{"id":"d1ab69c3-d86a-44de-82f7-a354742e0071","arxiv_id":"2507.02829","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Splitting a noisy GHZ sensor array into smaller independent GHZ sub-ensembles, with optimal sub-ensemble size set by the inverse error rate, maximizes the quantum Fisher information.","lead":"A quantum sensing paper shows that when entanglement generation is imperfect, it is better to split a large entangled sensor array into several smaller entangled groups rather than entangling everything together. The authors derive simple formulas for the optimal group size under different types of noise, giving experimentalists a clear rule for allocating sensors.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The closed-form optimal partition rule assumes the same per-gate fidelity k applies to every sub-ensemble and that sub-ensemble preparation errors are independent; correlated or crosstalk-dependent preparation noise would invalidate Eqs. (7) and (10).","rationale":"The paper's internal algebra is coherent: Eq. (4) correctly reduces to the known QFI for a depolarized GHZ state, the derivative calculations in Appendices A-C are internally consistent, and the approximate optimality conditions m + n ln k = 0 and 2m + n ln k = 0 follow from the stated model. The numerical figures, especially the integer-programming comparisons in Figs. 4 and 7, support the closed-form approximations within the model. The reader's conditional verdict is therefore appropriate. The most load-bearing threat is not the algebra but the noise-model assumption, because the headline formula takes a single input parameter k and assumes that parameter transfers unchanged from monolithic to partitioned preparation and that the sub-ensembles are statistically independent. The proposed gate-level simulation isolates this assumption by introducing crosstalk and correlated noise in a controlled way. If the assumption holds, the paper's closed-form guidance is a useful practical rule; if it fails, the quantitative claims are limited to the idealized independent-error model and the exponential advantage expressed in Eq. (10) may not be realized. The paper acknowledges the model is not unique, but does not quantify the sensitivity of the central result to violations of that model, which is exactly what a conditional acceptance should require.","tokens_in":47351,"tokens_out":11953,"duration_ms":152258,"concrete_test":"Simulate a fixed n = 100 sensor system on a gate-level noise model: prepare GHZ states with independent depolarizing two-qubit gate error epsilon, calibrate k from a single two-qubit GHZ, then repeat with (i) a crosstalk-dependent per-gate error epsilon_m = epsilon(1 + alpha(m-1)) for m concurrently prepared sub-ensembles and (ii) correlated Z noise shared across sub-ensembles. Enumerate integer m and compute the exact QFI of the full state; compare the optimal integer m with round(-n ln k) and the QFI ratio with Eq. (10). If the optimal m shifts by more than one sub-ensemble or the QFI ratio changes by more than 20% for modest alpha, the closed-form rule is not robust to realistic crosstalk.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Equations (7) and (10) are the paper's headline results. They follow from the state-preparation model in Eqs. (1)-(2): each n-qubit GHZ preparation has fidelity F(n) = k^{n-1} F, the same k applies to every sub-ensemble, and sub-ensemble errors are independent so that F_total = m F(F, k, n/m) in Eq. (5). The least secure link is the assumption that in a parallel m-partition preparation the per-gate error parameter k is independent of m and that no correlations build up between sub-ensembles. Real parallel GHZ generation shares control hardware, can have crosstalk between concurrently driven entangling gates, and may suffer from global noise; then the effective k for an ensemble of size n/m can differ from the monolithic k, and the joint state is not a product of independent depolarized GHZ states. In that case the derivative with respect to m in Appendix A2, approximated as dF/dm ≈ -F n^2 k^{n/m-1}(m + n ln k)/m^3, is not the correct optimality condition, m* ≈ -n ln k can fail, and the exponential ratio in Eq. (10) is not guaranteed. The paper explicitly acknowledges the model is 'not unique' in Sec. II.A, but it does not bound the error incurred when the model is violated, so the universality of the closed-form rule is unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies phase estimation with GHZ states under state-preparation noise, particle loss, and dephasing. It models an n-qubit GHZ preparation as a depolarized state with fidelity F(n)=k^{n-1}F (Eq. 2) and proposes partitioning the n sensors into m independent sub-ensembles of size n/m. The main results are closed-form approximations for the optimal partition number, m* ≈ -n ln k (Eq. 7), and for the QFI ratio between optimally partitioned and monolithic strategies, F(m*)/F(monolithic) ≈ -(1/e) k^{-n}/(n ln k) (Eq. 10). The paper also derives analogous formulas with particle loss and dephasing, studies QFI dynamics, analyzes the sequential scheme, and compares noisy GHZ Ramsey spectroscopy with spin-squeezed states.","tokens_in":47646,"tokens_out":14912,"duration_ms":156051,"significance":"If correct, the static partitioning rule m* ≈ n(1-k) provides a simple, practical guide for resource allocation in entanglement-enhanced sensing under realistic preparation errors. The paper's analytic derivations are explicit and the approximations are checked against exact integer programming in Figs. 4, 6, and 7, which is a notable strength. The closed forms for QFI with loss and dephasing extend the authors' earlier formula. However, the headline results are conditional on a specific noise model and on unquantified asymptotic approximations, and the sequential-scheme optimization in Sec. VI contains an internal inconsistency; these issues require revision before the claims can be considered fully established.","major_comments":[{"comment":"The central optimal-partition rule m*≈-n ln k and the exponential ratio Eq. (10) rest on the assumption, stated after Eq. (5), that sub-ensembles can be independently prepared with uniform errors and the same per-gate fidelity k for all m. The text acknowledges in Sec. II.A that the depolarized/exponential-fidelity model is 'not unique,' but it does not quantify the error incurred when the assumption fails. If parallel GHZ preparation shares control hardware, crosstalk or global noise can make the effective k depend on m or correlate errors across sub-ensembles; then the derivative condition dF/dm≈-F n^2 k^{n/m-1}(m+n ln k)/m^3 in Appendix A.2 is not the correct optimality condition and m* can differ from Eq. (7). The paper should either justify the independence microscopically for a concrete platform or provide a robustness bound, for example through a numerical study with correlated noise or an m-dependent k.","section":"Sec. II.A, Eqs. (1)-(2); Sec. III.C, Eq. (7)"},{"comment":"The closed-form expressions for n* and m* are obtained by dropping 'exponentially suppressed terms' (e.g., Eqs. (A1)-(A3) and (A5)-(A8)) without a bound on the remainder. The stated validity conditions, 'k, F close to 1' and 'n, n/m not small,' are not made quantitative, and the figures validate only isolated parameter points (k=0.99 in Fig. 4; k=0.995 in Fig. 7). Since the optimal partition numbers are the main quantitative output, a parameter-region validity statement or a numerical sweep over (k, p, q, n) is needed to establish the range in which Eqs. (7), (21), and (39) can be used.","section":"Appendix A; Eqs. (6)-(7), (19)-(21), (37)-(39)"},{"comment":"The sequential-scheme optimization is not consistent when t can be chosen freely. For fixed m, the best evolution time is t̃*=m/[n(2γ+η)] from Eq. (54); substituting this into Eq. (57) gives m̃*(t̃*)=m-n ln k > m, so the stationary-point equations for t and m cannot be satisfied simultaneously. In fact, the maximal QFI per unit time at the t-optimal point, Eq. (55), is monotone increasing in m, so the unconstrained joint optimum is m=n (separable sensors), not m≈-n ln k. The recommendation that m and n should be matched by m≈-n ln k is therefore valid only when the evolution time per cycle is bounded by tth; the text should state this restriction explicitly and the claimed advantage of partitioning in the sequential scheme should be framed accordingly.","section":"Sec. VI.C, Eqs. (54)-(58)"}],"minor_comments":[{"comment":"The symbol F is used both for the initialization fidelity (Eq. (2)) and for the quantum Fisher information (e.g., Eq. (4)); this is confusing, especially in Eqs. (8)-(11) and (24)-(26). Please use a different symbol, such as a script F, for one of the two quantities.","section":"Throughout"},{"comment":"There are several typographical errors that should be corrected: 'beacuse' (end of Sec. III.E), 'wiothout' and 'yileds' (Sec. IV.C), 'ocurred' (Sec. IV.A), 'costist' (Sec. V.A), and 'highest highest maximum' (Sec. VI.C). Also, in Sec. VI.C, 'm < k−1' should be 'm < l−1'.","section":"Secs. III.E, IV.A, IV.C, V.A, VI.C"},{"comment":"The asymptotic performance in Eq. (58), I → F nT/[e(2γ+η)], appears to omit a factor 1/k relative to Eq. (55), which gives F̃_peak ≈ F n k^{n/m}/[e k (2γ+η)]; in the limit k^{n/m}→1 the result should be F nT/[e k (2γ+η)] unless an additional approximation is intended. Please check and clarify.","section":"Sec. VI.C, Eq. (58)"},{"comment":"The QFI formula in Eq. (3) and its evaluation in Eq. (4) are taken from the authors' prior work [59]. A brief self-contained derivation in an appendix would make the paper more readable and would reduce the dependence on an external reference for a central quantity.","section":"Sec. II.D, Eq. (3)"}],"recommendation":"major_revision","confidential_remarks":"The paper's main new contribution relative to refs. [56-58] is the closed-form QFI expressions for loss and dephasing and the explicit dynamic analysis; the scaling m* ~ 1/(1-k) itself is not new. The referee should ensure that the authors position their contribution accordingly. The sequential-scheme inconsistency in Sec. VI.C is a substantive technical issue that should be addressed before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a useful analytical add-on to a known idea. The paper gives clean closed-form rules (m* ≈ -n ln k, with extensions for loss and dephasing) for partitioning GHZ states under depolarizing preparation noise, plus a genuinely new treatment of time-dependent QFI and the sequential scheme. If you work on ensemble sensors, the rule is worth having.\n\nWhat is actually new: the specific closed forms for optimal partition number, the QFI ratios (Eqs. 7, 10, 21, 27, 39, 40), and the dynamics in Sec. VI. The paper is honest that the qualitative inverse-error-rate scaling already appears in [56–58]; the contribution is the analytical sharpening. The central QFI expression (Eq. 4) is taken from the authors' own prior work [59], which is not a flaw by itself, but it means the partitioning principle is largely confirmed rather than discovered.\n\nWhere the soft spots are: the whole edifice rests on the state preparation model F(n) = k^{n-1}F, with the same k applied independently to each sub-ensemble. The stress-test concern about correlated or crosstalk-dependent preparation errors is fair. The authors acknowledge in Sec. II.A that the model is not unique, but they do not quantify how sensitive m* and the exponential advantage are to violations. That is the weakest link. The Appendix A derivations also drop exponentially suppressed terms without explicit error bounds; the agreement with exact integer programming in Figs. 4, 6, and 7 is the practical check and it holds well, but a bound would make the approximations load-bearing rather than suggestive. Minor: the no-loss-detection cases reduce to implicit equations, so the phrase “closed-form” is over-broad for those cases.\n\nWho it is for: experimental groups using GHZ-type states in clocks, NV ensembles, or atom arrays, who can take Eq. (7) directly into resource planning. The comparison with noiseless one-axis twisting is rough, but the authors flag that it is a strong benchmark.\n\nMy verdict: send it to peer review. It is not a breakthrough, but it is compact, mostly correct, and the closed-form rules will be cited. Request explicit error bounds on the approximations and a short discussion of correlated preparation errors; neither request should block publication.","headline":"Useful closed-form resource-allocation rules for noisy GHZ sensing, built on a model the authors state plainly; the formulas verify against numerics and deserve a referee, with requests for error bounds and a correlated-noise caveat.","tokens_in":48203,"tokens_out":1490,"would_cite":true,"duration_ms":21149,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"For a fixed number of noisy sensors, the best strategy is to partition them into m* ≈ −n ln k independent GHZ sub-ensembles, which beats one monolithic GHZ state by an exponential quantum Fisher information advantage.","keywords":["GHZ states","quantum Fisher information","entanglement-enhanced sensing","phase estimation","depolarizing noise","particle loss","dephasing","resource allocation"],"falsifier":"Measure the QFI (or estimation variance) of a fixed n-sensor array as a function of partition number m, using an independently characterized entangling-gate fidelity k; Eq. (7) predicts a peak at $m^*\\approx -n\\ln k$. If the measured peak is elsewhere, or if the measured GHZ fidelity as a function of n is not $F(n)=k^{n-1}F$, the derived closed-form partition rules do not hold.","tokens_in":47127,"feed_emoji":"🧩","tokens_out":7217,"duration_ms":79544,"temperature":0.7,"pith_summary":"The paper proposes and analyzes a resource-allocation strategy for entanglement-enhanced phase estimation under realistic noise: instead of preparing one GHZ state across all n sensors, split the sensors into m independent, smaller GHZ sub-ensembles. For state-preparation errors modeled by an entangling-gate fidelity k, the optimal number of sub-ensembles is $m^*\\approx -n\\ln k$, so each sub-ensemble holds about $1/(1-k)$ sensors. This partitioning turns what would otherwise be a noise-limited monolithic sensor into a sensor with an exponentially larger quantum Fisher information, by a factor roughly $-(1/e)\\,k^{-n}/(n\\ln k)$. The same logic is extended to particle loss, where $k$ is replaced by $kp$, and to dephasing, where $k$ is replaced by $k(2q-1)^2$, and it is applied to time-dependent and sequential sensing protocols. If correct, the results reduce noisy entanglement-enhanced sensing design to simple closed-form rules.","feed_headline":"Splitting a GHZ sensor into small pieces yields exponential gains","feed_subtitle":"The optimal sensor never entangles all qubits; a simple formula picks the best partition size from the error rate.","key_machinery":"The argument rests on the depolarized-GHZ preparation model $\\rho = V(n)|\\mathrm{GHZ}_n\\rangle\\langle\\mathrm{GHZ}_n| + [1-V(n)] I_n/2^n$ with fidelity $F(n)=k^{n-1}F$, together with the closed-form quantum Fisher information for GHZ-diagonal states under z-rotation encoding. From that, the paper derives the QFI of a monolithic n-qubit GHZ sensor and uses additivity of the QFI under tensor products to write the partitioned QFI as $m\\,F(F,k,n/m)$. The optimal partition number $m^*$ is then obtained by setting $\\partial F/\\partial m = 0$ and keeping leading-order terms for $k,F$ close to 1, giving $m^*\\approx -n\\ln k$. Inequality-based concavity of $F$ as a function of n is used to justify equal partition sizes.","core_discovery":"The central claim is that, with a fixed total number n of sensor qubits and a state-preparation fidelity that decays as $F(n)=k^{n-1}F$, the quantum Fisher information is maximized not by entangling all n qubits but by partitioning them into $m^*\\approx -n\\ln k$ equal GHZ sub-ensembles. At this partition, the ratio of the optimally partitioned QFI to the monolithic QFI is approximately $-(1/e)\\,k^{-n}/(n\\ln k)$, an exponential advantage in n. Equivalently, the optimal sub-ensemble size $n/m^*\\approx -1/\\ln k$ is set only by the error rate, so additional sensors should be used to create more parallel sub-ensembles rather than larger GHZ states. With sensor loss probability p per qubit, the rule becomes $m^*\\approx -n\\ln(kp)$; with dephasing probability q of no phase flip, $m^*\\approx -n[\\ln k + 2\\ln(2q-1)]$. The same partitioning viewpoint also yields closed-form QFI dynamics, peak times, optimal sensing bandwidth in the sequential scheme, and a parameter regime where noisy partitioned GHZ states outperform noiseless one-axis-twisting spin-squeezed states.","pith_inferences":["A broader reading is that the exponential-in-size fidelity penalty forces a crossover: Heisenberg scaling is only available up to a size set by the per-gate error, after which parallelism replaces entanglement as the optimal use of resources. This crossover could be tested by sweeping m on any platform with calibrated gate fidelity.","The same optimization logic should apply to other entangled probe families, such as W states or graph states, whenever their preparation fidelity decays exponentially with system size; verifying that would require a closed-form QFI analogous to Eq. (4) for each family.","Because the optimal sub-ensemble size is independent of n, the result suggests a calibration recipe: measure k once, set each sub-ensemble size to about $1/(1-k)$, then add as many sub-ensembles as sensors allow. This also provides a direct experimental test of Eq. (7) by measuring estimation variance versus m.","If preparation errors are correlated across sub-ensembles, the additivity assumption $mF(F,k,n/m)$ breaks down and the optimal partition would shift; quantifying that shift is a natural next step beyond the paper's independent-error model."],"forward_implications":["For any sensor array with entangling-gate fidelity k, the optimal GHZ sub-ensemble contains about $-1/\\ln k$ sensors; adding more sensors should create more sub-ensembles, not larger ones.","The QFI of an optimally partitioned sensor grows like $k^{-n}/(n\\ln k)$, exponentially in the total sensor count, whereas a monolithic GHZ sensor's useful size is capped near $-2/\\ln k$.","Particle loss and dephasing enter the same rule through effective fidelities $kp$ and $k(2q-1)^2$, so the optimal partition can be read off from independently measured error parameters.","In the sequential scheme, the optimal number of sub-ensembles also grows with the allowed evolution time, and partitioning raises the short-time QFI accumulation rate up to $m\\approx -n\\ln k$.","Noisy partitioned GHZ Ramsey spectroscopy can beat noiseless one-axis-twisting spin-squeezed Ramsey spectroscopy once preparation fidelity is high enough."],"supporting_citations":[{"why":"Supplies the exponential fidelity decay model $F(n)=k^{n-1}F$ and the QFI formula for GHZ-diagonal states.","marker":"[59]"},{"why":"Defines the sequential sensing scheme and the optimality conditions used in the time-dependent comparison.","marker":"[35]"},{"why":"Gives additivity of quantum Fisher information under tensor products, the step that turns the monolithic QFI into $mF(F,k,n/m)$.","marker":"[79]"},{"why":"Recent theoretical result cited as corroboration that optimal sub-ensemble size scales with inverse error rate.","marker":"[56]"},{"why":"Earlier numerical result for optimal entangled probe size under noise, used as corroboration for the scaling found here.","marker":"[57]"},{"why":"Tensor-network numerical approach supporting the inverse-error-rate scaling of the optimal partition.","marker":"[58]"},{"why":"Provides the Wineland spin-squeezing parameter used as the basis for the Ramsey spectroscopy comparison.","marker":"[23]"},{"why":"Provides the Kitagawa-Ueda squeezing parameter used as a lower bound in the spin-squeezed comparison.","marker":"[24]"},{"why":"Supplies the closed-form expression for the one-axis-twisting spin-squeezing parameter used to compare against noisy GHZ states.","marker":"[85]"}],"fun_headline_variants":["Partitioning GHZ states yields exponential gain","Split GHZ into small sets to beat noise","Optimal GHZ never entangles all qubits","Shrink GHZ for exponential QFI boost","Partition beats monolithic GHZ under noise"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that preparing an n-qubit GHZ state yields fidelity $F(n)=k^{n-1}F$ with a single per-gate parameter k that also applies identically to every sub-ensemble, and that preparation errors are independent across sub-ensembles.","fun_headline_variants_meta":{"raw":{"variants":["Partitioning GHZ states yields exponential gain","Split GHZ into small sets to beat noise","Optimal GHZ never entangles all qubits","Shrink GHZ for exponential QFI boost","Partition beats monolithic GHZ under noise"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000255,"raw_usage":{"total_tokens":1597,"prompt_tokens":998,"completion_tokens":599,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":614,"completion_tokens_details":{"reasoning_tokens":528}},"tokens_in":614,"tokens_out":599,"duration_ms":6597,"temperature":1.0,"reasoning_tokens":528,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T20:20:19.707493+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the QFI (or estimation variance) of a fixed n-sensor array as a function of partition number m, using an independently characterized entangling-gate fidelity k; Eq. (7) predicts a peak at $m^*\\approx -n\\ln k$. If the measured peak is elsewhere, or if the measured GHZ fidelity as a function of n is not $F(n)=k^{n-1}F$, the derived closed-form partition rules do not hold.","supporting_citations":[{"cited_title":"Multi-qubit gates and Schr\\\"odinger cat states in an optical clock","cited_arxiv_id":"2402.16289","evidence_quote":"Supplies the exponential fidelity decay model $F(n)=k^{n-1}F$ and the QFI formula for GHZ-diagonal states."},{"cited_title":"Fast and Accurate Greenberger-Horne-Zeilinger Encoding Using All-to-all Interactions","cited_arxiv_id":"2406.10336","evidence_quote":"Gives additivity of quantum Fisher information under tensor products, the step that turns the monolithic QFI into $mF(F,k,n/m)$."},{"cited_title":"Schioppo, R","cited_arxiv_id":null,"evidence_quote":"Earlier numerical result for optimal entangled probe size under noise, used as corroboration for the scaling found here."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Tensor-network numerical approach supporting the inverse-error-rate scaling of the optimal partition."},{"cited_title":"Petz and C","cited_arxiv_id":null,"evidence_quote":"Supplies the closed-form expression for the one-axis-twisting spin-squeezing parameter used to compare against noisy GHZ states."}],"review_version":1}