A minimal molecular dynamics driver using machine-learning interatomic potentials (MLIPs) via the ASE framework.

minimd.py runs a two-stage simulation on an input atomic structure:

1. Geometry relaxation — FIRE optimizer minimizes forces until convergence (fmax)
2. NVT thermalization — Langevin dynamics at a target temperature for a set number of steps

Trajectories are saved as both .traj (ASE binary) and .xyz files under a subdirectory named after the chosen model.

python minimd.py [-c CONFIG] [-m MODEL] [-f FMAX] [-s STEPS] [-t TEMPERATURE]

python minimd.py -c sio2.xyz -m mace -f 0.05 -s 5000 -t 1000

All output is written to <model>/:

• Python >= 3.11
• ASE >= 3.28
• PyTorch >= 2.11
• MACE-torch >= 0.3.15 (required for mace model; others are optional)

GPU acceleration is used automatically when CUDA is available.

This project uses a uv lockfile (uv.lock) to pin Python dependencies. Recommended steps to create a reproducible environment:

1. Ensure Python 3.11+ and pip are installed.

2. Install the uv CLI if not present:

   pip install uv

3. Sync the environment from the lockfile (creates and installs into a virtual environment):

   uv sync

4. Activate the created venv (if uv places it at .venv):

   • macOS / Linux: source .venv/bin/activate
   • Windows (PowerShell): ..venv\Scripts\Activate.ps1

If uv creates the environment at another path, follow uv's output to activate it. For GPU support, install the appropriate CUDA build of PyTorch following https://pytorch.org/ after syncing.

• MACE

  • Repository: https://github.com/ACEsuit/mace
  • Paper (NeurIPS 2022): https://openreview.net/forum?id=YPpSngE-ZU
  • Foundation model paper (JCP 2025): https://pubs.aip.org/aip/jcp/article/163/18/184110/3372267/A-foundation-model-for-atomistic-materials
  • Code / release DOI: https://doi.org/10.5281/zenodo.14103332

• NequIP

  • Repository: https://github.com/mir-group/nequip
  • Paper (Nature Communications 2022): https://doi.org/10.1038/s41467-022-29939-5
  • Code / release DOI: https://doi.org/10.5281/zenodo.18200066
  • Docs: https://nequip.readthedocs.io/

• Allegro

  • Repository: https://github.com/mir-group/allegro
  • Paper (Nature Communications 2023): https://doi.org/10.1038/s41467-023-36329-y
  • Docs: https://allegro.readthedocs.io/

• MatterSim

  • Project page: https://microsoft.github.io/mattersim/

• UPET (PET)

  • Paper (Nature Communications 2025): https://www.nature.com/articles/s41467-025-65662-7

• UMA

  • Publication: https://ai.meta.com/research/publications/uma-a-family-of-universal-models-for-atoms/

• SevenNet

  • Canonical project page / publication DOI: TBD