dolfin team mailing list archive

[Johan Hake <hake@xxxxxxxxx>]

On Friday 19 December 2008 10:08:40 Kristian Oelgaard wrote:
> Quoting Johan Hake <hake@xxxxxxxxx>:
> > On Thursday 18 December 2008 22:34:14 Anders Logg wrote:
> > > On Wed, Dec 17, 2008 at 11:42:47PM +0100, Johan Hake wrote:
> > > > On Wednesday 17 December 2008 22:57:48 Anders Logg wrote:
> > > > > On Wed, Dec 17, 2008 at 10:53:32PM +0100, DOLFIN wrote:
> > > > > > One or more new changesets pushed to the primary dolfin
> > > > > > repository. A short summary of the last three changesets is
> > > > > > included below.
> > > > > >
> > > > > > changeset:   5406:bc43cc830c11058dcabf239a5a7f878bb860fabb
> > > > > > tag:         tip
> > > > > > user:        Anders Logg <logg@xxxxxxxxx>
> > > > > > date:        Wed Dec 17 22:53:27 2008 +0100
> > > > > > files:       TODO demo/la/eigensolver/python/demo.py
> > > > > > demo/pde/optimization/python/demo.py description:
> > > > > > Fix remaining unchecked demos. Some demos still broken:
> > > > > >
> > > > > >   EE demo/pde/nonlinear-poisson/python/demo.py
> > > > > >   EE demo/pde/poisson1D/python/demo.py                    viper
> > > > > > problem with 1D EE demo/pde/waveguide/python/demo.py
> > > > > >
> > > > > >   results in C++ and Python differ EE
> > > > > > demo/pde/dg/advection_diffusion/python/demo.py
> > > > > >   EE demo/nls/nonlinearpoisson/python/demo.py
> > > > > >   EE demo/mesh/partitioning/python/demo.py
> > > > > >   EE demo/mesh/intersection/python/demo.py                strange
> > > > > > transformation EE demo/ode/aliev-panfilov/python/demo.py
> > > > > > EE demo/ode/lorenz/python/demo.py
> > > > >
> > > > > We may ignore the ODE demos here. Some work is needed to get the
> > > > > ODE solvers to function from Python but they have only worked in
> > > > > part before so it's not very important now.
> > > > >
> > > > > The following demos remain:
> > > > >
> > > > >    demo/pde/nonlinear-poisson/python/demo.py            ?
> > > > >    demo/pde/poisson1D/python/demo.py                    viper
> > > > > problem with 1D demo/pde/waveguide/python/demo.py                  
> > > > >  results
> >
> > in
> >
> > > > > C++ and Python differ
> > > > >    demo/pde/dg/advection_diffusion/python/demo.py       ?
> > > >
> > > > This demo is somewhat depricated as it stands now. I added it based
> > > > on the previous c++ demo. Since then has the c++ demo changed, to be
> > > > more simple, but including the builtin OutflowFacet special function.
> > > > This is not included in the python interface.
> > > >
> > > > To get it up and running in python we need to add OutflowFacet to the
> > > > python interface. This should be doable, and quite straightforward
> > > > with the new function interface. Please correct me:
> > > >
> > > > class OutflowFacet(ffc.Function,cpp.OutflowFacet):
> > > >     def __init__(self, vec_field):
> > > >         appropriate check on vec_field
> > > >         mesh = vec_field.function_space().mesh()
> > > >
> > > >         # Define the OutflowFacet form
> > > >         n = FacetNormal(mesh)
> > > >         a = ffc.dot(vec_field,n)*ffc.ds
> > > >         self._dolfin_form = Form(a)
> > > >
> > > >         domain  = dim2domain[mesh.topology().dim()]
> > > >         self._element = ffc.FiniteElement("Discontinuous Lagrange",
> > > > domain, 0)
> > > >
> > > >         ffc.Function.__init__(self, self._element)
> > > >         cpp.OutflowFacet(self, self._dolfin_form)
> > > >
> > > > It is a bastard wrt being a full fledged dolfin.Function, it does not
> > > > define its own function space. I do not know if this would be a
> > > > problem as the c++ counterpart neither initiate its base class with a
> > > > functionspace.
> > > >
> > > > Once this function is in place the rest should be quite straight
> >
> > forward.
> >
> > > > Johan
> > >
> > > It looks strange to me, and so does the C++ version (that a Function
> > > needs to be initialized with a form).
>
> That's what makes it a 'SpecialFunction' :)
>
> > > How about just evaluating the function at the midpoint of the facet
> > > and computing the inner product with the facet normal?
> >
> > Sounds resonable enough to me. I suppose an integral over the facet is
> > potentially more aqurate than just the midpoint evaluation.
>
> Yes, I guess this will be OK, and if it turns out that one needs the
> integral for accuracy reasons it is still possible to implement the
> function in the main.cpp file.
> We could do this for the CPP demo and use the 'new' OutflowFacet in the
> Python demo.

If noone(tm) does not object to this difference in c++ vs python version of 
the demo, I can update the python one, and take a try on the OutflowFacet 
function in SpecialFunction to a not-so-special function. This will then be 
instantiated using a FunctionSpace and the field function.

Johan