Surface characterization of cerium oxide catalysts using deep learning with infrared spectroscopy of CO

Abstract

Characterization of material surfaces is crucial for understanding their properties and behavior. In this work, we utilized a deep learning technique, along with infrared (IR) spectrum of CO as a probe molecule, to explore the surface properties of cerium oxide (CeO2) catalysts. Through systematic density functional theory (DFT) investigation of CO-derived adspecies on various CeO2 facets, we obtained an extensive dataset containing vibrational frequencies, intensities, and adsorption energies of CO on CeO2. This dataset was used to synthesize large quantities of complex IR spectra to train deep learning models for predicting surface structures, including the distribution of CeO2 facets, CO-derived adspecies, and binding energies. These models were successful in analyzing experimental IR spectra of CO adsorbed on different types of CeO2, and their predictions were consistent with experimental observations in most cases. This work provides a machine learning approach in understanding the morphology, local environmental arrangement, interaction behavior of probe molecules, and catalytic characteristics of diverse CeO2 materials.