Regensburg 2016 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 22: Topical session: Integrated computational materials engineering for design of new materials V
MM 22.3: Vortrag
Dienstag, 8. März 2016, 12:15–12:30, H39
An adaptive sparse grid approach for quantifying the error propagation in first principles kinetic Monte Carlo models — •Sandra Döpking1, Daniel Strobusch2, Christoph Scheurer2, and Sebastian Matera1 — 1FU Berlin, Germany — 2TU München, Germany
First principles kinetic Monte Carlo (1p-kMC) models utilize Density Functional Theory (DFT) based rate constants (RCs) as the input for stochastic mesoscopic reactivity models. Despite the success of DFT, the employed approximations can easily lead to uncertainties in the RCs of two orders of magnitude. We quantify the error propagation to the 1p-kMC simulation results by a global sensitivity analysis (GSA), i.e. we decompose the error in the result into contributions caused by each uncertain RC. The required high dimensional integrals over the space of the RCs are numerically evaluated using locally and dimension-adaptive sparse grids. With this approach, we can exploit that, typically, strongly non-linear behavior only appears in small sub-domains (local adaptivity) and on low-dimensional sub-spaces (dimension-adaptivity) of the whole parameter space. As a prototypical example, we consider the 1p-kMC model for the CO oxidation on a RuO2(110) surface. Comparing with local (linear) sensitivity analysis, we find qualitative differences between both approaches, but also in the global setting only a small portion of all RCs has an impact at all. Possible applications of the approach are the reduction of the number of expensive first principles simulations or the identification of descriptor sets in computational materials screening.