ACEsuit is a collection of Julia packages built around the Atomic Cluster Expansion (ACE) Model, an efficient parameterisation scheme for many-body interactions in particle systems. ACE models are defined in terms of body-ordered polynomial invariant features of (possibly decorated) point clouds. This github org collects a wide variety of Julia and Python packages that build in some way on ACE methodology.

ACE Potentials in Julia

ACEpotentials.jl provides a user-oriented interface to a collection of several Julia packages that interoperate to fit models for atomic cluster expansion (ACE) interatomic potentials.


MACE provides fast and accurate machine learning interatomic potentials with higher order equivariant message passing. Each message uses a variant of the ACE expansion to make the higher body order computationally efficient. The mace package is implemented in PyTorch.

Under Development

  • ACE.jl github repo : an extension of ACE1.jl (on which ACEpotentials.jl builds) to allow equivariant features

  • ACEatoms.jl - experimental interatomic potentials via ACE.jl

  • ACEds.jl - coarse-grained dynamical systems

  • ACEhamiltonians.jl : Hamiltonian operators with ACE, e.g., for tight binding

  • ACEpsi.jl : wave functions with ACE and VMC

  • Polynomials4ML.jl : new high-performance kernels for ACE models

  • EquivariantModels.jl : rewriting of ACE model architectures using Lux.jl