[Trilinos-Users] Questions about InsertGlobalValues

[Heroux, Michael A]

Scot,

Epetra doesn’t have support for reduced storage of a symmetric matrix.  We have talked about it from the very beginning, but never had quite enough time and priority to make it happen.  It’s actually fairly straight-forward to support it, but working through the details would take time.

I know there is some interest in having it.  Is it a top priority for anyone?

Mike


On 8/26/10 3:04 PM, "M. Scot Breitenfeld" <brtnfld at uiuc.edu> wrote:

>   Hi Nico, My method is a particle based method; I'm just using the
> FECrsMatrix class because it allows off-processor row assignment whereas
> CrsMatrix does not. The formulation therefore lends itself to row based
> assignment since I loop over nodes. But I'm thinking it would be better if I
> pass InsertGlobalValues at least 3 rows (corresponding to the x,y,and z dofs)
> then just one row at a time. I looked at the documentation but I can not
> find how do you tell Epetra the K matrix is symmetric? I don't want it to
> store the other half of the matrix. On 08/26/2010 02:09 PM, Nico Schlömer
> wrote: > Hi Scot, > > the Epetra_FECrsMatrix class was written to help
> creating FE matrix as > easily as possible, i.e., without having to worry
> about which matrix row > sits on which processor at insertion. > Generally for
> FEM systems, you would loop over the elements and create > the *local
> stiffness matrices (of size 3-by-3 for linear triangular > elements for
> example). Once you have one of these, you can insert the > whole thing into
> the stiffness matrix at once by > >    InsertGlobalValues (const
> Epetra_IntSerialDenseVector&indices, const > Epetra_SerialDenseMatrix&values
> ) > > -- There's no need to do that row-by-row. > > The same goes for the
> graph: You may want to use Epetra_FECrsGraph, and > then just loop over the
> elements to fill in the sparsity structure. > > Does that help? > > Cheers, >
> Nico > _______________________________________________ Trilinos-Users
> mailing
> list Trilinos-Users at software.sandia.gov http://software.sandia.gov/mailman/lis
> tinfo/trilinos-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://software.sandia.gov/pipermail/trilinos-users/attachments/20100827/9f67d4c8/attachment.html