The rational design of Cu‐based bimetallic catalysts for the electrochemical reduction of CO 2 into multicarbon (C 2 /C 2+ ) products critically depends on tuning the CO adsorption strength, which governs C–C coupling selectivity. However, systematically exploring the vast configurational space of CuM alloys through density functional theory (DFT) is computationally prohibitive. We developed a machine learning (ML) framework to predict CO adsorption behavior on Cu‐based bimetallic surfaces using a physically interpretable feature set, including geometric descriptors (nearest‐neighbor distance, coordination number) and elemental properties (electronegativity, ionization energy), which collectively determine the local electronic environment of the CO binding sites. A two‐step ML protocol, combining a Gradient Boosting Classifier to identify stable adsorption sites and a Gradient Boosting Regressor to predict their adsorption energies, was trained on DFT data for 15 CuM(111) and CuM(100) systems, resulting in a total of 1,515 structures. The models were then applied to screen 29 CuM alloys, encompassing about 91,000 adsorption sites across multiple surface concentrations and configurations of the alloying atoms. The screening identified CuAg, CuAl, CuAu, CuZn, CuIn, and CuGa as promising candidates for promoting CO–CO coupling, the rate‐determining step toward C 2 product formation. Among these, CuGa was selected for further validation as a representative system, having received far less attention in the CO 2 RR literature than other well‐studied alloys (CuAg, CuAl, and CuAu). Constant‐potential DFT calculations confirmed the ML predictions, revealing that CO dimerization on CuGa(100) proceeds with a more favorable reaction energy and an activation barrier about 0.2 eV lower than on pure Cu(100).
Salomone et al. (Thu,) studied this question.