Molecular dynamics study of local structure and migration properties of LiCl-Li2O-Li molten salts based on machine-learned deep potential

The local structure and physical properties of LiCl-Li₂O-Li molten salt, the reaction medium for lithium thermal and electrolytic reduction, are very important for the study of spent fuel pyroprocessing process. In this work, the machine-learned deep potential (MLDP) was trained using dataset based on first-principle molecular dynamics (FPMD) and was used to predict the changes in the physical properties of molten LiCl with the addition of different concentrations of Li₂O and Li between 923 K and 1323 K. Deep potential molecular dynamics (DPMD) calculations were performed for properties including shear viscosity, electrical conductivity, thermal conductivity, and specific heat capacity. It was revealed that the addition of Li significantly reduces the diffusion activation energies ($E_a$) of Li⁺ and Cl^- in the molten salt. By comparison with the experimental data of pure LiCl, it can be concluded that the MLDP can describe the inter-atomic interactions of molten salt correctly, overcome the problem of missing potential parameters in the classical inter-atomic empirical potentials. Finally, DPMD allows to simulate large systems with comparable accuracy of FPMD, thus provide theoretical guidance for the optimization of the pyroprocessing technology.