Predicting crystal structures and ionic conductivity in ternary halide solid electrolytes using machine learning interatomic potentials

Machine learning interatomic potentials (MLIPs) have transformed computational materials modeling by extending ab initio accuracy to much larger length and time scales. This capability enables realistic simulations of complex systems such as solid-state battery (SSB) components. The demand for safer, higher-performance Li-ion batteries drives the exploration of halide solid electrolytes (SEs), particularly Li₃MX₆ (M = trivalent metal, X = Cl, Br), which combine high ionic conductviity with excellent electrochemical stability. However, their diverse crystal symmetries and partially occuped sites in experimentally refined structures yield tens of thousands of symmetry-inequivalent atomic configurations, presenting major challenges for conventional modeling. Here, we employ MLIP-based simulations integrated with active-learning workflows to efficiently map composition-structure-property relationships in halide SEs for SSBs. The approach enables rapid estimation of formation energies, ranking of inequivalent configurations, and molecular dynamics simulations of ionic transport. Our results reveal how targeted cation and anion substitutions stabilize specific crystal symmetries and enhance Li-ion conductivity, providing design principles for next-generation SEs.