Odd orders in Shor's factoring algorithm

Shor's factoring algorithm (SFA) finds the prime factors of a number, $N=p_1 p_2$, exponentially faster than the best known classical algorithm. Responsible for the speed-up is a subroutine called the quantum order finding algorithm (QOFA) which calculates the order -- the smallest integer, $r$, satisfying $a^r \mod N =1$, where $a$ is a randomly chosen integer coprime to $N$ (meaning their greatest common divisor is one, $\gcd(a, N) =1$). Given $r$, and with probability not less than $1/2$, the factors are given by $p_1 = \gcd (a^{\frac{r}{2}} - 1, N)$ and $p_2 = \gcd (a^{\frac{r}{2}} + 1, N)$. For odd $r$ it is assumed the factors cannot be found (since $a^{\frac{r}{2}}$ is not generally integer) and the QOFA is relaunched with a different value of $a$. But a recent paper [E. Martin-Lopez: Nat Photon {\bf 6}, 773 (2012)] noted that the factors can sometimes be found from odd orders if the coprime is square. This raises the question of improving SFA's success probability by considering odd orders. We show that an improvement is possible, though it is small. We present two techniques for retrieving the order from apparently useless runs of the QOFA: not discarding odd orders; and looking out for new order finding relations in the case of failure. In terms of efficiency, using our techniques is equivalent to avoiding square coprimes and disregarding odd orders, which is simpler in practice. Even still, our techniques may be useful in the near future, while demonstrations are restricted to factoring small numbers. The most convincing demonstrations of the QOFA are those that return a non-power-of-two order, making odd orders that lead to the factors attractive to experimentalists.