# Miller-Rabin: Showing non trivial divisors of $n$

Let's assume $n$ doesn't pass the Miller-Rabin test and $b$ is a witness. Meaning, $b^{\frac{n-1}{2^r}} \equiv 1 \pmod{n}$ where $\frac{n-1}{2^r}$ is even, but $c = b^{\frac{n-1}{2^{r+1}}} \not\equiv \pm1 \pmod{n}$. Show that $\gcd(c+1,n),\gcd(c-1, n)$ are non-trivial divisors of $n$.

So first of all, for my convenience:

• Denote $n-1 = 2^ls$, $s$ is odd.
• Denote $k=l-r$.
• $c = b^{2^{k-1}s}$
• $c^2 = b^{2^{k}s}$
• $2^ks$ is even

It is given that $c^2 \equiv 1 \pmod{n} \implies (c-1)(c+1) \equiv 0 \pmod {n} \implies n\mid (c-1)(c+1)$

Now, I think we can assume that $n$ passed Fermat's theorem test (that's part of the initial tests of the Miller-Rabin algorithm).

Hence,

$$c^{r+1} = b^{2^l s} = b^{n-1} \equiv 1 \pmod {n}$$

but that isn't revealing anything new other than $r$ is odd (since $2$ must divide $r+1$).

What am I missing?

• An unusual definition of a witness. How did you arrive at $c^{r+1}$ ? We have $2^{r+1}$ in the denominator of the exponent defining $c$. – Peter Dec 7 '17 at 16:31
• @Peter, oh your're right. It suppose to be $c^{2^{r+1}}$ – Elimination Dec 8 '17 at 19:05

The general statement is that whenever we have a nontrivial square root of $1$ modulo $n$, a value $c$ such that $$c^2 \equiv 1 \pmod{n}, \qquad c \not\equiv \pm1 \pmod{n},$$ then we obtain a factorization of $n$.

If $c^2 \equiv 1 \pmod n$, then $(c+1)(c-1) \equiv 0 \pmod n$.

If $c+1$ were relatively prime to $n$, then it would have an inverse modulo $n$, and we could multiply by $(c+1)^{-1}$ to conclude that $c-1 \equiv 0 \pmod n$. But this is ruled out by our assumption that $c \not\equiv 1 \pmod n$.

If $c+1$ were divisible by $n$, then we would have $c+1 \equiv 0 \pmod n$. But this is ruled out by our assumption that $c \not\equiv -1 \pmod n$.

So the only remaining possibility is that $c+1$ shares some, but not all, prime factors with $n$: that $\gcd(c+1,n)$ is a nontrivial factor of $n$.

The same goes for the other factor $c-1$.

More generally, whenever we find $a$ and $b$ such that $a^2 \equiv b^2 \pmod n$ but $a \not\equiv \pm b \pmod n$, the same argument shows that $\gcd(a+b,n)$ and $\gcd(a-b,n)$ are nontrivial factors of $n$. This is the basis for many integer factorization algorithms starting from Fermat's factorization method and ending with the quadratic sieve.

From $b^{\frac{n-1}{2^r}}\equiv 1\pmod n$, we see that there exists an integer $m$ such that $$b^{\frac{n-1}{2^r}}=mn+1\tag1$$ From $b^{\frac{n-1}{2^{r+1}}}\not\equiv \pm 1\pmod n$, we see that there exist integers $k,r$ such that $$b^{\frac{n-1}{2^{r+1}}}=kn+r+1\tag2$$ where $$\text{1\le r\le n-1\quad with \quad r\not=n-2}\tag3$$

Since $\frac{n-1}{2^r}$ is even, we get, from $(1)(2)$, $$mn+1=b^{\frac{n-1}{2^r}}=b^{\frac{n-1}{2^{r+1}}\times 2}=\left(b^{\frac{n-1}{2^{r+1}}}\right)^2=(kn+r+1)^2=(k^2n+2kr+2k)n+(r+1)^2$$ implying $$1\equiv (r+1)^2\equiv r^2+2r+1\pmod n$$ which implies $$r(r+2)\equiv 0\pmod n\tag4$$

Here, supposing that $r+2=n+1$ implies $r\equiv 0\pmod n$ which is impossible.

So, from $(3)$, we have $1\lt r\lt n$ and $1\lt r+2\lt n$.

It follows from these and $(4)$ that both $r$ and $r+2$ are non-trivial divisors of $n$.

We have $b^{\frac{n-1}{2^{r+1}}}+1=kn+r+2$. Since $r+2$ is a non-trivial divisor of $n$, we see that $\gcd\left(b^{\frac{n-1}{2^{r+1}}} +1,n\right)$ is a non-trivial divisor of $n$.

We have $b^{\frac{n-1}{2^{r+1}}}-1=kn+r$. Since $r$ is a non-trivial divisor of $n$, we see that $\gcd\left(b^{\frac{n-1}{2^{r+1}}} -1,n\right)$ is a non-trivial divisor of $n$.