{"id":"6e147bec-51cc-4311-b0f8-665e9c825246","arxiv_id":"2607.29342","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Photonic-lantern wavefront sensing has a per-photon Fisher-information ceiling beta<=2, and the standard photon-noise sensitivity s_gamma equals only the Fisher diagonal, making it optimistic for mode-mixing devices.","lead":"A framework for computing the fundamental information limit of photonic-lantern wavefront sensors using Fisher and quantum-Fisher information. It produces a device-agnostic sensitivity metric and shows a commonly used photon-noise metric overstates performance when modes mix.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 3.5's 'averaged CRLB' inverts E[F(a0)], which by Jensen is an optimistic lower bound on the mean per-frame CRLB E[F(a0)^{-1}]; since Figs. 1–4 and the β_i comparison use this quantity without flagging the gap, the quantitative sensitivity claims may overstate lantern performance.","rationale":"The reader's weakest assumption is precisely the ensemble-average issue: inverting E[F(a0)] and calling it an 'averaged CRLB' is optimistic relative to E[F(a0)^{-1}]. My stress-test confirms this is the single most load-bearing concern for the quantitative claims. The theoretical contribution — the Fisher/CRLB framework, the s_gamma = F_ii identity, and the quantum bound — is mathematically sound and per-frame, so it is not invalidated. However, the illustrative application and figures present numbers that are not lower bounds on the average estimator variance, and the paper does not flag this. The reader already rendered a CONDITIONAL verdict, which appropriately captures the need to clarify or correct the averaged-CRLB interpretation before the quantitative sensitivity values are taken at face value. I therefore see no reason to change the reader's verdict; the concern strengthens the condition. The proposed test would settle the magnitude of the optimism and dictate whether relabeling or recomputation is necessary.","tokens_in":12402,"tokens_out":12166,"duration_ms":135925,"concrete_test":"Recompute the ensemble-averaged CRLB for the same set of operating points used to generate Figs. 1–4 by explicitly evaluating the sample mean of the per-frame inverse FIM diagonals, (1/K)Σ_k (F(a_k)^{-1})_{ii}, and compare with (\\bar F)^{-1}_{ii} for the displayed modes (e.g., Z2, Z4, Z7). If the fractional difference between these two quantities exceeds 10% for any low-order mode, the 'averaged CRLB' label and the β_i values should be revised to the per-frame-averaged bound or explicitly relabeled as an optimistic design metric.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's ensemble-averaged treatment (Sect. 3.5) defines \\bar F = E_{a0}[F(a0)] and labels diag[(\\bar F)^{-1}] as the 'averaged CRLB' (Figs. 1, 2, 4). For a parameter that varies frame to frame, the mean of the per-frame Cramér–Rao bounds is E[(F(a0)^{-1})_{ii}], not [(\\bar F)^{-1}]_{ii}. Jensen's inequality gives (\\bar F)^{-1} \\preceq E[F^{-1}], so the plotted curves are optimistic: they lie below the true average per-frame variance. The paper explicitly calls \\bar F an 'averaged design information measure' and notes its inverse can constrain more than N−1 modes, but it does not disclose the direction or magnitude of the discrepancy with the mean per-frame CRLB. Because the β_i values in Eq. (10) and the cross-sensor comparison in Fig. 3 are computed from this inverted average, the quantitative sensitivity claims are not guarded as optimistic. The per-frame relations (Eqs. 17–18) are unaffected, but the illustrative application and its figure set are presented without this caveat.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript develops a Fisher-information and Cramér–Rao framework for photonic-lantern wavefront sensing. The lantern is treated as a deterministic map from Zernike coefficients to N port intensities; Poisson and read-noise Fisher information matrices are derived, the CRLB is expressed through the diagonal of the inverse per-photon FIM, and a dimensionless sensitivity β is introduced. The paper then computes the pure-state quantum Fisher information matrix, proves joint saturability of the multi-parameter QCRB via vanishing mean Uhlmann curvature, and establishes exact relations to the Chambouleyron et al. photon-noise sensitivity s_γ and to the Haffert et al. quantum/classical limit. The framework is illustrated with a synthetic Gaussian-receiver lantern model for N=7,19,37,91 ports.","tokens_in":12738,"tokens_out":12786,"duration_ms":130744,"significance":"If the claims hold, the paper provides a useful algorithm-independent benchmark for lantern wavefront sensors and unifies three existing information-theoretic treatments on a common β scale. The exact identity s_γ² = \tilde F_ii and the ordering β_i ≤ √η s_γ are clean and correct, and the explicit QFI calculation with the joint-saturability argument is a genuine addition to the literature. The release of the analysis code is a strength. The main reservations concern the handling of the ensemble-averaged CRLB, which is used in all quantitative figures, and an inconsistency in the role of throughput η relative to N_ph; these affect the numerical sensitivity claims and the cross-sensor comparison.","major_comments":[{"comment":"The quantity diag[(E_{a0}[F(a0)])^{-1}] is labelled the 'averaged CRLB' and is used for the curves in Figs. 1, 2 and 4 and for the β_i values in Fig. 3. For a parameter that varies frame to frame, the mean of the per-frame bounds is E_{a0}[(F(a0)^{-1})_{ii}], not [(E_{a0}[F(a0)])^{-1}]_{ii}. By Jensen's inequality (or convexity of the inverse on positive-definite matrices), (E F)^{-1} ⪯ E[F^{-1}], so the plotted 'averaged CRLB' is an optimistic lower bound on the actual mean per-frame variance, and the paper does not disclose the direction or size of the gap. Please either replace it with a Monte Carlo average of per-frame inverse FIMs, or present ar F^{-1} strictly as a design-information measure with the caveat that it understates the true averaged variance, and quantify the discrepancy for the model.","section":"Sec. 3.5, Figs. 1–4"},{"comment":"The role of N_ph and throughput η is inconsistent. Eq. (3) calls N_ph 'total detected flux', but then writes µ_k = η N_ph p_k; Eq. (8) has F = η N_ph \tilde F; Eq. (10) defines β with a factor √η; yet Eq. (12) and Sect. 5.2 quote the quantum bound σ ≥ 1/(2√N_ph) with no η. Consequently the claimed ceiling β = 2 is only correct if η = 1. If N_ph is the incident photon number, the quantum bound should be 1/(2√(η N_ph)) and the ceiling becomes 2√η; if N_ph is the detected photon number, η should not appear in Eqs. (8) and (10). This ambiguity affects every comparison on the common β ≤ 2 scale and must be resolved before the quantitative claims can be accepted.","section":"Eqs. (3), (8), (10), (12), (18); Table 1"},{"comment":"All numerical results use a synthetic focal-plane Gaussian-receiver lantern that is explicitly 'not to reproduce a specific fabricated device'. The paper is honest about this, but the captions and conclusion still present the resulting parity structure and port-count saturation as lantern properties. Because a real lantern's inter-modal mixing can change the parity structure (as the paper notes), the reader cannot tell which numerical conclusions are robust. At minimum, the figure captions should state that the plotted curves are model-illustration results, and the discussion should explicitly avoid generalizing the even-mode weakness or the 'saturation at 19 ports' behavior beyond the model.","section":"Sec. 6, Figs. 1–4"}],"minor_comments":[{"comment":"The threshold λ_thr = 10^{-6} λ_max is arbitrary. The effective modal capacity M_eff depends on this choice; please either justify it, state the sensitivity of M_eff to the threshold, or present M_eff only as an illustrative convention.","section":"Sec. 3.5"},{"comment":"The comparison values for pyramid, Zernike and Shack–Hartmann sensors are shown without derivation or references in the caption. Please state where these values come from or how they were computed, so the comparison is reproducible.","section":"Fig. 3"},{"comment":"The model parameters (η, λ, aperture diameter, Gaussian receiver width, finite-difference step, N_ph ranges) are not given in the text or captions. The code is released, but the reader should be able to reproduce Figures 1–4 from the description alone.","section":"Sec. 6"},{"comment":"The notation for s_γ in Eq. (16) is ambiguous in the rendered text; it should be written explicitly as the squared L2 norm ||δI(φ_i)/√I_0||_2². Also, the definition of I_0 as p_k(0) should be stated before Eq. (16), not only in Appendix B.","section":"Eq. (16), Appendix B"},{"comment":"The statement that N_ph is 'total detected flux' is confusing when followed by the throughput factor η. Clarify the bookkeeping of incident versus detected photons, and make the symbols consistent in Eq. (3), Eq. (8), Eq. (10) and Eq. (12).","section":"Sec. 4.1"}],"recommendation":"major_revision","confidential_remarks":"The core statistical identities and the QFI calculation are correct, so rejection is not warranted. However, the ensemble-averaged CRLB problem is not merely cosmetic: it affects all quantitative figures and the central comparison to s_γ. The N_ph/η ambiguity must also be fixed before the β ≤ 2 ceiling and the cross-sensor comparison can be trusted. The paper is otherwise well structured and the author is appropriately candid about the synthetic model; the revision should focus on these two points."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this paper deserves a serious referee. The core theoretical contribution holds up, and the comparison with Chambouleyron and Haffert is genuinely clarifying.\n\nWhat is actually new: this is the first explicit Fisher/QFI treatment of the photonic-lantern WFS. The precise identity s_gamma^2 = F_ii (Eq. 17) and the inequality beta_i <= sqrt(eta) s_gamma (Eq. 18) are clean and correct. That distinction — diagonal of the FIM versus diagonal of its inverse — is the kind of observation that should be canon in the field. The joint-saturability argument via vanishing Uhlmann curvature is standard but the explicit statement for multiple phase modes is a useful addition. The capacity ceiling M_eff <= N-1 and the parity discussion in the synthetic model are also reasonable.\n\nThe classical Fisher and pure-state QFI derivations are internally consistent. The paper is honest about what is adopted versus new.\n\nThe biggest soft spot is exactly the one the stress-test note flags. Section 3.5 defines \\bar F = E[F(a0)] and calls diag(\\bar F^-1) an 'averaged CRLB', then plots it in Figs. 1-4. But the mean per-frame CRLB is E[F(a0)^-1], and Jensen gives E[F]^-1 <= E[F^-1]. So the plotted bound is optimistic, and the paper does not say so. The per-frame relations are unaffected, but the quantitative beta values and cross-sensor comparison inherit this optimism. This needs a caveat, and preferably the gap quantified for the synthetic model.\n\nA lesser but real concern: the numerical results use a synthetic Gaussian-receiver lantern, not a fabricated device, and the code is only available from the author with no repository or hash. The paper is upfront about this, and the framework is designed to ingest measured response matrices, so it does not invalidate the theory. But the concrete beta values in Figs. 3-4 should not be quoted as device performance.\n\nThe hand-chosen threshold lambda_thr = 1e-6 lambda_max is arbitrary but minor; it does not affect the central claims.\n\nBottom line: the theory is solid, the one quantitative omission is the Jensen gap, and the rest is cosmetic. A referee who knows Fisher information and AO will find this a useful, citable paper. I would send it to review and ask for a clarifying paragraph on the averaged CRLB before acceptance.","headline":"The first Fisher/quantum-Fisher treatment of the lantern WFS is mostly sound and worth engaging: the diagonal-vs-inverse distinction is the real contribution, and the main quantitative caveat is the un-flagged optimistic 'averaged CRLB'.","tokens_in":13232,"tokens_out":1196,"would_cite":true,"duration_ms":17553,"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 photonic lantern's wavefront-sensing precision is fully set by its per-photon Fisher information and capped by a quantum floor of half a radian per photon.","keywords":["photonic lantern","wavefront sensing","Fisher information","Cramer-Rao bound","quantum Fisher information","adaptive optics","astrophotonics","sensitivity metric"],"falsifier":"Measure the response matrix of a real photonic lantern across many phase realizations, compute per-frame Fisher matrices and the ensemble-averaged Fisher matrix, and compare the mean of per-frame CRLBs with the diagonal of the inverse averaged Fisher matrix. If the latter is smaller by more than the Jensen gap, the averaged-CRLB claim is falsified.","tokens_in":12262,"feed_emoji":"🔭","tokens_out":5967,"duration_ms":55199,"temperature":0.7,"pith_summary":"This paper shows that the photonic lantern wavefront sensor can be characterized, independent of any reconstruction algorithm, by the Fisher information of its port intensities. The attainable precision obeys the Cramér–Rao bound: it scales as one over the square root of the detected photon number and, mode by mode, cannot beat a quantum limit of one half radian root-mean-square per photon. The paper further proves that the multi-mode quantum bound is jointly attainable—estimating several Zernike modes at once carries no quantum penalty—and that a widely used sensitivity metric, which takes the diagonal of the Fisher matrix, is optimistic for a mode-mixing lantern because the rigorous bound uses the diagonal of the matrix inverse. The result is a common per-photon scale on which lanterns, pyramid, and Zernike-type sensors can be compared.","feed_headline":"Lantern wavefront sensing capped by quantum limit: 1/2 rad per photon","feed_subtitle":"A Fisher-information framework ranks all wavefront sensors on the same per-photon accuracy scale.","key_machinery":"The per-photon Fisher information matrix F̃_ij = Σ_k (1/p_k)(∂p_k/∂a_i)(∂p_k/∂a_j) built from the normalised port intensities p_k(a), along with its inverse. It is the pullback of the Fisher–Rao metric on the intensity simplex through the lantern map; its diagonal is the published s_γ sensitivity, its inverse diagonal sets the CRLB, and its quantum ceiling is fixed by the pure-state QFI F_Q=4 Cov(Ẑ_i,Ẑ_j) with vanishing mean Uhlmann curvature.","core_discovery":"The central claim is that the photonic lantern, treated as a deterministic map from aberration coefficients to N output intensities, has its estimator-independent precision governed by the per-photon Fisher information matrix. The Cramér–Rao lower bound for any unbiased estimator scales as N_ph^{-1/2} and is bounded mode-by-mode by the quantum Cramér–Rao bound, corresponding to a sensitivity of β=2, i.e. 1/(2√N_ph) rad rms. Because the phase generators are real and commuting, the mean Uhlmann curvature vanishes, so the multi-parameter quantum bound is jointly saturable: all low-order modes can reach the half-radian-per-photon limit simultaneously with no quantum incompatibility. The paper al","pith_inferences":["The paper's averaged CRLB uses the inverse of the ensemble-averaged Fisher matrix, but Jensen's inequality implies the mean of per-frame CRLBs is at least as large; the plotted closed-loop bounds are therefore optimistic as statements about average estimator variance.","The quantitative β values come from a synthetic Gaussian-receiver lantern model with a specific parity structure; a real fabricated lantern's mode mixing will likely alter even-mode sensitivity, so the numbers should be read as illustrating the framework, not as a device prediction.","The analysis suggests a natural design criterion: choose the lantern port layout that diagonalizes or concentrates the per-photon Fisher matrix, since off-diagonal information is what pushes the CRLB above the diagonal-only estimate.","Because the Fisher–Rao view makes sensing precision a geometric property, one could use this metric to optimize port count and geometry for a given set of target Zernike modes before building a device."],"forward_implications":["Any unbiased wavefront estimator reading lantern port intensities obeys σ ≥ 1/(2√N_ph) rad per mode; no reconstruction algorithm, learned or otherwise, can beat this floor.","The s_γ sensitivity used for Fourier-filtering sensors is exactly the diagonal of the per-photon Fisher matrix, so for a mode-mixing lantern it overestimates achievable precision; the full inverse matrix must be used.","A lantern with N single-mode outputs constrains at most N−1 aberration modes per exposure, so extra ports buy high-order capacity, not low-order precision once the PSF core is sampled.","The half-radian-per-photon quantum ceiling is jointly attainable for all low-order modes simultaneously, not just one mode at a time, because no quantum incompatibility exists between phase modes.","The framework yields a device-agnostic metric β on a common 0≤β≤2 scale, making lanterns, pyramid, Zernike and PIAA-ZWFS sensors directly comparable."],"fun_headline_variants":["Quantum limit reachable for all low-order modes at once","Half-radian-per-photon limit applies to every lantern mode","Lantern sensing hits 1/2 rad per photon quantum bound","Fisher framework unifies wavefront sensor accuracy scales"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that inverting the ensemble-averaged Fisher information matrix correctly gives the closed-loop sensitivity; since the inverse of an average is not the average of inverses, this can make the reported precision optimistic, and the concrete numbers also rest on a synthetic lantern model rather than a fabricated device.","fun_headline_variants_meta":{"raw":{"variants":["Quantum limit reachable for all low-order modes at once","Half-radian-per-photon limit applies to every lantern mode","Lantern sensing hits 1/2 rad per photon quantum bound","Fisher framework unifies wavefront sensor accuracy scales"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00086,"raw_usage":{"total_tokens":3667,"prompt_tokens":939,"completion_tokens":2728,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":683,"completion_tokens_details":{"reasoning_tokens":2660}},"tokens_in":683,"tokens_out":2728,"duration_ms":20897,"temperature":1.0,"reasoning_tokens":2660,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T08:57:17.836757+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the response matrix of a real photonic lantern across many phase realizations, compute per-frame Fisher matrices and the ensemble-averaged Fisher matrix, and compare the mean of per-frame CRLBs with the diagonal of the inverse averaged Fisher matrix. If the latter is smaller by more than the Jensen gap, the averaged-CRLB claim is falsified.","supporting_citations":[],"review_version":1}