Generative Discovery of Alloy Catalysts for Radical-Mediated Methane-to-Methanol Conversion.
Selective oxidation of methane to methanol under mild conditions remains a "holy grail" in chemical manufacturing, constrained not only by the inertness of the C-H bond but also by the difficulty of controlling highly reactive intermediates once methane is activated. Plasma activation provides a nonequilibrium reaction environment in which methyl radicals are generated independently of the catalytic surface, effectively bypassing surface-mediated C-H activation. However, this shifts the bottleneck to subsequent interfacial chemistry, where radical interception, product selectivity, and catalyst stability impose competing constraints. Here, we introduce a plasma-electrochemical reaction framework coupled with a generative catalyst discovery strategy to navigate this multiobjective landscape. By integrating graph-neural-network-accelerated evaluation with generative exploration across a multimetallic alloy space spanning ∼1029 possible configurations, the framework identifies transferable catalytic motifs rather than individual compositions. This process reveals an emergent catalytic architecture consisting of an Ag-rich matrix with isolated Pd/Pt sites, which provide localized reactivity for radical interception and C-O bond formation while maintaining weak oxygen affinity that suppresses overoxidation. Guided by this motif-level principle, we identify experimentally realizable Ag3Pd and Ag3Pt catalysts. In a hybrid plasma-electrocatalytic system, Ag3Pd achieves a methanol faradaic efficiency of up to 80.1% at ∼1.7 mA cm-2 with a productivity of 64.5 μmol·cm-2·h-1, outperforming previously reported plasma-assisted and electrochemical methane conversion systems. More broadly, this work demonstrates how generative algorithms can uncover emergent catalytic architectures across vast chemical spaces.