Harnessing the Potential of Machine Learning to Optimize the Activity of Cu-Based Dual Atom Catalysts for CO2 Reduction Reaction

Abstract

The electrochemical CO₂ reduction reaction (CO₂RR) paved the way to carbon neutrality while producing value-added chemicals and fuels. While Cu-based catalysts show potential, they suffer from inadequate faradaic efficiency. In this study, we explore Cu(100) surface-based dual atom alloy (DAA) catalysts for the CO₂RR to produce C₁ and C₂ products. Three distinct doping patterns involve two identical or different transition metals across 27 candidates. Machine learning (ML) based models were developed with high accuracy to predict the catalytic activity of unknown catalysts. The scaling relation between the adsorption energies of *CO and *CHO is circumvented by regulating the local environment with preferential dual atom doping. The integrated DFT+ML approach identifies 14 and 8 most suitable DAAs for C₁ and C₂ product formation, respectively. Feature importance analysis underscores the significance of valence d-orbital electrons in *CO adsorption. Additionally, PDOS analysis reveals atom-like electronic states in doped metals, characterized by highly localized d-states.