One of the major challenges in multiscale modeling of catalytic processes is the improvement of the accuracy of kinetic models for surface-catalyzed reactions. The kinetics of surface reactions are often not described well in standard mean-field kinetics, which can lead to substantial errors in cases where adsorbates are not well-mixed on the surface or spatial correlations limit the overall reaction kinetics. As a possible solution to solve this predicament, Kinetic Monte Carlo (KMC) Simulations based on Density Functional Theory (DFT) calculations have been proposed. However, as the results from KMC simulations are sensitive to the spatial arrangement of adsorbates on the surface, new errors emerge that stem from neglect or bad estimates of lateral interactions. Obtaining good estimates for lateral interactions and incorporating them in KMC are therefore vital for the improvement of multiscale modeling approaches for catalysis applications.<br/>Lateral interactions can be included in DFT-based KMC simulations using Cluster Expansion (CE) based approaches where the adsorption energy is expressed as a sum of interaction parameters. However, using the largest possible number of parameters is not always a good approach as the individual parameters become less reliable (overfitting) and the predictive capability of the model declines if too many parameters are used. We will demonstrate how to obtain meaningful lateral interaction parameters from DFT calculations using statistical descriptors and how to troubleshoot artifacts arising on the KMC level. In addition, CE models beyond first-nearest neighbor interactions tend to have many terms, which makes them slow to evaluate during the KMC simulations. To mediate this performance issue, we introduce the Supercluster approach, where parts of the CE are summed together and precalculated. With this performance boost, simulations with complex interaction models become computationally feasible to run in serial operation on HPC clusters or workstations.

Overcoming performance and accuracy challenges: How to incorporate lateral interactions of adsorbates into kinetic Monte Carlo simulations for kinetic modeling

ACS Fall 2020 Virtual Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.