The atomic cluster expansion (ACE) establishes a complete and orthonormal basis for the space of local atomic configurations. This enables one to recast many other interatomic potentials – classical as well as machine learning – in the form of ACE. The completeness of ACE further means that models of the atomic interaction can be converged systematically. ACE can be extended to include charge tra nsfer and magnetism, representation of tensorial properties and message passing networks. Relations to electronic structure calculations exist. Robust and automated software for the rapid creation of ACE models is available, which together with efficient i mplementations of ACE for CPU and GPU hardware enable large scale atomistic simulations. In my talk I will summarize ACE theory, briefly discuss its implementation and parameterization and show applications to the modeling of molecules and materials.
The atomic cluster expansion (ACE) establishes a complete and orthonormal basis for the space of local atomic configurations. This enables one to recast many other interatomic potentials – classical as well as machine learning – in the form of ACE. The completeness of ACE further means that models of the atomic interaction can be converged systematically. ACE can be extended to include charge tra nsfer and magnetism, representation of tensorial properties and message passing networks. Relations to electronic structure calculations exist. Robust and automated software for the rapid creation of ACE models is available, which together with efficient i mplementations of ACE for CPU and GPU hardware enable large scale atomistic simulations. In my talk I will summarize ACE theory, briefly discuss its implementation and parameterization and show applications to the modeling of molecules and materials.
