dolfin team mailing list archive

[Niclas Jansson <njansson@xxxxxx>]

Niclas Jansson <njansson@xxxxxxxxxx> writes:

> logg <logg@xxxxxxxxx> writes:
>
>> ** Changed in: dolfin
>>        Status: New => Confirmed
>>
>> ** Changed in: dolfin
>>      Assignee: (unassigned) => Niclas Jansson (njansson)
>>
>> -- 
>> Neumann boundary conditions not working in parallel
>> https://bugs.launchpad.net/bugs/426473
>> You received this bug notification because you are subscribed to DOLFIN.
>>
>> Status in DOLFIN: Confirmed
>>
>> Bug description:
>> Neumman boundary conditions (exterior facet integrals) don't seem to be working in parallel. To reproduce this error, go to test/system/parallel-assembly-solve/ and run
>>
>>   mpirun -n 2 python solver.py
>>
>> I get the following output:
>>
>> (unitsquare.xml.gz, 1):	*** ERROR (norm = 25.19288137326567, reference = 9.547454087327344, diff = 15.64542728593833)
>> (unitsquare.xml.gz, 2):	OK (norm = 18.42366670418566, reference = 18.42366670418527, diff = 3.872457909892546e-13)
>> (unitsquare.xml.gz, 3):	OK (norm = 27.29583104741751, reference = 27.29583104741712, diff = 3.907985046680551e-13)
>> (unitsquare.xml.gz, 4):	OK (norm = 36.16867128090711, reference = 36.1686712809094, diff = 2.295053036505124e-12)
>> (unitcube.xml.gz, 1):	*** ERROR (norm = 14.62062082718494, reference = 8.876490653853809, diff = 5.744130173331129)
>> (unitcube.xml.gz, 2):	*** ERROR (norm = 32.93811914105483, reference = 19.99081167299566, diff = 12.94730746805917)
>> (unitcube.xml.gz, 3):	OK (norm = 33.85477561286864, reference = 33.85477561286852, diff = 1.20792265079217e-13)
>> (unitcube.xml.gz, 4):	OK (norm = 49.97357666762994, reference = 49.97357666762962, diff = 3.197442310920451e-13)
>>
>> Strangely enough, it works for q = 2, 3, 4 in 2D but not q = 1, and works for q = 3, 4 in 3D but not q = 1, 2.
>
> Should work now. 
>
> BoundaryComputation needs facet connectivity and overlap, which are not
> computed for P1. Otherwise it applies boundary conditions to the
> "interior" boundary between processes.
>
> Niclas