Van Emden Henson Panayot Vassilevski Center for Applied Scientific Computing Lawrence Livermore National Laboratory Element-Free AMGe: General algorithms for computing interpolation weights in AMG International Workshop on Algebraic Multigrid Sankt Wolfgang, Austria June 27, 2000

VEH 2 CASC Overview l AMG and AMGe l Element-Free AMGe: an interpolation rule based on neighborhood extensions l Examples of extensions: — -extension, — -extension — extensions from minimizing quadratic functionals l Numerical experiments This work was performed under the auspices of the U. S. Department of Energy by the University of California Lawrence Livermore National Laboratory under contract number: W-7405-Eng-48. UCRL-VG

VEH 3 CASC AMG and AMGe l Assume a given sparse matrix l AMG, or Algebraic Multigrid is MG based only on the matrix entries. l Essential components of AMG: — a set,, of fine-grid degrees of freedom (dofs) — a coarse grid, ; typically a subset of — a prolongation operator — smoothing iterations; typically Gauß-Seidel or Jacobi — a coarse matrix given by

VEH 4 CASC Coarse-grid selection l There are several ways to select the coarse grid l is typically a maximal independent set l each fine-grid dof is typically interpolated from a subset of its coarse nearest neighbors

VEH 5 CASC Building the prolongation, P l Let be a fine-grid dof l Let be a neighborhood of l Let be the coarse-grid dofs used to interpolate a value at }}}}}} Examine rows of corresponding to.

VEH 6 CASC l is the minimal neighborhood of l Replace with modified version, — by adding to all off-diagonal entries in i th row that are weakly connected to — second, in all rows j for dofs strongly connected to i : –set –set off-diagonals to zero Prolongation in classical AMG l ith row of P is ith row of

VEH 7 CASC l Key idea of AMGe: We can use information carried in the element stiffness matrices to determine — the nature of the smooth error components — accurate interpolation operators — the selection of the coarse grids AMGe differs from standard AMG by using finite element information l Traditional AMG uses the following heuristic (based on M-matrices): smooth error varies slowest in the direction of “large” coefficients l New heuristic based on multigrid theory: interpolation must be able to reproduce a mode with error proportional to the size of the associated eigenvalue

VEH 8 CASC AMGe uses finite element stiffness matrices to localize the new heuristic l Local measure: where are canonical basis vectors, and are sums of local stiffness matrices l If all local measures are small, the global measure is bounded and small good convergence! l Then solving a small local problem yields a row of the optimal interpolation (for the given set of C- points).

VEH 9 CASC AMGe uses a small local problem to define prolongation l We can show that (for a given set of interpolation points), the “optimal” prolongation is the set of weights Q that satisfy the min/max problem: l Furthermore, solving the min/max problem is exactly equivalent to the following small matrix formulation

VEH 10 CASC Prolongation in AMGe l The neighborhood is the union of the elements having i as vertex l We use the assembled neighborhood matrix l Partition the neighborhood matrix as before l ith row of P is ith row of }}}}

VEH 11 CASC Prolongation in AMGe, cont. l Note that there is no need to modify l Knowledge of the element matrices (used to create the assembled neighborhood matrix) carries with it implicitly the correct assignment and treatment of “weak” and “strong” connection. This is the main contribution of AMGe methods l AMGe produces superior prolongation. The goal of this work is to accomplish the superior prolongation without the knowledge of the element matrices

VEH 12 CASC AMGe - Richardson vs Gauß-Seidel Richardson (1,0) cycle Gauß-Seidel (1,0) cycle

VEH 13 CASC Prolongation in Element-free AMGe: based on extensions l Let be the f-point to which we wish to interpolate

VEH 14 CASC Prolongation in Element-free AMGe: based on extensions l Let be the f-point to which we wish to interpolate l is the set of points in the neighborhood of

VEH 15 CASC Prolongation in Element-free AMGe: based on extensions l Let be the f-point to which we wish to interpolate l is the set of points in the neighborhood of l is the set of coarse nearest neighbors of i

VEH 16 CASC Prolongation in Element-free AMGe: based on extensions l Define, the set of “exterior” points for the neighborhood of : the set of points j such that j is connected to a fine point in the neighborhood of

VEH 17 CASC Prolongation in Element-free AMGe: based on extensions l Define, the set of “exterior” points for the neighborhood of : the set of points j such that j is connected to a fine point in the neighborhood of

VEH 18 CASC Prolongation in Element-free AMGe: based on extensions l We use the following window of the matrix A where we will only be interested in the blocks shown. everything else on grid

VEH 19 CASC Prolongation in Element-free AMGe: based on extensions l Assume that an extension mapping is available: i.e., we interpolate the exterior dofs (“X”) from the interior dofs f and c, by the rule

VEH 20 CASC Prolongation in Element-free AMGe: based on extensions l We construct the prolongation operator on the basis of the modified matrix that is, and

VEH 21 CASC Prolongation in Element-free AMGe: based on extensions l Then the ith row of the prolongation matrix P is taken as the ith row of the matrix l To make use of this method we must determine useful ways in which to build the extension operator

VEH 22 CASC Examples of extension: A-extension l Given defined on, we wish to extend it to defined on. l Let be an exterior dof and define to be entries of A to which is connected. l.The A-extension is:

VEH 23 CASC Examples of extension: -extension l Given defined on, we wish to extend it to defined on. l Let be an exterior dof and define to be entries of A to which is connected. l The -extension is (a simple average):

VEH 24 CASC Examples of extensions: based on minimizing a quadratic functional l This is the method Achi Brandt proposed in l Given defined on, find the extension by finding where l This is a “simultaneous” extension, and is more expensive than -extension or -extension

VEH 25 CASC Examples of extensions: minimizing a “cut-off” quadratic functional l Given defined on, find the extension that satisfies where and is a diagonal matrix. A good choice is the vector note that this is also a “simultaneous” extension (but less expensive than the “full” quadratic)

VEH 26 CASC Classical AMG viewed as extension l The classical (Ruge-Stüben) AMG corresponds to selecting and defining an -extension by setting if is weakly connected to, and setting if is strongly connected to

VEH 27 CASC Properties of the extensions l All the extensions described ensure that if is constant in then the extension is the same constant in, an important property for second-order elliptic PDEs. l For systems of PDEs we use the same extension mappings, based on the blocks of A associated with a given physical variable. That is, the extension to a dof corresponding to the physical variable k is based on dofs from that describe the same physical variable k.

VEH 28 CASC A simple example: the stretched quadrilateral l Consider the 2-D Laplacian operator, finite- element formulation on regular quadrialteral elements that are greatly stretched: l As, the operator stencil tends to:

VEH 29 CASC A simple example: the stretched quadrilateral: geometric approach l For this problem the standard geometric multigrid approach is to semicoarsen: l And the interpolation stencil is:

VEH 30 CASC A simple example: the stretched quadrilateral: the AMG stencil l A simple calculation shows that classical AMG yields the interpolation stencil:

VEH 31 CASC A simple example: the stretched quadrilateral: the neighborhood l Let l Define then N S NW SW SE NE E W

VEH 32 CASC A simple example: the stretched quadrilateral: the A-extension l For the A-extension the extension operators are: from which, and yielding the interpolation stencil:

VEH 33 CASC A simple example: the stretched quadrilateral: the -extension l A similar calculation for the weights using the extension yields the interpolation stencil:

VEH 34 CASC A simple example: the stretched quadrilateral: the cut-off quadratic min l For the cut-off extension the extension operators are: from which, and yielding the interpolation stencil:

VEH 35 CASC The stencil produced by various extension methods: l Classical AMG l A-extension l L2-extension l cut-off quadratic

VEH 36 CASC Numerical experiments l Second order elliptic operator — Unstructured triangular mesh (400, 1600, 6400, and fine-grid elements) — coarsened using agglomeration method of Jones & Vassilevski

VEH 37 CASC Element Agglomeration: Use graph theory to create coarse elements first, then select coarse-grid by abstracting geometric concepts of face, edge, vertex AMGe requires coarse-grid elements & stiffness matrices.

VEH 38 CASC Agglomeration coarsening l Finest grid — 1600 elements — 861 dofs — 5781 nonzero entries

VEH 39 CASC Agglomeration coarsening l 1st coarse grid — 382 elements — 330 dofs — 2602 nonzero entries

VEH 40 CASC Agglomeration coarsening l 2nd coarse grid — 93 elements — 143 dofs — 1397 nonzero entries

VEH 41 CASC Agglomeration coarsening l 3rd coarse grid — 33 elements — 64 dofs — 634 nonzero entries

VEH 42 CASC Agglomeration coarsening l 4th coarse grid — 15 elements — 32 dofs — 304 nonzero entries

VEH 43 CASC Agglomeration coarsening l 5th coarse grid — 7 elements — 16 dofs — 126 nonzero entries

VEH 44 CASC Agglomeration coarsening l 6th coarse grid — 3 elements — 8 dofs — 46 nonzero entries

VEH 45 CASC Agglomeration coarsening l coarsest grid — 1 elements — 4 dofs — 16 nonzero entries

VEH 46 CASC Coarsening history, unstructured 2nd order PDE l unstructured triangular grid l 2nd order PDE

VEH 47 CASC Convergence results, unstructured 2nd order PDE l 2nd order PDE problem — unstructured triangular grid — V(1,1) cycles — Gauß-Seidel smoothing

VEH 48 CASC Numerical experiments l 2d elasticity (including thin-body) — structured quadrilateral grid, — is the beam thickness — coarsened using agglomeration method — the same coarse-grid is used for AMGe (vertices of agglomerated elements)

VEH 49 CASC Coarsening history, elasticity problem l Structured rectangular grid l 2-d elasticity l d = 1

VEH 50 CASC Coarsening history, elasticity problem l Structured rectangular grid l 2-d elasticity l thin body — d = 0.05

VEH 51 CASC Convergence results, elasticity l 2d elasticity problem — d = 1 — V(1,1) cycles — Gauß-Seidel smoothing

VEH 52 CASC Convergence results, elasticity l 2d thin-body elasticity problem — d = 0.05 — V(1,1) cycles — Gauß-Seidel smoothing

VEH 53 CASC Conclusions l AMGe produces superior prolongation, but requires extra information, i.e., the element matrices l For purely algebraic problems, element-free AMGe provides a technique for building pseudo-element matrices based on several extension schemes l Preliminary experimental results suggest that element-free AMGe also produces superior prolongation, competitive to AMGe results l Systems of PDEs can be handled in a very natural way

VEH 54 CASC Some papers of interest l Pre/Re-prints Available at l V. Henson and P. Vassilevski, “Element-free AMGe: General Algorithms for computing Interpolation Weights in AMG,” submitted to SIAM Journal on Scientific Computing. l M. Brezina, A. Cleary, R. Falgout, V. Henson, J. Jones, T. Manteuffel, S. McCormick, and J. Ruge, “Algebraic Multigrid Based on Element Interpolation (AMGe),” to appear in SIAM Journal on Scientific Computing. l J. Jones and P. Vassilevski, “AMGe Based on Element Agglomeration,” submitted to SIAM Journal on Scientific Computing. l A. Cleary, R. Falgout, V. Henson, J. Jones, T. Manteuffel, S. McCormick, G. Miranda, and J. Ruge, “Robustness and Scalability of Algebraic Multigrid,” SIAM Journal on Scientific Computing, vol. 21 no. 5, pp , 2000 l A. Cleary, R. Falgout, V. Henson, J. Jones, “Coarse-Grid Selection for Parallel Algebraic Multigrid,” in Proc. of the Fifth International Symposium on: Solving Irregularly Structured Problems in Parallel, Lecture Notes in Computer Science, Springer-Verlag, New York, 1998.