On Wed, May 19, 2010 at 4:24 AM, Volker Braun <vbraun.n...@gmail.com> wrote:
> Hi Andrey,
>
> I've looked over your code and I think it would be good to extend the
> class hierarchy of cones. I'm thinking about
>
> Cone
> Cone_of_fan(Cone)
>  keeps reference of fan
> Cone_of_ToricVariety(Cone_of_fan)
>  keeps reference of toric variety
> Cone_of_DomainOfToricMorphism(Cone_of_ToricVariety)
>  knows the image cone (smallest cone containing image)
> Cone_of_RangeOfToricMorphism(Cone_of_ToricVariety)
>  knows the preimage cones
>
> etc...
>
> In order to allow this and future extensions, the
> RationalPolyhedralFan() must create new cones from a factory(**) that
> is passed to RationalPolyhedralFan.__init__(). The default factory
> would create Cone_of_fan objects. Do you agree?
>
> (**) One cannot use a virtual method of RationalPolyhedralFan,
> ToricVariety(RationalPolyhedralFan) since the constructor always calls
> base methods.


Hi Volker,

I do not support the idea of cones bound to toric varieties and
morphisms. Right now, for example, when you change the base field of a
toric variety it uses exactly the same fan and this is good not only
for speed and memory usage of this change itself, but also for cached
computations - working with any of the varieties different only by
fields will accumulate cone lattices, Gale transforms, etc. As far as
I know it would be very nice for certain applications to perform
computations over changing finite fields for the "same" toric variety.
If would be better if one did not have to reconstruct all fans and
cones each time, even if compatibility checks are suppressed.

On the other hand, I didn't think about it before but I definitely
agree with you that for morphisms we do need to store where do cones
go and where do they come from. What do you think about creating the
following:

- ToricLattice class.

- ToricLatticeMorphism, which is basically just a matrix and a pair of
lattices for domain/codomain  (I prefer codomain rather than range
since this is what is used in Sage now, plus by "range" one can mean
the actual image of the domain which can be smaller than the
codomain). No relation to cones/fans yet.

- While mathematically it is pretty much the same, for implementation
purposes we should create something like FanMorphism (a single cone
can always be turned into a fan). It should be either derived from
ToricLatticeMorphism or just store one inside (second is probably
better). In addition it has fans living in the domain and the codomain
of the ToricLatticeMorphism. The constructor should check that these
fans are compatible with the lattice morphism (unless check=False).
This class should have methods like image_cone and preimage_cones.
These functions will return cones you have mentioned, given a cone or
its index in the appropriate fan (cones are hashable and can be used
as keys if we implement it as a dictionary). This will allow the same
fans to be shared by different morphisms.

Note that with the approach above these morphisms have nothing to do
with schemes and toric varieties and can be incorporated exclusively
into sage.geometry. This will make framework of cones/fans quite
complete, polished, and stable. Even inside the geometry modules it
will be nicely structured:
* ToricLattice is "independent"
* ToricLatticeMorphism depends on ToricLattice
* Cone depends (slightly) on ToricLattice
* Fan depends on Cone
* FanMorphism depends on Fan and ToricLatticeMorphism
The variant allowing construction of Fans with cones designed for
morphisms creates circular dependencies in this diagram (or even ties
it further to toric varieties!) and it will be easier to introduce
bugs. Another technical argument against passing cone constructors is
that it may interfere with pickling and therefore fail
TestSuite(...).run(). (There is already a method in Cones that
discards associated polyhedron during pickling because something funny
is  going on there - it took me a few hours to figure out that this is
the problem, although I still don't know what happens in polyhedra).

Of course, in addition to all this talking I am also willing to
actually work and do this ;-)

FanMorphism should eventually induce morphisms between ToricVarieties
in the sense of general scheme morphisms in Sage (and I definitely
want to allow non-toric morphisms between toric varieties!).
Unfortunately, this part will likely require a lot of work on the
existing Sage morphisms. I am willing to do it as well but I very much
doubt that I can finish it by the end of August (surprisingly I got a
quite dense summer ahead ;-)). Fortunately, as I understand your
current code does not need it at all, as it relies on the
combinatorial description! So it should not be an obstacle for your
divisors/cohomology code for now and once "scheme morphisms of toric
varieties" are good enough, we can see if anything has to be "ported."

Having written the above arguments against deriving new classes and
recreating of objects I see that they may be applied also to my
classes ConeFace and Cone_of_fan. We may consider eliminating them,
but the main reason for their creation was to remember incidence
information in the face or cone lattice. They are also more likely to
be used interactively by users, so the ability to write
cone.fan_rays() instead of fan.ray_indices(cone) seems to be nice. As
I said, we may consider eliminating these classes if you agree with my
proposals above and think that they should be applied here as well.

Let me know what you think!
Andrey

-- 
To post to this group, send an email to sage-devel@googlegroups.com
To unsubscribe from this group, send an email to 
sage-devel+unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/sage-devel
URL: http://www.sagemath.org

Reply via email to