Database Reordering for Compact Grover Oracles with ESOP Minimization

TL;DR

This paper introduces a method to optimize quantum state preparation in Grover's algorithm by reordering database entries, using ESOP minimization and simulated annealing to significantly reduce circuit size.

Contribution

It proposes a novel database reordering technique combined with ESOP minimization and simulated annealing to minimize quantum circuit size for Grover oracles.

Findings

Reordering can reduce circuit size by up to 50%.

Simulated annealing achieves approximately 30% reduction compared to ESOP without reordering.

For larger databases, the method finds smaller circuits than random search.

Abstract

Grover's algorithm searches for data satisfying a desired condition in an unstructured database. This algorithm can search a space of size $N$ in $\sqrt{N}$ queries, thereby achieving a quadratic speedup. However, within the Grover oracle circuit that is repeatedly applied, the quantum state preparation circuit -- which embeds database information into quantum states -- suffers from a large gate count and circuit depth. To address this problem, we propose reducing the quantum state preparation circuit by reordering the database. Specifically, we consider a Quantum Read-Only Memory (QROM), where data are assigned to addresses, and assume that the address assignment of data can be freely permuted. By applying Exclusive Sum-of-Products (ESOP) minimization to the resulting truth table, we reduce the quantum circuit. Although the resulting circuit logic differs from the original, the state preparation remains correct in the sense that every desired datum is encoded at some address. Furthermore, we propose a proxy metric that estimates circuit size without compilation, and combine it with simulated annealing to efficiently find a near-optimal data ordering. In our experiments, an exhaustive search over all orderings for databases of size $N=8$ reveals that circuit size varies by up to approximately a factor of two depending on the ordering, demonstrating the utility of reordering. Compared with applying ESOP minimization without reordering, simulated annealing reduces the circuit size by approximately 30\% and yields circuits close to optimal. For $N=64$ and $128$, simulated annealing is shown to discover smaller circuits compared with random search.