Machine-Learning Methods Enable Exhaustive Searches for Active Bimetallic Facets and Reveal Active Site Motifs for CO2 Reduction

Abstract

Bimetallic catalysts are promising for the most difficult thermal and electrochemical reactions, but modeling the many diverse active sites on polycrystalline samples is an open challenge. We present a general framework for addressing this complexity in a systematic and predictive fashion. Active sites for every stable low-index facet of a bimetallic crystal are enumerated and cataloged, yielding hundreds of possible active sites. The activity of these sites is explored in parallel using a neural-network-based surrogate model to share information between the many density functional theory (DFT) relaxations, resulting in activity estimates with an order of magnitude fewer explicit DFT calculations. Sites with interesting activity were found and provide targets for follow-up calculations. This process was applied to the electrochemical reduction of CO2 on nickel gallium bimetallics and indicated that most facets had similar activity to Ni surfaces, but a few exposed Ni sites with a very favorable on-top CO ...