Machine learning molecular dynamics simulation identifying weakly negative effect of polyanion rotation on Li-ion migration

Abstract

Abstract Understanding the physical picture of Li ion transport in the current ionic conductors is quite essential to further develop lithium superionic conductors for solid-state batteries. The traditional practice of directly extrapolating room temperature ion diffusion properties from the high-temperature (>600 K) ab initio molecular dynamics simulations (AIMD) simulations by the Arrhenius assumption unavoidably cause some deviations. Fortunately, the ultralong-time molecular dynamics simulation based on the machine-learning interatomic potentials (MLMD) is a more suitable tool to probe into ion diffusion events at low temperatures and simultaneously keeps the accuracy at the density functional theory level. Herein, by the low-temperature MLMD simulations, the non-linear Arrhenius behavior of Li ion was found for Li3ErCl6, which is the main reason for the traditional AIMD simulation overestimating its ionic conductivity. The 1μs MLMD simulations capture polyanion rotation events in Li7P3S11 at room temperature, in which four [PS4]3− tetrahedra belonging to a part of the longer-chain [P2S7]4− group are noticed with remarkable rotational motions, while the isolated group [PS4]3− does not rotate. However, no polyanion rotation is observed in Li10GeP2S12, β-Li3PS4, Li3ErCl6, and Li3YBr6 at 300 K during 1μs simulation time. Additionally, the ultralong-time MLMD simulations demonstrate that not only there is no paddle-wheel effect in the crystalline Li7P3S11 at room temperature, but also the rotational [PS4]3− polyanion groups have weakly negative impacts on the overall Li ion diffusion. The ultralong-time MLMD simulations deepen our understanding of the relationship between the polyanion rotation and cation diffusion in ionic conductors at room environments.