Machine learning accelerates the screening of efficient metal-oxide catalysts for photocatalytic water splitting

Abstract

Photocatalytic water splitting represents a sustainable avenue for hydrogen production, yet the traditional trial-and-error approach for identifying efficient photocatalysts among numerous candidates is cumbersome. In this paper, we introduce a machine learning (ML)-driven approach for rapid screening of promising metal oxide photocatalysts for hydrogen generation. This methodology comprises three key steps: the construction of a comprehensive material library, ML-accelerated material screening, and Density Functional Theory (DFT) calculations for verification. Through this process, we successfully identify ten metal oxides with good energy levels and charge carrier transport properties from an initial library of 860 candidates. Notably, CsYO2 emerges as a standout candidate, exhibiting excellent photocatalytic water-splitting performance on its (001) crystal face, with Y atoms identified as active sites through DFT calculations. This study demonstrates the efficacy of applying machine learning to the precise and rapid screening of vast material databases, paving the way for discovering novel and efficient photocatalysts.