# Choosing good parameter for Lenstra's elliptic curve factorization

In Wikipedia, there is an article explaining Lenstra's factorization algorithm. As far as I got it, we choose some $e \in \mathbb{N}$ and a point $P$ on the curve and then calculate $eP$. While calculating, we must invert elements, and if that does not work, we are rewarded with a factorization.

So far, so good. Now, Wikipedia (and some other books) suggest that $e=B!$ for some small $B \in \mathbb{N}$. Can someone please explain why this is so good? Also, why is there a relation in the choice of $B$ to Pollard's $p-1$ algorithm?

-

It becomes clear once you understand what $p-1$ and ECM are doing. I'll start with $p-1$, since it is simpler.

Suppose the number you want to factor, $n$, is the product of two primes: $p$ and $q$. If we knew $p-1$, we could get $p$ from $n$ by doing

$$\gcd(a^{p-1} - 1 \bmod n, n) = p.$$

This works because by working modulo $n$, we're working "in parallel" modulo $p$ and modulo $q$. And modulo $p$, we get $a^{p-1} - 1 = 0 \bmod p$ (for nonzero $a$) by Fermat's little theorem. So $a^{p-1} - 1 \bmod n$ is going to be 0 modulo $p$, but not modulo $q$, and the $\gcd$ gives us a factorization of $n$.

However, we don't really know $p-1$ — if we did, we wouldn't be looking for $p$ — and that puts a dent in our plans. There is an alternative route, however: if we can find a multiple of $p-1$, that works too: $$a^{k(p-1)} - 1 \bmod p \equiv (a^{p-1})^k - 1 \bmod p \equiv 1^k - 1 \bmod p = 0 \bmod p.$$

So what Pollard's $p-1$ does is select an exponent that is likely to contain every single factor of $p-1$. An easy way to do this, that works as long as $p-1$ only has small-ish factors, is simply to make the exponent the product of all primes: $2\cdot3\cdot5\cdot7\ldots\cdot B_1$, where $B_1$ is our upper bound. This, however, misses power-of-prime factors, such as $2^2$ or $3^5$. So another exponent that catches everything is the factorial of $B_1$: that one is certain to catch every possible factor up to $B_1$!

Some observations about $p-1$ leads us to ECM. $p-1$ is the order of the group formed by the integers modulo $p$, denoted by $\mathbb{F}_p$. Further, if the order of $\mathbb{F}_p$ only has small factors, we can find $p$ by the aforementioned exponentiation. It turns out this generalizes to other mathematical groups. In particular, the group of points in an elliptic curve modulo $p$ is parameterized by two additional numbers $a$ and $b$:

$$E_{a,b} = \{ (x, y) : y^2 = x^3 + a x + b \pmod{p} \}$$

Each curve (parameterized by a and b) modulo $p$ has an order ($\#E$, analogous to $p-1$ above) in the range $$p + 1 - 2\sqrt{p} < \#E < p + 1 + 2\sqrt{p}.$$ Further, for each order in that interval, there is an $(a,b)$ tuple that yields it.

Suppose we pick a random elliptic curve $E_{a,b}$, and run it modulo $n$ (again, "in parallel" in both $p$ and $q$). Picking a random point in the curve, $P$, and multiplying it by $\#E$ results in the point at infinity (the elliptic curve analogous of 0):

$$\#E\cdot P = \infty$$

This special point, usually represented as $(0,1)$, triggers divisions by $0 \bmod p$ but not $\bmod~q$, and that nets us a factor. The logic for picking the exponent is the same as $p-1$: every prime and prime power up to $B_1$, and hope for a hit.

Besides the factorial, another good way to pick the exponent is the lowest common multiple of every integer up to $B_1$; this yields every prime power lower than $B_1$, and is much smaller than the factorial of $B_1$.

-
To add to the above correct answer, one problem that the factorial has is that it overrepresents small factors. For example, 10000! (which is a relatively small multiplier; it's limited to finding orders that have no prime factors > 10000) has $2^{9995}$ as a factor; obviously, the order cannot have that as a factor. Including the extra powers of 2 won't prevent us finding the factor; however, the additional EC additions it implies is a waste of time. Samuel's last suggestion is better; it doesn't waste time on large powers of small factors. –  poncho Dec 30 '12 at 1:14
@SamuelNeves Many Thanks! Just to be sure: In Pollard, we try a versatile (has many primes in it) exponent in order to get a multiple of $p-1$. In ECM, we try a versatile factor in order to get a multiple of $\#E$? Can you please check this? –  Johannes Jan 9 '13 at 16:50
I am unsure what you mean by "factor" in the ECM case; if you mean the integer multiplied by P, then yes, that's it. –  Samuel Neves Jan 9 '13 at 21:44
@SamuelNeves Yes, thanks. –  Johannes Jan 11 '13 at 15:37