发表机构
Instituto Milenio de Investigación en Óptica, Universidad de Concepción; Facultad de Ciencias Físicas y Matemáticas, Departamento de Física, Universidad de Concepción; ICFO - Institut de Ciencies Fotoniques, The Barcelona Institute of Science and Technology（孔塞普西翁大学米列尼奥光学研究所; 孔塞普西翁大学物理与数学科学学院物理系; 巴塞罗那科技大学光子科学研究所）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
提出一种自引导协议，通过变分优化和CSPSA算法最大化网络贝尔不等式违背，实现量子网络非局域性的设备无关认证，并在三角形网络上验证其有效性。
AI 中文摘要
贝尔定理表明量子理论与局部隐变量模型不相容。近年来，关于非局域性的研究已超越贝尔的原始场景，扩展到量子网络，其中多个独立源将物理系统分发给远距离各方，产生由非线性而非标准贝尔不等式认证的相关性。在此，我们引入一种自引导协议，该协议通过变分优化各方的测量，以最大化网络贝尔不等式的违背，每一步的违背值通过局部泡利经典阴影高效评估，搜索由复同步扰动随机逼近（CSPSA）算法驱动。收敛后，返回的测量设置直接实施，并在无阴影的情况下重新评估不等式。这种两阶段结构将设备相关的搜索与仅依赖于观测统计和网络因果结构的证书分开。我们通过在三角形网络上使用车轮不等式进行数值模拟验证该协议，恢复了Fritz分布所达到的违背，并将认证扩展到非最大纠缠和噪声态。
英文摘要
Bell's theorem shows that quantum theory is incompatible with local hidden-variable models. In recent years, research on nonlocality has moved beyond Bell's original scenario to quantum networks, where multiple independent sources distribute physical systems among distant parties, giving rise to correlations certified by nonlinear rather than standard Bell inequalities. Here, we introduce a self-guided protocol that variationally optimizes each party's measurement to maximize the violation of a network Bell inequality, with the violation evaluated efficiently at each step via local Pauli classical shadows and the search driven by the Complex Simultaneous Perturbation Stochastic Approximation (CSPSA) algorithm. Once converged, the measurement settings it returns are implemented directly and the inequality is re-evaluated without shadows. This two-stage structure separates a device-dependent search from a certificate that depends only on the observed statistics and on the causal structure of the network. We validate the protocol by numerically simulating it on the triangle network using the Wagon-Wheel inequality, recovering the violation achieved by the Fritz distribution, and extending the certification to non-maximally entangled and noisy states.
Comments10 pages, 3 figures
