Accelerated lattice constant prediction of perovskite materials (ABX3, A2BB′O6) using partial least squares and principal component regression methods

Abstract

Since device performance is greatly influenced by minute variations in the crystal structure of its constituent materials, reliably predicting the lattice parameters of perovskites can aid in minimizing structural mismatch and improve film integrity. Herein, we present the application of partial least squares regression (PLSR) and principal component regression (PCR) methods to predict the lattice constant of ABX3 and A2BB′O6 perovskites, with Shannon ionic radii (r), atomic number (Z), electronegativity (χ) and density (ρ) as model inputs. Our results revealed that, depending on the number of components used, PLSR and PCR estimates had better mean absolute percentage error than estimates from existing studies. Moreover, computed loadings can be used to describe the effect of each model input to the regression estimate. Overall, this study paves the way for rapid and simple elemental property-based perovskite material design and discovery.