(240f) Active Learning across Intermetallics Guides Discovery of Electrocatalysts for Carbon Dioxide Reduction and Hydrogen Evolution

Electrochemical reduction of carbon dioxide or evolution of hydrogen are promising methods for storing intermittent energy from renewable sources. Scale up of these reactions requires the discovery of new electrocatalysts, but current theoretical screening methods cannot search the entire experimentally relevant space of intermetallics. We present a fully automated screening method that uses active machine learning to guide density functional theory calculations of thermodynamic descriptors of electrocatalyst performance. The method created a surrogate model that predicts CO and H adsorption energies with mean absolute errors of 0.18 and 0.17 eV, respectively. We demonstrate feasibility of the method by screening various intermetallic combinations of 31 different elements, which encompasses 50% of the d-block elements and 33% of the p-block elements. This method has thus far identified 29 candidate intermetallics for carbon dioxide reduction and 118 candidates for hydrogen evolution. We also present qualitative, heuristic analyses to prioritize the top candidates.