[Dune] Re: ug intersection methods

[Peter Bastian]

Oliver Sander schrieb:
> Hi Sreejith!
> It seems to me that this is not an actual bug in UGGrid.  I think
> the reason is that, when implementing UGGrid, I interpreted an
> ambiguity in the definition of intersectionSelfLocal() and
> intersectionSelfGlobal() differently from the implementors
> of SGrid.
> 
> The specification says:
> 'intersectionSelfLocal () const
> geometrical information about this intersection in local coordinates of 
> the inside() entity. This method returns a Geometry object that provides 
> a mapping from local coordinates of the intersection to local 
> coordinates of the inside() entity.'
> 
> and analogously for intersectionNeighborLocal().  However, this
> is not completely precise.  First of all, it doesn't say anything
> about the orientation of the intersection wrt the element.  Secondly,
> with a given orientation I have to pick one cyclic permutation of
> the vertices of the intersection.
> 
> Since I've mostly worked with UG I tend to think in conforming meshes.
> Therefore, I implemented intersection...Local() such that the returned
> geometry corresponds to the corresponding subface of the element.
> In particular, this means the return value of intersectionSelfLocal()
> is oriented differently from the return value of 
> intersectionNeighborLocal().  That is why you don't get the same
> world coordinate for the local coordinate 0.25.  You would get the
> same if you chose 0.25 on one side and 0.75 on the other.
> 
> Note that this problem also occurs with intersectionGlobal().
> No orientation is specified.
> 
> This may be painful but I am afraid we have to come up with a
> precise specification of the intersection...() methods.
> 

It actually WAS my intention that the orientation is NOT specified. In 
our old DG implementation we really used a search algorithm that matches 
the integration points from both sides of a face.

I don't remember exactly but someone from Freiburg (Adrian, Andreas?) 
thought this was too inefficient and assumed an orientation (ie one has 
to assume that a local coordinate on a face corresponds to the same 
point on both sides). I do not know the current state of this and 
obviously this assumption never made it into the documentation. Could 
someone from FR clarify this point?


-- Peter
------------------------------------------------------------------
Peter Bastian, IWR,Uni Heidelberg, INF 348,R 020, 69120 Heidelberg
email: Peter.Bastian at iwr.uni-heidelberg.de   Tel: +49 6221 54 4984
WWW: http://www.iwr.uni-heidelberg.de/~Peter.Bastian Fax: ... 8860