ffc team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

On 08/07/10 13:13, Anders Logg wrote:
> On Thu, Jul 08, 2010 at 01:02:43PM +0100, Garth N. Wells wrote:
> > On 08/07/10 12:59, Anders Logg wrote:
> > > On Thu, Jul 08, 2010 at 12:00:38PM +0100, Garth N. Wells wrote:
> > > > On 08/07/10 11:49, Kristian Oelgaard wrote:
> > > > > On 8 July 2010 08:22, Garth N. Wells<gnw20@xxxxxxxxx>    wrote:
> > > > > > On Jul 8 2010, Anders Logg wrote:
> > > > > >
> > > > > > > On Wed, Jul 07, 2010 at 10:34:39PM +0100, Kristian Oelgaard wrote:
> > > > > > > > On 7 July 2010 20:22, Garth N. Wells<gnw20@xxxxxxxxx>    wrote:
> > > > > > > > > On 07/07/10 20:14, Anders Logg wrote:
> > > > > > > > > > On Wed, Jul 07, 2010 at 06:26:20PM +0100, Kristian Oelgaard wrote:
> > > > > > > > > > > Supporting CellVolume makes it possible to do:
> > > > > > > > > > >
> > > > > > > > > > > CG = FiniteElement("Lagrange", triangle, 2)
> > > > > > > > > > > DG = FiniteElement("DG", triangle, 0)
> > > > > > > > > > > v = TestFunction(DG)
> > > > > > > > > > > f = Coefficient(CG)
> > > > > > > > > > > vol = triangle.v
> > > > > > > > > >
> > > > > > > > > > Would it be better to call it vol or volume instead of v? Or does it
> > > > > > > > > > have to be a one-letter word?
> > > > > > > > >
> > > > > > > > > . . . or call it 'volume'.
> > > > > > > >
> > > > > > > > It can be whatever we want, I just followed what was already there.
> > > > > > > > Should we then rename 'd', 'n' and 'x' to 'geometric_dimension',
> > > > > > > > 'facet_normal', and 'spatial_coordinate' while we're at it?
> > > > > > >
> > > > > > > I think d, n, x are fine, but v does not necessarily look like a
> > > > > > > volume to me (it looks like a test function).
> > > > > >
> > > > > > I agree - d, n and x are all commonly used, but v for volume isn't.
> > > > >
> > > > > I changed 'v' to 'volume' and fixed a few bugs along the way.
> > > > >
> > > > > On a related note, should we implement CellDiagonal too?
> > > > > We use cell.diagonal() for MeshSize in SpecialFunctions.h of DOLFIN
> > > > > and since we often use MeshSize for the 'h' coefficient in DG forms
> > > > > e.g., PoissonDG we could remove the need for this function evaluation.
> > > > > The code to compute the CellDiagonal could just be copied from
> > > > > IntervalCell, TriangleCell and TetrahedronCell of DOLFIN and dumped in
> > > > > codesnippets.py.
> > > >
> > > > What if we call it 'Circumradius'?
> > > >
> > > > I recall that the agreed not have have CellSize because of the
> > > > ambiguity in its definition, but circumradius is unambiguous, so I
> > > > vote to add it.
> > >
> > > Isn't 'diameter' better? It is shorter and well-defined
> > > (2*circumradius).
> >
> > 'diameter' isn't well defined. We can use it, but we would need to
> > define it as being 2*circumradius.
> >
> > Garth
>
> Why isn't it well-defined? The diameter of a set T is
>
>    d = sup_{x,y \in T} ||x - y||
>
> and that should be the same as 2*circumradius unless I'm mistaken.

What's the set T? At a glance, the above is true if T is not the 
polygon, but the set of all points inside the circle inside which the 
polygonal can be inscribed.

Garth

> --
> Anders