A configuration sampling study of reaction intermediates constituting catalytic cycles for CO oxidation with Pt1/TiO2.

Abstract

We combine ab initio molecular dynamics (AIMD) simulations with an unsupervised machine learning approach to automate the search for possible configurations of CO oxidation reaction intermediates catalyzed by the atomically dispersed Pt1/TiO2 catalyst. Following the example of Roncoroni and co-workers [Phys. Chem. Chem. Phys. 25, 13741 (2023)], we employ t-distributed stochastic neighbor embedding and hierarchical density-based spatial clustering of applications with noise to reduce the dimensionality and cluster AIMD snapshots based on the local coordination environment of Pt. We identify new local minima, particularly in cases where CO2 is bound to the active site, because it can coordinate in various ways with both the metal and support. The new minima constitute additional elementary steps in some proposed pathways for CO oxidation, resulting in turnover frequencies that differ from prior estimates by several orders of magnitude. This work, therefore, demonstrates that configuration sampling is a necessary component of computational studies of catalytic cycles for atomically dispersed catalysts.