 2014 March 31 | 48.069 | 6.5 | 98 | 311 |
| 2014 April 01 | 49.055 | 6.8 | 120 | 235 |
| 2014 April 02 | 50.095 | 5.9 | 104 | 204 |
| 2014 April 03 | 51.048 | 6.9 | 124 | 277 |
| 2014 April 04 | 52.054 | 6.7 | 68 | 138 |
| 2014 April 05 | 53.077 | 6.1 | 98 | 225 |
| 2014 April 06 | 54.104 | 5.4 | 95 | 186 |
| 2014 April 07 | 55.03 | 3.4 | 60 | 117 |
| 2014 April 12 | 60.057 | 6.4 | 115 | 352 |
Fig. 1.- The upper panels represent a segment of the original real-time B- and V-band light curves of OO Dra observed on March 24 and April 6, 2014. The lowers are those with short-term pulsations subtracted.
## 3. The binary system
A total of 3 completely covered eclipses, including 2 primary and 1 secondary ones were recorded in our data. By using a least-squares parabolic fitting method, the epochs of these light minima were determined as given in Table 2. From the literatures (Lampens et al. 2010, Diethelm 2012, Hˇbscher et al. u 2013), we collected 4 minima times for the star. Based on these data, the orbital period of the eclipsing binary was analyzed with the classical O-C method. It yields an orbital period of 1.238378 days for the binary system and the following linear ephemeris is given:
$$H J D ( M i n I ) = 2 4 5 6 7 4 1. 2 6 0 7 ( 2 ) + 1. 2 3 8 3 7 8 ( 2 ) \times E$$
By using the newly derived linear ephemeris, phases of all the measurements were computed. The folded light curves due to eclipsing were formed as shown in Figure 3. The general feature of the light curves are typical of Algol-type eclipsing binaries with nearly total eclipses, suggesting a large inclination close to 90 o for the binary system. The depth of the primary eclipse was measured to be 0.46 mag. in B and 0.44 mag. in V. That of the secondary minima turned out to be 0.11 mag. in B, and 0.14 mag. in V, respectively.
We made use of the Wilson-Devinney method (Wilson & Devinney 1971, Wilson 1979) to obtain the first photometric solution for the binary system. The available B- and V-band light curves were simultaneously synthesized by applying the 2003 version of the WD code with the Kurucz atmospheres (Wilson 1990, Kallrath et al. 1998). A nonlinear limb-darkening law with the logarithmic form was applied in the light-curve synthesis. Considering of the short orbital period and the evolutionary status of classical Algols such as OO Dra, a circular orbit ( e = 0) and the synchronous rotation ( F 2 = F 2 = 1 0) assumption . were accepted. The temperature of the massive (and usually luminous ) primary star was set at 8500 K adopted from Dimitrov et al. (2008). The initial bolometric ( X 1 , X 2 , Y 1 , Y 2 )
Fig. 2.- Observed B- and V-band light curves of OO Dra and the theoretical synthesis
and monochromatic (x1, y1, x2, y2) limb-darkening coefficients of the components were taken from Van Hamme (1993). The gravity darkening exponents were set to be g 1 = 1 0 . for the primary and g 2 = 0 32 for the secondary component from Lucy (1967) according to . their temperatures. The bolometric albedos were taken as A 1 = 1 0 and . A 2 = 0 5 following . Rucinski (1969).
The adjustable parameters in computing the photometric solutions are the orbital inclination (i), phase shift, mass ratio (q). surface temperature of the secondary ( T 2 ), dimensionless luminosity of the primary ( L 1 ) and the potentials of the two components (Ω , 1 Ω ). 2 Since OO Dra is a single-line spectroscopic binary (Dimitrov et al. 2008), there is no mass ratio available from the radial-velocity solution. To search for an approximate mass ratio, we made a set of test solutions at the outset. The test solutions were computed at a series of assumed mass ratios with values ranging from 0.02 to 1.0. The mass ratio was photometrically determined to be 0.097 ± 0.001. The final best-fitting solution is given in Table 3. The synthesis of the observed B- and V-band light curves as well as the (O-C) residuals are shown in Figure 2.
The photometric solution reveals a detached configuration with the secondary nearly filling in its Roche lobe. The filling factor, defined as fraction of the stellar radius to that of the critical Roche lobe, of the two components turned out to be 55.8% and 97.5%. This suggests that the binary could also be in semi-detached. To check this possibility, we have modeled the light curves with Mode 5 (semi-detached model with the star 2 filling its Roche lobe). It yields a converged solution at q=0.075, which also gives a satisfied fitting to the observations. But the sum of residuals, Σ( O -C ) 2 , is about 10% larger than that of the solution with Mode 2 (detached model). We therefore adopted the results with detached configuration.
Based on the photometric solution, we have carried out a spectroscopic solution for
Table 2: Times of light minima of OO Dra and the residuals in respect to the derived ephemeris
| HJD | Epoch | (O-C) | Ref |
|---------------|---------|---------|---------------------------|
| (HJD2450000+) | | days | |
| 4861.4032 | -1518.0 | -0.0001 | Lampens et al. 2010 |
| 5944.9791 | -643.0 | -0.0047 | Diethelm 2012 |
| 6036.6300 | -569.0 | 0.0063 | Diethelm 2012 |
| 6061.3904 | -549.0 | -0.0009 | Hˇbscher u & Lehmann 2013 |
| 6741.2606 | 0.0 | -0.0001 | Present study |
| 6751.1675 | 8.0 | -0.0002 | Present study |
| 6754.2633 | 10.5 | -0.0004 | Present study |
Fig. 3.- Radial-velocity curve of OO Dra and the synthesis.
Table 3: Light-curve and radial velocity solutions for the binary OO Dra
| Parameter | Best-Fit Value | Formal Error |
|---------------------------|------------------|----------------|
| e* | 0 | |
| F 1 = F 2 * | 1.0 | |
| g 1 , g 2 * | 1.0, 0.32 | |
| A 1 , A 2 * | 1.0, 0.50 | |
| X 1 ,bolo , X 2 ,bolo * | 0.658, 0.639 | |
| Y 1 ,bolo , Y 2 ,bolo * | 0.139, 0.241 | |
| x 1 ,B , x 2 .B * | 0.785, 0.806 | |
| y 1 ,B , y 2 .B * | 0.301, 0.232 | |
| x 1 ,V , x 2 .V * | 0.676, 0.710 | |
| y 1 ,V , y 2 .V * | 0.276, 0.275 | |
| T 1 (K)* | 8500 | |
| i (deg) | 85.66 | ± 0.14 |
| T 2 (K) | 6452 | ± 8 |
| Ω 1 | 3.211 | ± 0.007 |
| Ω 2 | 1.986 | ± 0.002 |
| q= M 2 /M 1 | 0.097 | ± 0.002 |
| L 1 ( L 1 + L 2 ) B | 0.922 | ± 0.001 |
| L 1 ( L 1 + L 2 ) V | 0.899 | ± 0.001 |
| P orb (days) | 1.238378 | 0.000002 |
| a ( R /circledot ) | 6.27 | ± 0.35 |
| γ (km/s) | -55.92 | ± 0.02 |
| M 1 ( M /circledot ) | 1.97 | ± 0.25 |
| M 2 ( M /circledot ) | 0.19 | ± 0.03 |
| R 1 ( R /circledot ) | 2.04 | ± 0.09 |
| R 2 ( R /circledot ) | 1.17 | ± 0.05 |
| log ( L 1 /L /circledot ) | 1.27 | ± 0.03 |
| log ( L 2 /L /circledot ) | 0.31 | ± 0.01 |
*: assumed
OO Dra with the data published by Dimitrov et al. (2008). As it was indicated by the photometric solution that the less-massive secondary star contributes about 10 percent luminosity of the binary system, it is hard to be seen in the spectra. Radial velocities for only the primary component could be measured. By using the newly derived ephemeris, the radial velocity curve of the massive primary component was reformed. Two data points detected at HJD2454600.4282 and HJD2454600.4516 were not included into the analysis for the large scatterings. As a result, the radial-velocity curve modeling gives a separation between the two components of a = 6 27 . ± 0 35 . R /circledot with a systematic velocity of -55.92 km/s for the binary. With this, the absolute parameters of the components were then calculated as given in Table 3. The synthesis of the radial-velocity curve is shown in Figure 3.
With the absolute parameters derived for OO Dra, the ratio of orbital angular momentum to rotational angular momentum at equilibrium could be estimated for the binary system following Hut (1981):
$$\alpha = \frac { q } { 1 + q } \frac { 1 } { r _ { g } ^ { 2 } } ( \frac { a } { R _ { 1 } } ) ^ { 2 }$$
where r g is the gyration constant with a typical value of r 2 g = 0 1 for main-sequence stars . and varies negligibly during evolution. Taking the related parameters into Eq. (2), it gives a result of α ≈ 8 3. . It suggests that the rotational angular momentum is appreciable at equilibrium (Hut 1981) . This in turn supports the circular and synchronous assumptions we accepted during the solutions.
## 4. The intrinsic pulsations
In Figures 1 and 2, the pulsational light variability can be clearly seen in the phases of out-eclipses and the secondary minimum of the light curves. With the derived photometric solution, time-resolved theoretical light curves due to eclipsing were computed. Subtracting
the eclipsing light changes from the original observational data, we detected the pure pulsational light variations of the primary star. This enables us to make a detailed analysis of the pulsation nature of OO Dra.
The frequency analysis was carried out with the algorithm Period04 (Lenz & Breger 2005) based on the Fourier transform method. While doing that, we selected only those peaks with the signal to noise ratio (S/N) larger than 4.0 and appeared in both the B- and V-band data for further discussion. The noise levels were computed based on the residuals from the original data when all the trial frequencies were pre-whitened. Figure 4 illustrates the Fourier analysis of both the B- and V-band data. The spectral window and the step-by-step amplitude spectra were plotted. Each spectrum panel in the figure corresponds to the residuals with all the previous frequencies pre-whitened. The bottom panels show the final residuals of the B- and V-band data, wherein the dashed lines represent the confidence curves with 4 σ . The detected frequencies are listed in Table 4.
The periodograms show typical a feature of δ Sct stars with multi-periodicity. Two peaks at f 1 = 41.87 c/d and f 2 = 34.75 c/d are clearly shown in both the B- and V-band original amplitude spectra. The dominant peak is 41.87 c/d, corresponding to a period of 0.02388 days, has a whole amplitude of about 9.8 mmag in B and 8.4 mmag in V.
Table 4: Pulsational properties of OO Dra
| Filter | Frequency (c/d) | Amplitude (mmag) | Phase (rad) | S/N |
|----------|-------------------|--------------------|---------------|-------|
| B | 41.867 ± 0.001 | 4.9 ± 0.3 | 0.998 ± 0.008 | 8.1 |
| | 34.749 ± 0.002 | 3.4 ± 0.3 | 0.577 ± 0.012 | 4.7 |
| V | 41.865 ± 0.001 | 4.2 ± 0.2 | 0.063 ± 0.007 | 8.2 |
| | 34.753 ± 0.002 | 2.8 ± 0.2 | 0.185 ± 0.010 | 4.3 |
Fig. 4.- Spectral windows and the step-by-step amplitude spectra of the intrinsic oscillations of OO Dra..
The amplitude of the second peak is measured to be 6.8 mmag in B and 5.6 mmag in V. When the two frequencies were pre-whitened, a third frequency around 37 c/d arose. It is likely that one previously detected by Dimitrov et al. (2008). But it was not above the significance limit due to the high noise. This frequency is therefore not included into Table 4 for further discussion. In addition, there is no harmonics related to the rotation period detected, implying that the eclipses and proximity effects have been successfully removed from the data.
To check the result and discuss the pulsation nature further, we have computed the theoretical pulsation light curves using the adopted frequencies given in Table 4. After subtracted the pulsational light variations from the original observations, we detected the 'pure' light changes due to eclipsing. A segment of the eclipsing light curves were shown in the lower panels in Figure 1. It can be seen that the short-term light variations have been fairly removed. The totally eclipsing feature is clearly shown. A comparison between the two sets of light curves suggests that the intrinsic pulsations could be very probably from the hotter primary component of the system. This is identical with most eclipsing binaries with δ Sct variables. A discussion on the evolutionary status of the components in the next section also supports this possibility.
Because of the large uncertainty, the absolute parameters derived for the components were not suitable for detailed analysis of the pulsation nature. Following the method described by Zhang et al. (2013), we used only the results from the photometric solution to calculate the mean density of pulsating primary star. The result is ρ /ρ 1 /circledot = 0 232 . ± 0 003. . Taking this value into the well-known equation, Q = P pul ( ρ/ρ /circledot ) 1 2 / , the pulsation constants of about 0.0116 ± 0.0001 and 0.0139 ± 0.0001 days were calculated for 41.87 c/d and 34.75 c/d oscillations, respectively. By using the FAMIAS (Frequency Analysis and Mode Identification for Asteroseismology) program (Zima, 2008), we have tried to make a
preliminary mode identification. It suggests an angular quantum number of l = 0 or l = 1, implying that the star could be pulsating in radial mode or low degree non-radial modes. If this is the case, the detected frequencies could be further identified as the 5th and/or 4th overtone modes (Fitch 1981).
## 5. Summary and discussion
We have presented the simultaneous B- and V-band photometry of the Algol-type eclipsing binary OO Dra. It confirms δ Sct-type pulsation nature of the eclipsing binary. The first photometric solution and physical parameters of the binary system were determined. The intrinsic pulsational characteristics of the primary component were analyzed.
The light curves of OO Dra show totally eclipses. This enables us to obtain a reliable photometric solution for the binary system through light modeling by using the Wilson-Devinney method to determine the geometric parameters. The photometric solution reveals a very small mass ratio of 0.097. This could be the least mass ratio after KIC10661783 (Southworth et al. 2011) among the known eclipsing binaries containing of δ Sct-type pulsating components. The light-curve synthesis indicates a detached configuration for the binary system though the secondary component is very nearly filling its Roche lobe. It means that OO Dra could not be a member of the oEA (Mkrtichian et al. 2004) class. Since the number of δ Sct stars in confirmed detached Algols is very small (Liakos et al. 2012), this star could be an important object in studying of pulsating eclipsing binaries.
With results from the photometric solution, a radial-velocity synthesis was further performed with the measurements published by Dimitrov et al. (2008), and the absolute parameters of the components of the binary were calculated. The results show that the primary component is almost an un-evolved main-sequence star, the secondary is while quite
evolved in over-luminous and over-sized. Comparison with the mass-luminosity relations of Algols ( ˙ banoˇlu et al. I g 2006), we find that the mass and radius detected for the primary component of OO Dra match well with that of the detached Algols ( L 1 ∝ M 3 92 . ± 0 05 . 1 ) rather that the semi-detached ones ( L 1 ∝ M 3 20 . ± 0 25 . 1 ). This in turn supports the detached configuration as indicated by the photometric solution.
Based on the photometric solution, we have calculated the theoretical light curves due to eclipsing. Subtracting the theoretical eclipsing light changes from the observations, the intrinsic pulsational light variations from the hotter primary component were picked up. A frequency analysis of the remaining residuals showed multiple periodicities of the pulsating star. Two confident frequencies at 41.87 and 34.75 c/d were detected in both filters. The dominant pulsation period was determined to be 0.02388 days. The ratio of the pulsational to orbital period of the star turned out to be P pul /P orb = 0 0193. . It agrees well with the correlation between the orbital and pulsational periods preposed by Zhang et al. (2013) for eclipsing binaries containing of δ Sct-type components.
With the mean density of the primary component deduced from photometric solution, the pulsation constants of the detected frequencies were computed and a preliminary mode identification was made. The pulsation constant of the dominant frequency was computed to be about 0.012 days, which is very close to the mean value of δ Sct stars in eclipsing binaries (Zhang et al. 2013). It also suggests that the primary component of OO Dra could pulsate in fifth or fourth overtone assuming that it is in radial or low degree non-radial modes as most δ Sct stars do.
This work is supported by the National Natural Science Foundation of China (NSFC) and the NSFC/CAS Joint Fund of Astronomy through grants 11373037, U1231202 and 2013CB834900. LD would like to acknowledge partial support by National Key Basic Research Program of China 2014CB845703. The authors are grateful to the anonymous
referee for the valuable comments.
## REFERENCES
Biyalieva, N.A., Khruslov, A.V., 2007, PZP, 7, 17 Deng, L.C., Xin, Y., Zhang, X.B. et al., 2013, IAUS, 288, 318 Diethelm, R., 2012, IBVS, 6029 Dimitrov, D., Kraicheva, Z., Popov, V., 2008, IBVS, 5842 Fitch, W.S., 1981, ApJ, 249, 218. Hˇ ubscher, J., Lehmann, P.B., 2013, IBVS, 6070 Hut, P. 1981, A&A, 99, 126 ˙ banoˇlu, C., Soydugan, F., Soydugan, E., Dervi¸ soˇlu, A., 2006, MNRAS, 373, 435 I g g Kallarath, J., Milone, E.F., Terrell, D., Young, A.T. 1998, ApJ, 508, 308 Lampens, P., Kleidis, S., van Cauteren, P., Hambsch, F.J., Vanleenhove, M., Dufoer, S., 2010, IBVS, 5933 Lenz, P., Breger, M., 2005. Commun. Asteroseismol. 146, 53. Liakos, A., Niarchos, P., Soydugan, E., Zasche, P., 2012, MNRAS, 422, 1250. Liu, N., Zhang, X.B., Ren, A.B., Deng, L.C., Luo, Z.Q., 2012, RAA, 12, 671 Lucy, L.B. 1967, ZA, 65, 89 Mkrtichian, D. E., Kusakin, A. V., Rodr´ ıguez, E., et al., 2004, A&A, 419, 1015 Rucinski, S.M. 1969, Acta Astron, 19, 245 Southworth, J., Zima, W., Aerts, C. et al., 2011, MNRAS, 414, 2413
van Hamme, W. 1993, AJ, 106, 2096 Wilson, R.E., Devinney, E.J. 1971, ApJ, 166, 605 Wilson, R.E. 1979, ApJ, 234,1054 Wilson, R.E. 1990, ApJ, 356, 613 Zhang, X.B., Zhang, R.X., Li, Q.S., 2009, RAA, 9, 422 Zhang, X.B., Luo, C.Q., Fu, J.N., 2013, ApJ, 777, 77 Zhang, X.B., Deng, L.C., Wang, K. et al., 2014, AJ, 148, 40
Zima, W., 2008. Commun. Asteroseismol. 155, 17.
This manuscript was prepared with the AAS LT E X macros v5.2. A
|
10.1088/0004-6256/148/6/106
|
[
"X. B. Zhang",
"L. C. Deng",
"J. F. Tian",
"K. Wang",
"J. J. Sun",
"Q. L. Liu",
"H. Q. Xin",
"Q. Zhou",
"Z. Z. Yan",
"Z. Q. Luo",
"C. Q. Luo"
] |
2014-08-05T02:26:24+00:00
|
2014-08-05T02:26:24+00:00
|
[
"astro-ph.SR"
] |
Photometric study of the pulsating, eclipsing binary OO Dra
|
We present a comprehensive photometric study of the pulsating, eclipsing
binary OO Dra. Simultaneous B- and V-band photometry of the star was carried
out on 14 nights. Revised orbital period and a new ephemeris were derived from
the data. The first photometric solution of the binary system and the physical
parameters of the component stars are determined. It reveals that OO Dra could
be a detached system with the less-massive secondary component nearly filling
in its Roche lobe. By subtracting the eclipsing light changes from the data, we
obtained the intrinsic pulsating light curves of the hotter and massive primary
component. Frequency analysis of the residuals light yields two confident
pulsation modes in both B- and V-band data with the dominant frequency detected
at 41.865 c/d. A brief discussion concerning the evolutionary status and the
pulsation nature of the binary system is finally given.
|
1408.0852v1
|
## GAUGE THEORIES ON THE COULOMB BRANCH
John H. Schwarz 1 Walter Burke Institute for Theoretical Physics California Institute of Technology Pasadena, CA 91125, USA
## Abstract
We construct the world-volume action of a probe D3-brane in AdS 5 × S 5 with N units of flux. It has the field content, symmetries, and dualities of the U (1) factor of N = 4 U N ( + 1) super Yang-Mills theory, spontaneously broken to U N ( ) × U (1) by being on the Coulomb branch, with the massive fields integrated out. This motivates the conjecture that it is the exact effective action, called a highly effective action (HEA). We construct an SL (2 , /CI ) multiplet of BPS soliton solutions of the D3-brane theory (the conjectured HEA) and show that they reproduce the electrically charged massive states that have been integrated out as well as magnetic monopoles and dyons. Their charges are uniformly spread on a spherical surface, called a soliton bubble , which is interpreted as a phase boundary.
Lectures presented at the Erice 2014 International School of Subnuclear Physics
1 [email protected]
## Contents
| 1 2 | Introduction | Introduction | 2 |
|-------|-----------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------|-----|
| 1 2 | 2.1 | Background Material N = 4 super Yang-Mills theory . . . . . . . . . . . . . . . . . . . . . . . | 2 |
| | 2.2 | . . Type IIB superstring theory . . . . . . . . . . . . . . . . . . . . . . . . . . . | 2 4 |
| | 2.3 | . . . . | |
| | 2.4 | AdS/CFT duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . | 4 5 |
| | 2.5 | D3-branes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Coulomb branch . . . . . . . . . . . . . . . . . . . . . . . . . . . . | 5 |
| | 2.6 | . . . The probe approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . | 6 |
| | | . | |
| 3 | The Highly Effective Action | The Highly Effective Action | 7 |
| | 3.1 | General requirements for an HEA . . . . . . . . . . . . . . . . . . . | 7 |
| | 3.2 | . . . . . The AdS Poincar´ e patch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . | 8 |
| | 3.3 | The D3-brane in AdS 5 × S 5 . . . . . . . . . . . . . . . . . . . . . . . | 8 |
| | 3.4 | . . . . S duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . | 10 |
| | 3.5 | Summary and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . | 11 |
| 4 | BPS Soliton Solutions | BPS Soliton Solutions | 11 |
| | 4.1 | Solitons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . | 12 |
| | 4.2 | Comparison with the BPS 't Hooft-Polyakov monopole . . . . . . . . . . . . | 14 |
| | 4.3 | Black hole analogy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . | 14 |
| | 4.4 | Multi-soliton solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . | 15 |
| | 4.5 | . . | 15 |
| | | Previous related work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . | |
## 1 Introduction
Two recent papers [1] [2] explored the construction and properties of certain actions that describe superconformal gauge theories on the Coulomb branch. The examples that were studied are ones with a lot of supersymmetry. The reason for this choice is that the large symmetry gives us more confidence in the plausibility of the conjecture that we will describe.
Even though [1] discussed several examples, this article only describes one of them, namely N = 4 super Yang-Mills (SYM) theory. Our procedure is to begin by deriving the world-volume action of a probe D3-brane in an AdS 5 × S 5 background of type IIB superstring theory. As will be explained, the probe D3-brane action involves various approximations. However, the resulting formula has a number of exact properties, especially symmetries, that make it an attractive candidate for the exact solution to a different problem. Specifically, we conjecture that it is the exact effective action for N = 4 SYM theory on the Coulomb branch, called a highly effective action (HEA).
More generally, we consider the world-volume action of a probe p -brane in an AdS p +2 × M q background. M q is a q -dimensional compact space with N units of q -form flux, ∫ M q F q ∼ N . In addition to a D3-brane in AdS 5 × S 5 , ref. [1] also discussed
- · M2-brane in AdS 4 × S / 7 k
- · D2-brane in AdS 4 × CP 3
/CI
- · M5-brane in AdS 7 × S 4
The first two of these correspond to ABJM theory in three dimensions, whereas the third one corresponds to the (2 , 0) theory in six dimensions. As our understanding improves, we intend to extend these investigations to theories with less supersymmetry.
Section 2 provides background material in string theory and quantum field theory that is required to follow the rest of the story. People knowledgable in these subjects can skip this section. Section 3 constructs (the bosonic part of) the formula for the probe D3-brane action - the conjectured HEA - and discusses some of its properties. Section 4, based on [2], describes the construction of BPS soliton solutions of this action. We find exactly the spectrum of soliton solutions expected for the HEA. Moreover, they have an interesting structure in which the charge of the soliton resides on a spherical shell. Even though the theory is nongravitational, the solitons turn out to have unexpected analogies with black holes. This analogy suggests that it may be possible to associate an entropy to solitons with a large charge.
## 2 Background Material
## 2.1 N = 4 super Yang-Mills theory
When one says that a theory has N supersymmetries, one means that the conserved supercharges consist of N irreducible spinors, as appropriate for the spacetime dimension in question. N = 4 supersymmetry is the maximal amount that is possible for an interacting nongravitational quantum field theory in four-dimensional spacetime (4d). The four spinors
may be chosen to be two-component complex Weyl spinors. Thus, there are 16 real supercharges. It is believed that the only 4d nongravitational quantum field theories with this amount of supersymmetry are SYM theories. N = 4 SYM exists for any compact gauge group, but we will focus on the case of U N ( ), so all of the fields are N × N hermitian matrices.
N = 4 SYM was originally derived by first constructing SYM theory in 10d [3]. It contains a ten-vector gauge field A M and a Majorana-Weyl spinor Ψ, both in the adjoint representation of the gauge group. The 4d theory was then obtained by dimensional reduction, which simply means dropping the dependence of the fields on six of the (Euclidean) spatial dimensions. This gives a 4d theory containing a four-vector A µ , six scalars φ I , and four Weyl spinors ψ A . The 10d SYM theory is not a consistent quantum field theory (by itself), but the 4d SYM theory obtained in this way is a consistent quantum theory. It has an SU (4) ∼ SO (6) global symmetry, which corresponds to rotations of the six extra dimensions. Because this symmetry also rotates the four supercharges, Q A , it is called an R-symmetry.
N = 4 SYM theories are conformally invariant, which implies that they are ultraviolet (UV) finite. This was the first class of UV finite 4d quantum field theories to be discovered, but many more (with less supersymmetry) are now known. Combining all of the spacetime symmetries, which are Lorentz invariance, translation invariance, supersymmetry, scale invariance, conformal invariance, R-symmetry, and conformal supersymmetry, gives the superconformal supergroup called PSU (2 , 2 4). | (A supergroup has both bosonic and fermionic generators.) Its bosonic subgroup is SU (2 , 2) × SU (4). The first factor is the 4d conformal group and the second factor is the R-symmetry. Anticommutators of the supersymmetry charges give the momentum operators, which are generators of spacetime translations. Similarly, anticommutators of conformal supersymmetry charges give operators that generate conformal transformations. Mixed anticommutators give the rest of the bosonic generators. Counting both the Poincar´ and conformal supersymmetries, there are a total of 32 fermionic e generators.
The parameters of N = 4 SYM with gauge group U N ( ) are a YM coupling constant g , a vacuum angle θ , and the rank of the gauge group N . For large N and fixed 't Hooft parameter
$$\lambda = g ^ { 2 } N,$$
the theory has a 1 /N expansion (for large N ) with a nice topological interpretation. The leading term in this expansion (the planar approximation) has an additional symmetry called dual conformal invariance. This amount of symmetry is sufficient to make the theory completely integrable. It is not known whether this only applies to the planar approximation or whether it extends to the complete theory. Defining the complex parameter
$$\tau = \frac { \theta } { 2 \pi } + i \frac { 4 \pi } { g ^ { 2 } },$$
the U N ( ) theory has an SL (2 , ) duality symmetry under which
$$\tau \rightarrow \frac { a \tau + b } { c \tau + d },$$
/CI
where a, b, c, d are arbitrary integers satisfying ad -bc = 1. The transformation τ →-1 /τ is called S duality. When θ = 0, this gives g → 4 π/g , which allows one to relate strong coupling to weak coupling. S-duality is an exact nonabelian electric-magnetic equivalence.
## 2.2 Type IIB superstring theory
Type IIB superstring theory is one of the five distinct superstring theories in 10d. The IIB theory has two Majorana-Weyl supersymmetries of the same chirality, for a total of 32 real conserved supercharges. Its massless bosonic fields are
- · the 10d metric g MN
- · the dilaton σ
- · the Neveu-Schwarz (NS-NS) two-form B MN
- · Ramond-Ramond (RR) zero-, two-, and four-forms C C , MN , and C MNPQ
The four-form C 4 , constructed from C MNPQ , has a self-dual field strength F 5 = dC 4 + . . . .
Type IIB superstring theory has a few solutions that preserve all of the supersymmetry. The most obvious one is 10d Minkowski spacetime, i.e., g MN = η MN . In this solution σ and C are constants and the other fields vanish. A less obvious maximally supersymmetric solution has a metric describing the geometry AdS 5 × S 5 - we'll give the formula later. In this solution σ and C are again constants. However, now F 5 ∼ N [vol( AdS 5 ) + vol( S 5 )]. N is the number of units of five-form flux threading the five-sphere, ∫ S 5 F 5 ∼ N , where the coefficients depend on normalization conventions. The isometry of this solution is given by the supergroup PSU (2 , 2 4), the same supergroup as before! | In particular, SU (2 , 2) is the isometry of AdS 5 and SU (4) is the isometry of S 5 .
The value of exp( σ ) is the string coupling constant g s and the value of C is called χ . In terms of these one can form
$$\tau = \chi + i / g _ { s }.$$
Type IIB superstring theory has an exact SL (2 , ) symmetry under which
$$\text{as an exact $SL(2,\mathbb{Z}$) symmetry under which} \\ \tau \rightarrow \frac { a \tau + b } { c \tau + d },$$
as before. More generally, the fields σ and C transform in this manner, but we are concerned with the case when they are constants.
## 2.3 AdS/CFT duality
The preceding sections have been presented so as to make the AdS/CFT conjecture seem obvious. However, when it is was put forward by Maldacena [4], it came as quite a surprise. Specifically, he proposed that N = 4 SYM in 4d with U N ( ) gauge group is exactly equivalent ('dual') to type IIB superstring theory in an AdS 5 × S 5 background with N units of fiveform flux. There are many other analogous AdS/CFT pairs, some of which are relevant to the other examples of probe-brane theories listed in the introduction.
The evidence for AdS/CFT duality that we have presented so far is that both the gauge theory and the string theory have PSU (2 , 2 4) symmetry and | SL (2 , /CI ) duality. However, by now there is much, much more evidence, which we will not describe here. Even though the evidence is overwhelming, a complete proof is not possible. The reason for saying this is that we lack a complete nonperturbative definition of type IIB superstring theory (other than the one given by AdS/CFT duality). The best one can hope to do, in the absence of such a definition, is to show that everything we know about type IIB superstring theory in an AdS 5 × S 5 background with N units of five-form flux agrees with what can be deduced from N = 4 SYM in 4d with U N ( ) gauge group.
## 2.4 D3-branes
Superstring theories contain various supersymmetric (and hence stable) extended objects. Ones with p spatial dimensions are called p -branes. They carry a type of conserved current J that couples to a ( p +1)-form gauge field A with a ( p + 2)-form field strength F = dA . Those p -branes on which strings can end are called D p -branes. (D stands for Dirichlet here, since such strings satisfy Dirichlet boundary conditions in the directions orthogonal to the p -brane. They also satisfy Neumann boundary conditions in the directions parallel to the D -brane.) p
In type IIB superstring theory supersymmetric D p -branes exist for p = 1 3 5 7. Only the , , , D3-brane is invariant under SL (2 , /CI ) transformations. This theory has an infinite number of different kinds of strings. A ( p, q ) string arises as a bound state of p fundamental strings and q D-strings, provided p and q are relatively prime. These strings transform irreducibly under SL (2 , /CI ), and any one of them can end on a D3-brane. From the point of view of the world-volume theory of a single D3-brane, which is an abelian gauge theory, the end of the string appears to carry p units of electric charge and q units of magnetic charge.
The various p -branes also act as sources of gravitational and other fields. Maldacena was led to his conjecture by realizing that N coincident D3-branes give a 'black brane,' which is a higher-dimensional generalization of a black hole, whose near-horizon geometry is AdS 5 × S 5 with N units of 5-form flux on the sphere. In this way the branes are replaced by a 10d geometry with a horizon, and the 10d string theory is represented 'holographically' by a 4d quantum field theory.
A stack of coincident flat D p -branes has fields that are localized on its ( p +1)-dimensional world volume. They define a 'world-volume theory,' which is maximally supersymmetric when the background in which they are embedded is maximally supersymmetric. In particular, we will study the world-volume theory of a single D3-brane embedded in an AdS 5 × S 5 background geometry, which may be regarded as having been created by N other coincident D3-branes.
## 2.5 The Coulomb branch
A stack of N coincident flat D3-branes has a world-volume theory, which is a U N ( ) gauge theory. Consider starting with N coincident flat D3-branes and pulling them apart along one of the orthogonal axes (let's call it the x direction) into clumps N 1 + N 2 + . . . + N k = N .
Then, the world-volume gauge symmetry is broken spontaneously,
$$U ( N ) \to U ( N _ { 1 } ) \times U ( N _ { 2 } ) \times \dots \times U ( N _ { k } ).$$
The world-volume fields that arise as the lowest mode of a fundamental string with one end attached to the i th D3-brane and the other end attached to the j th D3-brane acquire the mass
$$m _ { i j } = | x _ { i } - x _ { j } | T,$$
where T is the fundamental string tension. This is similar to the Higgs mechanism. The main difference is that the scalar fields that are eaten by the gauge fields that become massive belong to the adjoint representation of the original gauge group (in contrast to the SU (2) Higgs doublet in the standard model). To emphasize this distinction, this phase of the gauge theory is called a Coulomb branch (rather than a Higgs branch).
Let us consider N = 4, d = 4 SYM theory with a U N ( ) gauge group. For most purposes one can say that a free U (1) multiplet decouples leaving an SU N ( ) theory. However, this 'decoupled' U (1) is needed to get the full SL (2 , /CI ) duality group rather than a subgroup [5][6]. In the special case of U (2), if we ignore the decoupled U (1), the remaining SU (2) theory on the Coulomb branch has unbroken U (1) gauge symmetry. Let us refer to the massless supermultiplet as the 'photon' supermultiplet and the two massive supermultiplets as W ± .
The U (2) N = 4 SYM theory on the Coulomb branch has a famous soliton solution: the 't Hooft-Polyakov monopole. This solution preserves half of the supersymmetry. One says that it is 'half BPS.' This monopole is part of an infinite SL (2 , /CI ) multiplet of half-BPS ( p, q ) states, with p units of electric charge and q units of magnetic charges, where p and q are coprime. The masses, determined by supersymmetry, are
$$M _ { p, q } = v g | p + q \tau | = v g \sqrt { \left ( p + \frac { \theta } { 2 \pi } q \$$
where 〈 φ 〉 = v is the vev of a massless scalar field. The W mass is M 1 0 , = gv and the monopole mass is M 0 1 , = 4 πv/g for θ = 0. As discussed above, a ( p, q ) dyon is introduced when a ( p, q ) string ends on a D3-brane.
## 2.6 The probe approximation
Consider a D3-brane embedded in a 10d spacetime. The probe approximation involves neglecting the back reaction of the brane on the geometry and the other background fields. Since the brane is a source for one unit of flux, this requires that the background flux N is large, so that the distinction between N and N +1 becomes negligible. The world-volume action of a single D-brane contains a U (1) field strength, F αβ = ∂ A α β -∂ A β α . Since, no derivatives of F are included in the formula, F is required to vary sufficiently slowly so that its derivatives can be neglected. A similar restriction applies to the other world-volume fields, as well.
Despite these approximations, the probe D3-brane action has some beautiful exact properties: It precisely realizes the isometry of the AdS 5 × S 5 background geometry as a worldvolume superconformal symmetry PSU (2 , 2 4). | Only the bosonic subgroup is taken into
account in the subsequent discussion, since fermi fields are omitted. (A version of the complete formula with fermions is given in [7].) The brane action also has the duality symmetry of the background, which is SL (2 , /CI ) for the D3-brane example. Furthermore, the D3-brane world-volume action is most naturally formulated with local symmetries, which are general coordinate invariance and a fermionic symmetry called kappa symmetry. It is only after implementing a suitable gauge choice that one is left with a conventional nongravitational quantum field theory in Minkowski spacetime.
In principle, one can integrate out the massive fields of the Coulomb branch theory exactly, thereby producing a very complicated formula in terms of the massless photon supermultiplet only. Very schematically,
$$\exp ( i S _ { \text{HEA} } ) = \int D W ^ { + } D W ^ { - } \exp ( i S ) \quad \quad \quad$$
If one could do this integral exactly, which is not possible in practice, the resulting action would capture the entire theory on the Coulomb branch, and it would be valid at all energies. This is in contrast with the more common notion of a low-energy effective action, which only includes the leading terms in a derivative expansion. We have proposed to call such an exact Coulomb branch action a highly effective action (HEA). Even though we cannot carry out such an exact computation, in some cases we know many of the properties that the HEA should possess. One can hope that they go a long way towards determining it.
## 3 The Highly Effective Action
## 3.1 General requirements for an HEA
An HEA should have all of the unbroken and spontaneously broken global symmetries of the original Coulomb branch theory with the massive W fields included. The conformal symmetry is spontaneously broken as a consequence of assigning a vacuum expectation value (vev) to a massless scalar field that has a flat potential. In other words, this vev spontaneously breaks spacetime symmetries (though not Poincar´ e symmetry) as well as gauge symmetry. An HEA should have the same duality properties as the Coulomb branch theory containing explicit W fields. The global symmetry and duality groups are PSU (2 , 2 4) | and SL (2 , ) in the example considered here.
/CI A further requirement for an HEA is that it should have the same spectrum of supersymmetry protected (or BPS) states as the original Coulomb branch theory. In particular, the W ± supermultiplets, which have been integrated out, should reappear in the HEA as solitons. By contrast, this would not be expected for a low-energy effective action. When we examine these soliton solutions explicitly, we will be led to a certain refinement of the interpretation of this requirement. Specifically, a complete understanding of the BPS soliton solutions also requires knowledge of the original conformal branch gauge theory for which the gauge symmetry and conformal symmetry are unbroken.
Ref. [1] conjectured that the probe D3-brane action, in an AdS 5 × S 5 background with one unit of flux ( N = 1), is precisely the HEA for the N = 4, 4d SYM theory with U (2) gauge symmetry on the Coulomb branch despite the fact that it is only an approximate
solution to a different problem. In fact, N = 1 is the worst possible case for the probe approximation. However, this shouldn't matter, because we are solving a different problem. Still, it is helpful to keep the 10d description in mind.
I am not certain that this conjecture is correct. On the one hand, the formula that we will obtain seems to be too simple to be the exact answer for such a complicated path integral. On the other hand, we will find that it has all of the expected properties of the HEA. If it not the conjectured HEA, then it would seem to define a new maximally supersymmetric 4d quantum field theory, whose existence is unexpected. So either conclusion is remarkable.
## 3.2 The AdS Poincar´ e patch
The AdS p +2 geometry with unit radius can be described as a hypersurface embedded in a ( p +3)-dimensional Lorentzian space of signature ( p +1 2): ,
$$y \cdot y - u \phi = - 1,$$
where y · y = -( y 0 ) 2 + ∑ p 1 ( y i ) 2 . The equation eliminates one of the two time directions leaving a manifold of signature ( p +1 1). The Poincar´ e-patch metric of radius , R is
$$d s ^ { 2 } = R ^ { 2 } ( d y \cdot d y - d u d \phi ).$$
Note that this preserves the SO p ( + 1 2) symmetry of the embedding equation. , Defining x µ = y µ /φ and eliminating u = φ -1 + φx · x ,
$$d s ^ { 2 } = R ^ { 2 } ( \phi ^ { 2 } d x \cdot d x + \phi ^ { - 2 } d v ^ { 2 } ).$$
It is more customary to write the formula in terms of a coordinate z = φ -1 , which has dimensions of length. However, the choice of φ is convenient for our purposes, since it will correspond to a scalar field, also called φ , with dimensions of inverse length in the D3-brane action.
The Poincar´ e patch is not geodesically complete. Rather there exists a different coordinate choice that describes a geodesically complete spacetime, called global AdS, for which the Poincar´ patch is just a region. e The Poincar´ e patch contains a horizon, which is not present in the global AdS metric. In this respect it is analogous to the horizon of Rindler spacetime, whose geodesic completion is Minkowski spacetime. The Poincar´ e patch of AdS is sufficient for our purposes.
The ten-dimensional AdS 5 × S 5 metric ds 2 = g MN ( x dx ) M dx N is
$$d s ^ { 2 } = R ^ { 2 } \left ( \phi ^ { 2 } d x \cdot d x + \phi ^ { - 2 } d \phi ^ { 2 } + d \Omega _ { 5 } ^ { 2 } \right ) = R ^ { 2 } \left ( \phi ^ { 2 } d x \cdot d x + \phi ^ { - 2 } d \phi \cdot d \phi \right ), \quad \ \ ( 1 2 ) \\ \text{where } d \Omega _ { 5 } ^ { 2 } \text{ is the metric of a round unit-radius five-sphere, and } \phi \text{ is now the length of the}$$
where d Ω 2 5 is the metric of a round unit-radius five-sphere, and φ is now the length of the six-vector φ I . It will be important later that φ cannot be negative.
## 3.3 The D3-brane in AdS 5 × S 5
According to the AdS/CFT dictionary, the value of the radius R , expressed in string units, satisfies
$$R ^ { 4 } = 4 \pi g _ { s } N l _ { s } ^ { 4 }.$$
The positive integer N is the number of units of five-form flux on the five-sphere. Also, ∫ S 5 F 5 = 2 πN for an appropriate normalization, and N is the rank of the U N ( ) gauge group of the dual CFT.
The probe D3-brane action is the sum of two terms: S = S 1 + S 2 . S 1 is a Dirac-BornInfeld (DBI) functional of the ten embedding functions x M ( σ α ) and a world-volume U (1) gauge field A β ( σ α ) with field strength F αβ = ∂ A α β -∂ A β α . S 2 is a Chern-Simons-like (CS) term, which is linear in the RR fields. S 1 and S 2 also depend of the fermi fields that we are omitting.
The general formula for the bosonic part of the DBI term is
$$S _ { 1 } = - T _ { D 3 } \int \sqrt { - \det \left ( G _ { \alpha \beta } + 2 \pi \alpha ^ { \prime } F _ { \alpha \beta } \right ) } \, d ^ { 4 } \sigma, \\ \intertext { < the induced } \, \text{$\Delta$ and world volume metric}$$
where G αβ is the induced 4d world-volume metric
$$G _ { \alpha \beta } = g _ { M N } ( x ) \partial _ { \alpha } x ^ { M } \partial _ { \beta } x ^ { N }.$$
As usual, α ′ = l 2 s and the D3-brane tension is
$$T _ { D 3 } = \frac { 2 \pi } { g _ { s } ( 2 \pi l _ { s } ) ^ { 4 } }.$$
In fact, only two dimensionless combinations of parameters occur in the brane action:
$$R ^ { 4 } T _ { D 3 } = \frac { N } { 2 \pi ^ { 2 } } \quad \text{and} \quad 2 \pi \alpha ^ { \prime } / R ^ { 2 } = \sqrt { \pi / g _ { s } N }. \\ y \, \text{depends on the dimensionless string coupling constant } g _ { s } \, \text{and the integer } N.$$
Thus, S 1 only depends on the dimensionless string coupling constant g s and the integer N . We shall see shortly that the g s dependence also drops out.
The nonvanishing Chern-Simons terms (aside from coefficients, which can be found in [1]) are
$$S _ { 2 } & \sim \int C _ { 4 } + \chi \int F \wedge F. \\ \text{:ential $C_{A}$ has a self-dual field strength $F_{\kappa}$ = $dC_{A}$.} \text{ as mentioned}$$
The RR four-form potential C 4 has a self-dual field strength F 5 = dC 4 . As mentioned previously,
F 5 ∼ N ( vol( S 5 ) + vol( AdS 5 ) ) , (19) and χ is the value of the RR 0-form C 0 . It is proportional to the theta angle of the gauge theory.
S 1 contains a potential term ∫ φ d x 4 4 , which should not appear in S , since there should be no net force acting on the brane when φ > 0. In fact, this term is canceled by the ∫ C 4 term in S 2 . To show this we consider the region M of AdS 5 for which the coordinate φ is less than the value of φ that specifies the position of the D3-brane. Then, using Stokes' theorem,
$$\int _ { D 3 } C _ { 4 } = \int _ { M } F _ { 5 } \sim \int _ { 0 } ^ { \phi } \int _ { D 3 ^ { \prime } } \text{vol} ( A d S _ { 5 } ) ^ { \prime },$$
where
$$\text{vol} ( A d S _ { 5 } ) ^ { \prime } \sim ( \phi ^ { \prime } ) ^ { 3 } d \phi ^ { \prime } \wedge d x ^ { 0 } \wedge d x ^ { 1 } \wedge d x ^ { 2 } \wedge d x ^ { 3 }$$
Thus,
The coefficients work perfectly.
$$\int _ { D 3 } C _ { 4 } \sim \int \phi ^ { 4 } d ^ { 4 } x.$$
The general coordinate invariance of the D3-brane action allows one to fix a gauge. A convenient choice for our purposes is the static gauge in which the four world-volume coordinates σ α are identified with the four Lorentzian coordinates x µ , which were introduced previously,
$$x ^ { \mu } ( \sigma ) = \delta _ { \alpha } ^ { \mu } \sigma ^ { \alpha }.$$
In this gauge φ I ( σ ) and A µ ( σ ) are the only remaining bosonic world-volume fields. Moreover, due to the gauge choice, they become functions of x µ .
The complete D3-brane action in the static gauge (aside from fermions), expressed in terms of canonically normalized fields φ I ( σ ) and A µ ( σ ), is where
$$S = \frac { 1 } { \gamma ^ { 2 } } \int \phi ^ { 4 } \left ( 1 - \sqrt { - \det M _ { \mu \nu } } \right ) \, d ^ { 4 } x + \frac { 1 } { 4 } g _ { s } \chi \int F \wedge F,$$
and
$$\gamma & = \sqrt { N / 2 \pi ^ { 2 } } & ( 2 5 ) \\. & \quad F _ { \dots }, \quad \dots \phi ^ { I } \partial _ { \dots } \phi ^ { I } & \dots$$
$$M _ { \mu \nu } = \eta _ { \mu \nu } + \gamma \frac { F _ { \mu \nu } } { \phi ^ { 2 } } + \gamma ^ { 2 } \frac { \partial _ { \mu } \phi ^ { I } \partial _ { \nu } \phi ^ { I } } { \phi ^ { 4 } },$$
where η µν denotes the Lorentz metric. It is an important fact that the first term in S is independent of g s and χ and that they only appear (as a product) in the coefficient of the second term.
The scale invariance of the action (24) is manifest, since all terms have dimension four and all parameters are dimensionless. On the Coulomb branch there is a scale, the vev of the scalar field φ . However, the full conformal symmetry is realized on the action. Only the choice of vacuum breaks the symmetry (and all choices are equivalent). The D3-brane action contains inverse powers of the scalar field. However, the vev of this field ensures that these are not singular. Individual terms in the action can be arbitrarily complicated and still end up with dimension four by including an appropriate (inverse) power of φ .
Regarded as an HEA, (24) should encode all the quantum effects that arise from integrating out the W ± supermultiplets. On the other hand, (24) should have its own loop expansion (or path integral) to take account of the quantum effects of the massless supermultiplet. The loop expansion is expected to be free of UV divergences, because of the (spontaneously broken) conformal symmetry. There are IR divergences, but they can be treated by standard methods. Rescaling all fields in (24) by γ brings the action to a form in which all the N dependence appears as an overall factor of N , so that S N ( ) = NS (1). This shows that the loop expansion of this theory is a 1 /N expansion.
## 3.4 S duality
The SL (2 , /CI ) transformation τ → -1 /τ duality is accompanied by an electric-magnetic transformation in the S-duality transformation of the gauge theory. This invariance has not
been proved in the nonabelian formulation with W fields, but it was proved for the abelian D3-brane action in [1]. Recall that τ = τ 1 + iτ 2 = χ + i/g s , but that the action only depends on g χ s = τ /τ 1 2 . When τ →-1 /τ , g χ s →-g χ s . To show invariance under this sign change, we must also examine the transformation that exchanges electric and magnetic fields.
The procedure is standard. It involves replacing the U (1) gauge field A µ by a new one such that the Bianchi identity
$$\partial _ { [ \mu } F _ { \nu \rho ] } = 0$$
becomes a field equation, and the field equation
$$\partial _ { \mu } \left ( \frac { \partial \mathcal { L } } { \partial F _ { \mu \nu } } \right ) = 0 \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\quad \dots$$
becomes a Bianchi identity. The nontrivial fact, which was proved in [1], is that the new action, obtained by this field replacement together with the sign change g χ s → -g χ s , is identical to the original one.
## 3.5 Summary and discussion
We have conjectured that the world-volume action of a probe D3-brane in an AdS 5 × S 5 background of type IIB superstring theory, with one unit of flux ( N =1), can be reinterpreted as the HEA of U (2) N = 4 SYM theory on the Coulomb branch. An explicit formula for the bosonic part of the action has been presented. It is likely that there is a generalization in which the formula with N > 1 plays a role in the Coulomb branch decomposition U N ( +1) → U N ( ) × U (1). However, when N > 1 certain issues still need to be clarified.
The evidence presented so far for the conjecture that the probe D3-brane action is the desired HEA is that the action incorporates all of the required symmetries and dualities: PSU (2 , 2 4) superconformal symmetry (when fermions are included) and | SL (2 , /CI ) duality. The next section will describe BPS soliton solutions of this action. Their properties will give further support to the conjecture. They will also lead us to refine the conjecture somewhat.
## 4 BPS Soliton Solutions
The previous section described the construction of the bosonic part of a world-volume action for a probe D3-brane in AdS 5 × S 5 . We also discussed N = 4 SYM theory, with gauge group U N ( +1), on the Coulomb branch. A suitable scalar field vev breaks the gauge symmetry so that
$$U ( N + 1 ) \rightarrow U ( 1 ) \times U ( N ).$$
The 2 N supermultiplets corresponding to the broken symmetries acquire mass gv . The probe D3-brane action was conjectured to be the HEA for the U (1) factor.
In this section we will derive classical supersymmetric soliton solutions of the action (24). Even though they are classical, they are supposed to take account of all quantum effects due to the massive fields that have been integrated out. Furthermore, they will turn out be BPS, which implies that they are protected from additional quantum corrections. The soliton
solutions will be independent of N (or γ ), since N is an overall factor of the action written in terms of rescaled fields. Thus, this parameter drops out of the classical field equations.
For the purpose of constructing the soliton solutions, it is sufficient to consider a D3brane that is localized at a fixed position on the S 5 . In this case, the S 5 coordinates do not contribute to the D3-brane action. The radial position in the AdS 5 space is encoded in the nonnegative scalar field φ . The dependence of the action on this field and the U (1) gauge field are all that are required in this section. The fermions are not required, since they vanish for these solutions.
## 4.1 Solitons
Let us begin by looking for spherically symmetrical static solutions, centered at r = 0, for which the action Eq. (24) is stationary. We require that /vector E and /vector B (the electric and magnetic parts of F µν ) only have radial components, denoted E and B . Also, E B , , and φ are functions of the radial coordinate r only. It then follows that
$$\begin{array} { c } \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \\ \cdot \cdot \cdot \cdot \cdot \\ - \det & \cdot \cdot \cdot \cdot \cdot \\ - \det & ( M _ { \mu \nu } ) = - \det \left ( \eta _ { \mu \nu } + \gamma ^ { 2 } \frac { \partial _ { \mu } \phi \partial _ { \nu } \phi } { \phi ^ { 4 } } + \gamma \frac { F _ { \mu \nu } } { \phi ^ { 2 } } \right ) \\ \quad & = \left ( 1 + \gamma ^ { 2 } \frac { ( \phi ^ { \prime } ) ^ { 2 } - E ^ { 2 } } { \phi ^ { 4 } } \right ) \left ( 1 + \gamma ^ { 2 } \frac { B ^ { 2 } } { \phi ^ { 4 } } \right ). \\ \text{lets in the Lagrangian density} \end{array}$$
This results in the Lagrangian density
$$\mathcal { L } = \frac { 1 } { \gamma ^ { 2 } } \phi ^ { 4 } \left ( 1 - \sqrt { \left ( 1 + \gamma ^ { 2 } \frac { ( \phi ^ { \prime } ) ^ { 2 } - E ^ { 2 } } { \phi ^ { 4 } } \right ) \left ( 1 + \gamma ^ { 2 } \frac { B ^ { 2 } } { \phi ^ { 4 } } \right ) } \, \right ) + g _ { s } \chi B E.$$
The equation of motion for A 0 is ∂ ∂r ( r D 2 ) = 0 , where
$$D = \frac { \partial \mathcal { L } } { \partial E } = E \sqrt { \frac { 1 + \gamma ^ { 2 } B ^ { 2 } / \phi ^ { 4 } } { 1 + \gamma ^ { 2 } [ ( \phi ^ { \prime } ) ^ { 2 } - E ^ { 2 } ] / \phi ^ { 4 } } } + g _ { s } \chi B.$$
For a soliton centered at r = 0, with p units of electric charge g and q units of magnetic charge g m , where g m = 4 π/g , we have
$$D = \frac { \ p g } { 4 \pi r ^ { 2 } } \quad \text{and} \quad B = - \frac { q g _ { m } } { 4 \pi r ^ { 2 } }.$$
The mass of the soliton is given by the Hamiltonian. H = ∫ H d x 3 = 4 π ∫ H r dr 2 . The Hamiltonian density is H = DE -L . Eliminating E in favor of D gives where
$$\mathcal { H } = \frac { 1 } { \gamma ^ { 2 } } \left ( \sqrt { ( \phi ^ { 4 } + \gamma ^ { 2 } ( \phi ^ { \prime } ) ^ { 2 } ) ( \phi ^ { 4 } + \gamma ^ { 2 } X ^ { 2 } ) } - \phi ^ { 4 } \right ),$$
and
$$X & = \sqrt { \tilde { D } ^ { 2 } + B ^ { 2 } } = Q / r ^ { 2 }. & & ( 3 5 ) \\ & \quad \tilde { \varkappa } \quad \varkappa \quad \varkappa & & \varkappa \varkappa \varkappa$$
$$\tilde { D } = D - g _ { s } \chi B.$$
Thus, ˜ D 2 + B 2 = Q /r , 2 4 where
$$\text{ere} \\ Q = \frac { g } { 4 \pi } | p + q \tau |.$$
As before,
$$\tau = \chi + i / g _ { s } = \frac { \theta } { 2 \pi } + i \frac { 4 \pi } { g ^ { 2 } }.$$
We want to find functions φ r ( ) that give BPS extrema of H with the boundary condition φ → v as r → ∞ . The BPS condition turns out to require that the two factors inside the square root in Eq. (34) are equal, which implies that H = ( φ ′ ) 2 . The proof goes as follows. One first writes the formula for H in the form
$$( \gamma ^ { 2 } \mathcal { H } + \phi ^ { 4 } ) ^ { 2 } = ( \gamma ^ { 2 } X | \phi ^ { \prime } | + \phi ^ { 4 } ) ^ { 2 } + \gamma ^ { 2 } \phi ^ { 4 } ( X - | \phi ^ { \prime } | ) ^ { 2 }.$$
Thus,
$$( \gamma ^ { 2 } \mathcal { H } + \phi ^ { 4 } ) ^ { 2 } \geq ( \gamma ^ { 2 } X | \phi ^ { \prime } | + \phi ^ { 4 } ) ^ { 2 },$$
which implies H ≥ X φ | ′ | . Saturation of the BPS bound is achieved for | φ ′ | = X and then
$$\mathcal { H } = X ^ { 2 } = ( \phi ^ { \prime } ) ^ { 2 } = Q ^ { 2 } / r ^ { 4 }.$$
The equation ( φ ′ ) 2 = Q /r 2 4 , together with the boundary condition φ → v as r →∞ , has two BPS solutions
$$\phi _ { \pm } ( r ) = v \pm Q / r,$$
where Q = g 4 π | p + qτ | . The φ + solution is similar to the flat-space case studied by Callan and Maldacena [8]. It describes a funnel-shaped protrusion of the D3-brane extending to the boundary of AdS at φ = + ∞ . This solution gives infinite mass (proportional to ∫ dr/r 2 ), and thus it is not the solution we are seeking.
The φ -solution is different. φ = 0 corresponds to the horizon of the Poincar´ patch of e AdS 5 . Thus, since φ is nonnegative, the φ -solution must be cut off at
$$r _ { 0 } = \frac { Q } { v }.$$
Then the masses of BPS solitons are given by
$$M = 4 \pi \int _ { r _ { 0 } } ^ { \infty } \mathcal { H } r ^ { 2 } d r = \frac { 4 \pi Q ^ { 2 } } { r _ { 0 } } = 4 \pi v ^ { 2 } r _ { 0 } = v g | p + q \tau |, \\ \int _ { - \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \ dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots\dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \delta \tau / \omega \sum \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \sigma \tau / \omega \sum \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \sigma \tau / \omega / \sum \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \delta \sigma \tau / \omega \sum \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \sigma \tau / \omega \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \salpha \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sig \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sg \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigma \sum \sigmoid \, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q. \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ eq, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ eq. \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ eq, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e g, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e g. \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \ e q, \$$
exactly as was expected. We have obtained a complete infinite SL (2 , /CI ) multiplet, which includes the W particles that were integrated out, as well as monopoles and dyons.
The charge of the φ -solution is uniformly spread on the sphere r = r 0 , which we call a soliton bubble . The interior of the bubble should not contribute to the mass of the soliton. So, how should we think about the interior of the bubble in the QFT? The only sensible interpretation is that the gauge theory is in the ground state of the conformal phase of U N ( + 1) inside the sphere. This implies that the bubble is a phase boundary . This interpretation has the advantage that the parameter τ is required to describe the U N ( +1) theory in the conformal phase. This would explain how the soliton solutions know what the values of g and θ are. The U (1) action does not contain all the required information. One also needs to know the nonabelian theory in the conformal phase.
## 4.2 Comparison with the BPS 't Hooft-Polyakov monopole
Let us compare the monopole solution of the D3-brane action that we have obtained with the corresponding BPS 't Hooft-Polyakov monopole solution. The single monopole solution of the nonabelian SU (2) gauge theory on the Coulomb branch (for θ = 0) has a triplet of scalar fields φ a whose internal symmetry is aligned with the spatial directions as follows : 2
$$\phi ^ { a } ( \vec { x } ) = \frac { x ^ { a } } { r } \phi ( y )$$
with
$$\phi ( y ) = v ( \coth y - 1 / y ),$$
where y = M r W and M W = gv . φ y ( ) is strictly positive for y > 0, and φ a ( /vector x ) is nonsingular at the origin. Thus, there is no sign of a soliton bubble in the nonabelian description. Both constructions give the correct mass and charge for the monopole, but the D3-brane solution gives a soliton bubble whereas the nonabelian solution does not. So, which formula more accurately describes what is happening?
Equation (46) differs from
$$\phi _ { - } ( r ) = v ( 1 - 1 / y ),$$
the D3-brane theory result for the monopole, by a series of terms of the form exp( -2 nM W r ), where n is a positive integer. In the context of the N = 4 theory, the effect of integrating out the fields of mass M W should be to cancel these exponential terms for y > 1 and to give φ = 0 for y < 1. After all, the U (1) HEA is supposed to incorporate all of the contributions due to W loops, and these exponentials are a plausible form for those contributions. Hence, we conclude that the bubble is real and that the D3-brane solution gives a more accurate description of what is happening than the usual classical solution (46) of the nonabelian theory.
## 4.3 Black hole analogy
We have found a universal formula relating the mass and radius of BPS soliton bubbles: M = 4 πv r 2 0 , which is valid for all ( p, q ). For comparison, the radius of the horizon of a 4d extremal Reissner-Nordstrom asymptotically Lorentzian black hole in four dimensions is r 0 = MG , where G is Newton's constant, for all ( p, q ). In the latter case the charge Q should be large for the classical analysis to be valid. Thus, the relation between mass and radius is the same in both cases, with (4 πv 2 ) -1 the analog of Newton's constant. The BPS condition ensures that the analogy extends to the relation between mass and charge.
This analogy is rather surprising, because the D3-brane theory is a nongravitational theory in 4d Minkowski spacetime. If one tries to pursue this analogy, there is a natural question: Does the Bekenstein-Hawking entropy of the black hole, which is proportional to Q 2 (for large Q ), have an analog for the solitons of the D3-brane theory? For example, is there an entanglement entropy between the inside and outside of the soliton bubble with
2 The N = 4 theory has six such triplets, but only one of them is utilized in the construction. This choice corresponds to the choice of a point on the five-sphere in the D3-brane construction.
this value. If the solitons can be shown to have a well-defined entropy of this sort (for large Q ), then one may be tempted to take the black-hole analogy seriously.
Even though the D3-brane theory is defined on a 4d Lorentzian spacetime, we know that the field φ can be interpreted as a radial coordinate in AdS 5 . From this point of view the soliton solutions have a nontrivial geometry induced from their embedding in AdS 5 . From the 5d (or 10d) viewpoint, the bubble is on the horizon of the Poincar´ patch of e AdS 5 , where it intersects the boundary of global AdS. This fact may be useful for understanding the origin of the black-hole analogy.
## 4.4 Multi-soliton solutions
It is easy to derive the generalization of φ -( r ) to the case of n solitons of equal charge. 3 Since supersymmetry ensures that the forces between them should cancel when they are at rest, their centers can be at arbitrary positions /vector x k , k = 1 2 , , . . . , n . Since φ satisfies Laplace's equation, the solutions can be superposed. The obvious guess, which is easy to verify, is
$$\phi ( \vec { x } ) = v - Q \sum _ { k = 1 } ^ { n } \frac { 1 } { | \vec { x } - \vec { x } _ { k } | }. \quad \begin{array} { c } \\ \\ \\$$
The surfaces of the bubbles, which are no longer spheres, are given by φ /vector ( x ) = 0. The fields /vector D and /vector B are then proportional to /vector ∇ φ , with coefficients determined by the charges. Their values at φ = 0 determine the charge densities on the bubble surface. This is much simpler than the usual multi-monopole analysis, which involves nonlinear equations! It should also be more accurate.
## 4.5 Previous related work
Soliton bubbles like those found here have appeared in the literature previously. We will briefly discuss the examples that we are aware of. There may be others.
Using attractor flow equations [9], Denef [10], [11] found similar structures to our soliton bubbles in the context of supergravity solutions in which a D3-brane wraps a cycle of a Calabi-Yau manifold that vanishes at a conifold point, where the central charge modulus is zero.
Gauntlett et al. [12] studied soliton solutions of a probe D3-brane in an asymptotically flat black D3-brane supergravity background. This problem is closely related to the one we have considered, since this geometry has AdS 5 × S 5 as its near-horizon limit. They identified half-BPS solutions, like those found here, 'in which a point charge is replaced by a perfectly conducting spherical shell.' In [13] the authors examined the DBI action of a D3-brane probe in F theory. They constructed a monopole solution containing a soliton bubble that coincides with a 7-brane. The formation of soliton bubbles may also be related to the enhan¸ con mechanism in [14]. This mechanism circumvents the appearance of a class of naked singularities, known as repulsons .
3 Pairs of solitons with different charges are mutually nonlocal and therefore difficult to describe.
In [15], Popescu and Shapere studied the low-energy effective action of N = 2 SU (2) gauge theory without additional matter in the Coulomb phase. The Seiberg-Witten lowenergy effective action [16] should only be valid below the mass of W bosons. However, the theory has a BPS monopole and a BPS dyon that become massless at points in the moduli space of vacua. Therefore, these particles should be obtainable as soliton solutions of the Seiberg-Witten action. Popescu and Shapere constructed these solutions and discovered that they exhibit a spherical shell of charge, just like what we have found.
There is some interesting related evidence for soliton bubbles in a nonsupersymmetric field theory context [17]-[20]. By considering multi-monopole solutions of large magnetic charge in the 4d SU (2) gauge theory with adjoint scalars on the Coulomb branch, Bolognesi deduced the existence of 'magnetic bags' with properties that are very close to those of the soliton bubbles obtained here. Bolognesi's magnetic bags do not have sharply defined surfaces, though they become sharp in the limit of large charge. Bolognesi also pointed out the analogy to black holes [19].
## 5 Conclusion
The action of a probe D3-brane in AdS 5 × S 5 is a candidate for the HEA for a U (1) factor of N = 4 SYM theory on the Coulomb branch. It incorporates all the required symmetries and dualities, and it gives the expected BPS soliton solutions. Even so, it might only be an approximation to the true HEA that is sufficient for computing susy-protected quantities. This would also be noteworthy, since then it would be a candidate for a new interacting UV finite abelian N = 4 quantum field theory in 4d. The existence of such a theory is unexpected. In either case, it is important to settle this question.
There is much more that remains to be explored. We need to understand the extent to which symmetry and other general considerations determine the HEA and whether the world-volume theory of a probe D3-brane should give this HEA. We should construct other analogous p -brane actions and explore their BPS soliton solutions. (BPS soliton solutions of the M5-brane example are discussed briefly in [2].) It should be illuminating to explore tree-approximation scattering amplitudes of the D3-brane theory. It seems likely that they will exhibit beautiful properties, maybe even a (spontaneously broken) Yangian symmetry. Finally, we would like to generalize the analysis to higher-rank gauge theories on the Coulomb branch, which have multiple abelian factors.
## Acknowledgment
The author wishes to acknowledges discussions and communications with Hee-Joong Chung, Frank Ferrari, Abhijit Gadde, Jerome Gauntlett, Sergei Gukov, Nicholas Hunter-Jones, Elias Kiritsis, Arthur Lipstein, Nick Manton, Hirosi Ooguri, Jaemo Park, Nati Seiberg, Savdeep Sethi, Yuji Tachikawa, David Tong, and Wenbin Yan.
This work was supported in part by DOE Grant # DE-SC0011632. The author acknowledges the hospitality of the Aspen Center for Physics, where he began working on this project in the summer of 2013, and where he wrote this manuscript in the summer of 2014. The ACP is supported by the National Science Foundation Grant No. PHY-1066293.
## References
- [1] J. H. Schwarz, 'Highly effective actions,' JHEP 1401 , 088 (2014) [arXiv:1311.0305 [hep-th]].
- [2] J. H. Schwarz, 'BPS soliton solutions of a D3-brane action,' arXiv:1405.7444 [hep-th].
- [3] L. Brink, J. H. Schwarz and J. Scherk, 'Supersymmetric Yang-Mills Theories,' Nucl. Phys. B 121 , 77 (1977).
- [4] J. M. Maldacena, 'The largeN limit of superconformal field theories and supergravity,' Adv. Theor. Math. Phys. 2 , 231 (1998) [hep-th/9711200].
- [5] D. Belov and G. W. Moore, 'Conformal blocks for AdS(5) singletons,' hep-th/0412167.
- [6] O. Aharony, N. Seiberg and Y. Tachikawa, 'Reading between the lines of fourdimensional gauge theories,' JHEP 1308 , 115 (2013) [arXiv:1305.0318 [hep-th]].
- [7] R. R. Metsaev and A. A. Tseytlin, 'Supersymmetric D3-brane action in AdS(5) x S**5,' Phys. Lett. B 436 , 281 (1998) [hep-th/9806095].
- [8] C. G. Callan and J. M. Maldacena, 'Brane dynamics from the Born-Infeld action,' Nucl. Phys. B 513 , 198 (1998) [hep-th/9708147].
- [9] S. Ferrara, R. Kallosh and A. Strominger, ' N = 2 extremal black holes,' Phys. Rev. D 52 , 5412 (1995) [hep-th/9508072].
- [10] F. Denef, 'Attractors at weak gravity,' Nucl. Phys. B 547 , 201 (1999) [hep-th/9812049].
- [11] F. Denef, 'Supergravity flows and D-brane stability,' JHEP 0008 , 050 (2000) [hepth/0005049].
- [12] J. P. Gauntlett, C. Kohl, D. Mateos, P. K. Townsend and M. Zamaklar, 'Finite energy Dirac-Born-Infeld monopoles and string junctions,' Phys. Rev. D 60 , 045004 (1999) [hep-th/9903156].
- [13] R. de Mello Koch, A. Paulin-Campbell and J. P. Rodrigues, 'Monopole dynamics in N = 2 super Yang-Mills theory from a three-brane probe,' Nucl. Phys. B 559 , 143 (1999) [hep-th/9903207].
- [14] C. V. Johnson, A. W. Peet and J. Polchinski, 'Gauge theory and the excision of repulson singularities,' Phys. Rev. D 61 , 086001 (2000) [hep-th/9911161].
- [15] I. A. Popescu and A. D. Shapere, 'BPS equations, BPS states, and central charge of N = 2 supersymmetric gauge theories,' JHEP 0210 , 033 (2002) [hep-th/0102169].
- [16] N. Seiberg and E. Witten, 'Electric-magnetic duality, monopole condensation, and confinement in N = 2 supersymmetric Yang-Mills theory,' Nucl. Phys. B 426 , 19 (1994) [Erratum-ibid. B 430 , 485 (1994)] [hep-th/9407087].
- [17] S. Bolognesi, 'Multi-monopoles, magnetic bags, bions and the monopole cosmological problem,' Nucl. Phys. B 752 , 93 (2006) [hep-th/0512133].
- [18] K. -M. Lee and E. J. Weinberg, 'BPS magnetic monopole bags,' Phys. Rev. D 79 , 025013 (2009) [arXiv:0810.4962 [hep-th]].
- [19] S. Bolognesi, 'Magnetic bags and black holes,' Nucl. Phys. B 845 , 324 (2011) [arXiv:1005.4642 [hep-th]].
- [20] N. S. Manton, 'Monopole planets and galaxies,' Phys. Rev. D 85 , 045022 (2012) [arXiv:1111.2934 [hep-th]].
| null |
[
"John H. Schwarz"
] |
2014-08-05T02:28:55+00:00
|
2014-08-05T02:28:55+00:00
|
[
"hep-th"
] |
Gauge Theories on the Coulomb branch
|
We construct the world-volume action of a probe D3-brane in $AdS_5 \times
S^5$ with $N$ units of flux. It has the field content, symmetries, and
dualities of the $U(1)$ factor of ${\cal N} =4$ $U(N+1)$ super Yang--Mills
theory, spontaneously broken to $U(N) \times U(1)$ by being on the Coulomb
branch, with the massive fields integrated out. This motivates the conjecture
that it is the exact effective action, called a `highly effective action'
(HEA). We construct an $SL(2,Z)$ multiplet of BPS soliton solutions of the
D3-brane theory (the conjectured HEA) and show that it reproduces the
electrically charged massive states that have been integrated out as well as
magnetic monopoles and dyons. Their charges are uniformly spread on a spherical
surface, called a `soliton bubble', which is interpreted as a phase boundary.
|
1408.0854v1
|
## Semi-Analytical Computation of Acoustic Scattering by Spheroids and Disks
Ross Adelman, Nail A. Gumerov, and Ramani Duraiswami
## Abstract
Analytical solutions to acoustic scattering problems involving nonspherical shapes, such as spheroids and disks, have long been known and have many applications. However, these solutions require special functions that are not easily computable. For this reason, their asymptotic forms are typically used since they are more readily available. We explore these solutions and provide computational software for calculating their nonasymptotic forms, which are accurate over a wide range of frequencies and distances. This software, which runs in MATLAB, computes the solutions to acoustic scattering problems involving spheroids and disks by semi-analytical means, and is freely available from our webpage.
## 1 Introduction
Despite the advent of efficient numerical solvers for many physical problems, analytical methods remain valuable. Many problems can be simplified to a point where analytical methods can be applied. These methods can provide insight into the problem and can help researchers when moving to a numerical solver. Numerical methods, on the other hand, have the advantage of being able to treat problems with arbitrary geometries. In order to do so, they return numerical approximations. The numerical error can usually be reduced by increasing the number of modeling elements. Without some kind of validation, though, there is no way to know whether the numerical solution converges to the correct one. Analytical methods can provide this validation.
The indirect boundary element method (IBEM) for the Helmholtz equation is a numerical method that requires this kind of validation [1]. The IBEM is capable of treating open, infinitely thin surfaces. These surfaces are good approximations to those often encountered in practice - those that are much smaller than a wavelength in one dimension, but span several in the other two. However, there is only one analytically tractable problem posed on such a surface that has an analytical solution and could be used to validate the IBEM: an acoustic wave scattering off a disk. The disk is actually the degenerate form of the oblate spheroid, so methods for solving scattering problems involving oblate spheroids can also be applied to the disk. This application is what lead us to the present problem. Scattering problems involving oblate spheroids, as well as the closely related prolate spheroids, have been studied for well over a century, and analytical expressions for computing their solutions have been documented extensively in the literature.
Figure 1. The prolate (left) and oblate (right) spheroidal coordinate systems. The three colored surfaces are isosurfaces for η = ± 1 / 2 (red), ξ = 3 2 for the prolate case and / ξ = 1 2 / for the oblate case (green), and φ = 0 (blue).
One of the earliest papers on the topic was by Lord Rayleigh in 1897 [2]. As discussed in [3], prior to the discovery and use of the spheroidal wave functions, the best solutions to these problems were approximations, either in frequency or distance. When separated in spheroidal coordinates, solutions to the Helmholtz equation can be written in terms of the spheroidal wave functions [4]. Because spheroids can be represented as isosurfaces in these coordinate systems, solutions to acoustic waves scattering off them can be written in terms of these special functions. The resulting expressions are accurate over a wide range of frequencies and distances. These expressions were applied to the disk in [5, 6], and they were validated experimentally in [7]. Spheroids of different sizes were studied in the following decades [8, 9], and much of this work was organized into an encyclopedic book [10], which also includes an extensive bibliography on the topic. More recently, these expressions were implemented and used to compute the solutions over a wide range of frequnecies and spheroid sizes [11, 12]. However, despite the immense amount of work that has been done on the topic, there are currently no publicly available libraries that implement these expressions. Also, in all of these references, the spheroids and disks were assumed to be entirely sound soft or sound hard, but the more general case of Robin boundary conditions was never considered in detail. This work represents our effort to correct this.
We have developed computational software for calculating the solutions to acoustic scattering problems involving prolate spheroids, oblate spheroids, and disks. This software is called scattering and runs in MATLAB. We have also developed software for computing the spheroidal wave functions required by scattering . This software is called spheroidal and is described in a separate paper [13]. Using spheroidal , we have precomputed many values of the spheroidal wave functions, which, along with scattering , are freely available for download from our webpage [14]. Together, they can be used to recreate the examples seen in this paper.
## 2 Spheroidal Coordinates
The prolate spheroidal coordinate system, shown in Figure 1, is related to the Cartesian coordinate system by [4]
$$x = a \left ( 1 - \eta ^ { 2 } \right ) ^ { 1 / 2 } \left ( \xi ^ { 2 } - 1 \right ) ^ { 1 / 2 } \cos \left ( \phi \right ), \quad y = a \left ( 1 - \eta ^ { 2 } \right ) ^ { 1 / 2 } \left ( \xi ^ { 2 } - 1 \right ) ^ { 1 / 2 } \sin \left ( \phi \right ), \quad z = a \eta \xi, \ \left ( 1 \right )$$
where 2 a is the interfocal distance. The Helmholtz equation, ∇ 2 V + k V 2 = 0, where k is the wavenumber, can be written in prolate spheroidal coordinates as
$$\left ( \frac { \partial } { \partial \eta } \left ( 1 - \eta ^ { 2 } \right ) \frac { \partial } { \partial \eta } + \frac { \partial } { \partial \xi } \left ( \xi ^ { 2 } - 1 \right ) \frac { \partial } { \partial \xi } + \frac { \xi ^ { 2 } - \eta ^ { 2 } } { \left ( 1 - \eta ^ { 2 } \right ) \left ( \xi ^ { 2 } - 1 \right ) } \frac { \partial ^ { 2 } } { \partial \phi ^ { 2 } } + c ^ { 2 } \left ( \xi ^ { 2 } - \eta ^ { 2 } \right ) \right ) V = 0, \quad ( 2 )$$
where c = ka . Applying the method of separation of variables yields three uncoupled ordinary differential equations, one for each coordinate:
$$\frac { \partial } { \partial \eta } \left ( \left ( 1 - \eta ^ { 2 } \right ) \frac { \partial } { \partial \eta } S _ { m n } \left ( c, \eta \right ) \right ) + \left ( \lambda _ { m n } - c ^ { 2 } \eta ^ { 2 } - \frac { m ^ { 2 } } { 1 - \eta ^ { 2 } } \right ) S _ { m n } \left ( c, \eta \right ) = 0,$$
$$\frac { \partial } { \partial \xi } \left ( \left ( \xi ^ { 2 } - 1 \right ) \frac { \partial } { \partial \xi } R _ { m n } \left ( c, \xi \right ) \right ) - \left ( \lambda _ { m n } - c ^ { 2 } \xi ^ { 2 } + \frac { m ^ { 2 } } { \xi ^ { 2 } - 1 } \right ) R _ { m n } \left ( c, \xi \right ) = 0,$$
$$\frac { \partial ^ { 2 } } { \partial \phi ^ { 2 } } \Phi _ { m } \left ( \phi \right ) + m ^ { 2 } \Phi _ { m } \left ( \phi \right ) = 0,$$
where m = 0 1 , , . . . and n = m,m +1 , . . . . The solutions to Eq. (3) are called the prolate spheroidal angle functions, and the solutions to Eq. (4) are called the prolate spheroidal radial functions. Collectively, they are called the prolate spheroidal wave functions. Any solution to Eq. (2) can be written as
$$V = \sum _ { m = 0 } ^ { \infty } \sum _ { n = m } ^ { \infty } S _ { m n } \left ( c, \eta \right ) \left ( A _ { m n } R _ { m n } ^ { \left ( 1 \right ) } \left ( c, \xi \right ) + B _ { m n } R _ { m n } ^ { \left ( 3 \right ) } \left ( c, \xi \right ) \right ) \cos \left ( m \phi \right ),$$
where the expansion coefficients, A mn and B mn , depend on the problem being solved.
The expression for an acoustic wave due to a point source is V ps = exp( ikr ) / (4 πr ), where r is the distance between the point source and evaluation point. When expanded in terms of the prolate spheroidal wave functions,
$$V ^ { p s } = \frac { i k } { 2 \pi } \sum _ { m = 0 } ^ { \infty } \sum _ { n = m } ^ { \infty } \frac { \epsilon _ { m } } { N _ { m n } \left ( c \right ) } S _ { m n } ^ { \left ( 1 \right ) } \left ( c, \eta _ { 0 } \right ) S _ { m n } ^ { \left ( 1 \right ) } \left ( c, \eta \right ) R _ { m n } ^ { \left ( 1 \right ) } \left ( c, \xi _ { < } \right ) R _ { m n } ^ { \left ( 3 \right ) } \left ( c, \xi _ { > } \right ) \cos \left ( m \phi \right ),$$
glyph[negationslash]
where ( η , ξ 0 0 , 0) and ( η, ξ, φ ) are the positions of the point source and evaluation point in prolate spheroidal coordinates, respectively (because of symmetry, φ 0 = 0 can be achieved by rotating the problem around the z axis), ξ < = min( ξ , ξ 0 ), and ξ > = max( ξ , ξ 0 ). Likewise, the expression for a plane wave is V pw = exp( i k · r ), where k is the wavevector and r is the position vector of the evaluation point. When expanded in terms of the prolate spheroidal wave functions,
$$V ^ { p w } = 2 \sum _ { m = 0 } ^ { \infty } \sum _ { n = m } ^ { \infty } \frac { \epsilon _ { m } \dot { \imath } ^ { n } } { N _ { m n } \left ( c \right ) } S _ { m n } ^ { \left ( 1 \right ) } \left ( c, \cos \left ( \theta _ { 0 } \right ) \right ) S _ { m n } ^ { \left ( 1 \right ) } \left ( c, \eta \right ) R _ { m n } ^ { \left ( 1 \right ) } \left ( c, \xi \right ) \cos \left ( m \phi \right ),$$
where, in this expression, k has been restricted to k = k (sin ( θ 0 ) , 0 cos ( , θ 0 )) (unrestricted values of k can be achieved by rotating the problem around the z axis).
Consider a prolate spheroid, which is described by the isosurface, ξ = ξ 1 . The prolate spheroid can be sound soft ([ V ] ξ = ξ 1 = 0), sound hard ([ dV/d n ] ξ = ξ 1 = 0), or Robin boundary conditions can be used ([ V + αdV/d n ] ξ = ξ 1 = 0 and α = constant). The Robin case lies somewhere between the other two: the prolate spheroid is sound soft when α = 0 and sound hard when α →∞ . Suppose an incident field, V i , due to either a point source or plane wave, is generated, which strikes the prolate spheroid. We wish to compute the scattered field, V s , from the incident field bouncing off the prolate spheroid. Depending on the boundary conditions and whether the incident field is due to a point source or plane wave, the exact method of solution is slightly different, but they all follow the same procedure: (1) the incident field is expanded in terms of the prolate spheroidal wave functions using either Eq. (7) or (8); (2) the same is done for the scattered field using Eq. (6); (3) the two expressions are added together to form an expression for the total field in terms of the prolate spheroidal wave functions; (4) the total field and normal derivative are evaluated at the boundary; and (5) by using the orthogonality of the prolate spheroidal wave functions, the resulting expression is used to determine the expansion coefficients, A mn and B mn , so that the boundary conditions are satisfied. For the Robin case and a point source,
$$A _ { m n } = 0, \ \ B _ { m n } = - \frac { i k } { 2 \pi } \frac { \epsilon _ { m } } { N _ { m n } \left ( c \right ) } S _ { m n } ^ { \left ( 1 \right ) } \left ( c, \eta _ { 0 } \right ) R _ { m n } ^ { \left ( 3 \right ) } \left ( c, \xi _ { 0 } \right ) \frac { R _ { m n } ^ { \left ( 1 \right ) } \left ( c, \xi _ { 1 } \right ) + \alpha R _ { m n } ^ { \left ( 1 \right ) \prime } \left ( c, \xi _ { 1 } \right ) } { R _ { m n } ^ { \left ( 3 \right ) } \left ( c, \xi _ { 1 } \right ) + \alpha R _ { m n } ^ { \left ( 3 \right ) \prime } \left ( c, \xi _ { 1 } \right ) }.$$
For the Robin case and a plane wave,
$$A _ { m n } = 0, \ \ B _ { m n } = - 2 \frac { \epsilon _ { m } i ^ { n } } { N _ { m n } \left ( c \right ) } S _ { m n } \left ( c, \cos \left ( \theta _ { 0 } \right ) \right ) \frac { R _ { m n } ^ { \left ( 1 \right ) } \left ( c, \xi _ { 1 } \right ) + \alpha R _ { m n } ^ { \left ( 1 \right ) \prime } \left ( c, \xi _ { 1 } \right ) } { R _ { m n } ^ { \left ( 3 \right ) } \left ( c, \xi _ { 1 } \right ) + \alpha R _ { m n } ^ { \left ( 3 \right ) \prime } \left ( c, \xi _ { 1 } \right ) }. \quad \left ( 1 0 \right )$$
For a prolate spheroid that is sound soft ( α = 0) or sound hard ( α →∞ ), these expressions reduce to those in [10].
The expressions arising in the oblate case are very similar to (and sometimes exactly the same as) those arising in the prolate case. In many cases, simply letting c, ξ →-ic, iξ provides a transformation from the prolate case to the oblate case [4]. Indeed, the preceeding paragraphs and equations for the prolate case can be transformed into those for the oblate case by using this transformation. The oblate spheroidal coordinate system is shown in Figure 1.
## 3 Scattering Routines
We have developed computational software for calculating the solutions to acoustic scattering problems involving prolate spheroids, oblate spheroids, and disks. This software is called scattering and runs in MATLAB. The spheroids and disks can be sound soft, sound hard, or Robin boundary conditions can be used, and the incident field can be due to either a point source or plane wave. Internally, scattering needs to compute the spheroidal wave functions of different order and degree for different values of their argument. We have developed
Figure 2. A plane wave scattering off a sound-hard prolate spheroid (left), oblate spheroid (center), and disk (right) for k = 10 (top) and k = 25 (bottom).
software for doing so. This software is called spheroidal and is described in a separate paper [13].
There are 18 routines in scattering , each one solving a different scattering problem. For example, one is called pro\_plane\_wave\_scat\_hard , which computes the scattered field from a plane wave striking a sound-hard prolate spheroid. They are organized into 18 separate MATLAB M-Files, one for each routine. At the top of each M-File is a detailed explanation of the calling convention, return values, and so on.
Each routine has a slightly different calling convention depending on what is being computed, but they all follow the same pattern. They all require k and a , as well as information about the incident field, either η 0 and ξ 0 for a point source or θ 0 for a plane wave. Routines that compute a scattered field also require ξ 1 . For example, to compute the scattered field from a plane wave striking a sound-hard prolate spheroid, the following MATLAB code fragment can be used:
## v\_scat = pro\_plane\_wave\_scat\_hard(10.0, 1.0, pi, 'saved', 1.5, x, y, z);
where, in this case, k = 10, a = 1, θ 0 = π , 'saved' is the directory in which the spheroidal wave functions have been precomputed and saved, and ξ 1 = 3 2. The variables, / x , y , and z , are row vectors, which contain the positions of the evaluation points in Cartesian coordinates, and v\_scat is a row vector, which will contain V s .
Figure 3. The two figures on the left show the field and normal derivative along the boundary of an oblate spheroid from a plane wave striking the oblate spheroid from directly above for five different choices of the boundary conditions. The one figure on the right shows that the boundary conditions have been satisfied in all five cases.
## 4 Conclusion
We have developed computational software for calculating the solutions to acoustic scattering problems involving prolate spheroids, oblate spheroids, and disks. While these problems have been studied for well over a century and analytical expressions for computing their solutions have been documented by many authors, there are currently no publicly available libraries that implement them. This paper gave an overview of these problems, derived their analytical solutions, described some theory behind the special functions required by them, and included several examples of running our software. We hope that our software will better enable research involving spheroids and disks.
## References
- [1] N. A. Gumerov and R. Duraiswami, 'A broadband fast multipole accelerated boundary element method for the three dimensional Helmholtz equation', J. Acoust. Soc. Am. 125(1) (2009).
- [2] Lord Rayleigh, 'On the passage of waves through aperatures in plane screens, and allied problems', Philos. Mag. 43(5) (1897).
- [3] C. J. Bouwkamp, 'On the freely vibrating circular disk and the diffraction by circular disks and apertures', Physica 16(1) (1950).
- [4] C. Flammer, Spheroidal wave functions (Dover Publications, Mineola) (2005).
- [5] R. D. Spence, 'The diffraction of sound by circular disks and aperatures', J. Acoust. Soc. Am. 20(4) (1948).
- [6] A. Storruste and H. Wergeland, 'On two complementary diffraction problems', Phys. Rev. 73 (1948).
- [7] A. Leitner, 'Diffraction of sound by a circular disk', J. Acoust. Soc. Am. 21(4) (1949).
- [8] G. Chertock, 'Sound radiation from prolate spheroids', J. Acoust. Soc. Am. 33(7) (1961).
- [9] T. B. A. Senior, 'The scattering from acoustically hard and soft prolate spheroids for axial incidence', Can. J. Phys. 41 (1966).
- [10] J. J. Bowman, T. B. A. Senior, and P. L. E. Uslenghi, Electromagnetic and acoustic scattering by simple shapes (Hemisphere Publishing Corporation, New York) (1987).
- [11] J. P. Barton, N. L. Wolff, H. Zhang, and C. Tarawneh, 'Near-field calculations for a rigid spheroid with an arbitrary incident acoustic field', J. Acoust. Soc. Am. 113(3) (2003).
- [12] J. A. Roumeliotis, A. D. Kotsis, , and G. Kolezas, 'Acoustic scattering by an impenetrable spheroid', Acoustical Physics 53(4) (2007).
- [13] R. Adelman, N. A. Gumerov, and R. Duraiswami, 'Software for computing the spheroidal wave functions using arbitrary precision arithmetic', (2014), preprint, http://arxiv.org/abs/1408.0074.
- [14] http://www.github.com/radelman/scattering.
|
10.1121/1.4901318
|
[
"Ross Adelman",
"Nail A. Gumerov",
"Ramani Duraiswami"
] |
2014-08-05T03:18:28+00:00
|
2014-08-05T03:18:28+00:00
|
[
"cs.MS",
"cs.SD",
"physics.comp-ph"
] |
Semi-Analytical Computation of Acoustic Scattering by Spheroids and Disks
|
Analytical solutions to acoustic scattering problems involving nonspherical
shapes, such as spheroids and disks, have long been known and have many
applications. However, these solutions require special functions that are not
easily computable. For this reason, their asymptotic forms are typically used
since they are more readily available. We explore these solutions and provide
computational software for calculating their nonasymptotic forms, which are
accurate over a wide range of frequencies and distances. This software, which
runs in MATLAB, computes the solutions to acoustic scattering problems
involving spheroids and disks by semi-analytical means, and is freely available
from our webpage.
|
1408.0855v1
|
## NUMERICAL SIMULATIONS OF IMPULSIVELY GENERATED ALFV ´ N WAVES IN SOLAR MAGNETIC E ARCADES
P. Chmielewski , K. Murawski , Z.E. Musielak 1 1 2 3 , , A.K. Srivastava 4 Draft version August 6, 2014
## ABSTRACT
We perform numerical simulations of impulsively generated Alfv´n waves in an isolated solar arcade, e which is gravitationally stratified and magnetically confined. We study numerically the propagation of Alfv´n waves along such magnetic structure that extends from the lower chromosphere, where the e waves are generated, to the solar corona, and analyze influence of the arcade size and width of the initial pulses on the wave propagation and reflection. Our model of the solar atmosphere is constructed by adopting the temperature distribution based on the semi-empirical VAL-C model and specifying the curved magnetic field lines that constitute the asymmetric magnetic arcade. The propagation and reflection of Alfv´n waves in this arcade is described by 2.5D magnetohydrodynamic equations that e are numerically solved by the FLASH code. Our numerical simulations reveal that the Alfv´n wave e amplitude decreases as a result of a partial reflection of Alfv´n waves in the solar transition region, and e that the waves which are not reflected leak through the transition region and reach the solar corona. We also find the decrement of the attenuation time of Alfv´n waves for wider initial pulses. e Moreover, our results show that the propagation of Alfv´n waves in the arcade is affected by spatial dependence e of the Alfv´n speed, which leads to phase-mixing that is stronger for more curved and larger magnetic e arcades. We discuss processes that affect the Alfv´n wave propagation in an asymmetric solar arcade e and conclude that besides phase-mixing in the magnetic field configuration, plasma properties of the arcade and size of the initial pulse as well as structure of the solar transition region all play a vital role in the Alfv´n wave propagation. e
Subject headings: Sun: atmosphere, Sun : corona, magnetohydrodynamics (MHD), magnetic fields
## 1. INTRODUCTION
Arcades are gravitationally stratified and magnetically confined structures in the solar atmosphere. Their physical properties are determined by flows of plasma along these arcades and the heating resulting from the energy dissipation by different waves ( ˇ adez et al. C 1994, Innes et al. 2003, McKenzie & Savage 2009). Investigations of magnetic topology of such arcades (Biskamp & Welter 1989, Miki´ et al. c 1989) and magnetohydrodynamic (MHD) wave propagation in these structures (e.g., Del Zanna et al. 2005, Selwa et al. 2005, Selwa et al. 2006, D´ ıaz et al. 2006) are potential areas of research that reveal both the wave heating and plasma dynamics. In more recent studies carried on by Gruszecki & Nakariakov (2011), the excitation of slow magnetoacustic waves
1 Group of Astrophysics, UMCS, ul. Radziszewskiego 10, 20031 Lublin, Poland
2 Department of Physics, University of Texas at Arlington, Arlington, TX 76019, USA
3 Kiepenheuer-Institut f¨r u Sonnenphysik, Sch¨ oneckstr. 6, 79104 Freiburg, Germany
4 Department of Physics, Indian Institute of Technology (Banaras Hindu University), Varanasi-221005, India
was investigated in a system of magnetic arcades. Specifically, the antisymmetric kink mode of magnetoacustic waves was modelled in the magnetic arcades and used for determination of the arcade magnetic field, which is often referred as a MHD seismology.
Theoretical predictions resulting from the above studies have been supplemented by several significant efforts to observe such waves in the magnetic arcades. Verwichte et al. (2004) detected the damped standing kink oscillations in the arcades using the Transition Region and Coronal Explorer (TRACE) observations. Verwichte et al. (2010) reported on the transversal kink oscillations in the coronal arcades obtained by SOHO/EIT and STEREOobservations, and used the results to invoke the spatial MHD seismology of these magnetic structures to estimate their local plasma conditions.
The previous studies of magnetoacoustic waves were also supplemented by investigations of the propagation of purely incompressible transverse (Alfv´n) waves in such e magnetic arcades in the solar atmosphere. Since the magnetic field serves as a guide for Alfv´n waves, it is e worth to study these waves in magnetic field structures
like arcades. Arregui et al. (2004) studied MHD waves, including Alfv´n, coupled Alfv´n and fast magnetoacouse e tic (fast, henceforth) waves, and the magnetic arcade oscillations connected with the propagation of these waves. In a cold plasma approximation, they showed that the linear coupling of discrete fast modes, which are characterized by a discrete spectrum of frequencies and a global velocity structure, and Alfv´n continuum modes, which e are characterized by a continuous spectrum of frequencies and a velocity perturbation confined to given magnetic surfaces, leads to modes with mixed properties that arise in the magnetic arcades.
Similarly, Oliver et al. (1996) also investigated the mixed fast magnetoacoustic and Alfv´n waves in the form e of MHD perturbations excited in the magnetic arcades. Due to the resonant coupling, Rial et al. (2010) found that the fast wave transfers its energy to Alfv´ enic oscillations localized around a particular magnetic surface within the arcade, thus producing the attenuation of the initial fast MHD mode. Moreover, they discussed another case in which the generated fast wavefront leaves its energy on several magnetic surfaces within the arcade. The system is therefore able to trap energy in the form of Alfv´ enic oscillations, which are subsequently phasemixed to smaller spatial scales.
Gruszecki et al. (2007) considered the energy loss by Alfv´ en waves for straight and curved magnetic fields, and found a relation between the energy loss and the amplitude of the initial wave pulse. Results clearly showed that Alfv´n waves in a magnetic arcade experience a dee crease of their amplitude and that this decrease is caused by partial reflection of the waves from the solar transition region (Gruszecki at al. 2007). The partial reflection of Alfv´n waves results from a steep gradient of Alfv´n e e speed in the transition region, because of a significant density drop in this region of the solar atmosphere.
An important problem of Alfv´ en waves chromospheric leakage in coronal loops was studied by Ofman (2002) in a normalized visco-resistive nonlinear 1.5D MHD model, for which the Alfv´n leakage time was computed. Moree over, Ofman & Wang (2007) constructed a MHD model that allowed them to examine the periodic oscillations in their coronal loop and compare the results those observed by Solar Optical Telescope (SOT) on the board of Hinode.
Del Zanna et al. (2005) performed numerical studies of Alfv´ en waves triggered by solar flare events in an isothermal and symmetric arcade model, and showed that these waves propagate back and forth along the arcade and rapidly decay. According to the authors, the fast decay of these waves explains the observed damping of the am- plitudes of oscillations, and the efficiency of Alfv´n waves e decay is related to the stratification of solar atmosphere as well as to a gradient of magnetic field along the arcade. Miyagoshi et al. (2004) investigated oscillations of a symmetric solar arcade and show that the amplitude of the loop oscillations decreases exponentially in time due to the energy transport by fast-mode MHD waves.
Based on the above results, we may conclude that curved magnetic fields, such as those existing in solar magnetic arcades, effectively increase the wave energy leakage. Moreover, the stronger Alfv´n wave pulse amplie tudes are the more efficient generation of fast magnetoacoustic waves occurs. Nevertheless, additional studies, in which the solar atmosphere model is extended to include the solar chromosphere and transition region and a broader range of physical parameters for an asymmetric magnetic field configuration is considered, are urgently needed. Therefore, the main goal of this paper is to perform such studies and significantly extend the previous work of Del Zanna et al. (2005) and Miyagoshi et al. (2004).
In our approach, we consider an asymmetric, stratified and non-isothermal solar arcade, assume that the temperature profile along the arcade is based on the VAL-C semi-empirical model of the solar atmosphere (Vernazza et al. 1981), and follow Low (1985) to construct a magnetic field model of the arcade. An important difference between our approach and that presented by Del Zanna et al. and Miyagoshi et al. is that we impulsively generate Alfv´n waves by launching them in the solar chromoe sphere, just above the solar photosphere, whereas in Del Zanna's work the waves are triggered by an onset of solar flare in the solar corona. Since flares are highly episodic transients that only occur in solar active regions, where even more complex loops can exist with highly changeable ambient medium, Del Zanna et al. (2005) model is designed to study short lived Alfv´n waves in post flare e loops, which are most dynamic and transient loop systems. Similarly, Miyagoshi et al. (2004) implemented transversal velocity pulses at the top of the coronal loop to excite the waves. In contrary, our model allows us to investigate the effects caused by trains of Alfv´n waves e resulting from wave reflection in the solar transition region in a stable and more realistic solar coronal arcade. As a consequence of this fundamental difference between these two models, we are able to explore different physical aspects of the Alfv´n wave propagation in the are cade than those studied by Del Zanna et al. (2005) and Miyagoshi et al. (2004).
In this paper, we use the publicly available numerical code FLASH to perform parametric studies of the prop-
agation of impulsively generated Alfv´n waves in the are cade by varying an initial pulse position and its width. Our model of solar arcade and the location of Alfv´ en wave sources in the solar chromosphere (just above the solar photosphere) allows us to investigate the efficiency of Alfv´n wave propagation through the solar transition e region, the resulting partial wave reflection, and the wave energy leakage through this region. We use our numerical results to identify the basic physical processes that affect the propagation of Alfv´n waves in our solar arcade e model.
The paper is organized as follows. Our model of solar magnetic arcades and description of our numerical method are introduced in Sects. 2 and 3, respectively. Results of our numerical simulations of the Alfv´n wave e propagation in magnetic arcades are presented and discussed in Sect. 4. Conclusions are given in Sect. 5.
## 2. NUMERICAL MODEL OF ALFV ´ EN WAVES
## 2.1. MHD equations
We consider a gravitationally stratified and magnetically confined plasma in a structure that resembles an arcade, which is described by the following set of ideal MHD equations:
$$\partial _ { \overline { \partial } t } ^ { \varrho } + \nabla \cdot ( \varrho V ) = 0, ( 1 ) \quad & \text{em} \\ \varrho \frac { \partial _ { \overline { \partial } t } } { \partial t } + \varrho ( V \cdot \nabla ) V = - \nabla p + \frac { 1 } { \mu } ( \nabla \times B ) \times B + \varrho g, ( 2 ) \quad & y = \\ \frac { \partial B } { \partial t } = \nabla \times ( V \times B ), ( 3 ) \quad & \text{em} \\ \nabla \cdot B = 0, ( 4 ) \quad & \text{field} \\ \frac { \partial p } { \partial t } + \nabla p = - \gamma p \nabla \cdot V, ( 5 ) \quad & \text{asyn} \\ p = \frac { k _ { B } } { m } \varrho T, ( 6 ) \quad & \text{as} = \beta \\ \text{where } \varrho \text{ is mass density, } p \text{ is gas pressure, } V, \text{$B$ and} \\ \varrho = ( 0 - a 0 ) \text{ represent the nlasma velocity, the max}$$
where /rho1 is mass density, p is gas pressure, V B , and g = (0 , -g, 0) represent the plasma velocity, the magnetic field and gravitational acceleration, respectively. In addition, T is a temperature, m is a particle mass, that was specified by mean molecular weight value of 1 24 (Os-. kar Steiner, private communication), k B is Boltzmann's constant, γ = 5 3 is the adiabatic index, / and µ is the magnetic permeability of plasma. The value of g is equal to 274 m s -2 .
Note that Del Zanna et al. (2005) worked in the framework of isothermal plasma assumption, γ = 1, which resulted in the elimination of energy equation and thermal pressure was described by the ideal gas law of Eq. (6). This approximation greatly simplifies numerical methods, particularly in regions of strongly magnetized plasma, where negative gas pressure can set in.
## 2.2. A model of the static solar atmosphere
We consider a model of the static ( ∂/∂t = 0) solar atmosphere with an invariant coordinate z ( ∂/∂z = 0), but allow the z -components of velocity ( V z ) and magnetic field ( B z ) to vary with x and y . In such 2.5 dimensional (2.5D) model, the solar atmosphere is in static equilibrium ( V e = 0 ) with the force-free and current-free magnetic field defined by
$$( \nabla \times B _ { e } ) \times B _ { e } = 0 \,, \quad \nabla \times B _ { e } = 0 \,. \quad \, ( 7 )$$
Here the subscript e corresponds to equilibrium quantities. We adopt a realistic magnetic flux-tube model originally developed for 3D by Low (1985), in which
$$\mathbf B _ { \mathrm e } ( x, y ) = \nabla \times \left ( A \hat { \mathbf z } \right ) \,,$$
is an equilibrium magnetic field with ˆ z being a unit vector along the z -direction and A denoting the magnetic flux function given by
$$A ( x, y ) = \frac { x ( y _ { \text{ref} } - b ) ^ { 2 } } { ( y - b ) ^ { 2 } - x ^ { 2 } } B _ { \text{ref} } \,. \quad \quad ( 9 )$$
Here b is a constant, which we choose and set equal to, b = -5 Mm, and B ref is the magnetic field at the reference level, y ref = 10 Mm.
Vectors of magnetic field, resulting from Eq. (8), are displayed in Fig. 1. The magnetic field is defined by the magnetic pole that is located at the point ( x = 0 Mm, y = -5 Mm). Note that at a given altitude y magnetic field is strongest around the line x = 0 Mm, where it is essentially vertical. However, further out the magnetic field declines with larger values of | x | , revealing its curved structure. Such magnetic field corresponds to an isolated asymmetric magnetic arcade.
As a result of Eqs. (2) and (7), the pressure gradient is balanced by the force of gravity,
$$- \nabla p _ { e } + \varrho _ { e } g = 0 \,. \quad \quad \quad ( 1 0 )$$
With the use of the ideal gas law given by Eq. (6) and the y -component of the hydrostatic pressure balance described by Eq. (10), we express the equilibrium gas pressure and mass density as
$$p _ { \text{e} } ( y ) = p _ { \text{ref} } \, \exp \left ( - \int _ { y _ { r } } ^ { y } \frac { d y ^ { ^ { \prime } } } { \Lambda ( y ^ { ^ { \prime } } ) } \right ) \,, \quad \ \ ( 1 1 )$$
$$\overset { \cdot } { \varrho } _ { \text{e} } ( y ) = \frac { p _ { \text{e} } ( y ) } { g \Lambda ( y ) } \,, \quad \ \ ( 1 2 )$$
where
$$\Lambda ( y ) = \frac { k _ { \text{B} } T _ { \text{e} } ( y ) } { m g } \, \quad \quad \quad ( 1 3 )$$
is the pressure scale-height, and p ref denotes the gas pressure at the reference level.
Fig. 1.Vectors of equilibrium magnetic field.
Fig. 2.Equilibrium profiles of the temperature along a fixed x (top panel) and the Alfv´n speed (bottom panel). e
We adopt a realistic plasma temperature profile given by the semi-empirical VAL-C model (Vernazza et al. 1981) that is extrapolated into the solar corona, (Fig. 2, top panel). In our model, the temperature attains a value of about 5 × 10 3 K at y = 1 5 Mm and . it increases to about 1 5 . × 10 6 K in the solar corona at y = 10 Mm. Higher up in the solar corona the temperature is assumed to be constant. The temperature profile determines uniquely the equilibrium mass density and gas pressure profiles. At the transition region, which is located at y /similarequal 2 7 Mm, . T e exhibits an abrupt jump (Fig. 2, top panel), but /rho1 e ( y ) and p e ( y ) experience a sudden fall off with the atmospheric height (not shown).
In this model the Alfv´n speed, e c A , varies in both the x and y directions and it is expressed as follows:
$$c _ { \text{A} } ( x, y ) = \frac { | \text{B} _ { \text{e} } ( x, y ) | } { \sqrt { \mu \varrho _ { \text{e} } ( y ) } } \,. \quad \quad ( 1 4 ) \\ \text{is displayed in Fig. 2 (bottom panel). Note} \\ \text{$a$} \, \text{and} \, \colon \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \, \text{$a$} \.$$
Its profile is displayed in Fig. 2 (bottom panel). Note that the Alfv´n speed is non-isotropic; higher values are e located near x 0 = 0 Mm, where the magnetic field is stronger, while c A decreases with larger values of x 0 . In the chromosphere c A ( x = 0Mm , y = 1 75 Mm) is about . 25 km s -1 . The Alfv´ en speed rises abruptly through the solar transition region, reaching a value of c A ( x = 0 Mm , y = 10 Mm) = 10 3 km s -1 (Fig. 2, bottom panel). The increase of c A ( x = 0 Mm , y ) with height results from a faster decrease of /rho1 e ( y ) than B e ( x = 0 Mm , y ) with the atmospheric height.
## 3. NUMERICAL SIMULATIONS OF MHD EQUATIONS
To solve Eqs (1)-(6) numerically, we use the FLASH code (Fryxell et al. 2000; Lee & Deane 2009; Lee 2013), in which a third-order unsplit Godunov-type solver with various slope limiters and Riemann solvers as well as Adaptive Mesh Refinement (AMR) (MacNeice et al. 1999) are implemented. The minmod slope limiter and the Roe Riemann solver (e.g., T´th 2000) are o used. We set the simulation box as (1 Mm 28 5 Mm) , . × (1 Mm 11Mm) and impose fixed in time boundary condi-, tions for all plasma quantities in the x - and y -directions, while all plasma quantities remain invariant along the z -direction; however, note that both V z and B z differ from zero.
In our present work, we use a static, non-uniform grid with a minimum (maximum) level of refinement set to 4 (6). We performed the grid convergence studies by refining this by a factor of two. As the numerical results remained essentially similar for the grid of maximum blocks levels 6 and 7, we adopted the former to get the results presented in this paper. Note that small size blocks of numerical grid occupy the solar transition region and the region located along Alfv´n wave propagae tion path (Fig. 3), and every numerical block consists of
Fig. 3.Numerical blocks used in the numerical simulations for the case of x 0 = 3 Mm.
Fig. 4.Spatial profiles of the velocity V z ( x, y, t ) near the point of initial perturbation for the case of x 0 = 2 Mm, w x = 0 1 Mm . at t = 5 s, t = 8 s, t = 11 s, t = 14 s, t = 18 s and t = 21 s, presenting propagation through the transition region and evolution of the initial pulse. A full-colour version of above figure and movie is available at www.pchmiel.republika.pl/store/Fig4.avi.
Fig. 5.The Alfv´ en speed, c A , along the magnetic line, which crosses the point ( x , y 0 0 ) = (2 Mm 1 75 Mm) (solid , . line) and the point ( x , y 0 0 ) = (3 Mm 1 75 Mm) (dashed line). , . 8 × 8 identical numerical cells. This results in an excellent resolution of steep spatial profiles and greatly reduces the
numerical diffusion in these regions.
## 3.1. Initial perturbations
We perturb initially (at t = 0 s) the equilibrium described in Sec. 2.2 by a Gaussian pulse in the z -component of velocity given by
$$V _ { z } ( x, y, t = 0 ) = A _ { \nu } e ^ { - \frac { ( x - x _ { 0 } ) ^ { 2 } } { w _ { x } ^ { 2 } } - \frac { ( y - y _ { 0 } ) ^ { 2 } } { w _ { y } ^ { 2 } } } \,, \quad ( 1 5 ) \quad \begin{matrix} \text{(pa)} \\ \text{th} \end{matrix} \,.$$
where A v is the amplitude of the pulse, ( x , y 0 0 ) is its initial position and w x and w y denote its widths along x -and y -directions, respectively. We set and hold fixed y 0 = 1 75 Mm, . A v = 3 km s -1 and w y = 0 1 Mm, but . allow x 0 and w x to vary. This shows that Alfv´n waves in e our model are generated in the solar chromosphere just above the photosphere, which makes our model significantly different than that considered by Del Zanna et al. (2005) and Miyagoshi et al. (2004), who launched the initial pulses in the solar corona. Physical consequences of these differences are described in Sect. 4. It is clear that our more realistic model significantly extends the model considered by Del Zanna et al. (2005) and provides a platform to pursue the modelling of Alfv´n wave phasee mixing in realistic solar atmosphere where an appropriate driver (dynamical phenomena in the solar chromosphere) is responsible for excitation of such waves.
Note that in our 2.5D model, the Alfv´n waves decoue ple from magnetoacoustic waves and it can be described solely by V z ( x, y, t ). As a result, the initial pulse triggers
Alfv´ en waves that in the linear limit are approximately described by the following wave equation:
$$\frac { \partial ^ { 2 } V _ { z } } { \partial t ^ { 2 } } = c _ { \mathrm A } ^ { 2 } ( x, y ) \frac { \partial ^ { 2 } V _ { z } } { \partial s ^ { 2 } } \,, \quad \quad ( 1 6 )$$
where s is the coordinate along a magnetic field line.
## 4. RESULTS OF NUMERICAL SIMULATIONS
We simulate small amplitude and impulsively excited Alfv´ en waves and investigate their propagation along magnetic field lines which are parallel to magnetic vectors shown in Fig. 1. This magnetic field configuration mimics an asymmetric solar arcade, which is more realistic than that considered by Del Zanna et al. (2005).
## 4.1. Initial stage of wave propagation
The Gaussian pulse of initial perturbation in the z -component of velocity, described by Eq. (15), decouples into two moving in opposite directions pulses. The downwardly propagating pulse fades very fast, however the second one propagates through the chromosphere and transition region to the solar corona (Fig. 4).
As a result of several effects such as finite size of the Alfv´ en wave pulse, inclination of the magnetic field, and highly inhomogeneous Alfv´ en speed, c A ( x, y ), the wave signal in our numerical simulations is deformed and elongated while passing in the arcade through the transition region. In Fig. 4, we observe the Alfv´n waves reaching e the transition region located at y /similarequal 2 7 . Mm, and accelerated due to a sudden increase of local Alfv´n speed e (panels a and b ). The wave signal penetrates gradually through the transition region. First, the initially circular wave profile is distorted to ellipsoidal shape because Alfv´ en speed slightly raise along the inclined magnetic field lines (Fig. 5), and next becomes significantly elongated above the transition region (Fig. 4, panel c ) due to sudden increase of the Alfv´n speed (Fig. 5). e
This clearly shows that the upper part of the wave signal, propagating along the magnetic field lines, penetrates the solar corona earlier than the middle and bottom parts of the signal (Fig. 4, panel b ). Moreover, the upper part of the wave follows along different (upper) magnetic field line than the right part of the wave that propagates over the lower magnetic field line, which results in deformation of the initial pulse and finally intensifies the phase-mixing that can be seen at the panels d and e of Fig. 4. Here, the upper part of the wave signal located at the upper magnetic field lines illustrated as a white patch (Fig. 4, panels d and e ) is in a different phase than the signal located at the lower magnetic field line, that already passed the transition region. This additional process of the wave deformation at the transition region
Fig. 6.Spatial profiles of transverse velocity V z ( x, y, t ) for x 0 = 2 Mm, w x = 0 2 Mm at . t = 16 s , t = 26 s, t = 36 s, t = 46 s, t = 56 s, t = 66 s, t = 76 s, t = 86 s, t = 96 s and t = 106 s (from top-left to bottom-right). A full-colour version of above figure and movie is available at www.pchmiel.republika.pl/store/Fig6.avi.
significantly influences Alfv´ en wave propagation in the curved coronal arcades, introducing different phases of the wave signal in the arcade at beginning stage. Notice, that the Alfv´n waves suffer from the partial reflection e from the transition region and a small signal propagates backward into the solar surface, which can be spotted at ( x, y ) /similarequal (2 Mm 1 8 Mm) in Fig. 4, panels , . e and f .
## 4.2. Different spatial positions of initial pulses
In our approach for a fixed value of y 0 , the length and curvature of magnetic field lines vary with x 0 . For a smaller value of x 0 a magnetic line is longer, more curved, and less inclined to the horizontal direction (Fig. 1) as well as Alfv´n speed e c A ( x, y ) experiences a larger structuring there (Fig. 2, bottom panel). While a larger value of x 0 corresponds to a shorter, less curved and more in-
Fig. 7.Spatial profiles of transverse velocity V z ( x, y, t ) for x 0 = 2 Mm, w x = 0 2 Mm at . t = 36 s, t = 40 s, t = 44 s, t = 48 s, t = 52 s, t = 56 s, t = 60 s and t = 64 s (from top-left to bottom-right), illustrating a partial reflection of Alfv´n waves from e the transition region.
Fig. 8.Spatial profiles of transverse velocity V z ( x, y, t ) for x 0 = 3 Mm, w x = 0 2 Mm at . t = 16 s, t = 20 s, t = 24 s, t = 28 s, t = 32 s, t = 36 s, t = 40 s, t = 44 s, t = 48 s and t = 52 s (from top-left to bottom-right). A full-colour version of above figure and movie is available at www.pchmiel.republika.pl/store/Fig8.avi.
Fig. 9.Vertical profiles of V z ( x = 12Mm) for the case of x 0 = 2Mm (top panel) at t = 22 s (solid line), t = 74 s (dashed line) and t = 124 s (dotted line), and V z ( x = 9 Mm) for the case of x 0 = 3Mm (bottom panel) at t = 20 s (solid line), t = 40 s (dashed line) and t = 60 s (dotted line).
clined magnetic field lines as well as less varying c A ( x, y ). Here, we consider two different cases of initial pulse position: x 0 = 2 Mm and x 0 = 3 Mm. In the first case, the Alfv´n waves propagate higher up along longer and e more curved magnetic field lines. However, in the case of x 0 = 3 Mm, the inclination of magnetic field lines to x -axis is much larger and this results in approximately two times smaller size of the arcade with less curved field lines. In Fig. 5, we notice that c A ( s ) varies less along the magnetic field line, s , for a small arcade, reaching a value of 2 Mm s -1 (dashed line), while for a large arcade it attains 2 4 . Mm s -1 (solid line). Here, a value of s = 0 corresponds to the point at which the initial perturbation was launched, ( x , y 0 0 ), equal to (2 Mm 1 75 Mm) , . for the large arcade (solid line) and (3 Mm , 1 75 Mm) for . the small arcade (dashed line).
## 4.2.1. Large arcade
Our numerical results are presented in Fig. 6, which illustrates spatial profiles of Alfv´n waves generated by the e initial Gaussian pulse given by Eq. (15). This pulse triggers counter-propagating Alfv´n waves (e.g., e Murawski & Musielak 2010). The wave that propagates downward enters a region of small Alfv´n speed, which results from e high mass density there. The presented results show that the evolution of the downwardly propagating waves (Fig. 4) is much less dynamic than the evolution of Alfv´n e
Fig. 10.Time-signatures of V z for w x = 0 2 Mm, . x 0 = 2 Mm(top panel), x 0 = 2 5 Mm (middle panel) and . x 0 = 3 Mm (bottom panel). Detection points are located at ( x = 12 Mm, y = 7 Mm) (top panel), ( x = 11 Mm, y = 5 1 Mm) (middle . panel) and ( x = 9 Mm, y = 4 2 Mm) (bottom panel). .
waves that already reach the upper regions of the solar atmosphere.
The upwardly propagating Alfv´ en waves, reaching the transition region, accelerate due to increase of local Alfv´ en speed (Fig. 2, bottom panel, and Fig. 5, solid line) and penetrate into the solar corona. In Fig. 6, panel a , the wave signal arrived to the level y = 9 Mm. At t = 26 s (panel b ), the Alfv´ en waves are at the apex while fanning its spatial profile as a result of diverging magnetic field lines (Fig. 1), and the spatial variation of the Alfv´n speed (Fig. 2, bottom panel). e This process becomes even more pronounced in time as the Alfv´ en waves penetrate the regions of more diverged field lines. Behind the Alfv´n wave signal, there is visible fanning of e a negative value of V z (violet-white patch), that is a remnant after the Alfv´n wave passing through the plasma e medium (Murawski & Musielak 2010).
Now, the violet patch seen in Fig. 6 becomes elongated
and it contains a white subregion located at x = 12 Mm, y = 7 Mm (panel b ). This white subregion corresponds to negative ( V z ≈ -2 1 km s . -1 ) values of transversal velocity while nearby laying positive velocity ( V z ≈ 6 4 km . s -1 ) region is represented by a red patch. This results from the finite-size of the initial pulse whose left and right-sides have to pass significantly different distances, because the magnetic field line along the left-side path is longer than the magnetic field line along the right-side in the considered arcade. Moreover, a strong fanning of the wave signal spatial profile is a consequence of the Alfv´n e speed, c A ( x, y ), which attains different values along different magnetic interfaces (Fig. 2, bottom panel). At t = 36 s, the already downwardly propagating Alfv´ en waves reach the transition region (panel c ) and they undergo partial reflection from this region.
The resulting wave reflection is well seen at the consecutive moments of time, e.g., at t = 46 s (panel d ) and at t = 56 s (panel e ), respectively. However, at later moments of time the Alfv´n waves profiles become e more complicated; we can notice a partial reflection of the wave signal with negative value of V z , illustrated by narrow elongated violet-white patch (panels e and f ). At t = 66 s the Alfv´n waves reflect again at the left side, e near x 0 = 2 Mm and propagate with positive phase velocity (red patch) at t = 76 s with amplitude V z ≈ 4 2 . km s -1 in right direction. We observe a long narrow signal seen as a red patch, which reaches the transition region near x 0 = 21 Mm at t = 86 s (panel h ). Reflections of the Alfv´n waves are well seen at e t = 86 s (panel h ) and t = 96 s (panel i ) as violet-white patches near x 0 = 21 Mm. Such reflections result in complex Alfv´n e waves spatial profiles. We provide details on the partial reflection in Sect. 4.4, where we evaluate the corresponding reflection coefficient.
Fig. 7 presents in some details the Alfv´n waves ree flection. The arriving waves (panel a , red patch) hit the transition region and start reflecting in there, decreasing its amplitude, which is well seen in consecutive moments of time. In panel b , the reflected, backwardly propagating wave is illustrated by a violet-white patch, superimpose on the arriving negative phase waves. The effect of superimposition of the wave signals is significant, because of fanning of the wave spatial profiles, which are a consequence of large arcade, where Alfv´ en speed gradient and diverged magnetic field lines are large. At later moments of time (panels e to h ), we also see the evolution of the lagging wave signals reflected from the transition region as a red patch under the backward propagating main signal (violet-white patch).
Note that there is clearly seen small partially non- reflected (transmitted) Alfv´n waves propagating downe wardly into the solar surface under the transition region. These waves are in phase and their amplitude is about 1 km s -1 .
It is worth to compare our results to those previously obtained by Del Zanna et al. (2005), who considered an isothermal symmetric solar arcade with the transverse signal undergoing reflection from the transition region. In our non-isothermal asymmetric arcade, the numerical results show how the Alfv´ en wave pulse is widened by the diverged magnetic field lines, becomes gradually reflected in the transition region, and creates a non-regular structure of the reflected wave signal (see Fig. 6, bottomright). Later on, the diverged magnetic field configuration, inhomogeneous Alfv´n speed, difference of magnetic e field lengths and the reflected waves interaction with the ongoing wave train complicates even more the spatial velocity profiles in the arcade.
## 4.2.2. Short arcade
Spatial profiles of Alfv´n waves, which result from the e initial Gaussian pulse given by Eq. (15), with x 0 = 3 Mm and w x = 0 2 Mm are presented in Fig. 8. . Initial pulse triggers the counter-propagating Alfv´n waves in a come paratively less curved magnetic arcade. In this case, process of initial deformation of the wave pulse during propagation throughout the transition region is more effective because magnetic field lines are more horizontal than in the case of x 0 = 3 Mm. The upper part of the signal propagates earlier than the rest of the pulse and strongly influences on phase-mixing. On the other hand, a smaller arcade results in less diverged magnetic lines and less ∇ c A ( x, y ), that in consequence significantly reduces the fanning process of the wave signal.
Such a smaller arcade causes that a wave signal ranges all arcade region. At t = 20 s (panel b ) and later on, the outer edge of the arcade with its large length clearly reveals a phase difference of transversal wave velocity profiles as compared to the same in the core of the arcade. We observe several wave signal reflections from the transition region in Fig. 8. The violet-white patch in panels d and e represents reflected Alfv´n waves that e propagate to the left-hand side. When they reach the transition region near x = 3 Mm, they become reflected and propagate again into the right-hand side (panels f and g ) to suffer reflection once again near x = 16 Mm, which is clearly seen at t = 44 s (panel h ), and go back into x = 3 Mm (panels i and j ). It should be noted that the initial pulse-width was assumed to be the same as in the case of x 0 = 2 Mm (Fig. 6).
In the presented vertical cross-sections of the Alfv´ en wave profiles, we observe difference in wave propagation for large (Fig. 9, top panel) and small arcades (Fig. 9, bottom panel). The cross-section for the case of x 0 = 2 Mm ( x 0 = 3 Mm) shows the region near the apex of the arcade, x = 12 Mm ( x = 9 Mm), and three moments of time, when the wave signal passing the apex: t = 22 s ( t = 20 s), t = 72 s ( t = 40 s) and t = 124 s ( t = 60 s), that are illustrated in the top (bottom) panel of Fig. 9 by solid, dashed and dotted line, respectively. The main difference between these two cross-sections is a compact shape of the Alfv´n wave in the short arcade during all e three moments of time in comparison to the large arcade. This is a result of less diverged magnetic lines, which spread the signal up to width equal 5 Mm in the case of x 0 = 2 Mm, while in the case of x 0 = 3 Mm a pulse width is about 2 Mm.
We also find a difference in negative phase amplitudes. The Alfv´ en wave cross-section in the large arcade (Fig. 9, top panel) already after one period exhibits minima with negative values of V z accompanying the main wave signal. At t = 72 s (dashed line), the first opposite-sign phase with amplitude V z /similarequal -2 2 km s . -1 at y /similarequal 8 8 Mm . is a part of the elongated main wave signal still following into the transition region located near x = 2 Mm to suffer reflection (Fig. 6, panel g , upper violet white patch), while the second opposite-sign phase with amplitude V z /similarequal -0 8 km s . -1 at y /similarequal 6 6 Mm is a signal that . lags behind the main Alfv´n signal, well seen in panel e g of Fig. 6 (lower violet patch). In later moment of time, at t = 124 s (dotted line) the negative value of V z has a similar form, however the amplitudes are V z /similarequal -1 4 km s . -1 at y /similarequal 7 6 Mm and . V z /similarequal -0 8 km s . -1 at y /similarequal 6 7 Mm. .
In the case of a short arcade (Fig. 9, bottom panel), for which divergence of magnetic lines and the gradient of c A are smaller, the opposite-sign phases at the cross-section states only 0 9% of the main wave signal amplitude equal . to V z /similarequal 5 6 km s . -1 at t = 20 s (solid line), 7% of the amplitude equal to V z /similarequal 3 8 km s . -1 at t = 40 s (dashed line) and 15 2% of the amplitude equal to . V z /similarequal 2 5 km s . -1 at t = 60 s (dotted line).
Figure 10 illustrates the time-signatures of V z collected at points near the apex of each arcade. The timesignatures for cases of x 0 = 2 Mm, x 0 = 2 5 Mm and . x 0 = 3 Mm are presented in top, middle and bottom panels, respectively. We set the detection point for the case of x 0 = 2 Mm to be at ( x = 12 Mm, y = 7 Mm), for the case of x 0 = 2 5 Mm to be at ( . x = 11 Mm, y = 5 1 . Mm) and for the case of x 0 = 3 Mm at ( x = 9 Mm, y = 4 2 Mm). . Time-signatures for smaller arcades ( x 0 = 2 5 Mm and . x 0 = 2 Mm) have a regu-
Fig. 11.Travel time vs. the horizontal initial position, x 0 , along the magnetic field line crossing the point ( x , 0 1 75 Mm). .
lar harmonic shape, but the time-signature for the large arcade exhibits action on fanning process that results in superimposition of incoming and reflected wave signals in V z and finally in non-regular shape of V z oscillations. Note that the amplitude of the Alfv´n waves decays with e time and that V z oscillates with its characteristic wave period equal to P = 50 7 s for . x 0 = 2 Mm, P = 27 2 s . for x 0 = 2 5 Mm and . P = 20 4 s for . x 0 = 3 Mm. These different wave periods result from the different arcade lengths and different Alfv´n speeds along the corresponde ing magnetic lines for x 0 = 2 Mm, x 0 = 2 5 Mm and . x 0 = 3 Mm.
We calculate a travel time of the Alfv´n waves, e t t , for the first bounce along few magnetic field lines, which cross the given point ( x , y 0 0 ),
$$t _ { t } & = \int _ { 0 } ^ { l } \frac { d s } { c _ { \text{A} } ( s ) } \,. \quad & ( 1 7 ) \\ \text{not a loan} \, \text{the magnetic line has the value}$$
Here, a coordinate along the magnetic line has the value of s = 0 at ( x , y 0 0 ), while for s = l the magnetic line ends at the transition region, y = 2 75 Mm. . In Fig. 11 the travel time is displayed depending on the initial point x 0 . Its value gradually decreases with x 0 from t t /similarequal 39 s at x 0 = 2 Mm, reaching t t /similarequal 26 s at x 0 = 3 Mm. Because of almost linear decrease of the magnetic field length from l = 23 4 Mm at . x 0 = 2 Mm up to l = 11 6 Mm at . x 0 = 3 Mm (not shown), these results show that the time of the Alfv´n wave traveling along longer magnetic field e lines is much bigger, than for shorter magnetic lines. This effect is especially important in case of the large arcade, where the travel time falls off rapidly, contributing to the wave signal profile fanning.
Our results show the complex picture of Alfv´n waves e caused by wave reflection from the transition region, and difficulties in direct studies of the interaction of counterpropagating waves in the asymmetric solar arcade. Because of these difficulties, we instead focus on evaluating a global attenuation time. This allows us to investigate
Fig. 12.The ratio of the attenuation time over wave period, τ / P , vs. pulse width w x for the case of x 0 = 2 Mm (top panel) and x 0 = 3 Mm (bottom panel).
Fig. 13.The value of arcade full width at half maximum (FWHM) vs. x 0 for the chosen x = 11 Mm.
integrally the attenuation processes affecting the transverse waves, like interactions of the counter-propagating waves, partial wave reflection from the transition region, and the wave energy leakage along a curved magnetic field lines of the arcade.
## 4.3. Different pulses width
For a wider initial pulse, w x , the process of phasemixing becomes stronger. The reason is that both sides
Fig. 14.The reflection coefficient, C r , vs. x 0 .
of such pulse experience a larger phase difference due to a larger difference in field lines curvatures on both magnetic interfaces and pulse deformation process in the transition region as compared to a narrower pulse. Hence, for the wider pulse more strongly Alfv´n waves e propagation is affected.
Basing on data signal of V z collected at point ( x = 12 Mm, y = 7 Mm) for the case of longer arcade ( x 0 = 2 Mm) and ( x = 9 Mm, y = 4 2 Mm) for the . case of shorter arcade ( x 0 = 3 Mm), we evaluate a wave period P using fast Fourier transform method and an attenuation time τ . The ratio of attenuation time τ to a wave period P with respect to the width of the initial pulse is presented in Fig. 12, which shows that τ/P first increases for small values of w x but then falls off slightly with w x .
When we look at the individual τ/P versus the pulse width profiles of the two types of arcades, they exhibit a similar trend that is some increment up to a certain pulse width and then decrement. In the case of the considered geometry of more curved and larger magnetic arcade, the τ/P first increases up to 100 km pulse width (Fig. 12, top panel), which means that up to this limit of the pulse width the attenuation of Alfv´n waves decreases in the e arcade.
Similarly in the case of the considered geometry of less curved and smaller magnetic arcade, the τ/P first increases up to 400 km pulse width (Fig. 12, bottom panel), which means that up to this limit of pulse width the attenuation of Alfv´n waves decreases in the arcade. e For w x > 400 km, the τ/P decreases (Fig. 12, bottom panel), which is the signature of the increment in the wave attenuation. In other words, for the Alfv´n waves launched e with the pulse widths larger than 400 km, the generated long wavelengths do not fit within the considered loop geometry and the wave leakage again causes a wave at-
tenuation (Gruszecki et al. 2007).
It is also found that the attenuation of the Alfv´n waves e is smaller (larger the attenuation time with respect to the wave period) in the more curved and larger magnetic arcades ( x 0 = 2 Mm), while it is larger (smaller the attenuation time with respect to the wave period) in the case of the less curved and comparatively smaller arcades ( x 0 = 3 Mm). We found τ/P = 1 75 s for the . core of x 0 = 2 Mm and τ/P = 1 89 s for . x 0 = 3 Mm, for w x = 0 2 Mm. .
The arcade full width at half maximum (FWHM) dependence on the position of initial pulse x 0 , and in consequence on a size of the arcade, is shown in Fig. 13, which clearly illustrates that a smaller arcade (bigger x 0 ) has a smaller width (FWHM); a smaller arcade consists of weakly diverged magnetic field lines.
If one considers the phase-mixing to be the only candidate for the wave dissipation (e.g., Heyvaerts & Priest 1983), then the strong phase-mixing in the more curved and larger arcades (cf., Fig. 6) must cause a greater attenuation as compared to the less strong phase-mixing in the less curved and smaller arcades (cf., Fig. 8). Nevertheless, the simulation results demonstrate a more complicated scenario of the wave attenuation, which simply implies that the phase-mixing may not be the only candidate responsible for the wave attenuation; actually, the magnetic field configuration and plasma properties of the arcade such as a spatial profile of the Alfv´n speed, e strength and dimension of the pulse, and the structure of the transition region, all play the vital role in this phenomenon.
## 4.4. Partial reflection from the transition region
Decay of the Alfv´ en wave amplitude results from the wave energy leakage caused by the curvature and divergence of magnetic field lines, inhomogeneous Alfv´ en speed c A ( x, y ) and a partial penetration of Alfv´n wave e signal into the chromosphere (Gruszecki et al. 2007). We clearly observe in our simulation that Alfv´n waves exe perience partial reflection from the transition region and Alfv´ en waves signal penetrates into solar atmospheric layers under the transition region (Figs. 6 and 7, panels d -h ). The amplitude of the waves penetrating into the lower region of the atmosphere drops from about V z = 1 km s -1 at t = 48 s just under the transition region (Fig. 7, panel d ), through V z = 0 5 km s . -1 after about 16 s (Fig. 7, panel h ) and finally disappears. We can evaluate the reflection coefficient,
$$C _ { r } = \frac { A _ { r } } { A _ { i } } \,, \quad \quad \ \ ( 1 8 ) \quad \ \ f$$
where A i ( A r ) is the amplitude of the incident (transmitted) waves. Substituting A i = 6 km s -1 and A r = 4 01 km s . -1 into the above formula we get C r = 0 668, . which means that 66 8% of the waves amplitude became . reflected in the transition region and 33 2% was transmit-. ted into lower atmospheric layers. The reflection coefficient, C r , vs. initial pulse position x 0 for the first reflection from the transition region is presented in Fig. 14. We expect that the amplitude of the wave signal reflected in the transition region is smaller for Alfv´n waves in larger e inclined magnetic field lines, which corresponds to larger values of x 0 .
In the case of the largely curved arcades, the lateral leakage of the Alfv´n wave energy may also be largely e dominant as compared to the one associated with less curved arcades (Gruszecki et al. 2007). The partial Alfv´ en wave reflection results from a steep gradient of Alfv´ en speed in the transition region because of a significant mass density drop in this region of the solar atmosphere.
It must be also noted that in case of linear Alfv´ en waves generated by small initial velocity pulses, V z = 3 km s -1 with the maximum amplitude reaching only V z ≈ 6 km s -1 , there are no associated density variations.
## 5. SUMMARY AND CONCLUSIONS
To determine the role played by Alfv´n waves in the e coronal heating and solar wind acceleration, a number of authors investigated physical processes responsible for the excitation and attenuation of Alfv´n waves in the soe lar atmosphere (e.g., Priest 1982; Ofman & Davila 1995; Ofman 2002; Miyagoshi et al. 2004; Dwivedi & Srivastava 2006; Ofman & Wang 2007; Chmielewski et al. 2013, and references therein). In general, Alfv´n waves are dife ficult to dissipate their energy and possible mechanisms involve the collisional dissipative agents, such as viscosity and resistivity, or non-classical plasma processes, such as mode-coupling and phase-mixing (cf., Heyvaerts & Priest 1983; Nakariakov et al. 1997; Zaqarashvili et al. 2006; Dwivedi & Srivastava 2006, and references there). Overall, the collisional dissipative processes are found to be less important for the Alfv´n wave dissipation than the e non-classical plasma processes, especially phase-mixing (e.g., Nakariakov et al. 1997). Extensive observational searches were performed to find signatures of the Alfv´n e waves dissipation in the solar corona (e.g., Banerjee et al. 1998; Harrison et al. 2002; O'Shea et al. 2005; Bemporad et al. 2012, and references therein). However, as of today, there is no convincing observational evidence for the existence of Alfv´n waves dissipation in the solar e
atmosphere.
In this paper, we simulated impulsively generated Alfv´ en waves in a stratified and magnetically confined solar arcade with the VAL-C temperature profile (Vernazza et al. 1981) as an initial realistic plasma condition in the curved magnetic field topology. Asymmetric solar magnetic arcades and the Alfv´ en waves propagation in these arcades were modeled by the time-dependent MHD equations that were solved numerically by the publicly available FLASH code (Lee & Deane 2009). We analyzed the effects of changing the horizontal position of the initial pulse and its width on the Alfv´n wave prope agation in the asymmetric solar arcade, the Alfv´n wave e deformation and phase-mixing resulting from inhomogeneous Alfv´n wave velocity, and different lengths and die vergence of magnetic field lines, and the partial wave reflection in the solar transition region.
We found that the more curved and larger arcade, then the stronger attenuation of the Alfv´n waves. Our results e also demonstrated attenuation of Alfv´n waves resulted e from the curvature and divergence of magnetic field lines, inhomogeneous Alfv´ en wave velocity, partial reflection in the solar transition region, as well as the decrement of the attenuation time of Alfv´n waves for wider initial pulses e in the given magnetic configuration of various types of arcades. Moreover, our numerical simulations also showed that Alfv´n waves, which are partially reflected in the e solar transition region return to the solar chromosphere as slowly downward propagating Alfv´ en waves.
The approach presented in this paper allowed us to investigate the effects caused by trains of Alfv´n waves e propagating in our arcade model as a result of the wave reflection in the solar transition region. Our results clearly show that the pulse deformation process in the transition region has a strong influence on the Alfv´ en waves propagation and on the wave damping in the curved magnetic field of the solar arcade together with the asymmetric magnetic field configuration, the plasma properties of the arcade, the horizontal size of the pulse, and the structure of the solar transition region.
In the previous work, Nakariakov et al. (1999) investigated the transverse oscillations of EUV loops as observed by the Transition Region and Coronal Explorer (TRACE), and demonstrated that the damping time was three times longer than the period of oscillations. They considered kink oscillations of coronal loops and their damping caused by non-classical viscosity and resistivity. Then, Ofman et al. (2002) showed that the dissipation of Alfv´ en waves due to chromospheric wave leakage was not sufficient to describe such fast damping of the observed transversal oscillations in the solar atmosphere. Finally,
Del Zanna et al. (2005) demonstrated that dissipation of Alfv´ en pulses within their coronal arcade led to a change in the local Alfv´n speed at various heights, and depende ing upon various localized conditions that determine the observed damping of these transverse oscillations.
In our work, the damping/attenuation time of the Alfv´ en waves are almost three times longer than that computed in both the longer and shorter arcades for a given pulse-width of 0.2 Mm. It should be noted that this damping time is due to the Alfv´ en wave dissipation caused by the phase mixing. The ratio of attenuation time and wave period is almost the same to what is observed for the transverse wave dissipation in coronal loops. An important point that must be mentioned is that transverse kink waves with their radial velocity perturbations in the cylindrical thin tube geometry (Roberts 2000) are different than Alfv´n waves studied in this pae per, which are essentially azimuthal. Nevertheless, there is a similar dissipation mechanism (e.g., phase-mixing) that may work for these two types of waves; its efficiency highly depends on the localized plasma, magnetic field configuration, nature of wave drivers, and Alfv´n veloce ity, and others (see Ofman et al. 2002, 2007; Del Zanna et al. 2005). Therefore, our parametric studies presented in this provide some likely physical scenario for dissipation of Alfv´n waves through phase-mixing. e
Finally, we would like to point out that our approach is significantly different from that developed by Del Zanna et al. (2005), who considered a symmetric coronal arcade, which was embedded in the isothermal plasma with the constant (in time) temperature specified by the hyperbolic tangent profile, and designed it in such a way that it described short lived Alfv´n waves in post flare e settings. Similarly, Miyagoshi et al. (2004) simulated Alfv´ en waves that were excited by perturbations imposed in the solar corona. By using our more realistic arcade model with the asymmetric magnetic field configuration, we were able to explore different physical aspects of the Alfv´ en wave propagation than those studied in the past. Essentially, our model significantly generalizes the model originally considered by Del Zanna et al. (2005) and by Miyagoshi et al. (2004). Nevertheless, further improvements of our model are possible and it is our hope that they would lead to even better understanding of the inherent physical complexity of phase-mixing process and its role in heating of the solar corona by Alfv´ en wave dissipation.
We thank the referee for his/her valuable suggestions that improved our manuscript considerably. P.Ch. expresses his cordial thanks to Piotr Konkol for his assis-
tance in drawing some numerical data. This work has been supported by NSF under the grant AGS 1246074 (K.M. & Z.E.M.), and by the Alexander von Humboldt
Foundation (Z.E.M.). The software used in this work was in part developed by the DOE-supported ASC/Alliance Center for Astrophysical Thermonuclear Flashes at the University of Chicago.
## REFERENCES
Arregui, I., Oliver, R., & Ballester, J. L. 2004, ApJ, 602, 1006 Banerjee, D., Teriaca, L., Doyle, J. G., & Wilhelm, K. 1998, A&A,
- 339, 208 Bemporad, A., & Abbo, L. 2012, ApJ, 751, 110 Biskamp, D., & Welter, H. 1989, Sol. Phys., 120, 49 ˇ adez, V. M., Oliver, R., & Ballester, J. L. 1994, A&A, 282, 934 C Chmielewski, P., Srivastava, A. K., Murawski, K., & Musielak, Z. E. 2013, MNRAS, 428, 4 Del Zanna, L., Schaekens, E., and Velli, M. 2005, A&A, 431, 1095 D´ ıaz, A. J., Zaqarashvili, T., & Roberts, B. 2006, A&A, 455, 709 Dwivedi, B. N., & Srivastava, A. K. 2006, Sol. Phys., 237, 143 Fryxell, B., Olson, K., Ricker, P., et al. 2000, ApJS, 131, 273 Gruszecki, M., Murawski, K., Selwa, M., & Ofman, L. 2006, A&A, 460, 887 Gruszecki, M., Murawski, K., Solanki, S. K., & Ofman, L. 2007, A&A, 469, 1117 Gruszecki, M., & Nakariakov, V. M. 2011, A&A, 536, A68 Harrison, R. A., Hood, A. W., & Pike, C. D. 2002, A&A, 392, 319 Heyvaerts, J., & Priest, E. R. 1983, A&A, 117, 220 Innes, D. E., McKenzie, D. E., & Wang, T. 2003, Sol. Phys., 217, 267 Lee, D. 2013, J. Comp. Phys., 243, 26 Lee, D., & Deane, A. E. 2009, J. Comp. Phys., 228, 952 Low, B.C. 1985, ApJ, 293, 31 MacNeice, P., Spicer, D. S., & Antiochos, S. 1999, 8th SOHO Workshop: Plasma Dynamics and Diagnostics in the Solar Transition Region and Corona, 446, 457
Mikic, Z., Schnack, D. D., & van Hoven, G. 1989, ApJ, 338, 114 Miyagoshi, T., Yokoyama, T., & Shimojo, M. 2004, PASJ, 56, 207 Murawski, K., & Musielak, Z. E. 2010, A&A, 518, A37 Nakariakov, V. M., Roberts, B., & Murawski, K. 1997, Sol. Phys.,
- McKenzie, D. E., & Savage, S. L. 2009, ApJ, 697, 1569
- 175, 93 Oliver, R., Hood, A. W., & Priest, E. R. 1996, ApJ, 461, 424 O'Shea, E., Banerjee, D., & Doyle, J. G. 2005, A&A, 436, L43 Ofman, L., & Davila, J. M. 1995, J. Geophys. Res., 100, 23427 Ofman, L. 2002, ApJ, 568, L135 Ofman, L., & Wang, T. 2007, AGU Fall Meeting Abstracts, 2 Priest, E. R. 1982, Dordrecht, Holland ; Boston : D. Reidel Pub. Co. ; Hingham,, 74P Rial, S., Arregui, I., Terradas, J., Oliver, R., & Ballester, J. L. 2010, ApJ, 713, 651 Robert, B., 2000, SoHO , 4. Selwa, M., Murawski, K., Solanki, S. K., Wang, T. J., & T´th, G. o 2005, A&A, 440, 385 Selwa, M., Solanki, S. K., Murawski, K., Wang, T. J., & Shumlak, U. 2006, A&A, 454, 653 Toth, G. 2000, J. Comp. Phys., 161, 605 Vernazza, J. E., Avrett, E. H., & Loeser, R. 1981, ApJS, 45, 635 Verwichte, E., Nakariakov, V. M., Ofman, L., & Deluca, E. E. 2004, Sol. Phys., 223, 77 Verwichte, E., Foullon, C., & Van Doorsselaere, T. 2010, ApJ, 717, 45 Zaqarashvili, T. V., Oliver, R., & Ballester, J. L. 2006, A&A, 456, L13
|
10.1088/0004-637X/793/1/43
|
[
"P. Chmielewski",
"K. Murawski",
"Z. E. Musielak",
"A. K. Srivastava"
] |
2014-08-05T03:23:38+00:00
|
2014-08-05T03:23:38+00:00
|
[
"astro-ph.SR"
] |
Numerical simulations of impulsively generated Alfvén waves in solar magnetic arcades
|
We perform numerical simulations of impulsively generated Alfv\'en waves in
an isolated solar arcade, which is gravitationally stratified and magnetically
confined. We study numerically the propagation of Alfv\'en waves along such
magnetic structure that extends from the lower chromosphere, where the waves
are generated, to the solar corona, and analyze influence of the arcade size
and width of the initial pulses on the wave propagation and reflection. Our
model of the solar atmosphere is constructed by adopting the temperature
distribution based on the semi-empirical VAL-C model and specifying the curved
magnetic field lines that constitute the asymmetric magnetic arcade. The
propagation and reflection of Alfv\'en waves in this arcade is described by
2.5D magnetohydrodynamic equations that are numerically solved by the FLASH
code. Our numerical simulations reveal that the Alfv\'en wave amplitude
decreases as a result of a partial reflection of Alfv\'en waves in the solar
transition region, and that the waves which are not reflected leak through the
transition region and reach the solar corona. We also find the decrement of the
attenuation time of Alfv\'en waves for wider initial pulses. Moreover, our
results show that the propagation of Alfv\'en waves in the arcade is affected
by spatial dependence of the Alfv\'en speed, which leads to phase-mixing that
is stronger for more curved and larger magnetic arcades. We discuss processes
that affect the Alfv\'en wave propagation in an asymmetric solar arcade and
conclude that besides phase-mixing in the magnetic field configuration, plasma
properties of the arcade and size of the initial pulse as well as structure of
the solar transition region all play a vital role in the Alfv\'en wave
propagation.
|
1408.0857v2
|
## Impact of the equation-of-state - gravity degeneracy on constraining the nuclear symmetry energy from astrophysical observables
Xiao-Tao He, 1, 2, ∗ F. J. Fattoyev, 1, † Bao-An Li, 1, ‡ and W. G. Newton 1, §
1 Department of Physics and Astronomy, Texas A&M University-Commerce, Commerce, TX 75429, USA 2 College of Material Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China (Dated: December 17, 2014)
There is a degeneracy between the equation of state (EOS) of super-dense neutron-rich nuclear matter and the strong-field gravity in understanding properties of neutron stars. While the EOS is still poorly known, there are also longstanding ambiguities in choosing Einstein's General Relativity (GR) or alternative gravity theories in the not-so-well tested strong-field regime. Besides possible appearance of hyperons and new phases, the most uncertain part of the nucleonic EOS is currently the density dependence of nuclear symmetry energy especially at supra-saturation densities. At the same time, the EOS of symmetric nuclear matter (SNM) has been significantly constrained at saturation and supra-saturation densities. To provide information that may help break the EOSgravity degeneracy, we investigate effects of nuclear symmetry energy within its uncertain range determined by recent terrestrial nuclear laboratory experiments on the gravitational binding energy and space-time curvature of neutron stars within GR and the scalar-tensor subset of alternative gravity models, constrained by recent measurements of the relativistic binary pulsars J1738+0333 and J0348+0432. In particular, we focus on effects of the following three parameters characterizing the EOS of super-dense neutron-rich nucleonic matter: (1) the incompressibility K 0 of symmetric nuclear matter (SNM), (2) the slope L of nuclear symmetry energy at saturation density and (3) the high-density behavior of nuclear symmetry energy. We find that the variation of either the density slope L or the high-density behavior of nuclear symmetry energy leads to large changes in both the binding energy and curvature of neutron stars while effects of varying the more constrained K 0 are negligibly small. The difference in predictions using the GR and the scalar-tensor theory appears only for massive neutron stars, and even then is significantly smaller than the differences resulting from variations in the symmetry energy. We conclude that within the scalar-tensor subset of gravity models, the EOS-gravity degeneracy has been broken by the recent relativistic pulsar measurements, and that measurements of neutron star properties sensitive to the compactness constrain mainly the density dependence of the symmetry energy at saturation and supra-saturation densities.
PACS numbers: 26.60.Kp, 21.65.Mn, 04.40.Dg
∗ [email protected]
† [email protected]
‡ [email protected]
§ [email protected]
## I. INTRODUCTION
Neutron stars (NSs) are among the most dense and neutron-rich objects in the Universe. They are thus a natural testing ground for fundamental theories of strong-field gravity and the nature of matter under extreme conditions. Our knowledge about super-dense matter in NSs remains quite poor, see, e.g., [1-3], and may be compounded by the possible break-down of Einstein's General Relativity (GR) in the strong-field regime [4, 5]. There is a possible degeneracy between the nuclear equation of state (EOS) for matter in NSs and the theory of gravity applied to describe their structure. How to break this degeneracy is a longstanding problem to which many recent studies have been devoted: see, e.g., [4-19]. A better understanding of the EOS will certainly help constrain the strong-field gravity theory, and vice versa. Besides the possible appearance of hyperons and quarks in super-dense matter in NSs, the most uncertain part of the nucleonic EOS is currently the nuclear symmetry energy at supra-saturation densities [20]. Ideally, effects of the uncertain nuclear EOS should be examined simultaneously within GR and a large sample of alternative gravity theories. This however requires much collaborative work of the whole community. In this work, we examine the extent to which a gravity-EOS degeneracy exists when models of gravity are restricted to GR and Scalar-Tensor (ST) theories, and when variations of the EOS are the result of variation of (1) the incompressibility K 0 of symmetric nuclear matter (SNM), (2) the slope L of nuclear symmetry energy at saturation density and (3) the high-density behavior of nuclear symmetry energy, within ranges constrained by terrestrial nuclear data. We find that the variation of either the density slope L or the high-density behavior of nuclear symmetry energy within their uncertainty ranges lead to significant changes in both the binding energy and curvature of NSs, while effects of varying K 0 are negligibly small. In particular, the variations are significantly greater than those that result from ST theories of gravity, leading to the conclusion that within those subset of gravity models, measurements of neutron star properties constrain mainly the EOS.
Neutron stars are very compact: the typical surface compactness parameter of NSs is of the order of η R ≡ 2 GM/c R 2 ≈ 0 4. . While it is common to measure the strength of the gravitational field with the compactness parameter, this leaves out most of the useful information related to its strength, in particular, when one would like to test the deviations of GR from alternative theories of gravity in the strong-field regime. It was shown in Ref. [21] that for the gravitational field to be considered as strong it is more natural to choose a parameter ξ ≡ 2 GM/c r 2 3 , which is related to the non-vanishing components of the Riemann tensor in vacuum, R 1 010 = -ξ . Notice that the curvature is the lowest order quantity of the gravitational field that cannot be set to zero by a coordinate transformation [21]. Recently, by using the venerable Akmal-Pandharipande EOS [22] the authors of Ref. [5] have studied the radial profile of compactness and curvature in neutron stars and have concluded that the GR is in a less tested regime than the EOS over the entire region of neutron star. In other work, by using a very stiff EOS with nuclear incompressibility of K 0 = 546 MeV [23], it was shown that the fractional gravitational binding energy of the PSR J0348+0432 with mass 2 01 . ± 0 04 . M glyph[circledot] [24] is twice as large as that of both NSs in the celebrated double pulsar system PSR J0737-3039, whose masses have been measured with a very high precision. Given that the observation of these pulsars resulted in some of the most precise tests of GR to date, it was concluded that the great difference in their fractional gravitational binding energies keeps PSR J0348+0432 in a special place far outside the presently tested binding energy range, and therefore this NS could serve as a testbed for theories of strong-field gravity. Particularly interesting is the study of the scalar-tensor theory of gravity, whose parameters have a well-defined constraint from a ten-year pulsar timing observation of PSR J1738+0333 [25].
Disentangling the EOS-Gravity degeneracy is important if we are to make use of astrophysical observables to measure neutron star properties. Potential observables directly related to the neutron star compactness include the binding energy [26, 27], surface redshift [28] and maximum spin frequency of a neutron star [29], and it is these observables that we shall focus on in this paper.
The nuclear symmetry energy S ρ ( ) represents the penalty imposed on the energy of the nuclear system as one departs from the symmetric limit of equal number of protons and neutrons at density ρ . Within the parabolic approximation of the nuclear EOS for neutron-rich matter, S ρ ( ) is simply the difference in specific energy between pure neutron matter (PNM) and SNM. The bulk parameters characterizing the EOS of SNM near nuclear saturation density ρ 0 are well constrained from fitting models to the ground state properties of finite nuclei, and to the breathing-mode energies of giant resonances in medium-to-heavy nuclei [30-37]. The incompressibility of nuclear matter is constrained in the far tighter range of K 0 = 240 ± 20 MeV, which is well below the value predicted three decades ago, for example, in Ref. [23]. Moreover, combining information from studying the collective flow and kaon production in relativistic heavy-ion collisions in several terrestrial nuclear physics laboratories the EOS of SNM at higher densities has also been limited in a relatively small range up to about 4 5 . ρ 0 [38] consistent with the latest observations of two-solar mass neutron stars [24, 39]. However, despite intensive efforts devoted to constraining the density dependence of the nuclear symmetry energy, its knowledge still remains largely uncertain even around nuclear saturation density ρ 0 [40-52]. It plays a vital role not only in describing the structure of rare isotopes and their reaction mechanisms [20, 53-55], but also determines uniquely the proton fraction essential for understanding the cooling mechanism and appearance
of exotic species in neutron stars [2, 56]. Moreover, it affects significantly the structure, such as the radii, moment of inertia, tidal polarizability and the core-crust transition density, as well as the frequencies and damping times of various oscillation modes of neutron stars (For a full review please refer to the topical issue on this subject matter [57]). Furthermore, the uncertainty in the high-density behavior of the symmetry energy is quite large with predictions from all varieties of nuclear models diverging dramatically [58-60]. Besides the well known difficulties of treating accurately quantum many-body problems, our poor knowledge about the spin-isospin dependence of three-body and many-body forces, the short-range behavior of nuclear tensor force and the resulting isospin-dependence of short-range nucloennucleon correlations, coupled with the lack of sensitive experimental probes are mainly responsible for the current situation. Whereas several observables have been proposed in the past [20], and some hints of the high-density behavior of nuclear symmetry energy have been reported recently [61, 62], there is still no widely accepted conclusion regarding the density dependence of nuclear symmetry energy [63]. Therefore, the symmetry energy is still regarded as the most uncertain part of the EOS of neutron-rich nuclear matter.
The paper has been organized as follows. First, in Sec. II we will overview the underlying nuclear EOS models used in this study. Special attention is paid to the uncertainties of the EOS of both SNM at saturation and the nuclear symmetry energy at both saturation and supra-saturation densities. Second, in Sec. III we provide the necessary details required to calculate the gravitational binding energy and discuss its sensitivity to the nuclear symmetry energy, in particular. Next, in Sec. IV we discuss the role of the full contraction of Riemann curvature tensor as a suitable choice of curvature. In particular, we will show that while the relative strength of the gravitational field could be several times larger from the surface to the core of the neutron star, it strongly depends on the choice of the underlying EOS. Then, in Sec. V we will briefly overview the necessary formula to calculate the mass versus radius relation and the gravitational binding energy in the scalar-tensor models of gravity. We will show that current observational constraints on parameters of this theory puts them at a disadvantage in breaking the degeneracy between the uncertainties in the EOS and alternative models of gravity. Later, in Sec. VI we discuss the possibility of determination of the nuclear symmetry energy from two particular astrophysical observables sensitive to the compactness of the star: the surface gravitational redshift of non-rotating NSs and the maximum spin frequency of rapidly rotating NSs. And finally in Sec. VII we offer our concluding remarks.
## II. THE EQUATION OF STATE OF STELLAR MATTER
The equation of state of neutron-rich nucleonic matter, in general, can be expressed as
$$E ( \rho, \alpha ) = E _ { 0 } ( \rho ) + \mathcal { S } ( \rho ) \alpha ^ { 2 } + \mathcal { O } ( \alpha ^ { 4 } )$$
where E ρ, α ( ) and E 0 ( ρ ) are the binding energy per nucleon in asymmetric nuclear matter of isospin asymmetry α = ( ρ n -ρ p ) /ρ and in SNM ( α = 0), respectively. Here S ( ρ ) is referred to as the nuclear symmetry energy, which represents the energy cost of departing from symmetric limit of equal neutrons and protons, and ρ = ρ n + ρ p is the total baryon density with ρ n ( ρ p ) being the neutron (proton) density. It is customary to characterize the EOS further in terms of bulk nuclear parameters by expanding both E 0 ( ρ ) and S ( ρ ) in Taylor series around nuclear saturation density
$$E _ { 0 } ( \rho ) = B _ { 0 } + \frac { 1 } { 2 } K _ { 0 } \chi ^ { 2 } + \mathcal { O } ( \chi ^ { 3 } )$$
$$\mathcal { S } ( \rho ) = J + \stackrel { - } { L } \chi + \frac { 1 } { 2 } K _ { s y m } \chi ^ { 2 } + \mathcal { O } ( \chi ^ { 3 } )$$
where χ ≡ ( ρ -ρ 0 ) / 3 ρ 0 quantifies the deviations of the density from its saturation value. As discussed in the Introduction the binding energy at saturation B 0 and the nuclear incompressibility coefficient K 0 are tightly constrained by terrestrial experimental data using ground state properties of finite nuclei and energies of giant resonances. On the other hand, while the symmetry energy at saturation J is more or less known, its density slope L is largely unconstrained. Moreover the value of the curvature of the symmetry energy at saturation K sym still has a huge uncertain range.
As our base models we employ two recently established EOSs for neutron-rich nucleonic matter within the IU-FSU Relativistic Mean Field (RMF) model and the SkIU-FSU Skyrme-Hartree-Fork (SHF) approach [49, 64, 65]. The IU-FSU EOS was obtained by adjusting the parameters of the RMF to satisfy the latest constraints from terrestrial nuclear experiments and astrophysical observations [65]. Moreover, its predictions closely match the state-of-theart nuclear many-body calculations for the EOS of low-density pure neutron matter (PNM). By construction, the SkIU-FSU model is a non-relativistic counterpart of the IU-FSU, and they have similar EOSs for both SNM with K 0 = 231 3 MeV, and PNM around and below . ρ 0 . In particular, the slope of the symmetry energy at saturation for both models is L = 47 2 MeV. At subsaturation densities, . S ( ρ ) is almost identical for these models. However, their
160
(MeV)
FIG. 1. (color online). Density dependence of the nuclear symmetry energy for the four models discussed in the text. The solid black line corresponds to the symmetry energy predictions in SkIU-FSU, the dashed black line corresponds to the original IU-FSU model, the dotted red and dash-dotted blue lines are symmetry energy predictions in the IU-FSU-like models with density slope L = 60 0 MeV (IU-FSU ) and . 1 L = 100 0 MeV (IU-FSU ), respectively. Notice that we did not display results for . 2 the IU-FSU-like models with incompressibility coefficients of K 0 = 220 MeV and K 0 = 260 MeV, since their predictions are almost identical to the one of the original IU-FSU model.
predictions for the symmetry energy at super-saturation densities of glyph[greaterorsimilar] 1 5 . ρ 0 significantly differ with the SkIU-FSU leading to a much softer S ( ρ ) at higher densities (See Fig. 1). For full details on these models we refer the reader to Refs. [49, 64, 65].
To test the sensitivity of the gravitational binding energy and space-time curvature of neutron stars to the variations of properties of neutron-rich nuclear matter around saturation density, we have further introduced four additional EOSs using the IU-FSU as the base model. By tuning two purely isovector parameters of the IU-FSU model one can efficiently modify the density dependence of the symmetry energy [64]. This tuning is carried out under the constraint that the value of the symmetry energy at a subsaturation density of ρ ≈ 0 103 . ρ 0 is fixed at 26 0 MeV. Such tuning . leaves the EOS of SNM unchanged and guarantees that predictions for the ground-state properties of finite nuclei will not deviate from their experimental values [66]. In this way we created two RMF models with density slopes of the symmetry energy at saturation density of L = 60 MeV and L = 100 MeV. Finally, by properly adjusting all parameters of the RMF model, so that their bulk properties match those of the IU-FSU baseline model, we have created two additional RMF models with differing incompressibility coefficients of K 0 = 220 MeV and K 0 = 260 MeV. We have ensured that predictions of these two models reproduce the experimental data on charge radii and binding energies of some selected closed-shell nuclei by slightly adjusting the binding energy at saturation. Shown in Fig. 1 are the density dependence of the symmetry energy for four of these models. Notice we did not display models with K 0 = 220 MeV and K 0 = 260 MeV, because the density dependence of the symmetry energy in these models are almost indistinguishable from that of the original IU-FSU.
## III. GRAVITATIONAL BINDING ENERGY OF NEUTRON STARS
We calculate the core EOS assuming a minimal model of neutrons, protons, electrons and muons in chemical equilibrium. We employ the EOS of Baym, Pethick, and Sutherland [67] for the outer crust. Because the quantities we will be calculating are not sensitive to the detailed composition of the inner crust, we simply use polytropic EOSs
that connect the uniform liquid core to the bottom layers of the outer crust defined by the neutron dripline. The stellar profiles are then determined by integrating the usual Oppenheimer-Volkoff (OV) equations [68]
$$\frac { d P ( r ) } { d r } = - \frac { G } { c ^ { 2 } r ^ { 2 } } \left [ \mathcal { E } ( r ) + P ( r ) \right ] \left [ M ( r ) + 4 \pi r ^ { 3 } \frac { P ( r ) } { c ^ { 2 } } \right ] \left [ 1 - \frac { 2 G M ( r ) } { c ^ { 2 } r } \right ] ^ { - 1 },$$
$$\frac { d M _ { \text{G} } ( r ) } { d r } = 4 \pi r ^ { 2 } \frac { \mathcal { E } ( r ) } { c ^ { 2 } }$$
where G is the gravitational constant, P r ( ), E ( r ) and M r ( ) are the pressure, energy density and mass profiles as a function of radial distance r . Given the boundary condition P (0) = P c , M (0) = 0, supplemented by an EOS, the radius of the neutron star R will then be determined from the condition of P R ( ) = 0. The corresponding total mass enclosed is referred to as the gravitational mass of the neutron star, which is
$$M _ { \text{G} } = \frac { 1 } { c ^ { 2 } } \int _ { 0 } ^ { R } 4 \pi r ^ { 2 } \mathcal { E } ( r ) d r,$$
One may also define the so-called baryonic mass of the neutron star as the mass of the star that has been disassembled into its component baryons, which is given by M B ≡ Nm B . Here m B is the baryon mass and N is the total number of baryons in the star that can be found by a volume integration of the baryon density ρ B ( r ) in the Schwarzschild geometry,
$$N = \int _ { 0 } ^ { R } 4 \pi r ^ { 2 } \rho _ { \text{B} } ( r ) \left [ 1 - \frac { 2 G M ( r ) } { c ^ { 2 } r } \right ] ^ { - 1 / 2 } d r,$$
Then the gravitational binding energy in mass units is simply defined as B ≡ M G -M B [69], which is always a negative quantity for a gravitationally bound system. It is also important to specify the baryon mass m B . If we set it to be the mass of a single nucleon, then B would be the total binding energy that contains the contribution from the nuclear binding energy as well. In order to compare the predicted binding energy with estimates of the baryon mass of the pre-collapse cores, the nuclear binding energy must be taken into account. It is therefore appropriate to set the baryon mass to the atomic mass unit instead. Therefore following Refs. [1, 26, 27, 70] we set it to m B ≡ 931 5 . MeV /c 2 . It his useful to express the fractional gravitational binding energy B /M G as a function of the compactness parameter η R [27, 71, 72], which for neutron stars is several orders of magnitude larger then that of regular stars such as our Sun [5, 71]. As a result, the binding energy contribution to the total mass of a neutron star is significant-up to 25% of its rest mass [72-74] for stars near the maximum mass-although the exact size of the binding energy strongly depends on the underlying EOS [27].
In Fig. 2 we display the fractional gravitational binding energy as a function of the total gravitational mass of a neutron star. Higher mass stars, and stars governed by softer EOSs (that is, softer symmetry energies or smaller values of incompressibility K 0 ) are more compact and have higher absolute values of B /M G (see also [26, 75]). Now let us examine effects of the uncertain symmetry energy in more details. First, notice that with almost the same low-density symmetry energy up to about 1 5 . ρ 0 , but different high-density behaviors, predictions for the B /M G in the SkIU-FSU and IU-FSU models are very different. This effect becomes more pronounced with the increase of the total gravitational mass. Indeed, larger neutron stars probe the symmetry energy at higher densities which is different in these two models due to their distinctly different high-density behaviors of the symmetry energy. Recall that by construction both the magnitude and the slope of the symmetry energy at saturation are identical in these models, i.e. J = 31 3 MeV and . L = 47 2 MeV, respectively. . Choosing the IU-FSU as a reference, then the relative changes in the fractional gravitational binding energies are 5 32%, 6 52% and 9 89% for 1 25 . . . . M glyph[circledot] , 1 4 . M glyph[circledot] and 1 9 . M glyph[circledot] neutron stars respectively (See Table. I). Thus, the fractional gravitational binding energy becomes increasingly sensitive to the the super-saturation behavior of the symmetry energy for increasing mass.
This is in contrast to the behavior of the binding energy in models with soft and stiff symmetry energies at saturation -e.g. , the IU-FSU with L = 47 2 MeV and the IU-FSU . 2 with L = 100 MeV. Again choosing the former as a reference model, then the relative changes in the fractional gravitational binding energies are -10 26%, . -8 85% . and -7 44% for 1 25 . . M glyph[circledot] , 1 4 . M glyph[circledot] and 1 9 . M glyph[circledot] neutron stars respectively. Thus the fractional binding energy is most sensitive to the saturation density slope of the symmetry energy for low mass neutron stars.
Finally, we note that, because the incompressibility coefficient of SNM, K 0 , is presently well constrained by terrestrial nuclear physics data, the effect on the fractional binding energy B /M G over the constrained range of K 0 is almost negligible.
The discovery of the double pulsar system PSR J0737-3039 has enabled the accurate determination of many pulsar properties including both pulsar's gravitational masses. In particular, the gravitational mass of the pulsar PSR J07373039B is determined to be M G = 1 2489 . ± 0 0007 . M glyph[circledot] [76], which is one of the lowest reliably measured mass for any
M
/
FIG. 2. (color online). The fractional gravitational binding energy versus the gravitational mass of a neutron star for a set of the EOSs discussed in the text: in the left panel (a) predictions are given for the EOSs discussed in Fig. 1, while in the right panel (b) predictions are given for the IU-FSU-like models with incompressibility coefficients of K 0 = 220 MeV (IU-FSU ) displayed 3 in green solid line, and K 0 = 260 MeV (IU-FSU ) displayed in dashed magenta line. 4 The constraints on the gravitational binding energy of PSR J0737-3039B, under the assumption that it is formed in an electron-capture supernova, are given by the vertical brown and purple lines.
neutron star up to date. Its low mass has led to the suggestion that this NS might have been formed as a result of the collapse of an electron-capture supernova from a progenitor of a O-Ne-Mg white dwarf [26]. Under this assumption, it was estimated that the baryonic mass of the precollapse O-Ne-Mg core should lie between 1 366 . M glyph[circledot] < M B < 1 375 . M glyph[circledot] [26] and 1 358 . M <M < . glyph[circledot] B 1 362 M glyph[circledot] [77]. Using these two sets of constraints, it was shown in Ref. [27] that the upper limit of the density slope at saturation should be L ≈ 70 MeV to be consistent with these observations. Notice that recently an even lower mass of M G = 1 230 . ± 0 007 . M glyph[circledot] has been reported through the long-term precision timing measurements of double neutron star system PSR J1756-2251 [78]. In Fig. 2 we confirm that only models with the soft symmetry energy are consistent with these set of constraints. Notably, models with even smaller values of L -such as IU-FSU 1 with L = 60 MeV-could still be inconsistent, if the subsequent density dependence of the symmetry energy at higher densities is stiff. Such low mass stars, via their binding energy, probe the symmetry energy slope at saturation; and, once given a constrained range for L , they may also probe the symmetry energy at higher densities.
Finally, we remark that the fractional gravitational binding energy in the massive neutron star PSR J0348+0432 is twice larger then that of the PSR J0737-3039B independent of the EOS model used [24]. As noted in that paper, and shown in Fig. 2, whereas uncertainties of the EOS due to variations of the density dependence of the symmetry energy are of the order of 20% for a given star, the substantial differences in the B /M G of canonical and two-solar mass neutron stars makes the latter an ideal laboratory for testing GR in the strong field regime.
Thus we can conclude that, given that the EOS of SNM is well constrained by terrestrial data, the binding energy of low mass neutron stars probes the saturation density symmetry energy primarily, and then the high density behavior of the symmetry energy. Given constraints from low mass neutron stars, the resulting range for the symmetry energy as a function of density constrains B /M G . For example, for the set of the EOSs used here we find that the fractional binding energy of a 1.9 solar-mass neutron star is between -0 1380 . ≤ B /M G ≤ -0 1639. There is no known . mechanism of extracting baryonic mass in massive neutron stars however. In the next section we will analyze this further by contrasting the effect of the density dependence of the symmetry energy with measures of the strength of the gravitational field.
TABLE I. Predictions for the radius R , surface compactness parameter η R , fractional gravitational energy B M G , full contraction of the Riemann curvature tensor at the surface K R of neutron stars with masses 1.25 M glyph[circledot] , 1.4 M glyph[circledot] and 1.9 M glyph[circledot] as predicted by the various models used in the text. Also relative changes of the gravitational binding energy and surface full contraction of the Riemann curvature tensor with respect to the predictions of the original IU-FSU model are provided in percentage. Finally we give the radius, binding energy and curvature of a star calculated using the scalar-tensor model described in the text is given for the IU-FSU EOS (indicated by IU-FSU (ST)), together with the relative change in binding energy compared with the general relativistic result.
| Model | L (MeV) | K 0 (MeV) | M G ( M glyph[circledot] ) | R (km) | η R | B M G | K R (10 - 2 km - 2 ) | ∆ B B (%) | ∆ K R K R (%) |
|-------------|-----------|-------------|------------------------------|-------------------|----------------------|----------------------------|------------------------|-----------------------|-------------------------|
| IU-FSU | 47.2 | 231.3 | 1 . 25 1 . 4 1 . 9 | 12.50 12.49 11.73 | 0.2953 0.3310 0.4785 | - 0.0861 - 0.0989 - 0.1491 | 0.6544 0.7347 1.2051 | - - - | - - - |
| IU-FSU 1 | 60 | 231.3 | 1 . 25 1 . 4 1 . 9 | 13.25 13.12 12.00 | 0.2786 0.3152 0.4676 | - 0.0840 - 0.0967 - 0.1468 | 0.5492 0.6345 1.1242 | - 2.48 - 2.21 - 1.57 | - 19.15 - 13.64 - 6.71 |
| IU-FSU 2 | 100 | 231.3 | 1 . 25 1 . 4 1 . 9 | 13.96 13.74 12.44 | 0.2644 0.3010 0.4512 | - 0.0781 - 0.0902 - 0.1380 | 0.4697 0.5524 1.0100 | - 10.26 - 8.85 - 7.44 | - 39.32 - 24.81 - 16.17 |
| IU-FSU 3 | 47.2 | 220 | 1 . 25 1 . 4 1 . 9 | 12.50 12.48 11.70 | 0.2954 0.3313 0.4799 | - 0.0863 - 0.0991 - 0.1496 | 0.6547 0.7362 1.2156 | +0.17 +0.21 +0.34 | +0.05 +0.22 +0.87 |
| IU-FSU 4 | 47.2 | 260 | 1 . 25 1 . 4 1 . 9 | 12.51 12.52 11.80 | 0.2951 0.3304 0.4757 | - 0.0857 - 0.0984 - 0.1480 | 0.6533 0.7307 1.1838 | - 0.44 - 0.51 - 0.77 | - 0.18 - 0.54 - 1.77 |
| SkIU-FSU | 47.2 | 231.2 | 1 . 25 1 . 4 1 . 9 | 11.81 11.71 10.85 | 0.3125 0.3531 0.5173 | - 0.0909 - 0.1054 - 0.1639 | 0.7756 0.8917 1.5226 | +5.32 +6.52 +9.89 | +15.63 +21.38 +26.35 |
| IU-FSU (ST) | 47.2 | 231.3 | 1 . 9 | 12.29 | 0.4565 | - 0.1509 | - | +1.21 | - |
## IV. CURVATURE AS THE STRENGTH OF THE GRAVITATIONAL FIELD
In the absence of the cosmological constant the vacuum solution of the Einstein's equations of GR, also known as the Schwarzchild exterior solution, suggests that it is natural to measure the strength of the gravitational field in terms of the compactness parameter, η ≡ 2 GM/c r 2 . It can take any value between 0 < η < 1, with η = 0 corresponding to the flat Minkowski space in special relativity, while η = 1 corresponds to the strongest gravitational field that can be observed by an observer at infinity, the region which is known as the black hole event horizon. However, as it was shown in Ref. [21] the compactness parameter η , which is directly related to the Newtonian gravitational potential, could not serve as a fundamental parameter in describing the gravitational field in GR. Indeed the strength of the gravitational field is not characterized by a strong field potential, but by a large curvature [21]. There are several quantities that can be used as a measure of the curvature of gravitational field. These could, for example, be the Ricci scalar, the Ricci tensor, and the Riemann tensor. However, both the Ricci scalar and all components of the Ricci tensor vanish in the exterior region of compact objects, which makes them unsuitable choice for the measure of curvature. On the other hand, there is a non-vanishing component of the Riemann tensor even in the vacuum, R 1 010 = -2 GM/c r 2 3 ≡ -ξ , which indicates that it should be a more relevant measure of curvature in neutron stars [5, 21]. Since there are totally 20 independent components of the Riemann tensor, it is best that one uses the full contraction of the Riemann tensor instead, which is a single scalar invariant. Therefore following the Ref. [5] we use the square root of the Kretschmann invariant, K ≡ √ R µνρσ R µνρσ , which is a full contraction of the Riemann tensor. For a spherically symmetric mass configuration it can be expressed in terms of the pressure P r ( ), energy density E ( r ), and mass M r ( ) profiles as [5],
$$\mathcal { K } ^ { 2 } = \kappa ^ { 2 } \left [ 3 \left ( \mathcal { E } ( r ) + P ( r ) \right ) ^ { 2 } - 4 \mathcal { E } ( r ) P ( r ) \right ] - \kappa \mathcal { E } ( r ) \frac { 1 6 G M ( r ) } { c ^ { 2 } r ^ { 3 } } + \frac { 4 8 G ^ { 2 } M ^ { 2 } ( r ) } { c ^ { 4 } r ^ { 6 } },$$
where κ ≡ 8 πG/c 4 . One can then readily evaluate the Kretschmann invariant while integrating the OV equations. Notice that outside the star this curvature has a desired form, K ≡ 2 √ 3 . ξ
It is common to measure the strength of gravity in neutron stars with respect to that of the solar system. At the surface of the Sun, the curvature is equal to K glyph[circledot] ≡ 3 0 . × 10 -17 km -2 , which may be regarded as a small quantity. On the other hand, neutron stars can have surface curvature of as large as 14 orders of magnitude greater. In Fig. 3 we display the ratio of the neutron-star surface curvature to the one of the Sun, K R / K glyph[circledot] , as a function of the neutron star mass. Clearly, this ratio becomes even larger for more massive neutron stars. Notice that for less massive stars
7
7
FIG. 3. (color online). The ratio of surface curvatures of neutron stars and the Sun, K R / K glyph[circledot] , as a function of the neutron stars gravitational mass for the EOSs considered in this work. Left panel (a) represents the uncertainty range of the nuclear symmetry energy, while the right panel (b) represents the uncertainty range of the incompressibility K 0 of SNM.
the uncertainty in the nuclear symmetry energy has a large effect on the curvature; with variation of ∼ 50% in the surface curvature among EOSs for stars less than 1 4 . M glyph[circledot] . Depending on the symmetry energy, a surface curvature of K R = 2 0 . × 10 14 K glyph[circledot] can be obtained for neutron stars with masses in the range of 1 0 . < M /M G glyph[circledot] < 1 5. . The knowledge about the symmetry energy thus becomes crucial in testing various models of strong-field gravity. The effects of the symmetry energy are less pronounced on the curvature of massive neutron stars, but still significant. A curvature of K R = 4 0 . × 10 14 K glyph[circledot] have been obtained for massive neutron stars in the range of 1 72 . < M /M G glyph[circledot] < . 1 95. Quantitatively, the relative variation in K R of a canonical 1.4 solar-mass NS predicted by the IU-FSU 2 and SkIU-FSU is 61%, and this change reduces to only 51% for a 1.9 solar-mass NS. This is comparable to the relative change between NSs with masses 1 4 . M glyph[circledot] and 1 9 . M glyph[circledot] within the two models: 83% and 71% for the IU-FSU 2 and SkIU-FSU, respectively (See Table I).
Coupled with the sudden increase in the surface curvature of neutron stars close to their maximum mass, this emphasizes the fact that lower mass neutron stars are more natural laboratories for testing the EOS, and that only once the EOS is significantly constrained can high mass neutron stars come into their own as laboratories for testing alternative models for gravity.
In the same spirit as in the previous section, we now compare the effect of the uncertainties due to the variations of density dependence of the symmetry energy at saturation density and at high densities. There is a 25% reduction in the value of the surface curvature of a 1 4 . M glyph[circledot] NS, when we compare models with the soft and stiff symmetry energies as controlled by their density slope at saturation of L = 47 2 MeV (IU-FSU) and . L = 100 MeV (IU-FSU ), 2 correspondingly (See Table I). Obviously, the curvature is much larger when the soft symmetry energy is used.
This can be compared with a reduction in stiffness in the high-density EOS: comparing predictions of the SkIU-FSU and IU-FSU, we see that the surface curvature further increases by about 21% for a 1 4 . M glyph[circledot] NS and 26% for a 1 9 . M glyph[circledot] NS. Thus the surface curvature is significantly more sensitive than the gravitational binding energy to the variation of the density dependence of the symmetry energy at both saturation densities and higher. Again, its variation with the incompressibility of SNM is negligible (See Table I and Fig. 3).
We next consider the strength of the gravitational field within the neutron star. Notice that towards the center of the star the compactness η would approach zero. On the contrary, the curvature stays non-zero throughout the neutron star [5]. In order to study the strength of the gravitational field within the NS, we normalize the curvature profile K ( r ) with respect to its value at the surface of the star K R . The normalized profile tells us how many times stronger the curvature K ( r ) gets as one approaches the central region. This also provides a convenient way to compare effects of the density dependence of the symmetry energy on the same basis. In Fig. 4 we plot the curvature profiles
/
(r)
FIG. 4. (color online). Normalized curvature profiles K ( r ) with respect to its value at the surface of the star K R for the EOSs considered.
for a 1.4 and 1.9 solar-mass neutron stars using the EOSs considered in this work. For a canonical 1 4 . M glyph[circledot] NS the central curvature is 2 8 . to 3 5 . times stronger then the surface curvature depending on the EOS used. This ratio becomes larger for massive neutron stars, and in particular for a 1.9 solar mass NS it can be as large as 3 6 to 4 2 . . times stronger. Notice that the ratio becomes especially larger for models with the stiff saturation-density symmetry energy (See Fig. 4), whereas the high density behavior of the symmetry energy has a significantly smaller effect on the ratio of curvatures. Again, we observe a very small effect due to the variation of K 0 .
Fig. 3 and Fig. 4 together give us a complete picture of the curvature K ( r ) throughout the neutron star. First of all, the curvature of massive neutron stars at the surface is much larger then that of canonical NSs. Secondly, as one moves towards the center the strength of gravity becomes even stronger. Whereas this explicitly demonstrates the important role of massive neutron stars in studying the strong-field gravity, the uncertainties of the EOS resulting from our poor knowledge of the density dependence of the symmetry energy cannot be overlooked. It was put forward in Ref. [24] that the two solar mass neutron star in PSR J0348+0432-which is the most massive neutron star observed up to date-can be used as a testbed for gravity in the strong field regime. This conclusion was based on the fact that its binding energy is far outside the presently tested binding energy range. While supporting this conclusion, we have also analyzed the effects of the density dependence of the symmetry energy on the strength of the gravitational field and have found that current uncertainties in the EOS are still very large to cleanly probe the fundamental physics in the extreme conditions. Nevertheless, massive NSs that are manifest as radio pulsars remain valuable probes of both models of gravity and theories of dense matter in the extreme conditions that are inaccessible to the terrestrial experiments. In the next section we will analyze our results for the case of a neutron star in the scalar-tensor theories.
where
$$\alpha ( \varphi ) \equiv \frac { \partial \ln A ( \varphi ) } { \partial \varphi } \.$$
At the limiting case of A ϕ ( ) = 1, and V ( φ ) = 0 one recovers the general relativistic OV Eqs. (5). Following the Ref. [79] we set V ( ϕ ) = 0 which can only appear in models of modified gravity, and consider a coupling function of the form
$$A ( \varphi ) = \exp \left ( \alpha _ { 0 } \varphi + \frac { 1 } { 2 } \beta _ { 0 } \varphi ^ { 2 } \right ) \.$$
Again once an EOS is supplemented, for a given central pressure P (0) = P c one can then integrate the equations above from the center of the star to r → ∞ . At the center of the star the Einstein frame boundary conditions are given as
$$P ( 0 ) = P _ { \text{c} } \, \quad \mathcal { E } ( 0 ) = \mathcal { E } ( \text{c} ) \, \quad \varphi ( 0 ) = \varphi _ { \text{c} } \, \quad \chi ( 0 ) = 0$$
while at infinity we demand cosmologically flat solution to agree with the observation
$$\lim _ { r \to \infty } \varphi ( r ) = 0 \.$$
The stellar coordinate radius is then determined by the condition of P r ( s ) = 0. The physical radius of a neutron star is found in the Jordan frame as R NS = A 2 [ ϕ r ( s )] r s . Notice however that the physical stellar mass as measured by an observer at infinity, also known as the Arnowitt-Deser-Misner (ADM) mass M ADM , matches with the coordinate
## V. NEUTRON STARS IN A SCALAR-TENSOR MODEL OF GRAVITY
Motivated by the fact that the properties of neutron stars are not only sensitive to the underlying EOS, but also to the strong-field behavior of gravity, in this section we will consider neutron stars in the simplest natural extension of the GR known as the scalar-tensor theories of gravity. According to these theories, in addition to the second-rank metric tensor, g µν , the gravitational force is also mediated by a scalar field, ϕ . The most general form of the action defining this theory is written as [79, 80]
$$S = \frac { c ^ { 4 } } { 1 6 \pi G } \int d ^ { 4 } x \sqrt { - g ^ { * } } \left [ R ^ { * } - 2 g ^ { * \mu \nu } \partial _ { \mu } \varphi \partial _ { \nu } \varphi - V ( \varphi ) \right ] + S _ { \text{matter} } \left ( \psi _ { \text{matter} } ; A ^ { 2 } ( \varphi ) g _ { \mu \nu } ^ { * } \right )$$
where R ∗ is the Ricci scalar curvature with respect to the so-called Einstein frame metric g ∗ µν and V ( ϕ ) is the scalar field potential. The physical metric, also known as Jordan frame metric , is defined as g µν ≡ A 2 ( ϕ g ) ∗ µν . Notice that the GR is automatically recovered in the absence of the scalar field. The relativistic equations for stellar structure then can be written as
$$\frac { d M _ { G } ( r ) } { d r } = \frac { 4 \pi } { c ^ { 2 } } r ^ { 2 } \mathcal { E } ( r ) A ^ { 4 } ( \varphi ) + \frac { r ^ { 2 } } { 2 } \left ( 1 - \frac { 2 G M ( r ) } { c ^ { 2 } r } \right ) \chi ^ { 2 } ( r ) + \frac { r ^ { 2 } } { 4 } V ( \varphi )$$
dM
B
(
r
)
2
3
dr
= 4
πm r ρ r A
B
(
)
(
ϕ
)
,
(11)
$$\frac { d \varphi ( r ) } { d r } = \chi ( r ) \, \quad \ \.$$
$$\frac { d r } { d r } & = \left [ 1 - \frac { 2 G M ( r ) } { c ^ { 2 } r } \right ] ^ { - 1 } \left \{ \frac { 4 \pi G } { c ^ { 4 } } A ^ { 4 } ( \varphi ) \left [ \alpha ( \varphi ) \left ( \varepsilon ( r ) - 3 P ( r ) \right ) + r \chi ( r ) \left ( \varepsilon ( r ) - P ( r ) \right ) \right ] \\ & \quad - \frac { 2 } { r } \left ( 1 - \frac { G M ( r ) } { c ^ { 2 } r } \right ) \chi ( r ) + \frac { 1 } { 2 } r \chi ( r ) V ( \varphi ) + \frac { 1 } { 4 } \frac { d V ( \varphi ) } { d \varphi } \right \}, \\ & \quad \vdash - 1 \.$$
$$\dots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \dots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \,
\frac { d P ( r ) } { d r } & = - \left ( \mathcal { E } ( r ) + P ( r ) \right ) \left [ 1 - \frac { 2 G M ( r ) } { c ^ { 2 } r } \right ] ^ { - 1 } \left \{ \frac { 4 \pi G } { c ^ { 4 } } r A ^ { 4 } ( \varphi ) P ( r ) + \frac { G } { c ^ { 2 } r ^ { 2 } } M ( r ) \\ & + \left ( 1 - \frac { 2 G M ( r ) } { c ^ { 2 } r } \right ) \left ( \frac { 1 } { 2 } r \chi ^ { 2 } ( r ) + \alpha ( \varphi ) \chi ( r ) \right ) - \frac { 1 } { 4 } r V ( \varphi ) \right \} \,,$$
mass, since at infinity the coupling function approaches unity. It was shown in [81] that measurement of the surface atomic line redshifts from neutron stars could be used as a direct test of strong-field gravity theories. In particular, as a representative of the scalar-tensor theory a model with α 0 = 0 and β 0 = -8 was used. Notice that however a significant improvement have been made since then in constraining the α 0 and β 0 parameters. For example, very recently the authors of Ref. [25] put the most stringent constraint on these parameters from the observations of the 10-year timing campaign on PSR J1738+0333. According to these latest constraints the quadratic parameter should take values of β 0 glyph[greaterorsimilar] -5 0. . Moreover, as shown first in Ref. [79] predictions by models with β 0 > -4 35 can not in . general be distinguished from the general relativistic resultsi.e. from the model with { α , β 0 0 } = { 0 0 , } -due to the so-called 'spontaneous scalarization' effect. Since we have already showed that the uncertainties in the SNM EOS due to the variation of K 0 has insignificant effects, in this section we will solely concentrate on the effects of the symmetry energy. Further, we will choose the upper bounds on the parameters of the scalar-tensor theory to be α 2 0 < . 2 0 × 10 -5 and β 0 > -5 0 as constrained by the observation [25]. .
In Fig. 5 we present our results for the fractional gravitational binding energy as a function of the stellar mass (Fig. 5a) and the mass versus radius relation (Fig. 5b). The general relativistic (GR) predictions are then compared with the predictions of the scalar-tensor theories using different EOSs. We observe that the GR predictions give a lower absolute value of the fractional binding energy for a given stellar mass then the scalar-tensor theory, in general. However the difference is quite negligible as far as the uncertainties in the EOS is concerned. Moreover, low-mass neutron-stars are indistinguishable in these two models of gravity, because the critical value for the so-called 'spontaneous scalarizaton' occurs only when NSs exceed masses of about 1 4 solar mass. . To further illuminate the effects of compactness on the binding energy we can look at the mass versus radius relation. We observe that there is a slight change in this relation; notably, massive neutron stars have smaller radii in GR. If one uses a different value of β 0 = -8 0 as in the Ref. [81]-which . has already been excluded by the observation-one can obtain very massive neutron stars. In particular, the IU-FSU EOS gives a maximum mass of 3 2 . M glyph[circledot] and a slightly larger radius of 14 5 . km. However, under present observational constraints, differences in predictions using the GR and the scalar-tensor theories are much smaller than those due to the uncertainties in the EOS. Thus measurements of gravitational binding energy and neutron star radii will lead to significant constraints on the neutron star symmetry energy at saturation and high densities, rather than constraining gravity models between GR and scalar-tensor theories.
## VI. FUTURE DETERMINATION OF THE NUCLEAR SYMMETRY ENERGY FROM THE SURFACE GRAVITATIONAL REDSHIFT AND THE MAXIMUM SPIN FREQUENCY
Having established that uncertainties in the EOS arising from the nuclear symmetry energy outweigh those within the set of Scalar-Tensor gravity theories, we now seek in this section for some possible astrophysical measurements (see Ref. [82] for review) that may help constrain the density dependence of nuclear symmetry energy. In Fig. 2 we showed the binding energy, which can be inferred from the masses of binary pulsars assuming a specific evolutionary pathway for which strong (albeit circumstantial) evidence exists for two systems in particular; currently such inferences favor a softer symmetry energy [26, 27]. It was shown by some of us previously that the pulsar moment of inertia is an observable sensitive to the density dependence of the nuclear symmetry energy [83, 84]. In particular, it was shown that knowledge of the moment of inertia for pulsar PSR J0737-3039A with a 10% accuracy could help discriminate among various equations of state. Moreover, it was recently shown by some of us that the observed symmetry energy effect on the tidal polarizability (also known as tidal deformability) parameter in binary neutron stars is very strong [49], and in particular it has been shown that future measurement of the tidal polarizability in binary mergers would stringently constrain the high-density behavior of the symmetry energy. Perhaps the simplest of all, the knowledge of stellar mass and radius even from a single neutron star would provide very useful information on the equation of state of neutron star matter as first pointed out by Ref. [85]. While the mass of the neutron star can be measured fairly accurately in binary systems, the measurement of its radius is usually very complicated. From the observational point of view, it is more relevant to describe stellar properties with the so-called radiation radius ,
$$R _ { \infty } = \frac { R } { \sqrt { 1 - \eta _ { R } } }$$
and the surface gravitational redshift,
$$z = \frac { 1 } { \sqrt { 1 - \eta _ { R } } } - 1 \.$$
Notice that the expression for redshift is only valid in the slow-rotation approximation, where the Doppler effect due to the rotation of the star and the Lense-Thirring effect have been neglected. The radiation radius R ∞ could be
M
B
M M
/
/
FIG. 5. (color online). The fractional gravitational binding energy (a) and the mass versus radius relation (b) calculated using the EOSs considered in this work. The upper observational bound on the scalar-tensor theory parameters of { α , β 0 0 } = { √ 2 0 . × 10 -5 , -5 0 . } have been used.
obtained from a combination of flux and temperature measurements from the neutron star's surface and distance to the source. There are however some major uncertainties involved in the determination of R ∞ such as the distance, interstellar absorption, and details concerning the composition of the atmosphere and its magnetic field strength and structure [2]. Another possible combination to uniquely determine the stellar mass and radius could come from the inference of the surface gravitational acceleration and the gravitational redshift as pointed out by Ref. [28]. The gravitational redshift can be measured from the possible shift of atomic absorption lines in spectra of the stars. One of the main difficulties in measuring the gravitational redshift of signals emitted from the surface of neutron stars is due to the existence of strong surface magnetic field. Indeed, a value of z = 0 35 was reported [86] based on an .
0.6
0.6
z
FIG. 6. (color online). The effect of the density dependence of the nuclear symmetry energy (a) and the incompressibility coefficient (b) on the surface gravitational redshift z of neutron stars.
analysis of the stacked bursts in the low-mass X-ray binary EXO 0748-676. Unfortunately, subsequent observations could not confirm the existence of such spectral lines [87]. Nevertheless, future developments of large-area instruments in X-ray spectroscopy missions could make it possible to detect the spectral lines from the surface [88].
In Fig. 6, we plot the surface gravitational redshift as a function of the stellar mass for various equations of state discussed in previous sections. As it is evident from the figure, the effect of the density dependence of the nuclear symmetry energy is quite significant. For a canonical neutron star, uncertainties resulting from the density dependence of the symmetry energy would allow a range of predictions for the surface redshifts spanning from 0 197 . < z < 0 244. . Interestingly, had the z = 0 35 redshift in EXO 0748-676 been confirmed, its mass would then have to lie in the range . of 1 71 . M glyph[circledot] < M < 1 89 . M glyph[circledot] similar to that found earlier in [89]. On the other hand, if future redshift measurements report z > 0 6 then all current models of the equation of state would be ruled out. . Obviously, the effect of rotation would significantly alter the predictions for the redshift, and therefore should be taken into account. Such a study is in progress and results will be reported elsewhere.
Next, we plot in Fig. 7 the surface gravitational redshift z as a function of the radiation radius R ∞ by varying only the density dependence of the symmetry energy. The radiation radius of the low-mass X-ray source, EXO 1745-248 has been determined to be R ∞ = 14 57 . ± 1 64 km (shaded band) . [90, 91]. It is seen that the IU-FSU model with L = 100 MeV is almost excluded by this observation regardless of the gravitational redshift for this source. Future precise determination of the radiation radius from the low-mass X-ray binaries would therefore more stringently constrain the density dependence of the symmetry energy.
A certain combination of the stellar mass and radii could alternatively be constrained by measuring the spin frequencies. The maximum spin frequency for a stellar structure is known as the Kepler frequency, which in Newtonian gravity can be calculated using f K = 1 2 π √ GM R 3 . The general relativistic expression for the maximum spin frequency is very complicated and can be found by solving an equation involving metric functions and their derivatives [93]. As first shown in Ref. [29], the maximum spin frequency for a maximum allowable mass configuration can be analytically written in terms of the stellar properties as f K max , = C ( M max M glyph[circledot] ) 1 / 2 ( R M max 10km ) -3 / 2 , where the M max and R M max are the maximum allowable mass and the corresponding radius for static configurations. The most recent and updated analysis gives a fitting coefficient of C = 1 08 kHz [94]. . While this expression is highly accurate in estimating the maximum spin frequency, it is limited to the maximum mass configuration only. Moreover, rotating stars assume an oblate shape thus allowing a larger maximum mass than their static counterparts for a given equation of state. Therefore, without relying on the approximate expression we have used the RNS code for rapidly rotating neutron stars [92]. In Fig. 8 we show the maximum spin rate as a function of the stellar mass. The mass of the fastest spinning
FIG. 7. (color online). Surface gravitational redshift z as a function of stellar radiation radii R ∞ are constructed from the four EOSs discussed in the text.
pulsar known today J1748-2446ad-with a frequency of f = 716 356 Hz-is currently unknown [95]. . Should it be a low-mass neutron star with M < 1 29 . M glyph[circledot] , models with larger density slope of the symmetry energy at saturation would be excluded. While not confirmed by subsequent observations, a weak evidence for an f = 1122 Hz signal has been reported during thermonuclear bursts from the XTE J1739285 neutron star [96]. Although this observation does not have strong statistical significance, our calculations indicate that its mass should be above 1 65 . M <M < . glyph[circledot] 2 12 M glyph[circledot] depending on the EOS model (see Fig. 8). Conversely, if such a fast-spinning neutron star had been found to have a mass of M = 1 7 . M glyph[circledot] or less, then only models with the very soft nuclear symmetry energy such as the SkIU-FSU would have survived. Although at this stage there is no known pulsars with larger spin frequencies, future measurements of such frequencies would be decisive in ruling out models predicting the stiff nuclear symmetry energy.
## VII. SUMMARY
There is a degeneracy between the EOS for super-dense matter and the strong-field gravity in understanding properties of neutron stars. Our goal is to provide information that may help break this degeneracy. Since the most uncertain part of the nucleonic EOS is currently the density dependence of nuclear symmetry energy especially at super-saturation densities, we have studied effects of the nuclear symmetry energy within its current uncertain range on the binding energy and curvature of neutron stars within GR and the scalar-tensor theory of gravity. We found that the gravitational binding energy is moderately sensitive to the underlying symmetry energy, with lower mass neutron stars glyph[lessorsimilar] 1 4 . M glyph[circledot] probing primarily the saturation-density stiffness of the symmetry energy, and higher mass stars being more sensitive to the high density behavior of the symmetry energy. For the most massive NSs discovered so far, we found an almost 20% change in their binding energies by varying the density dependence of the symmetry energy within its uncertainty range determined by recent terrestrial nuclear laboratory experiments. We also found that the binding energy is quite insensitive to the variation of the incompressibility K 0 of SNM EOS.
The curvature of neutron stars measured using the square root of the full contraction of the Riemann tensor was found to be significantly more sensitive to the variation of the density dependence of nuclear symmetry energy but not to that of the incompressibility K 0 , with variations in surface curvature of order 50% over the whole mass range.
Within the scalar-tensor theory of gravity which is the simplest natural extension of the GR, using upper bounds on the parameters of the scalar-tensor theory from observation, we find negligible change in the binding energy of
)
Hz
(
FIG. 8. (color online). Maximum spin frequencies as a function of stellar masses obtained calculated for the four EOSs discussed in the text using the publicly available code for rapidly rotating neutron stars [92].
neutron stars over the whole mass range, and significant changes in radii only for neutron stars well above 1 4 . M glyph[circledot] . Even then, the changes in radii are smaller than those that result from variation of the stiffness of the symmetry energy.
Restricting ourselves to the scalar-tensor modifications to gravity, we conclude that astrophysical measurements of quantities related to the compactness and curvature of neutron stars such as the gravitational binding energy, atomic line redshifts, and the maximum spin frequency will help constrain the neutron star EOS further rather than the model of gravity. However, the large range of curvatures predicted by varying the symmetry energy may still mean that current uncertainties in the density dependence of nuclear symmetry energy are still too large to clearly break the EOS-gravity degeneracy in determining properties of neutron stars for a much wider class of gravity models.
## VIII. ACKNOWLEDGEMENTS
This work is supported in part by the National Natural Science Foundation of China under Grant No. 11275098, 11275067 and 11320101004, the State Scholarship Fund of China Scholarship Council (CSC), the US National Science Foundation under Grant No. PHY-1068022 and the CUSTIPEN (China-U.S. Theory Institute for Physics with Exotic Nuclei) under DOE grant number DE-FG02-13ER42025.
- [1] J. M. Lattimer and M. Prakash, Astrophys. J. 550 , 426 (2001).
- [2] J. M. Lattimer and M. Prakash, Phys. Rept. 442 , 109 (2007).
- [3] J. M. Lattimer, Ann. Rev. Nucl. Part. Sci. 62 , 485 (2012).
- [4] P. Pani, E. Berti, V. Cardoso, and J. Read, Phys. Rev. D84 , 104035 (2011).
- [5] K. Y. Eksi, C. Gungor, and M. M. Turkoglu, Phys. Rev. D89 , 063003 (2014).
- [6] C. M. Will, Living Rev. Rel. 9 , 3 (2006).
- [7] T. Harada, Phys. Rev. D57 , 4802 (1998).
- [8] H. Sotani and K. D. Kokkotas, Phys. Rev. D70 , 084026 (2004).
- [9] P. D. Lasky, H. Sotani, and D. Giannios, Phys. Rev. D78 , 104019 (2008).
- [10] D.-H. Wen, B.-A. Li, and L.-W. Chen, Phys. Rev. Lett. 103 , 211102 (2009).
- [11] A. Cooney, S. DeDeo, and D. Psaltis, Phys. Rev. D82 , 064033 (2010).
- [12] M. Horbatsch and C. Burgess, JCAP 1108 , 027 (2011).
- [13] A. S. Arapoglu, C. Deliduman, and K. Y. Eksi, JCAP 1107 , 020 (2011).
- [14] C. Deliduman, K. Eksi, and V. Keles, JCAP 1205 , 036 (2012).
- [15] H. Sotani, Phys. Rev. D86 , 124036 (2012).
- [16] J. Xu, B.-A. Li, L.-W. Chen, and H. Zheng, J. Phys. G40 , 035107 (2013).
- [17] W. Lin, B.-A. Li, L.-W. Chen, D.-H. Wen, and J. Xu, J. Phys. G41 , 075203 (2014).
- [18] K. Yagi, L. C. Stein, N. Yunes, and T. Tanaka, Phys.Rev. D87 , 084058 (2013).
- [19] H. Sotani, Phys. Rev. D89 , 104005 (2014).
- [20] B.-A. Li, L.-W. Chen, and C. M. Ko, Phys. Rept. 464 , 113 (2008).
- [21] D. Psaltis, Living Rev. Rel. 11 , 9 (2008).
- [22] A. Akmal and V. R. Pandharipande, Phys. Rev. C56 , 2261 (1997).
- [23] P. Haensel, M. Kutschera, and M. Proszynski, Astron. Astrophys. 102 , 299.
- [24] J. Antoniadis, P. C. Freire, N. Wex, T. M. Tauris, R. S. Lynch, et al. , Science 340 , 6131 (2013).
- [25] P. C. Freire, N. Wex, G. Esposito-Farese, J. P. Verbiest, M. Bailes, et al. , Mon. Not. Roy. Astron. Soc. 423 , 3328 (2012).
- [26] P. Podsiadlowski, J. Dewi, P. Lesaffre, J. Miller, W. G. Newton, et al. , Mon. Not. Roy. Astron. Soc. 361 , 1243 (2005).
- [27] W. G. Newton and B.-A. Li, Phys. Rev. C80 ,
- 065809 (2009).
- [28] A. Majczyna and J. Madej, Acta Astron. 55 , 1 (2005).
- [29] P. Haensel and J. L. Zdunik, Nature 340 , 617 (1989).
- [30] D. H. Youngblood, H. L. Clark, and Y.-W. Lui, Phys. Rev. Lett. 82 , 691 (1999).
- [31] B. K. Agrawal, S. Shlomo, and V. K. Au, Phys. Rev. C68 , 031304 (2003).
- [32] G. Col`, N. Van Giai, J. Meyer, K. Bennaceur, o and P. Bonche, Phys. Rev. C70 , 024307 (2004).
- [33] U. Garg et al. , Nucl. Phys. A788 , 36 (2007).
- [34] S. Shlomo, V. M. Kolomietz, and G. Colo, Eur. Phys. J. A30 , 23 (2006).
- [35] J. Piekarewicz, J. Phys. G37 , 064038 (2010).
- [36] L.-W. Chen and J.-Z. Gu, J. Phys. G39 , 035104 (2012).
- [37] D. Patel, U. Garg, M. Fujiwara, H. Akimune, G. Berg, et al. , Phys. Lett. B718 , 447 (2012).
- [38] P. Danielewicz, R. Lacey, and W. G. Lynch, Science 298 , 1592 (2002).
- [39] P. B. Demorest, T. Pennucci, S. M. Ransom, M. S. E. Roberts, and J. W. T. Hessels, Nature 467 , 1081 (2010).
- [40]
- L.-W. Chen, C. M. Ko, and B.-A. Li, Phys. Rev. Lett. 94 , 032701 (2005).
- [41] D. V. Shetty, S. J. Yennello, and G. A. Souliotis, Phys. Rev. C76 , 024606 (2007).
- [42] A. Klimkiewicz et al. , Phys. Rev. C76 , 051603 (2007).
- [43] M. B. Tsang et al. , Phys. Rev. Lett. 102 , 122701 (2009).
- [44] M. Centelles, X. Roca-Maza, X. Vinas, and M. Warda, Phys. Rev. Lett. 102 , 122502 (2009).
- [45] C. Xu, B.-A. Li, and L.-W. Chen, Phys. Rev. C82 , 054607 (2010).
- [46] A. W. Steiner and S. Gandolfi, Phys. Rev. Lett. 108 , 081102 (2012).
- [47] J. M. Lattimer and Y. Lim, Astrophys. J. 771 , 51 (2013).
- [48] M. Tsang et al. , Phys. Rev. C86 , 015803 (2012).
[49]
F. J. Fattoyev, J. Carvajal, W. G. Newton, and B.-A. Li, Phys. Rev.
C87
,
015806 (2013).
- [50] B.-A. Li and X. Han, Phys. Lett. B727 , 276 (2013).
- [51] W. G. Newton, M. Gearheart, and B.-A. Li, Astrophys. J. Suppl. 204 , 9 (2013).
- [52] F. J. Fattoyev and J. Piekarewicz, Phys. Rev. Lett. 111 , 162501 (2013).
- [53] B.-A. Li, C. M. Ko, and W. Bauer, Int. J. Mod. Phys. E7 , 147 (1998).
- [54] V. Baran, M. Colonna, V. Greco, and M. Di Toro, Phys. Rept. 410 , 335 (2005).
- [55] L.-W. Chen, C. M. Ko, B.-A. Li, and G.-C. Yong, Front. Phys. China 2 , 327 (2007).
- [56] A. W. Steiner, M. Prakash, J. M. Lattimer, and P. J. Ellis, Phys. Rept. 411 , 325 (2005).
- [57] B.-A. Li, A. Ramos, G. Verde, and I. Vidana, eds., 'Topical Issue on Nuclear Symmetry Energy' , Eur. Phys. J. A50 , No. 2 (2014).
- [58] B. A. Brown, Phys. Rev. Lett. 85 , 5296 (2000).
- [59] A. Szmaglinski, W. Wojcik, and M. Kutschera, Acta Phys. Polon. B37 , 277 (2006).
- [60] A. E. L. Dieperink, Y. Dewulf, D. Van Neck, M. Waroquier, and V. Rodin, Phys. Rev. C68 , 064307 (2003).
- [61] Z. Xiao, B.-A. Li, L.-W. Chen, G.-C. Yong, and M. Zhang, Phys. Rev. Lett. 102 , 062502 (2009).
- [62] P. Russotto et al. , Phys. Lett. B697 , 471 (2011).
- [63] W. Trautmann and H. H. Wolter, Int. J. Mod. Phys. E21 , 1230003 (2012).
- [64] F. J. Fattoyev, W. G. Newton, J. Xu, and B.-A. Li, Phys. Rev. C86 , 025804 (2012).
- [65] F. J. Fattoyev, C. J. Horowitz, J. Piekarewicz, and G. Shen, Phys. Rev. C82 , 055803 (2010).
- [66] C. J. Horowitz and J. Piekarewicz, Phys. Rev. Lett. 86 , 5647 (2001).
- [67] G. Baym, C. Pethick, and P. Sutherland, Astrophys. J. 170 , 299 (1971).
- [68] J. R. Oppenheimer and G. M. Volkoff, Phys. Rev. 55 , 374 (1939).
- [69] S. Weinberg, 'Gravitation and Cosmology' (John Wiley & Sons, New York, 1972).
- [70] W. Arnett and R. Bowers, Astrophys. J. Suppl. 33 , 415 (1977).
- [71] J. M. Lattimer and M. Prakash, Phys. Rept. 333 , 121 (2000).
- [72] E. Alecian and S. M. Morsink, Astrophys. J. 614 , 914 (2004).
- [73] G. B. Cook, S. L. Shapiro, and S. A. Teukolsky, Astrophys. J. 424 , 823 (1994).
- [74] J. M. Lattimer and M. Prakash, (2010), arXiv:1012.3208 [astro-ph.SR].
- [75] M. Bagchi, Mon. Not. Roy. Astron. Soc. 413 , L47 (2011).
- [76] M. Kramer, I. H. Stairs, R. Manchester, M. McLaughlin, A. Lyne, et al. , Science 314 , 97 (2006).
- [77] F. Kitaura, H.-T. Janka, and W. Hillebrandt, Astron. Astrophys. 450 , 345 (2006).
- [78] R. D. Ferdman, I. H. Stairs, M. Kramer, G. H. Janssen, C. G. Bassa, B. W. Stappers, P. B. Demorest, I. Cognard, G. Desvignes, G. Theureau, M. Burgay, A. G. Lyne, R. N. Manchester, and A. Possenti, ArXiv e-prints (2014), arXiv:1406.5507 [astro-ph.SR].
- [79] T. Damour and G. Esposito-Farese, Phys. Rev. D54 , 1474 (1996).
- [80] S. S. Yazadjiev, D. D. Doneva, K. D. Kokkotas, and K. V. Staykov, (2014), arXiv:1402.4469 [gr-qc].
- [81] S. DeDeo and D. Psaltis, Phys. Rev. Lett. 90 , 141101 (2003).
- [82] M. C. Miller, (2013), arXiv:1312.0029 [astro-ph.HE].
- [83] A. Worley, P. G. Krastev, and B.-A. Li, Astrophys. J. 685 , 390 (2008).
- [84] F. J. Fattoyev and J. Piekarewicz, Phys. Rev. C82 , 025810 (2010).
- [85] L. Lindblom, Astrophys. J. 398 , 569 (1992).
- [86] J. Cottam, F. Paerels, and M. Mendez, Nature 420 , 51 (2002).
- [87] J. Cottam, F. Paerels, M. Mendez, L. Boirin, W. Lewin, E. Kuulkers, and J. M. Miller, Astrophys. J. 672 , 504.
- [88] S. Bhattacharyya, M. C. Miller, and F. K. Lamb, Astrophys. J. 644 , 1085 (2006).
- [89] B.-A. Li and A. W. Steiner, Phys. Lett. B642 , 436 (2006).
- [90] F. Ozel, T. Guver, and D. Psaltis, Astrophys. J. 693 , 1775 (2009).
- [91] J. M. Lattimer and A. W. Steiner, Eur. Phys. J. A50 , 40 (2014).
- [92] N. Stergioulas, 'Rapidly Rotating Neutron Star', http://www.gravity.phys.uwm.edu/rns/ .
- [93] F. Weber, 'Pulsars as astrophysical laboratories for nuclear and particle physics,' (Institute of Physics Publishing, Bristol, UK, 1999).
- [94] P. Haensel, J. L. Zdunik, M. Bejger, and J. M. Lattimer, Astron. Astrophys. 502 , 605 (2009).
- [95] J. W. Hessels, S. M. Ransom, I. H. Stairs, P. C. C. Freire, V. M. Kaspi, et al. , Science 311 , 1901 (2006).
- [96] P. Kaaret, Z. Prieskorn, J. 't Zand, S. Brandt, N. Lund, et al. , Astrophys. J. 657 , L97 (2007).
|
10.1103/PhysRevC.91.015810
|
[
"Xiao-Tao He",
"F. J. Fattoyev",
"Bao-An Li",
"W. G. Newton"
] |
2014-08-05T03:46:28+00:00
|
2014-12-16T02:05:07+00:00
|
[
"nucl-th",
"astro-ph.SR",
"gr-qc",
"nucl-ex"
] |
Impact of the equation-of-state -- gravity degeneracy on constraining the nuclear symmetry energy from astrophysical observables
|
There is a degeneracy between the equation of state (EOS) of superdense
neutron-rich nuclear matter and the strong-field gravity in understanding
properties of neutron stars. While the EOS is still poorly known, there are
also longstanding ambiguities in choosing General Relativity or alternative
gravity theories in the not-so-well tested strong-field regime. Besides
possible appearance of hyperons and new phases, the most uncertain part of the
nucleonic EOS is currently the density dependence of nuclear symmetry energy.
To provide information that may help break the EOS-gravity degeneracy, we
investigate effects of symmetry energy within its uncertain range determined by
terrestrial nuclear laboratory experiments on the gravitational binding energy
and spacetime curvature of neutron stars within GR and the scalar-tensor (ST)
theory of gravity. In particular, we focus on effects of the following
parameters characterizing the EOS of neutron-rich nucleonic matter: (1) the
incompressibility $K_0$ of symmetric nuclear matter, (2) the slope $L$ of
symmetry energy at saturation density and (3) the high-density behavior of
symmetry energy. We find that the variation of either the density slope $L$ or
the high-density behavior of symmetry energy leads to large changes in both the
binding energy and curvature of neutron stars while effects of varying the more
constrained $K_0$ are negligibly small. The difference in predictions using the
GR and the ST theory appears only for massive neutron stars, and is
significantly smaller than the differences resulting from variations in the
symmetry energy. We conclude that within the ST theory of gravity, the
EOS-gravity degeneracy has been broken by the recent relativistic pulsar
measurements, and that measurements of neutron star properties sensitive to the
compactness constrain mainly the density dependence of the symmetry energy.
|
1408.0858v2
|
## AN EXTENSION OF A THEOREM OF HARTSHORNE
MORDECHAI KATZMAN, GENNADY LYUBEZNIK, AND WENLIANG ZHANG
Abstract. We extend a classical theorem of Hartshorne concerning the connectedness of the punctured spectrum of a local ring by analyzing the homology groups of a simplicial complex associated with the minimal primes of a local ring.
## 1. introduction
R. Hartshorne proved in [Har62, Proposition 2.1] that, if ( R, m ) is a noetherian local ring whose depth is at least 2, then the punctured spectrum ( i.e. Spec( R ) \{ m } ) is connected. In this note we partially extend this result to the case depth ≥ 3. We express our extension in terms of a simplicial complex (which already appeared in [Lyu07, Theorem 1.1]) defined as follows.
/negationslash
Definition 1.1. Let R be a noetherian commutative ring and let { p 1 , . . . , p n } be the minimal primes of R . Assume that R is either local with maximal ideal m or that R is graded with m the ideal of elements of positive degrees. A simplicial complex ∆( R ) (or ∆ whenever R are clear from the context) is defined as follows: ∆( R ) is the simplicial complex on the vertices 1 , . . . , n such that a simplex { i 0 , . . . , i s } is included in ∆( R ) if and only if √ p i 0 + · · · + p i s = m .
Note that the punctured spectrum of R ( i.e. Spec( R ) \{ m } ) is connected (as a topological space under the Zariski topology) if and only if ˜ H 0 (∆( R , G ) ) = 0 for some (equivalently every) non-zero abelian group, where ˜ H 0 (∆( R , G ) ) is the 0-th reduced singular homology of the simplicial complex ∆( R ) with coefficients in G . Thus Hartshorne's theorem says that if depth R ≥ 2, then ˜ H 0 (∆( R , G ) ) = 0 . Our main result is the following extension of Hartshorne's result.
Theorem 1.2. Let ( R, m , k ) be a noetherian commutative complete local ring of characteristic p . Assume that the residue field k is separably closed and that depth( R ) ≥ 3 . Then
$$\widetilde { H } _ { 0 } ( \Delta ( R ) ; k ) = \widetilde { H } _ { 1 } ( \Delta ( R ) ; k ) = 0. \\ \text{graded analog of Theorem 1.2 over any se}.$$
We also have a graded analog of Theorem 1.2 over any separably closed field (not just in characteristic p ).
Theorem 1.3. Let R = ⊕ i ∈ N R i be a standard N -graded ring over a separably closed field k (i.e. R 0 = k and R is finitely generated by R 1 as a k -algebra). If
2010 Mathematics Subject Classification. 13D45, 13F55, 14B15.
M.K. gratefully acknowledges support from EPSRC grant EP/J005436/1; G.L. was partially supported by NSF grant DMS #1161783, and W.Z. by NSF grants DMS #1247354/#1405602 and an EPSCoR First Award grant. G.L. and W.Z. were also supported by NSF grant 0932078000 while in residence at MSRI.
$$\widetilde { H } _ { 0 } ( \Delta ( R ) ; k ) = \widetilde { H } _ { 1 } ( \Delta ( R ) ; k ) = 0. \\ \text{orem provides a sufficient condition for denth}$$
/negationslash
Hartshorne's theorem provides a sufficient condition for depth( R/I ) ≤ 1. Namely, if ˜ H 0 (∆( R ); k ) = 0, then depth( R/I ) ≤ 1. Our Theorem 1.3 provides a sufficient condition for depth( R/I ) 2. Namely, if H (∆( R ); k ) = 0, then depth( R/I ) 2.
We also show by an example (Example 3.4) that Theorem 1.3 (and hence also Theorem 1.2) does not necessarily hold if one replaces k in ˜ H ∗ (∆( R ); k ) with Z .
≤ ˜ 1 ≤
/negationslash
We do not know whether the assumptions that R is complete and the residue field is separably closed in Theorem 1.2 or the assumption that k is separably closed in Theorem 1.3 can be removed. Neither do we know whether the analogue of Theorem 1.2 in characteristic 0 holds.
Finally, it would be very interesting to know whether there exists a similar connection between higher values of the depth of R and the vanishing of higher homology groups of the complex ∆( R ). When R is a Stanley-Reisner ring we obtain a natural extension of Theorem 1.3 (Corollary 3.5) and show that if depth R ≥ d then ˜ H j (∆( R , k ) ) = 0 for all 0 ≤ j ≤ d -2.
## 2. Proof of Theorem 1.2
In this section, we will prove our main technical result that implies Theorems 1.2 and 1.3. To this end, we recall the definition of the cohomological dimension.
Definition 2.1. Let I be an ideal of a noetherian commutative ring R . Then the cohomological dimension of the pair ( R,I ), denoted by cd( R,I ), is defined by
$$\text{cd} ( R, I ) \coloneqq \max \{ j | H _ { I } ^ { j } ( R ) \neq 0 \},$$
/negationslash where H j I ( R ) is the j th local cohomology module of R supported at I .
We will also consider the following spectral sequence ([ ` MGLZA03, p. 39]) A
$$\cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdot \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \ \colon \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon \colon \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \oplus \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \row }, \, H _ { 1 } ^ { b - a } } ( R ) ) \Rightarrow H _ { 2 } ^ { b - a } } ( R ) ) \Rightarrow H _ { 3 } ^ { b - a } } ( R ) ) \rightarrow H _ { 4 } ^ { b - a } } ( R ) ) \rightarrow H _ { 5 } ^ { b - a } } ( R ) ).$$
where R is a noetherian commutative ring, I is an ideal of R , and { p 1 , . . . , p s } is the set of minimal primes of I .
To prove Theorem 1.2, we will need the following results which are completely characteristic-free; they hold even in mixed-characteristic.
Lemma 2.2. Let I be an ideal of a d -dimensional noetherian complete local domain ( R, m , k ) and let { p 1 , . . . , p n } be the set of minimal primes of I . Then
$$E _ { 2 } ^ { - t, d } \cong H _ { t } ( \Delta _ { n }, \Delta ( R / I ) ; H _ { \mathfrak { m } } ^ { d } ( R ) ),$$
where E -t,d 2 is the E 2 -term of the spectral sequence (2.0.1), ∆ n denotes the full n -simplex, and H t (∆ n , ∆( R/I ); H d m ( R )) denotes the singular homology of the pair (∆ n , ∆( R/I )) with coefficients in H d m ( R ) .
Proof. The spectral sequence (2.0.1) with b = d gives us a complex
$$( 2. 0. 2 ) & & \cdots \to E _ { 1 } ^ { - 3, d } \to E _ { 1 } ^ { - 2, d } \to E _ { 1 } ^ { - 1, d } \to \cdots$$
/negationslash
$$\text{Consequently, the complex (2.0.2) becomes} \\ ( 2. 0. 3 ) \quad & \cdots \to \bigoplus _ { \substack { j _ { 0 } < \cdots < j _ { t } \\ \sqrt { \mathfrak { p } _ { j _ { 0 } } + \cdots + \mathfrak { p } _ { j _ { t } } = m } } } H _ { \mathfrak { m } } ^ { d } ( R ) \to \bigoplus _ { \substack { i _ { 0 } < \cdots < i _ { t - 1 } ; \\ \sqrt { \mathfrak { p } _ { i _ { 0 } } + \cdots + \mathfrak { p } _ { i _ { t - 1 } } = m } } } H _ { \mathfrak { m } } ^ { d } ( R ) \to \cdots \\ \cdots \quad & \cdots$$
whose homology is E ∗ ,d 2 . By the Hartshorne-Lichtenbaum vanishing theorem ([Har68, Theorem 3.1]), for each ideal a of R , one has H d a ( R ) = 0 if and only if √ a = m . Consequently, the complex (2.0.2) becomes
Let C ∗ (∆ ) and n C ∗ (∆( R/I )) denote the complexes of singular chains of ∆ n and ∆( R/I ) with coefficients in H d m ( R ) respectively and let ˜ C ∗ denote the complex of (2.0.3). One can see that the condition imposed on √ p 1 + · · · + p t appearing in (2.0.3) is the opposite to the one imposed in the definition of ∆( R/I ); consequently, one has a short exact sequence
$$0 \to C ^ { * } ( \Delta ( R / I ) ) \to C ^ { * } ( \Delta _ { n } ) \to \widetilde { C } ^ { * } \to 0, \\ \text{complex of singular chains of the pair } ( C ^ { * } ( \Delta _ { n } ), C ^ { * } ( \zeta$$
i.e. ˜ C ∗ is the complex of singular chains of the pair ( C ∗ (∆ ) n , C ∗ (∆( R/I ))) with coefficients in H d m ( R ). Hence the homology of the complex (2.0.3) is the relative homology H ∗ (∆ n , ∆( R/I ); H d m ( R )), i.e.
$$E _ { 2 } ^ { - t, d } \cong H _ { t } ( \Delta _ { n }, \Delta ( R / I ) ; H _ { \mathfrak { m } } ^ { d } ( R ) ),$$
for each integer t .
/square
Theorem 2.3. Let I be an ideal of a d -dimensional noetherian complete local domain ( R, m , k ) . Assume that cd( R,I ) ≤ d -3 and cd( R, p j ) ≤ d -2 for each minimal prime p j of I . Then
$$\widetilde { H } _ { 0 } ( \Delta ( R / I ) ; H _ { \mathfrak m } ^ { d } ( R ) ) = \widetilde { H } _ { 1 } ( \Delta ( R / I ) ; H _ { \mathfrak m } ^ { d } ( R ) ) = 0 \\ _ { * } ( \Delta ( R / I ) ; H _ { \mathfrak m } ^ { d } ( R ) ) \ d e n o t e s \ t h e \ r e d u c e d \ s i n q u l a r \ h o m o l o q y \ o f \ \Delta ( R$$
where ˜ H ∗ (∆( R/I ); H d m ( R )) denotes the reduced singular homology of ∆( R/I ) with coefficients in H d m ( R ) .
Proof. Assume that I has n minimal primes p 1 , . . . , p n , and we will consider the spectral sequence (2.0.1).
Grothedieck's Vanishing Theorem ([BS98, 6.1.2]) asserts that H j I ( M ) = 0 for any ideal I of a d -dimensional noetherian commutative ring A , any A -module M , and any integer j > d . Therefore, for each r ≥ 2, one has
$$\text{merge $j>a.\ I\ I\ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I\ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I
I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ II \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I i I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I i I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I \ I I \ I I \ I I \ I I \ I I I \ I I \ I I \ I I \ I I \ I I I \ I I I \ I I I \ I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I \ I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I \ I I \ I I \ I I I \ I I \ I I I \ I I I \ I I \ I I \ I I I \ I I I \ I I I \ I I I I \ I I I \ I I I \ I I \ I I I \ I I \ I I \ I I I \ I I I \ I I I \ I I \ I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I \ I I \ I I I \ I I \ I I \ I I I \ I I \ I I \ I I \ I I \ I I I \ I I \ I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I \ I I I \ I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I \ I I I \ I I \ I I I \ I I \ I I I \ I I \ I I \ I I I \ I I I \ I I \ I I \ I I I \ I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I \ I I \ I I \ I I \ I I I \ I I \ I I I \ I I \ I I \ I I \ I I I \ I I \ I I I \ I I \ I I \ I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I \ I I \ I I \ I I I \ I I I \ I I I \ I I \ I I \ I I I \ I I I \ I I I \ I I I \ I I I \ I I \ I I I \ I I \ I I \ I I \ I I I \ I I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \I I \ I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ i I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I \ I I I$$
hence the (incoming) differential E - -2 r,d + r -1 r → E -2 ,d r is 0. Since E 0 ,d -1 1 = ⊕ j H d -1 p j ( R ) and cd( R, p j ) ≤ d -2, we have E 0 ,d -1 1 = 0 and hence the (outgoing) differential E -2 ,d -1 2 → E 0 ,d -1 2 is 0. If r ≥ 3, then -2 + r > 0 and clearly E -2+ r,d -r +1 r = 0. This shows that, for r ≥ 2, all incoming and outgoing differentials from E -2 ,d r are 0; consequently E -2 ,d 2 = E -2 ,d ∞ . On the other hand, since cd( R,I ) ≤ d -3, we have H d -2 I ( R ) = 0. Therefore E -2 ,d 2 = 0. Similarly, one can also show that E -1 ,d 2 = 0. On the other hand, it follows from the HartshorneLichtenbaum vanishing theorem ([Har68, Theorem 3.1]) that E 0 ,d 1 = 0. Thus, E 0 ,d 2 = 0.
Therefore, it follows from Lemma 2.2 that
H 2 (∆ n , ∆( R/I ); H d m ( R )) = ∼ E -2 ,d 2 = 0 and H 1 (∆ n , ∆( R/I ); H d m ( R )) = ∼ E -1 ,d 2 = 0 .
The long exact sequence of homology
$$\stackrel { \kappa } { \cdots } & \stackrel { \kappa } { \stackrel { \kappa } { \cdots } } H _ { j } ( \Delta _ { n } ; H _ { \mathfrak { m } } ^ { d } ( R ) ) \rightarrow H _ { j } ( \stackrel { \kappa } { \Delta _ { n } }, \Delta ( R / I ) ; H _ { \mathfrak { m } } ^ { d } ( R ) ) \rightarrow H _ { j - 1 } ( \Delta ( R / I ) ; H _ { \mathfrak { m } } ^ { d } ( R ) ) \\ & \rightarrow H _ { j - 1 } ( \Delta _ { n } ; H _ { \mathfrak { m } } ^ { d } ( R ) ) \rightarrow \cdots \\ \text{minlies that}$$
implies that
$$H _ { 1 } ( \Delta ( R / I ) ; H _ { \mathfrak { m } } ^ { d } ( R ) ) \cong H _ { 1 } ( \Delta _ { n } ; H _ { \mathfrak { m } } ^ { d } ( R ) ) = 0$$
$$H _ { 0 } ( \Delta ( R / I ) ; H _ { \mathfrak { m } } ^ { d } ( R ) ) \cong H _ { 0 } ( \Delta _ { n } ; H _ { \mathfrak { m } } ^ { d } ( R ) ) = H _ { \mathfrak { m } } ^ { d } ( R )$$
$$& \slash I \rangle ; H _ { \mathfrak m } ^ { d } ( R ) ) \cong H _ { 1 } ( \Delta _ { n } ; H _ { \mathfrak m } ^ { d } ( R ) ) = 0 \\ & \slash I \rangle ; H _ { \mathfrak m } ^ { d } ( R ) ) \cong H _ { 0 } ( \Delta _ { n } ; H _ { \mathfrak m } ^ { d } ( R ) ) = H _ { \mathfrak m } ^ { d } ( R ) \\ & \wedge \cdot \quad \tilde { \lambda } \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \ \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \, \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cd. \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cd. \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cd. \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cd> \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cd>< \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdocks \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cd.. \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cd< \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdors \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdjs \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdts \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cd>] \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdot \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \$$
where H 1 (∆ ; n H d m ( R )) = 0 since ∆ n is contractible. Therefore, we have
$$\widetilde { H } _ { 1 } ( \Delta ( R / I ) ; H _ { \mathfrak { m } } ^ { d } ( R ) ) & = 0 \\ \widetilde { H } _ { 0 } ( \Delta ( R / I ) ; H _ { \mathfrak { m } } ^ { d } ( R ) ) & = 0.$$
This finishes the proof of our theorem.
$$\widetilde { H } _ { 0 } ( \Delta ( R / I ) ; H _ { \mathfrak { m } } ^ { d } ( R ) ) = 0. \\ \text{if our theorem.}$$
/square
Remark 2.4 . By the Universal Coefficient Theorem ([Hat02, Theorem 3A.3]), for each topological space X and an abelian group G , there is a short exact sequence
$$0 \to H _ { i } ( X ; \mathbb { Z } ) \otimes _ { \mathbb { Z } } G \to H _ { i } ( X ; G ) \to \text{Tor} _ { 1 } ( H _ { i - 1 } ( X ; \mathbb { Z } ), G ) \to 0. \\ \cdots \dots \, \tau / \nu. \, \pi \, \colon = \dots \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \, \dots \, \tau \end{matrix}$$
Since H 0 ( X ; Z ) is always a free Z -module, we have Tor ( 1 H 0 ( X ; Z ) , G ) = 0 and hence H 1 ( X G ; ) = ∼ H 1 ( X ; Z ) ⊗ Z G . In particular, with notation as in Theorem 2.3, we have
$$H _ { 1 } ( \Delta ( R / I ) ; H _ { \mathfrak { m } } ^ { d } ( R ) ) & \cong H _ { 1 } ( \Delta ( R / I ) ; \mathbb { Z } ) \otimes _ { \mathbb { Z } } H _ { \mathfrak { m } } ^ { d } ( R ) \\ H _ { 1 } ( \Delta ( R / I ) ; k ) & \cong H _ { 1 } ( \Delta ( R / I ) ; \mathbb { Z } ) \otimes _ { \mathbb { Z } } k \\ \text{he assumptions of Theorem 2.3. for each nonzero element $r\in R$.}$$
$$\cdots \to H _ { \mathfrak { m } } ^ { d } ( R ) \stackrel { r } { \to } H _ { \mathfrak { m } } ^ { d } ( R ) \to H _ { \mathfrak { m } } ^ { d } ( R / ( r ) ) = 0, \\ 0 / / \pi \cdots = 0 \, \underset { \mathfrak { m } \, } \, \infty \, \underset { \mathfrak { m } \, } \, \infty \, \left ( \, \mathfrak { m } \, / \pi \cdots \, \geq \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, / \mathfrak { m } \, / \mathfrak { m } \, \mathfrak { m } \, \mathfrak { m } \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \cdots \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \mathfrak { m } \, \infty \, \complement \, \mathfrak { m } \, \infty \, \complement \, \mathfrak { m } \, \infty \, \complement \, \mathfrak { m } \, \complement \, \mathfrak { m } \, \complement \, \mathfrak { m } \complement \, \complement \, \complement \, \complement \, \complement \, \complement \, \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \; \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \components \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \implement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement\complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complementation \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \import \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \omension \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \complement \org.$$
Under the assumptions of Theorem 2.3, for each nonzero element r ∈ R , the short exact sequence 0 → R r - → R → R/ r ( ) → 0 induces an exact sequence where H d m ( R/ r ( )) = 0 since dim( R/ r ( )) < d . This shows that rH d m ( R ) = H d m ( R ) for each nonzero element r ∈ R . Therefore,
/negationslash
- (1) when R doesn't have any integer torsion ( i.e. when R contains Q or doesn't contain a field), H 1 (∆( R/I ); Z ) ⊗ Z H d m ( R ) = 0 if and only if H 1 (∆( R/I ); Z ) contains a copy of Z ;
/negationslash
- (2) when the characteristic of R is p , H 1 (∆( R/I ); Z ) ⊗ Z H d m ( R ) = 0 if and only if H 1 (∆( R/I ); Z ) contains either a copy of Z or a copy of Z /p Z .
Consequently, when R contains a field, it is clear that H 1 (∆( R/I ); Z ) ⊗ Z H d m ( R ) = 0 if and only if H 1 (∆( R/I ); Z ) ⊗ Z k = 0, i.e.
On the other hand, it should be clear that
$$\widetilde { H } _ { 1 } ( \Delta ( R / I ) ; H _ { \mathfrak { m } } ^ { d } ( R ) ) = 0 \Leftrightarrow H _ { 1 } ( \Delta ( R / I ) ; k ) = 0. \\ \text{ther hand, it should be clear that}$$
always holds.
$$\widetilde { H } _ { 0 } ( \Delta ( R / I ) ; H _ { \mathfrak m } ^ { d } ( R ) ) = 0 \Leftrightarrow H _ { 0 } ( \Delta ( R / I ) ; k ) = 0 \\ \text{lds.}$$
To prove Theorem 1.2, we also need the following result due to Peskine-Szpiro.
Theorem 2.5 (Remarque on p. 386 in [PS73]) . Let R be a d -dimensional regular ring of characteristic p and let I be an ideal of R . Then
$$\text{cd} ( R, I ) \leq d - \text{depth} ( R / I ).$$
Proof of Theorem 1.2. By the Cohen Structure Theorem, there is a complete regular local ring ( S, n , k ) that maps onto R . Let I denote the kernel of the surjection S → R . Then it is straightforward to check that, if { p 1 , . . . , p n } is the set of minimal primes of R , then the preimages of p i , { ˜ p 1 , . . . , ˜ p n } , are the minimal primes of I in S . It is also clear that
/negationslash
/negationslash and consequently ∆( R ) = ∆( S/I ).
$$s \, \text{also clear that} \\ \sqrt { \mathfrak { p } _ { i _ { 0 } } + \dots + \mathfrak { p } _ { i _ { s } } } \neq \mathfrak { m } & \Leftrightarrow \sqrt { \widetilde { \mathfrak { p } } _ { i _ { 0 } } + \dots + \widetilde { \mathfrak { p } } _ { i _ { s } } } \neq \mathfrak { n } \\ \text{unelly } \Delta ( R ) = \Delta ( S / I ).$$
Since each ˜ p j is a prime ideal in S , the punctured spectrum of S/ ˜ p j is connected. Hence cd( S, ˜ p j ) ≤ dim( S ) -2 by [PS73, Corollaire 5.5]. Furthermore, since depth( S/I ) ≥ 3, we know that cd( S, I ) ≤ dim( S ) -3 according to Theorem 2.5. Therefore, our conclusion follows directly from Theorem 2.3 and Remark 2.4. /square
To prove Theorem 1.3, the following result of Varbaro is needed.
Theorem 2.6 (Theorem 3.5 in [Var13]) . Let I be an homogeneous ideal of R = k x , . . . , x [ 1 d ] where k is a field of characteristic 0. If depth( R/I ) ≥ 3 , then cd( R,I ) ≤ d -3 .
Proof of Theorem 1.3. Since R is standard graded, one can write R = k x , . . . , x [ 1 n ] /I for a homogeneous ideal I of S = k x , . . . , x [ 1 n ]. Let p 1 , . . . , p t be the minimal primes of I in S . Then cd( S, p j ) ≤ dim( S ) -2 by [PS73, Corollaire 5.5], [Ogu73, Corollary 2.11], and [HL90, Theorem 2.9]. And one has cd( S, I ) ≤ dim( S ) -3 according to Theorem 2.5 (characteristic p ) and Theorem 2.6 (characteristic 0). Therefore, our conclusion follows directly from Theorem 2.3 and Remark 2.4. /square
We conclude this section with the following corollary.
Corollary 2.7. Let ( R, m , k ) be an equicharacteristic complete regular local ring whose residue field is separably closed. Assume that an ideal I of R can be generated by d -3 elements, where d = dim( R ) . Then
Proof. Since I can be generated by d -3 elements, cd( R,I ) ≤ d -3. Same as in the proof of Theorem 1.2, we have cd( R, p ) ≤ d -2 for each minimal prime of I . Therefore, our corollary follows directly from Theorem 2.3. /square
$$\widetilde { H } _ { 0 } ( \Delta ( R / I ) ; k ) & = \widetilde { H } _ { 1 } ( \Delta ( R / I ) ; k ) = 0. \\ \tan \, \text{be generated by } d - 3 \, \text{ elements, } c d ( R, I ) \leq d \,.$$
Remark 2.8 . In [Fal80, Theorem 6], it is proved that, if an ideal I of a d -dimensional complete local domain ( R, m ) can be generated by d -2 elements, then Spec( R ) \{ m } is connected . One may consider Corollary 2.7 as an extension of Faltings' connectedness theorem. Of course it would be very interesting to know whether the requirement that R be regular can be removed or at least weakened.
## 3. An application to Stanley-Reisner rings
Let S be a polynomial ring k x , . . . , x [ 1 n ] where k is any field, and let K be a simplicial complex with vertices { x , . . . , x 1 n } . In this section we explore the properties of ∆( R ) when R is the Stanley-Reisner ring R = k [∆] = S/I K ( ). We then apply this to construct Example 3.4 that shows that Theorem 1.2 does not hold if the reduced homology groups ˜ H ∗ (∆( R ); -) are computed with coefficients in Z .
We first recall a definition and a well known result.
Definition 3.1. Let K be a simplicial complex and let K ,. . . , K 1 t be subcomplexes of K . The nerve of K ,. . . , K 1 t , denoted N ( K ,. . . , K 1 t ), is the simplicial complex with vertices K ,. . . , K 1 t and faces consisting of sets { K ,...,K i 1 i /lscript } such that K i 1 ∩ · · · ∩ K i /lscript = . ∅
/negationslash
The usefulness of this construction derives from the following theorem (cf. [Bj¨95, o Theorem 10.6].)
Theorem 3.2. With the notation above, assume further that K 1 ∪·· ·∪ K t = K . If every non-empty intersection K i 1 ∩··· ∩ K i /lscript is contractible, K and N ( K ,. . . , K 1 t ) are homotopy equivalent (and hence have same homology groups.)
We can now prove the main result of this section.
Theorem 3.3. Let S = k x , . . . , x [ 1 n ] , let K be a simplicial complex with vertex set V = { x , . . . , x 1 n } , and let I = I ( K ) be the Stanley-Reisner ideal corresponding to K . Then ∆( S/I ) is homotopy equivalent to K .
Proof. Let K ,. . . , K 1 t be the facets (i.e., maximal faces) of K and note that since every intersection of facets is a face, Theorem 3.2 implies that K and N ( K ,. . . , K 1 t ) are homotopy equivalent.
The minimal primes of I are generated by complements of facets of K (cf. [BH93, Theorem 5.1.4]) thus ∆( S/I ) is the simplicial complex whose vertices are the complements of facets of K and whose faces consist of
$$\mathfrak { w } \quad \mathfrak { w } \quad \mathfrak { w } _ { i _ { 1 } }, \dots, V \, \mathfrak { w } \, \mathfrak { w } _ { i _ { \ell } } \} \, | \, ( V \, \mathfrak { w } \, \mathfrak { w } _ { i _ { 1 } } ) \cup \cdots \cup ( V \, \mathfrak { w } \, \mathfrak { w } _ { i _ { \ell } } ) \neq V \}. \\ \mathfrak { w } \quad \mathfrak { w } \quad \mathfrak { w } \quad \mathfrak { w } \quad \mathfrak { w } \quad \mathfrak { w } \quad \mathfrak { w } \quad \mathfrak { w } \quad \mathfrak { w } \quad \mathfrak { w } \quad \mathfrak { w }.$$
/negationslash
We conclude the proof by exhibiting the isomorphism of simplicial complexes Φ : N ( K ,. . . , K 1 t ) → K given by Φ( K i ) = V \ K i . /square
/negationslash
Example 3.4. Let K be Reisner's six-point triangulation of the projective plane (cf. [Rei76, Remark 3]) and let R = k K ( ) be the Stanley-Reisner ring associated to K . [Rei76, Theorem 1] shows that R is Cohen-Macaulay if and only if the characteristic of k is not 2 thus its depth of R is 3 when the characteristic of k is not 2. On the other hand the previous theorem implies that ˜ H 1 (∆( R ); Z ) = ˜ H 1 ( K ; Z ) = 0. We conclude that the analogue of Theorem 1.3 with homology computed with integer coefficients does not hold.
We can now extend Theorem 1.3 for Stanley-Reisner rings as follows.
Corollary 3.5. Let S = k x , . . . , x [ 1 n ] , let K be a simplicial complex with vertex set V = { x , . . . , x 1 n } , and let I = I ( K ) be the Stanley-Reisner ideal corresponding to K . If depth S/I ≥ d then ˜ H j (∆( S/I , k ) ) = 0 for all 0 ≤ j ≤ d -2 .
Proof. The Auslander-Buchsbaum theorem implies that the condition depth S/I ≥ d is equivalent to pdim S S/I ≤ n -d . This translates, using Hochster's formula (cf. [Hoc77], [MS05, Corollary 5.12]), to the vanishing of ˜ H # W n j -+ -1 ( K W , k ) for all 0 ≤ j ≤ d -1 and all W ⊆ V , where K W denotes the restriction of K to the subset of its vertices W . In particular, this holds for W = V , giving ˜ H j -1 ( K,k ) = 0 for all 0 ≤ j ≤ d -1. Theorem 3.3 now implies ˜ H j -1 (∆( S/I , k ) ) = 0 for all 0 ≤ j ≤ d -1. /square
We can also reinterpret Corollary 2.7 in the case of Stanley-Reisner ideals as follows. Let K be a simplicial complex with vertex set V = { x , . . . , x 1 n } , let
R = k x , . . . , x [ 1 n ] and let I = I ( K ) be the Stanley-Reisner ideal corresponding to K . The ideal I is generated by n -t elements if and only if the Alexander dual K ∗ has at most n -t facets. Also ˜ H j (∆( R/I , k ) ) = ˜ H K,k j ( ) = ˜ H n - -3 j ( K ,k ∗ ) where the first equality follows from Theorem 3.3 and the second from Alexander Duality (cf. [MS05, Theorem 5.6]). Now Corollary 2.7 implies the following: if a simplicial complex ∆ with n vertices has at most n -3 facets, then ˜ H n -3 (∆ , k ) = ˜ H n -4 (∆ , k ) = 0. This also leads to the following natural generalization of Corollary 2.7.
Proposition 3.6. For a simplicial complex ∆ with n vertices and at most µ facets, ˜ H i (∆ , k ) = 0 for all i ≥ µ -1 .
Corollary 3.7. Let K be a simplicial complex with vertex set V = { x , . . . , x 1 n } , let R = k x , . . . , x [ 1 n ] and let I = I ( K ) be the Stanley-Reisner ideal corresponding to K . If I is generated by n -t elements then ˜ H i (∆( R/I ); k ) = 0 for all 0 ≤ i ≤ t -2 .
Proof. Let K 1 ∪ · · · ∪ K s be the distinct facets of K and let N = N ( K ,. . . , K 1 s ). Note that dim N ≤ s -1 and that N is ( s -1)-dimensional precisely when it is a simplex. An application of Theorem 3.2 gives ˜ H i (∆ , k ) = ˜ H i ( N , k ). and this vanishes for all i ≥ s -1, and since s ≤ µ , ˜ H i (∆ , k ) = 0 for all i ≥ µ -1. /square
Proof. Using the previous discussion we reduce the problem to showing that ˜ H n - -3 j ( K ,k ∗ ) vanishes for 0 ≤ j ≤ t -2 and this follows from the application of Proposition 3.6 to K ∗ . /square
## Acknowledgements
We thank Vic Reiner for pointing out to us Theorem 3.2 and its relevance to Example 3.4. We also thank John Shareshian for pointing out Proposition 3.6 to us.
## References
[ ` MGLZA03] A J. Alvarez Montaner, R. Garc´ ıa L´pez, and S. Zarzuela Armengou ` o : Local cohomology, arrangements of subspaces and monomial ideals , Adv. Math. 174 (2003), no. 1, 35-56. 1959890 (2004a:13012)
[Bj¨95] o
A. Bj¨rner o : Topological methods , Handbook of combinatorics, Vol. 1, 2, Elsevier, Amsterdam, 1995, pp. 1819-1872.
[BS98]
M. P. Brodmann and R. Y. Sharp : Local cohomology: an algebraic introduction with geometric applications , Cambridge Studies in Advanced Mathematics, vol. 60, Cambridge University Press, Cambridge, 1998. MR1613627 (99h:13020)
[BH93]
W. Bruns and J. Herzog : Cohen-Macaulay rings , Cambridge Studies in Ad- vanced Mathematics, vol. 39, Cambridge University Press, Cambridge, 1993. MR1251956 (95h:13020)
[Fal80]
G. Faltings : Some theorems about formal functions , Publ. Res. Inst. Math. Sci. 16 (1980), no. 3, 721-737. 602466 (82h:14004)
[Har62]
R. Hartshorne : Complete intersections and connectedness , Amer. J. Math. 84 (1962), 497-508. 0142547 (26 #116)
[Har68]
R. Hartshorne : Cohomological dimension of algebraic varieties , Ann. of Math. (2) 88 (1968), 403-450. 0232780 (38 #1103)
[Hat02]
A. Hatcher : Algebraic topology , Cambridge University Press, Cambridge, 2002. 1867354 (2002k:55001)
[Hoc77]
M. Hochster : Cohen-Macaulay rings, combinatorics, and simplicial complexes , Ring theory, II (Proc. Second Conf., Univ. Oklahoma, Norman, Okla., 1975), Dekker, New York, 1977, pp. 171-223. Lecture Notes in Pure and Appl. Math., Vol. 26. 0441987 (56 #376)
| [HL90] | C. Huneke and G. Lyubeznik : On the vanishing of local cohomology modules , Invent. Math. 102 (1990), no. 1, 73-93. 1069240 (91i:13020) |
|----------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [Lyu07] | G. Lyubeznik : On some local cohomology modules , Adv. Math. 213 (2007), no. 2, 621-643. 2332604 (2008e:13021) |
| [MS05] | E. Miller and B. Sturmfels : Combinatorial commutative algebra , Gradu- ate Texts in Mathematics, vol. 227, Springer-Verlag, New York, 2005. 2110098 (2006d:13001) |
| [Ogu73] | A. Ogus : Local cohomological dimension of algebraic varieties , Ann. of Math. (2) 98 (1973), 327-365. 0506248 (58 #22059) |
| [PS73] | C. Peskine and L. Szpiro : Dimension projective finie et cohomologie locale. Applications ` a la d´monstration e de conjectures de M. Auslander, H. Bass et A. Grothendieck , Inst. Hautes ´ Etudes Sci. Publ. Math. (1973), no. 42, 47-119. MR0374130 (51 #10330) |
| [Rei76] | G. A. Reisner : Cohen-Macaulay quotients of polynomial rings , Advances in Math. 21 (1976), no. 1, 30-49. 0407036 (53 #10819) |
| [Var13] | M. Varbaro : Cohomological and projective dimensions , Compos. Math. 149 (2013), no. 7, 1203-1210. 3078644 |
(Katzman) Department of Pure Mathematics, University of Sheffield, Hicks Building, Sheffield S3 7RH, United Kingdom
E-mail address : [email protected]
(Lyubeznik) School of Mathematics, University of Minnesota, 207 Church Street, Minneapolis, MN 55455, USA
E-mail address : [email protected]
(Zhang) Department of Mathematics, University of Nebraska, Lincoln, NE 68588, USA
E-mail address : [email protected]
| null |
[
"Mordechai Katzman",
"Gennady Lyubeznik",
"Wenliang Zhang"
] |
2014-08-05T03:49:17+00:00
|
2014-10-08T04:47:01+00:00
|
[
"math.AC",
"13D45, 13F55"
] |
An extension of a theorem of Hartshorne
|
We extend a classical theorem of Hartshorne concerning the connectedness of
the punctured spectrum of a local ring by analyzing the homology groups of a
simplicial complex associated with the minimal primes of a local ring.
|
1408.0860v1
|
## Gravitational radiation from compact binaries in scalar-tensor gravity
## R N Lang
Department of Physics, University of Florida, Gainesville, Florida 32611, USA
E-mail:
[email protected]
Abstract. General relativity (GR) has been extensively tested in the solar system and in binary pulsars, but never in the strong-field, dynamical regime. Soon, gravitational-wave (GW) detectors like Advanced LIGO and eLISA will be able to probe this regime by measuring GWs from inspiraling and merging compact binaries. One particularly interesting alternative to GR is scalar-tensor gravity. We present progress in the calculation of second post-Newtonian (2PN) gravitational waveforms for inspiraling compact binaries in a general class of scalartensor theories. The waveforms are constructed using a standard GR method known as 'direct integration of the relaxed Einstein equations,' appropriately adapted to the scalar-tensor case. We find that differences from general relativity can be characterized by a reasonably small number of parameters. Among the differences are new hereditary terms which depend on the past history of the source. In one special case, binary black hole systems, we find that the waveform is indistinguishable from that of general relativity. In another, mixed black holeneutron star systems, all differences from GR can be characterized by only a single parameter.
## 1. Introduction
2015 marks the centennial of Einstein's theory of general relativity (GR). In the last 100 years, GR has withstood every test thrown at it. Nevertheless, we still expect GR to fail at some energy scale, since we have no consistent quantum theory of gravity. Modifications to gravity may also explain the phenomena of dark matter and dark energy. It is therefore important that we keep testing Einstein's theory as it moves into its second century. Notably, general relativity has not yet been tested in the strong-field, dynamical regime that exists in coalescing compact binaries made of neutron stars and/or black holes. These systems are the primary targets of current and future gravitational-wave (GW) detectors, such as Advanced LIGO and Advanced Virgo on the ground and eLISA in space. By carefully measuring the gravitational waveform of a compact binary with these detectors, and comparing it to templates from different theories, we can set limits on deviations from general relativity or possibly discover new physics.
A particularly important alternative to GR is the collection of scalar-tensor theories of gravity [1]. These theories have a long history, dating back over 50 years, and while they are perhaps an overly simple modification to Einstein's theory (with just the addition of some scalar fields to GR's metric tensor g µν ), they remain well-motivated. For instance, many socalled f ( R ) theories, which may explain the acceleration of the universe without dark energy, can be expressed in the form of a scalar-tensor theory [2]. Scalar-tensor theories are also potential low-energy limits of string theory [1].
Most analyses of gravitational waveforms in alternative theories of gravity include only the lowest-order effects in the waveform model. However, real templates used by gravitational-wave detectors must be calculated to high post-Newtonian (PN) order, an expansion in powers of the source velocity v/c . In this paper, we report on progress in calculating gravitational waveforms in a subclass of scalar-tensor theories to second post-Newtonian (2PN) order. The details of this work can be found in papers by Mirshekari and Will [3] (hereafter MW) and Lang [4] (hereafter L14). Throughout the paper, we use units in which c = 1. Greek indices run from 0-3, while Latin indices run from 1-3. Repeated indices are summed over.
We are specifically interested in theories with a single, massless scalar field φ . In the simplest and most familiar scalar-tensor theory, known as Brans-Dicke theory, the coupling function ω φ ( ) is a constant; here, however, we allow it to be any function with a Taylor expansion. We work in the 'Jordan frame,' in which the scalar field does not couple directly to the matter (though, of course, φ will indirectly impact the matter through its role in determining the metric). With all of these assumptions, the scalar-tensor field equations are given by
$$G _ { \mu \nu } = \frac { 8 \pi } { \phi } T _ { \mu \nu } + \frac { \omega ( \phi ) } { \phi ^ { 2 } } \left ( \phi _ {, \mu } \phi _ {, \nu } - \frac { 1 } { 2 } g _ { \mu \nu } \phi _ {, \lambda } \phi ^ {, \lambda } \right ) + \frac { 1 } { \phi } ( \phi _ { ; \mu \nu } - g _ { \mu \nu } \Box _ { g } \phi ) \,, \\ 1 / \Delta T _ { \mu } \quad d \omega \quad \dots \, \cdots \,.$$
$$\Box _ { g } \phi = \frac { 1 } { 3 + 2 \omega ( \phi ) } \left ( 8 \pi T - 1 6 \pi \phi \frac { \partial T } { \partial \phi } - \frac { d \omega } { d \phi } \phi _ {, \lambda } \phi ^ { \lambda } \right ) ^ { \cdot },$$
where G µν is the Einstein tensor, T µν is the stress energy of matter and nongravitational fields, and T ≡ g αβ T αβ is its trace. We use commas to denote ordinary derivatives. Semicolons denote covariant derivatives (taken using g µν in the usual way), and ✷ g ≡ g αβ ∂ α ∂ β .
Since scalar-tensor theories do not obey the strong equivalence principle, the motion and gravitational-wave emission of a binary depend on the internal composition of its constituent bodies. To handle this complication, we have adopted the approach of Eardley [5]. We treat the matter stress-energy tensor as a sum of delta functions located at the position of each compact object. However, instead of assigning each body a constant mass, we let the mass be a function of the scalar field, M A = M A ( φ ). This gives the matter an indirect dependence on φ [hence the term ∂T/∂φ in (1b)], even though we still work in the Jordan frame. In the final results, this dependence will appear as the 'sensitivity' of the mass to variations in the scalar field,
$$s _ { A } & \equiv \left ( \frac { d \ln M _ { A } ( \phi ) } { d \ln \phi } \right ) _ { 0 }, \\ \intertext { k i s. \dots \dots } \, \stackrel { \not \, T \in \dots \dots } { \sim } \, \stackrel { \not \, T \in \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \end{array}$$
as well as derivatives of this quantity. (The subscript 0 means that the derivative should be evaluated using the asymptotic value of the scalar field, φ 0 .) In the weak-field limit, the sensitivity is proportional to the Newtonian self-gravitational energy per unit mass of the body. For neutron stars, the sensitivity depends on the mass and equation of state of the star, with typical values 0.1-0.3 [6]. For black holes, s = 0 5, and all derivatives vanish. .
## 2. Solving the field equations
To solve the field equations, we use a method known as direct integration of the relaxed Einstein equations (DIRE). While DIRE was originally developed for GR by Will, Wiseman, and Pati [7-10], it can be easily adapted to scalar-tensor theory if the Einstein equations are replaced by the field equations (1a)-(1b).
The first step is to introduce new variables which will help us place the field equations in a 'relaxed' form. We assume that far away from the sources, the metric reduces to the Minkowski metric, η µν , and the scalar field tends to a constant φ 0 . We introduce a rescaled scalar field,
$$\varphi \equiv \frac { \phi } { \phi _ { 0 } } \,.$$
Next, we define a conformally transformed metric, ˜ g µν ≡ ϕg µν , with determinant ˜, g and a 'gravitational field,'
$$\tilde { h } ^ { \mu \nu } \equiv \eta ^ { \mu \nu } - \sqrt { - \tilde { g } } \tilde { g } ^ { \mu \nu } \,. & & ( 4 ) \\ \text{late this field from the gravitational field defined in general relativity,} \\ \text{$1$} \, \colon = \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \, \text{$1$} \}.$$
We use a tilde to differentiate this field from the gravitational field defined in general relativity, h µν , which has the same definition but with the normal metric g µν in place of the conformally transformed metric ˜ g µν . [See, for instance, (2.2) in [8].] We impose the Lorenz gauge condition
$$\tilde { h } ^ { \mu \nu } _ { \ \, \nu } = 0 \,.$$
Then the field equation (1a) reduces to
$$\Box _ { \eta } \tilde { h } ^ { \mu \nu } = - 1 6 \pi \tau ^ { \mu \nu } \,, \quad \quad \quad$$
where ✷ η ≡ η αβ ∂ α ∂ β is the flat-spacetime wave operator and the source is
$$\tau ^ { \mu \nu } \equiv ( - g ) \frac { \varphi } { \phi _ { 0 } } T ^ { \mu \nu } + \frac { 1 } {$$
Here g is the determinant of g µν . The first term of (7) represents the compact piece of the source. The quantity Λ µν represents the gravitational-field contribution to the stress energy. It is constructed so that it has the same functional form as in GR (except with components of ˜ h µν taking the place of components of h µν ). The quantity Λ µν s is an entirely new contribution to the stress energy which depends on ϕ . Both Λ µν and Λ µν s are given explicitly in L14, Eqs. (2.9)-(2.11). The scalar field equation (1b) can also be written as a flat-spacetime wave equation,
$$\Box _ { \eta } \varphi = - 8 \pi \tau _ { s } \,,$$
with source
$$\tau _ { s } \equiv - \frac { 1 } { 3 + 2 \omega } \sqrt { - g } \frac { \varphi } { \phi _ { 0 } } \$$
The wave equations (6) and (8) can be solved formally in all spacetime by using a retarded Green's function,
$$\tilde { h } ^ { \mu \nu } ( t, \mathbf x ) & = 4 \int \frac { \tau ^ { \mu \nu } ( t ^ { \prime }, \mathbf x ^ { \prime } ) \delta ( t ^ { \prime } - t + | \mathbf x - \mathbf x ^ { \prime } | ) } { | \mathbf x - \mathbf x ^ { \prime } | } d ^ { 4 } x ^ { \prime } \,, & ( 1 0 \text{a} ) \\ \nu. & \quad \sim \int \tau _ { s } ( t ^ { \prime }, \mathbf x ^ { \prime } ) \delta ( t ^ { \prime } - t + | \mathbf x - \mathbf x ^ { \prime } | ) _ { \mathbf d } \, & \nu \times \,.$$
$$\varphi ( t, \mathbf x ) = 2 \overset { \prime } { \int } \frac { \tau _ { s } ( t ^ { \prime }, \mathbf x ^ { \prime } ) \delta ( t ^ { \prime } - t + | \mathbf x - \mathbf x ^ { \prime } | ) } { | \mathbf x - \mathbf x ^ { \prime } | } d ^ { 4 } x ^ { \prime } \,. \quad \quad ( 1 0 b ) \\.. \ \cdot \ \cdot \ \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \,.$$
The delta function in both these equations restricts the integration to being over the past flatspacetime null cone emanating from the field point ( t, x ). To obtain explicit solutions, we divide the spacetime into two regions. The near zone is defined as the area with | x | = R < R , where R is the characteristic wavelength of gravitational radiation from the system. Everything outside the near zone ( R > R ) is the radiation zone .
We carry out the integrals (10a) and (10b) differently depending on the location of the field point x of interest (near zone or radiation zone). Once the field point is chosen, we then evaluate each integral in two separate pieces: one integral over source points x ′ in the near zone and another over source points in the radiation zone. This makes for a total of four different integration techniques, described in detail in [8] and L14. It has been shown (in GR) that all terms dependent on the boundary radius R in one-half of the integral are exactly canceled by pieces in the other half of the integral, leaving the final answer, as expected, independent of this arbitrary parameter [8, 9]. In our work, we simply assume this property and ignore any terms which depend on R .
## 3. Equations of motion
The first step in finding the gravitational waveform is to calculate the equations of motion for the compact binary. To do so, we must evaluate ˜ h µν and ϕ at field points in the near zone, where they influence the bodies. As stated above, this usually involves an integration over source points in both zones. It turns out, however, that only near-zone points contribute at the order to which we work.
The calculation of the near-zone fields is an iterative one. We begin by finding the lowest order pieces of ˜ h 00 and ϕ , which depend only on the compact piece of the source. These fields are then plugged back into (7) and (9) to generate the next-highest-order source. The process continues, alternating between calculating fields and plugging them back into the source expressions, until the fields are known to the desired post-Newtonian order. The details and results of this procedure can be found in MW.
Once we have the fields, we can use (3) and (4) to solve for the metric g µν . From the metric, we calculate Christoffel symbols just as in GR. The final equations of motion are a bit trickier, however, due to the violation of the strong equivalence principle. Using the Bianchi identity, we can write the equations of motion as
$$T ^ { \mu \nu } _ { \ \ ; \nu } = \frac { \partial T } { \partial \phi } \phi ^ {, \mu } \,.$$
A more useful form, involving the Christoffel symbols, is given in MW, Eq. (2.5). We plug in for the symbols, integrate over the extent of each body, and then convert to relative coordinates. The final 2.5PN equations of motion can be written as
$$a ^ { i } & = - \frac { G \alpha m } { r ^ { 2 } } \hat { n } ^ { i } + \frac { G \alpha m } { r ^ { 2 } } ( A _ { \text{PN} } \hat { n } ^ { i } + B _ { \text{PN} } \dot { r } v ^ { i } ) + \frac { 8 } { 5 } \eta \frac { ( G \alpha m ) ^ { 2 } } { r ^ { 3 } } ( A _ { 1. 5 \text{PN} } \dot { r } \hat { n } ^ { i } - B _ { 1. 5 \text{PN} } v ^ { i } ) \\ & \quad + \frac { G \alpha m } { r ^ { 2 } } ( A _ { 2 \text{PN} } \hat { n } ^ { i } + B _ { 2 \text{PN} } \dot { r } v ^ { i } ) \,. \\ \dashrightarrow \quad \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \,.$$
Here ˆ n is a unit vector pointing from body 2 to body 1, r is the orbital separation, v is the relative velocity, a is the relative acceleration, m ≡ m 1 + m 2 , and η ≡ m m /m 1 2 2 . The A and B coefficients contain all the details; they are given in MW. We have also defined
$$G \equiv \frac { 1 } { \phi _ { 0 } } \frac { 4 + 2 \omega _ { 0 } } { 3 + 2 \omega _ { 0 } } \,,$$
where ω 0 ≡ ω φ ( 0 ). This definition is chosen so that g 00 = -1 + 2 GU matches the expression in GR for a perfect fluid (i.e., with all sensitivities equal to zero). However, with this definition, the coupling in the equations of motion is not simply G , but rather Gα , with α ≡ 1 -ζ + (1 ζ -2 s 1 )(1 -2 s 2 ) and ζ ≡ 1 (4 + 2 / ω 0 ).
The even orders in the equations of motion (0PN, 1PN, and 2PN) correspond to conservative parts of the motion. These equations can be derived from a Lagrangian and admit conserved energy and momentum quantities; for details, see MW. The odd terms (1.5PN and 2.5PN) represent radiation-reaction terms which dissipate energy and cause the binary to coalesce. In
GR, radiation reaction begins at 2.5PN order with quadrupole radiation; the 1.5PN component in our results is due to the well-known phenomenon of dipole radiation.
## 4. Calculating the gravitational waves
Taking the limit of large R , the near-zone contribution to the tensor GWs, ˜ h ij N , is given by
To date, we have calculated the tensor gravitational waves ˜ h ij to 2PN order. The calculation requires integrating (10a) for field points in the radiation zone. To simplify matters, we need only be concerned with the results in a subset of the radiation zone, the far-away zone , where R >> R . Unlike for the equations of motion, we must consider source points in both the near and radiation zones.
$$\tilde { h } ^ { i j } _ { \mathcal { N } } ( t, \mathbf x ) = \frac { 2 } { R } \frac { d ^ { 2 } } { d t ^ { 2 } } \sum _ { m = 0 } ^ { \infty } \hat { N } ^ { k _ { 1 } } \cdots \hat { N } ^ { k _ { m } } I _ { \mathrm E W } ^ { i j k _ { 1 } \cdots k _ { m } } ( \tau ) \,, \\ \mathbf v / R \, \text{is the direction from the source to the dataator and } \tau = t - R \, \text{is the natural} d.$$
where ˆ N ≡ x /R is the direction from the source to the detector and τ ≡ t -R is the retarded time. The quantities I M +2 EW are known as 'Epstein-Wagoner' (EW) moments [11]. They make use of the near-zone fields calculated in MW and involve a variety of different integrals. It is useful to consider the post-Newtonian orders of the various EW moments. The lowest order moment, I ij EW , has a quadrupole character. As in general relativity, we define its lowest order piece to be 0PN order. It also has higher-order contributions, specifically at 1PN, 1.5PN, and 2PN orders. When substituted into (14), it generates GWs at all of these orders.
The three-index (octupole) moment I ijk EW begins at 0.5PN order but also comprises pieces at 1.5PN and 2PN orders. When plugged into (14), it generates GWs at all of these orders. Each successive EW moment begins one-half post-Newtonian order higher than the previous moment. The final moment we need is the six-index moment I ijklmn EW , which consists only of a 2PN contribution (and generates 2PN gravitational waves).
Once they have been calculated to the desired order for an arbitrary number of bodies, the EWmoments are simplified to the two-body case of interest and rewritten in relative coordinates. Finally, they are substituted back into (14), which involves taking time derivatives. Whenever an acceleration appears in one of these derivatives, we substitute in the equations of motion (12).
The tensor waves also include contributions from radiation-zone source points. The first step to finding them is to derive an expression for the source τ ij in the radiation zone. Since there are no compact sources at R > R , τ ij is composed purely of field terms. The details of this derivation, and of the subsequent calculation of the GWs, are found in L14.
It turns out that the radiation-zone integrals produce GWs at 1.5PN and 2PN orders. For the sake of clarity, we separate GWs generated by Λ ij from those generated by Λ ij s . The former produces terms which differ from the GR versions only by a scaling factor. For instance, at 1.5PN order, we find the contribution
$$\tilde { h } ^ { i j } & = \frac { 4 G ( 1 - \zeta ) m } { R } \left [ \frac { 1 1 } { 1 2 } \dddot { \mathcal { I } } ^ { i j } + \int _ { 0 } ^ { \infty } d s \ \overset { ( 4 ) } { \mathcal { I } } ^ { i j } ( \tau - s ) \ln \frac { s } { 2 R + s } \right ] \. \quad \ \quad ( 1 5 ) \\.. \quad \. \quad \. \quad \. \quad \. \quad \. \quad \. \quad \. \quad \. \. \. \. \. \. \. \. \. \. \. \. \. \. \.. \.. \..$$
Similar terms exist at 2PN order. The moment I ij is defined in L14, Eq. (6.3a). The first term in (15) is instantaneous , depending only on the state of the binary at the current (retarded) time. All near-zone contributions to the waveform are also instantaneous. The second term in (15), however, is hereditary , depending on the entire past history of the source. Specifically, the presence of the logarithm means that this is a 'tail' term, arising physically from the scattering of the waves off the background spacetime. This phenomenon is present in GR at this order; scalar-tensor theory only modifies the coefficient.
The field terms Λ ij s produce new pieces of the gravitational waveform not seen in GR. For instance, at 1.5PN order, we have
$$\tilde { h } ^ { i j } = \frac { ^ { ^ { \prime } } } { R } \left ( - \frac { 1 } { 1 2 } \ddot { \mathcal { I } } ^ { i j } _ { s } \right ) + \frac { 4 } { R } \frac { 1 - \zeta } { \zeta } \left ( \frac { 1 } { 6 } \int _ { - \infty } ^ { \tau } \ddot { \mathcal { I } } ^ { i } _ { s } ( \tau ^ { \prime } ) \ddot { \mathcal { I } } ^ { j } _ { s } ( \tau ^ { \prime } ) d \tau ^ { \prime } - \frac { 1 } { 6 } \dot { \mathcal { I } } ^ { i } _ { s } ( \ddot { \mathcal { I } } ^ { j } _ { s } ) - \frac { 1 } { 1 8 } \mathcal { I } ^ { i } _ { s } \dot { \mathcal { I } } ^ { i } _ { s } \right ) \,, \\ \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \ dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots\dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \delta \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \Dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots$$
where m s , I i s , and I ij s are defined in L14, Eqs. (6.10), (6.11a), and (6.11b). Similar terms exist at 2PN order. Eq. (16) contains a hereditary contribution which is not a tail. Terms like these-hereditary but without logarithms-do appear in GR, but not until 2.5PN order. Their appearance beginning at 1.5PN order in scalar-tensor theory results from the nonvanishing of the scalar dipole moment, I i s , the same effect which causes radiation reaction to appear at 1.5PN order in the equations of motion. The 1.5PN hereditary integral can be described as a 'gravitational-wave memory' term. It contains a zero-frequency, or DC, component, in addition to the usual oscillatory terms. (The new 2PN hereditary integral, not shown, does not.)
## 5. Discussion
To find the final tensor waveform, we add the contributions from the near and radiation zones. It can be written as a post-Newtonian expansion,
$$\dots \quad \tilde { h } ^ { i j } = \frac { 2 G ( 1 - \zeta ) \mu } { R } [ \tilde { Q } ^ { i j } + P ^ { 1 / 2 } Q ^ { i j } + P Q ^ { i j } + P ^ { 3 / 2 } Q ^ { i j } + P ^ { 2 } Q ^ { i j } + O ( \epsilon ^ { 5 / 2 } ) ] _ { \text{TT} } \,, \quad ( 1 7 )$$
where the superscripts on P denote the PN order of each term. 'TT' denotes the transversetraceless projection; to save time, we throw away pieces at any stage of the calculation which will not survive this final projection. Each term in the expansion can be found in L14; note that there the 1.5PN and 2PN terms have been divided into the contributions from the near zone and the contributions from the radiation zone [e.g., (15) and (16)].
The expressions are significantly more complicated than the GR equivalents, Eq. (6.11) of [8]. Nevertheless, all deviations can be characterized by a small number of parameters, including G ζ , , α , and several others (¯, γ ¯ , β i ¯ , δ i ¯ , χ i S + , and S -), given in or near MW Table I. All of these parameters, as expected, depend on the fundamental parameters of the theory (Taylor coefficients of the coupling function ω φ ( ) and the asymptotic scalar field φ 0 ) and/or the sensitivities and sensitivity derivatives of the sources. The same set of parameters characterizes the deviations of the equations of motion from the GR version.
In general, the tensor waves in scalar-tensor theory have the same general functional form as the GR waves, with only the addition of these parameters to the coefficients. In some cases, however, the existence of a nonvanishing scalar dipole moment generates entirely new terms. One example is the 1.5PN contribution from the quadrupole moment; in GR, this moment only contributes at 0PN, 1PN, and 2PN orders. Others include the new hereditary integrals.
For binary black holes, the expressions simplify considerably. The 2.5PN equations of motion and 2PN tensor waves turn out to have the exact same forms as in general relativity if the masses are rescaled by a factor 1 / (1 -ζ ). Since the masses of the bodies are defined by their Keplerian motion, this rescaling is unmeasurable. Therefore, to these orders, the motion and tensor GW emission of two black holes in scalar-tensor theory are indistinguishable from the motion and emission in GR. This result is not surprising. Hawking showed that stationary, asymptotically flat black holes in vacuum are identical in both theories [12], leading to conjectures that the same might be true for black hole binaries. The result is supported by other analytic studies (to all PN orders, but only leading order in the mass ratio) [13] and numerical studies [14]. Of course, it depends strongly on the assumptions we made (no potential for the scalar field, a time-independent φ 0 ). One unfortunate consequence of this result is that eLISA will be unlikely to provide information about these particular scalar-tensor theories; ground-based detectors, which will detect many inspirals involving neutron stars, are more suited to the task.
For a system containing one neutron star (say, body 1) and one black hole, the expressions also simplify somewhat. In this case, the motion and waves are identical to those in GR through 1PN order (after a mass rescaling). At 1.5PN order, deviations start to occur. However, the deviations are always parametrized by the single parameter Q = ζ (1 -ζ ) -1 (1 -2 s 1 ) 2 . Because Q contains no information on the derivatives of the coupling function ω φ ( ), we cannot, at this order, formally distinguish the waveform produced in the Brans-Dicke theory from that produced in a general scalar-tensor theory of the type we consider. One could imagine doing so by measuring masses and sensitivities of multiple neutron stars.
Work is underway on the calculation of the scalar gravitational waves. Some of the low-order pieces can be found in [15], Eqs. (101) and (102). The process is similar to that for the tensor waves; however, there are some unique complications. For instance, the near-zone calculation also relies on the computation of moments (like the EW moments). In the scalar case, however, these include monopole and dipole moments. Since we defined the quadrupole moment (and the waves it generated) as 0PN order, the monopole and dipole moments turn out to begin at -1PN and -0.5PN orders, respectively. These moments must then be calculated to relative 3PN and 2.5PN orders in order to find the complete 2PN gravitational waves. The number of integrals required to find the 3PN-accurate monopole moment is several times that needed for any of the tensor EW moments; furthermore, several of the integrals are extremely lengthy to compute. Things become even more difficult when considering the energy loss from the system, ultimately needed to determine the gravitational-wave phasing. While the 2PN energy loss from tensor waves can be found by knowing up to the 2PN form of the waves, the 2PN energy loss from scalar waves requires knowledge of certain 2.5PN pieces of the waveform. It also requires the 3PN equations of motion, one-half PN order beyond that calculated in MW. There is a still a great deal of work to do in calculating even the simplest scalar-tensor gravitational waveforms.
## Acknowledgments
We thank the organizers of the 10th International LISA Symposium for an excellent conference. We are grateful to Clifford Will and Saeed Mirshekari for their hard work on the first phase of this calculation and their useful insights on the rest. This work was supported in part by the National Science Foundation, Grants No. PHY-0965133, No. PHY-1260995, and No. PHY1306069.
## References
- [1] Fujii Y and Maeda K I 2003 The Scalar-Tensor Theory of Gravitation (Cambridge, UK: Cambridge University Press)
- [2] de Felice A and Tsujikawa S 2010 Living Rev. Relativity 13 3
- [3] Mirshekari S and Will C M 2013 Phys. Rev. D 87 084070
- [4] Lang R N 2014 Phys. Rev. D 89 084014
- [5] Eardley D M 1975 Astrophys. J. 196 L59-62
- [6] Will C M and Zaglauer H W 1989 Astrophys. J. 346 366-77
- [7] Wiseman A G 1992 Phys. Rev. D 46 1517-39
- [8] Will C M and Wiseman A G 1996 Phys. Rev. D 54 4813-48
- [9] Pati M E and Will C M 2000 Phys. Rev. D 62 124015
- [10] Pati M E and Will C M 2002 Phys. Rev. D 65 104008
- [11] Epstein R and Wagoner R V 1975 Astrophys. J. 197 717-23
- [12] Hawking S W 1972 Commun. Math. Phys. 25 167-71
- [13] Yunes N, Pani P and Cardoso V 2012 Phys. Rev. D 85 102003
- [14] Healy J, Bode T, Haas R, Pazos E, Laguna P, Shoemaker D M and Yunes N 2012 Classical Quantum Gravity 29 232002
- [15] Will C M 2014 Living Rev. Relativity 17 4
|
10.1088/1742-6596/610/1/012045
|
[
"Ryan N. Lang"
] |
2014-08-05T04:21:55+00:00
|
2014-08-05T04:21:55+00:00
|
[
"gr-qc"
] |
Gravitational radiation from compact binaries in scalar-tensor gravity
|
General relativity (GR) has been extensively tested in the solar system and
in binary pulsars, but never in the strong-field, dynamical regime. Soon,
gravitational-wave (GW) detectors like Advanced LIGO and eLISA will be able to
probe this regime by measuring GWs from inspiraling and merging compact
binaries. One particularly interesting alternative to GR is scalar-tensor
gravity. We present progress in the calculation of second post-Newtonian (2PN)
gravitational waveforms for inspiraling compact binaries in a general class of
scalar-tensor theories. The waveforms are constructed using a standard GR
method known as "direct integration of the relaxed Einstein equations,"
appropriately adapted to the scalar-tensor case. We find that differences from
general relativity can be characterized by a reasonably small number of
parameters. Among the differences are new hereditary terms which depend on the
past history of the source. In one special case, binary black hole systems, we
find that the waveform is indistinguishable from that of general relativity. In
another, mixed black hole-neutron star systems, all differences from GR can be
characterized by only a single parameter.
|
1408.0861v1
|
## Experimental research on the THGEM-based thermal neutron detector
YANG Lei( 杨雷 ) 1) Zhou Jian-Rong( 周健荣 ) 2;3,*) Sun Zhi-Jia( 孙志嘉 ) 2;3) Zhang Ying( 张莹 ) 4) Huang Chao-Qiang( 黄朝强 ) 4) Sun Guang-Ai( 孙光爱 ) 4) Wang Yan-Feng( 王艳凤 ) 2;3) Yang Gui-An( 杨桂安 ) 2;3) Xu Hong ( 许虹 ) 2;3) Xie Yu-Guang( 谢宇广 ) 2;3) Chen Yuan-Bo( 陈元柏 ) 2;3)
1 Dongguan University of Technology, Dongguan, 523808, China
2 Institute of High Energy Physics, Chinese Academy of Sciences, Beijing 100049, China;
3 State Key Laboratory of Particle Detection and Electronics ， Beijing 100049 ， China;
4 Institute of Nuclear Physics and Chemistry, China Academy of Engineering Physics, Mianyang, 621900 ， China
Abstract: A new thermal neutron detector with the domestically produced THGEM (THick Gas Electron Multiplier) was developed as an alternative to 3 He to meet the needs of the next generation of neutron facilities. One type of Au-coated THGEM was designed specifically for the neutron detection. A detector prototype had been developed and the preliminary experimental tests were presented, including the performance of the Au-coated THGEM working in the Ar/CO2 gas mixtures and the neutron imaging test with 252 CF source, which would provide the reference of experimental data for the research in future.
Key words: THGEM Boron convertor ， ， Thermal neutron detector ， Two-dimensional position sensitivity
PACS: 29.40.Gx, 29.40.Cs, 28.20.Cz
## 1. Introduction
Neutrons are used to investigate the structure and dynamics of a material. Many efforts have recently been devoted to the development of the next generation of neutron facilities, which include SNS in USA, J- PARC in Japan, ISIS in UK, CSNS (China Spallation Neutron Source) in China and ESS in Europe [1] . The neutron detector is one of the key components of the neutron scattering instruments. With the international development of the new generation neutron source, the traditional neutron detector based on He has not been able to satisfy very well the demand of the 3 application of high flux especially. And also facing the global crisis of He supply as 3 well [2] , the research on the new style of the neutron detector which can replace the 3 He based detection technology becomes extremely urgent.
As a good candidate, a boron-coated GEM became the focus of attention recently [3,4] , which was firstly designed by Martin Klein using CERN standard GEM in 2006 [4] . It has the outstanding and excellent characteristics, such as high counting rate capability (>10MHz/mm ), good spatial resolution and timing properties, radiation 2 resistance, flexible detector shape and readout patterns [5] . In 2011, IHEP [6] and UCAS [7] firstly developed successfully a kind of THGEM, manufactured economically by standard printed-circuit drilling and etching technology in China. Compared with the CERN standard GEM, THGEM has higher gain, sub-millimeter spatial resolution and the possibility of industrial production capability of large-area robust detectors [8] , which is very suitable for the application of neutron detection.
the provided by Xie Yuguang group of IHEP. It's a Au-coated THGEM with a thickness of 300 was made of FR4 glass epoxy substrate, with copper cladding on both sides and then Au coated on the copper, which had an active area of 50mm*50mm.In order to study its basic characteristics as the references for the development of this kind of THGEM based neutron detector, the performances of the counting rate plateau, the energy resolution and the gain had been measured in the different Ar/CO gas mixtures with the different high voltages. According to the tests, the working conditions optimized of the THGEM had been obtained. In this paper, it would be presented the experimental research on a new kind of neutron detector based on the domestically produced THGEM which was μ m, hole diameter of 250 μ m, pitch of 600 μ m and a rim of 80 μ m. The THGEM 2
to study the characteristics of the prototype including its data acquisition system and make a good preparation for the neutron beamline test in next step, a preliminary imaging measurement with to the THGEM-based neutron detector in future By using this kind of THGEM, a detector prototype had been developed. In order experiments, it would be very helpful for the design of this kind of 252 CF neutron source was carried out in IHEP. According [9,10] .
## 2. Design of the detector prototype
multiplication(~100), and the 2-D strips structure for signal readout. Fig.1.a shows the schematic view of the detector and the Fig.1.b is the image of the detector. The resistance chain is used to provide three channels of high voltage supply. With a layer coated on the drift cathode as the neutron convector, when a neutron is captured, either an its track a large number of Secondary Electrons (SE), which is proportional to the energy deposited in the drift gas volume (~3x10 SE for 1 MeV). THGEM is employed for SE multiplication and can provide an effective gain of several hundreds. And then, the signal will be induced on the 2-D strips structure for determining the spatial and timing information by the readout electronics and data acquisition. The detector is operated in flow mode with Ar/CO mixtures at atmospheric pressure. Continuous purge of cheap counting gas avoids ageing effects encountered in other detectors so as to get long term stability as well as long lifetime. Table 1 shows the design specifications of the detector prototype. The detector mainly consists of the neutron convertor, the single THGEM for gas α ion or a 7 Li ion is emitted into the drift gas volume, where it releases along 4 2 10 B
Fig.1. Schematic of the detector prototype
Table 1. Specifications of the detector prototype
| PARAMETER | SPECIFICATION |
|--------------------------|-----------------|
| Active Area | 50mm*50mm |
| Neutron Flux | <10 8 n/cm 2 .s |
| Spatial Resolution(FWHM) | <3mm |
| Timing Resolution | <1 μ s |
| [email protected]Å | ～ 4% |
| Max Counting Rate | >1MHz |
| Working mode | Real-time |
readout technology(Fig.2). The 2-D strips is followed directly and read out by a front-end electronics of 64 channels CIPix ASIC based daughter board. The strip period is 1.56mm with a gap of 0.26mm. For each dimension, there are 32 channels for readout. The CIPix ASIC was originally developed by the Heidelberg ASIC lab in 2000 for the DESY H1experiment. A single chip integrates 64 channels of a low noise charge sensitive preamplifier followed with a shaper and discriminator. In the typical neutron detection, no external trigger is available. On the contrary, individual neutron events are entirely uncorrelated. The FPGA is employed to realize the high bandwidth data processing. As one neutron is detected, a cloud of electron charges, drifting onto the readout dimensions (x,y)coordinates will be obtained by the means of simultaneous coincidence of signals on x- and y-readout channels. After coincidence, the time stamp of each neutron will be information(x,y,t) is completed. Finally, the data will be transferred to the PC for data acquisition and analysis. The data acquisition system of the detector is based on the analogue front-end ASIC chip and the structure, x added and will be and y. Thus, then an modern Fieldcollected the detection event reconstruction by electrode of Programmable-Gate-Array (FPGA) strips a neutron including corresponding to both at the the spatial corresponding and timing
Fig. 2. Schematic of the data flow and processing
## 3. Thermal neutron efficiency
Compared with the highly reactive and expensive Li, solid 6 10 B layer seems to be much more favourable and suitable for use as neutron converter. It can easily be produced in reasonable sizes using evaporation or sputtering techniques. 10 B (enrichment >99%) is commercially available. The enriched 10 B is coated on one surface of the copper cathode plate. Neutrons are detected by the following neutron reactions:
$$\begin{smallmatrix} n ^ { + ^ { 1 0 } } B \longrightarrow & \alpha ^ { + ^ { 7 } } L i + 2. 7 9 \L e V & & 7 \% & & E _ { \alpha } = 1. 7 8 \L e V & & E _ { L i } = 1. 0 \L e V \\ n ^ { + ^ { 1 0 } } B \longrightarrow & \alpha ^ { + ^ { 7 } } L i ^ { ^ { * } } + 2. 3 1 \L e V & & 9 3 \% & & \\ & & \alpha ^ { + ^ { 7 } } L i + \gamma ^ { + } + 2. 3 1 \L e V & & E _ { \alpha } = 1. 4 7 \L e V & & E _ { L i } = 0. 8 4 \L e V \end{smallmatrix}$$
When the thermal neutron induces the 10 B(n, α ) 7 Li reaction, 93 % of all the reactions lead to the first excited state of 7 Li , which decays spontaneously (~73fs * half-time) to the ground state of Li by emitting the 0.48 MeV gamma ray, and 7% of 7 the reactions result in the ground state of 7 Li. The thermal neutron (0.0253eV) cross section of the 10 B(n, α ) Li reaction is 3840 barn and it drops rapidly with increasing 7 the thermal neutron energy. The maximum conversion efficiency is about 5% for thermal neutrons through a single pure 10 B-layer with the thickness of 2.5 μ m. For charged ions α and 7 Li, the gas detector can reach its efficiency very close to 100%. As a result, the neutron efficiency can be considered as the conversion efficiency. For the thickness of 2.0 μ m used in the prototype, the thermal neutron efficiency calculated by Geant4 is 4.6%.
## 4. Results and discussions
## 4.1. Tests on THGEM foil
In order to study its basic characteristics as the references for the development of this kind of THGEM based neutron detector, the performances of the counting rate plateau, the energy resolution and the gain had been measured in the different Ar/CO 2 mixtures with the different high voltages. The THGEM foil was tested by using a 55 Fe X ray source (activity 10mCi) which was positioned in such a way that a collimated beam ( ф 1 hole) of X-rays perpendicularly entered the upper drift region. The signal was induced by a cathode pad and read out with an ORTEC 142IH preamplifier followed by an ORTEC 572A amplifier (shaping time t=2 μ s) and an ORTEC multi-channel analyzer (trump-usb-8k).
First of all, several THGEM foils were tested to check the stability of the gain in 12 hours. And then one of this foils which showed the best performance was chosen for the next tests below. After 200 minutes, the effective gain would get stable as shown in the Fig.3.a. Consequently, the THGEM warmed up about 200 minutes before every test. Fig.3.b shows the gain in the different Ar/CO mixtures with the 2 different high voltages. The THGEM can provide an effective gain range from 300 to 3000 in Ar/CO2 mixtures. For the gas mixture Ar/CO (90%/10%), the working 2 voltage will be lower to give the same gain.
Fig.3. Gain and its stability of the THGEM
In order to know the suitable working voltage of the detector in different Ar/CO 2 mixture, its counter plateau was measured in different total voltage. During the experiment, the total flow of Ar and CO2 gas is 50 SCCM to ensure the amount of effective working gas in the chamber. The counts were recorded in every one minute and the voltage of THGEM was increased by the increment of 20V until the spark discharge occurred. As the Fig.4 shows, it has a longer plateau in the Ar/CO 2 mixture ratio of 90%/10% and the THGEM woks in a lower voltage. For the gas mixture Ar/CO2 (90%/10%), the plateau range of the THGEM is from 2200V to 2400 V and its plateau slope is smaller than 3% / 100 V, as shown in the left of the Fig.4. This optimization would be helpful to know the working range of the THGEM and find the conditions at the lower HV.
Fig.4. Counting rate plateau with the total HV
Energy resolution is one of the most important parameters related to the detector performance. By using the 5.9 keV 55 Fe X-ray source, the energy resolution was measured. The much better energy resolution is obtained in the Ar/CO2 (90%/10%) gas mixture. Fig.5 shows the pulse height spectrum obtained with a 55 Fe source in the Ar/CO2 (90%/10%) gas mixture. For obtaining energy resolution, it's fitted with the Gaussian function. It indicates the energy resolution (FWHM) of the detector based on THGEM is about 34% at 2440V. With such an energy resolution, the detector can entirely separate the 3keV of Ar escape peak from the 55 Fe main X-ray peak located at 5.9keV.
Fig.5. A pulse height spectrum in the Ar/CO (90%/10%) gas mixture 2
## 4.2. Neutron tests in 252 CF source
In order to study the characteristics of the prototype including its data acquisition system and make a good preparation for the neutron beamline test in next step. In IHEP, a preliminary measurement was carried out with an isotopic neutron source 252 CF(~1 × 10 n/s), which was placed in a special shielding box with an open hole 6 with a diameter of 100 mm in the side. The detector was mounted onto the exit of the hole. For obtaining larger counts, no neutron moderation was used in the test.
The neutrons emitted from 252 CF source follow the fission spectrum and its average energy is 2.3MeV. The single cadmium mask cannot absorb all the neutrons. For the measurement of the spatial resolution, the mask with a slit of 3mm wide was made of 25mm thick boron plastic and 1mm thick cadmium. Due to the weak source, the test took 24 hours. The 2-D counts image is shown in the Fig.6.a and Fig.6.b is the projection on y axis. The counts distribution is fitted with the Gaussian function and the FWHM of spatial distribution is 4.2mm, which is a bit worse. The reason is because the neutrons emitted from 252 CF source have lots of energy and the mask with the slit cannot absorb all the neutrons completely. This is the main reason why a monoenergetic neutron beam line is needed for the accurate measurement.
Fig.6. Spatial distribution
For testing the imaging ability of the whole system, the mask with a letter E was used, which was made of 25mm thick boron plastic and 1mm thick cadmium. The test took 24 hours as well. Due to the same reason, the image (Fig.7) is not good.
However, it is sure that the detector prototype works normally as expected for neutron detection.
Fig.7. Neutron imaging
## 5. Summary
In this paper, a new type of neutron detector based on domestically produced Au-coated THGEM is introduced and the preliminary experimental study is presented, including the performance of the Au-coated THGEM working in the Ar/CO2 mixtures and the neutron test of the detector prototype using 252 CF source. This work is considered as the preparation of the accuracy measurement at the neutron beamline in the reactor, which will be reported in the near future.
## Acknowledgements
We are grateful to the supports from China Spallation Neutron Source, the State Key Laboratory of Particle Detection and Electronics, the Laboratory of Neutron Detection and Fast Electronics Technology in Dongguan University of Technology, the Laboratory of Neutron Physics in Institute of Nuclear Physics and Chemistry, China Academy of Engineering Physics.
## References
[1] Gebauer B. Towards detectors for next generation spallation neutron sources. Nucl Instr and Meth A, 2004, 535: 65-78.
[2] Dana A, Daniel M. The Helium-3 Shortage: Supply, Demand, and Options for Congress:Congressional Research Service, 2010.
- [3] Ohshita H, Uno S, Otomo T, et al. Development of a neutron detector with a GEM.Nucl Instr and Meth A, 2010,623:126-128.
- [4] Schmidt C, Klein M. The CASCADE neutron detector: A system for 2D position sensitive neutron detection at highest intensities.Neutron News, 2006, 17(1):12-15.
- [5] Klein M, Schmidt C. CASCADE neutron detectors for highest count rates in combination with ASIC/FPGA based readout electronics. Nucl Instr and Meth A, 2011, 628:9-18.
- [ 键入文字 ] [6] Xie Y.G, Lv J.G, Zhang A.W et al. Development of Au-coated THGEM for single photon, charged particle, and neutron detection. Nucl Instr and Meth A ,2013, 729:
## 809-815.
[7] Liu H.B, Zheng Y.H, Xie Y.G et al. The performance of thinner-THGEM. Nucl Instr and Meth A ,2011,659:237-241.
- [8] Breskin A, Alon R, Cortesi M et al.A concise review on THGEM detectors. Nucl Instr and Meth A ， 2009 ， 598:107-111.
- [9] Zhou J.R, Sun Z.J ， Liu B, et al. Neutron beam monitor based on a boron-coated GEM.Chinese Physics C,2011,35(7):668-674.
- [10] Wang Y.F, Sun Z.J, Zhou J.R, et al. Simulation study on the boron-coated GEM neutron beam monitor.Chinese Physics, Science China Physics,Mechanics & Astronomy, 2013, 56(10):1897-1902.
Submitted to 'Chinese Physics C'
| null |
[
"Yang Lei",
"Zhou Jian-Rong",
"Sun Zhi-Jia",
"Zhang Ying",
"Huang Chao-Qiang",
"Sun Guang-Ai",
"Wang Yan-Feng",
"Yang Gui-An",
"Xu Hong",
"Xie Yu-Guang",
"Chen Yuan-Bo"
] |
2014-08-05T04:35:30+00:00
|
2014-08-05T04:35:30+00:00
|
[
"physics.ins-det",
"nucl-ex"
] |
Experimental research on the THGEM-based thermal neutron detector
|
A new thermal neutron detector with the domestically produced THGEM (THick
Gas Electron Multiplier) was developed as an alternative to 3He to meet the
needs of the next generation of neutron facilities. One type of Au-coated THGEM
was designed specifically for the neutron detection. A detector prototype had
been developed and the preliminary experimental tests were presented, including
the performance of the Au-coated THGEM working in the Ar/CO2 gas mixtures and
the neutron imaging test with 252CF source, which would provide the reference
of experimental data for the research in future.
|
1408.0863v1
|
## Accurate projected augmented wave (PAW) datasets for rare-earth elements (RE=La-Lu)
M. Topsakal 1 and R.M. Wentzcovitch 1
1 Department of Chemical Engineering and Materials Science, University of Minnesota, Minneapolis, MN, USA (Dated: August 6, 2014)
We provide accurate projected augmented wave (PAW) datasets for rare-earth (RE) elements with some suggested Hubbard U values allowing efficient plane-wave calculations. Solid state tests of generated datasets were performed on rare-earth nitrides. Through density of state (DOS) and equation of state (EoS) comparisons, generated datasets were shown to yield excellent results comparable to highly accurate all-electron full-potential linearized augmented plane-wave plus local orbital (FLAPW+LO) calculations. Hubbard U values for trivalent RE ions are determined according to hybrid functional calculations. We believe that these new and open-source PAW datasets will allow further studies on rare-earth materials.
## I. INTRODUCTION
The lanthanide series of the periodic table comprises fifteen members ranging from Lanthanum (La) to Lutetium (Lu). Although they are more abundant than silver, and some of them are more abundant than lead, they are known as rare-earth (RE) elements. The 'rare' in their name refers to the difficulty of obtaining the pure elements, not to their abundances in nature. They are never found as free metals in the Earth's crust and do not exist as pure minerals. All having two electrons in the outermost 6 s orbital, they can form trivalent cations. Their chemistry is largely determined by the ionic radius, which decreases steadily along the series corresponding to the fillings of the 4 f -orbitals.
challenging for density functional theory (DFT) methods. It is well known that the narrow f -bands in RE compounds are not adequately described by standard local density approximation (LDA) 13 and generalized-gradient approximation (GGA) 14 due to strong electronic correlation effects. To cope with this insufficiencies, the DFT+U method 15 is often employed, with U fit to some spectral data. Another approach is to use a hybrid functional introducing a portion of exact Hartree-Fock type exchange into the exchange-correlation functional. 16-18 Both approaches improve the treatment of strongly correlated electrons. However DFT+U suffers from ambiguity of Hubbard U value while hybrid functionals suffer from extremely demanding computational costs.
One of the principal applications of the rare-earth elements in industry, involving millions of tons of raw material each year, is in the production of catalysts for the cracking of crude petroleum. 1 They are also commonly used in the glass and display industry. 2 Additionally, rare-earth oxides are regarded as potential candidates for high-k gate dielectrics because of their ability to form direct contact with silicon substrate. 3 Superconductors based on rare-earth oxypnictides 4-6 were also discovered with critical temperatures as high as 55 K. 7 Very recently it was shown that the rare-earth oxide ceramics are intrinsically hydrophobic and durable materials preventing water from spreading over a surface. 8 S. Bertaina et al. 9 revealed a new family of spin qubits based on RE ions having desired characteristics suitable for scalable quantum computation at 4 He temperatures. Adsorption of RE adatoms on graphene were investigated by several studies. In contrast to most of simple and transition metals, it was shown that RE adatoms induce significant electric dipole and magnetic moments on graphene. 10-12
Although RE elements constitute a significant portion of atoms in the periodic table and exhibit unique (and mostly unexplored) properties, it is interesting to note that the ab-initio electronic structure calculations with RE elements are very scarce. This is due to the fact that RE elements containing 4 f -electrons are particularly very
We realized that reliable DFT potentials for majority of RE elements are not provided by any open-source 19 DFT simulation environments such as ABINIT, 20 SIESTA, 22 and QUANTUM ESPRESSO. 23,25 The aim of this study is to provide projected augmented wave (PAW) potentials for RE elements with some suggested Hubbard U values allowing efficient plane-wave calculations. These potentials were developed to yield results comparable to highly accurate all-electron FLAPW+LO results using WIEN2K. 26 Hubbard U values are consistently determined according to all-electron hybrid functional calculations. The optimization of the potentials were performed on rare-earth nitrides in which RE element adopts trivalent ionic state.
The paper is organized as follows: In section II we present the details about the generation of RE PAW datasets. Next we investigate the electronic properties of RE-nitrides (REN's) which will be used as a reference for validation of our generated PAW datasets. In section IV we compare the performance of generated PAW datasets with all-electron FLAPW+LO results. In section V we present our procedure of obtaining Hubbard U values for feasible DFT+U calculations that match results from demanding hybrid functional (YS-PBE0) calculations. Our conclusions are summarized in section VI.
## II. METHODS AND CALCULATION DETAILS
We used ATOMPAW program 28-30 to generate projector and basis functions which are needed for performing first principles calculations based on the projector augmented wave (PAW) method. 31 For any element in the periodic table, the program inputs the atomic number, exchange-correlation functional, electronic configuration, choice of basis functions, and a cutoff radii. After initialization, the program can be instructed to output potential files which can be used by open-source density functional simulation environments such as QUANTUM ESPRESSO 23,24 and ABINIT. 20,21 It also outputs several files enabling wave function and logarithmic derivative plots.
Although Scandium (Sc) and Yttrium (Y) belong to rare-earth element family, we focus our attention in the lanthanides (La-Lu) since PAW datasets already exist for Sc and Y in the ATOMPAW repository. For each lanthanide we used neutral electronic configuration. 6 , s 5 , 5 s p , 5 d and 4 f orbitals are treated as valance states. Two projector and basis functions were used for each angular momentum channel (s,p,d,f). Two Rydberg reference energy was used to build each partial-waves. Since lanthanides are high-Z elements, scalar-relativistic wave equation was used for all-electron computations. VANDERBILT scheme 32 was used to pseudize partial-waves. The analytical form of shape functions used in compensation density was chosen according to the BESSELSHAPE option. 33 The generated dataset was carefully checked with logarithmic derivatives against ghost states. [Xe] 4 f ( n ) 5d 6s 1 2 atomic configuration was assumed for each RE in which n=0 for La, n=1 for Ce, n=2 for Pr, n=3 for Nd, etc.
Cutoff radius 28 (r cut ) of augmentation regions in the PAW formalism was determined according to solid state calculations performed on RE-nitrides (REN's). We first obtained equilibrium lattice constants (a 0 ) for each REN using the all-electron WIEN2K 26,27 program which employs highly accurate full-potential linearized augmented plane-wave + local orbitals (FLAPW+LO) method. Then we tuned cutoff radius of each PAW dataset to yield equilibrium lattice constant error not larger than ± 0.03 % relative to WIEN2K. Table I lists r cut values for each RE.
In all-electron (WIEN2K) calculations, the wave functions were expanded in spherical harmonics inside nonoverlapping atomic spheres of radius R MT (muffin-tin radii), and in plane waves for the remaining space of the unit cell. WIEN2K R MT defaults of each rareearth and nitrogen (N) atoms are reduced by %10. Corresponding R MT values of each RE are presented in Table I. The plane wave expansion in the interstitial region was made up to a cut-off wave vector chosen to be K max = 9.0/ R MT which is %28 larger than the WIEN2K default. Setting K max to 10.0/ R MT or 11.0/ R MT did not change equation of state results noticeably while significantly increasing computational time.
Table I: Muffin-tin radii values ( R MT ) used in WIEN2K calculations and cutoff radius (r cut ) of augmentation regions in PAW method used in ATOMPAW.
| Element | R MT (a.u) | r cut (a.u) |
|-----------|--------------|---------------|
| La | 2.47 | 2.45 |
| Ce | 2.35 | 2.335 |
| Pr | 2.41 | 2.49 |
| Nd | 2.43 | 2.45 |
| Pm | 2.4 | 2.45 |
| Sm | 2.39 | 2.516 |
| Eu | 2.4 | 2.559 |
| Gd | 2.37 | 2.417 |
| Tb | 2.33 | 2.37 |
| Dy | 2.26 | 2.392 |
| Ho | 2.32 | 2.37 |
| Er | 2.31 | 2.375 |
| Tm | 2.29 | 2.393 |
| Yb | 2.28 | 2.4 |
| Lu | 2.27 | 2.25 |
The Brillouin-zone is sampled at 12x12x12 k-points using the tetrahedron method. To improve the visibility, density of states (DOS) plots were smoothed using a Gaussian with 0.003 Ry width. Hybrid functional calculations are performed using the YS-PBE0 functional (where YS stands for Yukawa screened) as implemented in WIEN2K. This functional was shown to yield similar results 35 to the original HSE functional which is a screened hybrid functional. 16 One quarter of the PBE 36 short-range exchange is replaced by the exact exchange, while the full PBE correlation energy is included. The hybrid calculations were carried out using a 9x9x9 k-points mesh. For DFT+U calculations, the standard Duradev implementation 37 is used where onsite Coulomb interaction for localized orbitals is parametrized by U effective = U - J ( which we denote henceforth as U d and U f for d and f orbitals ). We select J = 0.
For verification of PAW datasets, we performed planewave calculations using QUANTUM ESPRESSO. Kpoints sampling, smearing parameters and DFT+U method was chosen similar to WIEN2K calculations while 50 (200) Ry wave function (charge density) cutoff was used. N.pbe-kjpaw.UPF potential (provided by QUANTUMESPRESSOweb page 34 ) was used for Nitrogen atom. In all WIEN2K and QUANTUM ESPRESSO calculations, spin polarized PBE exchange correlation functional was used.
For the rest of the paper, we shall simply refer to rare-earth as 'RE', rare-earth nitrides as 'REN's', allelectron FLAPW+LO as 'AE', plane augmented wave dataset as PAW, calculations with GGA-PBE functional as 'PBE', hybrid functional calculations as 'YS-PBE0' when applicable.
Figure 1: (Color online) Cubic unit cell of LaN. (b) Total and projected density of states (DOS) of LaN calculated using all-electron FLAPW+LO method. PBE exchange-correlation functional was used. Core-states are shaded. Positive (negative) y-axis corresponds to DOS of spin-up (spin-down) states. Zero of the energy was set to Fermi energy (E Fermi ). (c) DOS of GdN having seven f-orbitals filled and seven f-orbitals empty with total magnetic moment µ =7 µ B per primitive cell. (d) DOS of LuN having completely filled f-orbitals. (e) DOS for other REN's calculated using all-electron FLAPW+LO method.
## III. AE CALCULATIONS ON RARE-EARTH NITRIDES
The fidelity of the logarithmic derivatives of the pseudo wave functions in comparison with their all-electron counterparts is a requirement for developing a reliable
PAW dataset. However it is by no means sufficient. One needs to test the performance of generated dataset in realistic environments through solid-state calculations. In this section, we try to build our understanding of the electronic properties of rare-earth elements through reliable all-electron WIEN2K FLAPW+LO calculations to
which we will frequently refer in the following sections. In a similar manner, Lejaeghere et al. 38 have used WIEN2K as a reference to describe quantitatively the discrepancies in the equation of state of a wide set of elements in the periodic table (71 elements, excluding lanthanides). We also believe that our all-electron calculations in this section might provide useful information for other studies involving RE-elements.
We select rare-earth nitrides (REN's) as testing material due to their simple rocksalt structure and trivalent state of RE. As discussed in the introduction section, RE elements are never found as free metals in the Earth's crust and do not exist as pure minerals. They are very reactive with oxygen in the ambient atmosphere. Generally the trivalent state (RE +3 ) is the most stable form and therefore, sesquioxides 39 (RE 2 O 3 ), nitrides 40 (REN), cobaltates 41 (RECoO ) exist for rare-earth elements. We 3 believe that our generated RE datasets should also work equally well for other RE ions since they are generated from neutral electronic configuration.
The atomic structure of LaN is illustrated In Fig. 1 (a). The crystal system is cubic and belongs to the (Fm ¯ m) symmetry group. 3 Unless stated otherwise, we use experimental lattice constants 40,42 for any DOS and band structure calculation in this study. Total and local density of states for LaN is presented in Fig. 1 (b) calculated using standard PBE functional. Valance states near the Fermi level have mainly Nitrogen-2 p character while Nitrogen-2 s states lie approximately 11 eV below the Fermi level. The La-5 d and La-6 s orbitals don't have significant contribution in the valance states since they are ionized. The La-5 s semi-core state is approximately 31 eV below the Fermi level and lower lying states of La (4 d , 4 p , 4 ...) s and N-1 s orbitals are considered as corestates as depicted in Fig. 1 (b). Empty La-4 f states lie approximately 4 eV above the Fermi level. LaN is a nonmagnetic semi-metal and its PBE band structure is shown in 6 (b). Lowest energy conduction states have La-5 d orbital character.
Starting from Cerium (Ce), 4 f orbitals start to get occupied in the REN's. Gadolinium nitride (GdN) is the case in which seven 4 f orbitals are half occupied so it has a huge magnetic moment as µ =7 µ B per primitive cell. Total and local density of states for GdN are presented in Fig. 1 (c). Similar to LaN, the top of valance band has primarily Np states. Half occupied 4 f states are approximately 3.2 eV below the Fermi level and they interact with Np states. For the case of Lutetium nitride (LuN), each of f-orbitals are fully occupied so we have a nonmagnetic case. Total and local density of states for LuN is presented in Fig. 1 (d).
Fig. 1 (e) shows total and local density of states for the rest of REN's. One can easily track the increase of occupied spin-up electrons from Ce to Eu and spin-down electrons from Tb to Yb. In CeN, the single f-electron is distributed in spin-up and spin-down orbitals and it produces µ =0.27 µ B per primitive cell. Similarly YbN has only one empty 4 f orbital which is distributed in spin-up and spin-down orbitals and it produces µ =0.22 µ B per primitive cell. Others have almost integer magnetic moments, which are µ =1.97, 3.00, 4.00, 5.00, 6.00, 6.03, 5.01, 4.00, 3.00, 1.93 µ B for PrN, NdN, PmN, SmN, EuN, TbN, DyN, HoN, ErN, TmN respectively. Different from LaN, GdN, and LuN, the energies of f-states are very close to the Fermi level for REN's as shown in Fig. 1 (e).
## IV. TESTS OF GENERATED PAW DATASETS
In this section we test our generated PAW datasets with respect to their AE counterparts as presented in the previous section. For each RE, generated PAW datasets were converted into UPF format 43 and solid state calculations were performed using the QUANTUM ESPRESSO (QE). However it is also possible to perform these calculations using ABINIT since we also provide each PAW dataset in .abinit format readable by the ABINIT. Lattice constants, atomic positions, k-point sets and exchange-correlation functional parameters are kept identical for AE and PAW calculations.
The basic requirement is the matching of PAW total DOS with AE total DOS. Since we are using completely different software packages to get the DOS spectrum, very small differences are tolerable. In Fig. 2 (a) we compare total DOS of NdN and the agreement is fairly good. PAW (dashed) lines mostly overlap with AE (continuous) lines in the DOS. We believe that it is worthwhile to make similar comparison with previously generated RE pseudopotentials which are available on QUANTUM ESPRESSO repository 34 in order to reveal the importance of RE potentials on solid state DFT calculations. Nd.pbe-mt fhi.UPF is a norm-conserving (NC) type pseudopotential generated with Troullier-Martins (TM) method. 44 In Fig. 2 (b) total density of states obtained with this pseudopotential is compared with AE DOS. As apparently seen, there is a significant disagreement between NC and AE density of states - especially for f-states. In Fig. 2 (c), we compare another NC pseudopotential (Nd.pz-sp-hgh.UPF) which was generated using Hartwigsen-Goedecker-Hutter method (HGH). 45 Again the disagreement is quite pronounced. HGH potential even fails to produce a spin-polarized solution in contrast to PAW and TM results. This might be related with placement of f-electrons in the core region or failure of generated NC pseudopotential in trivalent ionic state. For GdN, generated PAW dataset agree excellently with AE as shown in Fig. 2 (d). On the other TM and HGH pseudopotentials ( Gd.pbe-mt fhi.UPF, Gd.pz-sphgh.UPF ) significantly fail as presented in Fig. 2 (e) and (f). All these results indicate that the choice of potentials are extremely important in DFT calculations.
In Fig. 2 (g) and (h) we compare the PBE band structure of NdN and GdN calculated with AE and PAW. Similar to DOS comparisons, band structures are almost identical between AE and PAW calculations for spin-up
Figure 2: (Color online) (a) Density of states (DOS) comparison of AE and PAW calculations for NdN. Positive (negative) y-axis corresponds to DOS of spin-up (spin-down) states. Zero of the energy was set to Fermi energy (E Fermi ). PBE exchangecorrelation functional was used. (b) DOS comparison of AE and Troullier-Martins (TM) norm-conserving (NC) pseudopotential calculations for NdN. (c) DOS comparison of AE and Hartwigsen-Goedecker-Hutter (HGH) norm-conserving (NC) pseudopotential calculations for NdN. (d) Same as (a) for GdN. (e) Same as (b) for GdN. (f) Same as (c) for GdN. (g) Band structure comparison of AE and PAW calculations for NdN. (h) Same as (g) for GdN.
and spin-down bands. For the rest of REN's, results of DOS and band structure tests are presented in Supplementary materials (see section IX) each showing excellent agreement with AE.
Upon achieving our first requirement that PAW total DOS and band structure should practically match with AE where each calculation was performed at experimental lattice constant, our next requirement is that similar agreement should also happen for different volumes where distances between atoms change. Equation of state
(EoS) calculation is a way of performing this test. The agreement of EoS for a generated potential with its allelectron counterpart is another measure of its reliability. For each REN's we have obtained their EoS calculated with AE and PAW.
Our strategy for EoS tests is as follows: First we obtain the AE energy versus volume curve for all RENs as exemplified in Fig. 3 (a). Results are fitted to a third order Birch-Murnaghan equation of state 46
$$E ( V ) = E _ { 0 } + \frac { 9 } { 1 6 } B _ { 0 } V _ { 0 } \left \{ \left [ \left ( \frac { V _ { 0 } } { V } \right ) ^ { 2 / 3 } - 1 \right ] ^ { 3 } B _ { 0 } ^ { \prime } + \left [ \left ( \frac { V _ { 0 } } { V } \right ) ^ { 2 / 3 } - 1 \right ] ^ { 2 } \left [ 6 - 4 \left ( \frac { V _ { 0 } } { V } \right ) ^ { 2 / 3 } \right ] \right \},$$
where E 0 is the equilibrium total energy, V 0 is the equilibrium volume, and B 0 and B 0 ′ are the bulk mod-
La Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er Tm Yb Lu
Figure 3: (Color online) (a) Equation of state (EoS) comparison of AE, PAW and NC calculations for NdN. Minimum of the energy was set to zero for each calculation. (b) Percent difference in AE versus PAW calculations for REN equilibrium lattice constants (a 0 ).
ulus and its pressure derivative, respectively. The corresponding cubic lattice constant (a 0 ) is obtained from V 0 . We repeat this procedure with QE PAW calculations and energy minima of each curve are set to zero. If AE and PAW V 0 values do not agree well, we change the cutoff (r cut ) of RE and generate a new PAW dataset. This procedure is repeated until the difference (% error) between AE and PAW lattice constants (a ) are less than 0 ± 0.03%. In Fig. 3 (a) AE and PAW EoS's are compared for NdN. Apparently, AE and PAW results are almost identical, similar to DOS and band structure tests presented in Fig. 2 (a). For sake of comparison, EoS of NdN calculated with Troullier-Martins (TM) normconserving (NC) type pseudopotential 34 is also presented in Fig. 3 (a). Since the electronic structure of NdN calculated using NC pseudopotentials is significantly different from PAW and AE calculations as shown in Fig. 2 (b), EoS curves are also significantly different as shown in Fig. 3 (a). For the rest of REN's, EoS tests are presented in Supplementary materials (see section IX). By taking the AE equilibrium lattice constant as reference, we can calculate % errors of lattice constants as depicted in Fig. 3 (b). Compared to PBE functional's typical overestimation 47 of lattice constants by about 1-2 %, we believe that there is no need to reduce the errors in Fig.
## V. SUGGESTED HUBBARD U VALUES
For the case of Cerium, J.L.F. Da Silva et al. 17 and P.J. Hay et al. 18 showed that hybrid functionals correctly predict Ce O 2 3 to be an insulator as opposed to the ferromagnetic metal predicted by the LDA and GGA. Calculated band gaps are also shown to be in close agreement with respect to available experimental data. For the case of GdN, M. Schlipf et al. 48 showed that FLAPW hybrid functional calculations attain good agreement with experimental data for band transitions, magnetic moments, and the Curie temperatures.
In Fig. 4, we present hybrid functional (YS-PBE0) density of states (DOS) results for all REN's (La-Lu). As compared to previous DOS calculations in Fig. 1, the YS-PBE0 functional yields significantly different DOS. RE 4 -states, which usually tend to lie near the Fermi f level in PBE calculations, get more localized and move away from the Fermi level. Especially in LaN, GdN, TbN, HoN, ErN, TmN, and LuN f-states are represented by distinct isolated peaks in the DOS. For CeN, PrN, SmN, EuN and YbN, f-states are still close to the Fermi level.
In section IV, we showed that DOS and EoS for all REN produced using our generated PAW datasets compare very well with those produced using FLAPWLO. However, hybrid functional results differ significantly from standard PBE results. Assuming that hybrid functional results are more reliable, the aim of this section is to provide suggested Hubbard U values for prospective users of RE PAW datasets that we generated. Of course it is always desirable to use hybrid functionals for any DFT calculation involving RE elements. However, in most cases this can be challenging. From our experience, a typical YS-PBE0 calculation on a REN can be 10 2 to 10 3 times more time consuming than a standard PBE calculation. With primitive cells containing 10-20 atoms, it is challenging even to start a YS-PBE0 calculation due to memory requirements. Another issue is related with the implementation of hybrid functionals into open-source density functional simulation environments. For example, at the time this paper is written, it is not possible to perform a PAW hybrid functional calculation using current version of QUANTUM ESPRESSO (QE 5.03). Therefore, it is desirable to perform DFT+U calculations that reproduce as closely as possible results from hybrid functional calculations. DFT+U calculations add little workload to standard DFT calculations. Sometimes U can be chosen to reproduce some well-established experimental spectroscopic data, which should in principle, be reproduced also by a hybrid functional calculation or using U values determined using linear response. 49 In this section, we try to define U values for some RE elements to allow us perform inexpensive DFT+U simulations giving results similar to YS-PBE0 calculations. These U values are suitable for trivalent ions in the highest spin state.
Figure 4: (Color online) Total and projected density of states (DOS) of REN's calculated using all-electron hybrid functional (YS-PBE0) FLAPW+LO method. Positive (negative) y-axis corresponds to DOS of spin-up (spin-down) states. Zero of the energy was set to Fermi energy (E Fermi ).
In principle, U varies with interatomic distances, atomic configurations, etc. However, these are small variations compared with those produced by different valence or spin states. 50-52
In the upper panel of Fig. 5 (a), valance DOS of NdN calculated by YS-PBE0 functional is shown. The peak due to f-states is located around -2 eV. This YSPBE0 DOS is considerably different from the PBE DOS shown in Fig. 1 (e), but can be well reproduced with the DFT+U method using U f = 3.1 eV for Nd (Fig. 5 (a)). Similarly good valence DOS agreement can be obtained with U f = 4.6 eV for Gd f-states in GdN, and U f =
Figure 5: (Color online) (a) Valance total density of states of NdN calculated with YS-PBE0 (upper) and PAW PBE+U f (lower) method. U =3.1 eV was used to match YS-PBE0 f DOS. Zero of the energy was set to Fermi energy (E Fermi ). (b) Same as (a) for GdN. U f =4.6 eV was used to match YSPBE0 DOS. (c) PAW PBE+U f calculations for the rest of REN's. Their YS-PBE0 counterparts are presented in Fig. 4 and the list of suggested U f values are presented in Table II.
4.6 for Lu f-states in LuN (see Fig. 5 (b) and (c)). For HoN, there are three distinct f-peaks in the valance region and it is more difficult to obtain an exact match between YS-PBE0 and PBE+U DOS. However, one can obtain a fairly good agreement for U f =4.95 eV. In Fig. 5 (c), we present PBE+U f DOS results that resemble their YSPBE0 counterparts shown in Fig. 4 for the remaining REN's. The list of suggested U f values are presented
Table II: Suggested Hubbard U f and U d values (see text).
| Element | U f (eV) | U d (eV) |
|-----------|------------|------------|
| La | 5.5 | 9.0 |
| Ce | 2.5 | - |
| Pr | 4 | - |
| Nd | 3.1 | 8.5 |
| Pm | 3.4 | 8.1 |
| Sm | 3.3 | - |
| Eu | 3 | - |
| Gd | 4.6 | 9.5 |
| Tb | 5 | 9.0 |
| Dy | 5 | - |
| Ho | 4.9 | 8.4 |
| Er | 4.2 | - |
| Tm | 4.8 | - |
| Yb | 3 | 8.0 |
| Lu | 5.5 | 8.2 |
in Table II. For YbN, CeN and PrN we are unable to find a U f value that closely match YS-PBE0 DOS. These cases are more subtle because of their orbital occupancies and non-zero J values (not considered in this study) are probably needed. Nevertheless, U f values provided here for trivalent high spin RE ions could be used in cases where no other alternative is available.
Similar to f-orbitals, RE 5 d orbitals, which constitute the low energy conduction bands of REN's, are also problematic in standard DFT. The first panel in Fig. 6 (a) shows the half-metal PBE band structure of GdN. However, experimental studies assert that GdN has a band gap around 1.3 eV 53 , which is taken as the average of majority and minority band gaps. The middle panel in Fig. 6 shows the YS-PBE0 bands with an average band gap of 1.40 eV, which is quite close to the experimental value. By taking YS-PBE0 band structure as a reference, one can obtain a similar band structure by applying U d = 9.5 eV to Gd-d orbitals as shown on the third panel of Fig. 6 (a). This procedure can be repeated for LaN and LuN as shown in Fig. 6 (b) and (c). U d = 9.0 (8.2) eV can be used to reproduce the YS-PBE0 band gap of LaN (LuN). The list of suggested U d values are shown in Table II. Due to the complex nature of states near Fermi level, we cannot safely determine U d values for some of REN's. In average, 8.5 eV U d is expect to work for most of them.
## VI. CONCLUSIONS
In conclusion, we have generated accurate and publicly available PAW datasets for RE elements (La-Lu) which can be used readily with open-source QUANTUM ESPRESSO and ABINIT DFT simulation packages. Similar to common trivalent RE ions found in na-
Figure 6: (Color online) Band structure of GdN calculated with PBE, YS-PBE0 and PBE+U d methods along Γ-X-W. Continuous and dashed lines correspond to spin-up and spindown bands respectively. U =9.5 eV was used to match YSd PBE0 band gap. (b) Same as (a) for LaN. U =9.0 eV was d used to match the YS-PBE0 band gap. (c) Same as (a) for LaN. U =8.2 eV was used to match YS-PBE0 band gap. d
ture, solid state tests and optimizations of PAW datasets were performed on rare-earth nitrides. We find that our optimized PAW datasets yield almost identical results to highly accurate all-electron full-potential linearized augmented plane-wave plus local orbital (FLAPW+LO)
- 1 D.H. Cornell, Pure Appl. Chem. 65, 1993, 2453.
- 2 K.A. Gschneidner, Jr., Am. Chem. Soc. 1981.
- 3 S. Coh, T. Heeg, J.H. Haeni, M.D. Biegalski, J. Lettieri, L.F. Edge, K.E. OBrien, M. Bernhagen, P. Reiche, R. Uecker, S. Trolier-McKinstry, D.G. Schlom and D. Vanderbilt, Phys. Rev. B 82, 2010, 064101.
- 4 Y. Kamihara, H. Hiramatsu, M. Hirano, R. Kawamura, H. Yanagi, T. Kamiya and H. Hosono, J. Am. Chem. Soc. 128, 2006, 012.
- 5 J. Yang, Z.-C. Li, W. Lu, W. Yi, X.-L. Shen, Z.-A. Ren, G.-C. Che, X.-L. Dong, L.L. Sun, F. Zhou and Z.-X. Zhao, Supercond. Sci. Technol. 21, 2008, 082001.
- 6 K. Ishida, Y. Nakai and H. Hosono, J. Phys. Soc. Jpn. 78,
calculations. All-electron hybrid functional calculations (YS-PBE0) were carried out to overcome limitations of standard PBE calculations and to be used as reference for the determination of Hubbard U values for PBE+U calculations. PBE results tend to place f-states very close to the Fermi level and 2.5 to 5.5 eV Hubbard U values are required to place f-states in positions similar to those produced by the YS-PBE0 hybrid functional. Nonzero Hubbard U value ( ∼ 8 5 . eV) on empty RE d-orbitals was also shown necessary to open a band gap of some REN's, which is consistent with experimental findings. We believe that these new PAW datasets - if used with suggested Hubbard U values (for trivalent high-spin ions) - will allow further studies on rare-earth materials.
## VII. ACKNOWLEDGMENTS
This work was supported primarily by NSF/DMR through the University of Minnesota MRSEC. RMW was also supported by NSF/EAR 1319361. We acknowledge fruitful discussions with Dr. Koichiro Umemoto, Prof. Natalie Holzwarth and Prof. David Vanderbilt. Computations were performed at the Minnesota Supercomputer Institute (MSI).
## VIII. APPENDIX A. ACCESS TO GENERATED PAW DATASETS
ATOMPAW input files used in the generation of LDA and GGA PAW datasets and corresponding output potential files which can be used with ABINIT and QUANTUM ESPRESSO are publicly available at http://www.vlab.msi.umn.edu/resources/repaw/index.shtml.
## IX. APPENDIX B. SUPPLEMENTARY MATERIAL
Supplementary material associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.commatsci.2014.07.030.
2009, 062001.
- 7 Z.A. Ren, G.C. Che, X.L. Dong, J. Yang, W. Lu, W. Yi, X.L. Shen, Z.C. Li, L.L. Sun, F. Zhou and Z.X. Zhao, Europhys. Lett. 83, 2008, 17002.
- 8 G. Azimi, R. Dhiman, H.-M. Kwon, A.T. Paxson and K.K. Varanasi, Nature Mater. 12, 2013, 315.
- 9 S. Bertaina, S. Gambarelli, T. Mitra, B. Tsukerblat, A. Mller and B. Barbara, Nature (London) 453, 2008, 203.
- 10 X. Liu, C.Z. Wang, M. Hupalo, Y.X. Yao, M.C. Tringides, W.C. Lu and K.M. Ho, Phys. Rev. B 82, 2010, 245408.
- 11 M. Alemani, A. Barfuss, B. Geng, C. Girit, P. Reisenauer, M.F. Crommie, F. Wang, A. Zettl and F. Hellman, Phys. Rev. B 86, 2012, 075433.
- 12 X. Liu, C.-Z. Wang, M. Hupalo, H.-Q. Lin, K.-M. Ho and M.C. Tringides, Crystals 3, 2013, 79.
- 13 W. Kohn and L.J. Sham, Phys. Rev. 140, 1965, A1133.
- 14 J.P. Perdew and Y. Wang, Phys. Rev. B 45, 1992, 13244.
- 15 1 V.I. Anisimov, J. Zaanen and O.K. Andersen, Phys. Rev. B 44, 1991, 943.
- 16 J. Heyd, G.E. Scuseria and M. Ernzerhof, J. Chem. Phys. 124, 2006, 219906.
- 17 J.L.F. Da Silva, M.V. Ganduglia-Pirovano, J. Sauer, V. Bayer and G. Kresse, Phys. Rev. B 75, 2007, 045121.
- 18 P.J. Hay, R.L. Martin, J. Uddin and G.E. Scuseria, J. Chem. Phys. 125, 2006, 034712.
- 19 Vienna ab-initio Simulation Package (¡http://www.vasp.at¿) provides PBE potential files for RE elements. But it is proprietary (not open-source). Testings and LDA versions of the potentials are not available.
- 20 X. Gonze, et al., Comput. Phys. Commun. 180, 2009, 2582.
- 21 ¡http://www.abinit.org¿.
- 22 J.M. Soler, E. Artacho, J.D. Gale, A. Garcia, J. Junquera, P. Ordejon and D. Sanchez-Portal, J. Phys.: Condens. Matter 14, 2002, 2745.
- 23 P. Giannozzi, et al., J. Phys.: Condens. Matter 21, 2009, 395502.
- 24 ¡http://www.quantum-espresso.org¿.
- 25 K.F. Garrity, J.W. Bennett, K.M. Rabe and D. Vanderbilt, Comput. Mater. Sci. 81, 2014, 446.
- 26 K. Schwarz and P. Blaha, Comput. Mater. Sci. 28, 2003, 259.
- 27 ¡http://www.wien2k.at/¿. h
- 28 N.A.W. Holzwarth, A.R. Tackett and G.E. Matthews, Comput. Phys. Commun. 135, 2001, 329.
- 29 A.R. Tackett, N.A.W. Holzwarth and G.E. Matthews, Comput. Phys. Commun. 135, 2001, 348.
- 30 ¡http://users.wfu.edu/natalie/papers/pwpaw/man.html¿.
- 31 P.E. Blchl, Phys. Rev. B 50, 1994, 17953.
- 32 D. Vanderbilt, Phys. Rev. B 41, 1990, 7892R.
- 33 G. Kresse and D. Joubert, Phys. Rev. B 59, 1999, 1758.
- 34 ¡http://www.quantum-espresso.org/pseudopotentials/¿.
- 35 F. Tran and P. Blaha, Phys. Rev. B 83, 2011, 235118.
- 36 J.P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev. Lett. 77, 1996, 3865.
- 37 S.L. Dudarev, G.A. Botton, S.Y. Savrasov, C.J. Humphreys and A.P. Sutton, Phys. Rev. B 57, 1998, 1505.
- 38 K. Lejaeghere, V. Van Speybroeck, G. Van Oost and S. Cottenier, Crit. Rev. Solid State Mater. Sci. 2014, 39.
- 39 S. Bernal, G. Blanco, J.M. Gatica, J.A. Perez-Omil, J.M. Pintado, H. Vidal G. Adachi, N. Imanak, Z.C. Kang (Eds.), Binary Rare Earth Oxides, Dordrecht, Netherlands, 2004.
- 40 W. Pies and A. Weiss, Landolt-Brnstein, Group III Condensed Matter vol. 7c1, 2006, Springer Verlag; Berlin, (Chapter VI.1.5.1).
- 41 M. Tachibana, T. Yoshida, H. Kawaji, T. Atake and E. Takayama-Muromachi, Phys. Rev. B 77, 2008, 094402.
- 42 Promethium (Pm) is a radioactive element and experimental lattice constant is not available for PmN. We take the average of NdN and SmN lattice constants for PmN.
- 43 ¡http://www.quantum-espresso.org/pseudopotentials/unifiedpseudopotential-format/¿.
- 44 N. Troullier and J.L. Martins, Phys. Rev. B 43, 1991, 1993.
- 45 C. Hartwigsen, S. Goedecker and J. Hutter, Phys. Rev. B 58, 1998, 3641.
- 46 F. Birch, Phys. Rev. 71, 1947, 809.
- 47 N. Staroverov, G.E. Scuseria, J. Tao and J.P. Perdew, Phys. Rev. B 69, 2004, 075102.
- 48 M. Schlipf, M. Betzinger, C. Friedrich, M. Lezaic and S. Blgel, Phys. Rev. B 84, 2011, 125142.
- 49 Cococcioni and S. de Gironcoli, Phys. Rev. B 71, 2005, 035105.
- 50 H. Hsu, K. Umemoto, M. Cococcioni and R.M. Wentzcovitch, Phys. Rev. B 79, 2009, 125124.
- 51 Hsu, P. Blaha, R.M. Wentzcovitch and Chris Leighton, Phys. Rev. B. 82, 2010, 100406.
- 52 H. Hsu, P. Blaha, M. Cococcioni and R.M. Wentzcovitch, Phys. Rev. Lett. 106, 2011, 118501.
- 53 H.J. Trodahl, A.R.H. Preston, J. Zhong, B.J. Ruck, N.M. Strickland, C. Mitra and W.R.L. Lambrecht, Phys. Rev. B 76, 2007, 085211.
| null |
[
"M. Topsakal",
"R. M. Wentzcovitch"
] |
2014-08-05T04:48:12+00:00
|
2014-08-05T04:48:12+00:00
|
[
"cond-mat.mtrl-sci"
] |
Accurate projected augmented wave (PAW) datasets for rare-earth elements (RE=La-Lu)
|
We provide accurate projected augmented wave (PAW) datasets for rare-earth
(RE) elements with some suggested Hubbard U values allowing efficient
plane-wave calculations. Solid state tests of generated datasets were performed
on rare-earth nitrides. Through density of state (DOS) and equation of state
(EoS) comparisons, generated datasets were shown to yield excellent results
comparable to highly accurate all-electron full-potential linearized augmented
plane-wave plus local orbital (FLAPW+LO) calculations. Hubbard U values for
trivalent RE ions are determined according to hybrid functional calculations.
We believe that these new and open-source PAW datasets will allow further
studies on rare-earth materials.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.