[petsc-dev] KSP monitor

[Barry Smith]

On Apr 26, 2011, at 6:50 AM, Nathan Collier wrote:

> Good day,
> 
> I am solving a linear problem using finite elements with PETSC and I
> want to look at the H1 norm of my true error as a function of KSP
> iterations. I registered my own monitor function with KSP which looks
> something like the last function in this example:
> 
> http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/src/ksp/ksp/examples/tutorials/ex9.c.html
> 
> where I get the current state of the solution vector using
> 
> KSPBuildSolution(kps,PETSC_NULL,&U)

   Try using KSPBuildSolution(ksp,work,0);   where you create the work vector in your monitor routine.

   There could be a bug in the usuage you are using.   Let us know if this works.


   Barry


> 
> Then I use U to compute the H1 norm. As per the documentation I do not
> Destroy U after. When I have this monitor function registered, my H1
> norm diverges while the KSP residual converges. Also, I compute the
> final H1 norm (outside of the monitor) and it is divergent, which
> matches the last iteration. If I remove my monitor function, things
> work well.
> 
> What am I screwing up? How can I access the current solution vector in
> a KSP monitor? Do I need to use a vector that I create from a DA since
> my FEM matrix assembly is based in DAs?
> 
> Thanks,
> NAte
> 
> -- 
> Nathan Collier
> Postdoctoral Researcher
> King Abdullah University of Science and Technology (KAUST)
> nathaniel.collier at gmail.com
> nathaniel.collier at kaust.edu.sa