Data-guided design of double-atom catalysts for enhanced electrocatalytic performance

Abstract

Double-atom catalysts (DACs) are promising electrocatalysts due to their synergistic metal–metal interactions and high atom utilization. However, the vast chemical space arising from diverse metal pairs and substrates presents a major challenge for rational design. Here, we combine high-throughput density functional theory (DFT) calculations with machine learning (ML) analysis to systematically investigate DACs for the CO₂ reduction reaction (CO₂RR), hydrogen evolution reaction (HER), and oxygen evolution reaction (OER). We establish a predictive ML framework capable of rapidly screening DAC candidates with near-DFT accuracy, enabling efficient evaluation across a wide range of substrates. Guided by ML and DFT approaches, we identify PtZn/N-C₃N₄ as a highly active OER catalyst with a theoretical overpotential of ∼0.15 eV, and CuNi/N-C₃N₄ as a top-performing bifunctional catalyst for overall water splitting. For CO₂RR, VTi/N-C₃N₄ shows a limiting potential approaching ∼0.15 V, close to the optimal volcano plot peak, along with strong HER suppression. In summary, this work offers key insights for the design of ACs, providing substantial time savings and demonstrating the immense potential of ML as a universally applicable tool in diverse energy-related fields.