[dune-fem] parallel solver issue

[Andreas Dedner]

Hi.
I can confirm this with the trunk version of dune-fem and dune-alugrid. 
Apparent;y the
slaves dofs are wrongly setup. On process 2 only one entry in the matrix 
is non zero (that is the one
on the middle of the process border) but all the other rows are zero. 
That means that the nodes on the
Dirichlet boundary are not correctly identified
t must be going wrong in this line
          double value = slaveDofs.isSlave( localDof )? 0.0 : 1.0;
where 0.0 is chosen instead of 1.0.
But I haven't found out why yet...
Andreas

On 11/04/15 12:56, Radcliffe, Alastair wrote:
> Hi fellow Dune-femers,
>
> I've come across an interesting issue with the non-convergence of the parallel solvers (istl or fem) for the (latest on the "stokes" branch) fem-howto example 04 when using Dirichlet boundary conditions on a dgf mesh i generated recently, see below.
>
> dune-fem version 1.4, and version 2.3 for the dune core modules
>
> Any feedback from someone who knows how the parallel solvers work would be gratefully received ...
>
> Thanks,
>
> Alastair
>
>
> ________________________________________
> From: Dedner, Andreas
> Sent: Friday, April 10, 2015 11:49 PM
> To: Radcliffe, Alastair
> Subject: Re: Fw: bug in alu or fem
>
> Hi Alastair.
> I can't figure out anything at the moment.
> You should perhaps report the bug to the dune-fem mailinglist with that
> dgf file.
> Andreas
>
>
> On 10/04/15 19:15, Radcliffe, Alastair wrote:
>> Hi,
>>
>> 'it' was a non-convergence of the solver, and the fem solvers didn't work either i think ... and of course, there are very few dune preconditioners that work in parallel to try ...
>>
>> the only way i've found to get it to 'solve' is to set the required residual to something e+3 or so ... then at least you can get paraview output to check it has "ranked" (ie: divided) the geometry up nicely ...
>>
>> i was hoping howto example 04 would do as a "minimal working example" for debugging ... but is it an alu or a dune-fem bug, so i can post it on the correct bug list ?
>>
>> Ta,
>>
>> Alastair
>>
>> ps: see you monday?, tuesday?, wednesday ? ...
>>
>> ________________________________________
>> From: Dedner, Andreas
>> Sent: Friday, April 10, 2015 4:18 PM
>> To: Radcliffe, Alastair
>> Subject: Re: Fw: bug in alu or fem
>>
>> Hi Alastair.
>> Yes I can conform it - whatever 'it' was in your case. You really have
>> to learn to write
>> meaningful and helpful error reports! In my case the program got stuck -
>> but is that the
>> same behaviour you saw? Also: what sort of tests did you do? For
>> example, did you see if
>> there is an endless loop in the solver (setting the verbosity flag to
>> 1)? I did do that and the
>> solver then printed "nan" for the residual. It starts off ok but then
>> goes bad. Did you test a
>> different solver, i.e., change the flag in Makefile.am to not use istl.
>> Anything else you tried/
>> Best
>> Andreas
>>
>>
>>
>> On 08/04/15 12:18, Radcliffe, Alastair wrote:
>>> mpiexec -np 2 main
>
> _______________________________________________
> dune-fem mailing list
> dune-fem at dune-project.org
> http://lists.dune-project.org/mailman/listinfo/dune-fem