Abstract In this paper we introduce the multiresolution LU factorization of non-standard forms (NS-forms) and develop fast direct multiresolution methods for solving systems of linear algebraic equations arising in elliptic problems. The NS-form has been shown to provide a sparse representation for a wide class of operators, including those arising in strictly elliptic problems. For example, Green's functions of such operators (which are ordinarily represented by dense matrices, e.g., of size N by N ) may be represented by −log ϵ· N coefficients, where ϵ is the desired accuracy. The NS-form is not an ordinary[fn9] matrix representation and the usual operations such as multiplication of a vector by the NS-form are different from the standard matrix–vector multiplication. We show that (up to a fixed but arbitrary accuracy) the sparsity of the LU factorization is maintained on any finite number of scales for self-adjoint strictly elliptic operators and their inverses. Moreover, the condition number of matrices for which we compute the usual LU factorization at different scales is O (1). The direct multiresolution solver presents, therefore, an alternative to a multigrid approach and may be interpreted as a multigrid method with a single V-cycle. For self-adjoint strictly elliptic operators the multiresolution LU factorization requires only O ((−log ϵ) 2 ·N ) operations. Combined with O ( N ) procedures of multiresolution forward and back substitutions, it yields a fast direct multiresolution solver. We also describe direct methods for solving matrix equations and demonstrate how to construct the inverse in O ( N ) operations (up to a fixed but arbitrary accuracy). We present several numerical examples which illustrate the algorithms developed in the paper. Finally, we outline several directions for generalization of our algorithms. In particular, we note that the multidimensional versions of the multiresolution LU factorization maintain sparsity, unlike the usual LU factorization.
[1]
G. Beylkin,et al.
A Multiresolution Strategy for Reduction of Elliptic PDEs and Eigenvalue Problems
,
1998
.
[2]
G. Schulz.
Iterative Berechung der reziproken Matrix
,
1933
.
[3]
Ronald R. Coifman,et al.
Wavelet-Like Bases for the Fast Solution of Second-Kind Integral Equations
,
1993,
SIAM J. Sci. Comput..
[4]
Y. Meyer.
Wavelets and Operators
,
1993
.
[5]
Adi Ben-Israel,et al.
On Iterative Computation of Generalized Inverses and Associated Projections
,
1966
.
[6]
S. Mallat.
Multiresolution approximations and wavelet orthonormal bases of L^2(R)
,
1989
.
[7]
Gregory Beylkin.
Wavelets, multiresolution analysis and fast numerical algorithms
,
1996
.
[8]
I. Duff,et al.
Direct Methods for Sparse Matrices
,
1987
.
[9]
R. Coifman,et al.
Fast wavelet transforms and numerical algorithms I
,
1991
.
[10]
G. Beylkin,et al.
A Multiresolution Strategy for Numerical Homogenization
,
1995
.
[11]
Andreas Rieder,et al.
Wavelets: Theory and Applications
,
1997
.
[12]
Charles K. Chui,et al.
Cholesky factorization of positive definite bi-infinite matrices
,
1982
.
[13]
G. Stewart,et al.
On the Numerical Properties of an Iterative Method for Computing the Moore–Penrose Generalized Inverse
,
1974
.
[14]
Stéphane Jaffard.
Propriétés des matrices « bien localisées » près de leur diagonale et quelques applications
,
1990
.