(732o) Deducing Subnanometer Heterogeneous Supported Catalyst Structure from Infrared Spectroscopy

"Adsorbate vibrational excitations, probed via infrared (IR) spectroscopy, are sensitive to adsorbate/metal interactions, accurate, and easily obtainable in-situ or operando. Most IR-based peaks, however, are typically assigned heuristically for relatively simple spectra following the gold standard of well-defined single crystals. Deducing the structure of highly dispersed heterogeneous catalysts consisting of single-atoms and ultra-small clusters directly from IR spectra has been lacking and is imperative for intelligent catalyst design for structure sensitive chemistries.

Here, we present a computational framework to characterize supported single-atoms and subnanometer clusters from adsorbate vibrational excitations determined from IR spectroscopy. We bypass the vast combinatorial space of brute-force spectra matching by determining a viable, low-energy ensemble of structures using machine-learned Hamiltonians, genetic algorithm optimization, and grand canonical Monte Carlo calculations that contributes maximally to the final spectral intensity. We obtain first-principles vibrations on this tractable ensemble and generate single-cluster primary spectra analogous to pure component gas-phase IR spectra. These spectra act as calibration standards that allow us to predict cluster size and shape distributions, and quantify uncertainty, within the Bayesian Inference (BI) framework directly from experimental and computational data. We demonstrate the efficacy of our BI deconvolution methodology on a case study of CO adsorption on Pd/CeO2(111) catalysts and find that our conclusions are consistent with those made from other characterization techniques. Finally, we discuss extensions for characterizing complex materials towards closing the materials gap. "