Hello Huan,
I see. I did similar tests some time ago. The only difference I did was
using constraints.is_constrained instead of
constraints.is_inhomogeneously_constraint. Can you try, if this works?
Anyway for me this method did not pay out. It took much longer than
assemling all in one function and saving the matrix, because most of the
time is - of course - spent on assembling the matrix. I think this
implementation is only favourable, if one has homogeneous Dirichlet
boundary conditions.
VectorTools::project () should also work for inhomogeneous boundary
conditions.
Best Regards,
Markus
Am 18.05.10 08:51, schrieb Michael Rapson:
Hi Huan,
The distribute_local_to_global modifies both the right hand side and
the matrix itself in how it preforms the algorithm. Hence it is not
enough just to give it a dummy matrix (in which case the modifications
will be incorrect) or even to give it a copy of the actual matrix, but
not use that matrix later.
If you look at step-22 in the deal-6.1.0 tutorial, you will notice
that the boundary conditions are implemented in a completely different
way, using VectorTools::interpolate_boundary_values and
MatrixTools::apply_boundary_values. This method is still available I
believe however the constraint matrix class seems to be preferred in
general.
Regards,
Michael
On Tue, May 18, 2010 at 8:12 AM, Huan Sun<[email protected]> wrote:
Hi Michael,
Thank you for the suggestions. I will look at the filtered matrix
class.
I understand that distribute_local_to_global() usually requires matrix
and righthand side at the same time. That's why I need an auxiliary
matrix matrix_for_bc.
I was basically following tutorial-34 and my Poisson test is much
simpler than the hydrodynamic system solved there. So I guess I am using
ConstraintMatrix in a legitimate way ... somehow something goes wrong.
Also I have one more question on the inhomogeneous boundary condition:
in this case is VectorTools::project() forbidden?
Thanks,
Huan
On Tue, 2010-05-18 at 04:12 +0200, Michael Rapson wrote:
Hi Huan,
I also sometimes separate the matrix and right hand side assembly.
However, care must be taken because the method
distribute_local_to_global uses both the right hand side matrix and
the system matrix. Basically, the algorithm scales the Dirichlet
boundary condition relative to the size of the terms in the matrix. In
generally it will not work to apply the method to matrix and right
hand side separately (hence the method which takes both) or even to
apply the method again on a matrix that has had some of the terms
assembled already.
One option in the library is to use the FilteredMatrix class. (Which
should work as long as you don't need to add terms to the matrix.)
Another is to keep a copy of the matrix without the boundary
conditions applied which you keep initializing from, completing with
any extra terms and then use to apply the boundary conditions to.
Regards,
Michael
On Mon, May 17, 2010 at 8:18 PM, Huan Sun<[email protected]> wrote:
Hi Markus,
Yes, it works to use the original local stiffness matrix. But I want to
later separate the assemblage of the matrix and the right-hand side,
similar to what tutorial 32 does. So I was testing this easier case.
Somehow it doesn't work with this auxiliary matrix. I am just wondering
if there is anything wrong or missing in the code.
Thanks,
Huan
On Mon, 2010-05-17 at 08:29 +0200, Markus Bürg wrote:
Hello Huan Sun,
why do you create another matrix for the boundary conditions? You have
already created the problem's matrix for this cell right above. Thus
you can hand over just this matrix:
constraints.distribute_local_to_global (local_rhs, local_dof_indices,
rhs, local_matrix);
Best Regards,
Markus
Am 17.05.10 08:12, schrieb Huan Sun:
Hi all,
I was following tutorial 22 and 34 to implement inhomogeneous Dirichlet
boundary conditions using the ConsraintMatrix class but couldn't get it
right. I think the problem mainly lies in the assemblage part (i was
solving the basic Poisson's eq.)
if(constraints.is_inhomogeneously_constrained
(local_dof_indices[i])){
for(unsigned int j=0; j<dofs_per_cell; ++j){
matrix_for_bc(j,i) +=
grad_basis_phi[i] * grad_basis_phi[j] *
fe_values.JxW(q);
}//j-loop
}//if
...
constraints.distribute_local_to_global(local_rhs,
local_dof_indices,
rhs,
matrix_for_bc);
However, when I comment out the if statement, the program seems to work
correctly. The code is attached in this email.
Your help will be greatly appreciated.
Thanks!
Huan
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
