Deep-Learning-Driven Prediction Strategy for the Phase Transition Behavior of Alkali Chloride XCl (X = Li, Na, or K).

In this study, we systematically investigated the phase transition characteristics of alkali metal chloride salts through a deep potential molecular dynamics (DPMD) strategy. The melting points are determined using the superheating-supercooling hysteresis method, while the phase transition behavior is comprehensively analyzed through the radial distribution function (RDF), mean-square displacement (MSD), self-diffusion coefficient (D), coordination number (CN), and molecular dynamics trajectory. The simulation captures the abrupt changes in the local environment of anions and cations during the melting of a molten salt. Molten salts do not complete phase transitions near the actual melting point. Instead, they undergo long-range ordered and long-range disordered transitions near the superheating and supercooling temperature, respectively. MSD, D, and CN analysis quantitatively demonstrate the hysteresis phenomenon during thermal cycling, accompanied by the recombination of ion coordination structures during the phase transition. This work reveals the solid-liquid phase transformation of molten salt and establishes a reliable computational framework for the high-precision prediction of thermal properties.