ibm-research/trajcast.datasets-arxiv2025

TrajCast Dataset arXiv 2025

This dataset contains atomic configurations of molecular and condensed matter systems used to train, validate, and test the TrajCast framework for forecasting molecular dynamics (MD) trajectories using autoregressive equivariant message-passing networks. Each entry represents a snapshot from a MD trajectory, defined by the chemical elements, instantaneous positions, and velocities of all atoms, along with information about their state at a slightly later time interval Δt.

Subdirectories

The dataset includes data for the models presented and benchmarked in our preprint, with time intervals varying for each system. Additionally, we provide a small example dataset to help you get started with our example notebooks.

• paracetamol: Dataset for gas-phase paracetamol (20 atoms) using a time interval of 7 fs.
• quartz: Data for crystalline quartz (162 atoms) using a time interval of 30 fs.
• water: Data for liquid bulk water (192 atoms) using a time interval of 5 fs.
• example: Subset of dataset for gas-phase paracetamol (20 atoms) using a time interval of 7 fs. About 5-10 % of the full dataset.

Dataset Features

The data comes in .extxyz format and we follow the units of style real from LAMMPS. Each entry correspondgs to an atomic configuration and includes the following fields:

• species: The type of atom (e.g., C for Carbon, O for Oxygen, ...).
• positions: 3D atomic positions (in Angstroms).
• velocities: 3D atomic velocities (in Angstroms per fs).
• displacements: 3D atomic displacements to the new positions after time interval Δt (in Angstroms).
• update_velocities: 3D atomic velocities at time interval Δt (in Angstroms per fs).
• timestep: Time interval Δt (in femtoseconds).
• pbc: Information about periodic boundary conditions (e.g., T T T).
• Lattice: The lattice vectors of the simulation box (in Angstroms).
• num_atoms: The total number of atoms in the system (first line).

Citation

If you decide to use this dataset, please consider citing our preprint

@misc{Thiemann2025Force-Free,
      title={Force-Free Molecular Dynamics Through Autoregressive Equivariant Networks}, 
      author={Fabian L. Thiemann and Thiago Reschützegger and Massimiliano Esposito and Tseden Taddese and Juan D. Olarte-Plata and Fausto Martelli},
      year={2025},
      eprint={2503.23794},
      archivePrefix={arXiv},
      primaryClass={physics.comp-ph},
      url={https://arxiv.org/abs/2503.23794}, 
}

Contact

For more information or if you like to contribute, please reach out to [email protected].