Optimization of two-level methods for DG discretizations of reaction-diffusion equations

Abstract

In this manuscript, two-level methods applied to a symmetric   interior penalty discontinuous Galerkin finite element discretization   of a singularly perturbed reaction-diffusion equation are analyzed.   Previous analyses of such methods have been performed numerically by   Hemker et al. for the Poisson problem.   The main innovation in this work is that explicit formulas for the   optimal relaxation parameter of the two-level method for the Poisson   problem in 1D are obtained, as well as very accurate closed form   approximation formulas for the optimal choice in the   reaction-diffusion case in all regimes.   Using Local Fourier Analysis, performed at the matrix level to make   it more accessible to the linear algebra community, it is shown that   for DG penalization parameter values used in practice, it is better to   use cell block-Jacobi smoothers of Schwarz type, in contrast to   earlier results suggesting that point block-Jacobi smoothers   are preferable, based on a smoothing analysis alone.   The analysis also reveals how the performance of the iterative   solver depends on the DG penalization parameter, and what value should   be chosen to get the fastest iterative solver, providing a new, direct   link between DG discretization and iterative solver performance.   Numerical experiments and comparisons show the applicability of the   expressions obtained in higher dimensions and more general geometries.