Understanding cation and surface charging effects at electrified interfaces using neural network interatomic potentials

Electrified interfaces are at the heart of electrochemical processes related to several technologically important applications, yet they remain poorly understood. This lack of understanding is primarily due to their complex and dynamic nature, that poses significant challenges in developing realistic models. To overcome some of these challenges, we train message-passing neural network interatomic potentials based on the MACE architecture to perform long time-scale molecular dynamics (MD) simulations (O(ns)) of electrified Au interfaces with near-first-principles accuracy. Specifically, we study the impact of surface charging and cation identity (Li+, Na+, K+, Cs+) to obtain detailed insights into various interfacial properties, including the structure and dynamics of (near surface) water molecules, cations, and the double layer capacitance. In general, we find that surface charging has a greater impact on these interfacial properties compared to that of the cation identity. Furthermore, we find that the estimated double layer capacitance from MACE/MD simulations (17-19 muF/cm2) at cathodic potentials is largely insensitive to the cation identity. Our work demonstrates the potential of message passing neural network potentials to obtain a detailed atomistic-level understanding of electrified interfaces.