Validation of machine-learned interatomic potentials via temperature-dependent electron thermal diffuse scattering

Abstract

Machine-learned interatomic potentials (MLIPs) show promise in accurately describing the physical properties of materials, but there is a need for a higher throughput method of validation. Here, we demonstrate using that MLIPs and molecular dynamics can accurately capture the potential energy landscape and lattice dynamics that are needed to describe electron thermal diffuse scattering. Using SrTiO$_3$ as a test-bed at cryogenic and room temperatures, we compare electron thermal diffuse scattering simulations using different approximations to incorporate thermal motion. Only when the simulations are based on quantum mechanically accurate MLIPs in combination with path-integral molecular dynamics that include nuclear quantum effects, there is excellent agreement with experiment\end{abstract}

Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.