Quantifying the error propagation in microkinetic modeling of catalytic reactions with model-predicted binding energies

Abstract

Energy prediction models, such as scaling relations, Brønsted–Evans–Polanyi relations and machine-learning prediction models, are widely employed to accelerate the calculations of energies of reaction intermediates and the rational design of catalysts. However, error propagation from predicted binding energies of reaction intermediates and transition states to the calculated reaction rates would result in the misidentification of optimal catalysts. In order to assess and quantify the error propagation in determining kinetic information for catalyst design, here we make energy error simulation based on DFT-calculated binding energies of reaction intermediates within the network of methane dry reforming reaction, then employ these simulated energies to microkinetic modeling. The results suggest that the microkinetic results would have different tolerance to the error indicators of predicted binding energy. Further detailed analyses show that binding energies with low variance are more likely to result in less significant error propagations during the calculation of reaction energy and activation energy, leading to more reliable kinetic information. Finally, several directions are discussed regarding the minimization of error propagation in microkinetic modeling based on predicted binding energies, and suggestions are provided for future studies.