{"id":"348346be-ab44-4040-9123-dc8f54c445ea","arxiv_id":"2502.01480","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Two photons sent into a parametric down-conversion crystal at gain 2 show destructive quantum interference that suppresses the probability of outgoing one-photon-pair events.","lead":"A team in Nanjing and Brussels sent two laser photons into a nonlinear crystal and watched a new kind of quantum interference: the crystal can either pass the photons or replace them with fresh ones, and those two possibilities cancel out. The effect is the nonlinear cousin of the Hong-Ou-Mandel effect, and could be used to build non-Gaussian states of light for photonic quantum computers.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The g=2 suppression is inferred, not measured: at g>1.2 P1 comes from model Eq. S45, whose |1,1> term already encodes (2-g)^2; the low-gain dip is real but does not verify the zero at g=2.","rationale":"The reader's weakest assumption identifies exactly the load-bearing point. The paper has two evidentiary layers: (i) direct low-gain data showing P1 for |~1,1> falling below |~1,0> as g grows and across delay scans, which is credible and largely model-independent; and (ii) the g=2.03 suppression extracted from the fitted model, which is not direct. The model's functional form, Eq. S45, already contains the (2-g)^2 interference factor, so reporting P1 at g=2.03 from that model does not provide independent evidence for the zero; it shows only that the experiment is consistent with the model under the model's assumptions. The auxiliary fits at g≈1.21 (Fig. 4a) validate internal consistency but do not validate extrapolation to g≈2, where multi-photon terms and any multimode or spectral-temporal correlations matter much more. Thus the claim of experimentally demonstrating suppression when the gain is tuned to 2 should remain conditional on a direct or model-independent high-gain reconstruction. This is a confirmation of the reader's concern rather than a new objection, so the verdict should stay CONDITIONAL. I would not escalate to rejection because the low-gain result and the agreement of the measured C_m with the fitted model at moderate gain are genuine supporting evidence, and no internal inconsistency was found in the derivation of Eq. S20 or Eq. S45.","tokens_in":21821,"tokens_out":5534,"duration_ms":51865,"concrete_test":"For the g=2.03 run, release the raw dead-time-corrected six-channel coincidence probabilities and reconstruct P0..P5 by direct maximum-likelihood inversion of the detector response probabilities in Eq. S31, using the measured per-detector efficiencies and bootstrap uncertainties, without imposing the Eq. S45 ansatz. Compare the reconstructed P1 at T=1 with the vacuum-input T=0 value; if the T=1 value is not below the T=0 value by more than the combined uncertainty, the claimed g=2 suppression is unsupported. If truncation is needed, bound the neglected n>=6 tail using the measured C6 rather than assuming the model's photon-number distribution.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline claim—that P1 vanishes at g=2—is not directly observed. At g=2.03 the only raw data are six-channel coincidence probabilities, and the reported P1 values (Fig. 4a and Fig. S11) are computed from the fitted model of Supplementary S5, not inverted directly from those coincidences. Equation S45 is constructed from the ansatz that the imperfect input is an incoherent mixture of |0,0>, |0,1>, |1,0> and |1,1> with weights (1-O1)(1-O2), (1-O1)O2, O1(1-O2) and O1O2, and that each component evolves under the ideal single-mode two-mode-squeezing unitary U_PDC_g. For the |1,1> component this ansatz produces the factor (n+1-g)^2, so the zero at n=1, g=2 is present by construction. Fitting g, O1 and O2 at low gain and substituting them into this functional form cannot independently confirm the g=2 zero; the agreement between measured and model C_m in Fig. S10 is a consistency check of a model whose predicted effect is already encoded. The direct low-gain data (Fig. 3) do show a genuine relative suppression of P1 when the second photon is added at g≈1.2, but a dip at g≈1.2 is weaker evidence for a zero exactly at g=2. The load-bearing assumption is that all mode mismatch is classical and factorizes into independent O1 and O2; if the 15-nm-filtered photons retain spectral-temporal correlations or partial coherence with the PDC modes, the true high-gain state is not the mixture of Eq. S44 and the extracted P1 is a model artifact. The Methods section explicitly states that for g>1.2 the authors 'create a model' and 'substitute the fitted parameters into the model to calculate P1', confirming that the high-gain suppression is not directly measured.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports an experimental study of two-photon interference in a parametric down-conversion (PDC) crystal, the nonlinear analogue of the Hong-Ou-Mandel effect. For an ideal |1,1> input, the probability P1 of outputting exactly one photon pair is predicted to vanish at parametric gain g=2, and the authors present data and analysis that they interpret as demonstrating this suppression. At low gain (g up to about 1.2), P1 is obtained from six-channel coincidence measurements by a truncated inversion, and delay scans show a dip when both injected photons are temporally matched. At high gain (g=2.03), the reported P1 values are not directly inverted from the measured coincidences but are computed from a fitted model (Supplementary Eq. S45) whose parameters are determined in auxiliary experiments. The paper also discusses reconstructed Wigner functions and potential applications to non-Gaussian state engineering.","tokens_in":22118,"tokens_out":4797,"duration_ms":46414,"significance":"The theoretical effect is clean, elegant, and analytically derived in the supplementary material; if experimentally established, it would constitute a genuinely new nonlinear quantum interference phenomenon with implications for the generation of non-Gaussian states. The low-gain data provide direct evidence of a relative suppression of P1 when a second photon is added, and the delay scan in Fig. 3(c) shows a dip, which is a meaningful first step. However, the paper's headline claim that P1 vanishes at g=2 is not directly measured: at g=2.03 it is inferred from a fitted model whose functional form already contains the (2-g)^2 factor. The central result therefore rests on the validity of assumptions that are not fully tested. The paper would be significantly strengthened by reporting model-independent high-gain data or by explicitly framing the high-gain result as consistent with the prediction rather than as a direct observation.","major_comments":[{"comment":"The central claim that P1 is suppressed at g≈2 is not directly measured. As the Methods section states, for g>1.2 the authors 'create a model' and deduce P1 from it; the six-channel coincidences are not inverted directly in that regime. The model of Supplementary Eq. S45 contains, for the |1,1> component, the factor (n+1-g)^2, so P1=(2-g)^2/g^3 vanishes at g=2 by construction. The agreement between the model and the measured C_m at g=1.21 (Fig. S10) is a useful consistency check, but it does not independently confirm the zero at g=2 because the prediction is already encoded in the fitted functional form. I recommend either reporting the raw six-channel C_m at g≈2 together with a model-independent bound on P1, or rephrasing the abstract and conclusion to say that the high-gain data are 'consistent with' the predicted suppression rather than that the suppression was 'observed' at g=2.","section":"Methods (Determination of P1); Supplementary S5.3, Eq. S45; Fig. 4(a)"},{"comment":"The high-gain extraction depends on the assumption that all input imperfections are described by an incoherent mixture with factorized, independent overlaps O1 and O2, and that each component evolves under the ideal single-mode unitary U_PDC_g. This assumption is load-bearing: if the 15-nm filtered photons retain spectral-temporal correlations, or if there is partial coherence between the |1,0> and |0,1> components of the input, the true high-gain state is not the mixture of Eq. S44 and the deduced P1 becomes a model artifact. The paper should provide a direct test of the factorization and single-mode assumptions, for example by comparing model predictions with two-photon-input measurements at intermediate gains (1.2<g<2) without refitting, or by measuring the spectral-temporal correlations and the effective number of modes of the interacting fields.","section":"Supplementary S5.2-S5.3, Eq. S44"},{"comment":"The parametric gain g is obtained by fitting the SPDC m-fold coincidences to a single-mode two-mode squeezed vacuum (Eq. S38). With a tightly focused pulsed pump and a 2.5-mm crystal, the PDC is likely multimode, and the fitted g is an effective parameter. The quantitative prediction P1=0 at g=2 assumes a single-mode model in which the same squeezing parameter r enters the (n+1-g)^2 interference term. If the source has multiple spatial or spectral modes with different squeezing parameters, the global output P1 need not vanish at the effective gain g=2. The authors should justify the single-mode effective description more carefully or quantify the number of modes participating in the interference.","section":"Fig. 2(d) and Supplementary S5.1"}],"minor_comments":[{"comment":"In the sentence describing the heralded V-polarized state, 'detector Tig-1' should be 'detector Trig-1' (and similarly 'Tig-2' should be 'Trig-2').","section":"Supplementary S5.4"},{"comment":"The notation 'P 5−detectors 1' and 'P 5−detect 1' is awkward and inconsistent; a compact notation such as P1^(5) and P1^(6) would improve readability.","section":"Main text and Supplementary, notation for P1"},{"comment":"The reconstructed Wigner functions are presented without a description of the reconstruction procedure or the exact quadrature definitions; please specify how the Wigner function was obtained from the fitted model and what the axes represent.","section":"Fig. 4(c) and Fig. S12"},{"comment":"The text writes 'g2(0)' in several places; this should be g^(2)(0) to avoid confusion with the parametric gain g.","section":"Supplementary S2"},{"comment":"The data availability statement says data are available 'upon request'; given the quantitative nature of the central claim, depositing the raw six-channel coincidence data and the fitted parameters would substantially increase confidence in the results.","section":"Data availability"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's central claim is experimentally important but currently rests on a fitted model that bakes in the predicted zero at g=2. The fact that two of the co-authors are the original theorists of the effect is not itself a problem, but it strengthens the need for the high-gain result to be either directly measured or at least validated by an independent, model-free observable. I would encourage the editor to require the authors to present raw high-gain coincidence data and a clear statement of what is directly measured versus what is deduced from the model."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a real experimental first and the low-gain data are good; the headline claim that P1 vanishes at g=2 is inferred from a model that already contains the predicted zero, not measured.\n\nThe new thing is the experiment. Two heralded single photons are injected into a high-gain PPKTP crystal, mode-matched at 0.65, and the output is analyzed with six detectors. The low-gain results in Fig. 3 are direct and convincing: adding a V photon lowers P1, and the P1 dip appears when the V photon is temporally matched. That is the first evidence for the Cerf-Jabbour version of two-photon interference, and it is clean enough to stand on its own. The authors also deserve credit for being transparent about their inversion: Methods states that for g>1.2 the direct five-detector estimate is invalid and a model is used.\n\nThe soft spot is the g=2 claim. At g=2.03, P1 is not inverted from the measured six-channel coincidences; it comes from Eq. S45 of the supplement. That equation is built from an ansatz in which the input is an incoherent mixture of |0,0>, |0,1>, |1,0>, and |1,1> with independent overlaps O1 and O2, each component evolving under the ideal PDC unitary. For the |1,1> component the P1 formula is (2-g)^2/g^3 by construction. Fitting g, O1, and O2 at low gain and plugging them into that form cannot independently verify the zero at g=2; the agreement with measured C_m in Fig. S10 is a consistency check of a model that already contains the effect. A dip at g=1.2 is real support for the interference, but it does not establish a zero exactly at g=2.\n\nThe load-bearing assumption is that all mode mismatch is classical and factorizes into O1 and O2. If the 15-nm-filtered photons keep spectral-temporal correlations or partial coherence with the PDC modes, the high-gain state is not the mixture of Eq. S44, and the extracted P1 could be a model artifact. There are also no raw high-gain data and essentially no error bars on the deduced P1; data are available only on request.\n\nThe self-citation is disclosed and not itself a problem. But because the theory comes from two of the co-authors, the fact that the high-gain evidence is model-generated makes the circularity worth a hard look in review.\n\nWho this is for: people working on nonlinear quantum interference, photon-number engineering, and non-Gaussian state generation. The low-gain observation deserves publication. The g=2 suppression should be presented as a model-supported prediction, not a direct measurement, unless raw data and a model-free inversion are released. I would accept this for serious peer review and send it back asking for that.","headline":"First real low-gain observation of Cerf-Jabbour interference, but the g=2 suppression is model-deduced rather than measured and should be framed that way.","tokens_in":22873,"tokens_out":2617,"would_cite":true,"duration_ms":23404,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81V80"],"pacs":["42.50.Ar","42.65.Lm"],"model":"deepseek-v4-flash","headline":"Two photons entering a nonlinear crystal can cancel each other when the crystal's gain is 2.","keywords":["Hong-Ou-Mandel effect","parametric down-conversion","Cerf-Jabbour interference","timelike indistinguishability","photon-number statistics","nonlinear quantum interference","non-Gaussian states","two-mode squeezing"],"falsifier":"Reconstruct $P_1$ at $g\\approx 2$ directly from six-channel coincidence counts without the fitted model, using a heralded $|1,1\\rangle$ input with verified near-unit mode overlap; the claim stands if the reconstructed $P_1$ is significantly below its low-gain value and falls as the input pair transmission $T$ approaches 1, and fails if no such dip appears.","tokens_in":21451,"feed_emoji":"⚛️","tokens_out":7767,"duration_ms":67085,"temperature":0.7,"pith_summary":"The Hong-Ou-Mandel effect is usually a linear process: two indistinguishable photons meet at a 50:50 beam splitter and never leave with one photon in each output. This paper claims that the same two-photon destructive interference occurs inside a parametric down-conversion crystal, where the beam splitter is replaced by a nonlinear medium and the role of the two spatial paths is played by the two directions of time. The crystal can either pass the two incident photons through or annihilate them and create two new \"reborn\" photons, and these two amplitudes interfere. When the parametric gain is tuned to $g=2$, the probability of outputting exactly one photon pair is predicted to vanish, $P_1=(2-g)^2/g^3=0$, and the paper reports experimental data consistent with this suppression. A sympathetic reader would care because this \"timelike indistinguishability\" is a new quantum mechanism and a practical handle on photon-number statistics.","feed_headline":"Two photons in a crystal can cancel at gain 2","feed_subtitle":"A PDC tuned to gain 2 suppresses one-pair output, showing timelike indistinguishability.","key_machinery":"The central object is the two-mode squeezing unitary of parametric down-conversion, $U_g^{\\mathrm{PDC}}=\\exp[r(\\hat a_H^\\dagger\\hat a_V^\\dagger-\\hat a_H\\hat a_V)]$ with parametric gain $g=\\cosh 2r$. Acting on $|1,1\\rangle$, this unitary creates the transmitted-versus-reborn superposition; the identity $P_1=(2-g)^2/g^3$ is what carries the argument, and it is the nonlinear analogue of the beam-splitter identity $P_{1,1}^{\\mathrm{HOM}}=(2T-1)^2$. The formal bridge is a partial-time-reversal duality that maps the beam-splitter transmittance $T$ onto the PDC gain $g$, turning spatial path distinguishability into temporal indistinguishability.","core_discovery":"At the heart of the paper is the Cerf-Jabbour interference: a two-mode squeezing operation acting on an ideal $|1,1\\rangle$ input produces a superposition of the transmitted photon pair and a pair that has been annihilated and recreated. In the one-pair sector these two amplitudes interfere destructively with probability $P_1=(2-g)^2/g^3$, which vanishes exactly at $g=2$. The experiment injects heralded H- and V-polarized photons into a high-gain PPKTP crystal, tunes the gain to $g\\approx 2.03$, and deduces from six-channel coincidences and a fitted model that $P_1$ is suppressed as the transmission of the input pair increases, reaching a residual value near 0.1 that is attributed to imperfect mode matching. The authors further show that the output state retains Wigner negativity and that at $g=3$ the two-pair component would be suppressed, extending the mechanism to higher-order destructive interference.","pith_inferences":["An extension not pursued in the paper: the $P_1$ dip could be used as a delay-scanning probe of temporal indistinguishability between two independent single-photon sources, in direct analogy to how a Hong-Ou-Mandel dip is scanned.","A testable consequence beyond the reported data: if the input-state model were replaced by a fully quantum description with spectral correlations, the exact $P_1=0$ might broaden or shift, so measuring $P_1$ versus filter bandwidth would separate the classical-mixture assumption from the unitary core.","The integer-gain structure suggests a practical recipe the paper does not develop: cascade PDC stages at $g=2,3,\\ldots$ to engineer targeted holes in the photon-number distribution of a multimode state."],"forward_implications":["At integer gain $g=3$, the same interference suppresses the two-pair component $P_2$, so the destructive mechanism extends to four-photon interference.","The output of the $|1,1\\rangle$ evolution is non-Gaussian and Wigner-negative, so cascading CJ stages with linear optical circuits is a route to multimode non-Gaussian states needed for photonic quantum computing.","The $P_1$ dip at $g=2$ is a time-domain witness of indistinguishability: only photons that match the PDC modes in spectrum, space, and time experience the cancellation, so the dip depth encodes temporal mode matching.","Because at low gain an extra injected photon increases $P_1$ while at high gain it decreases it, the effect explains why this interference was missed in previous moderate-power PDC experiments and marks a distinctly multi-photon nonlinear regime."],"supporting_citations":[{"why":"Supplies the theoretical prediction of two-boson quantum interference in time, including the $P_1=(2-g)^2/g^3$ cancellation at $g=2$ and the reborn-pair interpretation.","marker":"[30]"},{"why":"Defines the linear Hong-Ou-Mandel effect that the paper transposes to the nonlinear regime and serves as the baseline interference to compare against.","marker":"[15]"},{"why":"Introduces SU(1,1) nonlinear interferometers, the prior form of PDC-based interference from which this experiment distinguishes itself.","marker":"[2]"},{"why":"Provides the squeezing-theory relation between parametric gain and pump power used to calibrate $g$ from measured multi-channel coincidences.","marker":"[32]"},{"why":"Together with [39], supplies the method used to estimate the spectral purity of the filtered PDC photons that underlies the mode-matching parameter.","marker":"[29]"},{"why":"Quantifies spectral purity from unheralded second-order correlation measurements, used to characterise the indistinguishability of the injected photons.","marker":"[39]"}],"fun_headline_variants":["Photon pairs vanish at gain 2 in nonlinear crystal","Timelike interference suppresses one-pair output","Nonlinear quantum interference: cancellation at gain 2","Cerf-Jabbour effect: photon pair destructive interference"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The high-gain suppression at $g=2.03$ is not read directly from the measured coincidences; it comes from a model that assumes the real input is an incoherent mixture with independent mode overlaps $O_1$ and $O_2$ and that each mixture component evolves under the ideal two-mode PDC unitary. If the imperfect input is not a classical mixture, or if spectral-temporal correlations survive the 15-nm filter, the deduced $P_1$ dip could be an artifact of that model.","fun_headline_variants_meta":{"raw":{"variants":["Photon pairs vanish at gain 2 in nonlinear crystal","Timelike interference suppresses one-pair output","Nonlinear quantum interference: cancellation at gain 2","Cerf-Jabbour effect: photon pair destructive interference"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000367,"raw_usage":{"total_tokens":1974,"prompt_tokens":951,"completion_tokens":1023,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":567,"completion_tokens_details":{"reasoning_tokens":960}},"tokens_in":567,"tokens_out":1023,"duration_ms":8553,"temperature":1.0,"reasoning_tokens":960,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T15:07:53.640268+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Reconstruct $P_1$ at $g\\approx 2$ directly from six-channel coincidence counts without the fitted model, using a heralded $|1,1\\rangle$ input with verified near-unit mode overlap; the claim stands if the reconstructed $P_1$ is significantly below its low-gain value and falls as the input pair transmission $T$ approaches 1, and fails if no such dip appears.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the theoretical prediction of two-boson quantum interference in time, including the $P_1=(2-g)^2/g^3$ cancellation at $g=2$ and the reborn-pair interpretation."},{"cited_title":"Christ, K","cited_arxiv_id":null,"evidence_quote":"Quantifies spectral purity from unheralded second-order correlation measurements, used to characterise the indistinguishability of the injected photons."}],"review_version":1}