[dune-pdelab] Request for user input

Oliver Sander sander at igpm.rwth-aachen.de
Tue Jul 7 10:37:06 CEST 2015


A few more words on this.

> This is one aspect of the old complaint of the dune-fufem camp about the structure
> of dune-pdelab.  Basically, in my (Carsten's, etc) view, the pdelab operator system
> lacks one level of abstraction beneath the GridOperator.  It happens regularly that
> I need to assemble a stiffness or mass matrix, or a load functional, without that
> being part of an 'operator between functions space' or a residual formulation for
> a PDE.  Simply example: sometimes you need the Laplace stiffness matrix because you
> need the half-norm it creates for error control.  I'd like to be able to simply
> create such a matrix with no frills, no constraints, no magic FD approximations etc.
> 
> The l2volumefunctional was a small attempt to have something like this in pdelab.
> It does only one thing, but it is easy to understand.

Currently l2volumefunctional is only a half-baked solution.  To be able to use it I still
need a full GridOperator with two(!) spaces, constraints, and all.  What I would want
ideally is an abgespeckte version of GridOperator, which does only the calls to
lambda_volume, and receives only one FE space (the test space).  In dune-fufem,
this looks like this:

onefunction = // function object
std::vector<double> nodalWeights;   // result

FunctionalAssembler<FEBasis> functionalAssembler(feBasis);
L2FunctionalAssembler<GridType,Dune::FieldVector<double,1> > l2functional(onefunction);
functionalAssembler.assemble(l2functional,nodalWeights);

which I think is more readable than having to compute a residual of a nonlinear operator
at a dummy current iterate, just to test a given function against an fe basis.

:-)
Oliver

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 473 bytes
Desc: OpenPGP digital signature
URL: <https://lists.dune-project.org/pipermail/dune-pdelab/attachments/20150707/6ebcd40d/attachment.pgp>


More information about the dune-pdelab mailing list