Local Fourier k-grid (k=1,2,3) analysis for Navier-Stokes-type systems

R. Wienands and C.W. Oosterlee

GMD - Institute for Algorithms and Scientific Computing (SCAI)
Schloss Birlinghoven, D-53754 Sankt Augustin, Germany


Local Fourier one- (smoothing) and two-grid analysis (LFA) [1,2,6] are well-known tools for the quantitative analysis and the design of efficient multigrid methods for several problems. The two-grid analysis is the basis for classical asymptotic multigrid convergence estimates [6]. Moreover, it is the main analysis tool for nonsymmetric problems. For several multigrid components or cycle variants, however, the overall multigrid convergence cannot be approximated accurately by two-grid factors. For example, if one is interested in the varying performance of V- and W-cycles, of pre- and post-smoothing, of different smoothers on different grids or of different discretizations on different grids one has to consider at least one additional grid leading to a three-grid analysis [8].

Although LFA is often applied to scalar equations, two-grid values for systems of equations are rarely found in the literature and three-grid values are completely missing. To close this gap we developed a freely available, general Fourier analysis program - LFA00_2D - which will be presented in the first part of the talk. LFA00_2D is a set of Fortran77 subroutines to perform Fourier k-grid (k=1,2,3) analysis for two-dimensional systems of PDEs yielding measures of h-ellipticity [1], smoothing factors, two- and three-grid convergence factors and norms of the two- and three-grid operators. Several discretizations of well-known systems, like the biharmonic system, the Stokes equations, the Oseen equations (linearized Navier-Stokes equations) have already been implemented. Furthermore, an option to analyze your own linearized system is available. LFA00_2D supports a large variety of multigrid components including

It is, furthermore, possible to estimate convergence factors if the two- or three-grid operators are used as preconditioners for GMRES [7].

In the second part of the talk, we evaluate several non-staggered discretizations for Navier-Stokes-type systems (w.r.t. an efficient multigrid treatment) in order to demonstrate the usefulness of the analysis program. Here, we focus on second order discretizations. In particular, the Stokes and the Oseen equations are investigated, discretized by central differences with an artificial pressure correction term in the continuity equation (for low Reynolds numbers) or by higher order upwind schemes like Dick's flux difference splitting [4] using van Leer's kappa-scheme [5] (for high Reynolds numbers).
The Fourier results are confirmed by numerical experiments with the incompressible Navier-Stokes equations.

  1. A. Brandt, Multigrid techniques: 1984 guide with applications to fluid dynamics, GMD-Studie Nr. 85, Sankt Augustin, Germany (1984).
  2. A. Brandt, Rigorous quantitative analysis of multigrid, I: Constant coefficients two-level cycle with L2-norm, SIAM J. Numer. Anal., 31 (1994), pp. 1695-1730.
  3. J.E. Dendy Jr., Blackbox multigrid for nonsymmetric problems, Appl. Math. Comput., 13 (1983), pp. 261-283.
  4. E. Dick and J. Linden, A multigrid method for steady incompressible Navier-Stokes equations based on flux difference splitting, Int. J. Num. Meth. Fluids, 14 (1992), pp. 1311-1323.
  5. B. van Leer, Upwind-difference methods for aerodynamic problems governed by the Euler equations, In: Large Scale Computations in Fluid Mechanics, Lectures in Appl. Math. 22, II, B. Enquist, S. Osher, and R. Somerville (eds.), AMS, Providence, RI, 1985, pp. 327-336.
  6. U. Trottenberg, C.W. Oosterlee, and A. Schüller, Multigrid, Academic Press (2001).
  7. R. Wienands, C.W. Oosterlee, and T. Washio, Fourier analysis of GMRES(m) preconditioned by multigrid, SIAM J. Sci. Comput., 22 (2000), pp. 582-603.
  8. R. Wienands and C.W. Oosterlee, On three-grid Fourier analysis for multigrid, to appear in SIAM J. Sci. Comput. (2001).
  9. P.M. de Zeeuw, Matrix-dependent prolongations and restrictions in a blackbox multigrid solver, J. Comput. Appl. Math., 33 (1990), pp. 1-27.