Development of robust neural-network interatomic potential for molten salt

Abstract

Molten salts are a promising class of ionic liquids for clean energy applications, such as nuclear and solar energy. However, efficient and accurate evaluation of salt properties from a fundamental, microscopic perspective remains a challenge. Here, we apply artificial neural networks to atomistic modeling of molten NaCl to accurately reproduce the properties from ab initio quantum mechanical calculations based on density functional theory (DFT). The obtained neural network interatomic potential (NNIP) effectively captures the effects of both long-range and short-range interactions, which are crucial for modeling ionic liquids. Extensive validations suggest that the NNIP is capable of predicting the structural, thermophysical, and transport properties of molten NaCl as well as properties of crystalline NaCl, demonstrating near-DFT accuracy and 103× higher efficiency in atomistic simulations. This application of NNIP suggests a paradigm shift from empirical/semiempirical/ab initio approaches to an efficient and accurate machine learning scheme in molten salt modeling.