{"id":"e13f6b29-5fd4-4dc9-a0d4-af621de1c36d","arxiv_id":"2608.04476","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Under fixed conditional-probe energy and fixed gain, optimized success-weighted photon subtraction, addition, and catalysis all give less Fisher information than the optimized Gaussian input in an SU(1,1) interferometer.","lead":"This paper derives exact formulas for how much phase information photon subtraction, addition, and catalysis add to an SU(1,1) interferometer, and then shows that those apparent gains vanish once success probability and energy are counted. The result is a practical warning for experiments planning non-Gaussian inputs with parity detection.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Fixed-gain restriction is the load-bearing caveat: the paper's 'conservative' defense compares the Gaussian at g≈1.10 with non-Gaussian values at g=0.75, so gain optimization could still let PC (P_j F_Q = 104.290) overtake the gain-optimized Gaussian optimum (~123.8).","rationale":"The paper's core derivation is internally consistent: the finite-transmissivity Kraus map, moment generators, QFI formula Eq. (47), and parity CFI Eq. (70) are mutually compatible, and the loss-dressed observable Eq. (81) reduces correctly in the lossless limit. The strongest claim is carefully stated as a fixed-gain result, so there is no internal contradiction in reporting the non-Gaussian values below the Gaussian benchmark at g=0.75. However, the conclusion is load-bearing on the fixed-gain restriction for two reasons. First, the quoted Gaussian optimum F_Q^{(G,*)} = 107.569 is itself a maximum over r and α at fixed g; the paper notes that varying g raises it to ≈123.8, so the comparison denominator is gain-sensitive. Second, and more importantly, the Sec. V discussion asserts that this restriction is 'conservative' because g=0.75 captures 87% of the Gaussian optimum. That defense compares the Gaussian at g≈1.10 with non-Gaussian quantities evaluated at g=0.75; it provides no upper bound on P_j F_Q at g≈1.10. The high-variance PC branch in Table VI shows that conditional F_Q can reach 189.890 at fixed g, so the relevant question is whether P_j F_Q for PC (104.290 at g=0.75) grows or shrinks when g is reoptimized. This is exactly the test the paper leaves open. The PS/PA values 30.940 and 11.544 are far enough below 123.8 that they are less likely to be affected, but they should still be included for completeness. The finite T ≥ 0.65 and r ≤ 1.25 domains are a secondary caveat: the paper extends PC to r ≤ 2.4 and finds the high-variance branch, but it does not extend PS/PA, whose success probabilities are largest at low T; a quick scan suggests their per-attempt values remain well below the Gaussian optimum, but the paper does not show this. The recommended verdict is unchanged from the reader's CONDITIONAL: the analysis is self-consistent and valuable, but the headline no-advantage statement requires either a full gain-optimized check or an explicit acknowledgment that it applies only to g=0.75. The concrete numerical test above would settle whether the central claim survives.","tokens_in":29575,"tokens_out":13913,"duration_ms":125517,"concrete_test":"Run the Table V optimization with g as an additional variable: for m=1 and \\bar N_enc = 9, maximize P_j F_Q^{(j,1)} over 0.3 ≤ g ≤ 1.5, 0 ≤ r ≤ 2.4, and 0.65 ≤ T ≤ 0.995 for PS, PA, and PC, with |α| fixed by Eq. (38) at each (r,T,g). Compare each maximum with F_G^* = max_{g,r,|α|} F_Q^{(G)} at \\bar N_enc = 9, which the paper quotes as ≈123.8 at g≈1.10. If any non-Gaussian P_j F_Q exceeds F_G^*, the fixed-gain no-advantage claim does not extend to gain-optimized operation, and the Sec. V statement that the restriction is conservative is false. A useful subsidiary output is the location of any new PC optimum in (g,r,T), since this tests whether the high-variance branch of Fig. 8 survives success weighting at higher gain.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central no-advantage statement (Table V) is established only for g=0.75, r≤1.25, 0.65≤T≤0.995, and m=1. That scope is stated honestly in the abstract and conclusion, but the paper then weakens its own caveat in Sec. V: 'full gain optimization would raise the Gaussian QFI to ≈123.8 at g≈1.10; the fixed g=0.75 captures ≈87% of this value, so the no-advantage conclusion is conservative with respect to this restriction.' This inference is invalid. A bound on the Gaussian benchmark at g≈1.10 does not constrain the non-Gaussian metrics at g≈1.10. All terms in Eq. (47) carry C^4, S^4, or C^2 S^2, and P_j also depends on g through the resource constraint Eq. (38), so PC's P_j F_Q = 104.290 at g=0.75 could increase when g is reoptimized, just as the Gaussian value does. The high-variance PC branch in Table VI already has conditional F_Q = 189.890 at fixed g; only its success weighting keeps it below 107.569. Whether that branch crosses the gain-optimized Gaussian value is untested. Thus the headline result is conditional on g, and the paper's stated reason for treating that restriction as harmless is not supported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript analyzes an SU(1,1) interferometer in which mode a is prepared in a coherent state and mode b in a squeezed vacuum that is first sent through a heralded linear-optical module implementing photon subtraction, photon addition, or photon catalysis. It derives closed finite-transmissivity moment formulas, a three-moment pure-state QFI for parity-eigenstate inputs (Eq. 47), a dark-point parity CFI (Eq. 70), and a loss-dressed effective parity observable (Eq. 81). The authors compare conditional and success-weighted Fisher information under fixed-seed, fixed-total-energy, and fixed-arm-exposure protocols. The main quantitative result is Table V: for m=1, fixed gain g=0.75, Nbar_enc=9, and optimization over r<=1.25 and 0.65<=T<=0.995, the success-weighted QFI of photon subtraction, addition, and catalysis remains below the optimized Gaussian value 107.569, with P_j F_Q equal to 11.544, 30.940, and 104.290, respectively. An expanded catalysis scan shows a conditional branch with F_Q=189.890 but low dark-point parity extraction, which the paper interprets as a measurement mismatch rather than a state-preparation advantage.","tokens_in":29980,"tokens_out":9589,"duration_ms":96393,"significance":"The analytic results are a solid contribution: Eq. (47) unifies the three operations in a single moment formula, Eq. (70) gives a closed parity CFI with a transparent gap to the QFI, and Eq. (81) is an exact single-mode pulled-back observable for internal loss. The manuscript is parameter-free in the sense that all formulas follow from stated state models, and the numerics are cross-checked by Fock-cutoff convergence. The resource-accounting message—that conditional non-Gaussian QFI enhancement can disappear after success weighting and resource optimization—is valuable, concrete, and falsifiable. The significance is moderate: it sharpens the boundary for a specific near-term relevant setup, and the authors are mostly candid about the tested domain.","major_comments":[{"comment":"The statement that the fixed-gain no-advantage conclusion is \"conservative\" with respect to gain optimization is not supported by the paper's own formulas. The claim refers to the Gaussian QFI rising from 107.569 at g=0.75 to about 123.8 at g≈1.10, but Eq. (47) depends on g through C=cosh g and S=sinh g, and the resource constraint Eq. (38) makes |alpha| depend on g as well; the success-weighted metric P_j F_Q therefore also varies with g. The value 123.8 is only an optimized Gaussian benchmark at g≈1.10; it does not bound the non-Gaussian metrics at that gain. In particular, the PC conditional branch in Table VI has F_Q=189.890 at g=0.75, so whether P_PC F_Q at g≈1.10 lies above or below 123.8 is untested. Please provide a gain scan or gain optimization for the non-Gaussian operations over the same resource constraint, or remove the \"conservative\" claim and state explicitly that the no-advantage result is established only for g=0.75.","section":"V. Discussion"},{"comment":"The central no-advantage result is a finite-domain numerical maximization at fixed g=0.75, not an analytic bound. The abstract and conclusion do include the \"tested constraints\" qualifier, but the paper should also state this qualifier wherever Table V is summarized in the discussion. This matters because the PC row of Table V sits at the T=0.995 boundary where the catalysis map is close to the identity, so the table does not probe the low-transmissivity PC regime where Sec. III.C reports conditional QFI improvements. The paper should explicitly note that no claim is made for the success-weighted PC metric in the low-T windows, or it should extend the optimized success-weighted PC scan to those windows.","section":"III.D.4"}],"minor_comments":[{"comment":"The number 123.8 for the gain-optimized Gaussian QFI appears without derivation or a supporting table; if retained, the underlying optimization (g, r, |alpha|) should be reported.","section":"V. Discussion"},{"comment":"The phrase \"near its unconstrained optimum\" for g=0.75 is misleading in light of the paper's own statement that gain optimization raises the Gaussian QFI by about 15%; please rephrase to something like \"near the middle of the gain range used here.\"","section":"III.D.4"},{"comment":"In Eq. (B3), \"Rea>0\" should read \"Re a>0\".","section":"Appendix B"},{"comment":"The Fock cutoff d is used in Table VIII and in the convergence checks, but d is not defined in the main text; please define it as the photon-number cutoff of the single-mode Hilbert space.","section":"IV.D"},{"comment":"Panel (d) of Fig. 8 is labeled \"dark-point extraction\" but the vertical axis is not explicitly labeled; please label it as R_PC = F_C^Pi(0)/F_Q.","section":"Fig. 8"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the journal's scope and the central derivations appear sound. The main issue is the unsupported \"conservative with respect to gain optimization\" claim in Sec. V, which overstates the generality of the no-advantage result. The authors should either supply the missing gain scan or narrow the claim; this is fixable within the scope of the paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things you should know before reviewing. The analytic machinery is the real contribution and it is in good shape: a unified Kraus map for m-photon subtraction, addition, and catalysis; arbitrary-order finite-differential moment generators; a compact three-moment QFI expression (Eq. 47); the dark-point parity CFI (Eq. 70); and an exact loss-dressed effective parity observable (Eq. 81) from the IWOP reduction. The derivations are self-contained, there are no fitted parameters, and the ideal/lossless limits check out. Second, the headline negative result—that success-weighted Fisher information of PS/PA/PC stays below the optimized Gaussian at Nbar=9—is only as strong as a fixed-gain, finite-domain optimization at g=0.75, m=1. The paper states that scope honestly in the abstract and conclusion, but then weakens it in Sec. V with a 'conservative' argument that is not valid.\n\nWhat is actually new: earlier SU(1,1) photon-operation papers used different operation placements, measurements, or resource conventions, so their advantages were not comparable. This paper puts PS/PA/PC in one map, proves the single-photon PS/PA normalized-state equivalence while keeping their different success probabilities, and isolates a measurement mismatch: PC creates a conditional branch with F_Q=189.890 at g=0.75 but dark-point parity extracts only about 0.10 of it. The protocol distinctions—fixed seed, fixed total energy, fixed sensing-arm exposure—are careful, and conditional vs per-attempt information is kept separate. People working in CV metrology will get real value from this.\n\nSoft spots, in proportion. I found no error in the moment or kernel math, and the citation pattern is fine; same-group citations are used for context, not as a load-bearing prior. The real problem is the Sec. V claim that fixing g=0.75 makes the no-advantage conclusion conservative because optimizing g would raise the Gaussian QFI from 107.569 to about 123.8. That inference is unsupported. A gain-optimized Gaussian bound does not constrain the non-Gaussian metrics under gain re-optimization: every term in Eq. (47) carries C^4, S^4, or C^2S^2, P_j depends on g through the resource constraint, and PC's success-weighted 104.290 at g=0.75 has no tested value at g~1.10. The high-variance PC branch suggests real gain sensitivity. So the conclusion remains conditional; the 'conservative' language should be removed or backed by an actual g-scan. This is a revision issue, not a fatal flaw. Minor: the preprint posts no code or data, and the convergence statements refer to data supplied with the paper without being available.\n\nRecommendation: send it to a serious referee. The formulas deserve checking, and a referee should ask for the g-scan of PC success-weighted QFI before acceptance. It is solid enough that a desk reject would waste the community's time.","headline":"Useful and mostly sound analytic framework; the per-attempt no-advantage claim is conditional on a fixed gain, and the paper's claim that the restriction is conservative does not hold.","tokens_in":30437,"tokens_out":7082,"would_cite":true,"duration_ms":65030,"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":"At fixed gain and energy, heralded photon subtraction, addition, and catalysis yield less per-attempt Fisher information than the optimized Gaussian input in an SU(1,1) interferometer.","keywords":["quantum Fisher information","SU(1,1) interferometry","photon subtraction","photon addition","photon catalysis","parity detection","heralded non-Gaussian states","continuous-variable quantum metrology"],"falsifier":"Re-run the fixed-energy optimization with the gain left free, for instance over $g\\in[0.5,1.2]$; if any non-Gaussian operation's success-weighted Fisher information exceeds the Gaussian optimum at the same $\\bar N_{\\rm enc}=9$ (about $123.8$ when $g$ is optimized), the central no-advantage conclusion fails.","tokens_in":29381,"feed_emoji":"🎯","tokens_out":10342,"duration_ms":81980,"temperature":0.7,"pith_summary":"The paper asks whether putting photon-subtracted, photon-added, or photon-catalyzed squeezed states into an SU(1,1) interferometer improves phase estimation over the plain coherent-plus-squeezed Gaussian input, once the price of heralding is counted. It derives closed formulas for the conditional quantum Fisher information and the dark-point parity Fisher information, plus a loss-dressed effective parity observable. Its central result is that, at fixed interferometer gain $g=0.75$, fixed conditional-probe energy $\\bar N_{\\rm enc}=9$, and single-photon order $m=1$, the success-weighted Fisher information of all three operations stays below the optimized Gaussian benchmark ($F_Q^{(G,\\star)}=107.569$), with photon catalysis closest at $P_jF_Q=104.290$. The paper distinguishes conditional enhancement, which subtraction and addition do show in the high-transmissivity regime, from per-attempt precision, which they do not improve under this resource contract. A separate high-QFI catalysis branch exists but is poorly read out by dark-point parity, a measurement mismatch rather than a state-preparation failure.","feed_headline":"Non-Gaussian inputs trail optimized Gaussian in SU(1,1) metrology","feed_subtitle":"Photon subtraction, addition, and catalysis all fall short once heralding probability counts.","key_machinery":"The central object is the unified finite-transmissivity Kraus map $\\hat K_{\\mu,\\nu}(T)={}_c\\langle\\nu|\\hat B_{bc}(T)|\\mu\\rangle_c$, which mixes the squeezed vacuum with a Fock ancilla and projects onto a Fock outcome; the choices $(0,m)$, $(m,0)$, and $(m,m)$ give $m$-photon subtraction, addition, and catalysis. This map supplies closed success probabilities and the three moments $N_j$, $V_j$, $M_j$ that enter the pure-parity-eigenstate QFI formula $F_Q=4[C^4|\\alpha|^2+S^4V_j+C^2S^2(2|\\alpha|^2N_j+|\\alpha|^2+N_j+1+2\\Re(\\alpha^2M_j))]$, and the same moments feed the dark-point parity Fisher information. Internal loss is pulled back through the interferometer into a single effective parity observable $\\Omega_\\phi$, so loss changes the operator the prepared state is tested against rather than merely reducing a final contrast factor.","core_discovery":"On its own terms, the paper establishes a resource-accounted result: for single-photon ($m=1$) operations with gain $g=0.75$ and total conditional-probe energy $\\bar N_{\\rm enc}=9$, independently optimizing the coherent-squeezed allocation over $0\\le r\\le1.25$ and $0.65\\le T\\le0.995$ gives success-weighted Fisher information $P_jF_Q^{(j)}=11.544$ for photon subtraction, $30.940$ for photon addition, and $104.290$ for photon catalysis, all below the optimized Gaussian value $F_Q^{(G,\\star)}=107.569$; the corresponding dark-point parity values are lower still. The same closed formulas show that single-photon subtraction and addition do improve the conditional QFI over the Gaussian reference across most high-transmissivity settings, and that multi-photon catalysis creates low-transmissivity conditional windows that survive moderate internal loss. The paper therefore draws a boundary: these non-Gaussian operations act as conditional filters, not as per-attempt enhancements, under the tested resource contract.","pith_inferences":["The fixed-gain caveat invites a direct extension: optimizing $g$ as well, which the paper notes would raise the Gaussian QFI from $107.569$ to about $123.8$ at $g\\approx1.10$, could alter the ranking, and the same unified map could be scanned over $g$ to check whether any non-Gaussian branch overtakes.","Because the catalysis branch is variance-dominated, a readout sensitive to photon-number variance, such as number-resolving detection or a suited homodyne scheme, might convert that branch into a practical per-attempt advantage; the paper identifies the mismatch but does not optimize the alternative measurement.","The same resource-accounting methodology could be applied to output-port or internal photon operations; the paper notes that operation position does not commute with two-mode squeezing, so those settings require a fresh calculation rather than a simple extrapolation.","Adding failed-preparation energy to the resource contract could change the conclusion; the paper counts only successfully heralded probe energy, so a fixed-clock experiment without storage faces a different optimization."],"forward_implications":["At fixed gain $g=0.75$ and conditional-probe energy $\\bar N_{\\rm enc}=9$, single-photon subtraction, addition, and catalysis all fail to beat the optimized Gaussian per-attempt Fisher information, so parity-readout experiments in this regime cannot rely on these heralded inputs for a rate advantage.","Single-photon subtraction and addition remain conditional enhancers in the high-transmissivity regime: when failed heralds are discarded and only stored probes are used, their conditional QFI beats the unoptimized Gaussian reference.","Multi-photon catalysis is a low-transmissivity conditional filter, but its high-QFI branch is dominated by photon-number variance, which dark-point parity does not read out, so the available information sits in a state-measurement mismatch.","Internal loss does not simply reduce contrast: it changes the effective parity observable itself, and unequal arm losses can reverse the operation ranking, so loss must be specified per arm."],"supporting_citations":[{"why":"Supplies the beam-splitter-and-heralding state-engineering formalism from which the unified finite-transmissivity Kraus map is taken.","marker":"[26]"},{"why":"Establishes conditional photon subtraction through Fock-projected beam splitting, the experimental basis of the heralding modules.","marker":"[28]"},{"why":"Demonstrates photon addition and subtraction as conditional operations, supporting treating addition on the same linear-optical footing.","marker":"[29]"},{"why":"Provides the general SU(1,1) QFI construction for photon-operated inputs whose ideal limits the paper recovers, and the earlier positive-advantage result contrasted with the resource-accounted benchmark.","marker":"[33]"},{"why":"Introduces the probability-weighted sensitivity-difference diagnostic in passive interferometry that the paper adapts for its per-attempt comparisons.","marker":"[30]"},{"why":"Supplies the beam-splitter internal-loss model for SU(1,1) interferometers used in deriving the lossy effective parity observable.","marker":"[10]"},{"why":"Gives the Wigner-function parity treatment for lossy non-Gaussian phase measurements that the effective-observable derivation complements.","marker":"[42]"},{"why":"Provides the IWOP integration technique used to derive the closed pulled-back parity kernels in the ideal and lossy cases.","marker":"[43]"}],"fun_headline_variants":["Non-Gaussian inputs trail optimized Gaussian in SU(1,1) metrology","Success-weighted QFI: non-Gaussian ops lose to Gaussian","Photon catalysis, subtraction no per-attempt advantage","Heralded non-Gaussian probes underperform optimized Gaussian","Counting heralding: Gaussian wins in SU(1,1) interferometry"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central no-advantage conclusion rests on fixing the interferometer gain at $g=0.75$ and restricting the search to $0\\le r\\le1.25$ and $0.65\\le T\\le0.995$ for $m=1$; the paper itself notes that freeing $g$ would raise the Gaussian benchmark from $107.569$ to about $123.8$, so the comparison is gain-sensitive.","fun_headline_variants_meta":{"raw":{"variants":["Non-Gaussian inputs trail optimized Gaussian in SU(1,1) metrology","Success-weighted QFI: non-Gaussian ops lose to Gaussian","Photon catalysis, subtraction no per-attempt advantage","Heralded non-Gaussian probes underperform optimized Gaussian","Counting heralding: Gaussian wins in SU(1,1) interferometry"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000259,"raw_usage":{"total_tokens":1635,"prompt_tokens":1043,"completion_tokens":592,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":659,"completion_tokens_details":{"reasoning_tokens":502}},"tokens_in":659,"tokens_out":592,"duration_ms":5363,"temperature":1.0,"reasoning_tokens":502,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T14:39:39.860096+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the fixed-energy optimization with the gain left free, for instance over $g\\in[0.5,1.2]$; if any non-Gaussian operation's success-weighted Fisher information exceeds the Gaussian optimum at the same $\\bar N_{\\rm enc}=9$ (about $123.8$ when $g$ is optimized), the central no-advantage conclusion fails.","supporting_citations":[{"cited_title":"Hudelist, J","cited_arxiv_id":null,"evidence_quote":"Supplies the beam-splitter-and-heralding state-engineering formalism from which the unified finite-transmissivity Kraus map is taken."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes conditional photon subtraction through Fock-projected beam splitting, the experimental basis of the heralding modules."},{"cited_title":"Gupta, B","cited_arxiv_id":null,"evidence_quote":"Demonstrates photon addition and subtraction as conditional operations, supporting treating addition on the same linear-optical footing."},{"cited_title":"Dakna, T","cited_arxiv_id":null,"evidence_quote":"Provides the general SU(1,1) QFI construction for photon-operated inputs whose ideal limits the paper recovers, and the earlier positive-advantage result contrasted with the resource-accounted benchmark."},{"cited_title":"Frascella, E","cited_arxiv_id":null,"evidence_quote":"Introduces the probability-weighted sensitivity-difference diagnostic in passive interferometry that the paper adapts for its per-attempt comparisons."},{"cited_title":"Xin, Optics Express29, 43970 (2021)","cited_arxiv_id":null,"evidence_quote":"Gives the Wigner-function parity treatment for lossy non-Gaussian phase measurements that the effective-observable derivation complements."},{"cited_title":"Hou, J.-D","cited_arxiv_id":null,"evidence_quote":"Provides the IWOP integration technique used to derive the closed pulled-back parity kernels in the ideal and lossy cases."}],"review_version":2}