Algebraic multilevel iterative preconditioning methods for h-matrices (Q4951746)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 1442154
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | Algebraic multilevel iterative preconditioning methods for h-matrices |
scientific article; zbMATH DE number 1442154 |
Statements
Algebraic multilevel iterative preconditioning methods for h-matrices (English)
0 references
22 November 2000
0 references
iterative methods
0 references
preconditioners
0 references
algebraic multilevel iteration method
0 references
\(H\)-matrices
0 references
\(H\)-compatible splitting
0 references
numerical examples
0 references
diagonal compensation reduction
0 references
convergence
0 references
0.8434917330741882
0 references
0.8375199437141418
0 references
0.8370728492736816
0 references
0.8310337066650391
0 references
Algebraic multilevel iteration (AMLI) preconditioners for \(H\)-matrices are presented. The idea for defining these preconditioners is an extension of the idea proposed by \textit{O. Axelsson} and \textit{M. Neytcheva} [Numer. Linear Algebra Appl. 1, No. 3, 213-236 (1994; Zbl 0837.65024)] for Stieltjes matrices. An AMLI preconditioner \(M^{(\ell)}\) for a \((2 \times 2)\)-block matrix \(A^{(\ell)}= [A_{ij}^{(\ell)}]_{i,j=1}^2\) is recursively defined by \[ M^{(k+1)} = \left(\begin{matrix} (B_{11}^{(k+1)})^{-1} & 0\\ A_{21}^{(k+1)} & I\end{matrix} \right) \left(\begin{matrix} I & B_{11}^{(k+1)}A_{12}^{(k+1)}\\ 0 & S^{(k)}\end{matrix} \right), \;k = 0,1, \ldots,\ell-1, \;M^{(0)} = A^{(0)}, \] with \((S^{(k)})^{-1} = [I - p_{\nu_k}((M^{(k)})^{-1}A^{(k)})](A^{(k)})^{-1}\), where \(p_{\nu_k}\) is a polynomial of degree \(\nu_k\) derived from Chebyshev polynomials [see \textit{O. Axelsson} and \textit{P. S. Vassilevski}, SIAM J. Numer. Anal. 27, No. 6, 1569-1590 (1990; Zbl 0715.65091)]. The matrix \(B_{11}^{(k+1)}\) is an approximation for the inverse of \(A_{11}^{(k+1)}\). In the paper the following way for defining the matrix \(B_{11}^{(k+1)}\) is proposed and analysed. In a first step the matrix \(A_{11}^{(k+1)}\) is approximated by a symmetric positive definite \(H\)-matrix \(W_{11}^{(k+1)}\) which has a simpler sparse structure. Then, the matrix \((W_{11}^{(k+1)})^{-1}\) is approximated by a symmetric positive definite matrix \(B_{11}^{(k+1)}\). The main technique for defining these approximations are \(H\)-compatible splittings and diagonal compensation reduction. It is shown that the application of the proposed preconditioner leads to iterative solvers which are optimal with respect to the rate of convergence and the cost of arithmetical work supposed that the polynomial degree \(\nu_k\) is chosen appropriately. The presented numerical results confirm this theoretical result.
0 references