The physics and catalytic performance of clusters differ greatly from those of nanoparticles that have been studied extensively. The considerable heterogeneity of clusters resulting from the synthesis, the ensemble-averaged reaction rates obtained experimentally, and the inability to perform single-cluster measurements create a profound knowledge gap regarding the dependence of the reaction rate on cluster size. The asymmetric nature of clusters also raises the question of the relative activity of each site within a cluster. Here, we introduce a multiscale modeling framework applied to varying Pt cluster sizes. Our framework allows site- and cluster-specific reaction rate estimates using density functional theory calculations of reaction networks on small Pt1−4 clusters on γ-Al2O3(110), simple correlative machine-learning analysis, site-specific microkinetic modeling, and extension to Pt5−14 clusters with low-energy structures obtained using machine-learning potentials and basin-hopping optimization. We also demonstrate the framework for PtSn and PtZn bimetallic structures. We apply this framework to propane dehydrogenation (PDH) to propylene. Our results reveal strong correlations between transition state energies and specific adsorption energies. An expected volcano surface-like activity is observed. We find that Pt dimers and trimers are catalytically more active than other clusters by orders of magnitude. We reveal a profound site-specific activity in each cluster due to site-support interactions, which raises questions about the meaning of turnover frequency and the structural sensitivity of a reaction on nanoparticles versus small clusters. Actual catalysts may contain only a tiny fraction of clusters and sites that are catalytically relevant for PDH.
Liu et al. (Wed,) studied this question.