dolfin team mailing list archive

[Kristian Ølgaard <k.b.oelgaard@xxxxxxxxx>]

On 24 November 2011 20:33, Anders Logg <logg@xxxxxxxxx> wrote:
> On Thu, Nov 24, 2011 at 01:43:20PM +0100, Kristian Ølgaard wrote:
>> On 23 November 2011 10:54, Martin Sandve Alnæs <martinal@xxxxxxxxx> wrote:
>> > On 23 November 2011 09:37, Kristian Ølgaard <k.b.oelgaard@xxxxxxxxx> wrote:
>> >> On 22 November 2011 21:15, Kristian Ølgaard <k.b.oelgaard@xxxxxxxxx> wrote:
>> >>> On 22 November 2011 21:07, Anders Logg <logg@xxxxxxxxx> wrote:
>> >>>> On Tue, Nov 22, 2011 at 07:34:38PM +0000, Garth N. Wells wrote:
>> >>>>> On 22 November 2011 19:12, Kristian Ølgaard <k.b.oelgaard@xxxxxxxxx> wrote:
>> >>>>> > Hi,
>> >>>>> >
>> >>>>> > The IntervalCell::facet_area in DOLFIN currently returns 0.0, while
>> >>>>> > the facet determinant (a scale factor) 'det = 1.0;'
>> >>>>> > in the generated code for the tabulate_tensor function in 1D (facet integrals).
>> >>>>> >
>> >>>>> > Is 0.0 the correct return value, or should it return 1.0?
>> >>>>> >
>> >>>>>
>> >>>>> I would think 1.0.
>> >>>
>> >>> That's what I'm leaning towards too.
>> >>>
>> >>>> Facet is the thing of dimension one lower than the cell dimension so
>> >>>> in this case 1 - 1 so it would mean the area of an endpoint. So I
>> >>>> believe 0 is correct.
>> >>>
>> >>> I see your 'point' :), but is it consistent with how we evaluate
>> >>> integrals on endpoints?
>> >>
>> >> Running this script:
>> >>
>> >> from dolfin import *
>> >>
>> >> cells = [interval, triangle, tetrahedron]
>> >> meshes = [UnitInterval(1), UnitSquare(1,1), UnitCube(1,1,1)]
>> >>
>> >> for cell, mesh in zip(cells, meshes):
>> >
>> > btw, you can do
>> >    cell = mesh.ufl_cell()
>> >
>> >>    c = Constant(1, cell)
>> >>    print
>> >>    print cell
>> >>    print "volume:  ", assemble(c*dx, mesh=mesh)
>> >>    print "surface: ", assemble(c*ds, mesh=mesh)
>> >>
>> >> produces:
>> >>
>> >> <interval cell in R1>
>> >> volume:   1.0
>> >> surface:  2.0
>> >>
>> >> <triangle cell in R2>
>> >> volume:   1.0
>> >> surface:  4.0
>> >>
>> >> <tetrahedron cell in R3>
>> >> volume:   1.0
>> >> surface:  6.0
>> >>
>> >> If we want surface = 0.0 for the interval, 'det' must be set to zero
>> >> in the generated code.
>> >> However, doing so will mean that f*ds etc. becomes zero too, which is
>> >> not very useful.
>> >>
>> >> Kristian
>> >
>> > As Kristian shows here, I think defining it to 1.0 seems more useful,
>> > and consistent with the definition of 1.0*ds.
>>
>> I've pushed the change to DOLFIN, the output of the extended script:
>>
>> from dolfin import *
>>
>> for mesh in [UnitInterval(1), UnitSquare(1,1), UnitCube(1,1,1)]:
>>     c = Constant(1, mesh.ufl_cell())
>>     c0 = mesh.ufl_cell().facet_area
>>     c1 = FacetArea(mesh)
>>     print
>>     print mesh.ufl_cell()
>>     print "volume:  ", assemble(c*dx, mesh=mesh)
>>     print "surface: ", assemble(c*ds, mesh=mesh)
>>     print "facet_area ufl/ffc: ", assemble(c0*ds, mesh=mesh)
>>     print "FacetArea  dolfin:  ", assemble(c1*ds, mesh=mesh)
>>
>> is
>>
>> <interval cell in R1>
>> volume:   1.0
>> surface:  2.0
>> facet_area ufl/ffc:  2.0
>> FacetArea  dolfin:   2.0
>>
>> <triangle cell in R2>
>> volume:   1.0
>> surface:  4.0
>> facet_area ufl/ffc:  4.0
>> FacetArea  dolfin:   4.0
>>
>> <tetrahedron cell in R3>
>> volume:   1.0
>> surface:  6.0
>> facet_area ufl/ffc:  3.0
>> FacetArea  dolfin:   3.0
>>
>> Kristian
>
> ok.
>
> Can you make it into a unit test? (Maybe you already did, I haven't checked.)

I put the above script in
test/unit/function/python/SpecialFunctions.py (not pushed to repo)

but doing

mpirun -np 3  python ./SpecialFunctions.py

results in the error:

Process 0: Number of global vertices: 2
Process 0: Number of global cells: 1
EE
======================================================================
ERROR: testFacetArea (__main__.SpecialFunctions)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "./SpecialFunctions.py", line 29, in testFacetArea
    references = [(UnitInterval(1), 2, 2),]#\
  File "/home/oelgaard/software/fenics/lib/python2.7/site-packages/dolfin/cpp.py",
line 12959, in __init__
    _cpp.UnitInterval_swiginit(self,_cpp.new_UnitInterval(*args))
RuntimeError:

*** -------------------------------------------------------------------------
*** DOLFIN encountered an error. If you are not able to resolve this issue
*** using the information listed below, you can ask for help at
***
***     https://answers.launchpad.net/dolfin
***
*** Remember to include the error message listed below and, if possible,
*** include a *minimal* running example to reproduce the error.
***
*** -------------------------------------------------------------------------
*** Error:  Unable to compute mesh partitioning using ParMETIS.
*** Reason: ParMETIS cannot be used if a process has no cells (graph
nodes). Use SCOTCH to perform partitioning instead.
*** Where:  This error was encountered inside ParMETIS.cpp.
*** -------------------------------------------------------------------------


----------------------------------------------------------------------
Ran 1 test in 0.003s

FAILED (errors=1)

======================================================================
ERROR: testFacetArea (__main__.SpecialFunctions)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "./SpecialFunctions.py", line 29, in testFacetArea
    references = [(UnitInterval(1), 2, 2),]#\
  File "/home/oelgaard/software/fenics/lib/python2.7/site-packages/dolfin/cpp.py",
line 12959, in __init__
    _cpp.UnitInterval_swiginit(self,_cpp.new_UnitInterval(*args))
RuntimeError:

*** -------------------------------------------------------------------------
*** DOLFIN encountered an error. If you are not able to resolve this issue
*** using the information listed below, you can ask for help at
***
***     https://answers.launchpad.net/dolfin
***
*** Remember to include the error message listed below and, if possible,
*** include a *minimal* running example to reproduce the error.
***
*** -------------------------------------------------------------------------
*** Error:  Unable to compute mesh partitioning using ParMETIS.
*** Reason: ParMETIS cannot be used if a process has no cells (graph
nodes). Use SCOTCH to perform partitioning instead.
*** Where:  This error was encountered inside ParMETIS.cpp.
*** -------------------------------------------------------------------------


----------------------------------------------------------------------
Ran 1 test in 0.003s

FAILED (errors=1)


after which the process hangs.

Kristian

> --
> Anders
>