(234e) Improving the Efficiency of Kinetic Monte Carlo Simulations for Catalysis with a Parallel Caching Algorithm

Kinetic modelling methods have proven important for the fundamental understanding of underlying mechanisms that give rise to catalysis, but also for the prediction of catalytic performance metrics, such as activity, selectivity and stability (e.g. poisoning resistance). Kinetic Monte Carlo (KMC) approaches have shown exceptional promise in tackling complexities relevant to the effect of multiple adsorbate configurations, lateral interactions, adlayer structure, and the presence of spectator species, to name a few factors, on activity. The main issue with KMC simulations is that they are computationally expensive.

In this talk, we will discuss an efficient implementation of the graph theoretical KMC framework¹ that employs a caching scheme with OpenMP parallelisation to improve computational performance. This KMC framework incorporates Brønsted-Evans-Polanyi (linear) relations to calculate an activation barrier from the reaction energy, the latter evaluated as the difference in the initial and final state species of a given elementary event, in the presence of interactions with spectators. Thus, after a reaction has occurred, the removal of reactants and addition of products can affect the barriers (and thus rates) of other reactions in the neighbourhood of the one that just happened. The key concept behind the caching scheme is that, in order to update the barriers of the “affected reactions” one only needs to detect interactions between the product species of the “just-executed” reaction and the products of the “affected reactions”. Remaining interaction patterns, between the products of the “affected reactions” and other spectators are cached in a datastructure. Benchmarks on a model of the NO oxidation/NO₂ reduction reaction on Pt, show that the parallel caching scheme can exhibit up to 20x improvement in efficiency compared to a single processor run.

References

1. Nielsen, W., d’Avezac, M., Hetherington, J. and Stamatakis (2013). “Parallel kinetic Monte Carlo simulation framework incorporating accurate models of adsorbate lateral interactions”. The Journal of Chemical Physics, 139(22): 224706.