Parallelizing AMGe using domain decomposition

Panayot S. Vassilevski

Center for Applied Scientific Computing, Lawrence Livermore National Laboratory, 7000 East Avenue, Mail Stop 560, Livermore, CA 94550


In this talk we will present a natural domain decomposition strategy for parallelizing agglomeration based AMGe (algebraic multigrid for finite element problems exploiting fine-grid element matrices).

The method consists of the following simple steps:
(1) local subdomain solvers (preconditioners); i.e., each mesh subdomain (union of fine grid elements) is assigned to a processor and a sequential AMGe solver is built in there. That is, a sequence of coarse grids (coarse spaces) and respective interpolation matrices are constructed. The setup phase is based on the local bilinear (quadratic) forms (assembled from the subdomain element matrices). Since the local bilinear forms, in the case of model Laplace operator, correspond to a problem with natural (Neumann type) boundary conditions they are typically only semi-definite. Therefore, one builds the sequential AMGe solvers to be defined only for vectors in a space which is a hierarchical complement to a certain coarse space. This coarse space must contain the null-space components (so called rigid body modes) of the local quadratic forms. The version of the AMGe method which we exploit, called spectral agglomerate AMGe, allows for explicit construction of such hierarchical complements and ensures that the null-space components are fully represented on the coarse grid.

(2) a global coarse problem;

The overall preconditioner then is defined as the standard Neumann-Neumann type domain decomposition preconditioner (cf., [3], [2] or [4]), however, here the preconditioner is defined not for the interface Schur complements (as is in the original Neumann-Neumann case) but is applied to the original bilinear forms. As already mentioned, in addition to the subdomain solvers, one incorporates a global coarse solver.

The sequential subdomain coarsening AMGe algorithm consists of an agglomeration step and involves computing a few minimal eigenvectors of the corresponding assembled agglomerate stiffness matrix. The method (similarly to [1] and [5]) requires access to the individual element matrices (in order to assemble certain subdomain matrices). Based on the topological agglomeration algorithm (cf. [5]) we employed and the special interpolation rule chosen, one is able to define coarse elements and coarse element matrices thus allowing for recursion.

Numerical tests illustrating the parallel performance of the algorithm will be presented.


[1] M. Brezina A. J. Cleary, R. D. Falgout, V. E. Henson, J. E. Jones, T. A. Manteuffel, S. F. McCormick, and J. W. Ruge, "Algebraic multigrid based on element interpolation (AMGe)", SIAM J. Sci. Comput. 22(2000), 1570-1592.

[2] J. Mandel, Balancing domain decomposition, Comm. Appl. Numer. Methods 9 (1993) 233-241.

[3] Y.-H. De Roeck and P. Le Tallec, Analysis and a test of a local domain decomposition preconditioner, in: ``Fourth International Symposium on Domain Decomposition Methods for Partial Differential Equations'', (R. Glowinski, Y. Kuznetsov, G. Meurant, J. Periaux, and O. Widlund, eds.) SIAM, Philadelphia, PA 1991.

[4] M. Dryja and O. B. Widlund, Schwarz methods of Neumann-Neumann type for three-dimensional elliptic finite element problems, Comm. Pure Appl. Math. 42(1995), 121-155.

[5] J. E. Jones and P. S. Vassilevski, AMGe based on element agglomeration, SISC (to appear).

This work was performed under the auspices of the U. S. Department of Energy by University of California Lawrence Livermore National Laboratory under contract W-7405-Eng-48.