The study of ionic conductivity of the Li10GeP2S12 type Solid State Electrolyte by an extrapolation method and a deep-learning method

Abstract

An extrapolation method is usually applied when Ab initio molecular dynamics (AIMD) is applied to studying ionic conductivity in solid-state electrolytes in lithium-ion batteries. As the ions move slowly in solid-state electrolytes, the first-principles method typically involves computationally intensive calculations, and it can take significant time to obtain accurate results. First-principles method is too expensive for the time scale required at room temperature. The classical molecular dynamics method is typically applied to systems containing thousands of atoms and can simulate the system’s behavior over nanoseconds. During the simulation, the positions and velocities of the atoms are updated at discrete time intervals, allowing the system’s behavior to be studied over time. However, its accuracy depends on the empirical force-field libraries. Limited by the computational resource, the previous studies applied the extrapolation method to obtain the room temperature ionic conductivity, which was not accurate because the linear relationship in the Arrhenius equation was not valid in a wide range of temperatures. Deepmd-Kit is a tool that integrates these two different computational approaches. The extrapolation and Deepmd methods were applied to the materials Li10GeP2S12, Li10SiP2S12, Li10GePS12Cl, and Li10SiPS12Cl, respectively. Both methods showed that the lithium ions favor the c direction when diffusing in the LGPS-type solid-state electrolytes. The ionic conductivity is more accurate with the dependent method compared with experiments.