The interfacial structure and electronic properties of gold (Au) nanoclusters supported on anatase TiO₂ surfaces are key determinants of catalytic and photocatalytic activity for the Au/TiO₂ systems. Anatase TiO₂ exposes primarily the thermodynamically stable (101) surface and the more reactive but less stable (001) surface. Au nanoclusters are typically deposited onto these facets via methods such as impregnation, deposition–precipitation, or ligand exchange, often followed by annealing in air. These synthetic procedures give rise to facet-dependent interactions, including local surface reconstruction and interfacial charge transfer, which in turn influence the charge storage ability and overall catalytic behavior. As such, elucidating the atomic-scale structure of the Au/anatase TiO₂ interface is essential for understanding catalytic mechanisms and for rational catalyst design. Despite extensive experimental efforts, the precise nature of the Au/TiO₂ interface remains elusive due to the limited availability of in situ and surface-sensitive structural characterization techniques. On the theoretical side, the Au/anatase interface has been extensively studied using first-principles methods over the past two decades. Most studies focus on small clusters (typically <15 atoms) adsorbed on unreconstructed anatase surfaces within small unit cells, often neglecting interfacial oxygen and realistic reaction conditions.
To address these gaps, we conduct an atomistic study of realistic, subnano-size Au nanoclusters on reconstructed TiO₂(001) and (101) facets in ambient air conditions, bridging the gap between sub-nanometer models and experimental catalysts. Leveraging a machine learning potential, we perform large-scale global structure searches to efficiently sample interfacial configurations. We aim at revealing how cluster size and adsorbate adsorption jointly govern interfacial stability and charge redistribution, offering new atomic-level insights into the structure–function relationship of Au/TiO₂ catalysts under realistic conditions.