Advancing Kinetic Study of Catalytic Reaction: A Hybrid Modeling Approach for Predicting Effective Activation Energies

This study addresses limitations of traditional kinetic Monte Carlo (kMC) simulations, particularly their inability to capture latent surface dynamics on electrocatalysts due to complex many-body interactions among adsorbates, reactants, and intermediates. These shortcomings limit their predictive accuracy, often exacerbated by the separate limitations of density functional theory (DFT) in calculating activation energies (Ea) accurately. To overcome these challenges, we introduced a hybrid model that combines advanced machine learning (ML) with first-principles kMC. In this work, our approach defines “effective” activation energies incorporating nuanced physical phenomena that are absent in conventional kinetic models but evident in experiments. This hybrid model leverages these predictive effective activation energies to unveil underlying chemical phenomena occurring on catalyst surfaces. These phenomena include changes in product generation, surface coverage, and the dominant reaction mechanisms over time, which have been validated through experimental outcomes using MXene catalysts. Additionally, the ML component of our model not only provides an empirical fit but also infers underlying parameters that guide the subsequent DFT calculations based on changeable surface coverage over time.