fenics team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, Apr 01, 2011 at 11:43:19PM +0200, Joachim Berdal Haga wrote:
> Today seems like a good day to announce the cbc.block project -- Block
> operators for DOLFIN (in Python). Some of you attended my presentation
> earlier this year and know this work already (although it is now in a
> much better state). For everyone else, a short description:
>
> - Block operators may be defined from standard DOLFIN matrices and vectors
>     A = assemble(...); B = assemble(...); # etc
>     AA = block_mat([[A,B], [C,D]])
>
> - Preconditioners, inverses, and inner solvers are supported
>     AAprec = AA.scheme('gauss-seidel', inverse=ML)
>
> - A good selection of iterative solvers
>     AAinv = SymmLQ(AA, precond=AAprec)
>     x = AAinv*b
>
> - Matrix algebra is supported both through composition of operators...
>     S = C*ILU(A)*B-D
>     Spre = ConjGrad(S)
>
>   ...and through explicit matrix calculation via PyTrilinos
>     S = C*InvDiag(A)*B-D
>     Spre = ML(collapse(S))
>
> - And lots of demos! Have a look: http://launchpad.net/cbc.block
>
> (I will be on holiday the next weeks and may be slow to answer questions.)

Very nice!

--
Anders