Maximal Non-Classicality in a Multi-Setting Generalization of the CGLMP Inequality

Armin Tavakoli, Stefan Zohren, Marcin PawłowskiarXiv 2015 · arXiv

Abstract

The discrepancy between maximally entangled states and maximally non-classical quantum correlations is well-known but still not well understood. We aim to investigate the relation between quantum correlations and entanglement in a family of $N$-setting generalizations of the CGLMP inequality for $d$-outcome systems. Using analytical as well as numerical techniques, we derive both maximal quantum violations and violations obtained from maximally entangled states. Furthermore, we study the most non-classical quantum states in terms of their entanglement entropy for large values of $d$ and many measurement settings. Interestingly, we find that the entanglement entropy behaves very differently depending on whether $N=2$ or $N>2$: when $N=2$ the entanglement entropy is a monotone function of $d$ and the most non-classical state is far from maximally entangled, whereas when $N>2$ the entanglement entropy is a non-monotone function of $d$ and the most non-classical state converges to the maximally entangled state in the limit of large $d$.

Introduction: The Entanglement-Nonlocality Gap

In quantum information theory, entanglement and non-locality are often treated as synonymous markers of "quantumness." Intuitively, one might expect that a maximally entangled state—where particles are most strongly correlated—would also produce the most non-classical results in a Bell test. However, this intuition is famously challenged by the Collins-Gisin-Linden-Massar-Popescu (CGLMP) inequalities. For systems with more than two outcomes ($d > 2$), the state that maximally violates the CGLMP inequality is not maximally entangled.

This paper investigates whether this discrepancy is a fundamental property of high-dimensional quantum systems or merely a limitation of the standard two-setting ($N=2$) Bell test. By generalizing the CGLMP inequality to an arbitrary number of measurement settings $N$, the authors reveal a surprising phase shift in the relationship between entanglement and non-classicality.

Methodology

The study employs a multi-pronged approach combining analytical derivations with large-scale numerical optimization:

  • Chained Inequalities: The authors derive a simplified "chained" version of multi-setting Bell inequalities. This formulation is computationally more tractable while preserving the essential physics of the $N$-setting scenario.
  • Numerical Optimization: To identify the most non-classical states, the researchers used Semidefinite Programming (SDP) and "see-saw" optimization techniques to find the optimal measurement settings and quantum states.
  • High-Dimensional Scaling: For extremely large dimensions (up to $d = 500,000$), they utilized power iteration over Krylov subspaces to determine the maximum possible violation (the smallest eigenvalue of the Bell operator).
  • Analytical Limits: The behavior in the limit of infinite $N$ and $d$ was explored using Trigamma functions and Taylor expansions.

Results: The $N=2$ vs. $N>2$ Divide

The most significant finding is the discovery of a distinct change in behavior when the number of measurement settings exceeds two.

Entanglement Entropy

For the standard $N=2$ case, the entanglement entropy $E$ of the optimal state is a monotonically decreasing function of the dimension $d$. This means that as the system becomes more complex, the "most quantum" state becomes increasingly less entangled.

However, for $N \ge 3$, this trend reverses. The entanglement entropy becomes non-monotone: it initially decreases but then begins to climb back toward 1 (the value for a maximally entangled state) as $d$ increases.

Entanglement entropy $E$ vs. dimension $d$ for $N=2, 3, 10, 100$. Note the transition from monotone to non-monotone behavior.

Convergence to Maximality

To quantify how "far" the optimal state is from a maximally entangled state, the authors calculated the Kullback-Leibler (KL) divergence between their respective Schmidt coefficients. In the $N=2$ case, the KL divergence continues to grow with $d$, indicating that the states remain fundamentally different. In contrast, for $N \ge 3$, the KL divergence remains finite and eventually decreases as $N$ increases, suggesting that the most non-classical state converges to the maximally entangled state in the limit of many settings.

KL divergence between the most non-classical state and the maximally entangled state as a function of dimension $d$.

Discussion and Conclusion

The results suggest that the "glitch" where maximal non-classicality and maximal entanglement diverge is largely a consequence of restricted measurement settings. By allowing Alice and Bob more choices ($N > 2$), the most non-classical states naturally align with the most entangled ones as the dimensionality of the system grows.

While the numerical evidence is compelling up to $d \approx 10^6$, future work could focus on providing a rigorous global optimality proof for these measurements across all $N$. Additionally, extending this analysis to multipartite systems could reveal even more complex relationships between entanglement structures and non-local correlations. Experimentally, verifying these high-dimensional, multi-setting violations remains a frontier for quantum optics and trapped-ion systems.

Reference: Tavakoli, A., Zohren, S., & Pawlowski, M. (2021). Maximal Non-Classicality in Multi-Setting Bell Inequalities. arXiv:1506.04331v2 [quant-ph].

Summary updated

Abstract

The discrepancy between maximally entangled states and maximally non-classical quantum correlations is well-known but still not well understood. We aim to investigate the relation between quantum correlations and entanglement in a family of N-setting generalizations of the CGLMP inequality for d-outcome systems. Using analytical as well as numerical techniques, we derive both maximal quantum violations and violations obtained from maximally entangled states. Furthermore, we study the most non-classical quantum states in terms of their entanglement entropy for large values of d and many measurement settings. Interestingly, we find that the entanglement entropy behaves very differently depending on whether N=2 or N>2: when N=2 the entanglement entropy is a monotone function of d and the most non-classical state is far from maximally entangled, whereas when N>2 the entanglement entropy is a non-monotone function of d and the most non-classical state converges to the maximally entangled state in the limit of large d.
BibTeX
@misc{tavakoli2015maximal,
  author = {Armin Tavakoli and Stefan Zohren and Marcin Pawłowski},
  title = {{Maximal Non-Classicality in a Multi-Setting Generalization of the CGLMP Inequality}},
  year = {2015},
  eprint = {1506.04331},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1506.04331},
}

Loading related content...

Figure