Ultra-Fast Force Fields (UF3)

Tests

S. R. Xie, M. Rupp, and R. G. Hennig, “Ultra-fast interpretable machine-learning potentials”, preprint arXiv:2110.00624 (2021).

All-atom dynamics simulations have become an indispensable quantitative tool in physics, chemistry, and materials science, but large systems and long simulation times remain challenging due to the trade-off between computational efficiency and predictive accuracy. The UF3 framework is built to address this challenge by combinining effective two- and three-body potentials in a cubic B-spline basis with regularized linear regression to obtain machine-learning potentials that are physically interpretable, sufficiently accurate for applications, and as fast as the fastest traditional empirical potentials.

Documentation: https://uf3.readthedocs.io/

This repository is still under construction. Please feel free to open new issues for feature requests and bug reports.

Setup

conda create --name uf3_env python=3.7
conda activate uf3_env
git clone https://github.com/uf3/uf3.git
cd uf3
pip install wheel
pip install -r requirements.txt
pip install numba
pip install -e .

Getting Started

Please see the examples in uf3/examples/tungsten_extxyz for basic usage.

Overviews for individual modules can be found in uf3/examples/modules (WIP).

Standalone scripts and configuration generators/parsers are in development.

Optional Dependencies

Elastic constants:

pip install setuptools_scm
pip install "elastic>=5.1.0.17"

Phonon spectra:

pip install spglib
pip install seekpath
pip install "phonopy>=2.6.0"

LAMMPS interface:

conda install numpy==1.20.3 --force-reinstall
conda install -c conda-forge lammps --no-update-deps

Dependencies

  • We rely on ase to handle parsing outputs from atomistic codes like LAMMPS, VASP, and C2PK.

  • We use Pandas to keep track of atomic configurations and their energies/forces as well as organizing data for featurization and training.

  • B-spline evaluations use scipy, numba, and ndsplines.

  • PyTables is used for reading/writing HDF5 files.