Machine-learned force fields for thermal neutron scattering law evaluations

A new method is presented to use machine-learned interatomic potentials (MLPs) to generate material models for thermal neutron scattering laws (TSLs). MLPs are computationally efficient models of ab initio force fields that can be used in the creation of a vibrational spectrum as an input to TSL generation. MLP-based molecular dynamics introduces temperature effects into the vibrational spectrum, which have been neglected in most modern TSLs. Yttrium hydride ( YH x ) is used to illustrate this new MLP technique. The MLP approach is shown to predict temperature effects in the vibrational spectrum observed in experiment and improve on key features of the oscillatory scattering cross section of YH x when compared to current temperature-independent, ab initio techniques. • Machine-learned potentials method for accurate temperature-dependent phonons. • Machine-learned potential phonons improve thermal scattering law evaluations. • Yttrium hydride thermal scattering law in good agreement with experimental trends.