[petsc-users] Jacobian check is sensitive to the choice of differencing parameter

[Matthew Knepley]

On Sun, Jul 19, 2015 at 9:45 AM, Khai Pham <khaipham at utexas.edu> wrote:

> Dear Petsc-Users,
>
> The SNES only converges for a small problem. I did the jacobian check. The
> result is sensitive to the choice of differencing parameter.
>

This looks like you probably have the correct Jacobian, but the problem is
very sensitive. Then convergence might be
hard as you scale up because you have a bad initial guess. Its impossible
to say anything else without knowing more
about your problem.

   Matt


> With the option "-snes_type test":
>
> "Norm of matrix ratio 4.11569e-09 difference 7.64426e-07 (user-defined
> state)
> Norm of matrix ratio 2.05846e-08 difference 3.89655e-06 (constant state
> -1.0)
> Norm of matrix ratio 5.77446e-09 difference 1.09307e-06 (constant state
> 1.0)"
>
> with the option "-snes_type test -mat_fd_type ds" :
>
> "Norm of matrix ratio 1.05213e-07 difference 1.95418e-05 (user-defined
> state)
> Norm of matrix ratio 2.51463e-07 difference 4.76004e-05 (constant state
> -1.0)
> Norm of matrix ratio 2.81081e-08 difference 5.32071e-06 (constant state
> 1.0)".
>
> Could you give me any hint where should I check for the error?
>
> Khai
>
>
>



-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20150719/af86cead/attachment-0001.html>