{"id":"f789f890-b5d6-4168-8039-630a85c7c0d3","arxiv_id":"1908.08343","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A variational feedback algorithm using only global pulses and finite-range Rydberg interactions can prepare spin-squeezed states that outperform standard one-axis and two-axis twisting protocols in numerical simulations.","lead":"This paper proposes a variational quantum algorithm that optimizes pulse sequences on the quantum device itself to create spin-squeezed states in atom arrays for better precision measurements. A smart generalist should read it because it connects programmable quantum simulators with atomic clocks and could improve the sensitivity of future timekeeping and sensing without waiting for a full quantum computer.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Fig. 3's rotationally invariant ξ² is not the cost used in the on-device feedback loop, which minimizes N⟨J_y²⟩/⟨J_x⟩²; the claimed on-device advantage over OAT/TAT needs y-cost curves.","rationale":"The reader's weakest assumption correctly identifies a mismatch between the cost optimized on the device (N⟨J_y²⟩/⟨J_x⟩²) and the cost used in the exact benchmarks (the minimum transverse variance of Eq. (1)). I agree this is the most load-bearing point. However, the reader's formulation slightly mischaracterizes the direction of the problem: the y-variance is an upper bound on the minimum transverse variance, so a low y-cost guarantees a low true ξ²; the real issue is that the theoretical curves in Fig. 3 may be better than anything the feedback loop can achieve with the same circuit depth, because optimizing the rotationally invariant metric does not enforce the squeezed axis to be y. The circuit contains R_x rotations, but fixed-depth closure under such rotations is not proven, and the final D_z gate can tilt the squeezed quadrature. If the exact optimal states have their minimal variance direction away from y, then the feedback loop's cost will be higher (worse) than the plotted curves, and the claimed on-device advantage over OAT/TAT may be smaller or absent. This is a concrete, testable numerical question that does not invalidate the proposal outright, hence CONDITIONAL rather than REJECT. The paper should report the y-cost curves as the protocol-relevant performance or prove that the minimal y-cost equals the minimal Eq. (1) value at the depths studied.","tokens_in":38289,"tokens_out":18276,"duration_ms":183563,"concrete_test":"Recompute the exact optimizations of Fig. 3 (4×4 square, R_C/a ∈ {1, 1.5, 2, 3, 4, 5}, n = 1,...,5) using the on-device cost C_y(θ)=N⟨J_y²⟩/⟨J_x⟩² in place of the rotationally invariant ξ² of Eq. (1). Plot C_y against the fOAT/OAT/TAT benchmarks. Additionally, take the parameters that minimize Eq. (1) and evaluate (i) C_y and (ii) the angle of the minimal-variance direction in the y–z plane. If C_y differs from the reported ξ² by more than a few tenths of a decibel, or if the minimal-variance direction is not within a few degrees of y, the central comparison must be redone with the on-device cost; if no significant difference appears, the concern is resolved.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The on-device protocol in '2D arrays: single optimization run with shot noise' estimates the cost as ξ²(θ_i)=N⟨J_y²⟩/⟨J_x⟩² from projective measurements in the y and x bases. The exact results in Fig. 3, however, report the 'optimized squeezing parameter ξ²' and Fig. 4 explicitly labels it 'rotationally invariant spin squeezing', i.e., the minimum transverse variance of Eq. (1). Parity symmetry P_x guarantees ⟨J_y⟩=⟨J_z⟩=0, so the Bloch vector is pinned along x, but it does not make ΔJ_y² equal to the minimum eigenvalue of the y–z covariance matrix. The circuit contains R_x rotations, but the fixed-depth family S(θ) is not shown to be closed under arbitrary R_x rotations, so the minimal y-variance over S(θ) can be strictly larger than the minimal transverse variance. Since the feedback loop optimizes the y-variance, the curves in Fig. 3 do not directly bound what the on-device algorithm will achieve. If the squeezed quadrature lies away from y, the actual Ramsey sensitivity for the described π/2-pulse sequence (which probes the y component) is worse than the reported ξ². The paper does not flag this discrepancy or provide y-cost versions of the central benchmark plot.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a hybrid classical-quantum variational algorithm for generating spin-squeezed states on programmable Rydberg-dressed tweezer arrays. The central object is a shallow circuit S(θ)=U_n...U_1 with layers U_i=D_x(τ'_i)R_x(ϑ_i)D_z(τ_i), built from global rotation gates and Ising interaction gates with a finite-range Rydberg-dressed potential. The authors argue that the parity symmetry P_x keeps the collective Bloch vector along x, and that the number of variational parameters is independent of N. They present three sets of results: (i) a simulated on-device feedback loop with shot noise on a 4x4 array; (ii) exact numerical optimizations for 4x4 arrays as a function of interaction radius and circuit depth, reporting squeezing beyond finite-range OAT, infinite-range OAT, and TAT; and (iii) MPS simulations for 1D chains up to N=150, showing moderate improvement with system size and stronger improvement with depth. Robustness to stochastic filling and to Gaussian control noise is also studied. The supplemental material contains a detailed discussion of the pulse decomposition, Husimi-function visualization, measurement scaling, and a penalty-modified cost function.","tokens_in":38551,"tokens_out":11540,"duration_ms":123262,"significance":"If substantiated, the paper would establish a practical variational route to entangled-state preparation for metrology on tweezer-based optical clocks, with an experimentally concrete platform and a feedback loop that adapts to device noise. The work has genuine strengths: the exact small-system optimizations are a non-circular comparison against independent protocols (fOAT, OAT, TAT); the shot-noise emulation and the filling/noise robustness tests are concrete; the SM gives a full decomposition of each layer into Rydberg-dressing pulses and a useful measurement-scaling analysis; and the number of variational parameters scales as 3n, independent of N. However, the central on-device claim currently rests on an unverified identification between the experimentally estimated cost and the rotationally invariant squeezing parameter used in the benchmarks, and the 1D MPS results lack convergence evidence. Both issues are fixable in revision.","major_comments":[{"comment":"The simulated on-device feedback loop minimizes ξ²(θ_i)=N⟨J_y²⟩/⟨J_x⟩², whereas the exact optimization results in Figs. 3 and 4 are reported for the rotationally invariant spin-squeezing parameter of Eq. (1), as explicitly labeled in Fig. 4. Parity P_x only guarantees ⟨J_y⟩=⟨J_z⟩=0, not that the minimum transverse variance lies along y; the D_x gates twist the y–z plane and no final R_x rotation is included in the ansatz to align the squeezed quadrature with y. The device-level optimization therefore minimizes a different quantity from the one compared against OAT/TAT/fOAT, so the claimed on-device enhancement is not directly supported by the present figures. Please either add a final rotation and estimate the full transverse covariance matrix, or provide y-cost versions of all benchmark curves, including the reference protocols, so that the comparison is made on the same cost function.","section":"2D arrays: single optimization run with shot noise; Eq. (1); Fig. 4"},{"comment":"The MPS simulations for 1D chains (Fig. 3, top right panel, and SM Fig. 10) do not report bond dimensions or truncation errors. For finite-range Ising dynamics at R_C/a=3, entanglement can grow with N, and a fixed bond dimension could artifactually flatten the ξ²(N) curve and support the parameter-transfer claim. Please provide convergence checks of ξ² with respect to bond dimension for the largest system sizes studied, or state the truncation error for each data point.","section":"1D arrays; SM 'Measurement scaling with system size'"}],"minor_comments":[{"comment":"The statement that the allowed gates preserving the axis are 'R_z, D_x, D_z' conflicts with the main-text claim that gates commuting with P_x reduce rotations to R_x; R_z does not preserve the x-axis. Please correct this typo or clarify the intended set of allowed gates.","section":"Supplemental Material, 'Design of the variational circuit'"},{"comment":"The labels A and B for the OAT reference lines are not defined in the caption; please define them there or in a legend so that the figure is self-contained.","section":"Fig. 1 caption"},{"comment":"The relation between '100 runs for a single cost function evaluation' and the total budget of roughly 10^5 runs is not stated explicitly; adding one sentence would make the optimization trajectory in Fig. 1 easier to interpret.","section":"2D arrays: single optimization run with shot noise"},{"comment":"The claim that the number of measurements per cost evaluation does not increase with N is derived for a particular optimally squeezed state; for generic trial states with small ⟨J_x⟩, the relative error of the ratio can be large. Please qualify the statement accordingly.","section":"SM, 'Measurement scaling with system size'"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the cost-function mismatch between the on-device feedback loop and the rotationally invariant squeezing parameter used in the benchmarks is the main substantive issue and should be resolved before publication. The paper is otherwise within scope and likely of interest to both the quantum-sensing and variational-algorithm communities. The MPS convergence data should be requested as part of the revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Plain take: this is a real proposal, not a repackaging. The authors adapt variational hybrid classical-quantum optimization to prepare spin-squeezed states in Rydberg-dressed tweezer arrays, with a circuit of D_x, R_x, D_z layers using only global pulses. The new ingredient is the D_x gate: for finite-range Ising interactions, D_z alone generates transient non-Gaussian states that cannot be inverted; adding D_x lets the optimizer restore a squeezed Gaussian state at longer times. That design insight is the strongest part of the paper, and it is backed by exact numerics on 4x4 arrays and MPS simulations on 1D chains up to 150 sites, plus robustness checks against shot noise, filling disorder, and control noise. The benchmarks against fOAT, OAT, and TAT are honest and the improvement is clear in the studied geometries.\n\nThe main soft spot is the cost function used by the on-device feedback loop. The exact results in Fig. 3 and the QFI panels are computed with the rotationally invariant squeezing parameter, i.e. the minimum transverse variance of Eq. (1). But the simulated experiment in the feedback section estimates ξ² as N⟨J_y²⟩/⟨J_x⟩² from y-basis measurements. Parity symmetry pins the Bloch vector along x, but it does not guarantee that the minimal variance is along y. The circuit contains R_x rotations, so in principle the optimizer can rotate the squeezed quadrature, but the fixed-depth family is not shown to be closed under a final R_x; the last gate in each layer is D_x. So the y-cost could be strictly worse than the reported ξ². The paper does show a noisy feedback run in Fig. 1 that surpasses OAT using the y-cost, which is encouraging, but the systematic curves in Fig. 3 are not the right comparison for the on-device protocol. This needs either a short proof that optimal squeezing can always be oriented along y within the ansatz, or additional y-cost benchmark curves. It is an addressable gap, not a fundamental flaw.\n\nMinor issue: no code or data are released, so the numerics are not independently reproducible as-is. That is common for a proposal of this kind, but worth noting. The claim that Eq. (3) is the 'most general' sequence of allowed gates is argued from symmetry and pacing, not proven; it is not load-bearing.\n\nOverall, the paper is a clear step forward for practical entanglement-enhanced Ramsey metrology in tweezer clocks. I would bring it to a reading group and would cite it. It deserves full peer review; a referee should push on the y-cost issue, but the central claim is well supported.","headline":"A solid variational proposal for spin squeezing in tweezer clocks; the central existence claim holds, but the on-device cost function and the reported squeezing parameter are not the same quantity and need reconciling.","tokens_in":39123,"tokens_out":5542,"would_cite":true,"duration_ms":56677,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68"],"pacs":["03.67.-a","06.30.Ft"],"model":"deepseek-v4-flash","headline":"A programmable quantum sensor can use variational feedback to prepare spin-squeezed states that beat standard squeezing protocols, even with realistic noise.","keywords":["spin squeezing","variational quantum algorithm","programmable quantum sensor","Rydberg dressing","optical tweezer arrays","Ramsey interferometry","quantum metrology","one-axis twisting"],"falsifier":"Perform the feedback-loop optimization, then measure the collective-spin covariance matrix in the $y$-$z$ plane for the optimal parameters; if the minimum-variance direction is measurably away from $y$, the cost actually optimized on the device is not the spin-squeezing parameter quoted in the paper.","tokens_in":38036,"feed_emoji":"⏱️","tokens_out":7606,"duration_ms":75593,"temperature":0.7,"pith_summary":"This paper proposes a way to make a programmable quantum sensor an array of strontium atoms in optical tweezers with Rydberg-dressed interactions prepare its own optimally entangled states. The idea is a variational feedback loop: the device generates a trial squeezed state from a short sequence of global gates, measures a cost function that quantifies metrological gain, and a classical optimizer adjusts the gate parameters. The authors show numerically that in square and one-dimensional arrays this loop reaches spin-squeezing levels beyond finite-range one-axis twisting, infinite-range one-axis twisting, and two-axis twisting, and that the optimized states survive stochastic filling and control noise. Because the optimization runs on the real device, the final state automatically accounts for imperfections that a purely numerical optimization would miss.","feed_headline":"On-device feedback loop beats standard spin-squeezing protocols","feed_subtitle":"A variational circuit on Rydberg-dressed atom arrays prepares better squeezed states for clocks despite noise.","key_machinery":"The load-bearing object is the spin-squeezing parameter $\\xi^2(\\theta)=N(\\Delta J_{\\perp,\\min})^2/|\\langle\\mathbf{J}\\rangle|^2$, used as the variational cost function, together with the unitary squeezer $S(\\theta)=U_n\\cdots U_1$. Each layer combines two finite-range Ising interaction gates $D_z(\\tau_i)$ and $D_x(\\tau'_i)$, generated from Rydberg dressing by spin-echo sequences, with a global rotation $R_x(\\vartheta_i)$. This ordering is the most general gate sequence that preserves the collective spin direction along $x$ through parity symmetry, so $\\langle J_y\\rangle=0$ by construction and the cost can be estimated from $x$- and $y$-basis measurements alone. The $D_x$ gate is what lets the circuit escape the short-time limitation of pure one-axis twisting: numerical results show that $D_x$ unwinds the non-Gaussian, S-shaped Husimi distributions that finite-range $D_z$ dynamics generates irreversibly, allowing longer interaction times and better squeezing. The optimization is performed by a derivative-free search algorithm on the device, with the cost estimated from a number of measurements that does not grow with $N$.","core_discovery":"The central discovery is that spin-squeezed states for Ramsey interferometry can be prepared variationally from the finite-range Ising interactions available in Rydberg-dressed tweezer arrays, using a circuit $S(\\theta)=U_n\\cdots U_1$ with each layer $U_i=D_x(\\tau'_i)R_x(\\vartheta_i)D_z(\\tau_i)$. The circuit is deliberately built from global gates only, so the number of variational parameters, $3n$, is independent of the number of atoms $N$, and every gate commutes with parity in the $x$ direction, which keeps the collective Bloch vector along $x$ and removes the need to measure its direction. The paper claims that optimizing the spin-squeezing parameter $\\xi^2$ on the device itself, using a finite number of projective measurements per parameter update, yields states whose squeezing surpasses the values reachable with finite-range one-axis twisting, infinite-range one-axis twisting, and two-axis twisting in the geometries studied, and that the improvement persists under stochastic filling and Gaussian control noise.","pith_inferences":["Going beyond the paper, the cost-function alignment can be tested by computing, for the optimized states, the minimal transverse variance over all directions perpendicular to the $x$-axis and comparing it with the $y$-basis estimator; a mismatch would indicate the feedback loop should measure a rotated quadrature.","Going beyond the paper, the same layer structure could be run with the squeezing axis treated as a variational parameter, or with a penalty that enforces a chosen Bloch-vector length, extending the penalized-cost idea the paper only touches on.","Going beyond the paper, because the number of parameters is independent of $N$, the ansatz is a candidate for scalable variational metrology on other programmable platforms where finite-range interactions are available, such as molecule arrays or optical lattices."],"forward_implications":["If the claim is right, a tweezer-based clock can run the optimization loop at realistic repetition rates, around $10^5$ shots, and end with a state already adapted to the device's own noise without needing a perfect model.","Because the parameter count $3n$ is independent of $N$, the same pulse sequence can be transferred to larger arrays; parameters optimized for the bulk can be reused on bigger systems with improved squeezing.","Since the relative precision of the cost estimator saturates as $N$ grows, the feedback loop does not need more measurements per evaluation at larger system sizes.","The optimal circuit depth is set by the noise level: deeper circuits give better ideal squeezing, but noise-affected results identify a finite optimal depth $n$.","The method extends beyond squeezing to variationally optimizing other metrological cost functions, such as Fisher information, as the authors outline in their outlook."],"supporting_citations":[{"why":"Defines one-axis twisting and two-axis twisting, whose squeezing limits are the baselines the variational circuit is compared against.","marker":"[10]"},{"why":"Introduces finite-range one-axis twisting for Rydberg-dressed interactions, the known protocol the variational results must surpass.","marker":"[11]"},{"why":"Defines the spin-squeezing parameter and relates it to Ramsey phase sensitivity, supplying the cost function of the optimization.","marker":"[26]"},{"why":"Provides the experimentally demonstrated hybrid classical-quantum optimization loop and the derivative-free search algorithm used in the feedback simulation.","marker":"[19]"},{"why":"Supplies the review-level connection between spin squeezing, quantum metrology, and transient non-Gaussian states used in the analysis.","marker":"[23]"},{"why":"Shows that time-dependent twisting dynamics with $J_z^2$ and $J_x$ can reach Heisenberg-limited squeezing in the symmetric subspace, the performance target approached by the finite-range circuits.","marker":"[46]"},{"why":"Provides the matrix-product-state method used to simulate the one-dimensional arrays up to 150 sites.","marker":"[70]"}],"fun_headline_variants":["Variational squeezing beats standard protocols on atom arrays","On-device variational loop improves spin-squeezing beyond known protocols","Feedback-optimized squeezing surpasses standard metrology states","Variational squeezing on tweezers beats known limits","Rydberg-dressed arrays: on-device loop yields better squeezing"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The feedback loop assumes that the variance measured along the fixed $y$-axis is the same as the smallest transverse variance that defines the spin-squeezing parameter; the paper pins the Bloch vector along $x$ but never checks that the optimal squeezed axis is $y$.","fun_headline_variants_meta":{"raw":{"variants":["Variational squeezing beats standard protocols on atom arrays","On-device variational loop improves spin-squeezing beyond known protocols","Feedback-optimized squeezing surpasses standard metrology states","Variational squeezing on tweezers beats known limits","Rydberg-dressed arrays: on-device loop yields better squeezing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000657,"raw_usage":{"total_tokens":2971,"prompt_tokens":875,"completion_tokens":2096,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":491,"completion_tokens_details":{"reasoning_tokens":2013}},"tokens_in":491,"tokens_out":2096,"duration_ms":15650,"temperature":1.0,"reasoning_tokens":2013,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:42:00.838162+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Perform the feedback-loop optimization, then measure the collective-spin covariance matrix in the $y$-$z$ plane for the optimal parameters; if the minimum-variance direction is measurably away from $y$, the cost actually optimized on the device is not the spin-squeezing parameter quoted in the paper.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines one-axis twisting and two-axis twisting, whose squeezing limits are the baselines the variational circuit is compared against."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces finite-range one-axis twisting for Rydberg-dressed interactions, the known protocol the variational results must surpass."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the spin-squeezing parameter and relates it to Ramsey phase sensitivity, supplying the cost function of the optimization."},{"cited_title":"Pichler, T","cited_arxiv_id":null,"evidence_quote":"Shows that time-dependent twisting dynamics with $J_z^2$ and $J_x$ can reach Heisenberg-limited squeezing in the symmetric subspace, the performance target approached by the finite-range circuits."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the matrix-product-state method used to simulate the one-dimensional arrays up to 150 sites."}],"review_version":1}