Multigrid Preconditioners for the Discontinuous Galerkin Spectral Element Method : Construction and Analysis

Sammanfattning: Discontinuous Galerkin (DG) methods offer a great potential for simulations of turbulent and wall bounded flows with complex geometries since these high-order schemes offer a great potential in handling eddies. Recently, space-time DG methods have become more popular. These discretizations result in implicit schemes of high order in both spatial and temporal directions. In particular, we consider a specific DG variant, the DG Spectral Element Method (DG-SEM), which is suitable to construct entropy stable solvers for conservation laws. Since the size of the corresponding nonlinear systems is dependent on the order of the method in all dimensions, the problem arises to efficiently solve these huge nonlinear systems with regards to CPU time as well as memory consumption.Currently, there is a lack of good solvers for three-dimensional DG applications, which is one of the major obstacles why these high order methods are not used in e.g. industry. We suggest to use Jacobian-free Newton- Krylov (JFNK) solvers, which are advantageous in memory minimization. In order to improve the convergence speed of these solvers, an efficient preconditioner needs to be constructed for the Krylov sub-solver. However, if the preconditioner requires the storage of the DG system Jacobian, the favorable memory consumption of the JFNK approach is obsolete.We therefore present a multigrid based preconditioner for the Krylov sub-method which retains the low mem- ory consumption, i.e. a Jacobian-free preconditioner. To achieve this, we make use of an auxiliary first order finite volume replacement operator. With this idea, the original DG mesh is refined but can still be implemented algebraically. As smoother, we consider the pseudo time iteration W3 with a symmetric Gauss-Seidel type approx- imation of the Jacobian. Numerical results are presented demonstrating the potential of the new approach.In order to analyze multigrid preconditioners, a common tool is the Local Fourier Analysis (LFA). For a space- time model problem we present this analysis and its benefits for calculating smoothing and two-grid convergence factors, which give more insight into the efficiency of the multigrid method.