Decoding the Impact of Solvent Hydrogen-Bonding Networks on Nanozyme Activity via Machine Learning Potentials

Precise mechanistic insights into the chemical reactions at solid-liquid interfaces are crucial for advancing functional nanomaterials, such as enzyme-mimicking nanocatalysts known as nanozymes. However, conventional periodic density functional theory (DFT) calculations, usually performed under vacuum conditions, can systematically underestimate reaction barriers, leading to discrepancies between computational predictions and the catalytic performance of nanomaterials in aqueous environments. Here, we propose a deep potential (DP)-based strategy that integrates DFT, ab initio molecular dynamics (AIMD), and active learning to resolve solvent-mediated reaction kinetics. Using hydrogen peroxide dissociation on Au(111) as a prototypical model reaction for nanozyme activity, we develop a DP model that show comparable accuracy to DFT calculations, enabling efficient identification of transition states in explicit solvents. Calculations using this model have demonstrated that the aqueous hydrogen-bonding network significantly increases the energy barrier for $H_2O_2$ dissociation by 0.58 eV, the experimentally observed low activity of gold nanoparticles in aqueous media. This approach provides a robust framework to decode solvent effects on chemical reactions at solid-liquid interfaces, which is valuable for advancing the rational design of functional nanocatalysts such as nanozymes.