Thanks all, sorry, I will try to explain myself a little better.
I have several fully uncoupled systems of equations that correspond each one
to a finite element approximation for least squares formulation to solve a
differential equation.
This uncoupled systems are what I want to solve including all of them in the
same matrix to solve, that gives me a block sparsity pattern, but one on
which the only blocks that could have non-zero elements would be the blocks
on the full matrix diagonal (because they are uncoupled).
I am trying to define this particular sparsity pattern but when I do I get
off-diagonal blocks that are themselves diagonal matrices instead of "null"
or "empty" blocks without any non-zero elements, as I wished them to be. The
blocks in the diagonal are fine.
This is to say I want a sparsity pattern like this:
***
***
***
***
***
***
***
***
***
***
***
***
And I'm getting a sparsity pattern like this:
**** * *
*** * * *
*** * * *
* **** *
* *** * *
**** * *
* * ****
* * *** *
* **** *
* * * ***
* * * ***
* * ****
The code actually works, but it is working with non-zero element blocks
outside the diagonal, so it is being to expensive in memory when I solve
more than a hundred sub-systems.
I have tried moving the "compress" statement, defining a "null" sparsity
pattern as Timo suggested, by the off-diagonal matrices are still not empty.
Thanks again, the code I have until now is down here.
/*******************
/CODE
/*******************
GridGenerator::hyper_cube (triangulation, -5, 5);
triangulation.begin_active()->face(0)->set_boundary_indicator(0);
triangulation.begin_active()->face(1)->set_boundary_indicator(1);
triangulation.begin_active()->face(2)->set_boundary_indicator(2);
triangulation.begin_active()->face(3)->set_boundary_indicator(3);
triangulation.begin_active()->face(4)->set_boundary_indicator(4);
triangulation.begin_active()->face(5)->set_boundary_indicator(5);
triangulation.refine_global (2);
dof_handler.distribute_dofs (fe);
CompressedSparsityPattern c_sparsity(dof_handler.n_dofs());
// I have also tried here
// CompressedSparsityPattern null_sparsity(dof_handler.n_dofs());
DoFTools::make_sparsity_pattern (dof_handler, c_sparsity);
sparsity_pattern.reinit(nd,nd);
for (int i=0; i<nd; ++i) {
for (int j=0; j<nd; ++j) {
if (i==j) {
sparsity_pattern.block(i,j).copy_from(c_sparsity);
} else {
sparsity_pattern.block(i,j).reinit(dof_handler.n_dofs(),dof_handler.n_dofs(),0);
// I have also tried here
// sparsity_pattern.block(i,j).copy_from(null_sparsity);
// instead of the reinit without and making make_sparsity_pattern to this
null_sparsity block
}
}
}
sparsity_pattern.compress();
sparsity_pattern.collect_sizes ();
system_matrix.reinit (sparsity_pattern);
// .... and the rest of the code.
On Fri, Sep 24, 2010 at 9:54 AM, Guido Kanschat <[email protected]> wrote:
>
> On 9/23/2010 7:16 PM, fname lname wrote:
>>
>> Thanks Timo,
>>
>> I've tried that already, it gets me the same matrix as if I do:
>>
>>
sparsity_pattern.block(i,j).reinit(dof_handler.n_dofs(),dof_handler.n_dofs(),0);
>
> Can't you call sparsity_pattern.block(i,j).compress() right there?
>
> G
> _______________________________________________
> dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
