Screening of novel halide perovskites for photocatalytic water splitting using multi-fidelity machine learning

Abstract

Photocatalytic water splitting is an efficient and sustainable technology to produce high-purity hydrogen gas for clean energy using solar energy. Despite the tremendous success of halide perovskites as absorbers in solar cells, their utility for water splitting applications has not been systematically explored. A band gap greater than 1.23 eV, high solar absorption coefficients, efficient separation of charge carriers, and adequate overpotentials for water redox reaction are crucial for a high solar to hydrogen (STH) efficiency. In this work, we present a data-driven approach to identify novel lead-free halide perovskites with high STH efficiency (η_STH > 20%), building upon our recently published computational data and machine learning (ML) models. Our multi-fidelity density functional theory (DFT) dataset comprises decomposition energies and band gaps of nearly 1000 pure and alloyed perovskite halides using both the GGA-PBE and HSE06 functionals. Using rigorously optimized composition-based ML regression models, we performed screening across a chemical space of 150 000+ halide perovskites to yield hundreds of stable compounds with suitable band gaps and edges for photocatalytic water splitting. A handful of the best candidates were investigated with in-depth DFT computations to validate their properties. This work presents a framework for accelerating the navigation of a massive chemical space of halide perovskite alloys and understanding their potential utility for water splitting and motivates future efforts towards the synthesis and characterization of the most promising materials.