Accelerated Screening of Dual-Atom-Doped Copper Electrocatalysts for CO 2 -to-C 2+ Conversion via a Density Functional Theory-Guided Ten-Feature Interpretable Machine Learning Strategy

Electrochemical CO2 reduction (ECO2RR) to multi-carbon products (C2+) offers a promising approach to mitigate carbon emissions; however, the rational design of Cu electrocatalysts with high selectivity and low overpotential remains challenging and suffers from time-consuming trial-and-error experiments. This study introduces a density functional theory (DFT) and machine learning (ML) combined strategy to discover dual-atom-doped Cu electrocatalysts (A-B@Cu) for selective C2+ production. DFT calculations on the favored OC–COH dimerization reveal that doping induces lattice strain and regulates charge transfer from Cu to the OC–COH intermediate. With the A-B@Cu dataset expanded from 21 for DFT to 213 for ML, a gradient boosting regression (GBR) model, incorporating 10 features related to lattice strain and charge transfer, was developed to accurately predict the free energy of OC–COH dimerization (ΔGOC–COH). Moreover, three of the 10 features, Charge (the number of charge transfer), Φ (an atomic property derived from SISSO), and SF,ads (Fermi softness of Cu at adsorption sites), enable the rapid prediction of ΔGOC–COH with reduced accuracy requirements. This study developed a multi-feature, theoretically robust, computationally efficient ML strategy to expedite the identification of A-B@Cu electrocatalysts for enhanced C2+ production in ECO2RR.