arxiv: 2604.22859 · v1 · submitted 2026-04-22 · 🪐 quant-ph

Recognition: unknown

Bell Inequalities from Polyhedral Sampling

Christian Staufenbiel

Authors on Pith no claims yet

Pith reviewed 2026-05-09 23:56 UTC · model grok-4.3

classification 🪐 quant-ph

keywords Bell inequalitiesBell polytopefacet samplingquantum nonlocalitypolyhedral computationdevice-independent protocols

0 comments

The pith

Sampling adjacent facets of the Bell polytope identifies over 129 million distinct inequality classes in scenarios where full enumeration remains impossible.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces the Adjacency Sampling method to locate classes of Bell inequalities by exploring neighboring facets from known starting points instead of enumerating the entire polytope. This technique recovers every previously catalogued class in cases that have already been solved completely. When applied to larger unsolved scenarios it produces lists far longer than earlier partial results, including more than 1.29 times 10 to the 8 classes in L sub 3,3,3,3. Readers care because Bell inequalities certify the presence of quantum correlations that cannot be explained by classical models and are required for device-independent protocols. If the sampled classes are representative, the approach makes previously inaccessible scenarios available for study.

Core claim

The Adjacency Sampling method builds on the Adjacency Decomposition approach by deliberately sampling adjacent facets of the Bell polytope rather than continuing the enumeration to completion. On every Bell polytope that has already been fully solved the method recovers every known inequality class. In unsolved cases it produces substantially larger collections than any earlier partial enumeration, reporting over 1.29 times 10 to the 8 classes for L sub 3,3,3,3, 49,358 classes for L sub 4,5,2,2, and more than 4.3 million classes for L sub 4,6,2,2.

What carries the argument

The Adjacency Sampling method, a procedure that starts from known inequalities and generates new ones by examining adjacent facets of the Bell polytope without requiring exhaustive search.

If this is right

The method recovers every known class on all previously solved Bell polytopes.
In the L sub 3,3,3,3 scenario it finds over 1.29 times 10 to the 8 classes, more than 25 times the prior partial count.
In the L sub 4,5,2,2 scenario it yields 49,358 classes, nearly tripling the known list.
In the L sub 4,6,2,2 scenario it reports over 4.3 million classes.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The large numbers of classes found imply that the structure of Bell polytopes is considerably richer than what small fully enumerated cases had suggested.
The additional inequalities could be tested for their ability to tighten bounds on quantum correlations or to strengthen device-independent key-distribution protocols.
Guiding the sampling with information about which facets produce inequalities with particular properties might increase the proportion of useful classes obtained.
Statistical comparison of the sampled classes against known distributions from small polytopes could reveal whether the method captures the overall geometry of the polytope.

Load-bearing premise

The sampling of adjacent facets must produce a representative collection of inequality classes rather than systematically missing entire families or over-representing others.

What would settle it

A concrete falsifier would be the discovery of one specific, previously known Bell inequality class in any of the studied scenarios that repeated runs of the sampling method consistently fail to produce.

read the original abstract

Bell inequalities play a central role in certifying quantum correlations and underpin protocols such as device-independent quantum key distribution. However, enumerating all Bell inequalities for a given scenario remains intractable beyond the simplest cases, as it requires solving a computationally hard facet enumeration problem on the associated Bell polytope. We propose the Adjacency Sampling method, which builds on the Adjacency Decomposition method but sacrifices completeness for speed. On previously solved Bell polytopes, the method reproduces every known class of inequalities. For scenarios where no complete enumeration exists, it greatly exceeds existing partial results: in $\mathcal{L}_{3,3,3,3}$ we obtain over $1.29 \times 10^8$ classes, more than 25 times the previous count; in $\mathcal{L}_{4,5,2,2}$ we nearly triple the known list to 49\,358 classes; and for $\mathcal{L}_{4,6,2,2}$ we report over 4.3 million classes.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Adjacency sampling produces much larger partial lists of Bell inequality classes than before and matches all known results on solved polytopes, but the new counts lack checks against sampling bias.

read the letter

This paper gives a practical sampling variant of adjacency decomposition for Bell polytopes. It reproduces every known inequality class on the small cases that have been fully solved, then reports substantially bigger partial enumerations for three larger scenarios: over 129 million classes in L_{3,3,3,3}, nearly 50 thousand in L_{4,5,2,2}, and 4.3 million in L_{4,6,2,2}. Those numbers are the concrete advance and exceed prior partial results by large factors. The method deliberately trades completeness for speed by walking the adjacency graph instead of enumerating everything, which lets it reach these sizes in reasonable time. The validation on known polytopes is the strongest part of the work; matching all existing classes shows the code is at least consistent with prior exhaustive runs. For device-independent certification tasks that need more inequalities than the old lists provide, this is usable output. The main limitation is the sampling step itself. The paper does not include orbit-stabilizer analysis, facet-type distribution checks against exhaustive baselines, or any other test that would show whether the walks miss entire families or over-represent others. Without that, the reported class counts for the unsolved cases could be skewed, even though the method is sound on the small instances. This is for quantum information researchers who work with Bell polytopes and need larger partial lists for practical certification rather than a complete classification. The algorithmic nature and the reproduction of known results make the contribution clear enough to deserve referee time.

Referee Report

2 major / 2 minor

Summary. The paper introduces the Adjacency Sampling method, which extends the Adjacency Decomposition approach by performing random or guided walks over adjacent facets of the Bell polytope to enumerate inequivalent Bell inequality classes. It exactly reproduces every known class on all previously solved polytopes and, for unsolved scenarios, reports substantially larger partial enumerations: over 1.29 × 10^8 classes in L_{3,3,3,3} (more than 25 times prior counts), 49,358 classes in L_{4,5,2,2} (nearly triple prior), and over 4.3 million classes in L_{4,6,2,2}.

Significance. If the sampled classes prove representative, the work offers a practical advance for exploring Bell polytopes beyond exhaustive enumeration limits, directly aiding certification of quantum correlations in higher-dimensional scenarios. The exact reproduction of all known classes on solved instances, combined with the parameter-free algorithmic procedure, constitutes a clear methodological strength and provides a reproducible baseline for future comparisons.

major comments (2)

[§3] §3 (Adjacency Sampling procedure): the method description provides no orbit-stabilizer analysis, no facet-type distribution comparison against exhaustive baselines on solved polytopes, and no convergence diagnostics across independent runs; without these, the headline counts for unsolved polytopes (e.g., 1.29 × 10^8 in L_{3,3,3,3}) cannot be shown to be unbiased samples rather than artifacts of preferential traversal of certain cones in the adjacency graph.
[Results section] Results for unsolved scenarios (Table reporting L_{4,5,2,2} and L_{4,6,2,2} counts): the claim that the new lists 'greatly exceed' prior partial results rests on the untested assumption that adjacency sampling yields a representative collection of orbits; the reproduction of known classes on solved cases does not address whether the same procedure systematically misses or over-represents families when the polytope is larger and the graph is only partially explored.

minor comments (2)

[Abstract] Abstract and §1: the scenario notation L_{m,n,...} is used without an explicit reference to the standard (k,m,n,...) Bell scenario definition, which may confuse readers unfamiliar with the precise indexing.
[Figures] Figure captions (e.g., those illustrating sampled inequalities): axis labels and facet identifiers are occasionally abbreviated without a legend, reducing immediate readability.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the thorough review and positive assessment of the methodological strengths of Adjacency Sampling. We address the major comments below, with revisions planned where appropriate to clarify the scope and limitations of our results.

read point-by-point responses

Referee: [§3] §3 (Adjacency Sampling procedure): the method description provides no orbit-stabilizer analysis, no facet-type distribution comparison against exhaustive baselines on solved polytopes, and no convergence diagnostics across independent runs; without these, the headline counts for unsolved polytopes (e.g., 1.29 × 10^8 in L_{3,3,3,3}) cannot be shown to be unbiased samples rather than artifacts of preferential traversal of certain cones in the adjacency graph.

Authors: We acknowledge the absence of these analyses in the current manuscript. The exact reproduction of all known inequality classes on solved polytopes provides strong validation for the procedure in those instances. For unsolved cases, we do not assert that the samples are unbiased or representative, as the adjacency graph exploration is partial. We will revise §3 to explicitly discuss these limitations, include any available facet-type comparisons from solved cases, and note that the reported numbers are the result of extensive sampling runs rather than claims of completeness or unbiasedness. Full orbit-stabilizer analysis is beyond the scope of this work but could be pursued in future research. revision: partial
Referee: [Results section] Results for unsolved scenarios (Table reporting L_{4,5,2,2} and L_{4,6,2,2} counts): the claim that the new lists 'greatly exceed' prior partial results rests on the untested assumption that adjacency sampling yields a representative collection of orbits; the reproduction of known classes on solved cases does not address whether the same procedure systematically misses or over-represents families when the polytope is larger and the graph is only partially explored.

Authors: The referee correctly identifies that validation on solved polytopes does not automatically extend to larger ones. We will revise the results section to remove any implication of representativeness and instead emphasize that the method yields substantially larger partial enumerations than prior approaches, providing a practical tool for exploring high-dimensional Bell polytopes. The abstract will be updated accordingly to reflect this. revision: yes

standing simulated objections not resolved

Comprehensive orbit-stabilizer analysis and convergence diagnostics for the adjacency graph in large polytopes such as L_{3,3,3,3}, as performing these would require resources significantly beyond those used for the sampling itself.

Circularity Check

0 steps flagged

No circularity: algorithmic sampling procedure validated by reproduction on known cases

full rationale

The paper introduces Adjacency Sampling as a direct computational procedure extending the Adjacency Decomposition method to traverse the facet graph of the Bell polytope. It reports reproduction of all previously known inequality classes on solved polytopes (L_{2,2,2,2}, etc.) as empirical validation, then applies the same traversal to larger unsolved instances to obtain new counts. No equations define a quantity in terms of itself, no parameters are fitted to a subset and then called a prediction, and no load-bearing premise reduces to a self-citation chain or imported uniqueness theorem. The reported class counts are explicit outputs of the sampling algorithm rather than derived quantities that collapse to the input data by construction. The method is therefore self-contained as a heuristic enumeration tool whose correctness on new instances rests on the (unproven but stated) assumption of representative sampling, not on any internal circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The work rests on standard convex geometry and quantum correlation theory; no new free parameters or invented entities are introduced.

axioms (1)

domain assumption The set of classical correlations for a given Bell scenario forms a convex polytope whose facets are Bell inequalities.
Standard assumption in the Bell inequality literature.

pith-pipeline@v0.9.0 · 5457 in / 1162 out tokens · 37608 ms · 2026-05-09T23:56:34.922227+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

26 extracted references · 20 canonical work pages

[1]

A. Acín, N. Brunner, N. Gisin, S. Massar, S. Pironio, and V . Scarani. Device-independent security of quantum cryp- tography against collective attacks.Phys. Rev. Lett., 98: 230501, 2007. doi:10.1103/PhysRevLett.98.230501

work page doi:10.1103/physrevlett.98.230501 2007
[2]

Williams

D. Avis and K. Fukuda. Reverse search for enumeration. Discrete Appl. Math., 65:21–46, 1996. doi:10.1016/0166- 218X(95)00026-N

work page doi:10.1016/0166- 1996
[3]

D. Avis, D. Bremner, and R. Seidel. How good are con- vex hull algorithms?Comput. Geom., 7:265–301, 1997. doi:10.1016/S0925-7721(96)00023-5

work page doi:10.1016/s0925-7721(96)00023-5 1997
[4]

J. S. Bell. On the einstein podolsky rosen para- dox.Physics Physique Fizika, 1:195–200, 1964. doi:10.1103/PhysicsPhysiqueFizika.1.195

work page doi:10.1103/physicsphysiquefizika.1.195 1964
[5]

Brunner, D

N. Brunner, D. Cavalcanti, S. Pironio, V . Scarani, and S. Wehner. Bell nonlocality.Rev. Mod. Phys., 86:419–478,
[6]

doi:10.1103/RevModPhys.86.419

work page doi:10.1103/revmodphys.86.419
[7]

Christof and G

T. Christof and G. Reinelt. Decomposition and paralleliza- tion technique for enumerating the facets of combinatorial polytopes.Int. J. Comput. Geom. Appl., 11:423–437, 2001. doi:10.1142/S0218195901000560

work page doi:10.1142/s0218195901000560 2001
[8]

Cope and R

T. Cope and R. Colbeck. Bell inequalities from no- signaling distributions.Phys. Rev. A, 100:022114, 2019. doi:10.1103/PhysRevA.100.022114

work page doi:10.1103/physreva.100.022114 2019
[9]

Deza and M

M. Deza and M. Dutour Sikiri ´c. Enumeration of the facets of cut polytopes over some highly symmet- ric graphs.Int. Trans. Oper . Res., 23:853–860, 2015. doi:10.1111/itor.12194

work page doi:10.1111/itor.12194 2015
[10]

Deza and M

M. Deza and M. Laurent.Geometry of Cuts and Metrics. Springer, 1997. doi:10.1007/978-3-642-04295-9

work page doi:10.1007/978-3-642-04295-9 1997
[11]

Dutour Sikiri´c

M. Dutour Sikiri´c. permutalib. https://github.com/ MathieuDutSik/permutalib. Accessed: 2026-02-20

2026
[12]

Einstein , author B

A. Einstein, B. Podolsky, and N. Rosen. Can quantum- mechanical description of physical reality be consid- ered complete?Phys. Rev., 47:777–780, 1935. doi:10.1103/PhysRev.47.777

work page doi:10.1103/physrev.47.777 1935
[13]

Fukuda and A

K. Fukuda and A. Prodon. Double description method revisited. InCombinatorics and Computer Science, pages 91–111. Springer, 1996. doi:10.1007/3-540-61576-8_77

work page doi:10.1007/3-540-61576-8_77 1996
[14]

GAP – Groups, Algorithms, and Programming, version 4.11.1.https://www.gap-system.org, 2021

GAP. GAP – Groups, Algorithms, and Programming, version 4.11.1.https://www.gap-system.org, 2021

2021
[15]

Convex polytopes , SERIES =

B. Grünbaum.Convex Polytopes. Springer, 2003. doi:10.1007/978-1-4613-0019-9

work page doi:10.1007/978-1-4613-0019-9 2003
[16]

URLhttp://dx.doi.org/10.1038/nature15759

B. Hensen, H. Bernien, A. E. Dréau, A. Reiserer, N. Kalb, M. S. Blok, J. Ruitenberg, R. F. L. Vermeulen, R. N. Schouten, C. Abellán, W. Amaya, V . Pruneri, M. W. Mitchell, M. Markham, D. J. Twitchen, D. Elkouss, Preprint– BellInequalities fromPolyhedralSampling5 S. Wehner, T. H. Taminiau, and R. Hanson. Loophole- free bell inequality violation using elect...

work page doi:10.1038/nature15759 2015
[17]

Jesus and E

J. Jesus and E. Zambrini Cruzeiro. Tight bell inequalities from polytope slices.Phys. Rev. A, 108:052220, 2023. doi:10.1103/PhysRevA.108.052220

work page doi:10.1103/physreva.108.052220 2023
[18]

M. Joswig. Beneath-and-beyond revisited, 2002

2002
[19]

J. S. Leon. Permutation group algorithms based on par- titions, I: Theory and algorithms.J. Symb. Comput., 12: 533–583, 1991. doi:10.1016/S0747-7171(08)80103-4

work page doi:10.1016/s0747-7171(08)80103-4 1991
[20]

Y . Liu, H. Y . Chung, E. Z. Cruzeiro, J. R. Gonzales- Ureta, R. Ramanathan, and A. Cabello. Equiva- lence between face nonsignaling correlations, full nonlocality, all-versus-nothing proofs, and pseu- dotelepathy.Phys. Rev. Res., 6:L042035, 2024. doi:10.1103/PhysRevResearch.6.L042035

work page doi:10.1103/physrevresearch.6.l042035 2024
[21]

Lörwald and G

S. Lörwald and G. Reinelt. PANDA: a software for polyhe- dral transformations.EURO J. Comput. Optim., 3:297–307,
[22]

doi:10.1007/s13675-015-0040-0

work page doi:10.1007/s13675-015-0040-0
[23]

Schwarz, B

S. Schwarz, B. Bessire, A. Stefanov, and Y .-C. Liang. Bipartite bell inequalities with three ternary-outcome measurements—from theory to experiments.New J. Phys., 18:035001, 2016. doi:10.1088/1367-2630/18/3/035001

work page doi:10.1088/1367-2630/18/3/035001 2016
[24]

Staufenbiel

C. Staufenbiel. PANDA fork. https://github.com/ christian512/panda. Accessed: 2026-02-20

2026
[25]

Zapatero, T

V . Zapatero, T. van Leent, R. Arnon-Friedman, W. Liu, Q. Zhang, H. Weinfurter, and M. Curty. Advances in device-independent quantum key distribution.npj Quan- tum Inf., 9:10, 2023. doi:10.1038/s41534-023-00684-x

work page doi:10.1038/s41534-023-00684-x 2023
[26]

G. M. Ziegler.Lectures on Polytopes. Springer, 1995. doi:10.1007/978-1-4613-8431-1

work page doi:10.1007/978-1-4613-8431-1 1995