{"id":"32545ad3-6569-4eee-8bb7-6a50f21d77c7","arxiv_id":"2608.11811","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Randomized homodyne and heterodyne measurements are processed into shadow estimators of continuous-variable state moments, enabling entanglement and loss detection with a few thousand samples.","lead":"The authors show how to estimate many properties of continuous-variable quantum states from randomized homodyne and heterodyne measurements, using moment-generating functions instead of full state reconstruction. The protocol is tested on simulated entanglement detection and optical loss characterization.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The homodyne estimator as defined in Eqs. (19)-(20) is not unbiased: converting the Radon identity (18) into a sample average requires a factor pi (the inverse angle density), and Eq. (23) omits the (-1)^k factors from Eq. (15).","rationale":"The reader identified the normalization of P_theta as the weakest assumption, and that concern is real but under-specified. The more decisive issue is a concrete prefactor error in the estimator definition: Eq. (20) uses the integral-identity coefficient without dividing by the sampling density of theta, producing a systematic 1/pi bias, and Eq. (23) drops the sign factors required by Eq. (15). This affects the core unbiasedness claim directly, so it is load-bearing. I do not regard it as fatal: the numerical demonstrations almost certainly used the corrected kernel, since they match analytic values, and the fix is a matter of reporting the estimator correctly. The appropriate response is to keep the CONDITIONAL verdict with an explicit request to correct Eqs. (19)-(20) and (23), state the sampling density of theta, and add a Fock-state sanity check. This does not change the reader's verdict, hence UNCHANGED.","tokens_in":29486,"tokens_out":19880,"duration_ms":209895,"concrete_test":"Implement Eq. (20) for the Fock state |1> with k=1, l=0: sample theta uniformly on [0,pi), sample x from P_theta(x)=|<x|1>|^2, and average the single-shot estimator. With the printed coefficient 1/(4*pi) the average tends to 1/pi, not 1; with coefficient 1/4, and with the (-1)^k rule from Eq. (15) applied in Eq. (23), the average tends to 1. Re-running the Fig. 4 photon-subtracted simulation with the corrected kernel should reproduce the analytic SV determinant; with the printed kernel all estimated moments are scaled by 1/pi.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the homodyne MGF shadow yields unbiased moments depends on the prefactor in Eqs. (19)-(20). Eq. (18) gives the moment as C0 * integral_0^pi dtheta integral dx P_theta(x) H_{2k+l}(x) e^{-il theta}, with C0 = [pi sqrt(2^{2k+l}) binom(2k+l,k+l)]^{-1}. For randomized homodyne, theta is drawn uniformly from [0,pi), so the sample average of the integrand includes the angle density 1/pi; the unbiased per-sample kernel is pi*C0*H*e^{-il theta}, not C0*H*e^{-il theta} as written in Eq. (20). This introduces a systematic factor 1/pi in every estimated moment. For the Fock state |1> with k=1, l=0, Eq. (20) converges to 1/pi instead of <a^dagger a>=1. Separately, Eq. (15) relates moments to (-1)^k times derivatives of M_l; Eq. (23) drops these signs, biasing odd-k moments. These are internal inconsistencies in the estimator definition, not merely missing simulation details, and they would be caught by a single Fock-state test. The supplementary variance calculation inherits the same prefactor, so the sample-complexity claims are affected as well.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes a shadow-tomography protocol for estimating continuous-variable state moments from randomized homodyne and heterodyne measurements. For homodyne detection it defines a moment-generating function built from Hermite polynomials and derives single-shot estimators; for heterodyne detection it uses the characteristic function and a polynomial kernel. It gives Bernstein-based sample-complexity bounds, claims a variance that grows as a power of the Fock cutoff and exponentially in the moment order, and demonstrates entanglement detection via the Shchukin-Vogel criterion and optical-loss characterization with a few thousand simulated randomized measurements. The central claim is that the resulting shadows allow efficient concurrent estimation of many multimode moments.","tokens_in":29749,"tokens_out":19167,"duration_ms":211732,"significance":"The problem addressed is timely, and the idea of estimating normally ordered moments directly from randomized homodyne samples through a moment-generating function is natural and potentially useful; a correct version would be a worthwhile contribution to CV shadow tomography. The paper also deserves credit for connecting the framework to concrete applications, namely the Shchukin-Vogel entanglement criterion and optical-loss characterization, and for reporting simulation studies on non-Gaussian states. However, the estimator definitions in the homodyne section contain internal prefactor and sign inconsistencies, and the complexity argument relies on an unjustified support truncation. These issues affect the core unbiasedness claims and the sample-complexity guarantees, so the manuscript cannot be accepted in its current form; because the problems are local and potentially correctable, a careful revision could repair them.","major_comments":[{"comment":"The homodyne estimator as written is not unbiased. Eq. (18) expresses the moment as C times the double integral over theta in [0,pi) and x, with C = [pi sqrt(2^(2k+l) binom(2k+l,k+l))]^(-1). If theta is sampled uniformly from [0,pi), the expectation of a per-sample function f(theta,x) is (1/pi) times that integral, so the unbiased per-sample kernel must contain an explicit factor pi in the numerator. The k-th derivative of Eq. (20) has coefficient [(-1)^k pi sqrt(...)]^(-1) = (-1)^k / (pi sqrt(...)), which is too small by a factor pi and carries an extra (-1)^k relative to Eq. (18). Eq. (23) and Eq. (24) then take derivatives without the (-1)^k prefactor required by Eq. (15), so the final estimator is off by both a factor pi and a sign (-1)^k. For a Fock state |1> with k=1, l=0, the estimator converges to -1/pi instead of +1. The supplementary text is also internally inconsistent: Eq. (A25) uses C without the missing pi and without (-1)^k, while the variance calculation immediately after it inserts a factor pi. A single Fock-state calibration run would expose this error, and the claimed agreement of the simulations with analytic values cannot be checked without the exact estimator used in the code.","section":"Section III.B, Eqs. (18), (20), (23), (24)"},{"comment":"The Bernstein-based sample-complexity bound is not justified as stated. Bernstein's inequality requires the random variable Z to be almost surely bounded, with R a bound on |Z - E[Z]|. The kernels in Eqs. (20) and (30) are Hermite polynomials or polynomials in alpha, which are unbounded on the infinite phase space that homodyne and heterodyne measurements actually sample. The paper introduces a phase-space truncation Omega in Eqs. (37) and (39) and sets R = 2|C H(Omega)|, but it never states that outcomes outside [-Omega, Omega] have zero probability, nor does it provide a tail bound on P_theta or Q that would make the truncation valid. If such outcomes occur, R underestimates the true range and the confidence statement in Eq. (25) does not follow. As written, the required-measurement formula Eq. (27) is therefore an upper bound only under an additional support assumption that is neither stated nor controlled.","section":"Section III.C, Eqs. (26)-(27), (37), (39)"},{"comment":"The actual heterodyne estimator is left ambiguous. Eq. (12) defines moments as derivatives of chi_1 at alpha=0, and the text says these derivatives are obtained by numerical differentiation of stored characteristic-function values around the phase-space point alpha=0. The kernel estimator in Eqs. (29)-(30) and the variance bound in Eq. (36), however, do not involve any finite-difference step; they are direct expectation estimators over the Husimi-Q distribution. The finite-difference route introduces a systematic bias and a step-size-dependent variance that are not analyzed, and it is not connected to the complexity bound. Because the heterodyne numerical demonstrations and the few-thousand-measurements claim depend on which of these two estimators was actually used, the manuscript needs to state the estimator explicitly, give the step-size selection, and provide the associated bias-variance or complexity analysis.","section":"Section III.A, Eqs. (12), (29), and the numerical-differentiation discussion"}],"minor_comments":[{"comment":"The normalization of P_theta is not specified consistently with Eq. (18); the 1/2 prefactor in Eq. (5) should be reconciled with the quadrature distribution used in the Radon identity.","section":"Section II.A, Eq. (5)"},{"comment":"The multiple-testing issue for searching over many SV submatrices is mentioned but not addressed; the protocol-level claim that many submatrices can be evaluated without correction is not supported, even though the fixed-submatrix demonstrations are valid as illustrative examples.","section":"Section IV.A"},{"comment":"The statement that the required number of measurements grows exponentially in the moment order should be qualified by the fixed Fock truncation D and by the state-dependence of the variance bounds; the abstract's phrasing is too strong without this qualification.","section":"Section III.C and abstract"},{"comment":"The notation for the multimode MGF shadow is inconsistent: the text uses both M with a tilde and a typeset 'fM' for the same object; please standardize the notation.","section":"Eq. (22)"},{"comment":"The missing heterodyne data point for the photon-subtracted state is attributed to a limitation of Strawberry Fields; either explain why no alternative simulation method was used or provide a result from another simulator so the heterodyne comparison is complete.","section":"Figure 6"},{"comment":"No code or data repository is provided; given the estimator inconsistencies described above, the numerical agreement with analytic values cannot be independently audited, and the authors should make the simulation code available or at least state the exact estimator implemented.","section":"Reproducibility"}],"recommendation":"major_revision","confidential_remarks":"The central estimator equations are internally inconsistent enough that the numerical agreement claim is not verifiable as written. The authors should be asked to supply the precise estimator used in the simulations and to correct the prefactor and sign issues in Eqs. (20), (23), and (24), or to explicitly define a different convention. The relationship to the authors' related arXiv:2606.28698 should also be clarified, since the present manuscript cites it but does not discuss the overlap in contributions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The main thing you should know: as written, the homodyne estimator is not unbiased. Eq. (20) is missing a factor of π (the inverse angle density) and Eq. (23) drops the (-1)^k signs from Eq. (15). A Fock-state test would catch both immediately. The underlying idea is good, and the simulations suggest the code is correct, but the paper's own equations are misleading.\n\nThe useful part is real: replacing the singular characteristic function with an MGF whose sampling kernel is a finite Hermite sum (Eqs. 19–20) is a genuinely nice trick, and it avoids the pattern-function bottleneck. The variance calculation in the supplement, leading to the hypergeometric bound, is a genuine technical contribution. The two applications—SV entanglement detection and loss characterization—are well chosen, and the simulations match analytic values.\n\nThe problem is load-bearing. Eq. (18) is an integral over θ with unit weight; when θ is drawn uniformly from [0,π), the unbiased per-sample kernel acquires an extra π. Eq. (20) uses the bare prefactor, so every homodyne moment is low by 1/π—for the Fock state |1⟩, â†â would be 1/π rather than 1. Eq. (23) similarly omits the (-1)^k required by Eq. (15), giving the wrong sign for odd-k moments. These are internal inconsistencies, not missing simulation details. Because the simulations agree with exact values, I suspect the authors implemented the correct normalization and wrote the wrong equations; but readers will implement from the equations.\n\nSecondary issues: the Bernstein sample-complexity argument uses an unbounded estimator with a phase-space cutoff and no tail control, so the advertised bound is not established. The heterodyne variance bound relies on loose assumptions (all diagonal density-matrix entries set to 1, off-diagonals to 1/2) that are acknowledged but not rigorous. The novelty relative to the authors' own arXiv:2606.28698 is not clearly delineated, and no code is released.\n\nVerdict: conditional. The protocol deserves refereeing, but the normalization and signs must be fixed, the tail issue addressed or explicitly scoped, and code or precise pseudocode supplied. I'd send it to a serious referee, with a note asking the authors to reconcile Eqs. (20) and (23) with their own simulations.","headline":"Useful homodyne MGF shadow idea, but the written estimator is biased by a missing 1/π normalization and the (-1)^k signs; the simulations contradict the equations.","tokens_in":30305,"tokens_out":11974,"would_cite":false,"duration_ms":114328,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Randomized homodyne and heterodyne measurements can estimate continuous-variable state moments directly, without reconstructing the state.","keywords":["continuous-variable quantum states","shadow tomography","moment-generating functions","randomized homodyne measurements","heterodyne detection","Shchukin-Vogel criterion","optical loss characterization","sample complexity"],"falsifier":"Prepare a coherent state with known displacement $\\alpha$ and known moments, run the proposed homodyne protocol with a detector whose quadrature normalization is independently calibrated, and compare the estimated $\\langle \\hat a^{\\dagger k+l}\\hat a^k\\rangle$ against the analytic values for several orders; any systematic multiplicative bias, visible already at $k+l\\le 2$, would falsify the estimator's unbiasedness.","tokens_in":29263,"feed_emoji":"⚛️","tokens_out":5924,"duration_ms":56157,"temperature":0.7,"pith_summary":"The paper proposes a shadow-tomography protocol that estimates the moments of continuous-variable quantum states—expectation values of products of photon creation and annihilation operators—directly from randomized homodyne or heterodyne measurements. Instead of reconstructing the density matrix or the Wigner function, the protocol builds an unbiased estimator of a moment-generating function (MGF) from single-shot samples, so that many moments of many modes can be read off concurrently with minimal postprocessing. The authors prove the sample complexity grows exponentially in the moment order but only polynomially in the Fock truncation, and they demonstrate the protocol on entanglement detection and optical-loss characterization with a few thousand randomized measurements. A sympathetic reader would care because most CV simulation and verification tasks are written in the language of moments, and this offers a measurement-first, analyze-later route to them.","feed_headline":"A few thousand random measurements map CV state moments","feed_subtitle":"New estimator reads many multimode moments from random homodyne and heterodyne shots, then catches entanglement and loss.","key_machinery":"The load-bearing object is the single-shot MGF estimator $M_l^{(i)}(\\mu) = \\sum_{k=0}^{L} \\frac{\\mu^k}{k!} C_{k,l} H_{2k+l}(x_\\theta^{(i)}) e^{-il\\theta^{(i)}}$, where $H_n$ is a Hermite polynomial, $C_{k,l}$ is a closed-form constant, and $(x_\\theta, \\theta)$ is a single homodyne sample. It replaces the pattern-function-based sampling function $S_M(x_\\theta,\\mu,l)$ with an infinite series that can be truncated at the highest desired moment, making postprocessing nearly free and avoiding the singularity at $\\mu=0$. Judiciously, in multimode settings it works with the factorized representation $\\prod_m M_{l_m}^{(i)}(\\mu_m)$, whose memory footprint grows linearly in the number of modes.","core_discovery":"The central claim is that the full moment structure of a CV state can be recovered from randomized phase-space measurements without reconstructing the state. For heterodyne measurements the usual characteristic-function shadow suffices; for homodyne measurements, whose samples lie on one-dimensional slices and whose characteristic function would be a distribution, the paper introduces a new MGF $M_l(\\mu)$ that is regular everywhere and whose single-shot estimator is a sum of Hermite polynomials with explicit coefficients. From this MGF, normally-ordered moments $\\langle \\hat a^{\\dagger k+l}\\hat a^k\\rangle$ are obtained by taking derivatives at $\\mu=0$, and a multimode moment is an average of products of single-mode snapshots. The complexity analysis shows the estimator's variance grows as $O(D^{2k+l})$ with Fock truncation $D$ and moment order $2k+l$, so that low-order moments remain estimable even at high energy, and the two demonstrations show that a few thousand randomized measurements suffice to detect entanglement and to map single-mode loss.","pith_inferences":["If the normalization issue is benign, the same estimator can be adapted to any rotated quadrature sampling, including noisy or miscalibrated homodyne, by folding the detector's known response into the kernel.","The paper's own note about testing multiple SV submatrices suggests that applying the protocol to unknown states should be paired with a multiple-testing correction or a fixed pre-chosen submatrix; without it, apparent violations may occur by chance.","The exponential-in-order variance suggests a natural extension: importance-sampling the quadrature angle or the phase-space point to flatten the estimator variance, which could extend the reachable moment order at fixed sample count.","Because the moments are estimated without state reconstruction, the same shadows could feed moment-based Hamiltonian simulation or variational algorithms, bypassing the density-matrix reconstruction entirely."],"forward_implications":["Many moments of a multimode CV state can be estimated concurrently from a single batch of randomized measurements, with per-sample postprocessing that is essentially free.","The sample complexity grows exponentially in moment order, so low-order moments stay accessible for high-energy states.","Entanglement of Gaussian and non-Gaussian states can be certified with a few thousand randomized homodyne or heterodyne measurements using the SV submatrix determinants.","Optical loss on a specific mode of a photonic chip can be detected from the exponential decay pattern of the covariance matrix estimated via the protocol.","The factorized form of the shadow keeps memory linear in the number of modes, allowing the protocol to scale to many-mode systems."],"supporting_citations":[{"why":"Supplies the heterodyne characteristic-function shadow that the paper builds on and contrasts with for the homodyne case.","marker":"[16]"},{"why":"Provides the Radon-transform identity (Eq. 18) used to derive the homodyne moment estimator kernel.","marker":"[21]"},{"why":"Defines the original MGF $M_l(\\mu)$ and pattern-function sampling that the new estimator simplifies and extends.","marker":"[29]"},{"why":"Supplies the infinite-moment entanglement criterion (SV submatrix determinants) used in the demonstrations.","marker":"[19]"},{"why":"Provides Bernstein's inequality used for the sample-complexity bounds.","marker":"[33]"},{"why":"Supplies the characteristic-function and coherent-state machinery used throughout the derivations.","marker":"[18]"}],"fun_headline_variants":["Few thousand random shots read out CV moments","Random homodyne and heterodyne shots yield CV moments","Few thousand random measurements map CV moments","Randomized phase-space shots estimate CV moments","CV moment shadows from few thousand random shots"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The homodyne estimator is unbiased only if the quadrature distribution $P_\\theta(x_\\theta)$ sampled by the device is normalized exactly as in the Radon-transform identity used to derive the kernel; the paper does not fix that normalization, and a constant prefactor error would shift every estimated moment.","fun_headline_variants_meta":{"raw":{"variants":["Few thousand random shots read out CV moments","Random homodyne and heterodyne shots yield CV moments","Few thousand random measurements map CV moments","Randomized phase-space shots estimate CV moments","CV moment shadows from few thousand random shots"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.002105,"raw_usage":{"total_tokens":8185,"prompt_tokens":955,"completion_tokens":7230,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":571,"completion_tokens_details":{"reasoning_tokens":7161}},"tokens_in":571,"tokens_out":7230,"duration_ms":52989,"temperature":1.0,"reasoning_tokens":7161,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:28:52.181049+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Prepare a coherent state with known displacement $\\alpha$ and known moments, run the proposed homodyne protocol with a detector whose quadrature normalization is independently calibrated, and compare the estimated $\\langle \\hat a^{\\dagger k+l}\\hat a^k\\rangle$ against the analytic values for several orders; any systematic multiplicative bias, visible already at $k+l\\le 2$, would falsify the estimator's unbiasedness.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the heterodyne characteristic-function shadow that the paper builds on and contrasts with for the homodyne case."},{"cited_title":"Mølmer, Non-Gaussian states from continuous-wave Gaussian light sources, Physical Review A73, 063804 (2006)","cited_arxiv_id":null,"evidence_quote":"Provides the Radon-transform identity (Eq. 18) used to derive the homodyne moment estimator kernel."},{"cited_title":"Shchukin and W","cited_arxiv_id":null,"evidence_quote":"Defines the original MGF $M_l(\\mu)$ and pattern-function sampling that the new estimator simplifies and extends."},{"cited_title":"Quantum computation and simulation with vibrational modes of trapped ions","cited_arxiv_id":"2103.14299","evidence_quote":"Provides Bernstein's inequality used for the sample-complexity bounds."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the characteristic-function and coherent-state machinery used throughout the derivations."}],"review_version":1}