Leveraging DFT and Machine Learning for Accelerated Screening of Cluster-Modified Single-Atom Catalysts in Oxygen Reduction Reactions

Abstract Integrating metal atomic clusters (ACs) with single atoms (SAs) offers an effective strategy for designing high-performance catalysts for the oxygen reduction reaction (ORR); however, the immense combinatorial space of SA/AC configurations poses a formidable challenge for rational catalyst screening. To address this issue, we developed an integrated workflow combining the density functional theory, machine learning, and experimental validation, enabling the rapid and accurate identification of promising candidates. By employing the adsorption free energy of the OH* intermediate as a descriptor, 156 distinct SA/AC catalysts were systematically screened, leading to the identification of a Ru-cluster-modified Ir−N4 catalyst exhibiting high ORR activity, with a theoretical overpotential as low as 0.21 V. Electronic structure analysis reveals that Ru cluster modification lowers the Ir d-band center, reduces spin moments, and decreases orbital occupancy at the Ir site, collectively weakening OH* adsorption and thereby enhancing ORR kinetics. Guided by these theoretical predictions, we synthesized this catalyst, and its good ORR performance was confirmed (E1/2 = 0.841 V and robust durability: 16 mV degradation in E1/2). This study demonstrates a rational theory–experiment-integrated framework for the efficient discovery of high-performance SA/AC catalysts, providing a generalizable strategy to guide the development of advanced catalysts for a wide range of electrochemical reactions.