Insights into the local structure evolution and thermophysical properties of NaCl–KCl–MgCl2–LaCl3 melt driven by machine learning

Abstract

The acceleration of the development and breakthrough of molten salt electrolytic preparation for Mg–La alloys can be facilitated by gaining insights into the local structure evolution and thermophysical properties of NaCl–KCl–MgCl₂–LaCl₃ (NKML) melt. Herein, we developed interatomic potentials for the NKML melt using a concurrent learning strategy. Notably, this is the first time that machine learning methods have been employed for this purpose. The performance of the deep potential (DP) model was assessed by calculating the root mean square errors of energy and force. The maximum root mean square error observed for energy was 1.14 meV per atom, while for force, it was 41.48 meV Å⁻¹. These results indicate that a well-trained DP model is capable of accurately representing the potential energy surface of the NKML system. The local structure of NKML in short-range and intermediate-range order was predicted using DP model-driven molecular dynamics (DPMD) simulations. The evolution pattern of the NKML local structure was analyzed using various techniques, including the partial radial distribution function, potential of mean force, coordination number distribution, angular distribution function, and partial structure factor. A comprehensive analysis was conducted on the thermophysical properties that play a crucial role in the electrolysis process. These properties include density, self-diffusion coefficient, shear viscosity, and ionic conductivity. The analysis focused on their dependence on temperature and MgCl₂ concentration.