[petsc-users] Norm of RHS in BCGS

[Norihiro Watanabe]

Unfortunately -ksp_converged_reason prints the number of iterations
but no information about final errors.

Thanks,
Nori

On Tue, Mar 22, 2016 at 3:06 PM, Matthew Knepley <knepley at gmail.com> wrote:
> On Tue, Mar 22, 2016 at 9:02 AM, Norihiro Watanabe <norihiro.w at gmail.com>
> wrote:
>>
>> What I wanted to do is displaying final converged errors without using
>> -ksp_monitor. Because my problem includes a lot of time steps and
>> nonlinear iterations, log output from -ksp_monitor for each linear
>> solve is sometimes too much. But you are right. It doesn't make sense
>> to call the expensive function just for the log output.
>
>
> Maybe something like -ksp_converged_reason?
>
>   Thanks,
>
>      Matt
>
>>
>> Thanks,
>> Nori
>>
>> On Tue, Mar 22, 2016 at 2:45 PM, Matthew Knepley <knepley at gmail.com>
>> wrote:
>> > On Tue, Mar 22, 2016 at 8:25 AM, Norihiro Watanabe
>> > <norihiro.w at gmail.com>
>> > wrote:
>> >>
>> >> Thank you Matt!
>> >>
>> >> Actually I don't want to change a norm type used in a convergence
>> >> check. I just want to output a relative error which PETSc actually
>> >> used for a convergence check (for log output in my program without
>> >> -ksp_*) and thought I need to have a norm of a preconditioned RHS to
>> >> compute it by myself. Or is there any function available in PETSc
>> >> which returns the relative error or the tolerance multiplied by the
>> >> norm of a preconditioned RHS? I couldn't find it.
>> >
>> >
>> > If you want the action of the preconditioner, you can pull it out
>> >
>> >   KSPGetPC()
>> >
>> > and apply it
>> >
>> >   PCApply()
>> >
>> > but I still do not understand why you want this. Do you want to check
>> > the
>> > norms
>> > yourself? The PCApply() could be expensive to calculate again.
>> >
>> >   Thanks,
>> >
>> >      Matt
>> >
>> >>
>> >>
>> >> Best,
>> >> Nori
>> >>
>> >>
>> >> On Tue, Mar 22, 2016 at 1:51 PM, Matthew Knepley <knepley at gmail.com>
>> >> wrote:
>> >> > On Tue, Mar 22, 2016 at 2:50 AM, Norihiro Watanabe
>> >> > <norihiro.w at gmail.com>
>> >> > wrote:
>> >> >>
>> >> >> Hi,
>> >> >>
>> >> >> Is it correct that a norm of a preconditioned RHS vector is used to
>> >> >> compute a relative error in BCGS?
>> >> >
>> >> >
>> >> > Yes, but you can verify this using -ksp_view
>> >> >
>> >> >>
>> >> >> I'm testing BCGS + BoomerAMG. With "-info", PETSc says "initial
>> >> >> right
>> >> >> hand side norm" is 2.223619476717e+10 (see below) but an actual norm
>> >> >> of the RHS I passed is 4.059007e-02. If yes, is there any way to get
>> >> >> a
>> >> >> norm of a preconditioned RHS?
>> >> >
>> >> >
>> >> > Do you mean unpreconditioned? You can try
>> >> >
>> >> >
>> >> >
>> >> >
>> >> > http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/KSP/KSPSetNormType.html
>> >> >
>> >> > or use
>> >> >
>> >> >   -ksp_monitor_true_residual
>> >> >
>> >> >   Thanks,
>> >> >
>> >> >     Matt
>> >> >
>> >> >>
>> >> >> [0] KSPConvergedDefault(): Linear solver has converged. Residual
>> >> >> norm
>> >> >> 2.036064453512e-02 is less than relative tolerance
>> >> >> 9.999999960042e-13
>> >> >> times initial right hand side norm 2.223619476717e+10 at iteration 6
>> >> >>
>> >> >>
>> >> >> Regards,
>> >> >> Nori
>> >> >>
>> >> >> --
>> >> >> Norihiro Watanabe
>> >> >
>> >> >
>> >> >
>> >> >
>> >> > --
>> >> > 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
>> >>
>> >>
>> >>
>> >> --
>> >> Norihiro Watanabe
>> >
>> >
>> >
>> >
>> > --
>> > 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
>>
>>
>>
>> --
>> Norihiro Watanabe
>
>
>
>
> --
> 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



-- 
Norihiro Watanabe