Reactive Molecular Dynamics of Hydrogen Evolution at Charged Interfaces Using Effective Machine Learning Potentials

Abstract Understanding electrochemical reaction mechanisms requires atomistic simulations that simultaneously capture charged interfaces, explicit solvent, and long-time dynamics, which remain beyond the reach of conventional ab initio molecular dynamics. Here, we develop an equivariant machine-learning interatomic potential trained on density functional theory data for electrochemically activated MoS2/water interfaces under charged conditions. The resulting model reproduces first-principles energetics with an error of 7 meV/atom, while enabling nanosecond molecular dynamics simulations of systems containing hundreds of atoms. The simulations reveal that hydrogen evolution is strongly governed by surface hydrogen coverage, with higher coverages promoting faster H2 formation while preserving the interfacial water structure. Mechanistic analysis identifies the Heyrovsky-like pathway as the dominant hydrogen evolution mechanism, whereas Tafel recombination contributes only at high coverages. The emergence of sustained hydrogen evolution, stable interfacial populations, and coverage-dependent reaction pathways demonstrates that electrochemically informed MLIPs can reproduce chemically meaningful reactive dynamics at charged interfaces over nanosecond time scales.