Exploring Challenging C₂₊ Products During CO₂ Reduction via Machine Learning Acceleration

Although CO₂ reduction reaction (CO₂RR) has achieved significant progress in past years, the C₂₊ products are mainly limited to a few products, while many other products have rarely been reported in experiments with a limited understanding of the underlying mechanisms. Accordingly, in this work, machine learning (ML)-based theoretical investigations is conducted to uncover the reaction mechanisms for the conversion to challenging C₂₊ products (C₂H₆, CH₃OCH₃, CH₂CO, and C₂H₂) during CO₂RR on graphdiyne-supported atomic catalysts (GDY-ACs) with well-defined active sites. Using the first-principles machine learning (FPML) predictions, key factors limiting the diversity of C₂₊ products are identified. The conversions to C₂H₆ are mainly hindered by large rate-determining step (RDS) barriers (>4 eV). The formation of CH₂CO meets the competitive reactions due to similar reaction pathways with C₂H₆, which also undergoes further hydrogenation easily to other C₂₊ products. The CH₃OCH₃ formation is hindered by large dehydration barriers caused by steric hindrance induced by the neighboring adsorption of C₁ intermediates. FPML predictions also reveal the significance of binding configuration parameters in realizing efficient and accurate predictions. This work offers not only important references to the low selectivity of specific C₂₊ products but also critical theoretical insights into CO₂RR mechanisms. © 2025 Wiley-VCH GmbH.