On the use of MappingFEField

[Praveen C]

Dear all

I am trying to use MappingFEField on a parallel::distributed::Triangulation with TrilinosWrappers::MPI::Vector.

In this case, how should the euler vector be created ?

Should the euler vector use locally owned dofs or locally relevant dofs ?

I tried the following

      unsigned int degree = 4;

      const FE_Q<dim> fe(QGaussLobatto<1>(degree+1));

      const FESystem<dim> fesystem(fe, dim);

      DoFHandler<dim> dh_euler (triangulation);

      dh_euler.distribute_dofs (fesystem);

      IndexSet locally_owned_dofs;

      IndexSet locally_relevant_dofs;

      locally_owned_dofs = dh_euler.locally_owned_dofs ();

      DoFTools::extract_locally_relevant_dofs (dh_euler,

                                               locally_relevant_dofs);

      TrilinosWrappers::MPI::Vector euler_vector (locally_owned_dofs, MPI_COMM_WORLD);

      compute_mapping (degree, triangulation, dh_euler, euler_vector);

      MappingFEField<dim> map(dh_euler, euler_vector);

When “map" is destroyed, I get some error

An error occurred in line <128> of file </Users/praveen/Applications/deal.II/git/source/base/subscriptor.cc> in function

    void dealii::Subscriptor::check_no_subscribers() const

The violated condition was: 

    counter == 0

The name and call sequence of the exception was:

    ExcInUse (counter, object_info->name(), infostring)

Additional Information: 

Object of class N6dealii6VectorIdEE is still used by 1 other objects.

Thanks

praveen

On 30-May-2016, at 11:32 AM, Praveen C <cpra...@gmail.com> wrote:

Dear all

I have problems involving airfoil geometry where the cells have high aspect ratio in the boundary layer. Due to this, it is not enough to just use curved boundaries but the interior faces also need to be curved. See the attached figure gridq1.pdf

I have implemented a method based on winslow equations, see

http://persson.berkeley.edu/pub/fortunato15winslow.pdf

http://dl.acm.org/citation.cfm?id=2884302

Using this winslow method, I can generate an euler vector that gives valid cells, see gridqk.pdf which uses degree=4.

Using this euler vector, I want to create a MappingFEField and use that in my Navier-Stokes DG code.

I have some doubts.

When I use a MappingFEField, do I still need to attach manifold descriptions to the curved boundaries ? Assume that I am not doing any grid adaptation.

If I want to adapt the grid, is there a way to use the MappingFEField to place the new points ?

Are there any functions in deal.II to measure the quality of the mesh ?

The Winslow eqn approach seems quite useful and one can write a general class that may be useful for many people. At present I only have a serial implementation.

The approach is as follows:

Start with a Q1 mesh

Attach manifolds

Solve the winslow eqn to get the euler vector

Create a MappingFEField

Thanks

praveen

<gridq1.pdf><gridqk.pdf>

[Wolfgang Bangerth]

On 06/24/2016 11:57 PM, Praveen C wrote:
> MappingFEField<dim> map(dh_euler, euler_vector);
>
> When “map" is destroyed, I get some error
>
> An error occurred in line <128> of file
> </Users/praveen/Applications/deal.II/git/source/base/subscriptor.cc

> <https://urldefense.proofpoint.com/v2/url?u=http-3A__subscriptor.cc&d=CwMFaQ&c=ODFT-G5SujMiGrKuoJJjVg&r=bn1clvQAiDQrfAC4yKbN0PlSr7RWRs-U3WJ0zRBB2qM&m=Vf-2N0p6VmAXKK4t0RHa8AEgald8F3wLtzHtbbHTHDk&s=Q5qaqH8k_jfyOzP2rprL-TgG2uQe5NpA4zddILDSvuc&e=>>

> in function
> void dealii::Subscriptor::check_no_subscribers() const
> The violated condition was:
> counter == 0
> The name and call sequence of the exception was:
> ExcInUse (counter, object_info->name(), infostring)
> Additional Information:
> Object of class N6dealii6VectorIdEE is still used by 1 other objects.
>

What a coincidence -- I've explained this very error just 10 minutes to
someone sitting here next to me at the ASPECT hackathon :-)

The issue is that MappingFEField takes a dealii::Vector as second argument.
What you pass is a vector of a different class, so the constructor creates a
temporary vector to pass to the constructor, and then destroys it again after
the call. But the MappingFEField class has already taken its address, and so
you get the error above.

What you need to do is use the second template argument of the MappingFEField
class and set it to TrilinosWrappers::MPI::Vector. This should make it all work!

Best
Wolfgang

--
------------------------------------------------------------------------
Wolfgang Bangerth email: bang...@math.tamu.edu
www: http://www.math.tamu.edu/~bangerth/

[Praveen C]

Hello Wolfgang

I change to

MappingFEField<dim,dim,TrilinosWrappers::MPI::Vector> map(dh_euler, euler_vector);

But I get a linking error

Undefined symbols for architecture x86_64:
  "dealii::MappingFEField<2, 2, dealii::TrilinosWrappers::MPI::Vector, dealii::DoFHandler<2, 2> >::MappingFEField(dealii::DoFHandler<2, 2> const&, dealii::TrilinosWrappers::MPI::Vector const&, dealii::ComponentMask)", referenced from:
      _main in main.cc.o

Thanks

praveen

-- 
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en
--- You received this message because you are subscribed to the Google Groups "deal.II User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dealii+un...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Wolfgang Bangerth]

Praveen,

> I change to
>
> MappingFEField<dim,dim,TrilinosWrappers::MPI::Vector> map(dh_euler, euler_vector);
>
> But I get a linking error
>
> Undefined symbols for architecture x86_64:
> "dealii::MappingFEField<2, 2, dealii::TrilinosWrappers::MPI::Vector,
> dealii::DoFHandler<2, 2> >::MappingFEField(dealii::DoFHandler<2, 2> const&,
> dealii::TrilinosWrappers::MPI::Vector const&, dealii::ComponentMask)",
> referenced from:
> _main in main.cc.o

Indeed. You need this patch:
https://github.com/dealii/dealii/pull/2717
Best
W.