The Navascués--Pironio--Acín (NPA) hierarchy gives the standard semidefinite outer approximations to quantum behaviors. Whether \emph{any} finite level can already equal the quantum set has remained open even in the bipartite scenario with two binary measurements per party. We demonstrate that \emph{no finite level} is exact. For the symmetric doubly tilted CHSH functional $h_α=A_0B_0+A_0B_1+A_1B_0-A_1B_1+α(A_0+B_0)$, set $T=1-α$. Its quantum maximum satisfies $[ω_{\rm Q}(1-T)-(4-2T)]/T^3\to4/3$, whereas every fixed NPA level satisfies $[ω_L(1-T)-ω_{\rm Q}(1-T)]/T^3\to+\infty$. Under the corresponding boundary rescaling, an explicit expectation of the positive operator $ω_{\rm Q}(1-t^2)I-H_t$ converges to the Motzkin polynomial. A bounded fixed-level error would therefore make the Motzkin polynomial plus a nonnegative constant a sum of squares, which is impossible. Consequently, every standard NPA relaxation based on a fixed finite list of words in the measurement projectors strictly contains the complete quantum set, and its nonquantum behaviors accumulate at a local deterministic behavior. Thus, the finite-level exactness of CHSH and all one-sided tilted CHSH maxima does not extend to an exact finite-level description of the complete quantum set in the minimal scenario.
Comments
Submitted to Physical Review Letters on 14 July 2026. Closely overlapping preprints arXiv:2607.13762 and arXiv:2607.13774 appeared on 15 July 2026. This work was completed independently beforehand and gives a distinct fully analytical proof with broader behavior-set consequences
The Navascués--Pironio--Acín (NPA) hierarchy gives the standard semidefinite outer approximations to quantum behaviors. Whether \emph{any} finite level can already equal the quantum set has remained open even in the bipartite scenario with two binary measurements per party. We demonstrate that \emph{no finite level} is exact. For the symmetric doubly tilted CHSH functional hα=A0B0+A0B1+A1B0−A1B1+α(A0+B0), set T=1−α. Its quantum maximum satisfies [ωQ(1−T)−(4−2T)]/T3→4/3, whereas every fixed NPA level satisfies [ωL(1−T)−ωQ(1−T)]/T3→+∞. Under the corresponding boundary rescaling, an explicit expectation of the positive operator ωQ(1−t2)I−Ht converges to the Motzkin polynomial. A bounded fixed-level error would therefore make the Motzkin polynomial plus a nonnegative constant a sum of squares, which is impossible. Consequently, every standard NPA relaxation based on a fixed finite list of words in the measurement projectors strictly contains the complete quantum set, and its nonquantum behaviors accumulate at a local deterministic behavior. Thus, the finite-level exactness of CHSH and all one-sided tilted CHSH maxima does not extend to an exact finite-level description of the complete quantum set in the minimal scenario.
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 Bs=(1−s/2)(⟨A0⟩+⟨B0⟩)+CHSH we prove: for every NPA level k≥2 there are an explicit rational gk>0 and an sk∗>0 with ck(s)≥4−s+gks2 on (0,sk∗]; 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 a2>1/39, a3>1/188, a4>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 NkTΓ(y∗)Nk=ukukT at every level. The mechanism forces the sign: for k≥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.
Gigena et al. [npj Quantum Inf. 11, 82 (2025)] proved the exact quantum maximum of the doubly-tilted CHSH functional Bαβ=α⟨A0⟩+β⟨B0⟩+CHSH and observed that the NPA level required to reach it grows without evident bound toward the critical line α+β=2. We quantify the mechanism on the symmetric slice s=2−α−β: (i) the quantum value leaves the local bound cubically, cQ=4−s+s3/6−s4/36+O(s5); (ii) each NPA level overshoots quadratically, ck(s)=4−s+aks2+O(s3), with the almost-quantum coefficient computed exactly, a1+AB=3/64; (iii) the divergence of the required exact level is equivalent to positivity of the single sequence (ak) - proven for every k in the companion paper. We prove the supercritical side completely: for all α,β≥1 and every level the hierarchy is exact, via three explicit rational certificates realizing an affine identity. The hierarchy's exactness thus undergoes a phase transition at the critical line. On the subcritical side we certify the first four levels in exact arithmetic (rational pseudo-moments beating cQ, confirmed by Sturm's theorem). We identify the exact mechanism: rescaled to the critical corner, the limiting obstruction is the Motzkin polynomial, the classical nonnegative-but-not-sum-of-squares form, so the finite-level failure sits in the restricted-certificate regime. The phase boundary has a precise geometric reading via Nie's finite-convergence theorem and Marshall's boundary Hessian condition: a self-tested optimum is finitely NPA-certifiable whenever its boundary Hessian is nondegenerate (contact order two), which holds for the single tilt and fails exactly at the doubly-tilted cubic touch. Three verified errata in the published polynomial system of Gigena et al. are documented.
The conditional probabilities achievable by local measurements on a shared quantum state in a Bell scenario form the set of quantum behaviors, whose structure was studied by Tsirelson. In 1993, Tsirelson asked whether this set is semialgebraic, that is, describable by finite Boolean combinations of polynomial equations and inequalities. This question has remained open. We answer it in the negative: with four binary measurements per party, the set of finite-dimensional quantum behaviors, its closure, and the commuting-operator set are all nonsemialgebraic. More strongly, none admits a finite real-analytic description even locally near a particular classical behavior. These results rule out exact finite semidefinite representations and show that no finite level of the Navascués-Pironio-Acín hierarchy characterizes these sets exactly.
We give exact rational certificates for two results in the bipartite Bell scenario with three two-outcome measurements per party. First, the maximum of the I3322 (Froissart) functional over the almost-quantum set -- the 1+AB level of the Navascues-Pironio-Acin (NPA) hierarchy -- is certified to lie between 0.2514708838 and 0.2514709016 (width <2×10−8), in the normalization with classical bound 0 and quantum maximum ≈0.2508754; to our knowledge no value for this landmark had been published. Second, we exhibit a single exact rational behavior T∗ carrying certificates in both directions: a rational moment matrix proving T∗ almost-quantum, and a rational dual object proving that every NPA level-2 completion of T∗ requires slack at least q=4.5934×10−4, so that T∗ is not quantum. The strict inclusion Q⊊Q~ in the minimal three-setting scenario is thereby established end-to-end with no reliance on floating-point computation. Anatomically, the witness is locally quantum: all nine of its 2×2 correlator subtables pass the Tsirelson-Landau-Masanes criterion, so its non-quantumness is irreducibly a three-setting phenomenon, concentrated in ordering-dependent pseudo-moments coupling each party's third setting to the other's first two; the level-2 certificate is equivalently an exact sum-of-squares identity that any quantum state values nonnegatively and the witness is forced to value at −q. The certificates follow the rational rounding-projection-lifting programme recently developed for noncommutative optimization; for NPA moment matrices in canonical form the required projection reduces to a per-moment uniform shift, making certification essentially free at these sizes, and we work the programme in both the feasibility and infeasibility directions on one witness. Machine-checkable certificates and verifiers requiring only the Python standard library accompany the paper. Files in this record: the paper (PDF); exact rational certificates for both theorems (i3322_1AB_certificate.json, i3322_level2_infeasibility_certificate.json); standalone verifiers requiring only the Python 3 standard library (verify_i3322_certificate.py, verify_level2_certificate.py); and the generating scripts (certify_landmark.py, certify_level2.py). See README_ancillary.txt for which artifact proves which theorem and how to run the verifiers. v2: corrects the mechanism attribution in Sec. 5 (anatomy) and adds the archive reference to Data Availability; theorems, certificates, and verifiers unchanged.
The Navascués-Pironio-Acín (NPA) hierarchy provides a convergent sequence of semidefinite programming (SDP) relaxations for noncommutative polynomial optimisation, ubiquitous in quantum physics. However, its practical applicability is limited by the combinatorial growth in operator moments required at each level. Since not all moments contribute equally to bound tightness, selecting moments within a fixed computational budget is a relevant problem.
We reframe moment selection as combinatorial subset selection and show it is governed by strong higher-order synergistic interactions among moments, quantified through a marginal synergy diagnostic adapted from complex systems theory. We develop and compare three optimisation methods: Parallel Tempering (PT), an RBM-based reinforcement learning policy, and Bayesian Optimisation (BO). On the I3322 Bell inequality benchmark, all three substantially outperform greedy approaches at costs around two orders of magnitude below brute force, with the RBM achieving the closest approach to optimal throughout the hard transition regime.
We apply the framework to the 174 Bell inequalities in the (4,4,2,2) scenario, finding heterogeneous convergence behaviour across inequalities, and to the one-dimensional Heisenberg spin chain, demonstrating that physically motivated monomial bases are internally compressible and are not globally optimal in general. A budget-aware search over a broader pool improves certified bounds on long-range correlations by nearly two orders of magnitude. These results establish a scalable framework for moment selection in noncommutative polynomial optimisation, with broad applications across quantum physics and quantum information.
Almost-quantum correlations form the 1+AB level of the Navascués--Pironio--Acín hierarchy. They strictly contain the quantum set, are characterized by a single semidefinite program, arise in Sorkin's quantum measure theory, and satisfy several principles proposed to single out quantum theory. We show that they are nevertheless unboundedly stronger than quantum correlations in a zero-error communication task. Alice and Bob receive labels s,t∈[N], N≥3; after one bit from Alice, Bob must decide whether s=t, never erring but allowed to abstain. For every model of quantum correlations, from finite-dimensional tensor products to commuting operators, the optimal worst-case probability of a conclusive answer is exactly 1/N, the value attained with shared randomness, so the bound does not depend on the resolution of Tsirelson's problem. An explicit family of almost-quantum correlations attains 1/19 for every N. For certificates of equality from binary-outcome boxes, we solve the underlying semidefinite programs in closed form. The almost-quantum optimum equals N2/(17N2−80N+64) for N≥8 and tends to 1/17. The first level of the hierarchy, which coincides with the set singled out by macroscopic locality, attains N/[8(N−2)], which tends to 1/8; its optimal boxes lie outside the almost-quantum set for every N≥5. For longer messages, entanglement saves at most one bit over shared randomness, whereas almost-quantum correlations reduce the one-way zero-error cost of deciding equality of n-bit strings from n bits to at most 13. The quantum bound also yields a linear witness of post-quantumness on a face of the nonsignalling polytope.