Machine learning molecular dynamics for reaction rates of CO oxidation on Pt(111): Impact of site preference and oxygen coverage.

Abstract

The kinetics of elementary surface reactions lay the foundation of our understanding of heterogeneous catalysis. However, accurate determination of surface reaction rates is by no means a trivial task. A prominent example is the oxidation of CO on Pt(111), where accurate rate constants were determined recently by experiments, challenging the harmonic transition state theory based on the revised Perdew-Burke-Ernzerhof (RPBE) density functional. However, RPBE is known to predict the incorrect favorable adsorption site for CO on Pt(111), whereas a van der Waals-corrected functional (vdW-DF2) correctly describes this site preference. In addition, the experimental data were derived at a low oxygen coverage, which was not represented well by previous models using small supercells. In this work, we constructed a machine-learned potential energy surface (PES) using vdW-DF2 for the CO + O/Pt(111) system. Based on this vdW-DF2 PES, we performed reactive flux rate constant calculations through molecular dynamics with enhanced sampling and varying periodic supercell sizes. Our results show improved agreement with the experiment compared to earlier RPBE-based calculations. Furthermore, the oxidation rate constants are found to modestly decrease with the decreasing oxygen coverage due to minor increase in the barrier height.