10.12.2012 Views

Prime Numbers

Prime Numbers

Prime Numbers

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

6.2 Number field sieve 299<br />

polynomials do exist, and if we can find such polynomials, then the complexity<br />

of NFS is significantly lowered. The special number field sieve (SNFS) refers to<br />

the cases of NFS where we are able to find extraordinarily good polynomials.<br />

The SNFS has principally been used to factor many Cunningham numbers<br />

(these are numbers of the form b k ± 1 for b = 2, 3, 5, 6, 7, 10, 11, 12, see<br />

[Brillhart et al. 1988]). We have already mentioned the factorization of the<br />

ninth Fermat number, F9 =2 512 + 1, by [Lenstra et al. 1993a]. They used the<br />

polynomial f(x) =x 5 + 8 and the integer m =2 103 ,sothatf(m) =8F9 ≡ 0<br />

(mod F9). Even though we already knew the factor 2424833 of F9 (found by<br />

A. E. Western in 1903), this was ignored. That is, the pretty nature of F9<br />

itself was used; the number F9/2424833 is not so pretty!<br />

What makes a polynomial extraordinary is that it has very small<br />

coefficients. If we have a number n = b k ± 1, we can create a polynomial<br />

as follows. Say we wish the degree of f(x) tobe5.Writek =5l + r, wherer<br />

is the remainder when 5 is divided into k. Thenb 5−r n = b 5(l+1) ± b 5−r .Thus,<br />

we may use the polynomial f(x) =x 5 ± b 5−r ,andchoosem = b l+1 .Whenk<br />

is large, the coefficients of f(x) are very small in comparison to n.<br />

A small advantage of a polynomial of the form x d + c is that the order of<br />

the Galois group is a divisor of dϕ(d), rather than having the generic value<br />

d! for degree-d polynomials. Recall that the usefulness of free relations is<br />

proportional to the reciprocal of the order of the Galois group. Thus, free<br />

relations are more useful with special polynomials of the form x d + c than in<br />

the general case.<br />

Sometimes a fair amount of ingenuity can go into the choosing of special<br />

polynomials. Take the case of 10 193 − 1, factored in 1996 by M. Elkenbracht-<br />

Huizing and P. Montgomery. They might have used the polynomial x 5 − 100<br />

and m =10 39 , as suggested by the above discussion, or perhaps 10x 6 − 1and<br />

m =10 32 . However, the factorization still would have been a formidable. The<br />

number 10 193 − 1 was already partially factored. There is the obvious factor<br />

9, but we also knew the factors<br />

773, 39373, 561470969, 639701219449517, 4274417556076113498947,<br />

26409540111952717487908689681403.<br />

After dividing these known factors into 10 193 − 1, the resulting number n was<br />

still composite and had 108 digits. It would have been feasible to use either<br />

the quadratic sieve or the general NFS on n, but it seemed a shame not to<br />

use n’s pretty ancestry. Namely, we know that 10 has a small multiplicative<br />

order modulo n. This leads us to the congruence 10 64 3 ≡ 10 −1 (mod n),<br />

and to the congruence 6 · 10 64 3 ≡ 6 3 · 10 −1 ≡ 108 · 5 −1 (mod n). Thus,<br />

for the polynomial f(x) =5x 3 − 108 and m =6· 10 64 ,wehavef(m) ≡ 0<br />

(mod n). However, m is too large to profitably use the linear polynomial<br />

x−m. Instead, Elkenbracht-Huizing and Montgomery searched for a quadratic<br />

polynomial g(x) with relatively small coefficients and with g(m) ≡ 0(modn).<br />

This was done by considering the lattice of integer triples (A, B, C) with<br />

Am 2 + Bm + C ≡ 0(modn). The task is to find a short vector in this

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

Saved successfully!

Ooh no, something went wrong!