# The Universe of Discourse

Tue, 09 Oct 2007

Relatively prime polynomials over Z2
Last week Wikipedia was having a discussion on whether the subject of "mathematical quilting" was notable enough to deserve an article. I remembered that there had been a mathematical quilt on the cover of some journal I read last year, and I went to the Penn math library to try to find it again. While I was there, I discovered that the June 2007 issue of Mathematics Magazine had a cover story about the probability that two randomly-selected polynomials over Z2 are relatively prime. ("The Probability of Relatively Prime Polynomials", Arthur T. Benjamin and Curtis D Bennett, page 196).

Polynomials over Z2 are one of my favorite subjects, and the answer to the question turned out to be beautiful. So I thought I'd write about it here.

First, what does it mean for two polynomials to be relatively prime? It's analogous to the corresponding definition for integers. For any numbers a and b, there is always some number d such that both a and b are multiples of d. (d = 1 is always a solution.) The greatest such number is called the greatest common divisor or GCD of a and b. The GCD of two numbers might be 1, or it might be some larger number. If it's 1, we say that the two numbers are relatively prime (to each other). For example, the GCD of 100 and 28 is 4, so 100 and 28 are not relatively prime. But the GCD of 100 and 27 is 1, so 100 and 27 are relatively prime. One can prove theorems like these: If p is prime, then either a is a multiple of p, or a is relatively prime to p, but not both. And the equation ap + bq = 1 has a solution (in integers) if and only if p and q are relatively prime.

The definition for polynomials is just the same. Take two polynomials over some variable x, say p and q. There is some polynomial d such that both p and q are multiples of d; d(x) = 1 is one such. When the only solutions are trivial polynomials like 1, we say that the polynomials are relatively prime. For example, consider x2 + 2x + 1 and x2 - 1. Both are multiples of x+1, so they are not relatively prime. But x2 + 2x + 1 is relatively prime to x2 - 2x + 1. And one can prove theorems that are analogous to the ones that work in the integers. The analog of "prime integer" is "irreducible polynomial". If p is irreducible, then either a is a multiple of p, or a is relatively prime to p, but not both. And the equation a(x)p(x) + b(x)q(x) = 1 has a solution for polynomials a and b if and only if p and q are relatively prime.

One uses Euclid's algorithm to calculate the GCD of two integers. Euclid's algorithm is simple: To calculate the GCD of a and b, just subtract the smaller from the larger, repeatedly, until one of the numbers becomes 0. Then the other is the GCD. One can use an entirely analogous algorithm to calculate the GCD of two polynomials. Two polynomials are relatively prime just when their GCD, as calculated by Euclid's algorithm, has degree 0.

Anyway, that was more introduction than I wanted to give. The article in Mathematics Magazine concerned polynomials over Z2, which means that the coefficients are in the field Z2, which is just like the regular integers, except that 1+1=0. As I explained in the earlier article, this implies that a=-a for all a, so there are no negatives and subtraction is the same as addition. I like this field a lot, because subtraction blows. Do you have trouble because you're always dropping minus signs here and there? You'll like Z2; there are no minus signs.

Here is a table that shows which pairs of polynomials over Z2 are relatively prime. If you read this blog through some crappy aggregator, you are really missing out, because the table is awesome, and you can't see it properly. Check out the real thing.

 a0 a1 a2 a3 a4 a5 a6 a7 a8 a9 b0 b1 b2 b3 b4 b5 b6 b7 b8 b9 c0 c1 c2 c3 c4 c5 c6 c7 c8 c9 d0 d1 0 [a0] 1 [a1] x [a2] x + 1 [a3] x2 [a4] x2 + 1 [a5] x2 + x [a6] x2 + x + 1 [a7] x3 [a8] x3 + 1 [a9] x3 + x [b0] x3 + x + 1 [b1] x3 + x2 [b2] x3 + x2 + 1 [b3] x3 + x2 + x [b4] x3 + x2 + x + 1 [b5] x4 [b6] x4 + 1 [b7] x4 + x [b8] x4 + x + 1 [b9] x4 + x2 [c0] x4 + x2 + 1 [c1] x4 + x2 + x [c2] x4 + x2 + x + 1 [c3] x4 + x3 [c4] x4 + x3 + 1 [c5] x4 + x3 + x [c6] x4 + x3 + x + 1 [c7] x4 + x3 + x2 [c8] x4 + x3 + x2 + 1 [c9] x4 + x3 + x2 + x [d0] x4 + x3 + x2 + x + 1 [d1]

A pink square means that the polynomials are relatively prime; a white square means that they are not. Another version of this table appeared on the cover of Mathematics Magazine. It's shown at right.

The thin black lines in the diagram above divide the polynomials of different degrees. Suppose you pick two degrees, say 2 and 2, and look at the corresponding black box in the diagram:

 a4 a5 a6 a7 x2 [a4] x2 + 1 [a5] x2 + x [a6] x2 + x + 1 [a7]
You will see that each box contains exactly half pink and half white squares. (8 pink and 8 white in that case.) That is, exactly half the possible pairs of degree-2 polynomials are relatively prime. And in general, if you pick a random degree-a polynomial and a random degree-b polynomial, where a and b are not both zero, the polynomials will be relatively prime exactly half the time.

The proof of this is delightful. If you run Euclid's algorithm on two relatively prime polynomials over Z2, you get a series of intermediate results, terminating in the constant 1. Given the intermediate results and the number of steps, you can run the algorithm backward and find the original polynomials. If you run the algorithm backward starting from 0 instead of from 1, for the same number of steps, you get two non-relatively-prime polynomials of the same degrees instead. This establishes a one-to-one correspondence between pairs of relatively prime polynomials and pairs of non-relatively-prime polynomials of the same degrees. End of proof. (See the paper for complete details.)

You can use basically the same proof to show that the probability that two randomly-selected polynomials over Zp is 1-1/p. The argument is the same: Euclid's algorithm could produce a series of intermediate results terminating in 0, in which case the polynomials are not relatively prime, or it could produce the same series of intermediate results terminating in something else, in which case they are relatively prime. The paper comes to an analogous conclusion about monic polynomials over Z.

Some folks I showed the diagram to observed that it looks like a quilt pattern. My wife did actually make a quilt that tabulates the GCD function for integers, which I mentioned in the Wikipedia discussion of the notability of the Mathematical Quilting article. That seems to have brought us back to where the article started, so I'll end here.

[ Puzzle: The (11,12) white squares in the picture are connected to the others via row and column 13, which doesn't appear. Suppose the quilt were extended to cover the entire quarter-infinite plane. Would the white area be connected? ]