A Machine-Verified Proof of a Quantum-Optimization Conjecture (arxiv.org)

Researchers have successfully achieved a machine-verified proof of the Farhi, Goldstone, and Gutmann (FGG) conjecture concerning the Quantum Approximate Optimization Algorithm (QAOA). This conjecture, open for more than ten years, posited that the QAOA achieves an approximation ratio of \((2p+1)/(2p+2)\) for depth-$p$ on the ring of disagrees. Utilizing the large language model Claude Fable 5, the team formalized key components of the QAOA and leveraged the Lean 4 proof assistant for verification. This innovative methodology included a feedback loop that combined natural-language reasoning with mechanical verification, allowing the model to discover a hidden dynamical symmetry and streamline the proof construction. This development is significant for the AI and quantum information science communities as it demonstrates the potential of AI in solving complex mathematical conjectures, which could accelerate advancements in quantum optimization. By transitioning part of the problem-solving process to AI, the researchers showed that machine-generated proofs can be verified rigorously, thus enhancing the reliability of AI in formal mathematical reasoning. This successful intersection of AI, quantum theory, and formal verification elevates the prospects for addressing other long-standing problems in the field, paving the way for further breakthroughs in quantum computation.