Hi Lawrence, The indefinite one works very well with ‘minres’. Thanks again. David
On 10 Dec 2015, at 16:19, Angwenyi David <kipkoej@gmail.com> wrote:
Thank you!
On 10 Dec 2015, at 16:11, Lawrence Mitchell <lawrence.mitchell@imperial.ac.uk> wrote:
On 10/12/15 15:06, Angwenyi David wrote:
Dear David,
I was trying to solve the indefinite one.
Ah, OK. A few things:
1. Your operator is indefinite, so a ksp_type of 'cg' won't work (conjugate gradient only works for SPD operators), try using 'minres' instead (which works for symmetric indefinite systems).
2. The indefinite Helmholtz problem is notoriously ill-conditioned, and moreover, very difficult to precondition well. If you're not trying to solve problems that are very large, try using a direct solver:
{'pc_type': 'lu', 'ksp_type': 'preonly'}
Cheers,
Lawrence
_______________________________________________ firedrake mailing list firedrake@imperial.ac.uk https://mailman.ic.ac.uk/mailman/listinfo/firedrake