Mobility and Sintering of Silica-Supported Platinum Clusters via Reactive Neural Network Potentials

Supported Pt atoms and their sub-nanometre cluster counterparts are a promising avenue for the targeted, bottom-up development of high specific activity heterogeneous catalysts. However, they suffer from instability with respect to growth and deactivation via sintering. The atomistic processes which comprise the early stages of sintering are difficult to access, as the timescales are too long for traditional electronic structure calculation methods and too short for most experimental techniques to probe. Pinning via hydroxyl groups has been suggested as a means to stabilize small clusters, but the means and extent of this sintering resistance is currently unclear. In this work, neural network potentials (NNPs), trained at the dispersion-corrected metaGGA level of theory, are employed to probe the dynamic behaviour of sub-nanometre Ptx species (x = 1, 3, 15) on a model flat oxide support: silicatene. We observe that mobility is surprisingly lowest for the single Pt atom and highest for the larger cluster, which influences the sintering routes, leading to non-Ostwald ripening coalescence pathways. When a silanol defect is introduced to the surface, the clusters are stabilised by the defect via hydrogen transfer from the surface to the clusters, thereby suppressing sintering via electrostatic trapping. These findings provide a mechanistic underpinning of sintering routes and stabilisation mechanisms on a model non-reducible oxide support and provide insights into the mechanisms of sub-nanometre metal particle mobility at the atomistic scale.