Projective closure and affine patches for algebraic curves

[sagetrac-gjorgenson]

Given a curve in affine space, find its projective closure. Conversely, given a projective curve, find its affine patches with respect to the standard affine coordinate charts.

This could be done more efficiently by using the projective_embedding/affine patches functionality for affine/projective subschemes. Currently the projective_embedding function for affine subschemes doesn't always have the right codomain:

A.<x,y,z> = AffineSpace(3,QQ)
X = A.subscheme([y-x^2,z-x^3])
X.projective_embedding().codomain()

so this should be addressed here, along with creating a projective_closure function for affine subschemes.

Depends on #20697
Depends on #20698

CC: @bhutz @miguelmarco

Component: algebraic geometry

Keywords: gsoc2016

Author: Grayson Jorgenson

Branch/Commit: 91bc630

Reviewer: Ben Hutz

Issue created by migration from https://trac.sagemath.org/ticket/20676

[sagetrac-gjorgenson]

Branch: u/gjorgenson/ticket/20676

[sagetrac-git]

Commit: b3002ef

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

b3002ef

20676: First pass at implementation.

[miguelmarco]

comment:3

I have two questions:

1. Are you sure you really need to compute the groebner basis to compute the projective closue? If I am not missing something, just homogenizing the generators, you should get the generators of the projective ideal.

2. I think it would be better to keep the names of the variables when we compute the affine patches (or at least, to have an option to do so). It can be confusing to the user to force a change in the name of the coordinates. An option to choose the names of the new variables in the projective closure would be nice too.

[sagetrac-gjorgenson]

comment:4

Replying to @miguelmarco:

Thanks for looking at this. I think it's not always the case that a given set of generators is sufficient for finding generators for the ideal of the projective closure. One example I have seen referenced in some texts is the twisted cubic in A3 defined by the ideal (y-x^2, z-x^3). Its projective closure is (x^2 − wy, xy − wz, y^2 − xz), but if the given generators of the first ideal are homogenized we get (wy-x^2, zw<sup>2-x</sup>3) instead.

A proof is given in the book ideals, varieties, algorithms that if the generators of a groebner basis (with respect to a graded monomial ordering) for the defining ideal of an affine variety are homogenized, then the result is a set of generators for the ideal of the projective closure, so I tried to emulate that here.

I'll work on getting some flexibility with the variable names.

[bhutz]

comment:5

As Grayson pointed out the twisted cubic is the typical example for needing the gb.

For the variables. Actually it is not a good idea to name them the same as they really aren't the same variables. More importantly, you need to be careful to keep track of the embedding as well, since if you homogenize the affine patch, you would expect to get something back in the same projective space. Take a look at the functionality for projective space and projective subschemes (or in #16838) to see what I mean.

[miguelmarco]

comment:6

Thanks for the correction. I knew I was missing something.

About the names of the variables... right now they might be named the same or not deppending on what were the original names. So I still think that there should be at least on option to let the user decide how are the new variables named: the same way as before, maybe adding a "bar" at the end, or completely new ones.

And bhutz is totally right about the morphisms: that is the important thing to compute (either in this methods or in separated ones).

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

39dd215

20676: Attempt to keep track of ambient spaces.

[sagetrac-git]

Changed commit from b3002ef to 39dd215

[sagetrac-gjorgenson]

comment:8

Okay, I actually just finished up an attempt to better manage the creation of ambient spaces in the projective closure/affine patches computation. I tried piggy-backing off of the existing projective_embedding and affine_patch functionality for projective/affine spaces, and it seems to be working so far. Things like:

P.<x,y,z,w> = ProjectiveSpace(QQ,3)
C = Curve([y*z - x^2,w^2 - x*y])
C.ambient_space() == C.affine_patch(0).projective_closure().ambient_space()

and

A.<x,y,z> = AffineSpace(QQ,3)
C = Curve([y-x^2,z-x^3])
C.ambient_space() == C.projective_closure().affine_patch(0).ambient_space()

should now return true. Is this the right way to keep track of everything?

I also found something that seems strange:

A.<x,y,z> = AffineSpace(QQ,3)
C = Curve([y-x^2,z-x^3])
A == C.ambient_space()

returns false. As far as I can tell, this isn't an issue for projective space curves. Also, there seems to be an issue with the class structure of space curves vs plane curves. Right now there is a ProjectiveSpaceCurve_generic class, and a ProjectiveCurve_generic class, with the latter corresponding to plane curves, but the plane curve class does not inherit from the space curve class. Similarly for affine curves. Is there a reason why plane curves shouldn't inherit from the space curve class?

[miguelmarco]

comment:9

I would say it is a bug in Sage. Apparently, schemes don't have rich comparison implemented.

[bhutz]

comment:10

re: variable names.

The way this is handled for general affine and projective subschemes is to allow the user to specify the space that it will be embedded into (or dehomogenized into). This is better functionality than just specifying the variable names, since you can then embedded different curves into the same space.

re: ambient space

A.<x,y,z> = AffineSpace(QQ,3)
C = Curve([y-x^2,z-x^3])
A == C.ambient_space()

I haven't looked at the code, but it looks like Curve assumes you are passing in elements from a polynomial ring and doesn't see if they are actually from a particular affine space and hence creates a new affine space. This is a bug. If you pass in elements of the coordinate ring of an affine space, then the curve should be in that space. It may be that you need to include an optional parameter for the space and/or perhaps have an A.curve() function.

one last thought: Should these curve classes inherit from affine and projective subscheme as well? This could save you a fair amount of code duplication (such as for affine_patch and projective_embedding.

[sagetrac-gjorgenson]

Dependencies: #20697, #20698

[sagetrac-git]

Changed commit from 39dd215 to c6fe840

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

02349fe	20676: projective_embedding revision and projective_closure.
5de70c9	20697: plane curve classes now inherit from the space curve classes.
ef65aca	20676: Merge ticket 20697 into this ticket.
9cbdab3	20698: revised initialization of generic curves.
b718afa	20698: documentation and error formatting fixes.
d4eb8d4	20698: documentation spacing fixes.
3ffc3aa	20676: Merge ticket 20698 into this ticket.
c6fe840	20676: implemented curve-level functions.

[sagetrac-git]

Changed commit from c6fe840 to 0039d7b

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

058b909	20697: Change class names.
5a5fe9a	20697: change "plane_curves" folder name to "curves"
b3e8447	20697: Merge with latest beta.
3b5fdf8	20697: Merge in 20698 for access to better initialization of curves.
8541591	20697: Added documentation and made curve string labels match class names.
b918e77	20697: documentation adjustment, and attempt to fix pickle issue.
b380967	20697: Missed name changes.
54f4e62	20676: Merge again with 20697.
0039d7b	20676: Update example strings to match changes from 20697.

[bhutz]

Reviewer: Ben Hutz

[bhutz]

comment:15

• line 1916:

R.ideal([R(f) for f in self.defining_polynomials()])
could be replaced with
self.defining_ideal()

• projective closure should take projective space parameter

A.<x,y> = AffineSpace(QQ, 2)
P.<u,v,w>=ProjectiveSpace(QQ,2)
C = Curve([y-x^2], A)
D=C.projective_closure(1,P)

• Affine patch should take affine space parameter

A.<x,y> = AffineSpace(QQ, 2)
P.<u,v,w>=ProjectiveSpace(QQ,2)
C = Curve([u^2-v^2], P)
C.affine_patch(1,A)

• Affine patch should have a default patch and that default should match up with the projective closure default

A.<x,y,z> = AffineSpace(GF(3), 3)
C = Curve([y-x^2,z-x^3], A)
D=C.projective_closure()
D.affine_patch()==C

• you should mention in the doc for projective embedding that the image is the projective closure

[bhutz]

comment:16

err...having a default for affine_patch() doesn't really make sense, so ignore that part.

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

9262ae5

20676: changes from review, and spacing fixes.

[sagetrac-git]

Changed commit from 0039d7b to 9262ae5

[sagetrac-gjorgenson]

Changed keywords from none to gsoc2016

[sagetrac-git]

Changed commit from 9262ae5 to 91bc630

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

91bc630

20676: Merge this ticket with 7.3 beta3.

[bhutz]

comment:20

this looks good now

[vbraun]

Changed branch from u/gjorgenson/ticket/20676 to 91bc630