tolerance

[Matthew Knepley]

On Fri, Apr 17, 2009 at 12:58 AM, Yongcheng Zhou
<yongcheng.zhou at gmail.com>wrote:

> Hi there,
>
> I am wondering what is the real tolerance for stopping the Krylov
> iteations. In solving my nonlinear problem I have to call
> KSP solvers multiple time.  my tolerance setting is
>   KSPSetTolerances(ksp,1.e-5,1.e-5,PETSC_DEFAULT,200);
>
>  After several calls I found that the linear solver does not do
> anything, as shown by the convergence history:
>
>  35 KSP preconditioned resid norm 8.570481231422e-05 true resid norm
> 7.256019920381e+01 ||Ae||/||Ax|| 1.008233844002e-01
>  36 KSP preconditioned resid norm 5.210695852340e-06 true resid norm
> 7.448290646301e+00 ||Ae||/||Ax|| 1.034950123066e-02
>  0 KSP preconditioned resid norm 5.210622961414e-06 true resid norm
> 7.448436220158e+00 ||Ae||/||Ax|| 1.000000000000e+00
>  0 KSP preconditioned resid norm 5.210622961414e-06 true resid norm
> 7.448436220158e+00 ||Ae||/||Ax|| 1.000000000000e+00
>  0 KSP preconditioned resid norm 5.210622961414e-06 true resid norm
> 7.448436220158e+00 ||Ae||/||Ax|| 1.000000000000e+00
>
> where in the last three calls the KSP simply refused to do any
> interation. Is there any way I can force the KSP to check the
> convergence according to true resid or Ae/Ax  rather than the
> preconditoned resid?


I am not sure waht you are asking. You can write any tolerance check you
want
using KSPSetConvergenceTest().

   Matt


>
> Thanks!
>
> Rocky
>
-- 
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/20090417/c2d69513/attachment.htm>