Understanding the nature of heterogeneous catalysis is a challenging task, particularly when the catalyst is sensitive to operating conditions. The difficulty comes from the complex and elusive states which occur at atomic scales in both space and time. A significant challenge is identifying the active sites of Cu-based catalysts for methanol synthesis. In this study, we apply molecular dynamics driven by machine learning force fields (MLMD) to elucidate the dynamical states of Cu surfaces under the influence of relevant adsorbates. Specifically, we investigated the surface evolution involving defect sites, adsorbate-induced surface transformation, and the co-adsorption effects of the adsorbates. Our findings reveal that Cu surfaces undergo considerable reconstruction under the influence of CO and hydrogen. The force fields are trained using first-principles calculations to capture the interactions between surfaces and adsorbates. In the case of CO interacting with the surface, the simulations indicate the formation of small active clusters, which enhances the water-gas shift reaction[1]. Further, we observed that adsorbed hydrogen also induces surface reconstruction, particularly on pristine Cu(100) surfaces, where new active sites of three-fold hollow sites emerge [2]. These results help to elucidate the true active sites of Cu catalysts and its effect to catalytic properties under reaction conditions, contributing to realising the catalytic systems that enable energy-efficient, low-emission, and sustainable chemical transformations.

• ACS Catal. 2026, 16, 5, 5068–5079
• Phys. Chem. Chem. Phys., 2026, Advance Article