No finite level of the NPA hierarchy is exact for the doubly-tilted CHSH functional near the critical tilt

Anton Pakhunov

REVIEW 1 major objections 6 minor 1 cited by

No finite level of the NPA hierarchy is exact for the doubly-tilted CHSH functional near its critical tilt; every finite level overshoots the quantum value on a small interval.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 03:46 UTC pith:ZTRKLHYK

load-bearing objection A serious, carefully built proof that no finite NPA level is exact near the critical tilt; the central argument looks sound, but the load-bearing finite enumeration deserves independent scrutiny before full acceptance. the 1 major comments →

arxiv 2607.13762 v1 pith:ZTRKLHYK submitted 2026-07-15 quant-ph math.OC

No finite level of the NPA hierarchy is exact for the doubly-tilted CHSH functional near the critical tilt

Anton Pakhunov This is my paper

classification quant-ph math.OC MSC 81P4090C22 PACS 03.65.Ud03.67.-a

keywords NPA hierarchydoubly-tilted CHSHquantum valuesemidefinite programmingexactnesscritical tiltcomputer-assisted prooftangent cone

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to prove that the standard semidefinite relaxation hierarchy for quantum correlations fails, at every finite level, to reproduce the true quantum value of the doubly-tilted CHSH functional in any neighbourhood of its critical point. It constructs, for each level k≥2, an explicit feasible moment curve whose objective is 4−s+g_k s² while the true quantum value is 4−s+O(s³), so the level-k value strictly exceeds the quantum value on a small interval. The proof rests on a closed-form, level-independent witness moment function that produces a rank-one compression on the relation lattice at every level, together with exact-integer computer verification of a finite enumeration. If correct, the result answers an open question from the recent determination of the quantum maximum and shows the failure is genuinely non-quantum: the overshoot direction cannot come from any state or smooth family of quantum models.

Core claim

The central claim is that for every NPA level k≥2 there are explicit rational constants g_k>0 and s*_k>0 such that c_k(s) ≥ 4−s+g_k s² for all s in (0,s*_k]. Since the quantum value has the cubic expansion c_Q(s)=4−s+s³/6+O(s⁴), every finite level strictly overshoots the quantum value on some interval (0,ε_k]; no finite level is exact on any neighbourhood of the critical point. The lower bound is proven by a primal construction: a quadratic moment arc y_0+s y_1+s² y_2 that is exactly feasible at each level, built from a level-independent signed witness y* satisfying N_kᵀΓ(y*)N_k = u_k u_kᵀ. The author also shows that for k≥3 no quantum state and no smooth curve of quantum models can realize

What carries the argument

The proof is carried by three interacting objects: the relation lattice N_k, the closed-form witness y*, and a no-repair positivity lemma. N_k consists of integer vectors forced into the kernel of every optimal moment matrix at the critical point—trivial reductions and dressings—and has dimension 2k². The witness y* is a signed class function with the exact rank-one compression N_kᵀΓ(y*)N_k = u_k u_kᵀ at every level, where u_k is an explicit level-uniform vector. The no-repair lemma shows that once the first-order compressed form is rank-one and the second-order block has a strictly positive margin, the quadratic arc y_0+s y_1+s² y_2 is exactly feasible, giving the quadratic lower bound with

Load-bearing premise

The proof leans on the exhaustive enumeration of kernel-pair configurations in Lemma 4.9—that every fine regime is covered by a {0,1,2} grid within side length 20, so checking 1,413,721 pairs settles the witness identity—and on the published quantum-value expansion c_Q(s)=4−s+s³/6+O(s⁴) to convert the quadratic lower bound into strict overshoot.

What would settle it

Run the exact verifier against the closed-form witness: compute E(K₁,K₂)=Σ c_p c'_q y*(m(p,q))−u(K₁)u(K₂) for all 1,413,721 kernel-element pairs with side lengths ≤20; any nonzero value refutes the witness identity and the quadratic lower bound. Independently, fix a level k, compute c_k(s) at s=10⁻⁶ and 10⁻⁷ with high-precision semidefinite programming; if (c_k(s)−4+s)/s² tends to 0 as s→0, the claimed quadratic gap is false.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

For every fixed level k≥2 there is an explicit interval (0,ε_k] on which the level-k relaxation value strictly exceeds the quantum value; no finite level is exact in any neighbourhood of the critical tilt.
The level of the hierarchy required to reach the quantum value diverges as the tilt approaches the critical point: D(s)→∞ as s→0⁺.
The overshoot is at least quadratic for every k≥2, with unconditional numerical constants a₂>1/39, a₃>1/188, a₄>1/641; granted the companion's dual expansion, the exponent is exactly two and bounded above by 3/64.
The gain direction cannot be realized by any state or by any smooth family of quantum models, locating the finite-level failure in the non-quantum part of the NPA tangent cone.
All constants in the construction—g_k, s*_k, ε_k—are explicit rationals, and the proof is accompanied by exact-integer verification programs that can be rerun.

Where Pith is reading between the lines

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

If the proof is correct, the non-exactness is a structural property of the restricted certificate class: any relaxation that truncates monomial word length the same way will inherit the same uniform gap near this critical point.
The signed, level-independent witness suggests a template for other Bell functionals whose quantum maximum has cubic contact at a boundary point: look for a class-function witness with rank-one compression and use the same no-repair arc to prove finite-level inexactness.
A quantitative next step would be to measure the decay of a_k or of t_0(k): a uniform-in-k lower bound on g_k, or a level-to-level inequality, would turn positivity into a divergence rate for the required level.
Because the exponent-two upper bound rests on the companion note's conditional expansion, a clean verification of that dual arc would make the Θ(s²) statement unconditional—a sharper target than re-checking the whole chain.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

A serious, carefully built proof that no finite NPA level is exact near the critical tilt; the central argument looks sound, but the load-bearing finite enumeration deserves independent scrutiny before full acceptance.

read the letter

I read Pakhunov's proof that no finite NPA level is exact for the doubly-tilted CHSH functional near the critical tilt. If correct, this settles a concrete open question from Gigena et al. in the negative, and it does so with a genuinely interesting construction: a level-independent signed witness that compresses to a rank-one form at every level, an exactly feasible arc with a cubic objective, and a no-repair lemma that makes the guarantee uniform in k. That is a real result, not a repackaging.

What earns credit: the proof chain is unusually explicit. The witness identity is checked by exact integer arithmetic over a finite enumeration, with degree and coverage lemmas that turn the finite check into a proof. The independent re-verification—including a symbolic per-regime proof and a clean-room implementation—is a serious attempt to catch self-deception. The one external input is the published quantum value, cross-checked symbolically. This is how computer-assisted proofs in quantum information should be packaged.

The soft spots are real but not fatal, in my reading. The biggest is the pair of companion notes [3] and [4], same author, unpublished, which supply the almost-quantum expansion and the swap-symmetrization soundness lemma. The main theorem does not need everything in those notes, but the upper-bound half of the exponent statement does, and that part is explicitly conditional. More centrally, the proof of Theorem 4.4 depends on Lemma 4.9: every fine regime realizes a full {0,1,2}-grid within side length 20. That lemma is argued in a few lines, and the 1.4M-pair enumeration is the load-bearing step. I did not run the code, and the stress-test note has a fair point: the totality of the regime classifier is a property of the verifier, not a hand-proof. An independent exhaustive check of regime totality and grid coverage would remove most of my residual worry. That is a request for more evidence, not a claim that the proof is wrong.

Who should read this: anyone working on SDP hierarchies, Bell nonlocality, or sum-of-squares convergence. It deserves a serious referee—ideally someone willing to actually run the verification scripts and comb through Lemma 4.8–4.9. I would not desk-reject it. I would send it to peer review and ask for a high bar on the enumeration coverage before acceptance.

Referee Report

1 major / 6 minor

Summary. The paper claims that for the symmetric doubly-tilted CHSH functional B_s, no finite level of the NPA hierarchy is exact on any neighbourhood of the critical point s=0. For every level k≥2 it constructs explicit rational g_k>0 and s*_k>0 with c_k(s) ≥ 4−s+g_k s^2 on (0,s*_k], while the quantum value is 4−s+s^3/6+O(s^4), yielding a uniform non-exactness gap. The proof is a primal construction: an exactly feasible moment arc y(s)=y_0+s y_1+s^2 y_2 at each level, built from a level-independent signed witness y* satisfying N_k^T Γ(y*)N_k = u_k u_k^T, a uniform Slater point, and a no-repair positivity lemma that makes the arc feasible without o(s^2) corrections. The argument is computer-assisted in the strict sense, with exact-integer enumerations, a fully symbolic per-regime verification, and an independent clean-room re-implementation.

Significance. If correct, this is a substantial result: it answers in the negative an open question of Gigena et al. and demonstrates a genuine uniform failure of finite-level NPA exactness at an attained, self-tested critical point. The proof is unusually rigorous for a computer-assisted argument: exact rational certificates, a level-uniform Slater point, a no-repair lemma with certified interval endpoints, machine-checked enumeration, a symbolic per-regime proof, and an adversarial audit with a clean-room implementation. The mechanism is also conceptually interesting: the quadratic overshoot is shown to live in the non-quantum part of the NPA tangent cone, realized by a single level-independent signed witness. If the central verification step (Theorem 4.4) is correct, the paper establishes a sharp and surprising result about the NPA hierarchy.

major comments (1)

[§4.3, Lemmas 4.8–4.9 and completion of Theorem 4.4] The conclusive force of the witness identity N_k^T Γ(y*)N_k = u_k u_k^T rests on the completeness of the fine-regime classification and the grid-coverage lemma. The hand proof of Lemma 4.9 is a single arithmetical sentence ('6+4+6+4=20'), and the totality of the regime classifier is asserted only as a property of the verification code. Because the ≤20 exact-integer enumeration and the symbolic per-regime verifier share the same regime definitions, a missing regime or a failure of the degree/grid bound would invalidate both. This is the single most load-bearing step: without E≡0 for all kernel pairs, the feasible-arc construction of Sections 5–6, and with it Theorem 1.1, collapses. Please expand the proof of Lemma 4.8 to a complete case analysis (including the treatment of empty side words, for which 'first letter' is undefined), give a detailed derivation of the bounds in Lemma 4.9, and

minor comments (6)

[§A / Theorem 1.1] The numerical bounds a_2>1/39, a_3>1/188, a_4>1/641 are stated as unconditional but rely on certified ladder jets from the unpublished companion note [3]. Please state this dependency explicitly, or include the jets in the manuscript so the claims can be verified without access to the companion.
[§2.1] The swap-symmetrization soundness is cited to companion note [4]. Since it is a one-line averaging argument (any maximizer can be averaged with its party-swap), include the proof for self-containedness rather than referencing an unpublished companion.
[§4.3] The symbolic verifier is described as using 'clamp 3' while Lemma 4.8 states a clamp at 4. Clarify the relationship between these two thresholds and why the different clamp does not affect the totality of the classification.
[§4.3 / Lemma 4.5] The treatment of empty side words is not explicit. In the fine-regime data, the 'first letter' of an empty word is undefined. Specify how the regime representation handles the empty word and how Lemma 4.5 (iii) applies when one of the base words is empty.
[Remark 4.10] The 'germ stabilization law' is described but not formally stated or proven. Since it is not used in the proof, please label it as a heuristic remark or provide a precise statement and proof.
[§1.1] The value c_1(0)≈4.73 is stated without derivation or citation; add a footnote or reference.

Circularity Check

0 steps flagged

No circular reduction: the feasible-arc construction is self-contained; companion-note citations are peripheral or conditional, and the finite witness verification is an exact proof rather than a fitted prediction.

full rationale

The central derivation does not reduce to its inputs. Theorem 4.4's witness identity N_k^T Γ(y*) N_k = u_k u_k^T is established by explicit locality lemmas plus an exact finite-integer verification over a full interpolation grid; it is a genuine construction, not a parameter fitted to the target c_k(s). Theorem 6.3 then builds an exactly feasible arc y(s) whose objective is 4 - s + g_k s^2 + t_0 λ^2 s^3, giving c_k(s) ≥ 4 - s + g_k s^2 as a primal lower bound rather than a rearrangement of the conclusion. Non-exactness uses the external quantum-value expansion (1) from Gigena et al. [1], which is a published input and not a self-citation. The same-author companion notes [3] and [4] are invoked only for the conditional exact-exponent-2 upper bound and for swap-symmetrization soundness; the latter is justified in-line by averaging any maximizer with its swap, and the former is explicitly marked conditional and not needed for the non-exactness statements. The quantitative ladder bounds a2>1/39 etc. are backed by exact rational certificate programs listed in Appendix A. The only substantive risk is correctness of the finite regime enumeration (Lemma 4.8/4.9 completeness), which is a computational verification concern, not circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 5 axioms · 0 invented entities

No free parameters are fitted to data: g_k and s*_k are constructed from explicit rational data. The central proof rests on standard mathematics, the external quantum expansion, and same-author companion notes. The signed class function y* is a mathematical object, not a new physical entity.

axioms (5)

standard math Finite-dimensional SDP theory, Gram-matrix representations, and Schur complement criteria are applied correctly.
The proofs use PSD block arguments and Albert's generalized Schur criterion; these are standard background throughout.
domain assumption The quantum value expansion c_Q(s)=4-s+s^3/6+O(s^4) from Gigena et al. [1] is correct.
This is explicitly called the one external input; it is used to translate the lower bound 4-s+g_k s^2 into strict overshoot of the quantum value.
domain assumption The party-swap-symmetrized class-space formulation is sound and complete for the swap-symmetric functional.
The soundness lemma is cited to companion note [4] rather than proven in this paper.
domain assumption The almost-quantum expansion c_{1+AB}(s)=4-s+3/64 s^2+o(s^2) and the certified ladder bounds a_2>1/39, a_3>1/188, a_4>1/641 from companion note [3] are valid.
Used for the upper exponent and for the quantitative small-level instances; the companion note is unpublished and lacks an independent identifier.
ad hoc to paper The exact-integer verification suite correctly implements the stated enumerations and the fine-regime degree/coverage lemmas are valid.
The proof of Theorem 4.4 depends on a finite enumeration of kernel pairs and on Lemmas 4.8-4.9 making that enumeration conclusive for all levels.

pith-pipeline@v1.3.0-alltime-deepseek · 22694 in / 16438 out tokens · 160740 ms · 2026-08-02T03:46:53.644159+00:00 · methodology

0 comments

read the original abstract

Gigena, Panwar, Scala, Araujo, Farkas and Chaturvedi [npj Quantum Inf. 11, 82 (2025)] determined the quantum maximum of the doubly-tilted CHSH functionals, observed that the Navascues-Pironio-Acin level needed for exactness grows without evident bound toward the critical tilt, and asked whether any finite level suffices. We answer this in the negative. For the symmetric critical family $B_s=(1-s/2)(\langle A_0\rangle+\langle B_0\rangle)+\mathrm{CHSH}$ we prove: for every NPA level $k\ge 2$ there are an explicit rational $g_k>0$ and an $s^*_k>0$ with $c_k(s)\ge 4-s+g_k s^2$ on $(0,s^*_k]$; since the quantum value leaves the local bound only cubically, every finite level strictly overshoots on an interval: no finite level is exact on any neighbourhood of the critical point. Unconditionally $a_2>1/39$, $a_3>1/188$, $a_4>1/641$. The proof is a primal construction: an exactly feasible moment curve at each level, built from level-uniform structural laws and one level-independent signed witness -- a closed-form class function $y^*$ with $N_k^T\Gamma(y^*)N_k=u_k u_k^T$ at every level. The mechanism forces the sign: for $k\ge 3$ no quantum state and no smooth curve of quantum models can realize the gain direction, so the overshoot lives strictly in the non-quantum part of the NPA tangent cone. The proof is computer-assisted in the strict sense: finite exact-integer verifications with proven degree bounds are constituent parts of the argument; the chain has been re-verified against independent implementations, including a symbolic per-regime proof of the witness identity and a clean-room implementation written from the paper text alone. The one external input is the published quantum value of Gigena et al., cross-checked to twelve digits.

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

No Finite NPA Level Characterizes the Complete Quantum Set in the Simplest Bell Scenario
quant-ph 2026-07 accept novelty 7.0

In the (2,2,2) Bell scenario, every fixed finite NPA level strictly over-approximates the quantum correlation set, because any bounded error would force the Motzkin polynomial plus a constant to be a sum of squares, w...

Reference graph

Works this paper leans on

12 extracted references · 5 linked inside Pith · cited by 1 Pith paper

[1]

Gigena, E

N. Gigena, E. Panwar, G. Scala, M. Ara´ ujo, M. Farkas, A. Chaturvedi,Self-testing tilted strategies for maximal loophole-free nonlocality, npj Quantum Inf.11, 82 (2025); arXiv:2405.08743

Pith/arXiv arXiv 2025
[2]

Navascu´ es, S

M. Navascu´ es, S. Pironio, A. Ac ´ ın,A convergent hierarchy of semidefinite programs char- acterizing the set of quantum correlations, New J. Phys.10, 073013 (2008)

2008
[3]

Pakhunov,A phase transition in the exactness of the NPA hierarchy at the critical doubly-tilted CHSH functional, companion note (2026)

A. Pakhunov,A phase transition in the exactness of the NPA hierarchy at the critical doubly-tilted CHSH functional, companion note (2026)

2026
[4]

Pakhunov,A certified separation of the almost-quantum set from the quantum set in the minimal Bell scenario, companion note (2026)

A. Pakhunov,A certified separation of the almost-quantum set from the quantum set in the minimal Bell scenario, companion note (2026)

2026
[5]

Barizien, J.-D

V. Barizien, J.-D. Bancal,Quantum statistics in the minimal scenario, Nat. Phys.21, 577 (2025); arXiv:2406.09350

arXiv 2025
[6]

I. Klep, J. Levenson, S. McCullough,Fej´ er–Riesz factorization for positive noncommutative trigonometric polynomials, arXiv:2511.09267 (2025)

arXiv 2025
[7]

Nie,Optimality conditions and finite convergence of Lasserre’s hierarchy, Math

J. Nie,Optimality conditions and finite convergence of Lasserre’s hierarchy, Math. Program. 146, 97 (2014); arXiv:1206.0319

Pith/arXiv arXiv 2014
[8]

Marshall,Representations of non-negative polynomials, degree bounds and applications to optimization, Canad

M. Marshall,Representations of non-negative polynomials, degree bounds and applications to optimization, Canad. J. Math.61, 205 (2009);Positive Polynomials and Sums of Squares, Math. Surveys Monogr.146, AMS (2008)

2009
[9]

Bamps, S

C. Bamps, S. Pironio,Sum-of-squares decompositions for a family of CHSH-like inequalities and their application to self-testing, Phys. Rev. A91, 052111 (2015); arXiv:1504.06960

Pith/arXiv arXiv 2015
[10]

L. F. Vargas,On the hardness of deciding the finite convergence of Lasserre hierarchies, Numer. Algebra Control Optim.16, 35 (2026); arXiv:2401.12613

Pith/arXiv arXiv 2026
[11]

M. A. Dritschel,Factoring non-negative operator valued trigonometric polynomials in two variables, Math. Ann.391, 515 (2024); arXiv:1811.06005. 22

Pith/arXiv arXiv 2024
[12]

Scheiderer,Sums of squares on real algebraic surfaces, Manuscripta Math.119, 395 (2006);Non-existence of degree bounds for weighted sums of squares representations, J

C. Scheiderer,Sums of squares on real algebraic surfaces, Manuscripta Math.119, 395 (2006);Non-existence of degree bounds for weighted sums of squares representations, J. Complexity21, 823 (2005). 23

2006

[1] [1]

Gigena, E

N. Gigena, E. Panwar, G. Scala, M. Ara´ ujo, M. Farkas, A. Chaturvedi,Self-testing tilted strategies for maximal loophole-free nonlocality, npj Quantum Inf.11, 82 (2025); arXiv:2405.08743

Pith/arXiv arXiv 2025

[2] [2]

Navascu´ es, S

M. Navascu´ es, S. Pironio, A. Ac ´ ın,A convergent hierarchy of semidefinite programs char- acterizing the set of quantum correlations, New J. Phys.10, 073013 (2008)

2008

[3] [3]

Pakhunov,A phase transition in the exactness of the NPA hierarchy at the critical doubly-tilted CHSH functional, companion note (2026)

A. Pakhunov,A phase transition in the exactness of the NPA hierarchy at the critical doubly-tilted CHSH functional, companion note (2026)

2026

[4] [4]

Pakhunov,A certified separation of the almost-quantum set from the quantum set in the minimal Bell scenario, companion note (2026)

A. Pakhunov,A certified separation of the almost-quantum set from the quantum set in the minimal Bell scenario, companion note (2026)

2026

[5] [5]

Barizien, J.-D

V. Barizien, J.-D. Bancal,Quantum statistics in the minimal scenario, Nat. Phys.21, 577 (2025); arXiv:2406.09350

arXiv 2025

[6] [6]

I. Klep, J. Levenson, S. McCullough,Fej´ er–Riesz factorization for positive noncommutative trigonometric polynomials, arXiv:2511.09267 (2025)

arXiv 2025

[7] [7]

Nie,Optimality conditions and finite convergence of Lasserre’s hierarchy, Math

J. Nie,Optimality conditions and finite convergence of Lasserre’s hierarchy, Math. Program. 146, 97 (2014); arXiv:1206.0319

Pith/arXiv arXiv 2014

[8] [8]

Marshall,Representations of non-negative polynomials, degree bounds and applications to optimization, Canad

M. Marshall,Representations of non-negative polynomials, degree bounds and applications to optimization, Canad. J. Math.61, 205 (2009);Positive Polynomials and Sums of Squares, Math. Surveys Monogr.146, AMS (2008)

2009

[9] [9]

Bamps, S

C. Bamps, S. Pironio,Sum-of-squares decompositions for a family of CHSH-like inequalities and their application to self-testing, Phys. Rev. A91, 052111 (2015); arXiv:1504.06960

Pith/arXiv arXiv 2015

[10] [10]

L. F. Vargas,On the hardness of deciding the finite convergence of Lasserre hierarchies, Numer. Algebra Control Optim.16, 35 (2026); arXiv:2401.12613

Pith/arXiv arXiv 2026

[11] [11]

M. A. Dritschel,Factoring non-negative operator valued trigonometric polynomials in two variables, Math. Ann.391, 515 (2024); arXiv:1811.06005. 22

Pith/arXiv arXiv 2024

[12] [12]

Scheiderer,Sums of squares on real algebraic surfaces, Manuscripta Math.119, 395 (2006);Non-existence of degree bounds for weighted sums of squares representations, J

C. Scheiderer,Sums of squares on real algebraic surfaces, Manuscripta Math.119, 395 (2006);Non-existence of degree bounds for weighted sums of squares representations, J. Complexity21, 823 (2005). 23

2006