Optimal Complex Relaxation Parameters in Multigrid for Complex-Shifted Linear Systems

Robert Hocking and Chen Greif

Complex-valued spatially-varying damping parameters for a model problem (example from Section 6.5 in the paper), visualized using a color wheel.

Abstract:

We derive optimal complex damping parameters minimizing smoothing factors associated with multigrid using red-black successive over-relaxation or damped Jacobi smoothing applied to a class of linear systems arising from discretized linear partial differential equations with a complex shift. Our analysis yields analytical formulas for smoothing factors as a function of the complex damping parameter, which may then be efficiently numerically minimized. Our results are applicable to second-order discretizations in arbitrary dimensions, and generalize earlier work of Irad Yavneh on optimal damping parameters in the real case. Our analysis is based on deriving a novel connection between the performance of SOR as a smoother and as a solver, and is validated by numerical experiments on problems in two and three spatial dimensions, using both vertex- and cell-centered multigrid, with both constant and variable coefficients. In the variable coefficient case we assign different damping parameters to different grids points, which our framework allows us to do efficiently.

Paper

MATLAB code