Universal Neural Network Potential Study of the Pt/CO Heterogeneous Catalytic System

The high computational costs of DFT methods limit their applicability in studying heterogeneous catalytic systems. Neural network potentials present a promising approach for modelling such systems. In this study, we first validate the use of a universal neural network potential for the Pt(111)/CO system, whose theoretical approach is still under debate, by reproducing experimental CO IR spectra and providing insights into the vibrational modes of adsorbed CO. We further study temperature-dependent single CO adsorption dynamics and oxidation barriers on a Pt201 nanoparticle. The results show that low barriers require on-top CO and O adsorption at ridge sites, which may be hindered by CO poisoning or strong bridge oxygen adsorption, providing a possible explanation for the experimentally observed slower reaction kinetics at ridges compared to facets. These findings underscore the need for consideration of site distribution on nanoparticles beyond the adsorption and activation properties calculated from slab systems.