Vibrationally Accurate Interatomic Potentials

Thermal transport properties in covalent solids are largely dictated by collective atomic vibrations known as phonons. Understanding the underlying phenomena, or vibrations, giving rise to thermal transport is the first step in designing properties, but currently no general phonon transport theory exists for all systems. Molecular dynamics (MD) simulations, however, provide a way to generally study phonon transport in any system. The heart of MD simulations is the mathematical representation of potential energy between atoms, termed the interatomic potential, from which the forces and dynamics are calculated. The use of MD simulations to generally predict and describe thermal transport has not been fully realized due to the lack of accurate interatomic potentials for a variety of systems, and obtaining accurate interatomic potentials is not a trivial task. Furthermore, it is not clear in the heat transfer community how to make potentials that are guaranteed to accurately predict phonon properties; this is the main roadblock hindering progress in the atomistic study of heat transfer. This work solves the problem of making potentials optimized for the study of phonon transport, which is accomplished by training potentials to mimic ab initio quantities such as forces. The method uses a genetic algorithm to directly fit the empirical parameters of the potential to the key properties that determine whether the atomic level dynamics and most notably the phonon transport is properly described.