Dynamics of gold nanocluster on MgO surface with F-center defect

Gold nanoclusters supported on oxide surfaces exhibit enhanced catalytic activity due to charge redistribution at defect sites and strong metal–support interactions. In this study, we employ machine‑learned interatomic potential-based simulations to investigate the dynamics of Au$_8$ nanoclusters adsorbed on an oxygen‑vacancy (F‑center) defected MgO(100) surface. on-the-fly probability-enhanced sampling (OPES) simulations driven by a graph neural network–based collective variable reveal the low‑energy conformational landscape of Au$_8$ and the preferred binding site of CO, while machine-learned Bader charges uncovers an inverse correlation between Au–Au coordination number and localized negative charge in undercoordinated Au atoms. The most stable Au$_8$ conformer was then used to probe CO adsorption, demonstrating that CO preferentially binds to the most negatively charged undercoordinated Au sites. These findings highlight how defect‑mediated charge transfer and cluster morphology together dictate adsorption behavior and catalytic functionality on oxide supports.