Unlocking Enhanced Ionic Transport: A Machine Learning-Driven AIMD Study on Doping, Defects, and Strain in Antiperovskite Solid-State Electrolytes
C. X. Lin, Zhang Lin, Yi Dong
Abstract
Ab initio molecular dynamics (AIMD) simulations have been employed to investigate doped antiperovskite solid-state electrolyte (AP SSE) structures, specifically formulated as Li 3 OCl x Br 1– x ( x = 0, 0.25, 0.50, 0.75, 1). Various defects, including lithium vacancies, interstitials, as well as Schottky and Frenkel defects, were analyzed under elastic biaxial strain to simulate practical conditions. Machine learning (ML) was then applied to the data from AIMD simulations, targeting lithium diffusivity and ionic conductivity. Our results show that the most significant enhancement in lithium diffusivity occurs when the Cl/Br ratio is 0.5/0.5, and the defect type is the double lithium ion interstitial. Lithium diffusivity and conductivity are mainly governed by the vibration amplitude and number of lithium ions. Although biaxial strain offers a slight promotion effect, it is less influential compared to doping and defect structures. Additionally, SHAP analysis was conducted to assess the relative importance and interactions of each feature descriptor, offering novel insights into the design strategies for enhanced AP SSEs.