Defect-Enhanced Low-Temperature CO Oxidation on Graphene-Supported Pt-Au Nanoclusters

The catalytic properties of bimetallic nanoclusters are governed by a complex interplay between their size, composition, chemical ordering, and interaction with the support. Here, we perform a comprehensive data-driven computational study using the USPEX evolutionary algorithm for global structure prediction of Pt-Au nanoclusters (up to 24 atoms), followed by high-throughput screening for O and CO adsorption energy with MACE foundation model and simulations of chemical reactor. We combine about 12 000 density functional theory (DFT) calculations with a machine learning workflow that adds over 80 000 additional data points to enable extensive sampling of ground-state and metastable isomers. Our analysis reveals that adsorption energies are primarily determined by local atomic composition of the active site rather than by the precise global-minimum structure. Consequently, a representative set of low-energy isomers provides the same adsorption information as the most stable clusters, which is crucial for experiments where monodisperse samples are difficult to obtain. We further demonstrate that a graphene substrate with a single carbon monovacancy strongly anchors cluster, enhances $O_2$ adsorption, and lowers CO oxidation barriers by activating traditionally inert Au atoms. Molecular dynamics simulations of CO saturation show that the supported cluster undergoes adsorption-induced reconstruction, with Pt atoms segregating to the surface, and that the support significantly alters the saturation profile compared to the free-standing cluster. This work establishes that rational design of bimetallic catalysts must take into account for local site chemistry, dynamic synergy with the support, and practical relevance of isomer ensembles.