This research uses a hierarchical machine learning approach to predict the band gap (Eg) of double perovskite oxides. Two machine learning models—a classification model to identify wide band gap materials (Eg ≥ 0.5 eV) and a regression model to predict the Eg values of those materials—are trained on a large dataset of DFT calculations. This process efficiently screens a vast chemical space of 5.2 million potential double perovskite compositions, identifying 310 high-probability candidates for experimental investigation. The models are analyzed to gain insights into band gap prediction, and design maps illustrate the band gap variation with element substitution.

Anjana Talapatra, Blas Pedro Uberuaga, Christopher Richard Stanek, Ghanshyam Pilania