Machine Learning-Predicted Ternary Molybdenum Chalcogenophosphides for High-Efficiency Hydrogen Evolution Catalysis

The search for efficient and cost-effective alternatives to platinum-based catalysts for the alkaline hydrogen evolution reaction (HER) remains a formidable challenge, driving the need for innovative materials. In this study, we employed machine learning-driven moment tensor potentials in conjunction with particle swarm optimization to predict a new family of ternary molybdenum chalcogenophosphides, specifically Mo2SP and Mo3SP. Our calculations show that these materials exhibit robust thermodynamic, kinetic, and thermal stability at room temperature, as indicated by their low formation energies, phonon dispersion curves free of imaginary frequencies, and molecular dynamics simulations. Advanced HSE06 calculations further confirm their metallic nature in the bulk phase. We also investigated their size-dependent nucleation behavior, finding that Mo2SP and Mo3SP possess distinct crystallization pathways, with Mo2SP showing a lower nucleation barrier. Notably, the (100) facet of Mo3SP displays optimal hydrogen adsorption energies and lower water dissociation barriers compared to the (110) facet of Mo2SP, highlighting its superior catalytic efficiency in the Volmer step of HER. These findings provide a crucial foundation for the development of high-performance, low-cost catalysts based on ternary molybdenum chalcogenophosphides, offering a promising alternative for sustainable hydrogen production.