Data-Driven Design of Single-Atom Electrocatalysts with Intrinsic Descriptors for Carbon Dioxide Reduction Reaction

Abstract The strategic manipulation of the interaction between a central metal atom and its coordinating environment in single-atom catalysts (SACs) is crucial for catalyzing the CO 2 reduction reaction (CO 2 RR). However, it remains a major challenge. While density-functional theory calculations serve as a powerful tool for catalyst screening, their time-consuming nature poses limitations. This paper presents a machine learning (ML) model based on easily accessible intrinsic descriptors to enable rapid, cost-effective, and high-throughput screening of efficient SACs in complex systems. Our ML model comprehensively captures the influences of interactions between 3 and 5d metal centers and 8 C, N-based coordination environments on CO 2 RR activity and selectivity. We reveal the electronic origin of the different activity trends observed in early and late transition metals during coordination with N atoms. The extreme gradient boosting regression model shows optimal performance in predicting binding energy and limiting potential for both HCOOH and CO production. We confirm that the product of the electronegativity and the valence electron number of metals, the radius of metals, and the average electronegativity of neighboring coordination atoms are the critical intrinsic factors determining CO 2 RR activity. Our developed ML models successfully predict several high-performance SACs beyond the existing database, demonstrating their potential applicability to other systems. This work provides insights into the low-cost and rational design of high-performance SACs.