lam-group / plip

Linear Machine learning Interatomic Potential for atomistic simulations

PLIP (Physical LassoLars Interactions Potential) is a hybrid python/C++ package for machine learning interatomic potentials for materials modelling. Physical Lasso-Lars Interaction Potential (PLIP) has been developed to address the lack of interpretability of MLIPs. PLIP uses a physically motivated mathematical formulation for the potential and a constrained linear regression. The MLIP is constructed as a linear combination of classical interactions, including two-body, three-body, and many-body interactions. These interactions are made up of simple functions such as Lennard-Jones, Gaussian, and Slatter-type orbitals, as well as explicit angular functions and embedded atom-like interactions.

Documentation

The full documentation is available at plip.readthedocs.io.

Tutorial

Colab Python notebook tutorial: https://colab.research.google.com/github/LAM-GROUP/PLIP/blob/main/PLIPwithLAMMPS.ipynb

Installation

git clone https://github.com/LAM-GROUP/PLIP.git
cd  PLIP
pip install -e .

Dependencies

• Python 3.8+
• Support for C++17
• Compiler gcc 6.0 and above
• Python dependencies: see pyproject.toml
• LAMMPS

Authors

The PLIP package was developed by Julien Lam and Akshay Krishna AK

License

The PLIP package is distributed under terms of GNU GPLv3.

plip's Issues

Coordinates

The code needs the input files to be in cartesian coordinates (poscar). Giving reduced coordinates (direct keyword in poscar) will most likely give bad results. Could be fixed by either doing conversion if necessary or simply raise an error if the inputs are wrong.

JOSS Paper equations are missing some definitions

Hello! I've been tapped as a reviewer for your recent submission to JOSS!
I've noticed that the PDF proof linked at the JOSS review issue thread doesn't contain definitions of all the symbols in the equations at the top of page 2:

I think the paper should include definitions of all the symbols used in these equations, as you have in your docs. That is, it's not clear what $f_c$, $f_s$, or $f_n$ are in the paper alone.

Species

The fact atomic species are hardcoded in mod_XYZ.cpp can easily lead to some issues. Also, the code is not capable of reordering the species (ie if atom A is labeled 1 and atom B is labeled 2 then all poscar files given as input must have atom A first and atom B second.

Bash commands

There is a bash command in doXMT.ccp that is system dependant (line 62). "ls -l" can return different number of strings per line depending on the distribution/parameters of the system.

Please add community guidelines

In connection with your ongoing JOSS review, please add a section/subsection to your docs regarding community guidelines (contributing, reporting issues, and seeking support).