Dresden 2017 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

DY: Fachverband Dynamik und Statistische Physik

DY 25: Posters - Soft Particles, Microswimmers, Microfluidics

DY 25.3: Poster

Dienstag, 21. März 2017, 18:15–21:00, P3

Reduced-order hybrid multiscale method combining the regularized data from MD simulations with the macroscopic flow solver — Nehzat Emamy¹, Leonid Yelash¹, Mária Lukácová¹, •Stefanie Stalter², and Peter Virnau² — ¹Institut für Physik, JGU Mainz, Staudingerweg 7, 55128 Mainz — ²Institut für Mathematik, JGU Mainz, Staudingerweg 9, 55128 Mainz

We introduce a reduced-order method to simulate the dynamics of complex materials (e.g. polymer/colloid systems) combining the continuum and atomistic descriptions. We follow the framework of heterogenuous multi-scale method (HMM), separating the scales to macro- and micro-levels. On the macro-level, the governing equations of the incompressible flow are the continuity and momentum equations, which are solved using a high-order accurate discontinuous Galerkin Finite Element Method (dG) and implemented in the BoSSS code. The missing information on the macro-level to solve the momentum equation is the stress tensor, which is computed from molecular Dynamics (MD) simulations on the micro-level. The data obtained from the MD simulations underlie relatively large stochastic errors, which can be controlled by means of the least square approximation. In order to reduce a large number of MD runs for the coupled simulations, we split the computations into an offline phase of expensive training and an online phase of fast multiple queries. In the training phase, we use the Greedy sampling algorithm as a model reduction technique to replace the nonlinear functionality of the stress tensor on the strain by a smooth low-dimensional reliable approximation.