(411d) Characterization of Supported Subnanometer Clusters Via Computational Infrared Spectroscopy | AIChE

(411d) Characterization of Supported Subnanometer Clusters Via Computational Infrared Spectroscopy


Vlachos, D., University of Delaware - Catalysis Center For Ener
Cohen, M., University of Delaware
Wang, Y., University of Delaware
Heterogenous single-atom catalysts (SACs) have been extensively studied due to their potential for high performance catalytic activity, while simultaneously minimizing noble metal usage. Determining experimentally the stability and morphology of single atoms and subnanometer clusters of few atoms remains a challenge, especially under working conditions. Adsorbate vibrational excitations are often used to determine structural properties, as they are selective to adsorbate/metal interactions. Infrared (IR) spectra associated with activating adsorbate vibrational modes are accurate and can be obtained quickly in situ or operando. Experimental infrared peaks are often assigned heuristically, and are often the gold standards for well-defined single crystals, but they provide few details on structural characterization of subnanometer catalysts.

Here, we present a computational framework to characterize supported single-atoms and subnanometer clusters from adsorbate vibrational excitations determined from IR spectroscopy. We combine data-based approaches with physics-driven surrogate models to generate realistic synthetic IR spectra from first-principles vibrational calculations of carbon monoxide on palladium clusters supported on ceria. Due to the relative expense of first-principles vibrational calculations, their direct use for exploring the vast combinatorial space to directly match experimental spectra is beyond our current computational capabilities. Rather, we utilize calculations of energetically viable structures and apply realistic peak broadening to generate single-cluster primary spectra, analogous to pure component spectra in gas-phase IR spectroscopy. Finally, we perform peak deconvolution of complex synthetic and experimental spectra under the Bayesian Inference framework to predict cluster size distributions and quantify uncertainty. We discuss extensions of this computational methodology for characterization of other complex materials under realistic working conditions.