19.07.2014 Views

Contents - Student subdomain for University of Bath

Contents - Student subdomain for University of Bath

Contents - Student subdomain for University of Bath

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

4.2. POLYNOMIALS IN TWO VARIABLES 127<br />

For the lemma to be false, the last g.c.d. has to be non-trivial. This implies<br />

that the resultant Res x (A y−v /C y−v , B y−v /C y−v ) vanishes, by proposition 58 <strong>of</strong><br />

the Appendix. This resultant is the determinant <strong>of</strong> a Sylvester matrix 3 , and<br />

|M y−v | = (|M|) y−v , <strong>for</strong> the determinant is only a sum <strong>of</strong> products <strong>of</strong> the coefficients.<br />

In the present case, this amounts to saying that Res x (A/C, B/C) y−v<br />

vanishes, that is that y − v divides Res x (A/C, B/C). But the hypotheses <strong>of</strong> the<br />

lemma exclude this possibility.<br />

Definition 71 If gcd(A y−v , B y−v ) = gcd(A, B) y−v , we say that the evaluation<br />

<strong>of</strong> this problem at v is good, or that y − v is <strong>of</strong> good reduction. If not, we say<br />

that y − v is <strong>of</strong> bad reduction.<br />

This lemma implies, in particular, that there are only a finite number <strong>of</strong> values v<br />

such that gcd(A y−v , B y−v ) does not have the same degree as that <strong>of</strong> gcd(A, B),<br />

that is the y − v which divide the g.c.d. <strong>of</strong> the leading coefficients and the y − v<br />

which divide the resultant <strong>of</strong> the lemma (the resultant is non-zero, and there<strong>for</strong>e<br />

has only a finite number <strong>of</strong> divisors). In particular, if A and B are relatively<br />

prime, we can always find a v such that A y−v and B y−v are relatively prime.<br />

4.2.3 G.c.d. in Z p [x, y]<br />

By Gauss’ Lemma (Theorem 5),<br />

gcd(A, B) = gcd(cont x (A), cont x (B)) gcd(pp x (A), pp x (B)),<br />

and the real problem is the second factor.<br />

Observation 9 While the content <strong>of</strong> A = ∑ m<br />

i=0 a ix i can be computed as<br />

gcd(a m , gcd(a m−1 , gcd(a m−2 , . . . , a 0 ) . . .),<br />

the following process is more efficient in practice (asymptotically, it’s only worth<br />

a constant factor on what is asymptotically the cheaper operation, but in practice<br />

it is worthwhile), and is valid over any g.c.d. domain R with a concept <strong>of</strong> ‘size’<br />

equivalent to degree.<br />

Algorithm 16 (Content)<br />

Input: A = ∑ m<br />

i=0 a ix i ∈ R[x].<br />

Output: cont x (A)<br />

S := {a i } # Really a set, as duplicates don’t matter<br />

g := minimal degree element <strong>of</strong> S; S := S \ {g}<br />

if g is a unit<br />

3 There’s a subtle point here. The resultant <strong>of</strong> polynomials <strong>of</strong> degrees m and n is the<br />

determinant <strong>of</strong> an (m + n) 2 matrix. Hence if y − v divides neither leading coefficient, the<br />

Sylvester matrix <strong>of</strong> A y−v and B y−v is indeed the reduction <strong>of</strong> the Sylvester matrix <strong>of</strong> A and<br />

B. If y −v divides one leading coefficient, but not the other, the Sylvester matrix <strong>of</strong> A y−v and<br />

B y−v is smaller, and the reader should check that this only makes a difference <strong>of</strong> a product<br />

<strong>of</strong> that leading coefficient which doesn’t vanish when v is substituted <strong>for</strong> y.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!