The key issue for my application is whether the make_trilinos_map method of 
the IndexSet class is called with overlapping equal to true or false. At 
the moment, when make_trilinos_map is called from the reinit method for 
Vector in the TrilinosWrappers∷MPI namespace, overlapping is always set to 
true.


I believe that, at lease with my simple grid and running on only one 
processor, that overlapping can be set to false. My question is, does the 
following code change make any sense? In my case, has_ghosts is evaluating 
to false and so when I make this code change, I have linear epetra vectors. 
I'll need to think about how, or if, I could make this work with Ifpack and 
Hypre when there are Ghost elements.


*Current code*
void
    Vector::reinit (const IndexSet &parallel_partitioner,
                    const MPI_Comm &communicator,
                    const bool      /*omit_zeroing_entries*/)
    {
      nonlocal_vector.reset();

      Epetra_Map map = parallel_partitioner.make_trilinos_map (communicator,
                                                               true);

      vector.reset (new Epetra_FEVector(map));

      has_ghosts = vector->Map().UniqueGIDs()==false;de here...


*My change*
void
    Vector::reinit (const IndexSet &parallel_partitioner,
                    const MPI_Comm &communicator,
                    const bool      /*omit_zeroing_entries*/)
    {

      nonlocal_vector.reset();

      has_ghosts = vector->Map().UniqueGIDs()==false;

      Epetra_Map map = parallel_partitioner.make_trilinos_map (communicator,
                    has_ghosts);

      vector.reset (new Epetra_FEVector(map));de here...


-- 
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+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to