(569j) Unraveling Dynamics of Interfacial Electrocatalysis with Foundational Deep Learning

Electrocatalysis at solid-electrolyte interfaces is fundamental to various future technologies for clean energy, such as electrolyzers, metal-air batteries, and fuel cells. Under biased potentials, not only the distribution of ionic species within the electrical double layer but also the electrode materials dynamically evolve via atomistic events, e.g., dissolution, deposition, and catalytic reactions. Despite its practical significance, our understanding of interfacial dynamics in electrocatalysis is very limited. This knowledge gap has severely compromised the predictive power of structure-performance relationships derived from ex situ material properties and catalytic outcomes.

Large-scale molecular simulations powered by foundational deep learning potentials (MLP) is promising to unravel the dynamic mechanisms of interfacial electrocatalysis, shedding light on strategies to modulate underpinning processes for improved catalytic performance. To establish benchmarks for our methodology, we are focusing on carbon dioxide (CO₂) electrolysis. Driven by the increasing accessibility and affordability of renewable energy, electrochemical CO₂ reduction reaction (CO2RR) into value-added chemicals and fuels has emerged as a promising route for shifting our society towards sustainability. Copper (Cu) is the only elemental metal capable of catalyzing CO2RR to C2+ products, albeit with high overpotentials. In-situ/operando characterization has shown structural evolution of Cu nanoparticles during CO2RR, mediated by transient Cu complexes, e.g., [Cu(CO)(H₂O)_n]⁺. However, the underlying dynamics of structural evolution and its relevance to CO2RR still remain elusive, largely due to the intrinsic complexities of electrified solid-electrolyte interfaces. Atomistic simulations of structural transformations at near quantum-chemical accuracy, with a rigorous inclusion of the solvation, ions, and potential bias, are appealing but have been inaccessible with today's computing infrastructures. With recent advances in deep learning and open materials informatics, large-scale molecular simulations powered by MLP are poised to effectively bridge the length and time scales in comparison with experimental data, providing unprecedented insights into dynamics of Cu electrocatalysts under CO2RR.