Grand Canonical Monte Carlo Simulations for Hydrogen Adsorption on Metal Surfaces Using Neural Network Potentials

Hydrogen adsorption on metal surfaces (Pt, Pd, and Ir) is investigated by the grand canonical Monte Carlo simulations using density functional theory-based neural network potentials. By introducing the modified cavity bias method, generalized hybrid Monte Carlo method, and replica exchange method, atomic configurations across various temperatures and pressures are efficiently sampled. To ensure accurate predictions of adsorption energy of hydrogen atoms in diverse configurations, the neural network potentials are refined using additional loss terms that capture energy differences between systems with and without hydrogen atoms. These combined techniques enable us to analyze hydrogen density profiles on metal surfaces with reliable accuracy. Our results highlight the influence of temperature and pressure on hydrogen atom configurations and their interactions with the surfaces. This approach not only elucidates the intricate interactions between hydrogen atoms under specific thermodynamic conditions, which are critical for optimizing catalytic and hydrogen storage materials, but also demonstrates the efficacy of neural network potentials in providing high-fidelity simulations that are computationally feasible for complex surface phenomena.