Compact Bell Inequalities

• Compact Bell inequalities are a class of nonlocality tests that reduce experimental and descriptive overhead while maintaining a clear quantum-classical separation.
• They employ minimal exclusivity graphs, sparse multipartite correlators, and canonical reduction to simplify traditional Bell inequality structures.
• Their design facilitates efficient self-testing, entanglement-depth certification, and practical implementations in loophole-free quantum experiments.

Searching arXiv for the specified papers to ground the article in current bibliographic records. Compact Bell inequalities are Bell inequalities designed to minimize descriptive or experimental overhead—typically the number of probabilities, correlators, settings, or nonzero coefficients—while preserving a nontrivial separation between local hidden-variable models and quantum theory. In the literature, compactness appears in several technically distinct but related forms: minimal exclusivity structures in bipartite scenarios, sparse CHSH-type or Mermin-type multipartite correlator inequalities, canonical reduced representatives modulo relabelings and liftings, and state-tailored inequalities obtained from sums-of-squares constructions (Sadiq et al., 2011, He et al., 2014, Wu et al., 2013, Rosset et al., 2014, Barizien et al., 2023).

1. Compactness as a structural property of Bell inequalities

Compactness is not a single formal criterion. In the bipartite exclusivity-graph approach, it refers to using the smallest graph that already exhibits a quantum-classical gap: the pentagon $C_5$, yielding inequalities with only five probabilities (Sadiq et al., 2011). In multipartite CHSH-type constructions, compactness refers to reducing the support from the $2^n$ terms of MABK- or WWZB-type families to roughly $2^{n/2}$ terms, or even to a constant four-term form in a specific eight-party construction (He et al., 2014, Wu et al., 2013). In polytope-based classification, compactness also denotes choosing one reduced representative per equivalence class after removing normalization, no-signaling, relabeling, lifting, and composite degeneracies (Rosset et al., 2014, Cope et al., 2018).

These usages are related by a common objective: reducing redundancy without sacrificing the operational role of the inequality. A compact inequality may remain facet-defining, may be tailored to a target state, or may be preferable experimentally because fewer correlations must be estimated. This suggests that compactness is best viewed as a constrained optimization over Bell descriptions rather than as a property tied to a single formalism.

Family	Scenario	Compactness feature
Pentagonal $C_5$ inequalities	Bipartite	5 probabilities
Paired-CHSH Mermin-type operators $B_n$	$n$-qubit multipartite	$O(2^{n/2})$ terms
Eight-party CHSH-type inequality	8 parties, 2 settings, 2 outcomes	4 correlators
Canonical representatives	General Bell scenarios	Degeneracy-free reduced form

2. Minimal bipartite compact inequalities and the pentagon $C_5$

In the graph-theoretic approach of Cabello-Severini-Winter, each event $ab|xy$ is represented by a vertex, and edges encode exclusivity. Among graphs whose classical independence number is strictly below the quantum Lovász-theta bound, the 5-cycle $C_5$ is the smallest. An exhaustive combinatorial argument shows that there are exactly three nonequivalent ways to label the five vertices of $2^n$0 with bipartite Bell events so that edges match bipartite exclusivity, yielding three pentagonal Bell inequalities with local bound $2^n$1 and quantum violation (Sadiq et al., 2011).

The three inequalities are

$2^n$2

with the explicit representatives

$2^n$3

$2^n$4

$2^n$5

Their quantum maxima are $2^n$6 and $2^n$7. The second inequality is algebraically equivalent to CHSH through

$2^n$8

while the third appears as a subcomponent of $2^n$9. The pentagonal structure is therefore not merely minimal in graph size; it is also a logical kernel underlying more familiar Bell inequalities.

The experimental realization used polarization-entangled photons produced via type-II SPDC in BBO, with measured values $2^{n/2}$0, $2^{n/2}$1, and $2^{n/2}$2, all violating the local bound by more than $2^{n/2}$3 standard deviations. The reported tests were free of the compatibility loophole. In this setting, compactness means that a genuine quantum-classical gap already appears with only five events arranged in a pentagon, which is the minimum identified in that framework.

3. Sparse multipartite constructions from CHSH and Mermin-type architectures

A major multipartite use of compact Bell inequalities is to reduce the exponential growth of correlator support. He, Ding, Yan, and Gao construct $2^{n/2}$4-qubit Bell operators by pairing CHSH partitions. For parties $2^{n/2}$5, define

$2^{n/2}$6

Then, for even $2^{n/2}$7,

$2^{n/2}$8

and for odd $2^{n/2}$9,

$C_5$0

Any LHVM satisfies $C_5$1. The number of distinct correlation terms is $C_5$2 for odd $C_5$3 and $C_5$4 for even $C_5$5, replacing the $C_5$6 terms of MABK and WWZB by roughly $C_5$7 (He et al., 2014).

On pure $C_5$8-qubit GHZ states, the maximal quantum values depend on $C_5$9: $B_n$0 for $B_n$1, and $B_n$2 for the other congruence classes under the measurement choices listed in the construction. For generalized GHZ states

$B_n$3

the same settings give $B_n$4 for $B_n$5, and $B_n$6 otherwise. Hence violation requires $B_n$7 in the first case and $B_n$8 in the others. For $B_n$9, with

$n$0

one obtains $n$1, four-tangle $n$2, and therefore

$n$3

Violation of the compact inequality is equivalent to $n$4, whereas the generalized Svetlichny inequality requires $n$5. In that specific sense, the compact $n$6 is more sensitive to four-qubit GGHZ entanglement.

A distinct sparse multipartite route is given by the correlation-polytope construction of Wu et al. Starting from CHSH and adding parties inductively via

$n$7

one can choose parent facets so that the number of terms remains four at every step. For eight parties this yields the tight CHSH-type inequality

$n$8

The comparison with standard 8-party constructions is explicit: the full universal inequality involves up to $n$9 correlation functions, whereas the compact inequality uses $O(2^{n/2})$0. Using the estimate from the eight-photon GHZ experiment of Huang et al., $O(2^{n/2})$1 terms at $O(2^{n/2})$2 hours each corresponds to $O(2^{n/2})$3 years, while $O(2^{n/2})$4 terms correspond to $O(2^{n/2})$5 days; the four-term inequality also has critical visibility $O(2^{n/2})$6, compared to $O(2^{n/2})$7 for the 9-term witness used there (Wu et al., 2013).

4. Canonical reduction, facet enumeration, and systematic generation

Compactness is also a classification problem. Rosset, Bancal, and Gisin analyze the main degeneracies in Bell expressions: normalization and no-signaling redundancies, relabelings of parties, settings, and outcomes, liftings by irrelevant parties or settings, and composite tensor-product structure. They project the coefficient tensor onto the normalized no-signaling dual space, fix the scale by making the gcd of all entries equal to $O(2^{n/2})$8, and define a canonical representative

$O(2^{n/2})$9

under lexicographic order, together with an index $C_5$0 inside the relabeling orbit (Rosset et al., 2014). This provides a degeneracy-free reference form for comparing inequalities and underlies the faacets.com compendium.

The computational significance is that compact representatives need not be found by ad hoc simplification. They can emerge from exact symmetry reduction and orbit minimization. The same perspective appears in facet-enumeration methods that operate under constraints. Cope and Colbeck exploit extremal no-signalling vertices and the dual of the local-weight linear program to derive compact facet inequalities in low-dimensional scenarios. In the $C_5$1 scenario they find all $C_5$2 Bell inequality classes, and in the $C_5$3 scenario they find $C_5$4 classes, keeping one compact representative per equivalence class after removing relabelings and liftings (Cope et al., 2018). One representative new facet in $C_5$5, $C_5$6, has only $C_5$7 nonzero $C_5$8 blocks, each with two ones; one representative new facet in $C_5$9, $ab|xy$0, has $ab|xy$1 nonzero blocks.

A further systematic route is to impose structural constraints before facet enumeration. In the cone-projection method, the local polytope is embedded as a slice of a cone, the facet normal is constrained by homogeneous linear equations, the cone is projected to the corresponding kernel, and facets are enumerated in the reduced space before being lifted back. Applied to fully symmetric three-party generalizations of $ab|xy$2, this yields $ab|xy$3 inequivalent inequalities. The three simplest,

$ab|xy$4

contain only $ab|xy$5 or $ab|xy$6 distinct symmetric correlators plus a small number of marginals and constants (Bernards et al., 2020). In this literature, compactness is therefore inseparable from symmetry reduction and constrained optimization over facet normals.

5. State-tailored compact inequalities and device-independent uses

Compact Bell inequalities need not be generic. The formal sums-of-squares framework constructs Bell functionals adapted to a target pure state by selecting nullifiers $ab|xy$7 satisfying $ab|xy$8 and enforcing

$ab|xy$9

If the expansion contains no monomials of local degree $C_5$0, then $C_5$1 is a bona fide Bell polynomial of local degree $C_5$2, $C_5$3 for every quantum realization, and saturation requires all nullifier conditions (Barizien et al., 2023). This gives the Tsirelson bound and the self-testing statement simultaneously.

For partially entangled multipartite GHZ states

$C_5$4

the method uses nullifiers

$C_5$5

together with a one-parameter measurement family on party $C_5$6. The resulting Bell operator $C_5$7 satisfies

$C_5$8

and equality forces exactly the target GHZ state and the specified measurements. The same framework constructs compact self-tests for the maximally entangled two-qutrit state and for partially entangled two-qubit states with a two-parameter family of settings. It also shows that the same quantum point can be exposed by infinitely many distinct Bell inequalities, while certain limit points lie on flat segments of the quantum set and cannot be exposed by a single linear functional. This is a structural statement about the geometry of the quantum set, derived from compact state-specific inequalities rather than from large universal families.

Compact inequalities also play a direct role in detecting multipartite nonlocality depth. Bernards and Gühne derive a family

$C_5$9

which detects nonlocality depth at least $2^n$00 in an $2^n$01-party system (Bernards et al., 2022). For $2^n$02,

$2^n$03

and for $2^n$04,

$2^n$05

The same work gives complete facet lists, under symmetry and full-body-correlator constraints, for four- and five-party hybrid models, with several inequalities using only $2^n$06 to $2^n$07 correlator terms plus a constant. Compactness here is operational: it reduces the number of full-body correlators required to witness entanglement depth in a device-independent manner.

6. Experimental significance, limitations, and recurrent misconceptions

The main experimental attraction of compact Bell inequalities is the reduction in measurement overhead. This is explicit in the eight-party four-term construction and in the paired-CHSH family with only two settings per qubit and $2^n$08 correlators (Wu et al., 2013, He et al., 2014). The pentagonal inequalities likewise reduce the test to five probabilities and were implemented in a photonic setup with results close to the ideal quantum values (Sadiq et al., 2011). Compactness therefore often means that a Bell test becomes experimentally realistic in regimes where full-support inequalities are prohibitive.

A recurrent misconception is that compactness automatically implies stronger nonlocality detection. The available results are more nuanced. For generalized $2^n$09-qubit GHZ states, the compact Mermin-type inequalities have violation thresholds $2^n$10 or $2^n$11, whereas the full MABK family has threshold $2^n$12 (He et al., 2014). This shows that an inequality may be experimentally simpler while being less sensitive in that state family. Conversely, in the four-qubit GGHZ case the compact $2^n$13 detects entanglement already at $2^n$14, while the generalized Svetlichny inequality needs $2^n$15. Compactness is therefore not monotonic with respect to detection strength; it trades against the specific geometry of the tested correlations.

Another misconception is that a sparse form is necessarily primitive. Classification results show that many inequalities are liftings or composites of lower-dimensional kernels, and canonical reduction is required to distinguish genuinely new families from rewritten old ones (Rosset et al., 2014, Cope et al., 2018). In $2^n$16, for example, apart from positivity facets only CHSH, $2^n$17, and one genuinely new $2^n$18 family appear; the rest are liftings. Compactness must therefore be interpreted together with equivalence under relabeling and lifting.

A further point is that compactness can be compatible with facet-tightness and with advanced device-independent tasks. The SOS constructions prove self-testing for several compact inequalities, and constrained facet-enumeration methods deliver compact inequalities that remain facet-defining (Barizien et al., 2023, Bernards et al., 2020). This suggests that sparse support does not merely reflect heuristic simplification; it can coincide with extremality in the relevant local polytope or with exact characterization of a target quantum realization.

Compact Bell inequalities thus occupy a broad methodological space: they may arise from minimal exclusivity graphs, inductive CHSH constructions, symmetry-constrained cone projections, no-signalling-guided facet searches, or SOS-based state engineering. Across these approaches, the unifying principle is the removal of superfluous structure while retaining a rigorous local bound, a quantum-classical gap, and, in many cases, direct utility for multipartite nonlocality, self-testing, entanglement-depth certification, or loophole-resistant experiments.