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.

4.5 The primality test of Agrawal, Kayal, and Saxena (AKS test) 209<br />

so every other member of J, in particular p, is congruent modulo r to some<br />

power of n. (The reader might note the similarity of this argument to Theorem<br />

4.3.3.)<br />

In the proof of Theorem 4.5.2 we have x + a ∈ G for each integer a with<br />

0 ≤ a ≤ ϕ(r)lgn, but all that we used is that this condition holds for<br />

0 ≤ a ≤ √ d lg n. We certainly have this latter condition currently. So now<br />

everything matches up, and the proof may be concluded in exactly the same<br />

way as in Theorem 4.5.2. ✷<br />

The preceding proof used some ideas in the nice survey paper [Granville<br />

2004a].<br />

With Theorem 4.5.2 we were constrained by the fact that while we<br />

conjectured that there are suitable values of r that are fairly close to lg 2 n,all<br />

that we could prove was that r ≤ lg 5 n (Theorem 4.5.3), though by ineffective<br />

methods this upper bound for r could be brought down to O(ln 3 n). But with<br />

Theorem 4.5.4 we are liberated from just looking at polynomials of the form<br />

x r − 1. We now have the complete freedom of looking at any and all monic<br />

polynomials f(x), as long as the degree exceeds lg 2 n and (4.27) and (4.28) are<br />

satisfied. Note that if n is prime, then by Theorem 2.2.8, a polynomial f(x)<br />

satisfies (4.27) and (4.28) if and only if f(x) is irreducible in Zn[x]. And it<br />

is easy to show that there are plenty of monic irreducible polynomials of any<br />

given degree (see (2.5) and Exercise 2.12). So why not just let d = ⌊lg 2 n⌋ +1,<br />

choose a polynomial of degree d that would be irreducible if n were prime,<br />

and be done with it?<br />

Unfortunately, things are not so easy. Irreducible polynomial construction<br />

over Fp, wherep is prime, can be done in expected polynomial time by the<br />

random algorithm of just choosing arbitrary polynomials of the desired degree<br />

and testing them. This is exactly the approach of Algorithm 4.3.4. But what<br />

if one wants a deterministic algorithm? Already in the case of degree 2 we<br />

have a known hard problem, since finding an irreducible quadratic in Fp[x] is<br />

equivalent to finding a quadratic nonresidue modulo p. Assuming the ERH,<br />

we know how to do this in deterministic polynomial time (using Theorem<br />

1.4.5), but we know no unconditional polynomial-time method. In [Adleman<br />

and Lenstra 1986] it is shown how to deterministically find an irreducible<br />

polynomial of any given degree in time polynomial in ln p and the degree,<br />

again assuming the ERH. They also consider an unconditional version of<br />

their theorem in which they allow a small “error.” That is, if the target<br />

degree is d, they find unconditionally and in time polynomial in ln p and d an<br />

irreducible polynomial modulo p of degree D, whered ≤ D = O(d ln p). In the<br />

paper [Lenstra and Pomerance 2005] this last result is improved to finding an<br />

irreducible polynomial modulo p with degree in [d, 4d], once p is sufficiently<br />

large (the bound is computable in principle), and assuming d>(ln p) 1.84 .<br />

(If one does not insist on effectivity, the lower bound for d may be relaxed<br />

somewhat.) Further, the number of bit operations to find such a polynomial<br />

is bounded by Õ(d8/5 ln n) (the notation Õ being introduced in the preceding<br />

subsection).

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

Saved successfully!

Ooh no, something went wrong!