Hi All,
I am currently using svn revision 22915, and I think I have spotted a bug in
one of the TrilinosWrappers::SparseMatrix::reinit() functions, specifically the
one that takes an Epetra_CrsMatrix as an input. I observe weird behavior in
which the diagonal entry for a given row in the Epetra_CrsMatrix would be
assigned to every non-zero column in the TrilinosWrappers::SparseMatrix, while
all the other off-diagonal entries in the Epetra_CrsMatrix would be ignored.
On line 684 of trilinos_sparse_matrix.cc, my current revision reads:
for (unsigned int i=0; i<my_nonzeros; ++i)
values[i] = in_values[0];
But I think it should instead be
for (unsigned int i=0; i<my_nonzeros; ++i)
values[i] = in_values[i];
That is, values[i] and in_values[i] should be walked through in tandem, rather
than just assigning the diagonal entry in_values[0] throughout.
Does this make sense, and have other people observed this behavior? I ran into
it originally when calling the TrilinosWrappers::SparseMatrix::mmult() function.
Best,
Josh
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
