RSA public key cryptosystem

I got stuck on a homework question. If anyone could help me with this certain problem, I would be grateful. I'll state what the problem say and some relevant theorem (i believe) that I used to partly prove the problem.

Problem: A decryption exponent for an RSA public key $(N,e)$ is an integer $d$ with the property that $a^{de} \equiv a \mod N$ for all integers $a$ such that $\gcd(a,N) = 1$. Note that $N = pq$ but $p,q$ is unknown where $p,q$ are distinct primes and $e$ is the encryption exponent.

Suppose I can find a decryption exponent for a given $(N,e)$ where $N$ is a fixed modulus and for a large number $e$ that is not $N$. How can I factor $N$?

Attempt: According to the RSA public key cryptosystem, primes $p$, $q$ are picked such that they are distinct and an encryption exponent $e$ is chosen such that $\gcd(e, (p-1)(q-1)) = 1$.

So here's what I did: Suppose that I am given two public keys with both public keys having the same modulus, so I have $(N, e_1)$ and $(N,e_2)$. By hypothesis, I can find $d_1$ and $d_2$ such that $e_1d_1 \equiv 1 \mod (p-1)(q-1)$ and $e_2d_2 \equiv 1 \mod (p-1)(q-1)$. By definition of congruences, there exists $k_1, k_2 \in \mathbb{Z}$ such that $e_1d_1 - 1 = k_1[(p-1)(q-1)]$ and $e_2d_2 - 1 = k_2[(p-1)(q-1)]$. Then I took the $\gcd(e_1d_1-1,e_2d_2 -1) = \gcd(k_1,k_2)[(p-1)(q-1)]$.

Consider the case when $\gcd(k_1,k_2) = 1$. Then we found the value of $(p-1)(q-1)$, so $(p-1)(q-1) = pq - (p+q) + 1 = N - (p+q) + 1$. This implies that $(p+q) = N + 1 - (p-1)(q-1)$. So we use the quadratic formula to solve $X^2 - (p+q)X + N$ since this equation equals $(X-q)(X-p)$. Hence, I found the factors of $N$.

But if the case that $\gcd(k_1,k_2) > 1$ seems to got me stuck. If anyone can help me with this by providing a way to think about this problem, that'll be great.

-

I would try finding non-trivial square roots of $1$ modulo $N$. This is a bit probabilistic, but works in practice reasonably quickly I think.

You know that $de-1=\ell\,\mathrm{lcm}\{(p-1),(q-1)\}$. Note that both $p-1$ and $q-1$ are even, so they are not coprime. Write $$de-1=2^km,$$ with $m$ odd (this is easy as you can just keep dividing by two). Let $a$ be a random integer. Most likely it is coprime to $N$ (check with Euclid- if not , then you found a factor of $N$ and can quit). Compute the power $z=a^m$ modulo $N$. Keep squaring $z$ (modulo $N$). We know that $z^{2^k}\equiv 1\pmod N$, because $$z^{2^k}\equiv a^{2^km}\equiv 1\pmod N,$$ as $2^km$ is divisible by both $p-1$ and $q-1$. Let $k_1$ be the smallest integer, $0\le k_1\le k$, such that $z^{2^{k_1}}\equiv 1\pmod N$. If $k_1=0$, then we are out of luck, and must try another $a$. Otherwise let us examine $$x\equiv z^{2^{k_1-1}} \pmod N.$$ Then $x^2\equiv 1\pmod N$. If $x\equiv -1$, then, again we are out of luck, and should try another $a$. If not, then we are done, because $(x-1)$ (and $(x+1)$) will have a non-trivial common factor with $N$ that we can again find with Euclid's algorithm.

Can we be denied success by a string of bad luck? Not really! Let $U_{q,2}$ be the group of residue classes modulo $q$ of order that is a power of two (=2-Sylow subgroup of $\mathbb{Z}/q\mathbb{Z}^*$), and similarly $U_{p,2}$. Then $z$ is equally likely to land on any element of $U_{q,2}$ (resp. $U_{p,2}$), and by CRT these two choices are independent from one another. The process fails, if and only if $z$ has the same order $2^{k_1}$ in both groups, because that is when $z^{2^{k_1-1}}\equiv-1$ modulo both $p$ and $q$ resulting in $x\equiv -1\pmod{N}$. The chance of this happening is at most one half. If $|U_{q,2}|=|U_{p,2}|$, then we succeed at least when $a$ is a quadratic residue modulo exactly one of the factors $p$ and $q$, because $z$ is in the maximal proper subgroup of $U_{q,2}$ (resp. $U_{p,2}$), iff it is a quadratic residue modulo $q$ (resp. modulo $p$). If $|U_{q,2}|\neq|U_{p,2}|$, the situation is even better, because success is guaranteed, when $z$ is of maximal order in the bigger group.

So with at least an even money chance of success in each round, we will succeed sooner rather than later.

-
So a bit similar to Rabin-Miller primality testing. I think that the penultimate paragraph can be simplified (the business with quadratic residue vs. non-residue), but can't wrap my head around it now. –  Jyrki Lahtonen Jul 9 '12 at 11:42
@MathNewbie: In your other question square roots of integers modulo $pq$ were also studied. Here it is a special case of looking for square roots of one. Except this time we use the square roots to find factors, not the other way around as was the case there. –  Jyrki Lahtonen Jul 10 '12 at 6:09
@MathNewbie: This question is an even better match! In fact, they could all come from the same chapter of a textbook - one exercise logically leading to the next. –  Jyrki Lahtonen Jul 10 '12 at 6:12
Ah I see how I can apply it. Thanks. ;) –  MathNewbie Jul 10 '12 at 9:40