[petsc-users] Keep symbolic factorization

[Mathieu MORLIGHEM]

Hi,

I am using an augmented Lagrangian approach that requires to solve multiple times a system with the same Stiffness matrix but different right hand sides:

K U1 = F1
K U2 = F2
…
K UN = FN

I am using MUMPS right now, and I would like to do the analysis and factorization only once (to compute U1), and then only call the back substitution in order to speed up the convergence.
Is there any option in PETSc to do that? If yes, would it work with other direct solvers (cholmod, Pardiso)?

Thanks a lot for your help
--
Mathieu Morlighem

Assistant Professor
Department of Earth System Science
University of California, Irvine
3218 Croul Hall, Irvine, CA 92697-3100
Mathieu.Morlighem at uci.edu

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20140925/406fc063/attachment.html>