Classical Attacks on Peaked Quantum Circuits
Peaked random circuit sampling (Aaronson–Zhang) and heuristic peaked circuit benchmarks (HQAP, arXiv:2510.25838) were designed as quantum supremacy challenges where random-looking circuits concentrate output probability onto a secret target bitstring. We developed structural de-obfuscation and compressed tensor-network methods that recover the secret peak classically without reading target bitstrings at runtime.
Using these tools, we solved all 12 problem instances on the BlueQubit Peak Portal challenge (2,570/2,570 points, first-time correct). On instance P9 (56 qubits), our middle-out operator solver finds the peak in approximately 3 seconds on a laptop, compared to approximately 1 hour on an NVIDIA A100 GPU reported by prior published methods (Kremer & Dupuis, arXiv:2604.21908). On P6 (62 qubits), we solved the instance by discovering ring-partitioned star tensor-network contractions where gap layers factor into ring-local clusters.
Methodology: The approach exploits approximate mirror symmetries, groups of commuting blocks, and causal backward light-cone marginal evaluation. Outer layers are contracted through cached tensor-network paths, while the approximate core is solved via threshold-free frontier gate optimization.