Skip to content

kups.potential.mliap

Machine learning interatomic potentials (MLIAPs).

This module provides interfaces to machine learning models for computing atomic energies and forces. MLIAPs offer quantum-mechanical accuracy at classical force field computational cost, enabling accurate simulations of complex systems.

Available Models

  • tojax: Generic jaxified MLFF models (exported JAX)
  • local: Local MLIAP with single message passing and incremental updates
  • torch: PyTorch MLFF models (MACE, UMA) via TorchModuleWrapper
  • direct: Direct-gradient MLIAP potential factory (make_direct_mliap_potential) — used by the torch bridge

MLIAPs are trained on ab initio data and can capture complex many-body interactions, bond breaking/forming, and reactive chemistry that classical force fields cannot.