Machine learning-guided discovery of stable low-work-function perovskite oxides for advanced catalysis and energy technologies

Perovskites with a low work function find extensive applications across various fields. However, the diverse nature of perovskite compounds presents challenges for conventional trial-and-error material screening methods. In this study, we introduced an effective target-driven approach that integrates machine learning (ML) and density functional theory (DFT) calculations. Initially, we explored an exhaustive chemical space encompassing ABO₃-type single and A₂BB'O₆-type double perovskite oxides to identify stable compounds with work functions of AO-terminated (001) surfaces below 2.5 eV via a trained ML model. By employing high-precision calculations, we subsequently narrowed the selection to 27 stable perovskite oxides from the initial pool of 23,822 candidate materials. Two promising compounds, Ba₂TiWO₈ and Ba₂FeMoO₆, were then successfully synthesized and characterized experimentally. Furthermore, the first synthesized Ba₂TiWO₈ was found to exhibit catalytic activities for both NH₃ synthesis and NH₃ decomposition under mild conditions with Ru loading, suggesting its future application in catalysis. Moreover, as a Li-ion battery electrode material, Ba₂FeMoO₆ exhibited long-term cycling stability at a current density of 10 A∙g⁻¹ (10,000 cycles), revealing many possibilities for sustainable electrochemical applications of perovskites. Our work demonstrates the efficacy and efficiency of the ML-assisted method in establishing a reliable structure–property relationship for mapping work functions.