Machine learning molecular dynamics simulations unraveling No paddle-wheel effect in Li2B12H12 solid-state electrolyte at room temperature

Abstract

Understanding the fast ion migration mechanism in ionic conductors by the ab-initio molecular dynamics (AIMD) simulations at the atomic scale plays a crucial role in the further optimization and development of superionic conductors. Sometimes, the room temperature (RT) diffusion properties extrapolated from the elevated temperature AIMD simulation with short simulation time and limited supercell size by the linear Arrhenius assumption are problematic. The recently emerging machine learning interatomic potential models enabling the ultra-long time and ultra-large size molecular dynamics simulations and maintaining the accuracy of the density functional theory calculation have caused great attentions. Herein, we utilized the deep neural network model to develop the machine learning interatomic potential for Li₂B₁₂H₁₂ and conducted the large-scale machine learning molecular dynamics (MLMD) simulations, investigating the effects of Li vacancy concentration and temperature on [B₁₂H₁₂]^2- anion group rotation. [B₁₂H₁₂]^2- anion groups display the limited vibrational motion rather than remarkable rotation in Li₂B₁₂H₁₂ without any Li vacancy defect at RT. Moreover, our MLMD simulations demonstrate the absence of the "paddle-wheel" effect even in the Li vacancy-rich Li₂B₁₂H₁₂ at RT. Oppositely, the rotational [B₁₂H₁₂]^2- anion groups have a negative impact on Li ion diffusion. These MLMD simulation results deepen our understandings of the relationship between anion rotation and cation diffusion in solids at RT.