{"id":"43f9160c-eb3f-4570-b096-929eccb65f5d","arxiv_id":"2509.08230","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A six-node Mach-Zehnder interferometer network fed by one split squeezed-vacuum state estimates multiple phases with 4.36 dB of noise suppression below the standard quantum limit.","lead":"This paper reports a working network of six Mach-Zehnder interferometers that share a single squeezed light source, measuring multiple phases with 4.36 dB less noise than the standard quantum limit. The result matters because it shows a scalable, reconfigurable design for distributed quantum sensing using far less quantum hardware than prior approaches.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Loss bookkeeping in Eq. (3) lumps pre-MZI distribution loss with output loss; the theory curves and robustness thresholds rely on an unvalidated single-η model.","rationale":"The reader's weakest_assumption correctly identifies the loss model as the load-bearing element. The entire quantitative structure of the paper—the 4.36 dB suppression, the loss thresholds, the SQL-to-HL crossover curve, and the scalability fit with Λ=0.14—passes through Eq. (3). If the single-η bookkeeping is not the correct physical model for where the losses occur, every one of these numbers can shift. I checked whether a more glaring concern exists in the claimed Heisenberg-limit crossover, since the experiment never reaches the n_T^{-2} asymptote; that is a real overstatement, but it is a secondary claim and does not affect the main sub-SQL distributed-sensing result. The loss model is more fundamental because it also underlies the crossover fit. The concern is not that the authors are careless; it is that the Supplement's Eq. (17) is not a derivation of the efficiency-product formula, and the difference between pre-MZI and output losses is not negligible in the loss-threshold regime. The proposed check is analytical and should be quick to perform. If it shows only small conservative shifts, the current verdict and claims can stand. If it shows larger shifts, the theory lines and robustness thresholds need to be redone. Therefore the existing CONDITIONAL verdict is appropriate and no change is needed.","tokens_in":19565,"tokens_out":24864,"duration_ms":318386,"concrete_test":"Analytically re-derive Eq. (3) using the actual loss topology: η_dis on each squeezed mode between the QC and the MZI, η_MZI as balanced loss in both interferometer arms, and η_m per mirror reflection, with independent vacuum modes for each loss port. Compare the resulting phase variance, the K=1 and K=5 loss thresholds for beating the SQL, and the predicted suppression at η=0.88 with the single-η formula. If the thresholds shift by more than ~5 percentage points or the predicted suppression changes by more than ~0.2 dB, the single-η model in Eq. (3) is inadequate and the reported robustness and theory lines require revision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central sensitivity formula, Eq. (3), is Δ²(ν⊤θ) = (e^{-2r} + Λ)/(K n_T) with Λ = 1/η − 1 and η = η_dis η_MZI η_m^(2K−1). This treats every inefficiency as a single beam-splitter loss on the detected output mode. In the actual setup, the distribution efficiency η_dis is physically located before the MZI, on the squeezed-vacuum modes only, while the coherent-state inputs do not pass through the QC and therefore do not suffer η_dis. The Supplement's loss model, Eq. (17), also couples both interferometer arms to one common vacuum mode, which is not a standard independent-loss description. These two models are not equivalent when the squeezed quadrature variance V = e^{-2r} differs from 1. A correct pre-MZI-loss treatment gives, for equal arm losses, Δ² ≈ [η_s V + 1 − η_s]/(η_out n_c) with η_s = η_out η_dis, whereas Eq. (3) gives (V + 1/η_s − 1)/n_c. The difference is small at η ≈ 0.88 but grows as η_dis decreases, so the claimed loss thresholds (η_dis ≳ 65% for K=1, η_dis ≳ 20% for K=5) and the fitted curves in Figs. 3 and 4 depend on which loss topology is assumed. The manuscript does not report an independent check of this bookkeeping, only calibrated efficiencies.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a six-interferometer distributed sensing network in which a single squeezed-vacuum state is split by a linear quantum circuit into six entangled modes, each injected into one input of a Mach-Zehnder interferometer whose second input is a coherent state. Balanced homodyne detection at the dark ports, with joint processing, is used to estimate linear combinations of the six phase shifts. The central analytic formula, Eq. (3), gives the optimized phase variance (e^{-2r}+Λ)/(K n_T) for the high-coherent-intensity regime, with losses collected into Λ=1/η−1. Experimentally, the authors report 4.36±0.35 dB joint noise suppression below the SQL, phase uncertainty near 1.4×10^{-9}, scaling of sensitivity with the number of sensors d, robustness against distribution losses down to η_dis≈65% (K=1) and ≈20% (K=5), and a low-intensity optimized behavior that they describe as a crossover from SQL to the Heisenberg limit. The Supplement derives the sensitivity formula from the input state, BHD measurement, and a single-beam-splitter loss model.","tokens_in":19903,"tokens_out":13395,"duration_ms":166705,"significance":"If the central derivation and loss bookkeeping are correct, the result is significant: it would demonstrate that one nonclassical resource, distributed passively, suffices for sub-SQL multiparameter phase estimation across a scalable interferometric network, with explicit formulas, analytical optimization, and a direct resource comparison against d independent squeezed states. Strengths of the manuscript include a self-contained analytic derivation in the Supplement, a clear experimental layout, systematic scans over photon number, loss, multipass number, number of sensors, and weight vector ν, and honest statements about remaining limitations (e.g., n_s≲1 in the low-intensity runs). The main technical risk is the loss model: the single-efficiency formula in Eq. (3) treats all losses as if they occurred at the output, while the experimentally varied η_dis is physically a pre-MZI loss on the squeezed mode only. This affects the claimed loss thresholds and the fitted theory curves. The abstract additionally overstates the low-intensity demonstration as a crossover to the Heisenberg limit when the data follow a loss-limited curve that asymptotes to Λ/(K n_T).","major_comments":[{"comment":"The loss model used to derive Eq. (3) inserts a single beam splitter with efficiency η on both input quadratures and couples them to the same vacuum mode q0. At the working point θ=0 this is equivalent to an output loss, not to a distribution loss η_dis located on the squeezed mode before the MZI. Since the coherent mode a_j does not pass through the QC, η_dis should not multiply the signal derivative ∂⟨q_j⟩/∂θ. With two-stage loss (η_dis on b_j only, η_out on the measured output), the optimized variance becomes η_dis[V+(1−η_dis η_out)/(η_dis η_out)]/(K n_c) times (Σ|ν_j|)^2, with V=e^{-2r}, rather than [V+1/(η_dis η_out)−1]/(K n_c). For η_out=0.89 and r=0.75, the K=1 sub-SQL threshold becomes η_dis≳16% instead of 65%. Because the loss thresholds in Fig. 3(b) and the theory lines in Figs. 3–4 are load-bearing, please re-derive Eq. (3) with the actual loss topology or justify why the loss","section":"Supplement Eq. (17); main text Eq. (3) and Fig. 3(b)"},{"comment":"The claim of a demonstrated crossover from the SQL to the Heisenberg limit is not supported by the data. The experimental dots in Fig. 4 follow the red solid line, which is the lossy optimization of Eq. (3); for n_T≫1 that line asymptotes to √(Λ/(K n_T)), not to the HL 1/(√K n_T). The violet dashed HL line is the lossless case, and the text itself states that reaching the HL would require lower losses and n_s≳1. Moreover, for n_T=3.29 the employed n_s=0.93 is far below the optimal n_s=n_T/2 required for the HL regime of Eq. (4). Please rephrase the abstract and conclusions to describe a loss-limited transition region, with the HL as a lossless asymptotic target rather than an experimentally demonstrated scaling.","section":"Abstract and Fig. 4"},{"comment":"The optimization step from Eq. (26) to Eq. (33) uses the Cauchy-Schwarz inequality in a direction that depends on the sign of (Δ²q)−1. For squeezed vacuum this prefactor is negative, so the lower bound is obtained by maximizing the cross term; the final result in Eq. (33) appears algebraically inconsistent with the intermediate Eq. (28) unless the δ_jk term in Eq. (22) is carried through the calculation. Please check the derivation and state explicitly whether Eq. (3) includes the vacuum-noise contribution ηδ_jk correctly; otherwise the fitted value of the squeezing term and the loss term Λ cannot be separated from the data.","section":"Supplement Eq. (33) and main text Eq. (3)"}],"minor_comments":[{"comment":"Typo: 'transitivity' should be 'transmissivity'.","section":"Supplement Eq. (17)"},{"comment":"The sentence 'The squeezing is distributed to six modes with of 5.30 ± 0.10 dB beyond standard quantum limit (SQL)' is grammatically incomplete; also 'beyond' should likely be 'below' when referring to squeezing noise suppression.","section":"Methods, first paragraph"},{"comment":"The notation ∥ν∥_2^2/3 in Eq. (5) is ambiguous; please define explicitly which norm and power are used, and verify the stated bounds [1,d] for the example ν_ave.","section":"Main text, Eq. (5)"},{"comment":"The caption says 'Dots represent experimental results' but does not specify error bars, number of repetitions, or how the total photon numbers were calibrated; this information is needed to judge the goodness of the fit to the lossy curve.","section":"Fig. 4 caption"}],"recommendation":"major_revision","confidential_remarks":"The high-intensity 4.36 dB result is likely sound and the paper is potentially publishable, but the loss-topology issue is load-bearing for the robustness thresholds and the fitted curves. I would ask for an explicit multi-stage loss derivation and a corrected abstract before considering acceptance. The Cauchy-Schwarz sign point in the Supplement should also be checked, as it affects the interpretation of the optimized formula."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe experimental result is real. They split a single squeezed-vacuum state across six Mach-Zehnder interferometers and see 4.36 ± 0.35 dB of joint noise suppression below the SQL at high coherent power, with an absolute phase uncertainty near 10^-9. That is a legitimate demonstration that one squeezed source is enough to run a reconfigurable multiparameter sensing network. The Supplement derives the central sensitivity formula from the input state, measurement, and a loss model rather than fitting it to the data, and the high-intensity points in Fig. 3(a) follow the theory line with η = 88%.\n\nThe issues are in proportion. First, the abstract's 'crossover from the SQL to the Heisenberg limit' oversells what the data show. In Fig. 4 the experimental points track the loss-limited SQL asymptote Λ/(K n̄_T), not the n̄_T^-2 Heisenberg scaling. The paper itself concedes that reaching the HL would require fewer losses and more squeezing than they have. So call it a demonstration of optimized low-photon performance in the loss-limited SQL regime, not a crossover to the HL.\n\nSecond, the loss bookkeeping. Eq. (3) writes Λ = 1/η −1 with η = η_dis η_MZI η_m^(2K−1). That applies the distribution efficiency to the coherent input modes, but those modes never pass through the QC. The correct treatment leaves the coherent amplitude unattenuated by η_dis and replaces the squeezed variance V by η_dis V + 1 − η_dis. The two models differ by a factor (1−η_dis)(V−1 + 1/(η_dis η_out)), which grows as η_dis drops. The paper's model is the pessimistic one: the true sensitivity is better, so the robustness thresholds (η_dis above 65% at K=1, above 20% at K=5) are conservative. Still, the model is not physically exact, and they never validate it with an independent check. The Supplement's Eq. (17) also uses one common vacuum bath for both interferometer arms; at the working point θ=0 the correlation cancels and it matches independent-loss behavior, so that part is less concerning.\n\nThird, most figures have no error bars and no raw data are provided. The error bars they do quote (4.36 ± 0.35 dB) are useful but the rest of the scatter is unquantified. That will need fixing before this is fully reproducible.\n\nNet: it's a substantial experimental paper with a coherent theory backing. It deserves a serious referee, ideally one who will push on the loss model and ask for raw data. I would not desk-reject it. The overclaim in the abstract is fixable with a few sentences.","headline":"Real experimental result—one squeezed source drives six MZIs with 4.36 dB sub-SQL joint noise suppression—but the abstract oversells a Heisenberg-limit crossover, and the loss bookkeeping is approximate, conservatively so.","tokens_in":20452,"tokens_out":5320,"would_cite":true,"duration_ms":59697,"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":"A single squeezed-vacuum state, split across six Mach-Zehnder interferometers, beats the standard quantum limit by 4.36 dB at 10^-9 phase uncertainty.","keywords":["distributed quantum sensing","squeezed vacuum","Mach-Zehnder interferometer","multiparameter estimation","standard quantum limit","Heisenberg limit","balanced homodyne detection","photon losses"],"falsifier":"Vary where the loss is introduced—before the first beam splitter, inside one arm, after the second beam splitter—while keeping the total transmission η fixed, and check whether the measured sub-SQL joint noise suppression (4.36 dB at K=1) stays constant. Eq. (3) predicts only η matters; any dependence on loss placement or modulation frequency would falsify the loss model and shift the quoted thresholds.","tokens_in":19429,"feed_emoji":"🔬","tokens_out":8899,"duration_ms":93991,"temperature":0.7,"pith_summary":"This paper sets out to show that distributed multi-parameter phase sensing can be made scalable without distributing many nonclassical resources: a single squeezed-vacuum state, split by a linear optical circuit, entangles a whole array of Mach-Zehnder interferometers. The central result is a compact formula for the phase variance of any linear combination of the local phases, and the experiment realizes it with six interferometers beating the standard quantum limit by 4.36 dB at an absolute phase uncertainty of about 10^-9. If correct, one nonclassical source, rather than one per sensor, is enough for quantum-enhanced networks, with sensitivity improving as 1/sqrt(d) and tolerating realistic losses. The paper also claims that, in the low-intensity regime and with optimal allocation of photons between squeezed and coherent inputs, the network crosses from standard-quantum-limit scaling to Heisenberg-limit scaling.","feed_headline":"One squeezed state drives six interferometers 4.36 dB below the SQL","feed_subtitle":"Distributed sensing with a single shared nonclassical resource reaches 10^-9 phase precision and scales as 1/sqrt(d).","key_machinery":"The machinery is the multinomial splitting of one squeezed-vacuum state by a linear quantum circuit: a single nonclassical photon-number distribution is fractioned into d modes with probabilities P_j=|ν_j|, producing mode entanglement from one nonclassical input. Each mode b_j enters one port of a Mach-Zehnder interferometer (a two-beam interferometer with two beam splitters), the other port receives a coherent state |α_j| with |α_j|^2/n_c=|ν_j|/Σ|ν_j|, and balanced homodyne detection reads one output quadrature. The carrying identity is the optimized phase variance Eq. (3), together with its optimization over squeezing, Eq. (4): it predicts SQL behavior at n_s=0, sub-SQL sensitivity when Λ<","core_discovery":"The central claim is that a configurable optical network of d Mach-Zehnder interferometers can estimate any linear combination of local phases with variance (e^{-2r}+Λ)/(K n_T) when the joint photon budget is dominated by coherent states, provided the d nodes share a single squeezed-vacuum state split through a linear quantum circuit. This formula carries the argument: squeezing suppresses common noise, losses add a vacuum-noise term Λ=1/η-1, multipass interactions multiply sensitivity by K, and the total photon number n_T sets the shot-noise scale. The experiment realizes this with six MZIs, reporting joint noise suppression 4.36±0.35 dB below the standard quantum limit at phase uncertainty","pith_inferences":["A natural next test is gradient estimation: since Eq. (3) holds for any weight vector ν, choosing ν linear in the node index would let the same six-node network estimate first and second spatial derivatives of an inhomogeneous field; the paper demonstrates only sign-structured vectors, not continuous weight profiles.","The single-effective-transmission loss model will probably need refinement for deployed fiber networks, where losses are frequency-dependent and uneven across nodes; testing the 4.36 dB suppression with loss inserted asymmetrically or at different Fourier frequencies would reveal whether the gain depends only on total transmission.","The resource-allocation formula n_s,opt ≈ n_T/(1+√(1+4Λ n_T)) reads as a control algorithm: a network operator could tune squeezing and coherent power in real time from a measured efficiency and target photon budget, without recomputing the full covariance matrix.","Because the entanglement comes from linear splitting of a Gaussian state, the same architecture should port to other platforms with Gaussian nonclassical resources, such as spin-squeezed atomic ensembles, where the coherent state and quadrature readout have platform-specific analogues."],"forward_implications":["The network reaches the same sensitivity as a separable scheme with d independent squeezed sources, using one squeezed-vacuum state; the nonclassical-resource count drops from d to one.","Increasing the number of sensors d improves the uncertainty on the average phase as 1/√d for fixed per-node coherent power, so the architecture scales without additional nonclassical resources.","Any linear combination of phases—average, staggered, antisymmetric—can be measured with the same optimized sensitivity by setting the quantum-circuit splitting and coherent phases, making gradient or parity-modulated field sensing accessible.","In the low-intensity regime with optimal photon allocation, the phase variance crosses from (1+Λ)/(K n_T) SQL-type scaling to 1/(K n_T^2) Heisenberg scaling when squeezed and coherent photon numbers are balanced; losses turn the asymptotic scaling back to Λ/(K n_T).","Multipass interaction multiplies the phase signal by K and, at 99.99% mirror efficiency, improves sensitivity and loss robustness, with the SQL beaten down to distribution efficiencies of about 65% (K=1) and 20% (K=5)."],"supporting_citations":[{"why":"Supplies the theoretical proposal for distributed quantum sensing with squeezed-vacuum light in a configurable array of Mach-Zehnder interferometers that this experiment realizes.","marker":"[32]"},{"why":"Provides the quantum Cramér-Rao bound and optimal local measurement analysis that the paper uses to claim saturation of the bound.","marker":"[33]"},{"why":"Contains the derivations of Eq. (3), the optimization over squeezing, and the comparison with the separable network.","marker":"[42]"},{"why":"Gives the multiparameter error-propagation and covariance machinery used to write Eq. (2) for the phase uncertainty.","marker":"[41]"},{"why":"Underpins the statement that losses break Heisenberg scaling in quantum-enhanced metrology, leading to the asymptotic Λ/(K n_T) floor.","marker":"[45]"},{"why":"Provides the companion result on the elusive Heisenberg limit under noise, supporting the loss-limited SQL asymptote in Eq. (4).","marker":"[46]"},{"why":"Defines the standard quantum limit baseline in Mach-Zehnder interferometry against which the 4.36 dB suppression is reported.","marker":"[7]"},{"why":"Establishes the coherent-plus-squeezed-vacuum Mach-Zehnder configuration and its Heisenberg-limit behavior, the single-interferometer ancestor of the network's low-intensity crossover.","marker":"[11]"}],"fun_headline_variants":["One squeezed state runs six MZIs 4.36 dB below SQL","Single entangled resource scales MZI network to Heisenberg limit","Distributed sensing: six interferometers from one squeezed vacuum","Shared squeezed vacuum achieves 4.36 dB joint noise suppression","MZI network with one nonclassical resource crosses Heisenberg limit"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"All photon losses are modeled as one effective transmission efficiency η with vacuum noise entering both quadratures; the sub-SQL region, the loss thresholds (65% and 20%), and the loss-limited SQL floor all follow from that single-beam-splitter model, which the paper calibrates but does not independently validate.","fun_headline_variants_meta":{"raw":{"variants":["One squeezed state runs six MZIs 4.36 dB below SQL","Single entangled resource scales MZI network to Heisenberg limit","Distributed sensing: six interferometers from one squeezed vacuum","Shared squeezed vacuum achieves 4.36 dB joint noise suppression","MZI network with one nonclassical resource crosses Heisenberg limit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000647,"raw_usage":{"total_tokens":2797,"prompt_tokens":719,"completion_tokens":2078,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":463,"completion_tokens_details":{"reasoning_tokens":1990}},"tokens_in":463,"tokens_out":2078,"duration_ms":15552,"temperature":1.0,"reasoning_tokens":1990,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T21:00:11.772190+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Vary where the loss is introduced—before the first beam splitter, inside one arm, after the second beam splitter—while keeping the total transmission η fixed, and check whether the measured sub-SQL joint noise suppression (4.36 dB at K=1) stays constant. Eq. (3) predicts only η matters; any dependence on loss placement or modulation frequency would falsify the loss model and shift the quoted thresholds.","supporting_citations":[],"review_version":1}