Exploring lithium diffusion in LiF with machine learning potentials: from point defects to collective ring diffusion

Lithium fluoride (LiF) is a fundamental inorganic component of the solid electrolyte interphase (SEI) in lithium-ion batteries, yet its intrinsic ionic transport properties remain poorly understood, since there is a lack of direct experimental data, and standard computational approaches have difficulties in modeling these phenomena. Here, we train a machine learning based interatomic potential (MLIP) to perform large-scale Molecular Dynamics simulations of lithium diffusion in bulk LiF with near-DFT accuracy. Our study reveals not only conventional vacancy and interstitial transport mechanisms, but also a novel collective ring diffusion process involving six lithium ions. This mechanism arises under high interstitial concentrations, and it is stabilized by partial charge delocalization. Given the dynamic evolution of the SEI under electrochemical cycling, we suggest that such correlated diffusion events may be transiently active in LiF-rich regions. More broadly, our results demonstrate the power of MLIPs to access long timescales and complex dynamical behavior, providing a robust framework for future multiscale modeling of the SEI.