(754e) Computational Modeling of Operando Infrared Spectroscopy for Site-Specific Catalyst Characterization

The need for in-situ and operando methods to characterize promising catalyst structures with adsorption site-level resolution and evaluate structure stability under operating conditions has never been more pressing. Adsorbate vibrational excitations are a unique descriptor of adsorbate/surface interactions. The infrared spectra (IR) associated with activating vibrational excitations can be experimentally measured with high accuracy, captures details of most vibrational modes, and can be obtained operando. Carbon monoxide (CO) is commonly used as a probe molecule because its normal modes are associated with visually identifiable frequencies and its binding geometry to surfaces is well understood. In this work, we discuss the theory and methods of modeling the IR of adsorbates, using carbon monoxide on transition metal surfaces as an example. We explain the structure effects to CO frequencies via d-band theory and the site-projected density of states. The structural descriptors for which IR can provide information arise naturally from this theory. Due to the indirect effects of temperature and pressure on frequencies via lateral interactions, there is also significant interest in understanding coverage shifts to frequency on complex surfaces. Experimental isotopic labeling studies of carbon monoxide have tried to elucidate coverage effects on IR spectra resulting from occupation of multiple sites, yet computational studies to systematically understand these effects are lacking. The large number of possible adsorptions sites required to simulate real, complex spectra prohibits their direct modeling for characterization purposes via first-principles. We therefore combine the theory developed here with ideas from isotopic studies to develop a physics-based surrogate model to generate IR spectra associated with high coverage of CO on complex surfaces. Our approach for generating complex IR spectra enables physics-based constraints to be embedded in machine learning models for predicting local structure of adsorption sites and their occupancy from experimental IR spectra with very high accuracy.