On Exact Sizes of Minimal CNOT Circuits

Abstract: Computing a minimum-size circuit that implements a certain function is a standard optimization task. We consider circuits of CNOT gates, which are fundamental binary gates in reversible and quantum computing. Algebraically, CNOT circuits on $n$ qubits correspond to $GL(n,2)$, the general linear group over the field of two elements, and circuit minimization reduces to computing distances in the Cayley graph $G_n$ of $GL(n,2)$ generated by transvections. However, the super-exponential size of $GL(n,2)$ has made its exploration computationally challenging. In this paper, we develop a new approach for computing distances in $G_n$, allowing us to synthesize minimum circuits that were previously beyond reach (e.g., we can synthesize optimally all circuits over $n=7$ qubits). Towards this, we establish two theoretical results that may be of independent interest. First, we give a complete characterization of all isometries in $G_n$ in terms of (i) permuting qubits and (ii) swapping the arguments of all CNOT gates. Second, for any fixed $d$, we establish polynomials in $n$ of degree $2d$ that characterize the size of spheres in $G_n$ at distance $d$, as long as $n\geq 2d$. With these tools, we revisit an open question of [Bataille, 2022] regarding the smallest number $n_0$ for which the diameter of $G_{n_0}$ exceeds $3(n_0-1)$. It was previously shown that $6\leq n_0 \leq 30$, a gap that we tighten considerably to $8\leq n_0 \leq 20$. We also confirm a conjecture that long cycle permutations lie at distance $3(n-1)$, for all $n\leq 8$, extending the previous bound of $n\leq 5$.