Screening of single-atom catalysts for CO2 electroreduction to CH4 using DFT calculations and machine learning

Abstract

The development of highly active and selective electrocatalysts for the CO₂ reduction reaction (CO₂RR) is critical for renewable energy applications, particularly in the sustainable synthesis of fuels such as methane (CH₄). Single-atom catalysts (SACs) have emerged as promising candidates for the CO₂RR, yet the lack of efficient predictive methods hinders their systematic exploration. Herein, we combined density functional theory (DFT) calculations with machine learning (ML) to investigate a series of 3d and 4d transition metal atoms supported on C₅N substrates with five different defect types as potential CO₂RR catalysts. Through a rigorous five-step screening strategy, we identify nine stable TM@C₅N SACs that exhibit superior catalytic activity and selectivity compared to the conventional Cu (211). Notably, our screening protocol incorporates *OH reduction, an often-overlooked factor, thereby enhancing the prediction accuracy. Among the candidates, Pd@C₅N_C2 demonstrates exceptional performance, achieving a remarkably low limiting potential of 0.42 V. For ML-assisted analysis, we trained XGBoost and Random Forest models using features derived from the thermodynamic, electronic, and geometric properties. Feature importance analysis highlights intrinsic catalyst parameters, such as the d-electron (dⁿ), first ionization energy (IE₁), d-band center (ε_d), and atomic radius (r), as dominant factors governing the CO₂RR performance. By elucidating atomic-level reaction mechanisms and advancing a high-throughput screening framework, this study establishes key structure–activity relationships through ML interpretability, offering valuable insights for the rational design and accelerated discovery of high-performance SACs for the CO₂RR.