{"id":"a1dd6e77-6079-4e05-947f-ea386d097944","arxiv_id":"2608.06888","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For nuisance-aware quantum sensor networks, global effective Fisher information is capped by the weakest weighted local capacity, and an exact phase map shows when correlations help, saturate, or overcorrelate.","lead":"This paper derives an exact map of how quantum correlations distribute estimation information between local sensors and a global target in a quantum sensor network, and identifies when extra correlations hurt both. It also defines a privacy condition under which every local parameter is hidden while the desired global quantity stays measurable, with explicit qubit states realizing it.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Bottleneck-saturating QFIM designs are not proven physically realizable; the paper's own Discussion leaves physical-realizability and measurement-compatibility as separate conditions, so the engineering conclusions rest on an unproved step.","rationale":"The QFIM-level mathematics is internally consistent: Theorem 1's Cauchy–Schwarz proof is correct, the phase map Eq. (6) follows by block inversion, and Theorem 3's range criterion is the standard pseudoinverse estimability condition. I find no algebraic error in the recursive construction, and the qubit privacy states in Sec. VII are explicit and checkable. The load-bearing gap is the passage from 'a PSD matrix with fixed diagonals' to 'a physical probe with local generators and the same local capacities'. For pure states Q_ij = 4Cov(H_i,H_j), so this is a covariance-realizability problem; PSD alone is necessary but not sufficient when the local generators have fixed discrete spectra, because commuting ±1 observables obey additional moment constraints. The paper supplies explicit physical states only for privacy (Sec. VII) and not for the bottleneck-saturating Eq. (140) family, and its own Discussion concedes that implementations must additionally satisfy physical-realizability and measurement-compatibility conditions. That is an honest limitation, but it is exactly the missing link for the headline 'engineering optimal network states' conclusion. The verdict should remain CONDITIONAL: the theorems hold as QFIM-level results, while the engineering claim should be read as conditional on a realizability construction that is not in the paper. The proposed feasibility test would settle whether qubit platforms can support the canonical optimal family, and a CV realization would clarify how much additional squeezing or energy is required for fixed q_i.","tokens_in":29979,"tokens_out":23912,"duration_ms":275351,"concrete_test":"Test the canonical family in Eq. (140) for N=3 qubit sensors with H_i = Z_i and random admissible (q_i, w_i). For each instance, solve the linear feasibility problem over distributions p(z) on {±1}^3 with constraints p ≥ 0, ∑p = 1, and second moments ∑p(z) z_i z_j = m_i m_j + Q*_{ij}/4, where m_i are chosen so that Var(Z_i) = q_i/4. If any instance is infeasible, the canonical optimal QFIM is not realizable by N-qubit sensors, and the engineering claim must be restricted to platforms with a proven realization. If all sampled instances are feasible, construct the explicit probe state and verify numerically that its QFIM equals Q* and E_w = κ.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central results are statements about abstract positive-semidefinite QFIMs. Theorem 1's tightness and the recursive optimal family construct Q* = κ w w^T + diag(q_i − κ w_i^2) in Eq. (140), asserting only that this matrix has the prescribed diagonals and E_w = κ. For the paper's engineering conclusion to follow, this family must be realizable as a QFIM of a physical probe state with local generators H_i and unchanged local capacities q_i. For pure states Q_ij = 4 Cov(H_i,H_j), so the problem is covariance realizability, not just matrix positivity. On fixed discrete-variable platforms, PSD is necessary but not sufficient: for qubit sensors, the values of commuting ±1 observables satisfy moment-polytope constraints beyond positive semidefiniteness. On continuous-variable platforms, many PSD covariance matrices can be realized by Gaussian squeezing, but the required squeezing energy is not counted in q_i. The paper supplies explicit physical states only for the privacy constructions in Sec. VII, not for the bottleneck-saturating family; the Discussion states that implementations 'must additionally satisfy physical-realizability and measurement-compatibility conditions.' Thus the QFIM-level theorems stand, but the headline methodology for engineering optimal network states rests on an unproven realizability step.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies distributed quantum sensing of a single weighted linear combination θ_w = w^T θ when all N local parameters are unknown. It works with the effective quantum Fisher information E_w = (w^T Q^{-1} w)^{-1} (with Moore–Penrose pseudoinverse in singular cases) and the local EQFIs E_i. The main results are: Theorem 1, a \"barrel-effect\" bottleneck E_w ≤ min_i q_i/w_i^2, tight at the QFIM level; Theorem 2, an exact node–subnetwork phase map that organizes correlations into trade-off, overcorrelated, and matched-singular regimes; a recursive QFIM construction that saturates the bottleneck; a permutation-invariant example showing that the projected QFI can badly overestimate nuisance-aware precision; and Theorem 3, an intrinsic-local-privacy condition characterized by e_i ∉ range(Q) for all i while w ∈ range(Q), together with explicit qubit state constructions for ranks 1 to N−1. The analytical derivations are mostly clean and self-contained, and the privacy constructions are explicit and checkable.","tokens_in":30362,"tokens_out":27424,"duration_ms":296137,"significance":"The QFIM-level mathematics is sound: Theorem 1 follows from a Cauchy–Schwarz inequality, Eq. (6) from exact block inversion, and Theorem 3 from the Moore–Penrose estimability criterion. The paper credits Ref. [2] for the node-wise inequality and provides explicit qubit states for the privacy construction; no free parameters or numerical fitting enter the derivations. The intrinsic-local-privacy notion is a useful generalization of functional privacy. The main unresolved issue is the gap between abstract QFIM designs and physically realizable probe states, which limits the operational scope of the engineering claims.","major_comments":[{"comment":"The bottleneck-saturating family Q* = κ w w^T + diag(q_i − κ w_i^2) is constructed as an abstract positive-semidefinite QFIM, but the paper's stated goal of \"engineering optimal network states\" requires that this matrix be realizable as the QFIM of a physical probe state with local generators H_i and unchanged local capacities q_i. For a pure probe, Q_ij = 4Cov(H_i,H_j), so the existence of a probe is a covariance-realizability problem, not merely a matrix-positivity problem. Positive semidefiniteness is not sufficient on discrete-variable platforms, where covariance matrices of commuting ±1 observables must satisfy moment-polytope constraints, and on continuous-variable platforms a Gaussian realization may require squeezing energy that is not counted in q_i. Explicit physical states are supplied only for the privacy constructions in Sec. VII, not for the saturating QFIM family. Since the abstract and Section IV present this as a methodology for engineering optimal network states, this gap is load-bearing. I ask the authors either to provide explicit state families (or a general realizability theorem) for the saturating QFIMs, or to explicitly reframe the contribution as QFIM-level bounds and designs, leaving physical state engineering as an open condition.","section":"Sec. IV, Eq. (140); Discussion"},{"comment":"The text alternates between treating E_w and E_i as precision bounds and describing them as the attainable precision or as an actual metrological penalty, for example in the phrase \"additional correlation becomes a metrological penalty rather than a resource\" and in the abstract's claim that excessive correlations \"actively degrade both local and global performance.\" The quantum Cramér–Rao bound is not always saturable for mixed-state multiparameter models, and the attainability conditions are postponed to the final Discussion paragraph. The main text should state explicitly which statements are about Fisher information and which are about achievable variances, and should give the sufficient attainability conditions (for instance, pure states with commuting generators) at the point where E_w is first used.","section":"Main text, Local–Global Phase Map section; Discussion"},{"comment":"Theorem 3 states an if-and-only-if phase-map characterization, Eq. (14), for singular QFIMs, with the singular case handled through a \"nonsingular limiting sequence\" explanation. The supplement does not supply a complete proof that the regularized limits in Eq. (182) are equivalent to e_i ∉ range(Q) for every i when the complementary submatrix Q_{\\not i} is itself singular or when w_i = 0. Since Theorem 3 is central to the privacy claim, either provide the limiting proof or state Eq. (14) as an interpretation valid under additional regularity assumptions.","section":"Sec. VI.A, Theorem 3"}],"minor_comments":[{"comment":"The displayed formula for eE_w in Eq. (12) is missing parentheses; it should read eE_w = 1 + [(N−1)γ(eα_S − γ)]/(1 + γR). Please correct the typesetting.","section":"Main text Eq. (12)"},{"comment":"The tightness witness Q* is asserted to satisfy E_w = κ in one line. When multiple bottleneck nodes make diag(q_i − κ w_i^2) singular, this is not immediate; please add a short pseudoinverse or block-inversion computation.","section":"Theorem 1 proof"},{"comment":"The symbol c^(k) is used both for the correlation vector in Eq. (9) and for the scalar amplitude in Eq. (123) of the supplement; please use distinct notation for the vector and its magnitude.","section":"Sec. IV and Supplement Sec. IV.B"},{"comment":"The claim E_w = N^2/r for the general qubit construction is stated without derivation. Because the vectors v^(j) are not orthogonal for N > 4, a short pseudoinverse computation would help the reader verify the result.","section":"Sec. VII.B"},{"comment":"The Discussion's caveat that implementations must additionally satisfy physical-realizability and measurement-compatibility conditions should be introduced earlier, in Section IV, so that the recursive QFIM construction is not read as an explicit state-engineering recipe.","section":"Discussion"}],"recommendation":"major_revision","confidential_remarks":"The main technical content is sound; my recommendation hinges on the authors' response to the physical-realizability point. If they can supply explicit saturating states or clearly downgrade the engineering claims, the paper could be acceptable. I see no inappropriate citation behavior: Ref. [6] is the authors' own preprint, cited as context, and the priority to Ref. [2] for the node-wise bound is acknowledged."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, here is the short version. The paper is honest about what is new: the bottleneck bound in Theorem 1 is explicitly credited to Eldredge et al. [2]. What is genuinely new is the exact node–subnetwork phase map, the overcorrelated regime, the recursive construction of saturating QFIMs, and the intrinsic local privacy characterization with explicit qubit states at every rank. The math at the QFIM level is clean: Cauchy-Schwarz for the bottleneck, block-inversion for the phase map, and pseudoinverse range conditions for estimability and privacy. No fitted parameters, no circularity.\n\nThe soft spot is exactly the one the stress-test flags: the optimal QFIM family is not shown to be realizable by physical probe states with local generators and unchanged local capacities. The canonical Q* is proven positive semidefinite, but positive semidefiniteness is not the same as being a covariance matrix of local observables on an N-partite state. For qubit sensors there are moment-polytope constraints; for CV platforms the squeezing energy would eat into the resource budget. The paper's own Discussion says implementations \"must additionally satisfy physical-realizability and measurement-compatibility conditions,\" so this is a stated limitation, not a hidden one. Still, the phrase \"principled methodology for engineering optimal network states\" overreaches relative to what is proved. The QFIM-level inequalities and the phase map stand on their own.\n\nOne minor presentation issue: Eq. (12) in the main text is ambiguously formatted; the supplementary version is consistent.\n\nVerdict: this deserves a serious referee. The gap between QFIM design and physical realization is real, but the authors are honest about it, and the phase-map framework is likely to be useful for the distributed sensing community. I would cite this for the privacy construction and the trade-off/overcorrelated picture, and I would bring it to a reading group session.","headline":"Clean QFIM-level theory with genuinely new map and privacy results; the engineering claim overreaches slightly because the optimal designs are not shown to be physically realizable.","tokens_in":30789,"tokens_out":2010,"would_cite":true,"duration_ms":21627,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45","81P50"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves a 'barrel-effect' ceiling for quantum sensor networks: with all local parameters unknown, global precision is capped by the weakest weighted local sensing capacity, and an exact phase map shows when correlations help or…","keywords":["quantum sensor networks","effective quantum Fisher information","nuisance parameters","quantum Fisher information matrix","distributed quantum sensing","local privacy","Heisenberg scaling","correlation geometry"],"falsifier":"Take a two-node network with equal local capacities and uniform target weights, and sweep the off-diagonal QFIM correlation from zero toward its maximal value, reconstructing $E_1$ and $E_w$ from quantum state tomography of the probe. The phase map predicts $E_w$ rises to a finite maximum at $\\mu^* = \\min\\{s/r, r/s\\}$ and then falls into the overcorrelated branch while $E_1$ decreases monotonically, whereas the projected QFI $F_w$ keeps increasing to the singular endpoint. Observing monotone growth of $E_w$ all the way to the endpoint, or no decline past $\\mu^*$, would refute the paper's overcorrelated-regime claim.","tokens_in":29756,"feed_emoji":"📡","tokens_out":9738,"duration_ms":99069,"temperature":0.7,"pith_summary":"Quantum sensor networks usually aim to estimate a single weighted linear combination of many unknown local parameters; the remaining $N-1$ combinations act as nuisance parameters. This paper establishes that, once these nuisance parameters are properly accounted for, the global effective quantum Fisher information obeys $E_w \\le \\min_{i: w_i\\neq 0} q_i/w_i^2$, so the weakest weighted local sensing capacity is a fundamental precision ceiling that is tight at the QFIM level. It then derives an exact local–global phase map, $E_i = q_i(1-\\mu_i^2)$ and $E_w = \\frac{1}{A_{\\not i}}[1+(s_i-r_i\\mu_i)^2/(1-\\mu_i^2)]^{-1}$, showing that correlations first transfer information from local to global targets and, beyond an optimal strength, degrade both—an overcorrelated regime. The same map yields QFIM designs that saturate the ceiling and identifies intrinsic local privacy, the condition under which every local parameter is unidentifiable while the global combination remains estimable. The upshot is a closed-form solution to the resource-constrained network design problem: given local capacities and a target, the attainable precision and the required correlation geometry are known exactly.","feed_headline":"Weakest node caps quantum sensor network precision","feed_subtitle":"An exact phase map shows when correlations help, when they backfire, and how to hide local data.","key_machinery":"The central object is the effective quantum Fisher information, $E_w = (w^T Q^{-1} w)^{-1}$, defined through the inverse quantum Fisher information matrix so that the $N-1$ unwanted parameter directions are integrated out as nuisance parameters, together with its local counterpart $E_i = [(Q^{-1})_{ii}]^{-1} = q_i - c_i^T Q_{\\not i}^{-1} c_i$. The paper's main analytical device is the node–subnetwork phase map, which decomposes the coupling vector $c_i$ into a signed amplitude through three dimensionless coordinates: $\\mu_i$ (normalized correlation strength), $r_i$ (geometric overlap between the correlation axis and the complementary target direction), and $s_i$ (normalized local weight). This map carries the argument because it converts the matrix bottleneck inequality into a one-dimensional function of $\\mu_i$ whose stationary points, break-even point, and singular endpoint classify all correlation regimes and yield the recursive saturating constructions and the privacy boundary.","core_discovery":"The paper's central claim is Theorem 1: for an $N$-node network with fixed diagonal sensing capacities $q_i$ and target $\\theta_w = w^T\\theta$, the global effective quantum Fisher information satisfies $E_w = (w^T Q^{-1} w)^{-1} \\le \\kappa_N = \\min_{i: w_i\\neq 0} q_i/w_i^2$, and this bound is tight at the QFIM level, witnessed by $Q^* = \\kappa_N w w^T + \\mathrm{diag}(q_i - \\kappa_N w_i^2)$. Beneath this ceiling, Theorem 2 gives an exact phase map for a single node coupled to its subnetwork: in terms of the normalized correlation strength $\\mu_i$, target overlap $r_i$, and normalized weight $s_i$, the local and global EQFIs obey $E_i = q_i(1-\\mu_i^2)$ and $E_w = \\frac{1}{A_{\\not i}}[1+(s_i-r_i\\mu_i)^2/(1-\\mu_i^2)]^{-1}$. Along the aligned branch, increasing correlation first transfers information from local to global (the trade-off zone), then past the optimum $\\mu_i^*$ both quantities fall (the overcorrelated zone), and at the matched-singular boundary $|s_i|=r_i$ the local EQFI vanishes while $E_w \\to 1/A_{\\not i}$. Theorem 3 identifies intrinsic local privacy, $E_i=0$ for all $i$ with $E_w>0$, with the simultaneous matched-singular boundary, equivalently $e_i\\notin\\mathrm{range}(Q)$ for all $i$ and $w\\in\\mathrm{range}(Q)$, and supplies explicit qubit probes realizing it at every QFIM rank from $1$ to $N-1$.","pith_inferences":["A reader should not infer from the QFIM-level designs that every optimal matrix is physically preparable: the paper supplies explicit probe states only for the qubit privacy constructions, so the general engineering claim still needs a physical-realizability check for arbitrary $q_i$ and $w_i$.","The phase map suggests a practical tuning rule for real networks: correlation strength should be increased only up to $\\mu_i^*$, since anything beyond that actively worsens both local and global precision; this could be tested by scanning entanglement strength in a small spin-squeezed or continuous-variable network.","Intrinsic local privacy is statistical non-identifiability rather than cryptographic secrecy: it means no locally unbiased estimate of an individual $\\theta_i$ exists, but the collective mode $w^T\\theta$ remains readable, so its privacy value depends on the adversary's allowed measurements and prior knowledge.","The weight-uniformity fragility result implies that matching the probe symmetry to the target weight profile is a design constraint rather than a convenience; small asymmetries in the target weights can convert Heisenberg scaling back to standard-quantum-limit scaling."],"forward_implications":["For any fixed local capacities, no amount of inter-node correlation can push the global EQFI past the weakest weighted node; improving precision requires raising $\\min_i q_i/w_i^2$.","The bound is tight: the canonical QFIM $Q^* = \\kappa_N w w^T + \\mathrm{diag}(q_i - \\kappa_N w_i^2)$ and the more general recursive family achieve $E_w = \\kappa_N$, and with $q_i=\\Theta(1)$ and $|w_i|=\\Theta(1/N)$ this yields Heisenberg scaling $\\Delta^2\\hat{\\theta}_w = \\Theta(N^{-2})$.","In a permutation-invariant network, the projected QFI $F_w$ is monotone in correlation and can greatly exceed the EQFI; the nuisance-aware EQFI peaks at a finite correlation and then declines, and Heisenberg scaling survives only if the target weights approach uniformity at the rate $1-\\tilde{\\alpha}_S = O(1/N)$.","Intrinsic local privacy—zero local EQFI for every node while the global target remains estimable—is achievable for any QFIM rank from $1$ to $N-1$, with the GHZ state as the rank-one endpoint and $E_w = N^2/r$ for the qubit construction."],"supporting_citations":[{"why":"First derived the node-wise inequality $E_w \\le q_i/w_i^2$ that Theorem 1 builds on.","marker":"[2]"},{"why":"Defines quantum state estimation with nuisance parameters, giving the effective quantum Fisher information quantity $E_i$ used for local targets.","marker":"[18]"},{"why":"Develops quantum semiparametric estimation and the effective-information treatment that underlies the nuisance-parameter penalty.","marker":"[20]"},{"why":"Analyzes distributed quantum metrology with linear networks, providing the SQL/Heisenberg comparison and the projected-QFI baseline.","marker":"[3]"},{"why":"Introduces functional privacy, the stronger privacy notion that the paper shows implies its intrinsic local privacy.","marker":"[23]"},{"why":"Supplies the quantum Fisher information matrix formalism and the multiparameter Cramér–Rao bound used throughout.","marker":"[29]"}],"fun_headline_variants":["Too much correlation degrades quantum sensor networks","Quantum sensor networks: weak node caps global precision","Exact geometry shows when quantum correlations help or hurt","Quantum network privacy: hide local data, keep global estimate","Overcorrelation regime: quantum sensor networks lose precision"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is physical realizability: the paper treats arbitrary positive-semidefinite QFIMs with prescribed diagonal capacities—including the canonical saturating $Q^*$ and the recursive family—as achievable network designs, while explicit probe states are supplied only for the qubit privacy constructions.","fun_headline_variants_meta":{"raw":{"variants":["Too much correlation degrades quantum sensor networks","Quantum sensor networks: weak node caps global precision","Exact geometry shows when quantum correlations help or hurt","Quantum network privacy: hide local data, keep global estimate","Overcorrelation regime: quantum sensor networks lose precision"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000766,"raw_usage":{"total_tokens":3479,"prompt_tokens":1109,"completion_tokens":2370,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":725,"completion_tokens_details":{"reasoning_tokens":2296}},"tokens_in":725,"tokens_out":2370,"duration_ms":20885,"temperature":1.0,"reasoning_tokens":2296,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:39:57.926148+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a two-node network with equal local capacities and uniform target weights, and sweep the off-diagonal QFIM correlation from zero toward its maximal value, reconstructing $E_1$ and $E_w$ from quantum state tomography of the probe. The phase map predicts $E_w$ rises to a finite maximum at $\\mu^* = \\min\\{s/r, r/s\\}$ and then falls into the overcorrelated branch while $E_1$ decreases monotonically, whereas the projected QFI $F_w$ keeps increasing to the singular endpoint. Observing monotone growth of $E_w$ all the way to the endpoint, or no decline past $\\mu^*$, would refute the paper's overcorrelated-regime claim.","supporting_citations":[{"cited_title":"Tsang, F","cited_arxiv_id":null,"evidence_quote":"Develops quantum semiparametric estimation and the effective-information treatment that underlies the nuisance-parameter penalty."}],"review_version":2}