Physics-Informed Gaussian Process Inference of Liquid Structure from Scattering Data
Abstract
Here we present a nonparametric Bayesian framework to infer radial distribution functions with uncertainty quantification from experimental scattering measurements using non-stationary Gaussian processes. The Gaussian process prior mean and kernel functions are designed to resolve well-known problems with the Fourier transform of scattering data, including discrete measurement binning and detector windowing, while encoding fundamental yet minimal physical knowledge of atomistic representations of liquid structure. We apply the methodology to liquid Ar and the site-site partial structure factors of liquid water to provide radial distribution functions with rigorous uncertainty quantification that can serve as a vital benchmark for molecular modeling.
