(234c) Liquid Phase Modeling in Heterogeneous Catalysis | AIChE

(234c) Liquid Phase Modeling in Heterogeneous Catalysis


Saleheen, M. - Presenter, University of South Carolina
Heyden, A., University of South Carolina
The role(s) of solvent in affecting the intrinsic activity and selectivity of active sites, as well as having a significant impact on the reaction mechanism of heterogeneously catalyzed reactions is well-recognized nowadays. Liquid phase processing is attractive for the conversion of highly functionalized lignocellulosic biomass, considering the aqueous condition of the feedstock and their high-water solubility, reactivity, and thermal instability. However, the ability to include the effect of a liquid phase environment on chemical reactions occurring at solid-liquid interfaces poses a distinctive challenge to computational catalysis. Describing such a system from first principles is computationally very demanding, often requiring configuration space sampling of an aqueous phase environment consisting of hundreds if not thousands of molecules.

The main objective of our study is to gain a better understanding of the chemistry of heterogeneously catalyzed reactions at solid-liquid interfaces using both implicit and explicit solvation approaches. As a prototypical reaction with relevance to biomass catalysis, we chose to investigate the hydrodeoxygenation (HDO) of ethylene glycol over Pt(111) and Ru(0001) under aqueous phase processing conditions by employing both our iSMS1 (Implicit Solvation for Metal Surfaces) and eSMS2 (Explicit Solvation for Metal Surfaces) solvation models. Our results indicate that implicit solvation models are inherently restricted by their inability to characterize strong directional hydrogen bonding while the hybrid QM/MM models can predict reasonably accurate solvent effects with an adequate balance between computational expense and chemical accuracy of the reaction system.3 To describe the metal-water interaction at DFT level accuracy, we developed a high dimensional neural network potential using supervised machine learning techniques for representing the interaction between Ru(0001) and water. Mapping a high dimensional continuous potential energy surface allows us to calculate energies and forces for an arbitrary metal-water structure4 and thus, this neural network potential can be used in our hybrid QM/MM solvation formalism for modeling chemical processes at Ru-water interfaces.

(1) Faheem, M.; Suthirakun, S.; Heyden, A. J. Phys. Chem. C 2012, 116, 22458-22462.

(2) Faheem, M.; Heyden, A. J. Chem. Theory Comput. 2014, 10, 3354-3368.

(3) Saleheen, M.; Heyden, A. ACS Catalysis 2018, 8, 2188-2194.

(4) Behler, J. J Phys-Condens Mat 2014, 26, 183001:183001-183024.