Beyond Classical Molecular Dynamics with Layered Interatomic Potentials

Any prediction that classical molecular dynamics simulations offer is, strictly speaking, a function of the specific Born-Oppenheimer potential used. These are reduced order models of the full electronic-ionic Hamiltonian and leave out important electronic effects that modify the potential energy surface that ions evolve on for the purpose of computational efficiency. Adaptations to the training protocol of machine learned interatomic potentials offer a unique solution to recover these lost degrees of freedom. Sommerfeld (temperature-dependent) potentials are generated that accurately capture high energy neutron damage events in plasma facing materials, a simulation capability that is only possible if efficient interatomic potentials are available. It is found that inclusion of the temperature-dependent potentials yields less peak and residual damage in Tungsten which is attributed to the inelastic scattering of ions that these new potentials capture.