Machine Learning for the Expedited Screening of Hydrogen Evolution Catalysts for Transition Metal-Doped Transition Metal Dichalcogenides

Abstract

Two-dimensional transition metal dichalcogenides (TMDs) have gained attention as potent catalysts for the hydrogen evolution reaction (HER). The traditional trial-and-error methodology for catalyst development has proven inefficient due to its costly and time-intensive nature. To accelerate the catalyst development process, the Gibbs free energy of hydrogen adsorption ( $\Delta G_{{\text{H}}^{\ast }}$ ), computed using the density functional theory (DFT), is widely used as the paramount descriptor for evaluating and predicting HER catalyst performance. However, DFT calculations for $\Delta G_{{\text{H}}^{\ast }}$ are time-consuming and thus pose a challenge for high-throughput screening. Herein, we devise a predictive model for $\Delta G_{{\text{H}}^{\ast }}$ within transition metal-doped TMD systems using a machine learning (ML) framework. We calculate DFT $\Delta G_{{\text{H}}^{\ast }}$ values for 150 TM-doped MX2 (CrS2, MoS2, WS2, MoSe2, and MoTe2) and apply various ML algorithms. We validate the universality of our model by constructing 15 new external test sets. The prediction results show a high correlation coefficient of $R^2=0.92$ . Based on feature analysis, the three most important parameters are the number of valence electrons of the doped transition metal, the distance of the valence electrons of the doped transition metal, and the electronegativity of the doped transition metal. Our DFT-based ML model provides a useful guideline for the material development process through $\Delta G_{{\text{H}}^{\ast }}$ prediction and facilitates the efficient design of transition metal dichalcogenide catalysts that exhibit superior HER activity.