Sticking of Methane on Palladium Oxide: a Computational Approach

Abstract

The catalytic properties of palladium oxide for the combustion of methane have been studied extensively in recent years. The rate-determining step of this reaction is believed to be the dissociation of methane on the surface. The rate of the event is dependent on both the active sites of the catalyst and the energy and orientation of the incoming methane molecules. The dependence of energy and orientation is often summarized in a sticking coefficient. Here, we will address the challenge of calculating the sticking coefficient from firstprinciples. However, due to the large number of trials and large time scales required to study this event, ab initio molecular dynamics would be too computationally expensive to perform, and an alternative approach using neural networks is applied. The adsorption position on the active sites and the activation energy of the dissociation process are studied using density functional theory. To determine the probability of a sticking event, a neural network is trained to predict the multidimensional potential energy surface, which is used to perform molecular dynamics. The density functional theory calculations confirm that the active sites of the catalyst are the under-coordinated palladium atoms, with an apparent activation energy of 0.2 eV for the dissociation reaction. The neural network is able to predict the energies of the system five orders of magnitude faster than regular density functional theory calculations, with an MAE of 0.02 eV. The molecular dynamics suggest that the previously believed most probable transition path might be dominated by the sum of the other, less likely, transition paths. The hope is that the results and understanding obtained from this computational study can be used to assist in the discovery of more efficiently designed catalysts in the future.