Progress Towards a Quantum-Accurate Classical SNAP ML Interaction Potential for Gold

Classical molecular dynamics is a powerful method of theoretically describing the properties of a material but is often inconsistent with quantum molecular dynamics. Quantum molecular dynamics while more accurate are much more computationally intensive and not feasible for systems of more than 1000 atoms. Machine learning is used to bridge the gap between quantum accuracy and the scalability of classical interatomic potentials. Here, we report on progress made testing the performance of a quantum-accurate classical interatomic potential for gold against state-of-the-art classical potentials in describing experimentally verified properties of gold using LAMMPS (Large Scale Atomic/Molecular Massively Parallel Simulator) Preliminary results show close agreement with quantum molecular dynamics simulations, while being computationally less expensive. Future work includes extending this study to simulate rare events in a large, such as the herringbone reconstruction of the Au(111) surface, that are not accurately described by current classical potentials.