# How to write a classical version of Shor's algorithm

For learning purposes, I would like to write a classical version of Shor's algorithm. From what I have read, what makes this algorithm fast is the quantum FFT, which is used to find the period of the function $$a^k \bmod N$$ with the ultimate goal of finding the k that solves $$a^k \bmod N = 1$$.

Acknowledging that it would be impractically slow, I would like to write a version that uses the classical FFT. Certainly such an algorithm could factor small numbers.

What confuses me is that when I calculate the values of $$a^k \bmod N$$ to feed into the FFT, it is not that much harder skip the FFT and just find $$a^k \bmod N = 1$$ by brute force (similar to this question).

What am I missing here? Alternatively, if I had a black box that could calculate FFTs instantaneously, how would this change Shor's algorithm?

• Hi @Chuck. Welcome to QCSE. What do you mean "by brute force?" Do you mean keeping a large table and searching for such a $k$? Remember $N$ is meant to be very large. – Mark S Oct 19 '20 at 18:44
• Hi @ Mark S. By brute force, I mean starting with k=1 and doing the calculation. Then increment k by 1 and repeating until a^k mod N = 1. I realize that in practice N can be very large. But what if N is around 10,000? How would I use an FFT to solve that problem? – Chuck Oct 19 '20 at 19:51
• The quantum Fourier transform (QFT) is performed in quantum superposition. You may be able to classically envision such a superposition as a time-series of a lookup table that stores pairs $(k, a^k\bmod N)$, and then perform a (classical) Fourier transform (FFT) on the lookup table, to find the period $r$ of the table. If I were you, to get intuition on the QFT, I would build such a table without hunting for a particular $k$. This "building the table" in a blind manner is close to the idea of quantum superposition. – Mark S Oct 19 '20 at 20:52

In Shor's (quantum) algorithm, you compute the modular exponent only a small number of times—far smaller than $$N$$. The algorithm is probabilistic, and if you're lucky you may compute it only once.
After the modular exponentiation step, in principle the qubits encode the value of $$a^k\text{ mod }N$$ for every $$k$$, but you can't just search through the wave function for a value of $$1$$. If you could do that (sometimes called "postselection"), you could do far more impressive things than factoring numbers, such as solving Circuit SAT in linear time by postselecting on the circuit's output being $$1$$. The problem is that you don't get to choose the outcome of your measurement, and with overwhelming probability you'll get an outcome that you didn't want.