Machine Learning-Assisted Design of Dual-Atom Catalysts and the "Convergence-Selective Transfer" Charge Regulation Mechanism for ORR

Abstract

The oxygen reduction reaction (ORR) is a crucial cathodic process in fuel cells, but the unclear catalytic mechanism, low screening efficiency, and poor environmental adaptability led to slow reaction kinetics, hindering the development of high-performance electrocatalysts to replace scarce platinum-based materials. A novel design strategy for high performance dual atom ORR catalysts is proposed, leveraging the synergistic integration of machine learning (ML) and first-principles calculations to construct 53 Fe/CoM@N6-G dual-atom catalysts supported on defective graphene, and thus to achieve high precision prediction of ORR overpotentials for 378 dual-atom catalysts. Among them, FeZn@N6-G, FeIr@N6-G, IrCu@N6-G exhibit excellent catalytic performance. Moreover, we have innovatively proposed a "convergence-selective transfer" charge regulation mechanism in nitrogen-coordinated diatomic catalysts, where charges are regulated by differences in metal centers, selectively converging at active sites and transferring to key reaction intermediates, thereby optimizing the adsorption strength of oxygen-containing species and the O-O bond cleavage process, ultimately enhancing ORR catalytic performance. Based on the proposed charge regulation mechanism, the key descriptors j =(〖ΔG〗_1^a+〖ΔG〗_4^a )+abs(〖ΔG〗_1^a-〖ΔG〗_4^a ) for evaluating catalytic performance are determined with a high coefficient of determination R² = 0.99.