(560ig) Enhancing Ab Initio Microkinetic Models with Machine Learning

Mean-filed microkinetic models (MF-MKM) based on energies derived from density functional theory (DFT) are widely used to study heterogenous catalytic system, in terms of elucidation of mechanisms and quantification of reaction fluxes of catalytic systems. However, the model usually deviates from kinetic experiments, due to the intrinsic error of DFT calculation, and over-simplified mean-field approximation, etc.

Here, we present two examples of the application of statistical learning in kinetic modeling. (I) We demonstrate the improvement of a microkinetic model by statistically calibration the intrinsic error of DFT energies, and further improvement by incorporating less experimental data. (II) After accounting the influence of the adsorbate-adsorbate interaction with Monte Carlo simulation and statistical learning, MF-MKM captures missing phenomenon (bistability of oxidation system) missing from mean-field approximation.

We show how this approach systematically improve the prediction of MF-MKM combining machine learning model. This work demonstrates the promising performance of statistical learning in the application of kinetic modeling of heterogenous catalytic system.