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.

4.1. GCD IN ONE VARIABLE 119<br />

be made effective, but the results tend to be unduly pessimistic, since it<br />

is extremely likely that all the bad primes would be clsutered just above<br />

2M.<br />

• In theory, the division tests could yield very large numbers if done as tests<br />

<strong>of</strong> the remainder being zero: <strong>for</strong> example the remainder on dividing x 100<br />

by x − 10 is 10 100 . This can be solved by a technique known as “early<br />

abort” trial division.<br />

Proposition 44 If h, <strong>of</strong> degree m, is a factor <strong>of</strong> f <strong>of</strong> degree n, the coefficient<br />

<strong>of</strong> x n−m−i in the quotient is bounded by ( )<br />

n−m 1<br />

i lc(h) ||f||.<br />

This is basically Corollary 22. Hence, as we are doing the trial division,<br />

we can give up as soon as we find a coefficient in the quotient that exceeds<br />

this bound, which is closely related to M (the difference relates to the<br />

leading coefficient terms).<br />

For example, if we take p = 7 in the example at the start <strong>of</strong> this chapter,<br />

we find that the g.c.d. <strong>of</strong> A 7 and B 7 is x + 10 (it could equally well be<br />

x + 3, but x + 10 makes the point better). Does x + 10 divide B? We<br />

note that ||B|| ≈ 23.92. Successive terms in the quotient are 3x 5 (and 3<br />

is a permissible coefficient), −30x 4 (and 30 < ( 5<br />

1)<br />

× 23.92) and 305x 3 , at<br />

which point we observe that 305 > ( 5<br />

2)<br />

× 23.92 = 239.2, so this cannot be<br />

a divisor <strong>of</strong> B. Hence 7 was definitely unlucky.<br />

With this refinement, it is possible to state that the numbers dealt with<br />

in this algorithm are “not much larger” than 2M, though considerable<br />

ingenuity is needed to make this statement more precise.<br />

If we apply this algorithm to the polynomials at the start <strong>of</strong> this section,<br />

√ ∑8<br />

we deduce that<br />

i=0 a2 i = √ √ ∑6<br />

113,<br />

i=0 b2 i = 2√ 143, and hence corollary 8<br />

gives a bound <strong>of</strong><br />

( )<br />

√113,<br />

2 6 2√ min<br />

143 ≈ 510.2, (4.4)<br />

3<br />

so our first prime would be 1021, which is indeed <strong>of</strong> good reduction. In this<br />

case, corollary 9 gives a bound <strong>of</strong><br />

( )<br />

1 √<br />

2 6 2 √<br />

min 113, 143 ≈ 72.8, (4.5)<br />

5 21<br />

so our first prime would be 149. In general, we cannot tell which gives us the<br />

best bound, and it is normal to take the minimum.<br />

Open Problem 4 (Improving Landau–Mignotte <strong>for</strong> g.c.d.) A significant<br />

factor in the Landau–Mignotte bound here, whether (4.4) or (4.5), was the<br />

2 min(8,6) contribution from the degree <strong>of</strong> the putative g.c.d. But in fact the<br />

exponent is at most 4, not 6, since the g.c.d. cannot have leading coefficient<br />

divisible by 3 (since A does not). Hence the g.c.d. must have at most the degree

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

Saved successfully!

Ooh no, something went wrong!