[petsc-users] how to speed up convergence?

[Jed Brown]

On Thu, Nov 10, 2011 at 11:07, Konstantinos Kontzialis <
ckontzialis at lycos.com> wrote:

> 1. I use the conservative variables.
>
> 2. No limiter.
>
> 3. Boundary conditions are weakly imposed (to solve the riemann problem).
>
> 4. mat_fd_type ds is worse for the current case
>

It sounds like your system is poorly scaled. If you were consistent about
using PetscScalar in your code, then you can configure
--with-precision=__float128 and try that. It should remove the differencing
errors. It is very important to scale the system carefully when using
coloring or matrix-free finite differencing. It's a waste of time to dwell
on the other issues until you know that you have an accurate Jacobian.


> 5. sub_pc_type lu did a better work but not to a satisfactory level
>

In terms of time or iteration count? This test is for iteration count.


>
> 6. MATCOLORINGSL is used.
>
> For the preconditioner I need to do some changes.
>
> Please help me first to understand the coloring types in order to choose
> the best one (any web pages, links etc.?).
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20111110/c81d47a8/attachment.htm>