Polarizability models for simulations of finite temperature Raman spectra from machine learning molecular dynamics

While the efficacy of machine learning (ML) force fields in simulating molecular dynamics (MD) trajectories has already been well established, simulating Raman spectra from them requires polarizability models which are much less explored. In this work, three polarizability models are compared using three widely different materials, namely boron arsenide, 2D molybdenum disulfide and inorganic halide perovskites. The Raman spectra are obtained in combination with ML MD and compared to experiments, allowing us to highlight the advantages and shortcomings of each model.