Of course, for large meshes the opposite happens.  For a UnitSquareMesh(n,
n), the number of vertices (and hence dimension of the space) is (n+1)^2,
while the number of cells is 2n^2.

On 20 April 2015 at 17:44, Anders Logg <[email protected]> wrote:

>  The way I see it, the global space would have 4 basis functions on the
> mesh consisting of two triangles, but the dimension of the space is still
> 2, meaning in particular that the mass matrix would be singular. (Which
> does not necessarily imply that this type of "element" would not be useful
> for anything.")
>
>  --
> Anders
>
>
> mån 20 apr. 2015 kl 18:31 skrev Jan Blechta <[email protected]>:
>
>> On Mon, 20 Apr 2015 18:25:03 +0200
>> Jan Blechta <[email protected]> wrote:
>>
>> > On Mon, 20 Apr 2015 16:00:52 +0000
>> > Anders Logg <[email protected]> wrote:
>> >
>> > > Yes, it might actually be that simple: The elements would be
>> > > standard P1 elements with the only difference that the value of the
>> > > basis functions are always one (each triangle has 3 basis function
>> > > and each is = 1) and all derivatives are zero.
>> >
>> > This element does not make a sense with this definition. Imagine 2D
>> > mesh of two triangles. There are 4 "basis" functions while the
>> > dimension of the space is 2. In the other words, these "basis"
>> > functions are not linearly independent hence they don't form a basis.
>>
>> In other words, span of suggested "basis" functions is exactly the
>> usual space of piece-wise constants. That's why it is not used as a FE
>> space. Rather as a transform for operator lumping.
>>
>> Jan
>>
>> >
>> > Jan
>> >
>> > >
>> > > Might be possible to "hack" by modifying
>> > > _create_fiat_element(ufl_element) in ffc/fiatinterface.py.
>> > >
>> > > --
>> > > Anders
>> > >
>> > >
>> > > mån 20 apr. 2015 kl 17:44 skrev Martin Sandve Alnæs
>> > > <[email protected]>:
>> > >
>> > > > Doesn't sound that hard. Basically dofmaps like CG1 elements with
>> > > > basis functions replaced by 1.0 on the entire support?
>> > > >  On 20 April 2015 at 15:37, Joakim Bø <[email protected]> wrote:
>> > > >
>> > > >>  Thanks for answering!
>> > > >>
>> > > >>
>> > > >>  Anders got it right, discontinous and overlapping basis
>> > > >> functions with the same global support as P1 tent functions.
>> > > >> Sorry to hear that it would be hard to implement, but it came as
>> > > >> no surprise...
>> > > >>
>> > > >>
>> > > >>  Thanks anyway!
>> > > >>
>> > > >>
>> > > >>  Joakim
>> > > >>
>> > > >>
>> > > >>  --
>> > > >> Joakim Bø
>> > > >> Prosjektleder ENT3R UiO
>> > > >> Tlf.: 915 24 326
>> > > >>
>> > > >> http://www.ENT3R.no/OSLO
>> > > >>   ------------------------------
>> > > >> *From:* Anders Logg <[email protected]>
>> > > >> *Sent:* 20 April 2015 13:46
>> > > >> *To:* Andrew McRae; Jan Blechta
>> > > >> *Cc:* Joakim Bø; [email protected]
>> > > >> *Subject:* Re: [FEniCS] Implement a new finite element type for
>> > > >> testing purposes?
>> > > >>
>> > > >>   If I understand correctly, you want discontinuous and
>> > > >> overlapping basis functions with the same global support as the
>> > > >> P1 tent functions. Unless you find a clever trick for how to
>> > > >> treat this (perhaps via some linear algebra using P0 elements in
>> > > >> combination with some suitable constraints), this looks
>> > > >> difficult to implement in FEniCS. We assume each element is
>> > > >> defined locally on triangles/tetrahedra.
>> > > >>
>> > > >>  --
>> > > >> Anders
>> > > >>
>> > > >>
>> > > >> mån 20 apr. 2015 kl 13:14 skrev Andrew McRae
>> > > >> <[email protected]>:
>> > > >>
>> > > >>>  I interpret it as a DG0, but where nodes are associated with
>> > > >>> vertices.  Related to mass-lumping, I guess.
>> > > >>>
>> > > >>> On 20 April 2015 at 12:07, Jan Blechta
>> > > >>> <[email protected]> wrote:
>> > > >>>
>> > > >>>> On Fri, 17 Apr 2015 10:21:33 +0000
>> > > >>>> Joakim Bø <[email protected]> wrote:
>> > > >>>>
>> > > >>>> > Hi!
>> > > >>>> >
>> > > >>>> >
>> > > >>>> > I am in need of a new type of basis function for testing
>> > > >>>> > purposes. It is much similar to the basis functions of the
>> > > >>>> > Taylor-Hood P1 element, the difference is that the functions
>> > > >>>> > are piecewise constant equal to 1 in this "local
>> > > >>>> > domain" (similar for 1D and 3D):
>> > > >>>> >
>> > > >>>> >
>> > > >>>> > [http://www.fsz.bme.hu/~szirmay/radiosit/rad10.gif]
>> > > >>>> >
>> > > >>>> >
>> > > >>>> > and zero in the rest of the domain. In general, phi_i = 1 for
>> > > >>>> > "local domain of dof i", 0 else.
>> > > >>>>
>> > > >>>> If I understand your explanation correctly (it does not seem to
>> > > >>>> match with the figure!), it is Discontinuous Lagrange element
>> > > >>>> of degree 0, which is implemented.
>> > > >>>>
>> > > >>>> Jan
>> > > >>>>
>> > > >>>> >
>> > > >>>> >
>> > > >>>> > Would it be possible to implement this without too much work?
>> > > >>>> > Or would it require a lot of effort?
>> > > >>>> >
>> > > >>>> >
>> > > >>>> > Thanks!
>> > > >>>> >
>> > > >>>> > Joakim
>> > > >>>> >
>> > > >>>> >
>> > > >>>> > --
>> > > >>>> > Joakim Bø
>> > > >>>> > Prosjektleder ENT3R UiO
>> > > >>>> > Tlf.: 915 24 326
>> > > >>>> >
>> > > >>>> > http://www.ENT3R.no/OSLO
>> > > >>>>
>> > > >>>> _______________________________________________
>> > > >>>> fenics mailing list
>> > > >>>> [email protected]
>> > > >>>> http://fenicsproject.org/mailman/listinfo/fenics
>> > > >>>>
>> > > >>>
>> > > >>>
>> > > >> _______________________________________________
>> > > >> fenics mailing list
>> > > >> [email protected]
>> > > >> http://fenicsproject.org/mailman/listinfo/fenics
>> > > >>
>> > > >>
>> >
>> > _______________________________________________
>> > fenics mailing list
>> > [email protected]
>> > http://fenicsproject.org/mailman/listinfo/fenics
>>
>> _______________________________________________
>> 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