Comprehensive Molecular Representation from Equivariant Transformer

CMRET provides spin (S) and/or molecular charge (Q) awared machine learning force fields. Note this is still an experimental project.

Requirements

torch>=2.0.0
ase>=3.22.0
lightning>=2.1.2

Usage

See examples:

train and test on singlet/triplet CH₂ dataset

train and test on ISO17 dataset

running molecular dynamic simulation

Known issue

• Using a compiled TorchScript Module for MD simulation (ASE-based) will lead to RuntimeError.

Cite

@misc{2023cmret,
      title={Comprehensive Molecular Representation from Equivariant Transformer}, 
      author={Nianze Tao and Hiromi Morimoto and Stefano Leoni},
      year={2023},
      eprint={2308.10752},
      archivePrefix={arXiv},
      primaryClass={physics.comp-ph}
}