Causal Analysis of Factors for Li Ionic Conductivity in Olivine-Type LiMXO4 Materials Using LiNGAM

All-solid-state batteries are increasingly examined to improve the safety of lithium-ion batteries. Considerable efforts are being made to develop solid electrolyte materials with high ionic conductivity to realize such batteries. Recently, materials informatics (MI), which uses machine learning to quantify the relationships between descriptors that represent the composition, structure, and properties of materials, has been increasingly utilized. However, MI has primarily focused on evaluating the correlations between descriptors, with few examples analyzing hierarchical cause-and-effect relationships. This study analyzes the causal relationships among the descriptors of lithium-ion conductivity in olivine-type LiMXO4 materials using linear non-Gaussian acyclic model (LiNGAM), a method of causal discovery. We previously compiled a data set for these materials using first-principles calculations and performed machine learning regression analyses. A key aspect of this research is the systematization of knowledge regarding the composition, structure, and ionic conductivity of materials. Clarifying the causal relationships can lead to the discovery of factors that truly control ionic conductivity. Therefore, we qualitatively interpret the causal relationship diagram and provide specific guidelines for controlling the lithium-ion conductivity by comparing the results of previous machine learning regression analyses with those of LiNGAM.