Ask Your Question
2

Polynomial division in quotient rings?

asked 2020-03-29 00:49:43 +0100

hshackle gravatar image

I have a working example shown below to help explain my question.

R.<x> = GF(2)[];
I = R.ideal([x^2 - 1])
S.<u> = R.quotient_ring(I)
f = x^2 + x
g = x
print(f.quo_rem(g))
f = u^2 + u
g = u
print(f.quo_rem(g))

The first quo_rem call works fine, but the second does not. Is polynomial division not supported for quotient rings, or am I calling it the wrong way? I'm specifically interested in doing this for multivariate polynomials, but I can't seem to get it to work for a single variable either.

Related to the question, I am interested in understanding this division in quotient rings better. If the operation isn't supported in sage, any sort of references/algorithms on how to do such a calculation would be very helpful and appreciated.

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted
3

answered 2020-03-29 20:27:56 +0100

rburing gravatar image

updated 2020-03-30 12:21:24 +0100

The problem is that in such rings the quotient and remainder are not well-defined in general.

Consider the following example: divide $f = x^2 y + xy^2 + y^2$ by $y^2-1$ in the ring $\mathbb{Q}[x,y]/\langle xy-1\rangle$.

On the one hand, $$f = (x+1)(y^2-1) + 2x + 1 + x(xy-1),$$ so $(x+1)$ is "a quotient" and $2x+1$ is "a remainder" (and $x(xy-1)$ is zero in this ring).

On the other hand, $$f = 1\cdot(y^2-1) + x + y + 1 + (x+y)(xy-1),$$ so $1$ is "a quotient" and $x+y+1$ is "a remainder" (and $(x+y)(xy-1)$ is zero in this ring).

The problem is basically that division in a polynomial ring by several polynomials ($y^2-1$ and $xy-1$ in the example) is a priori not well-defined. For one, the problem is underdetermined. Choosing a monomial ordering and an ordering of the polynomials to be divided by allows you to define an unambiguous (multivariate) division algorithm. Unfortunately its output will depend on the orderings chosen. Furthermore, while a remainder of zero will mean that the input belongs to the ideal generated by the polynomials to be divided by, the converse is not necessarily true. Groebner bases were invented to address these issues. See for example the book Ideals, Varieties, and Algorithms by Cox, Little, and O'Shea. In fact I took the example from there; I just (slightly artificially) rephrased it to be about a quotient ring. If you are only interested in "a remainder" (any), then you can use the division algorithm of Theorem 3 in that book. If you want more, you have to read more. Good luck.


SageMath can find the remainder using the multivariate division algorithm:

sage: R.<x,y> = PolynomialRing(QQ,2,order='lex')
sage: (x^2*y+x*y^2+y^2).reduce([y^2-1,x*y-1])
2*x + 1
sage: R.<x,y> = PolynomialRing(QQ,2,order='invlex')
sage: (x^2*y+x*y^2+y^2).reduce([y^2-1,x*y-1])
y + x + 1

To find the quotient(s), I'm not sure if there is built-in method in SageMath. But it's not difficult to implement the division algorithm in SageMath, because things like monomial orderings and getting the leading term/coefficient, etc. are already there. There is also a trick to get the quotient(s) by adding extra variables to track what happens in the reduction. It goes something like this:

sage: R.<x,y,c1,c2> = PolynomialRing(QQ,4,order='lex(2),lex(2)')
sage: rem = (x^2*y+x*y^2+y^2).reduce([y^2-1-c1,x*y-1-c2])
sage: rem.coefficient(c1), rem.coefficient(c2)
(x + 1, x)
sage: rem.subs({c1:0,c2:0})
2*x + 1
sage: R.<x,y,c1,c2> = PolynomialRing(QQ,4,order='invlex(2),invlex(2)')
sage: rem = (x^2*y+x*y^2+y^2).reduce([y^2-1-c1,x*y-1-c2])
sage: rem.coefficient(c1), rem.coefficient(c2)
(1, y + x)
sage: rem.subs({c1:0,c2:0})
y + x + 1
edit flag offensive delete link more

Comments

Thanks! This is very helpful. Do you if the division algorithm they reference (or in general, division by several polynomials) is implemented in Sage?

hshackle gravatar imagehshackle ( 2020-03-30 06:08:02 +0100 )edit

You're welcome! I added details about the implementation in SageMath.

rburing gravatar imagerburing ( 2020-03-30 12:22:37 +0100 )edit

That's much appreciated, thank you. One final question on the extra variables that you add, just to make sure I'm understanding - the reduce function does the reduction in a reduced Groeber basis, and this procedure should cause problems if the original basis isn't a Groeber basis, correct? It seems that if the original basis isn't a Groebner basis, we get higher powers of c1 and c2 in the remainder (although there's always only one unique power of c1 and c2 involved in the remainder). Taking the coefficients corresponding to these higher powers of c1 and c2 seem to still give sensible answers for the quotients in the cases that I've considered, but I haven't been able to convince myself that this is always the correct thing to do.

hshackle gravatar imagehshackle ( 2020-03-31 01:00:57 +0100 )edit

The reduce method does division by a list of polynomials when you pass a list; the "problems" that can occur are the ones I mentioned before. A (reduced) Groebner basis is used only when you pass an ideal instead, or a list which happens to be such a basis. In the "trick", the monomial ordering is chosen such that the $c_i$ are smaller than all the original variables, so in the division algorithm (which eliminates leading terms) the $c_i$ will be introduced in favor of the originals. In the result actually arbitrary powers of $c_i$ can appear (just try reducing an expression which is a polynomial in the original generators), so this is more appropriate for a subalgebra test. It is easier just to program the division algorithm.

rburing gravatar imagerburing ( 2020-03-31 13:19:46 +0100 )edit

By the way, the quotients are generally non-unique in the sense that syzygys between the generators can exist.

rburing gravatar imagerburing ( 2020-03-31 13:36:33 +0100 )edit

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools

1 follower

Stats

Asked: 2020-03-29 00:49:43 +0100

Seen: 2,200 times

Last updated: Mar 30 '20