{"id":"da026bd0-4cdc-4e76-bb37-3c1496eb2fdc","arxiv_id":"2501.18256","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Using two spin-squeezed atom interferometers and ellipse fitting, differential phase sensitivity can scale as N^{-2/3} per measurement, a gain of N^{1/6} over the standard quantum limit, even under full 2π common-mode noise.","lead":"Physicists show that two squeezed atom interferometers can measure a small differential phase even when a huge shared noise swamps both signals, using an ellipse-fitting analysis of the noisy data. The best squeezing strength gives a precision gain that grows as the sixth root of the atom number, with a strongly reduced bias, a practical route to beat the standard quantum limit in realistic noisy devices.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Omitted second-order delta-method term may set the asymptotic bias to 1/N rather than the claimed 1/N^{4/3} at the bias-cancellation point.","rationale":"The reader's weakest_assumption correctly identifies the first-order Taylor expansion as the fragile point, but frames it mainly as a finite-sample shift of the zero-bias point. My concern is sharper: even at the large-N, τ = τ∗ operating point, the first-order truncation omits a second-order covariance term of order 1/N. Since the claimed 1/N^{4/3} bias scaling arises precisely from making the first-order term small, the omitted term can dominate asymptotically and invalidate the scaling law. The sensitivity part of the central claim — N^{-2/3} variance scaling and N^{1/6} gain — is supported by independent numerical fits and by the comparison with the Cramér-Rao bound and does not rely on the bias cancellation. The bias half of the claim, however, is load-bearing for the word 'accurate' and for the 'eliminating bias' statement. The paper deserves credit for a transparent derivation and extensive numerical comparisons, including multiple fitting methods; the issue is not a logical inconsistency but an unverified cancellation. A concrete computation of the second-order term, or high-N numerical confirmation, would settle whether the asymptotic bias scaling is 1/N^{4/3} or 1/N. Until then, the verdict should be conditional: the central sensitivity result can stand, but the bias-scaling claim needs the additional check.","tokens_in":23094,"tokens_out":6892,"duration_ms":71214,"concrete_test":"Compute the omitted second-order bias term B2 = (1/2)Σ_{j,l}(∂²f/∂G_j∂G_l) Cov(G_j, G_l) using the moments derived in Appendix B.b, evaluated at τ = τ∗ and δϕ fixed, and compare its large-N scaling with the first-order term Eq. (30). If B2 ∼ 1/N with a finite coefficient, the claimed 1/N^{4/3} bias scaling is not asymptotic and the claim should be revised. As a numerical cross-check, run the one-parameter fit Monte Carlo at N = 10^5 and 10^6 points per ellipse (using the exact moment expressions to avoid sampling cost) and fit log|B| versus log N; if the fitted slope moves from ≈1.3 toward ≈1, the Fig. 5 result is a finite-N crossover rather than the asymptotic scaling.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The bias scaling result in Sec. III C, Eq. (30), is derived in Appendix B.b as a first-order Taylor expansion of the estimator f about the unbiased point g∞. It contains only mean-shift terms, (∂f/∂g_l)(⟨G_l⟩ − g∞_l). The true estimator bias also has a second-order delta-method contribution (1/2)Σ_{j,l}(∂²f/∂G_j∂G_l) Cov(G_j, G_l), which is generically O(1/N) because Cov(G_j, G_l) = O(1/N). At τ = τ∗, the first-order term is suppressed to O(1/N^{4/3}) through H0 ∼ σ_z² ∼ 1/N^{4/3} and H2 = 0. But unless the second-order covariance term vanishes identically or is also subleading at τ = τ∗ — which the paper does not show — the asymptotic bias is O(1/N), not O(1/N^{4/3}). The numerical fits in Fig. 5 are restricted to 300 < N < 1000 and cannot distinguish a true 1/N^{4/3} tail from a crossover between a 1/N^{4/3} transient and a 1/N asymptotic term. The one-parameter analytical curves in Fig. 5 use the same first-order formula Eq. (27)/(30), so they do not independently confirm the exponent. This matters because the abstract's claim of 'eliminating bias' and the improved bias scaling are central to the paper's accuracy contribution; if the second-order term dominates, squeezed states still reduce bias at practical finite N, but the claimed asymptotic scaling advantage over coherent states is not established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper theoretically analyzes differential phase estimation with two atom interferometers using spin-squeezed states in the presence of large common-mode phase noise spanning the full 2π range. The authors propose model-free ellipse fitting as the estimator and derive an optimal squeezing strength τ* that equalizes the phase-dependent variance contributions. They claim that, at τ*, the single-point differential phase variance scales as N^{-2/3} (a gain of N^{1/6} over the SQL) and that the estimation bias scales as B ~ 1/N^{4/3}, compared with 1/N for coherent states. The analytical scaling laws are derived from first principles in Sec. II and the Appendix, and are compared with Monte Carlo simulations for several fitting algorithms, the Cramér-Rao bound, and a hybrid approach with a classical sensor.","tokens_in":23366,"tokens_out":7277,"duration_ms":64493,"significance":"If the claims hold, this is a valuable contribution to quantum-enhanced interferometry: it shows that spin squeezing can simultaneously improve precision and accuracy in a realistic large-noise scenario using readily available states and a calibration-free estimator. The paper provides concrete, falsifiable scaling predictions and benchmarks against the CRB, and it explicitly compares different fitting methods. The variance scaling N^{-2/3} and the corresponding gain N^{1/6} are well supported by both analytic calculations and simulations, making the core metrological result credible and of direct interest to the atom interferometry and quantum metrology communities.","major_comments":[{"comment":"The claimed bias scaling B ~ 1/N^{4/3} at τ = τ* is derived from a first-order delta-method expansion of the estimator f about the unbiased point g∞. This expansion retains only the mean-shift terms (∂f/∂G_l)(⟨G_l⟩ − g∞_l), while the second-order contribution (1/2)Σ_{j,l}(∂²f/∂G_j∂G_l) Cov(G_j, G_l) is omitted. Because the G_l are sample means over N independent measurements, Cov(G_j, G_l) = O(1/N), so the second-order term is generically O(1/N). At τ = τ* the first-order term is suppressed to O(1/N^{4/3}) via H0 ∼ σ_z² ∼ N^{-4/3} and H2 = 0, but unless the second-order term vanishes identically or is also subleading at τ*, which the manuscript does not demonstrate, the asymptotic bias is O(1/N), not O(1/N^{4/3}). This is load-bearing because the abstract's claim of 'eliminating bias' and the improved asymptotic accuracy relative to coherent states are central to the paper's contribution. The authors should compute or bound the second-order delta-method term, or verify the scaling with Monte Carlo at larger N using an estimator that does not rely on the first-order formula.","section":"Sec. III C, Appendix B.b, Eq. (30)"},{"comment":"The numerical evidence for the 1/N^{4/3} bias scaling is not conclusive. The power-law fits are restricted to the range 300 < N < 1000, and the one-parameter analytical curves shown as solid lines in Fig. 5 are obtained from the same first-order formula Eq. (27)/(30), so they do not independently confirm the exponent. The error bars on the smallest bias values are also large. To distinguish a true asymptotic 1/N^{4/3} tail from a crossover between a 1/N^{4/3} transient and a 1/N asymptotic term, the authors should extend the range of N (e.g., to 10^5 or 10^6) or provide an independent numerical evaluation of the bias, for instance by directly solving the cubic equation at each Monte Carlo realization without the delta-method approximation.","section":"Fig. 5, Table I"},{"comment":"The phrase 'eliminating bias inherent in ellipse fitting methods' overstates the result, since the analysis itself shows a residual bias B(δϕ_est) ≠ 0 at τ = τ*, with a scaling that is at best N^{-4/3} (and possibly O(1/N) if the second-order term dominates). The conclusion in Sec. IV similarly claims that spin squeezing 'can remarkably suppress the bias' without specifying the residual scaling. The authors should rephrase to 'strongly suppress' and explicitly report the residual bias and its asymptotic behavior, keeping the claim consistent with the analysis.","section":"Abstract and Sec. IV"}],"minor_comments":[{"comment":"The symbol N is used both for the number of atoms per interferometer and for the number of measurement points per ellipse (e.g., Eq. (7) vs. Fig. 5). This is confusing, especially in Sec. II C and Appendix C. Consider introducing separate notation such as N_at for the atom number and N_pts for the sample size.","section":"Notation throughout"},{"comment":"The expression σ_SQL_δφ = √2 N^{-1/2} N^{-1/2} appears to contain an extra factor N^{-1/2}; clarify which N is the atom number and which is the number of measurements, or combine them into a single symbol with an explicit definition.","section":"Eq. (15)"},{"comment":"The shorthand notations σ²_0 and σ²_π/2 are introduced in the main text but the equations in Appendix B would benefit from a restatement of these definitions for self-containedness, particularly because the calculation in Eq. (B10) heavily relies on them.","section":"Appendix B.b, Eq. (B10)"},{"comment":"The legend in Fig. 4 refers to 'one parameter' while the caption calls it 'algebraic one-parameter'; unify the terminology to avoid confusion with the two-parameter algebraic fits.","section":"Fig. 4"},{"comment":"The reference has a typo: 'Pezz`e ans Smerzi' should be 'Pezzè and Smerzi'.","section":"Reference [33]"},{"comment":"The notation N is again used for the number of points in an ellipse while in the main text N denotes the atom number; please use distinct symbols and clarify the axes in Fig. 7.","section":"Appendix C, Fig. 7"}],"recommendation":"major_revision","confidential_remarks":"The variance scaling result (N^{-2/3} per measurement, gain N^{1/6}) appears solid and is a useful contribution. The main weakness is the bias-scaling claim, which is derived from a first-order delta-method expansion and may be incorrect at the asymptotic level due to the omitted second-order term. The authors should be asked to address this point explicitly, either by computing the second-order term or by providing more extensive numerical evidence. If the second-order term indeed dominates, the paper can still be accepted after revising the abstract and conclusions to claim a finite-N bias reduction rather than an asymptotic 1/N^{4/3} scaling."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Friend,\n\nThe paper has a genuine result: under full 2π common-mode phase noise, two spin-squeezed states with the variance-equalizing strength τ* give a per-shot differential phase sensitivity scaling as N^{-2/3}, a N^{1/6} gain over the SQL, and this is confirmed numerically and compared to the CRB. That part is solid and worth building on.\n\nThe soft spot is the bias scaling. The paper claims B ~ 1/N^{4/3} at τ*, derived from a first-order Taylor expansion (Eq. 30, Appendix B.b). That approximation keeps only the mean shift of the G's and omits the second-order delta-method term involving their covariances. Because the data points spread along the ellipse, Cov(G_j, G_l) is O(1/N) even for squeezed states, so the second-order term is generically O(1/N). At τ*, the first-order term is O(1/N^{4/3}), so the second-order term dominates asymptotically unless it vanishes by symmetry, which the paper does not show. The numerical fits in Fig. 5 only go up to N=1000, so they can't tell a true 1/N^{4/3} tail from a crossover to a 1/N asymptote. The one-parameter analytical curves use the same first-order formula, so they don't independently settle the exponent. In short, the improved bias scaling over coherent states is not established. The abstract's \"eliminating bias\" is also stronger than the main text's \"nearly unbiased,\" a minor overclaim.\n\nThis doesn't sink the sensitivity claim, which is the paper's main novelty. The comparison across fitting methods and the hybrid analysis are useful, and the derivations are transparent. The missing code is a reproducibility limitation, not a correctness flaw.\n\nThis paper deserves a serious referee, but the referee should push on the bias analysis. If the second-order term is indeed O(1/N), the accuracy advantage of τ* over coherent states is a constant factor rather than a scaling advantage, and the title's emphasis on 'accurate' needs rethinking. For those working on squeezed-state atom interferometry, the N^{-2/3} sensitivity result is a useful reference regardless.\n\nRecommendation: send to peer review, with a request for a corrected or qualified bias analysis before acceptance.\n\nBest","headline":"The sensitivity scaling N^{-2/3} is real and worth building on, but the claimed N^{-4/3} bias suppression is likely an artifact of dropping the second-order delta-method term.","tokens_in":23999,"tokens_out":8058,"would_cite":true,"duration_ms":69033,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Spin-squeezed states in both interferometers let differential phase sensing beat the standard quantum limit even when common-mode phase noise covers the full $2\\pi$ range.","keywords":["spin squeezing","atom interferometry","differential phase estimation","ellipse fitting","standard quantum limit","common-mode phase noise","quantum metrology","one-axis twisting"],"falsifier":"Simulate the distribution of Eq. (6) for two squeezed states at $\\tau=\\tau_*$, with atom number $N$ between $10^2$ and $10^4$ and many sampled points per ellipse, and estimate $\\delta\\phi$ with the trace-constrained and one-parameter fits. The claim fails if the per-shot uncertainty does not approach the $N^{-2/3}$ scaling relative to the SQL, or if the bias decays as $1/N$ rather than $1/N^{4/3}$ as $N$ grows. A direct check is to confirm that the empirical bias crosses zero near $\\tau_*$ and that the crossing point converges to Eq. (18) as $N$ increases.","tokens_in":22857,"feed_emoji":"⚛️","tokens_out":12095,"duration_ms":99306,"temperature":0.7,"pith_summary":"Two atom interferometers that share a noisy laser usually lose their squeezed-state advantage when common-mode phase noise randomizes the phase over a full circle. The paper claims that feeding both interferometers with spin-squeezed states of a particular strength $\\tau_*$ restores it: estimating the differential phase by fitting the correlated outputs to an ellipse gives a phase uncertainty that scales as $N^{-2/3}$ per shot, a gain of $N^{1/6}$ over the standard quantum limit, while the fitting bias falls as $N^{-4/3}$ instead of the $1/N$ found for coherent states. The protocol requires no calibration of the output statistics and uses squeezed states that are already available in experiments. If correct, it would make entanglement useful under exactly the large common-mode noise that differential atom sensors are built to reject.","feed_headline":"Squeezed atoms beat the quantum limit under full-range noise","feed_subtitle":"At one special squeezing strength, ellipse fitting gives N^{-2/3} phase uncertainty and suppresses fitting bias.","key_machinery":"The central machinery is the one-axis-twisted spin-squeezed state $|\\psi_{\\mathrm{Squ}}\\rangle = e^{-i\\nu\\hat{J}_x}e^{-i\\tau\\hat{J}_z^2}|\\psi_{\\mathrm{Coh}}\\rangle$ sent into each of two Ramsey interferometers, together with the ellipse-fitting estimator acting on the joint output distribution $P(z_A,z_B|\\delta\\phi)=\\int_0^{2\\pi} d\\phi_{\\mathrm{cn}}\\,P_0(z_A|\\phi_A)P_0(z_B|\\phi_B)/(2\\pi)$. The argument turns on the variance-balance condition $\\sigma^2_z|_{\\phi=0}=\\sigma^2_z|_{\\phi=\\pi/2}$, which selects $\\tau_*$, and on a first-order Taylor (delta-method) expansion of the one-parameter fit, whose cubic equation in $h=\\cos\\delta\\phi$ yields the bias formula used to show $H_2=0$ at $\\tau_*$. Four fitting variants are compared--trace-constrained algebraic, ellipse-specific algebraic, geometric, and one-parameter--and the geometric fit comes closest to the Fisher-information bound at small differential phase.","core_discovery":"With a common phase $\\phi_{\\mathrm{cn}}$ uniformly distributed over $[0,2\\pi]$, the mean outputs of the two interferometers trace an ellipse whose shape encodes the differential phase $\\delta\\phi$. The paper shows that at the squeezing strength $\\tau_* \\simeq (2/N^5)^{1/6}$, set by equating the projection-noise variances at $\\phi=0$ and $\\phi=\\pi/2$, the noise around that ellipse becomes phase-independent. At this strength, ellipse fitting extracts $\\delta\\phi$ with a per-shot standard deviation $\\sigma_{\\delta\\phi} \\sim N^{-2/3}$ and a bias whose leading term is $\\sim N^{-4/3}$; the compact bias formula $B(\\delta\\phi_{\\mathrm{est}}) \\approx -4\\cot\\delta\\phi\\,(H_0+H_2 h^2)/(1+2h^2)$, with $h=\\cos\\delta\\phi$, has $H_2=0$ at $\\tau_*$, removing the dominant coherent-state bias. The resulting sensitivity lies within about 1.5 dB of the Cramér-Rao bound for the same states, and the $N^{1/6}$ quantum gain holds over the whole range $0\\lesssim\\delta\\phi\\lesssim\\pi/2$ for the ellipse method, while a hybrid classical-sensor fringe fit remains complementary near $\\delta\\phi=0$.","pith_inferences":["Inference: The finite-sample shift of the zero-bias point away from $\\tau_*$ suggests an adaptive version of the protocol could estimate the optimal squeezing strength from early data and then hold it, trading some measurement time for lower bias.","Inference: Since the geometric fit already approaches the Cramér-Rao bound at small $\\delta\\phi$, a maximum-likelihood estimator on the same squeezed states may close the remaining 1.5 dB gap, at the price of the calibration step the paper avoids.","Inference: The variance-balance condition that defines $\\tau_*$ suggests a design principle for other probe states: make the projection noise isotropic over the Bloch sphere, which could extend the scaling gain to other entangled or variational states if they become experimentally available."],"forward_implications":["A differential gravimeter or gradiometer subjected to large vibration or laser-phase noise can gain a factor $N^{1/6}$ over the standard quantum limit by tuning the squeezing strength to $\\tau_*$, with no calibration of the readout distribution.","The bias reduction means long averaging is not dominated by a systematic fitting offset: the leading bias scales as $N^{-4/3}$ instead of $1/N$.","Because the sensitivity at $\\tau_*$ is essentially independent of the differential phase, the protocol works across the wide range $0\\lesssim\\delta\\phi\\lesssim\\pi/2$.","Near $\\delta\\phi=0$ the hybrid classical-sensor fringe-fitting method remains the better choice, so the two approaches are complementary rather than interchangeable."],"supporting_citations":[{"why":"Defines one-axis twisting and the spin-squeezed states used as probe states, including the variance formulas that set the optimal squeezing strength.","marker":"[11]"},{"why":"Introduces ellipse-specific fitting for phase extraction between coupled atom interferometers, the estimation procedure at the centre of the paper.","marker":"[38]"},{"why":"Supplies the differential interferometer model with common phase noise and the ellipse equation connecting the two outputs.","marker":"[39]"},{"why":"Establishes the geometry and statistical bias of static conic fitting, which the paper extends to quantum projection noise.","marker":"[72]"},{"why":"Provides the atomic coherent-state generating function used in the Appendix to compute squeezed-state moments.","marker":"[77]"},{"why":"Gives the asymptotic expansion of functions of statistics that justifies the delta-method bias and variance formulas.","marker":"[91]"},{"why":"Analyzes how phase noise degrades squeezed-clock sensitivity and identifies the $N^{1/6}$ gain recovered here for a single squeezed state.","marker":"[30]"}],"fun_headline_variants":["Squeezed atoms achieve N^{-2/3} phase precision under 2π noise","Squeezed ellipse fit yields sub-SQL sensitivity under full-range noise","Squeezed differential sensing surpasses SQL despite full 2π noise","Bias-free squeezed ellipse fit delivers N^{-2/3} phase precision","Squeezed states give noise-immune N^{-2/3} differential sensing"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The bias and variance formulas assume that the sample moments feeding the cubic fit are close to their mean values, so a first-order Taylor expansion around the large-$N$, zero-squeezing limit captures the estimator's behaviour; with finite data the zero-bias point shifts away from $\\tau_*$ and a residual bias remains.","fun_headline_variants_meta":{"raw":{"variants":["Squeezed atoms achieve N^{-2/3} phase precision under 2π noise","Squeezed ellipse fit yields sub-SQL sensitivity under full-range noise","Squeezed differential sensing surpasses SQL despite full 2π noise","Bias-free squeezed ellipse fit delivers N^{-2/3} phase precision","Squeezed states give noise-immune N^{-2/3} differential sensing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001926,"raw_usage":{"total_tokens":7563,"prompt_tokens":996,"completion_tokens":6567,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":612,"completion_tokens_details":{"reasoning_tokens":6464}},"tokens_in":612,"tokens_out":6567,"duration_ms":43207,"temperature":1.0,"reasoning_tokens":6464,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T00:10:00.388300+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the distribution of Eq. (6) for two squeezed states at $\\tau=\\tau_*$, with atom number $N$ between $10^2$ and $10^4$ and many sampled points per ellipse, and estimate $\\delta\\phi$ with the trace-constrained and one-parameter fits. The claim fails if the per-shot uncertainty does not approach the $N^{-2/3}$ scaling relative to the SQL, or if the bias decays as $1/N$ rather than $1/N^{4/3}$ as $N$ grows. A direct check is to confirm that the empirical bias crosses zero near $\\tau_*$ and that the crossing point converges to Eq. (18) as $N$ increases.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces ellipse-specific fitting for phase extraction between coupled atom interferometers, the estimation procedure at the centre of the paper."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the differential interferometer model with common phase noise and the ellipse equation connecting the two outputs."},{"cited_title":"Gessner, L","cited_arxiv_id":null,"evidence_quote":"Establishes the geometry and statistical bias of static conic fitting, which the paper extends to quantum projection noise."},{"cited_title":"Ridley and A","cited_arxiv_id":null,"evidence_quote":"Provides the atomic coherent-state generating function used in the Appendix to compute squeezed-state moments."},{"cited_title":"Buˇzek, R","cited_arxiv_id":null,"evidence_quote":"Gives the asymptotic expansion of functions of statistics that justifies the delta-method bias and variance formulas."},{"cited_title":"Anders et al., Momentum Entanglement for Atom Interfer- ometry, Phys","cited_arxiv_id":null,"evidence_quote":"Analyzes how phase noise degrades squeezed-clock sensitivity and identifies the $N^{1/6}$ gain recovered here for a single squeezed state."}],"review_version":1}