Wolfgang,
thank you so much. Since, I need the "inner" matrices as part of a block
matrix the second more expensive approach sounds like an option. Maybe, I
can get my head around how to use the matrix with the diagonal ones in my
algorithm.
Cheerio,
Martin
On Thu, Dec 4, 2008 at 12:30 AM, Wolfgang Bangerth
<[EMAIL PROTECTED]>wrote:
>
> Martin,
>
> > I am writing a code for an optimal control problem based on dealii. For
> > that purpose I need the mass and stiffness matrices only on the interior
> > nodes, i.e.,
> > after calling apply_boundary_values() I have the matrix K and I want
> > \hat{K}=K(index,index) (in Matlab notation) where index indicates the
> > interior nodes. I then only want to work with \hat{K}.
>
> The matrix K after calling apply_boundary_values is essentially what you
> want: all rows and columns not in your index set of interior nodes are set
> to zero except for the diagonal element. In other words, if you imagine
> having numbered your degrees of freedom such that all interior nodes come
> first, and then the boundary nodes, then the top left part of the matrix
> is the matrix \hat K, and the bottom right one is a scaling of the
> identity matrix. You may want to try to build your algorithms around this.
>
> Alternatively, you could create a ConstraintMatrix object that sets the
> boundary nodes to zero. The ConstraintMatrix has set of functions that
> condense a matrix not by zeroing out rows and columns but by copying
> everything into a new matrix (which would be your \hat K). These functions
> aren't usually used because it is so expensive to generate a second
> sparsity pattern and matrix that people typically are happier working with
> a matrix that has a few zero rows and columns.
>
> Best
> W.
>
> -------------------------------------------------------------------------
> Wolfgang Bangerth email: [EMAIL PROTECTED]
> www:
> http://www.math.tamu.edu/~bangerth/<http://www.math.tamu.edu/%7Ebangerth/>
>
>
_______________________________________________
