Improving ionic conductivity of doped Li7La3Zr2O12 using optimized machine learning with simplistic descriptors

Abstract The dawn of machine learning methods brings a possible solution to efficiently get through the vast design space of doped Li7La3Zr2O12 (LLZO) solid-state electrolytes. In this work, a machine learning model to classify the ionic conductivity of doped LLZO is developed using features derived from molecular, structural, and electronic descriptors. Meticulous model selection, validation, and optimization yielded a classifier based on the Light Gradient Boosting Machine algorithm with a leave-one-out cross-validation accuracy score of 0.903. Two key aspects were identified to obtain doped LLZO with high ionic conductivity, namely electrolyte's relative density and Li site dopant’s electronegativity. This study illustrates the role of powerful data-driven methods with easily obtainable features in accelerating the process of novel solid-state electrolyte design.