High-throughput discovery of surface-functionalized MXenes for photocatalytic water splitting via integrated machine learning and DFT screening

MXenes, as highly promising two-dimensional (2D) materials, have attracted significant attention in the field of photocatalytic water splitting. However, the enormous diversity of material structures and the substantial costs associated with experimental validation have hindered the exploration of MXenes' potential. In this work, we obtained property data for 213 different MXene structures through density functional theory (DFT) calculations and subsequently developed an atomic descriptor suitable for MXenes. Building upon this foundation, we established two machine learning (ML) prediction models, used the predicted data to perform high-throughput screening of numerous MXene structures, and finally conducted theoretical validation using the Heyd–Scuseria–Ernzerhof (HSE06) method and ab initio molecular dynamics simulations. The results show that the coefficient of determination (R2) of the two optimal prediction models reached 95% and 93%, respectively. Through high-throughput screening, 21 photocatalyst candidates were identified from 23 857 MXene structures, among which YTaC(OCl)(NCS) and Y2CFH exhibit high optical absorption coefficients (>1 × 105 cm−1), solar-to-hydrogen (STH) efficiencies (>14%), and thermodynamic stability. This work establishes an integrated strategy combining ML, DFT, and high-throughput screening, providing an effective approach for discovering and investigating potential MXene photocatalysts.