Article ID Journal Published Year Pages File Type
6933897 Journal of Computational Physics 2013 18 Pages PDF
Abstract
In this work, we develop a parallel minimum action method for small random perturbations of Navier-Stokes equations to solve the optimization problem given by the large deviation theory. The Freidlin-Wentzell action functional is discretized by hp finite elements in time direction and spectral methods in physical space. A simple diagonal preconditioner is constructed for the nonlinear conjugate gradient solver of the optimization problem. A hybrid parallel strategy based on MPI and OpenMP is developed to improve numerical efficiency. Both h- and p-convergence are obtained when the discretization error from physical space can be neglected. We also present preliminary results for the transition in two-dimensional Poiseuille flow from the base flow to a non-attenuated traveling wave.
Related Topics
Physical Sciences and Engineering Computer Science Computer Science Applications
Authors
,