[Libmesh-users] Pble with MONOMIAL const element

[Buffat M. <bu...@uf...>]

Dear Ben,
thank you very much for your fast help!

I am evaluating libmesh for a CFD research project using a Finite Volume 
formulation.
I have found some problems with the library which I have solved, and which may 
be of interest for  other users:
- writing element data in the solution file (in GMV format): it is not to 
difficult, even in //e

-a bug in the distributed vector  init function
here is the patch for the file distributed_vector.h
440a441
>   // BUG MPI ?
441a443
>   std::vector<int> local_sizes_send(n_proc, 0);
443c445
<   local_sizes[proc_id] = n_local;
---
>   local_sizes_send[proc_id] = n_local;
445c447
<   MPI_Allreduce (&local_sizes[0], &local_sizes[0], local_sizes.size(),
---
>   MPI_Allreduce (&local_sizes_send[0], &local_sizes[0], local_sizes.size(),
462c464,465
<   assert (sum == static_cast<int>(size()));
---
>   // BUG Set the initialized flag not set
>   //assert (sum == static_cast<int>(size()));

the  MPI_Allreduce needs a different buffer for the send and the receive and 
we cannot call the size() function here because the vector is not initialized 
!
- I also add a public access to the PetSC vector in the PetscVector class 
 i.e.
// acces au vecteur Petsc
  Vec & PetscVec() { return vec; }
and in the class PetscMatrix i.e.
// acces a la matricePetsc
  Mat & PetscMat() { return mat; }
in order to be able to use all the PETSC lib, specially the print function.
Indeed, in //e the print function of the numeric_vector class works only for 
the processor 0 and gives erroneous values for the other proc.

Thank you very much again

	Marc