{"id":"fc53370d-550c-4edf-9145-513fb0312caf","arxiv_id":"2508.01071","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A d-input, d-outcome Bell inequality with an explicit sum-of-squares certificate robustly self-tests every maximally entangled qudit state, with O(sqrt(epsilon)) state and measurement error.","lead":"This paper proves a device-independent test that certifies high-dimensional entangled states purely from the observed correlations of a Bell experiment, with an explicit error bound. The test works for every dimension, using Heisenberg-Weyl observables that generalize the CHSH test from qubits to qudits.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 1's proof does not establish uniqueness of Eq (22): M is d×d² with full row rank, so ker M has dimension d(d−1); the Appendix C route's diagonal Kronecker rows span only d dimensions of C^{d²}.","rationale":"The reader's weakest_assumption is well-targeted. After checking Eq (20)–(22) and Appendix C, the linear-algebraic step that is supposed to force B_kB_l|ψ> = ω^{2^{-1}(k-l)}B^2_{2^{-1}(k+l)}|ψ> does not follow. Eq (22) is a d×d² system; full row rank of M gives consistency and a d-dimensional image, but the kernel has dimension d(d−1). The text's statement 'Because the matrix is of maximal rank, the solution is uniquely determined' is false. The Appendix C route is no better: Eq (C12) provides, for each j, a single equation whose coefficients are the j-th diagonal Kronecker row u_j⊗u_j; these d rows are orthonormal but span only d dimensions of C^{d²}, so R=0 does not follow. The subsequent claim 'since g(j,k,n)≠0' is insufficient. This gap is load-bearing because Lemma 1's relations (17) are the only input to the Mayers-Yao isometry in Appendix D and to every robustness bound in Appendices E–G. Without (17), the theorem does not have a proof even in the ideal case. I did not find a further independent flaw that would change the reader's verdict. The composite-dimension extension is indeed only sketched, and the absence of an explicit LHV bound for d>3 is a secondary weakness, but both are subordinate to the Lemma 1 gap. The concern is likely fixable—e.g., by using the full set of equations over all n,n' or by exploiting unitarity of the B_k—but as written the proof is incomplete. Hence the conditional verdict stands.","tokens_in":29322,"tokens_out":20902,"duration_ms":252680,"concrete_test":"For d=5, build the d×d² matrix M with entries M_{j,(k,l)}=g(j,k,1)g(j,l,1) using the paper's g from Eq (10), and numerically compute the nullspace of M. If dim ker M = d(d−1)>0, exhibit a nonzero c with M c=0, refuting the claimed uniqueness. Then test whether any such c can define B_kB_l|ψ>=Σ_m c^m_{kl}B^2_m|ψ> that is consistent with unitarity of B_kB_l (norm one and consistency of the Gram matrix); if a unitarity-compatible alternative exists, Lemma 1 is false. Separately, check whether the d vectors {row_j(G_1)⊗row_j(G_1)} span C^{d²}; they span only d dimensions, confirming that the Appendix C inference R=0 is invalid.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Lemma 1's twisted commutation relations (17) are the load-bearing step for the whole self-test. The proof reduces to the d×d² linear system M c(n)=g_n in Eq (22). The paper argues that because M has full row rank d, the solution is unique. This is incorrect: a rank-d matrix with d² columns has a d(d−1)-dimensional kernel, so Eq (22) determines at most the projection of c(n) onto the row space. The alternative route in Appendix C has the same problem: Eq (C12) asserts Σ_{k,l} g(j,k,1)g(j,l,1)R_{k,l}=0 for each j, but the coefficient vectors are the d diagonal Kronecker rows u_j⊗u_j of G_1⊗G_1. These rows are orthonormal yet span only a d-dimensional subspace of C^{d²}; they cannot force R=0. The remark 'since g(j,k,n)≠0' does not justify termwise vanishing. Unless additional constraints—unitarity of the products B_kB_l, or the full family of equations over all n,n'—are invoked, the relations (17) are not proven. Since Appendix D's isometry and the robustness analyses in Appendices E–G rely on (17), the central claim is unsupported at its core.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a device-independent certification protocol for maximally entangled states of every finite local dimension. For each odd prime d, it defines a d-input, d-output Bell operator built from non-diagonal Heisenberg-Weyl observables and a non-Clifford phase gate, proves an exact sum-of-positive-operators decomposition, and claims that near-maximal violation forces the underlying state and measurements to be close, up to local isometries, to the ideal rotated Bell state and canonical Heisenberg-Weyl operators. The main theorem states an O(√ε) robustness bound for prime d, and a tensor-factor argument is sketched to extend the result to composite d. The proof structure has two stages: first, a Lemma establishing twisted commutation relations from the SOPO decomposition, and second, a Mayers-Yao isometry construction. The robustness analysis in the appendices follows the ideal-case proof with error bounds.","tokens_in":29530,"tokens_out":9444,"duration_ms":114133,"significance":"If the proof gaps are repaired, this would be a significant advance: it provides a single, analytic, noise-tolerant self-testing protocol covering all finite dimensions, with explicit robustness bounds and simple Heisenberg-Weyl measurements, in contrast to earlier constructions that required tailored inequalities, number-theoretic conditions, or lacked robustness. The explicit SOPO decomposition and the detailed isometry and robustness calculations in Appendices B-G are valuable and largely self-contained. The paper also gives a fair account of prior work on qudit self-testing. However, the central algebraic step of Lemma 1 is not proven as written, and the claimed extension to every composite dimension is only sketched, so the main theorem is currently unsupported at a load-bearing point.","major_comments":[{"comment":"The uniqueness claim for the solution of Eq. (22) is incorrect. The matrix M is d×d² with full row rank d, so its kernel has dimension d(d−1); full row rank only makes the row-space projection of c(n) unique, not the full coefficient vector C^n_{kl}. Consequently the conclusion C^n_{kl} = ω^{2^{-1}(k−l)} δ_{n,2^{-1}(k+l)} does not follow from the displayed linear system. Since the twisted commutation relations (17) are derived from this uniqueness, Lemma 1 is not established as stated. The proof needs an explicit argument using the full family of equations over n,n′ or additional constraints from unitarity of the operators B_kB_l, rather than the rank of M alone.","section":"Proof of Lemma 1 for d>3, Eq. (22)"},{"comment":"The step from the weighted sum Σ_{k,l} g(j,k,n)g(j,l,n′)[...]|ψ⟩ = 0 to termwise vanishing of each bracket [...]|ψ⟩ is a non sequitur. The coefficients for fixed j form the diagonal Kronecker rows u_j^{(n)} ⊗ u_j^{(n′)} of G_n ⊗ G_{n′}; these rows span only a d-dimensional subspace of C^{d²}, so they cannot force a general vector in C^{d²} to vanish. The remark that g(j,k,n) ≠ 0 does not justify the conclusion unless the coefficient matrix has full column rank, which it does not. A rank or linear-independence argument over the full set of equations is required; the current text does not provide one.","section":"Appendix C, Eqs. (C12)-(C14)"},{"comment":"The proof expands B_kB_l|ψ⟩ in the basis {B^2_m|ψ⟩} and then projects with ⟨ψ|(B^2_n)^†, but the orthonormality of {B^2_m|ψ⟩} is not proven before this point. The SOPO relations and unitarity of the B_m do not by themselves imply that these vectors form an orthonormal basis of the grade-2 sector. If this fact is intended to follow from other parts of the argument, the appendix containing that proof should be cited explicitly; I did not find such a proof.","section":"Proof of Lemma 1 for d>3, expansion in Eq. (22)"},{"comment":"The extension to every composite dimension is only sketched and is not a rigorous theorem. The prime-dimension result covers odd primes only, so a composite d with a factor 2 is not covered by the stated theorem. Moreover, decomposing C^d as a tensor product of prime-power subsystems does not automatically imply that the Bell operator B_d or the Heisenberg-Weyl observables A_j, B_j factor accordingly; the protocol and the robustness bound for the joint system are not derived. Since the title and abstract claim self-testing in every finite dimension, this gap must be addressed with a explicit tensor-product construction and proof.","section":"Discussion, tensor-factor argument"}],"minor_comments":[{"comment":"The definition of µ_3 is inconsistent with Appendix G: Theorem 1 writes µ_3 = 9√ε(√3+2), while Eq. (G3) defines µ_3 = 9(√3+2). The current definition would make δ(ε) scale as O(ε) for d=3, contradicting the abstract's O(√ε). Please correct the typo.","section":"Theorem 1 statement"},{"comment":"The phrase \"this specual case\" should read \"this special case\".","section":"Introduction, after Eq. (7)"},{"comment":"Several typos occur: \"Form here onward\" should be \"From here onward\" and \"If follows that\" should be \"It follows that\" in the first paragraphs of Appendices E and F.","section":"Appendices E and F"},{"comment":"The sentence \"We conjecture that self-testing most of the algebra separates this Bell test from previous works\" is grammatically incomplete and unclear; please rephrase.","section":"Discussion"},{"comment":"The isometry circuit in the main text defines X := A^†_0 and Z := ω^{-2}A^†_0 A_1, while Appendix D uses expressions such as ω^{-2^{-1}j}A^†_0 A_j; please ensure the notation is consistent and the constants are defined unambiguously.","section":"Appendix D and Fig. 2"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses an important problem and contains a promising construction, but the missing uniqueness proof for Eq. (22) and the unjustified termwise conclusion in Appendix C are central, and the composite-dimension extension is only a sketch. I believe the technical gaps are potentially repairable, but they require substantial additional argument, not just local editing. The authors should also be asked to clarify the status of d=2 and even composite dimensions in the tensor-factor claim."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper has a genuinely novel construction: a d-input, d-outcome Bell operator built from non-diagonal Heisenberg–Weyl observables, with an explicit sum-of-positive-operators decomposition and analytic O(sqrt(epsilon)) robustness bounds. The appendices are detailed and the authors are honest about what is conjectural versus proven. If the central lemma were fixed, this would be a useful addition to device-independent quantum information.\n\nThe problem is Lemma 1. The proof that optimal violation forces the twisted commutation relations (17) reduces to the linear system M c(n) = g_n in Eq. (22). The paper claims uniqueness because M has full row rank d, but M is d-by-d^2, so a rank-d matrix has a d(d-1)-dimensional kernel. Full row rank does not give a unique solution. The alternate route in Appendix C has the same issue: Eq. (C12) gives sum_{k,l} g(j,k,n) g(j,l,n') R_{k,l}=0 for each j, but the coefficient vectors are the diagonal Kronecker rows u_j ⊗ u_j, which are orthonormal yet span only a d-dimensional subspace of C^{d^2}. The remark that g(j,k,n) ≠ 0 does not justify termwise vanishing. Without the twisted commutation relations, the isometry in Appendix D has no premise, and the robustness analyses in Appendices E–G inherit the gap. The composite-dimension extension and the missing explicit LHV bound for d>3 are secondary; the lemma is the load-bearing step.\n\nThis is not a matter of a small typo; the argument as written is invalid. However, the gap may be fixable by using the full family of equations over all n, n' together with unitarity of the products B_k B_l, and the rest of the paper suggests the authors know the intended structure. So I would not reject the paper outright, but I would not accept it in the current form either.\n\nWho should read it: anyone working on high-dimensional self-testing, because the Bell operator and SOPO construction are interesting even if the proof needs repair. A serious referee should engage with it, but the authors must supply a correct linear-independence argument or an alternative proof of Lemma 1 before the central claim can be trusted.","headline":"New Bell operator and explicit SOPO decomposition for qudit self-testing, but the proof of the central Lemma 1 has a linear-algebra gap that, as written, leaves the main claim unsupported.","tokens_in":30119,"tokens_out":2824,"would_cite":false,"duration_ms":34236,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P40","81P68"],"pacs":[],"model":"deepseek-v4-flash","headline":"A single Bell experiment robustly certifies maximal entanglement in every finite dimension.","keywords":["self-testing","device-independent certification","maximally entangled states","qudit Bell inequalities","Heisenberg-Weyl operators","robustness bounds","sum-of-positive-operators decompositions","CHSH generalization"],"falsifier":"For $d=5$, compute the kernel of the matrix $M$ whose rows are the products $g(j,k,1)g(j,l,1)$; if any nonzero vector $c(n)$ satisfies $M c(n) = g(j,n,2)$, then the uniqueness claim in Lemma 1 fails and the twisted commutation relations are not secured by the proof as written.","tokens_in":29037,"feed_emoji":"⚛️","tokens_out":11833,"duration_ms":128813,"temperature":0.7,"pith_summary":"This paper tries to establish that a single $d$-input, $d$-outcome Bell experiment can certify, without trusting the devices, that a bipartite quantum system of any finite dimension $d$ carries a maximally entangled state, and that the certification tolerates noise. The certificate generalizes the CHSH test to qudits by using non-diagonal Heisenberg–Weyl observables, and the Bell operator comes with an exact sum-of-positive-operators decomposition that gives the quantum maximum $d(d-1)$ in closed form. The main theorem says that if the observed Bell value is within $\\epsilon$ of that maximum, then, up to local unitaries and an auxiliary factor, the state is within trace distance $\\mathcal{O}(\\sqrt{\\epsilon})$ of the ideal maximally entangled state and the measurements are close to the canonical Heisenberg–Weyl operators $T(1,j)$. The proof handles every odd prime dimension directly, and a tensor-factor argument lifts the result to every composite dimension. If correct, this is the first single protocol that is both robust and universal in dimension, which earlier high-dimensional self-tests lacked.","feed_headline":"One Bell test certifies maximal entanglement in any finite dimension","feed_subtitle":"Epsilon-half error bounds make high-dimensional device-independent certification practical in photonic and atomic setups.","key_machinery":"The load-bearing object is the Bell operator $B_d$, defined as a sum over non-diagonal Heisenberg–Weyl terms $g(j,k,n)\\, A_j^n \\otimes B_k^n$, summed over $j,k \\in \\mathbb{Z}_d$ and $n \\in \\mathbb{Z}_d^*$, with coefficients $g(j,k,n)$ built from a non-linear phase function $\\nu$. The key identity is the sum-of-positive-operators decomposition $d(d-1)\\mathbb{1} - B_d = \\sum_{n=1}^{(d-1)/2} \\sum_{j \\in \\mathbb{Z}_d} C_{n,j}^\\dagger C_{n,j}$ with $C_{n,j} = A_j^n - \\sum_k g(j,k,n)^* (B_k^n)^\\dagger$. This decomposition turns near-maximal violation into small-norm conditions $C_{n,j}|\\psi\\rangle \\approx 0$, from which the proof derives twisted commutation relations and the relations linking $A_j$ to $B_j$. The second carrying mechanism is a local isometry, defined by a qudit SWAP circuit that applies Fourier transforms and the operators $A_0, A_1$ on Alice's side and $B_0, B_1$ on Bob's side; it maps the physical state to a rotated maximally entangled state times an auxiliary factor and maps the physical observables to the canonical $T(1,j)$.","core_discovery":"For every odd prime $d \\geq 3$, the paper constructs a Bell operator $B_d$ (with a separate operator $B_3$ for $d=3$) from non-diagonal Heisenberg–Weyl observables $A_j$ and $B_j$, and shows that near-maximal violation of the inequality $\\langle B_d \\rangle \\leq d(d-1)$ forces the state and measurements to the ideal ones. Concretely, Theorem 1 states that whenever $|\\langle \\psi | B_d | \\psi \\rangle - d(d-1)| \\leq \\epsilon$, there exist local unitaries $V_A, V_B$ and an auxiliary state such that the rotated state is within trace distance $\\sqrt{\\epsilon}\\, d(d-1)\\,(\\mu_d(4+1/d)+1)$ of the maximally entangled state, with $\\mu_d = \\sqrt{d}(\\sqrt{d}+2)$ for $d>3$ and a slightly different constant for $d=3$, and under these unitaries the observables map to canonical Heisenberg–Weyl operators $T(1,j)$. The proof has two stages: maximal violation forces twisted commutation relations $(A_jA_k - \\omega^{j-k}A_kA_j)|\\psi\\rangle = 0$ and the analogue for $B$, and then a local isometry built from the operators $A_0, A_1, B_0, B_1$ extracts the entangled state and implements the observable mapping. Because every finite dimension factorizes into prime-power blocks, the prime-dimensional result yields a robust self-test for every composite dimension, with errors accumulating at most linearly in the number of blocks.","pith_inferences":["The tensor-factor argument suggests a modular experimental certification strategy for large local dimension: certify each prime-power factor independently and combine the certificates; this is a testable extension of the paper's robustness analysis.","The $\\mathcal{O}(\\sqrt{\\epsilon})$ robustness bound is exactly the ingredient a device-independent randomness or key-rate analysis would need, so folding this self-test into an entropy-accumulation argument is a plausible next step that the paper does not carry out.","Whether the $d$-input overhead can be reduced while keeping explicit analytic measurement forms is left open; a constant-setting protocol with comparable robustness would be a natural competitor, but the paper does not claim one."],"forward_implications":["Correlations within $\\epsilon$ of $d(d-1)$ certify the high-dimensional maximally entangled state up to trace distance $\\mathcal{O}(\\sqrt{\\epsilon})$, so the self-test is noise-tolerant rather than ideal-case only.","The same self-test certifies the measurements: the local observables are mapped to canonical Heisenberg–Weyl operators $T(1,j)$, not just the state.","Since every finite dimension decomposes into prime-power factors, a single protocol now covers all dimensions, and the composite-dimension robustness degrades only linearly in the number of prime-power blocks.","The explicit SOPO decomposition proves the Cirelson bound $d(d-1)$ in closed form and shows analytically that the correlations are nonlocal for every odd prime $d$, where previous work only had numerical evidence for small dimensions.","The required operations are Heisenberg–Weyl displacements and diagonal non-Clifford phase gates, so the protocol maps directly onto high-dimensional photonic and atomic experimental platforms."],"supporting_citations":[{"why":"supplies the qubit local-isometry construction whose qudit generalization extracts the state and observables in Theorem 1.","marker":"[3]"},{"why":"is the earlier arbitrary-dimension self-test without robustness that the present protocol upgrades with explicit error bounds.","marker":"[38]"},{"why":"introduces the CHSH-type qudit Bell operator and the coefficients $g(j,k,n)$ that the present experiment refines by dropping diagonal terms.","marker":"[40]"},{"why":"provides the quantum Bell bound whose closed form the SOPO decomposition reproduces.","marker":"[41]"},{"why":"supplies the family of non-Clifford diagonal unitaries $U_\\nu$ whose non-polynomial phase is needed for the non-classical correlations.","marker":"[53]"},{"why":"shows that stabilizer states admit classical simulation, which is why the phase $\\nu$ cannot be quadratic or linear for the self-test to work.","marker":"[54]"}],"fun_headline_variants":["Robust self-test for maximal entanglement in all finite dimensions","Bell test certifies any-dimension maximal entanglement, noise-tolerant","Generalizing CHSH: robust self-testing of all maximally entangled states","Robust finite-dimensional self-testing of all maximally entangled states","Robust Bell test for maximal entanglement in any dimension"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The twisted-commutation step assumes that the linear system fixing the products $B_k B_l |\\psi\\rangle$ has a unique solution; the paper's row-rank argument does not by itself prove uniqueness, since the matrix involved has more columns than rows, so an additional linear-independence argument is needed.","fun_headline_variants_meta":{"raw":{"variants":["Robust self-test for maximal entanglement in all finite dimensions","Bell test certifies any-dimension maximal entanglement, noise-tolerant","Generalizing CHSH: robust self-testing of all maximally entangled states","Robust finite-dimensional self-testing of all maximally entangled states","Robust Bell test for maximal entanglement in any dimension"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000858,"raw_usage":{"total_tokens":3816,"prompt_tokens":1127,"completion_tokens":2689,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":743,"completion_tokens_details":{"reasoning_tokens":2602}},"tokens_in":743,"tokens_out":2689,"duration_ms":22419,"temperature":1.0,"reasoning_tokens":2602,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T05:53:53.443068+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For $d=5$, compute the kernel of the matrix $M$ whose rows are the products $g(j,k,1)g(j,l,1)$; if any nonzero vector $c(n)$ satisfies $M c(n) = g(j,n,2)$, then the uniqueness claim in Lemma 1 fails and the twisted commutation relations are not secured by the proof as written.","supporting_citations":[{"cited_title":"In Ap- pendix A, we calculate explicitly that χ[(U ⊗ 1) |Φ⟩](x1,z1),(x2,z2) = 1 d δx1=x2 X s∈Zd ω(z1+z2)s+νs+2−1 x1 −νs−2−1 x1","cited_arxiv_id":null,"evidence_quote":"supplies the qubit local-isometry construction whose qudit generalization extracts the state and observables in Theorem 1."},{"cited_title":"Self-testing quantum systems of arbitrary local dimension with minimal number of measurements","cited_arxiv_id":null,"evidence_quote":"is the earlier arbitrary-dimension self-test without robustness that the present protocol upgrades with explicit error bounds."},{"cited_title":"Quan- tum bounds for compiled xor games and d-outcome chsh games, 2024","cited_arxiv_id":null,"evidence_quote":"supplies the family of non-Clifford diagonal unitaries $U_\\nu$ whose non-polynomial phase is needed for the non-classical correlations."},{"cited_title":"Constant-sized correlations are sufficient to self-test maximally entangled states with unbounded di- mension","cited_arxiv_id":null,"evidence_quote":"shows that stabilizer states admit classical simulation, which is why the phase $\\nu$ cannot be quadratic or linear for the self-test to work."}],"review_version":1}