dolfin team mailing list archive

["Martin Sandve Alnæs" <martinal@xxxxxxxxx>]

2008/8/20 Anders Logg <logg@xxxxxxxxx>:
> On Wed, Aug 20, 2008 at 08:43:08AM -0600, Ostien, Jakob T wrote:
>> Hi,
>>
>> I need to be able to assemble an active set of integration points.  Essentially, to determine the set I loop over cells and then loop again over the integration points in the cell and determine if that integration point is active with some criteria.  Then I'd like to be able to assemble that set.
>>
>> This is a problem because currently the element_tensor does not break down into integration points, and I need to take derivatives, so the QuadratureElement in FFC is also ruled out.  I suppose I could calculate derivatives and then project on the QuadratureElement, but that seems sort of unclean.
>>
>> It seems like the recent discussion about integrating at a point (on the UFC list) might help me out here.
>>
>> Any other thoughts on how I might go about this?
>>
>> Jake
>
> I don't know yet. I'm thinking about how much of this should go into
> the compiler and how much should be left to the user.
>
> We're working on similar things (and I know Garth is too). I think a
> common denominator would at least be to be able to evaluate a form at
> a given point.
>
> I like Kent's suggestion earlier about adding a new UFC class to go
> along with the other three integral classes, something like
>
>  point_integral
>
> with tabulate_tensor taking the coordinates for a point as additional
> input.
>
> --
> Anders

I liked your version with adding a new function to each *_integral class better.
The reason is that each *_integral object will correspond to one
foo*dx(i) or bar*ds(j)
in the form definition, and each *_integral object will be able to
compute both an
integral over some cell and the corresponding integrand.

Although a separate point_integral might make sense for something I
don't know about,
matching them to corresponding cell_integral, exterior_facet_integral
and interior_facet_integral
objects will become messy.

--
Martin