Spoofing cross entropy measure in boson sampling

TL;DR

This paper introduces a classical heuristic algorithm that surpasses current boson sampling experiments in cross entropy benchmarking, challenging claims of quantum advantage by efficiently approximating ideal distributions and spoofing large-scale systems.

Contribution

The authors develop a classical algorithm that achieves higher cross entropy scores than current boson sampling experiments and can scale to larger systems, questioning the reliability of XE as a quantum advantage metric.

Findings

Classical algorithm outperforms recent boson sampling XE scores.

The method scales to larger fermion sampling systems.

Analytic evidence suggests the algorithm can spoof noisy boson sampling efficiently.

Abstract

Cross entropy (XE) measure is a widely used benchmarking to demonstrate quantum computational advantage from sampling problems, such as random circuit sampling using superconducting qubits and boson sampling (BS). We present a heuristic classical algorithm that attains a better XE than the current BS experiments in a verifiable regime and is likely to attain a better XE score than the near-future BS experiments in a reasonable running time. The key idea behind the algorithm is that there exist distributions that correlate with the ideal BS probability distribution and that can be efficiently computed. The correlation and the computability of the distribution enable us to post-select heavy outcomes of the ideal probability distribution without computing the ideal probability, which essentially leads to a large XE. Our method scores a better XE than the recent Gaussian BS experiments when implemented at intermediate, verifiable system sizes. Much like current state-of-the-art experiments, we cannot verify that our spoofer works for quantum advantage size systems. However, we demonstrate that our approach works for much larger system sizes in fermion sampling, where we can efficiently compute output probabilities. Finally, we provide analytic evidence that the classical algorithm is likely to spoof noisy BS efficiently.