{"id":"1c6bdf24-dfd1-416b-82f5-5e8a1975d72a","arxiv_id":"2506.07384","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For non-degenerate two-photon absorption, intensity-correlation measurements with double-seeded two-mode squeezed light achieve the best precision scaling (1/n^3.5), while normalized intensity correlation trades away quantum enhancement for robustness to photon loss.","lead":"This paper compares three ways of measuring how strongly a material absorbs two photons at once, using specially squeezed pairs of light beams. It finds that measuring the product of the intensities in the two beams gives the most precise result, unless losses are high, in which case a normalized version is safer but gives up the quantum advantage.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The g(1,1)-robust/G(1,1)-best trade-off is only tested for identical losses η1=η2 (Eq. 11); unequal losses at the two non-degenerate frequencies are physically generic and could change the comparison.","rationale":"The reader's weakest_assumption is exactly the equal-loss assumption, and I agree it is the most load-bearing gap. The paper's headline claims depend on three asymptotic scalings: G(1,1) best with 6/n_T^3.5 for double seeding, g(1,1) robust at 1/n_T^2, NRF worst at 1/n_T^2. The first and second are derived from formulas in an unprovided companion file; that is a verifiability problem, but not by itself a technical defect. The equal-loss assumption, by contrast, is an explicit modeling choice in Eq. (11) whose failure mode is physically expected in the very non-degenerate setting the paper targets. A single-η model cannot capture the different spectral responses of real optics. The proposed check—varying δ at fixed mean loss—would settle whether the trade-off survives unequal losses. If it does survive, the paper's qualitative conclusions hold; if not, the central claim should be substantially qualified. Because this is an addressable but untested condition, the conditional verdict stands.","tokens_in":15460,"tokens_out":8470,"duration_ms":104256,"concrete_test":"Generalize Eq. (11) to d1 = √η1 c1 + √(1−η1) u1, d2 = √η2 c2 + √(1−η2) u2, set η1 = η(1+δ), η2 = η(1−δ) with δ = 0.1 and 0.5 at η = 0.7, 0.9, and recompute the optimized Δε² for G(1,1), g(1,1), NRF for the double-seeded state at n_T = 100, 500, 1000 using the companion Mathematica code. If the normalized loss ratio for g(1,1) departs from 1 by more than a few percent, or if G(1,1) no longer beats g(1,1) and NRF by the reported margins, the abstract's trade-off must be restricted to balanced losses.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section II.A models all single-photon loss with a single transmittivity η (Eq. (11), Fig. 1 caption). For non-degenerate TPA, ω1≠ω2, so beam splitters, spectral filters, and detectors generally present different losses to the two modes. This is not a harmless rescaling: the observables are nonlinear functions of detected photon numbers, and the loss factors enter the covariance matrices in Eqs. (B8) and (C5) in different powers. In particular, ∂⟨G(1,1)⟩/∂ε ∝ η1η2 because the TPA signal requires one photon from each mode to survive, while Var(G(1,1)) contains terms of order η1, η2 and η1η2; hence the G(1,1) error can degrade significantly faster than the g(1,1) error when one mode is lossy. The g(1,1) ratio itself cancels η1η2 at the level of mean values, but the variance of g(1,1) and its ε-derivative depend on η1 and η2 separately, so the claimed 'robust to loss' behavior of g(1,1) (Fig. 3, Section III.E) is not guaranteed away from η1=η2. Since the comparison in Table I and the abstract's trade-off statement are made for the non-degenerate case, this equal-loss assumption is load-bearing and untested.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes the precision with which the two-photon absorption (TPA) absorbance can be estimated from transmission measurements, comparing three observables: the noise reduction factor (NRF), the intensity correlation G(1,1), and the normalized intensity correlation g(1,1). The input light is a two-mode squeezed state, considered in three configurations: squeezed vacuum, single-seeded squeezed coherent state, and double-seeded squeezed coherent state. Single-photon losses are modeled by beam splitters with a common transmittivity. The authors report that G(1,1) measurements give the best precision, with an optimized double-seeded state yielding Δϵ²_G(1,1) ≈ 6/n_T^3.5, while g(1,1) gives Δϵ² ≈ 1/n_T^2 but is robust to loss, and NRF is the worst observable. The paper includes closed-form expressions for the squeezed-vacuum case, asymptotic scaling laws for seeded states, and a discussion of the trade-off between loss robustness and quantum enhancement.","tokens_in":15714,"tokens_out":6881,"duration_ms":85072,"significance":"If the reported scalings are correct, the paper provides practically useful guidance for choosing both the input state and the measurement observable in quantum-enhanced TPA experiments. The strength of the paper is that the squeezed-vacuum results are given in closed form (Eqs. 21-24) and the error-propagation framework in Appendices A-C is clearly formulated, making the comparison falsifiable. The central claim, however, rests on seeded-state scaling laws and optimization results that are not contained in the manuscript, and the loss model assumes identical losses for the two non-degenerate modes. These issues currently limit the verifiability and generality of the main conclusions.","major_comments":[{"comment":"The central scaling laws for the seeded states, including the headline Δϵ²_G(1,1) ≈ 6/n_T^3.5 in Eq. (29), are stated to be derived in a companion Mathematica file that is not accessible from the manuscript. Neither the optimized expressions nor the asymptotic expansions are given, so the reader cannot verify that these scalings follow from the stated master equation and beam-splitter transformations. Because Table I and the abstract's trade-off conclusion depend on these results, this is a load-bearing reproducibility gap. The derivations (or a permanent, reviewed supplement) should be included in the paper.","section":"Section III.C-III.D, Eqs. (25)-(30)"},{"comment":"The loss model uses a single transmittivity η for both modes. For non-degenerate TPA, where ω1 ≠ ω2, losses in the two arms are generally different (η1 ≠ η2). The claims that g(1,1) is robust to loss and that G(1,1) remains the best under loss are only tested for η1 = η2. Since the error-propagation formulas in Appendices B and C are nonlinear functions of the detected photon-number moments, the η1 and η2 dependences do not cancel in general; for example, the TPA signal derivative in G(1,1) scales as η1η2 while the variance can contain terms with different powers of η1 and η2. The equal-loss assumption is therefore load-bearing and should be relaxed or explicitly justified for the experimental regime of interest.","section":"Section II.A, Eq. (11), and Fig. 3"},{"comment":"The optimization over the double-seeded state is not fully specified. The total photon number n_T depends on α1, α2, r, and the relative phase, but the text only states that the squeezing parameter and the relative phase are optimized. It is not stated whether α1 and α2 are fixed, varied, or constrained (e.g., α1 = α2). Without the full optimization domain, the reported 'optimized' scalings in Eqs. (28)-(30) and the phase diagrams in Fig. 6 are not well-defined. This ambiguity is central to the claimed advantage of double seeding and should be clarified.","section":"Section III.D"}],"minor_comments":[{"comment":"There are several typographical issues, including 'transmitivity' for 'transmittivity' and 'stander deviation' for 'standard deviation' in Appendix A; the paper should be proofread.","section":"Throughout"},{"comment":"G(1,1) is called the 'first-order intensity correlation function', but ⟨n1 n2⟩ is a second-order intensity correlation; the nomenclature should be aligned with standard quantum-optics usage.","section":"Section II.B"},{"comment":"The color bar is labeled 'Log Δϵ²_NRF' but the axes are not labeled in the reproduction; please add axis labels and a more complete caption.","section":"Fig. 4"},{"comment":"The reference contains a spurious space ('V ol. 2'); several other references have similar spacing or formatting inconsistencies.","section":"Reference [53]"}],"recommendation":"major_revision","confidential_remarks":"The absence of the companion Mathematica file is a serious verifiability issue for the central scaling claims. The equal-loss assumption should be addressed before acceptance; if the authors can show that the ranking is robust under η1 ≠ η2, or specify a concrete experimental regime where equal loss is a good approximation, the main conclusions would be on much firmer ground."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The short version: this is a solid theory paper that gives experimental groups a concrete comparison of three correlation measurements for non-degenerate two-photon absorption with two-mode squeezed light. The main new results are the n_T^-3.5 scaling for double-seeded G(1,1) and the phase-optimization maps, and the central claim that G(1,1) beats g(1,1) and NRF holds up under the model they use. But before relying on the trade-off story, you should know about one load-bearing simplification and one verification gap.\n\nWhat the paper does well: it extends the same group's earlier single-mode squeezed-light formalism to two-mode squeezed states and compares squeezed vacuum, single-seeded, and double-seeded inputs for NRF, G(1,1), and g(1,1). The squeezed-vacuum formulas in the text (Eqs. 21-24) are consistent, and I could verify those. The optimization procedure is systematic, and the abstract and conclusions match the actual results. The observation that g(1,1) is loss-robust but loses the quantum enhancement, while G(1,1) gives the best precision in the lossless case, is clearly stated and plausible.\n\nWhere the soft spots are: first, the loss model assumes identical transmittivity η for both modes (Eq. 11). For non-degenerate TPA, ω1 ≠ ω2, so different losses are physically generic. The stress-test note is right: the observables are nonlinear in the detected photon numbers, and the covariance matrices in Appendices B and C contain different powers of η1 and η2. So the claimed loss-robustness of g(1,1) and the full ranking of G(1,1) versus g(1,1) are only established for equal losses. This is not a nitpick; it is a real gap that could change the comparison in practice. That said, it does not obviously break the central argument, and it is addressable.\n\nSecond, the most important scaling laws for the seeded and double-seeded cases (Eqs. 25-30) are deferred to a companion Mathematica file that is not accessible. So the headline quantitative claims cannot be independently checked from the paper itself. The derivations may well be correct, but the paper needs to let referees see the key expressions.\n\nMinor point: the error-propagation formula is standard, and evaluating at ε→0 is fine. The NRF being insensitive to ε for squeezed vacuum is a known feature, not a flaw.\n\nWho this is for: experimental groups working on quantum-enhanced TPA spectroscopy or sensing, and theorists who care about optimal measurement strategies with two-mode squeezed light. It deserves a serious referee. My recommendation: send to peer review, but ask the authors to include or otherwise share the companion file and to test the ranking under unequal losses.","headline":"A useful theory comparison for non-degenerate TPA with two-mode squeezed light, but the loss model treats both modes as identical and the key scaling laws sit in an inaccessible companion file.","tokens_in":16321,"tokens_out":1739,"would_cite":true,"duration_ms":21846,"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":"The paper argues that the raw intensity correlation G(1,1) with optimized double-seeded squeezed light gives the best two-photon absorption precision, scaling as 6/n_T^3.5, while normalized g(1,1) loses the quantum gain but keeps loss…","keywords":["non-degenerate two-photon absorption","two-mode squeezed light","intensity correlation G(1,1)","normalized intensity correlation","noise reduction factor","quantum metrology","photon loss","squeezed coherent states"],"falsifier":"Measure the G(1,1) estimation error for the optimized double-seeded squeezed state at several total photon numbers in a lossless setup; if the log-log slope is not -3.5, the central scaling claim fails. Then insert unequal loss for the two modes to test whether g(1,1) remains loss-immune and whether G(1,1) stays the best readout.","tokens_in":15238,"feed_emoji":"🎯","tokens_out":8827,"duration_ms":85451,"temperature":0.7,"pith_summary":"The paper compares three ways of reading out a transmission measurement of non-degenerate two-photon absorption (a process in which one photon from each of two different frequency modes is absorbed together), using two-mode squeezed light as the probe. It claims that the unnormalized intensity correlation G(1,1), optimized over squeezing, seeding, and phase, gives the most precise estimate of the absorbance, with estimation error scaling as about 6/$n_T^{3}$.5 at large photon number. The normalized correlation g(1,1) and the noise reduction factor scale only as 1/$n_T^{2}$, and under linear photon loss g(1,1) keeps its precision but no longer exhibits any quantum advantage, whereas G(1,1) remains the best despite some degradation. If correct, this tells experimentalists which observable and which squeezed input state to use, and it exposes a trade-off between loss robustness and quantum enhancement.","feed_headline":"Intensity correlation wins for two-photon absorption","feed_subtitle":"Double-seeded squeezed light reaches 6/n^3.5 error; normalized correlation stays loss-immune but loses the quantum edge.","key_machinery":"The calculation builds on a Markovian master equation for the two-photon absorption sample, whose Lindblad operator removes one photon from each mode only when the two frequencies add up to the resonance. The smallness of two-photon absorption cross sections justifies expanding the evolution operator to first order in the absorbance, and photon loss is inserted as a beam splitter of transmittivity η acting identically on both modes. Each observable's error is obtained from error propagation, Δε² = Var(O)/(∂⟨O⟩/∂ε)² at ε=0, with the variance computed through a covariance matrix of photon-number moments. Optimization over the squeezing parameter r, the coherent seeding amplitudes, and the relative phase between seeding and squeezing is what produces the different scaling exponents collected in Table I.","core_discovery":"For non-degenerate two-photon absorption probed by two-mode squeezed light in a transmission geometry, the paper establishes that the raw intensity correlation G(1,1) is the best measurement. With a double-seeded two-mode squeezed state and optimal squeezing and phase, the error in estimating the absorbance scales as Δε²_G(1,1) ≈ 6/$n_T^{{3.5}}$ in the lossless large-photon-number limit, compared with Δε² ≈ 1/$n_T^{2}$ for both g(1,1) and the noise reduction factor, and with 4/$n_T^{3}$ for classical coherent-state G(1,1). Under single-photon loss modeled by a common transmittivity η on both modes, G(1,1) degrades but remains the most precise; g(1,1) is practically unchanged by loss but its precision no longer reflects a non-classical advantage; and the noise reduction factor becomes sensitive to the loss itself, which the paper uses to switch two-photon absorption detection on and off. The paper concludes that the optimal choice of observable is a trade-off between robustness to imperfection and the size of the quantum enhancement.","pith_inferences":["The identical-loss assumption (one transmittivity η for both modes) is the most fragile point; with unequal losses the normalization in g(1,1) will not cancel perfectly, so its loss immunity and the ranking of G(1,1) versus g(1,1) should be re-examined with two loss parameters.","The paper's error-propagation analysis is restricted to the method of moments; a full quantum Fisher information treatment of the same probe states could reveal whether the 6/n_T^{3.5} scaling is the ultimate limit or whether even better estimators exist.","The n_T^{-3.5} advantage of raw coincidence counting suggests that in future nonlinear-interferometer versions of two-photon absorption metrology, unnormalized intensity correlations, not normalized ones, are the readout most likely to preserve a quantum advantage.","A simple experimental check would place frequency-dependent attenuators in the two arms and map the region where g(1,1) loses its immunity; the paper's predictions are only guaranteed where both modes lose equally."],"forward_implications":["With optimized double-seeded two-mode squeezed light, G(1,1) transmission measurements yield the smallest estimation error for two-photon absorption absorbance, scaling as 6/n_T^{3.5}, and beat all classical coherent-state strategies.","g(1,1) forfeits the quantum scaling advantage, but its precision is essentially unchanged by linear photon loss, making it the safer readout in lossy or poorly characterized setups.","The noise reduction factor is the least precise observable, yet it is the only one that can distinguish single-photon losses from two-photon absorption, acting as a built-in loss monitor.","The optimal probe state depends strongly on the observable: G(1,1) prefers weak squeezing with strong seeding, g(1,1) prefers a small fixed amount of seeding, and the noise reduction factor requires balanced squeezing and seeding.","Squeezing improves precision over coherent states for every observable considered, but the gain is much larger for G(1,1) than for g(1,1) or the noise reduction factor."],"supporting_citations":[{"why":"Supplies the prior treatment of two-photon absorption precision under single-photon losses that this paper extends to two-mode squeezing and seeding.","marker":"[22]"},{"why":"Provides the optical parametric amplifier transformation and the beam-splitter loss model used throughout the calculation.","marker":"[44]"},{"why":"Gives the Markovian master-equation form for multiphoton absorption used to model the two-photon absorption sample.","marker":"[46]"},{"why":"Defines the noise reduction factor and the classical-versus-quantum correlation criterion behind that observable.","marker":"[40]"},{"why":"Gives the definitions of the intensity correlation G(1,1) and the normalized correlation g(1,1).","marker":"[43]"},{"why":"Supplies the error-propagation (method of moments) formula used to convert observable variances into estimation errors.","marker":"[50]"},{"why":"Provides the error-propagation and covariance-matrix basis underlying the estimation-error expressions.","marker":"[53]"}],"fun_headline_variants":["Intensity correlation wins for two-photon absorption","Squeezed light: best precision, but loss trades off","Quantum edge vs loss: Correlation choice matters","Two-photon absorption: Precision versus robustness","Best two-photon measurement: Intensity or robustness?"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central claim assumes the two light colours involved in non-degenerate two-photon absorption lose the same fraction of photons in the optics, so the recommended measurement could change if one colour is attenuated more than the other.","fun_headline_variants_meta":{"raw":{"variants":["Intensity correlation wins for two-photon absorption","Squeezed light: best precision, but loss trades off","Quantum edge vs loss: Correlation choice matters","Two-photon absorption: Precision versus robustness","Best two-photon measurement: Intensity or robustness?"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000112,"raw_usage":{"total_tokens":1045,"prompt_tokens":912,"completion_tokens":133,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":528,"completion_tokens_details":{"reasoning_tokens":61}},"tokens_in":528,"tokens_out":133,"duration_ms":2328,"temperature":1.0,"reasoning_tokens":61,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T05:35:29.000134+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the G(1,1) estimation error for the optimized double-seeded squeezed state at several total photon numbers in a lossless setup; if the log-log slope is not -3.5, the central scaling claim fails. Then insert unequal loss for the two modes to test whether g(1,1) remains loss-immune and whether G(1,1) stays the best readout.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the prior treatment of two-photon absorption precision under single-photon losses that this paper extends to two-mode squeezing and seeding."},{"cited_title":"Bondani, A","cited_arxiv_id":null,"evidence_quote":"Provides the optical parametric amplifier transformation and the beam-splitter loss model used throughout the calculation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the Markovian master-equation form for multiphoton absorption used to model the two-photon absorption sample."},{"cited_title":"Schaffrath, D","cited_arxiv_id":null,"evidence_quote":"Defines the noise reduction factor and the classical-versus-quantum correlation criterion behind that observable."},{"cited_title":"Enhancing entangled two-photon absorption of Nile Red via temperature-controlled SPDC","cited_arxiv_id":"2406.01075","evidence_quote":"Gives the definitions of the intensity correlation G(1,1) and the normalized correlation g(1,1)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the error-propagation (method of moments) formula used to convert observable variances into estimation errors."}],"review_version":1}