Symmetric CGLMP Inequalities in Quantum Systems

• Symmetric CGLMP inequalities are Bell inequalities for (2,2,d) scenarios that enforce party-exchange invariance and set canonical constraints for quantum nonlocality.
• They reveal a unique facet in the symmetric subspace of the Bell polytope and are crucial for device-independent tests and experimental validations.
• Optimal quantum violations using symmetric strategies highlight the interplay of non-maximal entanglement, Hilbert space dimension, and practical implementation.

Symmetric Collins–Gisin–Linden–Massar–Popescu (CGLMP) inequalities are a distinguished family of Bell inequalities central to the study of quantum nonlocality beyond the qubit (d>2) regime. They provide the canonical constraints for bipartite scenarios with two measurement settings per party, each with d outcomes, and are characterized by explicit invariance under the exchange of parties (party‐permutation invariance, PPI). Their symmetric forms make them ideal for both theoretical investigations of the quantum boundary and practical implementation of device-independent certification tasks.

1. Definition and Symmetric Formulation

The symmetric CGLMP inequality is defined for the (2,2,d) scenario: two parties (Alice and Bob), each with two measurement settings $x,y \in \{0,1\}$, and d possible outcomes per measurement ($a,b \in \{0,\dots,d-1\}$). Symmetry under party exchange is enforced at the level of both the Bell functional and the allowed correlations: $$P(a,b\mid x,y) = P(b,a\mid y,x) \quad \forall a,b,x,y.$$ A widely used symmetric form of the CGLMP inequality is: $$I_{22dd} = \sum_{(a,b): a+b\leq d-2} P(a,b\mid 0,0) -\sum_{(a,b): a+b\geq d-2} P(a,b\mid 1,1) +\sum_{(a,b): a+b\geq d-2}\left[P(a,b\mid 0,1) + P(a,b\mid 1,0)\right] -\sum_{a=0}^{d-2} P(a\mid 0) -\sum_{b=0}^{d-2} P(b\mid 0) \leq 0,$$ where marginal probabilities are defined (using no-signaling) as $P(a\mid x) = \sum_b P(a,b\mid x,y)$, independent of y. This form is manifestly party-symmetric at both coefficient and functional levels, and outcome relabellings combined with no-signaling complete the symmetrization (Hsu et al., 6 Jan 2026).

The standard affine rescaling relates $I_{22dd}$ to the historical CGLMP value $I_d$: $I_d = \frac{2d}{d-1}I_{22dd} + 2$ with local-realist bound $I_d \leq 2$.

2. Polytope Structure and Facet Uniqueness

Projecting the full (2,2,d) Bell polytope onto its party-symmetric subspace yields a polytope whose facets correspond to all party-invariant Bell inequalities. For general d, the symmetric CGLMP inequality forms the unique non-trivial facet of this subpolytope, with the rest being trivial positivity constraints. Explicitly, using the symmetrizing map $\Pi = \tfrac12(e + \pi)$, all local probabilistic assignments map into d² symmetrized vertices. The facet structure is then revealed by standard polytope theory; the CGLMP inequality stands as the only essential symmetric constraint for (2,2,d) (Bancal et al., 2010).

The explicit PPI CGLMP inequality in an alternative compact form is: $a,b \in \{0,\dots,d-1\}$0 where

$a,b \in \{0,\dots,d-1\}$1

For d = 2, this reduces to the CHSH inequality; for d = 3, to the original three-outcome CGLMP form; for d > 3, new symmetric facets may appear but CGLMP remains a key facet (Bancal et al., 2010).

3. Quantum Violations and Optimal Strategies

Quantum nonlocality is demonstrated by violations of the symmetric CGLMP inequalities. For (2,2,d), the maximal quantum value

$a,b \in \{0,\dots,d-1\}$2

is achieved using a symmetric quantum strategy (SQS): the same dimension and measurement structure on both parties. The optimal state is the eigenvector corresponding to the largest eigenvalue of the symmetric Bell operator, with measurements constructed using specific unitaries derived from mutually unbiased bases or Fourier transforms, depending on d (Hsu et al., 6 Jan 2026).

Numerically, for d ranging from 2 to 7, the maximal values are:

$a,b \in \{0,\dots,d-1\}$3	$a,b \in \{0,\dots,d-1\}$4
2	$a,b \in \{0,\dots,d-1\}$5
3	2.914854
4	2.972698
5	3.015710
6	3.049700
7	3.077648

(Hsu et al., 6 Jan 2026). These values coincide with those obtained using asymmetric strategies, indicating no enhancement of maximal violation through either symmetry or explicit asymmetry. For d ≤ 5, analytic results are available; for d > 5, results confirm that the maximal value converges just below 3 as d increases (Bancal et al., 2010, Hsu et al., 6 Jan 2026).

Notably, maximal quantum violation does not always require maximal entanglement. For d ≥ 3, the state maximizing the violation can be non-maximally entangled (Markiewicz et al., 2016). In particular, for d = 3: $a,b \in \{0,\dots,d-1\}$6 attains the Tsirelson bound 2.915 (Markiewicz et al., 2016).

4. Symmetric Qutrit–to–Qubit Mappings and Insight from Operator Representation

An essential insight is gained by representing qutrit (spin-1) systems in terms of symmetric subspaces of higher-qubit systems. Any qutrit operator decomposes into a hermitian basis involving spin-1 matrices and is mapped one-to-one onto the three-dimensional symmetric subspace of two qubits. For instance: $a,b \in \{0,\dots,d-1\}$7 where $a,b \in \{0,\dots,d-1\}$8 (Markiewicz et al., 2016).

The two-qutrit CGLMP Bell operator, when written in symmetric two-qubit language, decomposes into a weighted combination of four CHSH operators and a four-qubit Mermin operator: $a,b \in \{0,\dots,d-1\}$9 revealing that the optimal state for CGLMP maximal violation is not maximally entangled from the qutrit perspective because it must negotiate the trade-off between these competing sub-correlations (Markiewicz et al., 2016).

For the three-qutrit case, the corresponding Bell operator maps to a six-qubit operator, whose unique optimal vector is a superposition of a generalized GHZ state and three Bell pairs, displaying similar non-maximal entanglement structure (Markiewicz et al., 2016).

5. Symmetry, Dimension, and Resource Requirements

A central question addressed in recent work is whether maximally violating symmetric (PPI) Bell inequalities demands higher local Hilbert space dimension, i.e., if symmetry is a costly resource. For the family of symmetric CGLMP inequalities, it has been shown numerically that for all d up to at least 19, symmetric quantum strategies in dimension d already saturate the quantum bound: $$P(a,b\mid x,y) = P(b,a\mid y,x) \quad \forall a,b,x,y.$$0 (Hsu et al., 6 Jan 2026). Thus, there is no trade-off: enforcing party-exchange invariance does not raise the minimal Hilbert space dimension for maximal violation.

For other Bell inequalities, however, there exist facets where maximal violation in minimal dimension can only be achieved with asymmetric strategies. In such cases, symmetry in minimal dimension leads to suboptimal violation and “flat” faces of the quantum correlation set, precluding self-testing (Hsu et al., 6 Jan 2026). This property distinguishes the CGLMP family as especially suited for device-independent tasks requiring sharp boundary points.

6. Experimental Realization and Device-Independent Certification

The symmetric nature of CGLMP inequalities enables transparent and loophole-free experimental Bell tests with authentic multi-outcome projective measurements. In recent experimental work, e.g., for d=4 (ququart) systems, the apparatus routes each possible outcome to physically distinct detectors, closing the binarisation loophole that afflicts high-dimensional tests when outcomes are emulated through sequential binary (click/no-click) responses. Exceeding dimension-witness SDP bounds using the symmetric CGLMP inequality certifies the presence of entanglement in the full d-dimensional local spaces (Miao et al., 5 Jan 2026).

Symmetric CGLMP inequalities also serve as the primary tool for device-independent certification of high-dimensional entanglement, self-testing, and nonlocality, with maximal quantum violations providing direct witnesses for the underlying nonclassical state and measurements.

7. Generalizations and Related Facets

While the CGLMP inequality is the unique non-trivial symmetric facet for (2,2,d) up to d=3, for d > 3, additional symmetric facets may appear. For d=4, seven new symmetric inequalities have been identified beyond the standard CGLMP facet (Bancal et al., 2010). Their relevance for nonclassicality, resource requirements, and boundary geometry warrants further investigation, but the CGLMP family retains its central role as the canonical symmetric facet throughout dimensions.

The method of decomposing CGLMP-type Bell operators into combinations of lower-dimensional operators (such as CHSH and Mermin terms) via symmetric mappings reveals the structure of quantum nonlocality in higher dimensions and its relation to entanglement monogamy and complementarity constraints (Markiewicz et al., 2016).

---

References:

• (Markiewicz et al., 2016): Investigating nonclassicality of many qutrits by symmetric two-qubit operators
• (Bancal et al., 2010): Looking for symmetric Bell inequalities
• (Hsu et al., 6 Jan 2026): Trading symmetry for Hilbert-space dimension in Bell-inequality violation
• (Miao et al., 5 Jan 2026): Binarisation-loophole-free observation of high-dimensional quantum nonlocality