⚛️ quantum physics

Symmetric multipartite Bell inequalities via Frank-Wolfe algorithms

This paper leverages the symmetry of GHZ state correlation tensors and Frank-Wolfe algorithms to efficiently derive new symmetric multipartite Bell inequalities that improve upon existing bounds for nonlocality robustness and detection efficiency, including demonstrating nonlocality activation in star networks with finite measurements.

Original authors: Sébastien Designolle, Tamás Vértesi, Sebastian Pokutta

Published 2026-10-07

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Sébastien Designolle, Tamás Vértesi, Sebastian Pokutta

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Quantum mechanics describes a world where particles can be linked in ways that defy our everyday experience. When two or more particles are "entangled," a measurement performed on one instantly influences the state of the others, no matter how far apart they are. This phenomenon, known as nonlocality, is not just a theoretical curiosity; it is a fundamental feature of nature that distinguishes the quantum world from the classical one. To prove that a system is truly quantum and not just behaving according to hidden, local rules, scientists use mathematical tests called Bell inequalities. If a system violates these inequalities, it confirms the presence of genuine quantum connections. However, as the number of particles in a system grows, the complexity of these tests explodes. The mathematical space required to describe the possible behaviors of ten particles is so vast that it has historically been impossible to calculate the limits of what is possible with classical physics, leaving a gap in our understanding of how robust these quantum links are against noise and imperfections.

A team of researchers has now bridged this gap for systems involving up to ten parties by developing a new way to navigate this immense mathematical landscape. They focused on a specific type of quantum state, known as the Greenberger-Horne-Zeilinger state, which involves multiple particles prepared in a highly entangled configuration. In their experiments, each particle is measured using a set of directions arranged like the points of a regular polygon on a circle. The researchers realized that this specific arrangement creates a powerful symmetry: the relationships between the particles repeat in a predictable pattern. By exploiting this symmetry, they were able to drastically shrink the size of the problem they needed to solve. Instead of trying to calculate every single possible outcome for ten particles, which would require tracking billions of billions of combinations, they reduced the problem to a manageable size by grouping identical outcomes together.

Using a sophisticated mathematical tool known as a Frank-Wolfe algorithm, which is designed to find optimal solutions by taking iterative steps rather than checking every possibility, the team calculated the precise limits of how much noise these quantum systems can tolerate before they lose their nonlocal properties. They found that for systems with three to ten parties, they could determine the exact threshold of noise that breaks the quantum connection. These results provide the best-known upper bounds on the robustness of these states, meaning they tell us exactly how "pure" the quantum state must be to remain detectable. For a system with ten parties, the calculation that would have been impossible without their symmetry tricks was reduced to a task that took ten days on a powerful supercomputer, whereas the unoptimized version would have been computationally intractable.

One of the most striking outcomes of this work is a general improvement on a famous inequality known as Mermin's inequality. For systems where each party performs four measurements, the researchers discovered a pattern that allows them to construct a new inequality that is more resistant to noise than any previously known method, regardless of the number of parties involved. This means that for larger networks of entangled particles, it is now possible to detect quantum behavior even when the signal is significantly degraded by environmental interference. Furthermore, the team calculated the efficiency required for detectors to successfully observe these quantum effects. They found that their new inequalities allow for the detection of nonlocality with fewer measurements and lower detector efficiency than before, making these experiments more feasible for real-world applications.

The study also demonstrated a phenomenon called the activation of nonlocality in star-shaped networks. In such a network, a central node shares entangled pairs with several surrounding nodes. Previous research suggested that nonlocality could only be activated in such networks if an infinite number of measurements were used. The researchers showed that with their new method, this activation can be demonstrated with a finite and relatively small number of measurements, specifically for a network with ten surrounding parties. This proves that complex quantum networks can exhibit nonlocal behavior under conditions that are much more practical than previously thought. The work stands as a proof of concept that symmetry is not just a mathematical curiosity but a powerful computational tool. By recognizing and utilizing the repeating patterns in quantum data, scientists can solve problems that were once considered too large to tackle, opening the door to a deeper understanding of how quantum networks function and how they might be used in future technologies.

Technical Summary: Symmetric Multipartite Bell Inequalities via Frank-Wolfe Algorithms

Problem Statement
The paper addresses the computational intractability of characterizing the nonlocality of multipartite quantum states, specifically the Greenberger-Horne-Zeilinger (GHZ) state. In multipartite Bell scenarios, the parameter space scales exponentially with the number of parties ( $N$ ) and measurement settings ( $m$ ). Consequently, determining the "nonlocality robustness" (the critical visibility $v$ at which a quantum state can no longer be distinguished from local correlations) and computing the corresponding local bounds for Bell inequalities becomes impossible using standard linear programming or exhaustive enumeration for even moderate values of $N$ and $m$ . The authors focus on a specific scenario where $N$ parties perform $m$ planar measurements forming a regular polygon in the $XY$ plane of the Bloch sphere.

Methodology
The authors propose a framework that combines two primary tools to overcome the exponential scaling:

Symmetry Exploitation: They identify and exploit the specific symmetries inherent in the GHZ state and the regular polygon measurement settings. These symmetries are not ambient to the scenario but embedded in the specific instance. The authors define a group action generated by permutations of parties, cyclic shifts of measurement indices, and sign flips. By applying the Reynolds operator (an averaging operator over the group), they project the high-dimensional correlation tensor and the local polytope onto a significantly smaller symmetric subspace.
- This reduces the dimension of the search space from $m^N$ to $\lceil m/2 \rceil$ .
- It reduces the number of deterministic strategies to enumerate from $2^{(N-1)(m-1)}$ to a combinatorial term involving the number of binary necklaces, specifically $\binom{u_m + N - 2}{N-1}$ , where $u_m$ is the number of orbits under cyclic shifts.
Symmetric Frank-Wolfe Algorithm: To find the Bell inequality (a separating hyperplane) that maximizes the violation, the authors employ the Frank-Wolfe algorithm (a conditional gradient method). Instead of enumerating all vertices of the local polytope, the algorithm iteratively solves a Linear Minimization Oracle (LMO) to find the "worst-case" deterministic strategy relative to the current gradient.
- The algorithm operates within the reduced symmetric subspace.
- It utilizes an "active set" to recycle previously found vertices, reducing the number of expensive LMO calls.
- For the LMO, they use a heuristic (alternating minimization) for large $N$ and exact orbit enumeration for smaller $N$ (specifically $N \ge 8$ ), leveraging the symmetry to restrict the search to orbits of deterministic strategies.

Key Contributions

Algorithmic Acceleration: The paper demonstrates how symmetry reduction drastically accelerates the computation of Bell inequalities and local bounds. It allows the authors to solve instances ( $N=10, m=9$ ) that would otherwise require enumerating $\sim 10^{21}$ strategies and handling tensors with $\sim 10^9$ elements.
New Bell Inequalities: The method generates facets of the symmetrised local polytope. For $N=3$ to $10$, these inequalities give the best known upper bounds on the nonlocality robustness of the GHZ state.
Generalization for $m=4$ : The authors identify a pattern in the inequalities obtained for $m=4$ measurements. They generalize this to any number of parties $N$ , deriving closed-form expressions for the local bounds. These bounds form Lucas sequences and improve upon the asymptotic noise robustness given by Mermin's inequality.
Detection Efficiency Analysis: The paper computes the critical detection efficiency thresholds for these inequalities, showing improvements over previous results (e.g., for $N=4$ , the threshold improves from 0.7706 to 0.7544 as $m$ increases).

Results

Critical Visibilities: The authors provide numerical values for the critical visibility $v_{GHZ_N}^m$ for $N \in [3, 10]$ and various $m$ . For example, for $N=10$ and $m=9$ , the critical visibility is approximately $0.02301$.
Non-monotonicity: The results show that the critical visibility is non-monotonic with respect to $m$ for odd $N$ , attributed to the sub-optimality of using identical regular polygon measurements for all parties in asymmetric configurations.
Lower Bounds on Robustness: By constructing explicit local models (convex decompositions) via the Frank-Wolfe algorithm, the authors derive rigorous lower bounds on the nonlocality threshold for projective measurements in the $XY$ plane ( $v_{GHZ_N}^{XY}$ ). For $N=10$ , they establish a lower bound of approximately $0.01974$.
Activation of Nonlocality: Using the derived bounds, the authors demonstrate the activation of nonlocality in star networks for $N=10$ parties. They show that with a finite number of measurements ( $m=8$ or $9$), nonlocality can be activated in a network where the initial two-qubit states are local, a result previously only known to hold with an infinite number of measurements.

Significance and Claims
The paper positions itself as a proof of concept that symmetry can be effectively leveraged within Frank-Wolfe algorithms to solve high-dimensional constrained optimization problems in quantum information. The authors claim that their method allows them to go well beyond standard scenarios tractable by linear programming or previous Frank-Wolfe applications.

The significance lies in:

Providing the tightest known bounds on GHZ nonlocality robustness for up to 10 parties.
Demonstrating that finite measurement settings are sufficient to witness nonlocality activation in star networks, refining previous results that required infinite settings.
Offering a computational framework that reduces the complexity of Bell inequality generation from exponential to a manageable level for specific symmetric instances.

The authors remain modest regarding the generalizability of their specific inequality patterns, noting that while they found a pattern for $m=4$ , the general case for arbitrary $m$ remains an open question. They also note that the integer coefficients of the inequalities can become very large, which is a known algebraic property but contrasts with the small integers often found in literature. The work is presented as a step toward identifying other symmetric cases (such as bipartite tensors) that could further improve bounds on constants like the Grothendieck constant.

More like this