Lars Schlieper

We study quantum period finding algorithms such as Simon and Shor (and its variants Ekerå-Håstad and Mosca-Ekert). For a periodic function $f$ these algorithms produce -- via some quantum embedding of $f$ -- a quantum superposition $\sum_x \ket{x}\ket{f(x)}$, which requires a certain amount of output qubits that represent $\ket{f(x)}$. We show that one can lower this amount to a single output qubit by hashing $f$ down to a single bit in an oracle setting. Namely, we replace the embedding of $f$ in quantum period finding circuits by oracle access to several embeddings of hashed versions of $f$. We show that on expectation this modification only doubles the required amount of quantum measurements, while significantly reducing the total number of qubits. For example, for Simon's period finding algorithm in some $n$-bit function $f: \mathbb{F}_2^n \rightarrow \mathbb{F}_2^n$ our hashing technique reduces the required output qubits from $n$ down to $1$, and therefore the total amount of qubits from $2n$ to $n+1$. We also show that Simon's algorithm admits real world applications with only $n+1$ qubits by giving a concrete realization of a hashed version of the cryptographic Even-Mansour construction. Our oracle-based hashed version of the Ekerå-Håstad algorithm for factoring $n$-bit RSA reduces the required qubits from $(\frac 3 2 + o(1))n$ down to $(\frac 1 2 + o(1))n$. In principle our hashing approach also works for the Mosca-Ekert algorithm, but requires strong properties of the hash function family. A hashed version of Mosca-Ekert with as few as $\mathcal{O}(\log n)$ qubits would imply classical polynomial time factoring. keywords: Quantum period finding, Simon, Even-Mansour, Shor, Ekerå -Håstad, Mosca-Ekert, minimizing qubits

Let $f: \mathbb{F}_2^n \rightarrow \mathbb{F}_2^n$ be a Boolean function with period $\vec s$. It is well-known that Simon's algorithm finds $\vec s$ in time polynomial in $n$ on quantum devices that are capable of performing error-correction. However, today's quantum devices are inherently noisy, too limited for error correction, and Simon's algorithm is not error-tolerant. We show that even noisy quantum period finding computations lead to speedups in comparison to purely classical computations. More precisely, we implemented Simon's quantum period finding circuit on the $15$-qubit quantum device IBM Q 16 Melbourne. Our experiments show that with a certain probability $\tau(n)$ we measure erroneous vectors that are not orthogonal to $\vec s$. We propose new, simple, but very effective smoothing techniques to classically mitigate physical noise effects such as e.g. IBM Q's bias towards the $0$-qubit. After smoothing, our noisy quantum device provides us a statistical distribution that we can easily transform into an LPN instance with parameters $n$ and $\tau(n)$. Hence, in the noisy case we may not hope to find periods in time polynomial in $n$. However, we still obtain quantum advantage even for large errors $\tau(n)$ close to $\frac 1 2$. Thus, period finding does not necessarily require full quantum error correction capability. keywords: Noise-tolerant Simon period finding, IBM Q 16, LPN algorithms, quantum advantage