Mechanism-Guided, Data-Driven Discovery of a Dinuclear Gold Catalyst for Promoting Oxidative Addition

Gold catalysis is frequently constrained by the limited accessibility of Au(I)/Au(III) redox pathways, particularly for the direct oxidative addition (OA) of aryl halides. Here, we present a mechanistically guided and machine learning-accelerated strategy to design dinuclear gold complexes capable of facile OA with aryl iodides. Mechanistic DFT calculations reveal a favorable cationic Au(III)-Au(I) OA pathway localized at a single gold center within an electronically coupled bimetallic framework. Guided by this insight, 42 398 bidentate ligands have been screened using high-throughput virtual screening, multiobjective Bayesian optimization and DFT refinement. This approach identifies pyridine-phosphine (di-PN) ligands as privileged scaffolds, which can dramatically reduce the OA activation barrier and render the reaction exergonic. Interpretable machine learning and energy decomposition analyses elucidate that the enhanced reactivity arises from a synergy of geometric pre-distortion, axial electronic polarization, and adaptive Au-Au interactions. A representative predicted dinuclear gold catalyst has been synthesized and experimentally validated in a model sulfonylation reaction of iodobenzene, supporting the practical relevance of the computationally identified di-PN scaffold. This work establishes a mechanism-guided, data-driven workflow for evaluating ligand effects in dinuclear gold redox catalysis, with broader implications for multinuclear transition-metal catalyst development.