Design of Graphdiyne and Holey Graphyne-Based Single Atom Catalysts for CO2 Reduction With Interpretable Machine Learning

Abstract

Using electrochemical CO2 reduction reaction (CO2RR) to synthesize value-added hydrocarbons provides a useful solution for environmental issues and energy crisis. However, this process is impeded by the low activity and selectivity of electrocatalysts toward targeted products. Employing density functional theory computations, the graphdiyne and holey graphyne supported single-atom catalysts (SACs, M/GDY and M/HGY) are demonstrated to be the promising candidates for the CO2RR. By taking full elemental diversity of metal sites across the periodic table, 25 catalysts are found to effectively activate CO2 and inhibit competitive hydrogen evolution, and 8 of them show higher activity for CH4 production than Cu(211). Remarkably, changing supports are found to greatly affect limiting potentials and reaction pathways, even leading to an “inert-active” transition for some metal centers. The resulting SACs, including Mn/GDY, Co/HGY, Ru/GDY, and Os/GDY, can achieve high activity with low limiting potentials of ≈ −0.22 to −0.58 V. Machine learning enables to identify the critical role of the polarized charge and magnetic moment of metal atoms in affecting the activity. The built machine learning model also shows an interpretable capability to predict the activity of the other types of SACs, offering a great promise to quick screening of high-performance SACs.