Enhancing Predictive Accuracy in Catalysis: A Hybrid Modeling Approach for Dynamic Surface Configuration Analysis

One inherent limitation of density functional theory (DFT) is its inability to precisely capture changes in surface configurations and activation energies, both of which are essential for interpreting reaction kinetics effectively. These shortcomings are further exacerbated when combined with the challenges traditional kinetic Monte Carlo (kMC) simulations face in capturing latent reaction mechanisms on electrocatalysts due to complex many-body interactions among adsorbates, reactants, and intermediates. To tackle these issues, we have developed a hybrid model that integrates machine learning (ML) with first-principles model. This model clearly tracks the changes in surface configurations over time, integrating complex physical phenomena often overlooked by conventional DFT models but observable in experiments. Our approach not only boosts the predictive accuracy of these models but also broadens their applicability in testing catalyst performance under diverse conditions. Intriguingly, the ML component of our model does more than fitting empirical data; it identifies critical parameters that refine subsequent DFT calculations to better match real-world surface conditions and coverages.