Linking stability with molecular geometries of perovskites and lanthanide richness using machine learning methods

Abstract

Oxide perovskite materials of type ABO3 have a wide range of technological applications, such as catalysts in solid oxide fuel cells and as light-absorbing materials in solar photovoltaics. These materials often exhibit differential structural and electrostatic properties through lanthanide or non-lanthanide derived A- and B- sites. Although, experimental and/or computational verification of these differences are often difficult. In this paper, we thus take a data-driven approach. Specifically, we run three analysis using the dataset Li, Jacobs, and Morgan [2018a] applying advanced machine learning tools to perform nonparametric regressions and also to produce data visualizations using latent factor analysis (LFA) and principal component analysis (PCA). We also implement a nonparametric feature screening step while performing our high dimensional regression analysis, ensuring robustness in our results