Anders wrote:
> I would like the names to be as consistent with the poster as
> possible. After all, we are printing the UFL names in the poster next
> to the element names.
>
> I see two options:
>
> 1. Either we keep the UFL names as suggested:
>
> P dP
> P RTe/RTf dP
> P N1e N1f dP
> ...
>
> 2. Or, as suggested but all uppercase:
>
> P DP
> P RTE/RTF DP
> P N1E N1F DP
> ...
> I think I would vote for the second option.
I tend not to agree. If we have a notation that
uses subscripts and mixed case for the poster but
FEniCS translates that algorithmically by switching
everything to upper case and moving the scripts
to the line, it does not seem like a problem to me.
But I don't feel very strongly, and can't really
tell how it would affect the look of the poster without
seeing the alternatives.
Also, how do you propose to handle elements like
RTCE and DPC in UFL? Will it be ('DPC', quadrilateral, deg)
or ('DP', quadrilateral, deg) with the understanding that
the quadrilateral implies the C, since C means the cubical
variant. I prefer the latter. Insisting on typing the
C when it adds no meaning, doesn't seem right. Another possibility
is to make the DPC the basic name, but allow just DP,
which is unambiguous, as an alias. But I don't think
the unambiguous ('DP', quadrilateral, deg) should lead to an error.
One more issue of this sort: Is 'RT' going to be acceptable
as an alias for 'RTF', or will the user
be responsible for specifying RTF or RTE? ('BDM' should
then be treated consistently, and it should work the same
on simplices and cubes.)
Martin wrote:
> I was a bit quick there, the two names Q/S do of course carry
> information beyond the cell being different from the P cases. Can we
> perhaps make an alias Lagrange -> P/Q (DP/DQ for degree 0)?
Not only do Q and S spaces both use the same cell (cubes),
but P can be used on cubes as well (though not for continuous spaces).
For example, in the famous Q2-P1 Stokes element P1 (discontinuous)
is used for the pressure on cubes.
The notion of Lagrange elements is not completely standard. 1. Some
people call a Lagrange element any element for which all the DOFs
are nodal values (e.g., I just saw that in the book of Bernardi,
Maday and Rapetti). 2. Some people call the continuous P_r family
on simplices (r=1, 2, ...) the Lagrange finite elements. I think
this is most common. 3. Some people also throw in
discontinuous P_0. Perhaps convenient, but it is somewhat illogical
and leads to a lot of exceptions. E.g., "the Lagrange finite
element spaces belong to H^1, except for r=0", "the Lagrange finite
elements have one DOF at each vertex, except for r=0", "the
barycentric coordinates of the nodes for a Lagrange finite element
are of the form j/r, except for r=0...", etc. I'm don't think that we
should encourage this. It would be analogous to declaring
that RT0 is the space of DG0 vectors. No good reason to do
that except that there is no RT0 element so the name is
available. 4. Some people use Lagrange elements
for continuous P_r on simplices and continuous Q_r elements on
cubes. I don't see a problem with this. In this view, there are
simplicial Lagrange elements and cubic Lagrange elements,
just like there are simplicial RT elements and cubic RT elements.
So I guess my vote is Lagrange is an alias for P/Q, but I would
tend to discourage it going to DP/DQ for r=0.
- DNA
On 03/03/2014 07:23 AM, Martin Sandve Alnæs wrote:
I was a bit quick there, the two names Q/S do of course carry
information beyond the cell being different from the P cases. Can we
perhaps make an alias Lagrange -> P/Q (DP/DQ for degree 0)?
A couple of UFL implementation sidenotes:
All aliases should be applied in UFL just as we do today with short/long
names, not in the form compiler, otherwise the form signatures will
differ depending on what names you use.
If we deprecate (some of) the old names, the deprecation period should
be long because this touches a lot of code and keeping old aliases does
not carry any real cost to code maintenance.
Martin
On 3 March 2014 12:04, Anders Logg <[email protected]
<mailto:[email protected]>> wrote:
Both these options could be handled:
- Allow r = 0 for P elements and handle the switch from continuous to
discontinuous inside the form compiler, which will anyway need a lot
of if-else checking since many of the names are different in FIAT.
- Introduce shortcuts in some cases, but note that both the "Q" elements
and the "S" elements are essentially different from the "P" elements
so it doesn't make sense that "P" should work for all of them. In the
case of discontinuous elements, we would need to introduce an alias
for "DP" --> "DPC".
--
Anders
On Mon, Mar 03, 2014 at 12:30:50PM +0100, Garth N. Wells wrote:
>
> On 3 Mar 2014, at 12:09, Martin Sandve Alnæs <[email protected]
<mailto:[email protected]>> wrote:
>
> > I prefer the second option here over the first.
> >
> > But I don't like not being able to switch between (continuous)
piecewise linear and piecewise constant elements with just the
degree. There's a lot of code snippets a'la 'family = lamda degree:
"DG" if degree == 0 else "CG"' around in user code.
> >
>
> Agree. This can be annoying.
>
> > For similar reasons, I don't like having to select e.g. P,Q,S
depending on the cell type, when we already have the cell type
specified otherwise. That will make writing mesh-type independent
code more difficult when we get support for quads.
> >
>
> Agree.
>
> Garth
>
> > I'm pretty sure a family alias which maps to P/Q/S depending on
the cell for degree>0 and DP/DQ/DS for degree==0 would quickly
become the most popular one.
> >
> > Martin
> >
> >
> > On 3 March 2014 09:47, Anders Logg <[email protected]
<mailto:[email protected]>> wrote:
> > I would like the names to be as consistent with the poster as
> > possible. After all, we are printing the UFL names in the
poster next
> > to the element names.
> >
> > I see two options:
> >
> > 1. Either we keep the UFL names as suggested:
> >
> > P dP
> > P RTe/RTf dP
> > P N1e N1f dP
> >
> > P dP
> > P BDMe/BDMf dP
> > P N2e N2f dP
> >
> > Q dQ
> > Q RTce/RTcf dQ
> > Q Nce Ncf dQ
> >
> > S dPc
> > S BDMce/BDMcf dPc
> > S AAe AAf dPc
> >
> > 2. Or, as suggested but all uppercase:
> >
> > P DP
> > P RTE/RTF DP
> > P N1E N1F DP
> >
> > P DP
> > P BDME/BDMF DP
> > P N2E N2F DP
> >
> > Q DQ
> > Q RTCE/RTCF DQ
> > Q NCE NCF DQ
> >
> > S DPC
> > S BDMCE/BDMCF DPC
> > S AAE AAF DPC
> >
> > I think I would vote for the second option.
> >
>
_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics
