FLARE: Fast Learning of Atomistic Rare Events

FLARE is an open-source Python package for creating fast and accurate atomistic potentials. Documentation of the code can be accessed here: https://flare.readthedocs.io/

Major Features

• Gaussian Process Force Fields

  • 2- and 3-body multi-element kernels
  • Maximum likelihood hyperparameter optimization

• On-the-Fly Training

  • Coupling to Quantum Espresso, CP2K, and VASP DFT engines

• Mapped Gaussian Processes

  • Mapping to efficient cubic spline models

• ASE Interface

  • ASE calculator for GP models
  • On-the-fly training with ASE MD engines

• Module for training GPs from AIMD trajectories

Prerequisites

1. To train a potential on the fly, you need a working installation of Quantum ESPRESSO or CP2K.
2. FLARE requires Python 3 with the packages specified in requirements.txt. This is taken care of by pip.

Installation

FLARE can be installed in two different ways.

1. Download and install automatically:

   pip install mir-flare
   

2. Download this repository and install (required for unit tests):

   git clone https://github.com/mir-group/flare
   cd flare
   pip install .
   

Tests

We recommend running unit tests to confirm that FLARE is running properly on your machine. We have implemented our tests using the pytest suite. You can call pytest from the command line in the tests directory to validate that Quantum ESPRESSO or CP2K are working correctly with FLARE.

Instructions (either DFT package will suffice):

pip install pytest
cd tests
PWSCF_COMMAND=/path/to/pw.x CP2K_COMMAND=/path/to/cp2k pytest

References

If you use FLARE in your research, or any part of this repo (such as the GP implementation), please cite the following paper:

[1] Jonathan Vandermause, Steven B. Torrisi, Simon Batzner, Yu Xie, Lixin Sun, Alexie M. Kolpak, and Boris Kozinsky. On-the-fly active learning of interpretable Bayesian force fields for atomistic rare events. https://arxiv.org/abs/1904.02042