Introduction to Scientific Computing II Multigrid Miriam Mehl, Michael Bader
Multigrid – Algorithm iterate (GS) on the fine grid restrict residual to the coarse grid solve coarse grid equation for the error interpolate error to the fine grid correct fine grid solution
Multigrid – Things to Choose smoother relation step sizes coarse – fine grid transfer operators restriction interpolation processing order of grid levels
Multigrid – Convergence two grid analysis h-independent convergence for ‚good‘ components
Two Grid – Multigrid Example: 2D Poisson 5-point-stencil h two-grid analysis V-cycle 1/32 0.042 1/64 0.044 1/128 1/256 0.043 1/512 1/1024 1/2048
Multigrid – Some Rules smoother optimal smoothing not(!) optimal convergence small number of smoothing iterations!
Multigrid – Some Rules grid coarsening standard: doubling of h exceptions: anisotropic operators adaptively refined grids unstructured grids/general SLEs
Multigrid – Some Rules restriction/interpolation order of restriction + order of interpolation > order of discretisation
Multigrid – Some Rules V-cycle faster W-cycle more robust
Multigrid – Parallelisation parallel smoothing parallel restriction and interpolation parallel stopping criteria
Multigrid – Parallel Smoothing Gauss Seidel Jacobi-like operation!!! processor 1 processor 2
Multigrid – Parallel Smoothing Gauss Seidel processor 1 processor 2
Multigrid – Parallel Smoothing Gauss Seidel processor 1 processor 2
Multigrid – Parallel Smoothing Gauss Seidel Jacobi-like operation!!! processor 1 processor 2
Multigrid – Parallel Smoothing Gauss Seidel different result than sequential GS!!! processor 1 processor 2
Multigrid – Parallel Smoothing Alternatives: Red-Black Gauss-Seidel processor 1 processor 2
Multigrid – Parallel Smoothing Alternatives: Red-Black Gauss-Seidel processor 1 processor 2
Multigrid – Parallel Smoothing Alternatives: Red-Black Gauss-Seidel processor 1 processor 2
Multigrid – Parallel Smoothing Alternatives: Red-Black Gauss-Seidel processor 1 processor 2
Multigrid – Parallel Smoothing Alternatives: Red-Black Gauss-Seidel processor 1 processor 2
Multigrid – Parallel Smoothing Alternatives: Red-Black Gauss-Seidel
Multigrid – Parallel Smoothing Alternatives: damped Jacobi processor 1 processor 2
Multigrid – Parallel Smoothing Alternatives: damped Jacobi processor 1 processor 2
Multigrid – Parallel Smoothing Alternatives: damped Jacobi processor 1 processor 2
Multigrid – Parallel Smoothing red-black GS: robust and fast smoothing further reading: Irad Yavneh, Multigrid smoothing factors for red-black Gauss-Seidel relaxation applied to a class of elliptic operators, SIAM Journal on Numerical Analysis, 32 (4), 1995 Jun Zhang, Acceleration of five-point red-black Gauss-Seidel in multigrid for Poisson equation, Applied Mathematics and Computation, 80(1), 1996 damped Jacobi: good smoothing
Ferienakademie, Sarntal, Sep 23 – Oct 5, 2012 Universität Erlangen-Nürnberg Technische Universität München Universität Stuttgart Ferienakademie, Sarntal, Sep 23 – Oct 5, 2012 Course 4: Scales and Scalability as Challenges for CSE (Bader, Schweitzer, Wellein) Deadline: Yesterday