dolfin team mailing list archive

[Nuno David Lopes <ndl@xxxxxxxxxxxxxx>]

On Tuesday 17 June 2008, you wrote:
> Nuno David Lopes wrote:
> > I hope it works.
> >
> > On Tuesday 17 June 2008, Nuno David Lopes wrote:
> >> For time-dependent problems were we don't change the system
> >> Matrix, it makes sense to apply Boundary conditions only to
> >> right-hand vector b?
> >> I've simplified the function:
> >> void apply(GenerecMatrix& A,GenericVector& b, const GenericVector* x,
> >>                const DofMap& dof_map, const ufc::form& form);
> >> and added
> >> void apply(GenericVector& b, const GenericVector* x,
> >>                const DofMap& dof_map, const ufc::form& form);
> >> so inside a loop we don't need to touch in the system matrix again.
>
> As far as I know, setting dirichlet conditions only changes the matrix
> the first time. The row that corresponds to the boundary dof is zeroed
> out and a one is put on the diagonal. Does it really do this again the
> second time around? Is there actually any computational cost associated
> with it?
Probably right...now i get no time differences in my tests 
i was thinking that when 
// Modify linear system (A_ii = 1)
 A.ident(boundary_values.size(), dofs);
and 
A.apply()
are called that the matrix was changed again...
(And I really don't understand why don't, but i'm at a superficial level of 
the code)

On the other hand saving results on the raw format, saves much time. 
in 50 iterations on my laptop with a 40121x40121 system i get:
110seconds wrinting 25 iterations of 14Mb vtu files and
96seconds  wrinting the same 25 iterations with 592k files.

> In my code I assemble the matrix once and factorize it (using UMFPACK).
> In the time-loop I only need to assemble the right-hand side and solve
> with the pre-computed factors (which takes no time at all).
>
> Anyway, I'm guessing that Anders will want to sort out the bug in the
> dirichlet bcs before he takes in your patch. I'll have use for it when
> that happens :)
>
> /Dag (PhD student, KTH (Sweden))
>
> >> I'll create a bundle and send it, also with  the RAWFile.h and
> >> RAWFile.cpp, for Xd3d  format.
> >
> > ------------------------------------------------------------------------
> >
> > _______________________________________________
> > DOLFIN-dev mailing list
> > DOLFIN-dev@xxxxxxxxxx
> > http://www.fenics.org/mailman/listinfo/dolfin-dev



-- 
Nuno David Lopes

e-mail:ndl@xxxxxxxxxxxxxx        (FCUL/CMAF)
           nlopes@xxxxxxxxxxxxxxx    (ISEL)
http://ptmat.ptmat.fc.ul.pt/%7Endl/ 

Tue Jun 17 13:38:41 WEST 2008