Microkinetic modeling with machine learning predicted binding energies of reaction intermediates of ethanol steam reforming: The limitations

Abstract

Machine learning based methods have shown great potential in binding energy predictions for surface species, and root mean squared error (RMSE) has been considered as an important indicator to assess the accuracy of the model. In this work, error propagation from energetic prediction to kinetic properties evaluation, e.g. dominant reaction pathways and reaction activities, was assessed and quantified for ethanol steam reforming reaction. The results show that the identification of the dominant pathway is independent of the RMSE of the energetic prediction model but closely related to the predicted energetics of competing steps. The reaction rate is influenced by two factors, namely energetics of the rate-determining step and free site coverage at steady state. Interestingly, the rate-determining step alone can dominate the reaction activity within high temperature range, whereas a global error indicator regarding all surface intermediates must be taken into account in the low temperature range. This work put a straightforward example of error propagation, showing that the RMSE of the energy prediction model cannot guarantee the accuracy of evaluations of kinetic properties.