(659d) Bayesian Experimental Design and Mean Field Microkinetic Modeling of Heterogeneous Catalytic Systems

Reaction kinetics experiments are valuable in providing mechanistic insights of a catalytic system, however, they can be time-consuming and expensive and may not be able to discriminate between plausible reaction pathways. Microkinetic models, using ab initio energetics, can in principle provide information of a catalytic system a priori; however, the intrinsic errors in modern ab initio methods (such as density functional theory) can result in significant errors in prediction of rates (a factor of 10 – 100 and sometimes higher). We propose a modeling framework that integrates experimental kinetic data and microkinetic modeling, to sequentially identify and select the most “informative” data to build a robust mechanistic model.

Our framework uses a Gaussian Process (GP) model to quantify the uncertainty in energies of the chosen functional (e.g. PW91) by benchmarking with experimental adsorption energies, and propagates the errors through a microkinetic model formulated using DFT energies using the same functional. Then, Bayesian experimental design is employed to rank order candidate experimental conditions, and sequentially pick informative experiments to refine the model through Bayesian inference.

In this talk, the formulation of the GP model, the error propagation, and the experimental design strategies will be presented and discussed in the context of an illustrative example involving the low temperature water gas shift (WGS) reaction on copper.