You lost the list here, others may have alternative solutions.

Martin
1. apr. 2014 16:46 skrev "Miroslav Kuchta" <[email protected]> følgende:

>  On 04/01/2014 04:21 PM, Martin Sandve Alnæs wrote:
>
> The normals used for dof definitions are chosen such that elements on both
> sides of a facet agree by design in FEniCS. As a consequence half of the
> facet normals you get should point inwards as you experience.
>
>
> So the normals that I see on the interior edges are the same regardless or
> whether I use Raviart-Thomas
> or the discontinous Raviart-Thomas?
>
>
>  What do you need this for? The symbolic term for what you ask for would
> be
>
>
> I'd like to compute divergence of vector u in each cell from the Gauss law
> and so I need to integrate normal
> fluxes over the cell boundary. If u was linear, then I could do the
> integration exactly by midpoint rule and this
> is where n*edge_length comes in.
>
>    n = FacetNormal(mesh)
>   a = FacetArea(mesh)
>    term = a*n
>
>  which you can use in forms directly but not (yet) interpolate.
>
>  Martin
>
>  Thanks, I'll think about how to put this to use.
>
> Miro
>
>
> On 1 April 2014 16:08, Miroslav Kuchta <[email protected]> wrote:
>
>> Hi everyone,
>>
>> for all the cells of my mesh, I'd like to compute the term [outer normal
>> of edge]*[edge length]
>> for each edge of the cell. It seemed that this term could be obtained by
>> interpolating Constant(1, 0)
>> and Constant(0, 1) to DRT1 space which would yield n_x*[edge length] and
>> n_y*[edge length] and
>> then putting the two together. The motivation for this is
>>
>> l_i((1, 0)) = \int_{e_i} n_x dl = n_x |e_i|, where l_i is the degree of
>> freedom of DRT1 on edge ei.
>>
>> Unfortunately this approach only works for some edges. Moreover, for
>> edges that are shared by
>> two cells I obtained the result vectors that point in the same direction.
>> Based on the form of l_i,
>> I imagine that since the normals are pointing in opposite directions, the
>> signs in the result should
>> be opposite as well. Is this assumption wrong? It seems that the normal
>> used in the integral is not
>> the always the outer one.
>>
>> Thanks for answer, Miro
>> _______________________________________________
>> fenics mailing list
>> [email protected]
>> http://fenicsproject.org/mailman/listinfo/fenics
>>
>
>
_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics

Reply via email to