[dune-pdelab] P2 in parallel - small correction

[Volker Schauer]

Hallo Jö,

yes I set the true parameter, so my gridoperator is defined in the way

typedef
Dune::PDELab::GridOperator<GFS,GFS,PLOP,MBE,double,double,double,CC,CC,true> PGOS;

The matrix I extract from the GridOperator also really is "consistent"
i.e. in the matrix assembling the ghost entities were not assembled. (I
checked this some time ago at least when I used P1 elements by giving a
warning message in the local operator if a local matrix was assembled
from a ghost element).


It would be helpful if someone says that he has already used the
communication from PDELab to set up the Poisson or any other PDE
successfully with P2 elements (I suppose this was tested in PDELab when
the parallel use was implemented?). Would it help you or anybody else if
I extract the Poisson problem (which is embedded in a larger code doing
some more stuff) and send it to the mailing list ?

Thanks

Volker



Am Donnerstag, den 15.12.2011, 16:39 +0100 schrieb Jö Fahlke:
> Am Thu, 15. Dec 2011, 13:11:35 +0100 schrieb Volker Schauer:
> > Am Donnerstag, den 15.12.2011, 10:27 +0100 schrieb Volker Schauer:
> > > Maybe I should reformulate my question and give some additional
> > > information about the way I am setting up the problem.
> > > 
> > > I use the P2 elements in a nonoverlapping setting, given by the alugrid.
> > > In principal I use the solver 
> > > 
> > > Dune::PDELab::ISTLBackend_NOVLP_CG_SSORk<PGOS> 
> > > 
> > > to solve the problem, where I find that the solver converges, but the
> > > tested residuum is not zero.
> > > 
> > > As input to the solver, I give an "consistent vector" which has the same
> > > value on processor boundary nodes on all processes but arbitrary values
> > 
> > I don't have the ghost dofs with arbitrary values, but values are set to
> > zero by
> > 
> > Dune::PDELab::set_constrained_dofs(cc,0.0,rhs_poiss);
> > 
> > where the constraints container cc contains also the ghost nodes.  
> 
> I'm assuming your using PDELab's GridOperator(Space) to assemble the matrix
> (or compute the Jacobian on the fly).  Did you pass the template parameter
> nonoverlapping_mode=true?
> 
> Bye,
> Jö.
> 
> > 
> > 
> > Sorry for increasing your mail box needlessly
> > 
> > Volker
> > 
> > > on the ghost nodes (is that correct or should these values also be the
> > > same as their interior complements ?) and I also expect to get such an
> > > consistant vector in the solution.
> > > 
> > > So my question is, whether the upper assumptions are correct or if there
> > > might be a reason in pdelab why the P2 don't work, e.g. due to
> > > missing/incorrect communication.
> > > 
> > > I hope this explains my problem sufficiently
> > > 
> > > Thanks
> > > 
> > > Volker
> > > 
> > > Am Mittwoch, den 14.12.2011, 12:22 +0100 schrieb schauer:
> > > > Dear dune-pdelab,
> > > > 
> > > > does the parallel implementation of 3-dimensional P2 elements work fine
> > > > in pdelab ?
> > > > 
> > > > I am solving the 3D Poisson problem in parallel using alugrid. With P1
> > > > elements in parallel everything works fine and gives me the same
> > > > solution as on a single process. When switching to P2 (where I only
> > > > change the gridfunction space) the solver still tells me that the
> > > > residual is zero, but when I actually calculate it by hand it is quite
> > > > large. This is often the case if things on the boundary/ghost nodes goes
> > > > wrong.
> > > > 
> > > > Does anybody have experience with 3D P2 elements ?
> > > > 
> > > > Thanks for your answers
> > > > 
> > > > Volker
> > > > 
> > > > 
> > > > _______________________________________________
> > > > dune-pdelab mailing list
> > > > dune-pdelab at dune-project.org
> > > > http://lists.dune-project.org/mailman/listinfo/dune-pdelab
> > > 
> > 
> > -- 
> > 
> > _/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
> > 
> > Volker Schauer (Dipl.Phys)
> > 
> > Institute of Applied Mechanics (Civil Engineering), Chair 1
> > University of Stuttgart
> > Pfaffenwaldring 7
> > D-70569 Stuttgart
> > GERMANY
> > 
> > Phone ++49 (0)711-68560044
> > email: schauer at mechbau.uni-stuttgart.de
> > 
> > 
> > _______________________________________________
> > dune-pdelab mailing list
> > dune-pdelab at dune-project.org
> > http://lists.dune-project.org/mailman/listinfo/dune-pdelab
> > 
> 

-- 

_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/

Volker Schauer (Dipl.Phys)

Institute of Applied Mechanics (Civil Engineering), Chair 1
University of Stuttgart
Pfaffenwaldring 7
D-70569 Stuttgart
GERMANY

Phone ++49 (0)711-68560044
email: schauer at mechbau.uni-stuttgart.de