Correction to Density Functional Theory Calculations and Machine Learning Interatomic Potentials for Molten Salts to Achieve Experimental Accuracy

Abstract

Despite the considerable success of density functional theory (DFT) in a broad class of materials, there are no exchange–correlation functionals or dispersion corrections that can systematically achieve high accuracy in molten salt simulations; for example, the density is often significantly underestimated. This study proposes a method to construct a correction potential that can fill the difference between DFT and experiments, using KCl as a test case. First, a machine learning interatomic potential (MLIP) with DFT accuracy was constructed. Subsequently, a correction potential was prepared to remove residual stresses brought by the MLIP at experimental densities. It was found that a small cation–anion pairwise correction potential is sufficient to significantly improve not only the density but also other material properties, suppressing calculation errors to a level comparable to the deviation of experimental data. This method is versatile and is expected to help realize experimental accuracy in molten salt simulations.