ffc team mailing list archive

[Kristian Oelgaard <k.b.oelgaard@xxxxxxxxxx>]

Quoting Patrick Riesen <priesen@xxxxxxxxxxxxxxx>:

> i suppose the problem lies not in VTKFile.cpp.
>
> you consider a tensor function to have a rank >= 2,
> but for my tensor function the rank returned by e.g.
>
> e.function_space().element().value_rank()
>
> is 1. So it is rather a vector with 4 components having rank 1.
> possibly value_rank() in the FiniteElement class should return 2 even
> for a 2d tensor function, then your code in VTKFile is fine, but i
> didn't look into the FiniteElement classes.

The problem is in ffc/fem/createelement.py where a tensor element is treated as
a MixedElement which has rank 1. A simple workaround might be to attach
len(ufl_element._shape) as the rank of a mixed element and use this value when
we generate code for the value_rank() function.
What do you think Anders?

Kristian

> patrick
>
>
> -------- Original Message --------
> Subject: Re: [FEniCS-users] [Fwd: Re:  tensor function]
> Date: Wed, 30 Sep 2009 11:30:45 +0200
> From: Patrick Riesen <priesen@xxxxxxxxxxxxxxx>
> CC: fenics <fenics-users@xxxxxxxxxx>
> References: <4AC31DB1.2040001@xxxxxxxxxxxxxxx>
> <4AC320F4.1000600@xxxxxxxxxxxxxxx> <4AC32181.8050202@xxxxxxxxx>
>
> seems that the tensor function for a 2d problem doesn't get to the
> output structure for a tensor function but to that for a vector function
> because it only has rank = 1. the write out for tensor is only invoked
> for a (3x3) tensor with rank = 2.
>
> Garth N. Wells wrote:
> >
> >
> > Patrick Riesen wrote:
> >> Hello,
> >>
> >> I don't know if this is a problem (dolfin0.9.2):
> >> When i was writing the TensorFunction e to a vtk-file after a solve
> >> by just using
> >>
> >> file << e
> >>
> >> the results i got were absolutely nonsense. I got three components,
> >> but none of them made any sense. This is why i was thinking i'm doing
> >> something completely wrong in my equations/code,
> >>
> >
> > VTK output of tensor functions is untested. I would be very surprised if
> > it did work. Dig around in VTKFile.cpp to see if you can find the problem.
> >
> > Garth
>
>
>
>
>
> >
> >> But if i first separate all components and write them to separate
> >> files as
> >>      Function e11 = e[0];
> >>      Function e12 = e[1];
> >>      Function e21 = e[2];
> >>      Function e22 = e[3];
> >>
> >>      fileE11 << e11;
> >>      fileE12 << e12;
> >>      fileE21 << e21;
> >>      fileE22 << e22;
> >>
> >> the solutions look reasonable (uff!..) and e12 and e21 are the same.
> >>
> >> regards,
> >> patrick
> >>
> >>
> >> Patrick Riesen wrote:
> >>> sorry, something went wrong, this would be the message:
> >>>
> >>> -------- Original Message --------
> >>> Subject: Re: [FEniCS-users] tensor function
> >>> Date: Wed, 30 Sep 2009 10:52:02 +0200
> >>> From: Patrick Riesen <priesen@xxxxxxxxxxxxxxx>
> >>> To: Anders Logg <logg@xxxxxxxxx>
> >>> References: <4AC0A8D5.6040207@xxxxxxxxxxxxxxx>
> >>> <20090928122033.GH2528@olorin>    <4AC0AD56.50608@xxxxxxxxxxxxxxx>
> >>> <20090928140134.GA24982@olorin> <4AC0C346.7020309@xxxxxxxxxxxxxxx>
> >>>
> >>> or maybe not..
> >>>
> >>> i can not do
> >>>
> >>> Function e11 = e[0][0];
> >>>
> >>> i get
> >>>
> >>> error: no match for 'operator[]' in
> >>> 'dolfin::Function::operator[](dolfin::uint)(0u)[0]'
> >>>
> >>>
> >>> but i can get components doing
> >>>
> >>> e11 = e[0];
> >>> e12 = e[1];
> >>> e21 = e[2];
> >>> e22 = e[3];
> >>>
> >>>
> >>> i suppose for dolfin 0.9.3 it is the case as described below and for
> >>> dolfin 0.9.2 the case as above therefore ?
> >>> (which is what i am still using currently)
> >>>
> >>> greetings,
> >>> patrick
> >>>
> >>> Patrick Riesen wrote:
> >>>> this helped to sort things out, thanks again.
> >>>> patrick
> >>>>
> >>>> Anders Logg wrote:
> >>>>> On Mon, Sep 28, 2009 at 02:34:30PM +0200, Patrick Riesen wrote:
> >>>>>> Hi,
> >>>>>>
> >>>>>> thans for the quick response,
> >>>>>>
> >>>>>> therefore when V is the function space of the (e, v, p) element and
> >>>>>>
> >>>>>>> e[0] = (a, b)
> >>>>>>> e[1] = (c, d)
> >>>>>>> e[0][0] = a
> >>>>>>> e[0][1] = b
> >>>>>>> e[1][0] = c
> >>>>>>> e[1][1] = d
> >>>>>> therefore doing
> >>>>>>
> >>>>>> fs = new SubSpace(V,0,0)
> >>>>>>
> >>>>>> corresponds to the subspace for the two components (a, b) of the
> >>>>>> element  and not just for a, which then would be SubSpace(V,0,0,0) ?
> >>>>> Yes.
> >>>>>
> >>>>> --
> >>>>> Anders
> >>>> _______________________________________________
> >>>> FEniCS-users mailing list
> >>>> FEniCS-users@xxxxxxxxxx
> >>>> http://fenics.org/mailman/listinfo/fenics-users
> >>>
> >>> _______________________________________________
> >>> FEniCS-users mailing list
> >>> FEniCS-users@xxxxxxxxxx
> >>> http://fenics.org/mailman/listinfo/fenics-users
> >>
> >> _______________________________________________
> >> FEniCS-users mailing list
> >> FEniCS-users@xxxxxxxxxx
> >> http://fenics.org/mailman/listinfo/fenics-users
> >
> >
>
> _______________________________________________
> FEniCS-users mailing list
> FEniCS-users@xxxxxxxxxx
> http://fenics.org/mailman/listinfo/fenics-users
> _______________________________________________
> DOLFIN-dev mailing list
> DOLFIN-dev@xxxxxxxxxx
> http://www.fenics.org/mailman/listinfo/dolfin-dev
>