mir-group/nequip

NequIP is a code for building E(3)-equivariant interatomic potentials

What it solves

NequIP provides a framework for building E(3)-equivariant interatomic potentials, which are machine learning models that predict the energy and forces of atoms in a system. It addresses the need for data-efficient and accurate potentials that respect the physical symmetries of 3D space (rotation, translation, and inversion).

How it works

The project implements E(3)-equivariant graph neural networks to model interatomic interactions. It offers a family of foundation potentials—pre-trained models covering most of the periodic table—that can be used directly or fine-tuned on specific datasets. The framework is designed to be extensible, allowing users to build custom architectures (such as the strictly local architecture implemented in the Allegro package) or implement new training techniques.

Who it’s for

It is designed for researchers and scientists in computational chemistry, materials science, and biomolecular simulations who need high-performance interatomic potentials for production simulations.

Highlights

  • Foundation Potentials: Pre-trained, wide-purpose models covering most of the periodic table.
  • High Performance: Supports compiled training and inference, multi-GPU training, and GPU kernel accelerations via OpenEquivariance and CuEquivariance.
  • Integration: Seamlessly integrates with the Atomic Simulation Environment (ASE) and LAMMPS.
  • Extensible Architecture: Flexible framework allowing for the development of extension packages and custom architectures.

Related

  • Project
    libAtoms/QUIP

    A collection of software tools for molecular dynamics simulations that integrates various interatomic potentials and tight-binding quantum mechanics, including machine-learning-based GAP potentials.

  • Project
    ACEsuit/mace

    A machine learning framework for creating fast and accurate interatomic potentials using higher-order equivariant message passing neural networks.

  • Project
    deepmodeling/deepmd-kit

    A deep learning package that turns quantum-mechanical reference data into fast, scalable interatomic potentials for molecular and materials science simulations.

  • Project
    materialyzeai/matgl

    A graph deep learning library for materials science that provides various GNN architectures to predict material properties and simulate potential energy surfaces.

  • Project
    brucefan1983/GPUMD

    A high-performance GPU-accelerated molecular dynamics package that supports empirical and neuroevolution machine-learned potentials for materials simulation.