Molecular dynamics simulation for phonon angular momentum Hall effect via ab initio-based machine learning force field

Abstract

Simulating the first-principles-based molecular dynamics encounters limitations in capturing the movements of thousands of atoms within 100 picoseconds. Machine learning is employed to extract fields based on atomic data obtained from first-principles molecular dynamics to extend molecular dynamics simulations over longer timeframes. Subsequently, classical molecular dynamics simulations are conducted using the extracted fields to analyze the dynamics of molecules on a large scale. Building upon this, the investigation focuses on the phonon angular momentum Hall effect. This effect refers to the perpendicular accumulation of heat in the direction of temperature change when atoms possess angular momentum. Analogous to the phonon thermal Hall effect, where the topological property Chern number, breaking the time-reversal symmetry of the system, determines the thermal conductivity, it is hypothesized that the phonon angular momentum Hall effect is also related to the Chern number of the molecule. Therefore, the aim is to confirm whether the phonon angular momentum Hall effect is influenced by the molecule's Chern number by comparing the results for BeAu with a Chern number and Au without a Chern number.