Integrating Density Functional Theory Calculations and Machine Learning to Identify Conduction Band Minimum as a Descriptor for High-Efficiency Hydrogen Evolution Reaction Catalysts in Transition Metal Dichalcogenides

Identifying efficient and physically meaningful descriptors is crucial for the rational design of hydrogen evolution reaction (HER) catalysts. In this study, we systematically investigate the HER activity of transition metal dichalcogenide (TMD) monolayers by combining density functional theory (DFT) calculations and machine learning techniques. By exploring the relationship between key electronic properties, including the conduction band minimum (CBM), pz band center, and hydrogen adsorption free energy (ΔG*H), we establish a strong linear correlation between the CBM and ΔG*H, identifying the CBM as a reliable and physically meaningful descriptor for HER activity. Furthermore, this correlation is validated in vacancy-defected TMD systems, demonstrating that the CBM remains an effective descriptor even in the presence of structural defects. To enable the rapid and accurate prediction of the CBM, we develop an interpretable three-dimensional model using the Sure Independence Screening and Sparsifying Operator (SISSO) algorithm. The SISSO model achieves a high predictive accuracy, with correlation coefficients (r) and coefficients of determination (R2) reaching 0.98 and 0.97 in the training and 0.99 and 0.99 in the validation tests, respectively. This study provides an efficient computational framework that combines first-principles calculations and machine learning to accelerate the screening and design of high-performance TMD-based HER catalysts.