Dynamic Surface Reconstruction in Machine-Learning-Predicted Cu₃MoP Governs Selective CO Electroreduction to C₂⁺Products

Dynamic surface reconstruction has emerged as a pivotal strategy for enhancing both activity and selectivity in electrochemical CO reduction (eCOR) to multicarbon (C₂⁺) products, key intermediates in sustainable fuel synthesis. In this study, we introduce a physics-informed, machine-learning-driven framework that integrates moment tensor potentials with a symmetry-guided ABC algorithm to systematically explore the structural landscape of ternary alloys. This data-centric approach identifies Cu₃MoP as a thermodynamically favorable metallic phase, exhibiting robust dynamical stability as validated by phonon dispersion analysis and long-time-scale molecular dynamics simulations. Explicit modeling of the solid–liquid interface confirms that Cu₃MoP retains structural integrity under experimentally relevant electrochemical conditions. Mechanistic investigation of the Cu₃MoP(100) surface reveals a low onset potential of 0.19 eV for ethanol production and a moderate C–C coupling barrier of 0.31 eV, indicating kinetically accessible pathways toward C₂⁺ product formation. Under aqueous conditions, dynamic surface reconstruction induces Mo clustering and the emergence of Cu–Mo motifs, which modulate the electronic structure and redirect product selectivity from ethanol to ethylene. This interfacial restructuring also reorients water molecules, forming a structured hydration shell that stabilizes key reaction intermediates. Collectively, these findings establish Cu₃MoP as a dynamically adaptive and highly selective electrocatalyst and demonstrate the effectiveness of integrating machine learning with atomistic simulations to accelerate the discovery of next-generation multicarbon eCOR systems.