Coordination-Engineered Double-Atom Catalysts with Inverse Sandwich Structures for CO ₂ Reduction: A Combined DFT and Machine Learning Study

Coordination engineering provides a promising route to enhance the activity and selectivity of electrocatalysts for CO₂ reduction reaction (CO₂RR). Here, we established a density functional theory (DFT)-machine learning (ML) framework to accelerate the discovery of Cu-based double-atom catalysts (DACs) with inverse sandwich structures. A four-step screening protocol (stability → CO₂ adsorption → selectivity → activity) identified 18 candidates among 162 structures, all exceeding the performance of Cu(111) and Cu-N₄, highlighting the benefits of coordination-tuned geometries. We further developed an interpretable XGBoost model based on five key descriptors to predict catalytic activity. Applying this model to 162 Ag-based and 837 Cu-based DACs with mixed C/N/B coordination yielded 9 and 153 promising candidates, respectively. DFT validation of selected candidates confirmed the model's reliability. This study highlights the potential of coordination-engineered DACs for efficient CO₂RR and demonstrates a robust, transferable DFT-ML strategy for accelerating catalyst discovery.