From: KIRK, BENJAMIN (JSCEG) (NASA) <benjamin.kirk1@na...>  20050214 15:34:07

I have looked at the PETSc matrices in some detail, and it appears there is no specialization for a simple symmetric matrix. They do have a symmetric block matrix, but if you are solving a scalar problem this isn't really applicable. You are solving a scalar Poisson system, right? So, we can't really limit the matrix size using PETSc. You may, however, be able to increase the performance of the iterative solver by using ICCpreconditioned conjugategradient as the linear solver. Additionally, you can only assemble 1/2 of the elementlevel matrix in your assembly routine and then copy it to the other 1/2 before inserting it into the global system matrix. This should cut the matrix assembly time in 1/2. Ben Original Message From: libmeshusersadmin@... [mailto:libmeshusersadmin@...] On Behalf Of Michael Povolotskyi Sent: Monday, February 07, 2005 4:45 AM To: libmeshusers@... Subject: [Libmeshusers] symmetric matrix Hello everybody, I'm implementing libmesh for a boundary value problem for the Poisson equation. It is possible to reduce a PDD problem to a linear system problem with a symmetric matrix. My question: is it possible to exploit somehow the symmetry of the matrix in order to reduce the RAM occupation? Thank you very much, Michael.   Michael Povolotskyi, Ph.D. University of Rome "Tor Vergata" Department of Electronic Engineering Viale Politecnico, 1  00133 Rome  Italy Phone + 39 06 72597367 Fax + 39 06 2020519 http://www.optolab.uniroma2.it/pages/moshe/moshe.html  