Constant-Potential MD with Neural Network Potentials Reveals Cation Effects on CO₂ Reduction at Au-Water Interfaces

Understanding the evolution of electrified solid-liquid interfaces during electrochemical reactions is critical but poses significant challenges due to the difficulty in capturing their dynamic behavior with high temporal resolution over extended time scales. Here, we present a constant potential reactor framework that enables ab initio-accurate simulations of electrochemical reactions, providing real-time, atomic-scale insights into the evolution of electrified interfaces. By integrating an enhanced-sampling active learning protocol, our approach leverages scalable neural network potentials trained on high-throughput density functional theory computations within an explicit-implicit hybrid solvent model. This framework uncovers key mechanistic insights, such as the intrinsic role of alkali metal cations in promoting CO₂ adsorption while suppressing hydrogen evolution reaction, reconciling prior experimental observations and clarifying ambiguities. By bridging gaps between experiments and computations, our framework establishes a powerful tool for studying the dynamic interplay between interfacial structure and reactivity in realistic electrochemical environments, paving the way for future advances in electrochemistry.