Atomistic Insights into CO Oxidation on Pt Single Crystals with Machine-Learned Force Fields and Spectroscopy

Dynamic mechanistic insights into CO oxidation are crucial for optimizing catalyst design and controlling chemical conversion processes, especially for reactions like water-gas-shift (WGS). Several experimental investigations have demonstrated oscillatory behavior in this reaction, wherein the local coverage of chemisorbed O or CO can lead to markedly different reactivity. However, atomistic insight to confirm this hypothesis is inaccessible given the spatial and temporal resolutions of experimental techniques. Coupling this with the poor scaling of ab initio methods means that the mechanisms governing this disparate behavior cannot be determined with (1) atomic resolution, and (2) under proper time- and length scales for comparison to experiment. Understanding CO oxidation in this context is essential for gaining fundamental insights into the mechanisms governing chemical conversion at surfaces, and ultimately providing guidance for effective catalyst design and operating procedures. Therefore, accelerated computational methods that retain high accuracy are necessary to resolve the structure-performance relationship of these catalytic systems.<br/><br/>In this work, we investigate the mechanisms of CO oxidation on Pt(111) and Pt(100) surfaces using reactive molecular dynamics (MD) simulations powered by machine-learned force fields coupled with spectroscopic experiments. We employ the FLARE code, which accelerates sampling of expensive ab initio training data via its active-learning module, which can then be used to train a MLFF for use in production MD. The model is first validated against theoretical benchmarks on the coverage effects and site dependency of CO adsorption on Pt(111) and Pt(100), and subsequently used to study the transient kinetics of CO oxidation reactions as a function of temperature and relative coverage of CO and O. In addition to revealing the timescales of these processes, our simulations provide direct insight into the dominating mechanisms when either CO or oxygen is covering the surface. Moreover, we show that our MLFF is capable of capturing mesoscopic reconstruction of the Pt(100) surface, which is also modulated via exposure to these adsorbates. <br/><br/>We compare our findings with experimental results from operando spectroscopy and spectrometry, where the oscillatory behavior in CO oxidation is observed through pulsing CO on Pt(111) and Pt(100) with O₂ in the background. Our work thus provides atomistic understanding of the structure-performance relationship of the Pt(111) and Pt(100) surfaces in CO oxidation, which is crucial for optimizing the design of catalysts and developing better control strategies for optimal reaction conditions.