Atomic interactions

We develop atomic interactions based on the Atomic Cluster Expansion (ACE). ACE is a leading method for accurate and efficient parameterization of atomic interactions and can be employed in large-scale atomistic simulations on parallel CPU and GPU architectures. ACE models are trained using advanced machine learning algorithms to reproduce energies and forces of electronic structure calculations. Active learning capabilities facilitate automated construction of training datasets and provide uncertainty indication during production runs. Since ACE is general, atomic interaction models are available for metallic, covalent, ionic and mixed types of chemical bonding as well as inter-molecular interactions.