Prediction of Vibrational Spectra of CO Adsorbed on PdPt(111) Using Machine Learning and Monte Carlo Simulations

Abstract

In this work, we combine density functional theory (DFT), Gaussian Process Regression (GPR), and Monte Carlo simulations to investigate surface segregation and CO adsorption on Pd/Pt alloy surfaces. The DFT dataset is used to train a sparse GPR model that efficiently predicts formation energies, CO adsorption energies, and vibrational frequencies across a wide range of alloy configurations. Key structural features are encoded by local atomic descriptors, capturing both bulk and surface environments as well as adsorption sites for CO. Our simulations predict that Pd initially segregates to the surface at low temperatures, transitioning to a more random alloy configuration at elevated temperatures. CO adsorption strongly influences segregation, driving Pt toward the surface while altering the local environment and vibrational properties of adsorbed CO. The predicted surface segregation and the composition of the binding site can be connected directly to the observable vibrational spectrum and binding energy of CO. This allows us to predict central quantities in catalysis where the stretch frequency of CO is used for characterization, while the CO-adsorption energy acts as a central descriptor of catalytic activity when using BEP- or scaling relations. Many important processes involve CO, such as Fischer-Tropsch synthesis, steam reforming, the reverse water-gas shift, and oxidation reactions. Overall, the combination of GPR and Monte Carlo simulations not only provides a powerful framework for modeling large, complex systems under variable reaction conditions but also showcases a scalable, data-driven strategy for predicting segregation phenomena, adsorption energetics, and vibrational signatures in alloy materials.