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 used to model the energy and forces of atoms in a system. It aims to provide high-performance training and inference for these deep equivariant models, making them more data-efficient and accurate for molecular simulations.
How it works
NequIP uses E(3)-equivariant graph neural networks to represent the geometric properties of atomic systems. It is designed as a flexible framework that allows users to build custom architectures and implement new training techniques. It integrates with common scientific software like ASE (Atomic Simulation Environment) and LAMMPS for molecular dynamics simulations.
Who it’s for
Researchers in computational chemistry, materials science, and biomolecular simulations who need accurate, data-efficient interatomic potentials for simulating atomic-scale interactions.
Highlights
- Compiled training and inference for increased speed.
- Multi-GPU training support.
- GPU kernel accelerations via OpenEquivariance and CuEquivariance.
- Integration with ASE calculator and LAMMPS pair styles.
- Extensible architecture allowing for custom extension packages like Allegro.