Skip to content

Overview

mlx-atomistic is an experimental Apple Silicon-native atomistic simulation runtime built on MLX and Metal. It targets Python 3.13 through uv and uses MLX for local GPU execution on Apple Silicon.

mlx_atomistic is the primary trajectory generator and product runtime in this repo. OpenMM, LAMMPS, and the source trees under vendors/ are reference or validation surfaces; they do not replace the MLX runtime path.

  • Density Functional Theory — a legacy Γ-point teaching/reference path, plus a separate periodic PBE-GTH path with Monkhorst-Pack SCF, block-Davidson solves, frozen-density bands, and periodic forces. Verified claims stay limited to the published silicon, carbon, and MgO workloads.
  • Molecular Mechanics — Lennard-Jones, Coulomb, harmonic bonds/angles, periodic + Ryckaert–Bellemans torsions, bounded PME, NVE and Langevin NVT.

Lightweight DFT building blocks, validation notebooks, and visualization utilities rather than a heavy production DFT engine. Small, validated examples before broader chemistry coverage.

Terminal window
uv run --no-project --python 3.13 --with mlx-atomistic \
python -c "import mlx_atomistic as ma; print(ma.__version__)"

Extras are opt-in for checkout workflows: prep for topology/prep imports, viz for visualization, and notebook for Jupyter.

Terminal window
uv venv --python 3.13
uv sync --extra notebook --extra prep --extra viz
uv run python -m ipykernel install --user --name mlx-atomistic --display-name "mlx-atomistic"
uv run jupyter lab

If uv cannot use the home cache in a sandboxed run, use a writable cache:

Terminal window
UV_CACHE_DIR=/tmp/mlx-atomistic-uv-cache uv sync --extra notebook --extra prep --extra viz