[Dune] Fwd: Re: Navier-Stokes parallel with UG and YaspGrid in 2D

[manuel.hofmann at iwr.uni-heidelberg.de]

Hi,

it does not work if i use a linear solver with preconditioner( i get  
the errors as described in mz first email ). If I use e.g. bicgstab  
without a preconditioner it works. Unfortunately it does not converge.  
Yes i use the latest svn trunk of dune-grid.

Regards,
Manuel

> Hi,
>
>> you're right, 2d ALUGrid is not parallel. It looks like I really have to
>> find the time to publish SPGrid.
>>
>
> That would be much appreciated anyways.  We could finally retire  
> SGrid and YaspGrid.
>
>
>> Implementing edge communication into YaspGrid basically shouldn't be too
>> hard (it was trivial with SPGrid). Unfortunately, the codim 1 entities
>> would have to be implemented first. I really cannot say anything about UG.
>>
>
> It should really work in UGGrid (caveat: unless you refine your grid).
> I take it you are using the svn trunk of dune-grid?  Maybe I forgot
> the all / all interface.
>
> Best,
> Oliver
>
>> If you're not out for polynomial order, you could try using the mini
>> element.
>>
>> On 03/20/2012 02:25 PM, manuel.hofmann at iwr.uni-heidelberg.de wrote:
>>> Hi Martin,
>>>
>>> yes thats right I use Taylor-Hood Elements Q2/Q2 or P2/P1. As far as i
>>> know ALUGrid does not support parallel mode in 2D. Is it hard to
>>> program edge communication? How do I do that? Im not familiar with
>>> parallel computing.
>>>
>>> Regards,
>>>
>>> Manuel
>>>
>>>
>>> Zitat von Martin Nolte <nolte at mathematik.uni-freiburg.de>:
>>>
>>>> Hi Manuel,
>>>>
>>>> I guess you're using a Taylor-Hood element with 2nd order Lagrange
>>>> polynomials for the velocity and first order ones for the pressure. The
>>>> 2nd order polynomials have a degree of freedom on each edge of the
>>>> grid. To use such a code in parallel, you need to communicate data on
>>>> the edges residing on the process borders.
>>>>
>>>> Unfortunately, as far as I know neither UG nor YaspGrid support
>>>> communication of edge data. Have you tried ALUGrid?
>>>>
>>>> Best,
>>>>
>>>> Martin
>>>>
>>>> On 03/20/2012 10:10 AM, manuel.hofmann at iwr.uni-heidelberg.de wrote:
>>>>> Hi,
>>>>>
>>>>> i tried to run the navier-stokes (cgstokes_instat) example in
>>>>> pdelab-howto in
>>>>> parallel mode. However i get with YaspGrid and UGGrid some interface
>>>>> communication errors. I used UG-3.9.1-patch7.
>>>>>
>>>>> [findDDDInterfaces_:/home/mhofmann/diplomarbeit/dev/dune/dune-grid/dune/grid/uggrid.hh:968]:
>>>>>
>>>>>
>>>>> Edge communication not supported for interfaces of type all / all
>>>>> interface
>>>>>
>>>>> [comm:/home/mhofmann/diplomarbeit/dev/dune/dune-grid/dune/grid/yaspgrid.hh:2439]:
>>>>>
>>>>>
>>>>> interface communication not implemented
>>>>>
>>>>> I tried ISTLBackend_OVLP_BCGS_ILU0 and ISTLBackend_NOVLP_CG_NOPREC
>>>>> as linear
>>>>> solver and newton solver of pdelab as nonlinear sovler. Both with
>>>>> the same
>>>>> error message. I am using the latest dune-grid module
>>>>>
>>>>> This is how i load the grid with UG8(if thats important?):
>>>>> Dune::MPIHelper& helper = Dune::MPIHelper::instance(argc, argv);
>>>>> [...]
>>>>> typedef Dune::UGGrid<2> GridType;
>>>>> GridType grid;
>>>>> std::string grid_file = "grids/cylinder.msh";
>>>>> Dune::GridFactory<GridType> factory(&grid);
>>>>> if(helper.rank()==0) {
>>>>> Dune::GmshReader<GridType>::read(factory,grid_file,true,false);
>>>>> }
>>>>> factory.createGrid();
>>>>> [...]
>>>>> grid.globalRefine(parameters.domain_level);
>>>>> if(!Dune::MPIHelper::isFake) grid.loadBalance();
>>>>>
>>>>> ---
>>>>>
>>>>> Dune::FieldVector<double,2> L(1.0);
>>>>> Dune::FieldVector<int,2> N(1);
>>>>> Dune::FieldVector<bool,2> B(false);
>>>>> int overlap=1;
>>>>> Dune::YaspGrid<2> grid(helper.getCommunicator(),L,N,B,overlap);
>>>>> typedef Dune::YaspGrid<2>::LeafGridView GV;
>>>>>
>>>>>
>>>>> Am i doing something wrong or is this still not implemented?
>>>>>
>>>>> Regards
>>>>> Manuel
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Dune mailing list
>>>>> Dune at dune-project.org
>>>>> http://lists.dune-project.org/mailman/listinfo/dune
>>>>
>>>> -- 
>>>> Dr. Martin Nolte <nolte at mathematik.uni-freiburg.de>
>>>>
>>>> Universität Freiburg phone: +49-761-203-5630
>>>> Abteilung für angewandte Mathematik fax: +49-761-203-5632
>>>> Hermann-Herder-Straße 10
>>>> 79104 Freiburg, Germany
>>>
>>>
>>>
>>>
>>> ----- Ende der weitergeleiteten Nachricht -----
>>>
>>>
>>>
>>> _______________________________________________
>>> Dune mailing list
>>> Dune at dune-project.org
>>> http://lists.dune-project.org/mailman/listinfo/dune
>>
>
> _______________________________________________
> Dune mailing list
> Dune at dune-project.org
> http://lists.dune-project.org/mailman/listinfo/dune