Neural Network Potentials

This repository summarizes popular neural network potentials (NNPs) that serve as efficient replacements for ab initio calculations.

Note: This repository collects and summarizes the NNP models we have used. For the most up-to-date information, please refer to each model’s specific repository.

• Webpage: https://ishikawa-group.github.io/neural_network_potentials/

List of NNP Models

1. M3GNet
   • A versatile interatomic potential leveraging graph neural network architectures.
2. CHGNet
   • A universal neural network potential designed for accurate energy and force predictions.
3. MatterSim
   • A deep learning atomistic model for simulating materials across different elements, temperatures, and pressures.
4. SevenNet
   • A scalable, equivariance-enabled neural network for efficient parallel molecular dynamics simulations using LAMMPS.
5. MACE
   • A fast and accurate machine learning interatomic potential based on higher-order equivariant message passing.
6. ORB
   • Pretrained neural network potentials for atomic simulations, optimized for scalability and speed.
7. OpenCatalystProject (Fairchem)
   • A framework focused on generating and using catalyst reaction potentials with NNPs.

Other Model Training Approaches

DeepMD-kit is a package written in Python/C++, designed to minimize the effort required to build deep learning-based models of interatomic potential energy and force field and to perform molecular dynamics simulations.

• DeepMD-kit

  • A toolkit for molecular dynamics simulations using deep neural network potentials.