Divide-and-conquer embedding for QUBO quantum annealing

TL;DR

This paper introduces a divide-and-conquer embedding method for quantum annealing that intentionally worsens embedding quality measures to enhance partial solutions, significantly improving performance on complex problems.

Contribution

It proposes a problem-focused division strategy for embedding in quantum annealing, demonstrating substantial performance gains over traditional methods.

Findings

Improved embedding performance by orders of magnitude.

Effective for irregular and geometrically frustrated systems.

Enhances partial solutions despite worse embedding quality.

Abstract

Quantum annealing promises to be an effective heuristic for complex NP-hard problems. However, clear demonstrations of quantum advantage are wanting, primarily constrained by the difficulty of embedding the problem into the quantum hardware. Community detection methods such as the Girvin--Newman algorithm can provide a divide-and-conquer approach to large problems. Here, we propose a problem-focused division for embedding, deliberately worsening typical measures of embedding quality to improve the partial solutions we obtain. We apply this approach first to the highly irregular graph of an integer factorisation problem and, passing this initial test, move on to consider more regular geometrically frustrated systems. Our results show that a problem-focused approach to embedding can improve performance by orders of magnitude.