Theoretical and Machine Learning Exploration of Electronic Factors Governing Ni-Centered CO2 Reduction Catalysts

Abstract

Carbon-based electrocatalysts are promising candidates for CO2 reduction due to their intrinsic redox properties. However, achieving an in-depth understanding and rational design of the nickel site coordination environment and active site density remains a significant challenge. In this study, a single-atom catalyst supported on graphene was designed to reduce CO2 to CO or HCOOH, based on density functional theory. The catalytic activity of the Nin-CxNy-d monolayer was systematically evaluated through charge density, density of states, and molecular dynamics analyses, verifying the conductivity and stability. Furthermore, an analysis of the Gibbs free energy pathway and electronic structure revealed that Ni2-C3N1-1 exhibits excellent catalytic performance for CO production in the CO2 reduction reaction, while Ni2-C2N2-1 demonstrates superior performance for HCOOH, with relatively low limiting potentials of -0.27 V and -0.09 V, respectively. Molecular orbital theory analysis underscores the critical role of bonding states in explaining the adsorption energy of intermediate products. Moderate adsorption energy is shown to effectively suppress hydrogen evolution reactions, thereby enhancing both reaction activity and product selectivity. Leveraging the best-performing machine learning XGBoost model, the feature importance between HCOOH product and the Ni single-atom catalyst structure was predicted to be 0.568, allowing for the identification of optimal tuning strategies to achieve superior catalytic performance. This study provides novel theoretical insights and technological strategies for advancing sustainable CO2 reduction.