Machine Learning-Based Exploration of Dopants in Li7La3Zr2O12 in Reference to Lithium-Ion Conductivity

Abstract

A detailed evaluation of various parameters that influence the lithium (Li)-ion conductivity in Li₇La₃Zr₂O₁₂ is undertaken based on data from the literature. In particular, the importance of the dopant on the Li site, the ionic radius of the dopant, and the relative density of the compound are evident. The relative density can only be obtained from experimental measurements, which restrict the evaluation of unexplored dopants and their associated stoichiometry. The element embedding is utilized to generate 200D element representations that can obviate the need for hard-to-obtain descriptors. Different machine learning methods are evaluated for the prediction of superionicity of the compound for unknown dopants on the Li site and the F1 score of 0.81 using the K-nearest neighbor classifier. Based on this analysis, new dopants and associated stoichiometry are suggested.