Size- and Facet-Dependent Behavior of Electron Storage at the Au/Anatase TiO ₂ Interface under Ambient Conditions from Machine Learning

The interfacial structure and electronic properties of gold (Au) nanoclusters supported on anatase TiO2 surfaces are key determinants of catalytic and photocatalytic activity for Au/TiO2 systems. Here, we present an atomistic study of realistic, ∼1.5 nm Au nanoclusters on reconstructed TiO2(001) and (101) facets under ambient air conditions, bridging the gap between small supported cluster models and experimental catalysts. We focus on the ability of Au/TiO2 to store photogenerated electrons that can be used later in dark conditions. Leveraging a high-dimensional Au–Ti–O neural-network potential coupled with stochastic surface walking and DFT refinement, we exhaustively sampled atomic configurations over 120,000 minima. Two distinct interfacial motifs emerge─Type (i), with oxygen adsorbed on Ti sites near the Au nanocluster, and Type (ii), with oxygen occupying Au3 hollow sites. While both motifs stabilize clusters on both facets, Type (ii) is notably more prevalent on (101), giving rise to deeper-lying O 2p states and a larger fraction of fully oxidized Auδ+ (δ ≥ 1) that can passivate the cluster. In contrast, Type (i) optimally balances neutral Au0 and partially oxidized Auδ+ (0 < δ < 1), enabling more than 0.90 vacant state per interfacial Au atom and enhancing electron-storage capability. Coupled with facet-dependent behavior and oxidation-state distributions, these findings deliver direct, experimentally relevant design principles for tailoring the electronic functionality of Au/TiO2 catalysts.