From flat to stepped: active learning frameworks for investigating local structure at copper–water interfaces

Understanding processes at solid-liquid interfaces at the atomic level is important for applications such as electrocatalysis. Here we explore the effects of different step densities on the structure of interfacial water at the copper-water interface. Utilizing spatially resolved uncertainties, we develop an active learning framework and train a machine-learning force field (MLFF) based on dispersion-corrected density functional theory data. Using molecular dynamics simulations, we investigate structural properties of water molecules in the contact layer, including density profiles, angular distributions, and 2D pair correlation functions. In accordance with previous studies, we observe the formation of two sublayers within the contact layer at the Cu(111) surface, whereas the structure on surfaces with a high step density is dominated by the undercoordinated ridge atoms. By systematically decreasing the step density, we identify the cross-over to when the behavior observed at the flat surface can be locally recovered. Using dimensionality reduction, we identify four distinct types of Cu environments at the interfaces, providing insights into analyzing less idealized surfaces with MLFFs.