Path: news.mathworks.com!not-for-mail
From: "Etienne " <balmes.nospam@sdtools.com.nospam>
Newsgroups: comp.soft-sys.matlab
Subject: Re: FEM substructuring  (static condensation) -> out ot memory; ANSYS can do it
Date: Thu, 24 Jul 2008 16:47:02 +0000 (UTC)
Organization: Ecole Centrale Paris
Lines: 13
Message-ID: <g6abm6$8gu$1@fred.mathworks.com>
References: <g1rovt$8ep$1@fred.mathworks.com> <g1se12$sc2$1@fred.mathworks.com> <g1u2da$can$1@fred.mathworks.com> <g1utuj$bje$1@fred.mathworks.com> <g25rli$8e8$1@fred.mathworks.com> <g25vqp$t7r$1@fred.mathworks.com>
Reply-To: "Etienne " <balmes.nospam@sdtools.com.nospam>
NNTP-Posting-Host: webapp-03-blr.mathworks.com
Content-Type: text/plain; charset="ISO-8859-1"
Content-Transfer-Encoding: 8bit
X-Trace: fred.mathworks.com 1216918022 8734 172.30.248.38 (24 Jul 2008 16:47:02 GMT)
X-Complaints-To: news@mathworks.com
NNTP-Posting-Date: Thu, 24 Jul 2008 16:47:02 +0000 (UTC)
X-Newsreader: MATLAB Central Newsreader 91143
Xref: news.mathworks.com comp.soft-sys.matlab:481579




The way it is done in SDT www.sdtools.com/sdt

is to use a factored matrix object for K_bb^-1 (very 
critical if you want to have 100 000 DOFs). Then, you solve 
for Kbb^-1*Kba by piece by block. Then you project the 
matrices.

You have to realize that 1e5*5000*8/1023^3= 3.7 GB 
If you want that to happen on a 32 bit machine you need to 
go out of core (like SDT and ANSYS do).