Path: news.mathworks.com!not-for-mail
From: "Alessio Rucci" <alessio.rucci@libero.it>
Newsgroups: comp.soft-sys.matlab
Subject: Re: Solving Ax=b for large sparse matrix(1e+6 X 1e+5) and cholinc failure
Date: Thu, 8 Nov 2007 14:41:51 +0000 (UTC)
Organization: The MathWorks, Inc.
Lines: 36
Message-ID: <fgv77f$hib$1@fred.mathworks.com>
References: <fguq3f$ig1$1@fred.mathworks.com> <fgv6cb$2c5$1@fred.mathworks.com>
Reply-To: "Alessio Rucci" <alessio.rucci@libero.it>
NNTP-Posting-Host: webapp-05-blr.mathworks.com
Content-Type: text/plain; charset="ISO-8859-1"
Content-Transfer-Encoding: 8bit
X-Trace: fred.mathworks.com 1194532911 17995 172.30.248.35 (8 Nov 2007 14:41:51 GMT)
X-Complaints-To: news@mathworks.com
NNTP-Posting-Date: Thu, 8 Nov 2007 14:41:51 +0000 (UTC)
X-Newsreader: MATLAB Central Newsreader 1176818
Xref: news.mathworks.com comp.soft-sys.matlab:436585


"
> >  
> 
> My last post seems to have gotten lost.  Or not.
> I clicked "submit"  but I got sent to the "update your
> profile" page.  So this might be a duplicate.
> 
> Upload the matrix to http://www.cise.ufl.edu/~web-gfs
> and let me take a crack at it.  I'd like to add it to
> my collection (http://www.cise.ufl.edu/research/sparse).
> Send A itself, not A'*A, as a mat file.
> 
> x=(A'*A)\(A'*b) is not a good idea, since it is very
> ill conditioned.  The failure here is CHOLMOD running
> out of memory, however.  It's puzzling that backslash
> then tries MA57.  CHOLMOD and MA57 both use AMD, and
> the latter tends to use a tad bit more memory.  So if
> CHOLMOD+AMD runs out of memory, MA57+AMD almost certainly
> will.
> 
> You should use sparse QR, although that doesn't use
> the BLAS and is thus slower than sparse Cholesky.
> 
> Try using METIS.  There is an interface to the METIS
> ordering method in SuiteSparse on the File Exchange.
> METIS can give better orderings than AMD, although it
> can segfault sometimes :-(
> 
> 
> 
> 
> 
> 

Ok,I'll send you the matrix