[petsc-users] schur complement and condition number

[Jed Brown]

On Thu, Mar 31, 2011 at 07:59, Gong Ding <gdiso at ustc.edu> wrote:

> No, not a singular system. Just bad conditioned.
> I mean if I can eliminate the bad conditioned rows by building schur
> complement?
>

How are you enforcing the integral constraint in the boundary conditions?
There is probably a way to write your boundary condition so that it is well
scaled.

The Schur complement of your strip is an awkward thing to compute and will
significantly affect sparsity unless that part of the matrix is already
diagonal in which case there are good ways to scale correctly during
assembly or eliminate a-priori, or have PCREDISTRIBUTE do it automatically.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20110331/7042a8ed/attachment.htm>