From: spellucci@mathematik.th-darmstadt.de (Peter Spellucci)
Newsgroups: sci.math.num-analysis
Subject: Re: linear sparse matrix solver
Date: 25 Mar 1998 12:32:02 GMT

In article <35180B4A.6D69@appl2.hrz.uni-siegen.de>, "R. Kaczmarcik" <Kacz@appl2.hrz.uni-siegen.de> writes:
|> Hello all,
|> I'm searching for source Code in C/C++ to solve a large set of linear
|> equations (A*x=B), where A is in worst case a 90.000 by 90.000 matrix.
|> The Matrix is symmetric, has no explicit band-structure and is very
|> sparse (only 10E-4 of its positions are unequal zero).
for lu-decomposition use the meschach library in netlib.
there exists also the possibility to use iterative methods, e.g. 
conjugate gradient with preconditioning or some code from the qmr-suite
(e.g. from netlib/linalg). for cg you need positive definiteness (you didn't
mention that). the problem with iterative solvers is that they usually are 
(very) slow without a preconditioner and in order to get a useful preconditioner
you need some (a lot of) knowledge concerning the properties of A.
hope this helps
peter