Dynamic Polaronic Control of Metal Cluster Adaptability on Reducible Oxides

Metal-oxide interactions are ubiquitous in many technological applications and involve a complex interplay between the oxide support and the metal nanoparticle. Particularly, it has been proposed that in strong metal-support interaction, the defect chemistry affects the metal cluster morphology. Here we develop a physics-guided machine learning framework to decode these interactions using Pt₇ and Pt₁₃ representative of planar and tridimensional clusters, analyzing the impact of across oxygen vacancy concentrations of CeO2-x = 0-12.5% (528 configurations). Our models (R² > 0.97) reveal that polaron swarms, rather than defect concentrations, predominantly control cluster shape and charge through size-dependent pathways. The framework yields quantitative design principles for defect-driven catalyst optimization and provides a general methodology for systematic mechanisms of metal-support interactions across diverse catalyst systems.