{"id":"cde75d87-37b8-4a5b-acaa-e88e7a22b066","arxiv_id":"2608.02977","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For N single photons in a linear-optical interferometer with photon-number-resolving detection, the classical Fisher information at zero phase is exactly k(N-k)/N, maximized by a balanced split between phase and reference modes.","lead":"This paper derives a simple formula for how much phase information a linear-optical interferometer can extract when its probe is N single photons split between phase-encoding and reference paths. It shows that the best configuration splits the photons equally, and compares this detection scheme with local homodyne measurements under photon loss.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed linear loss advantage of PNR over homodyne detection is not controlled: Table II gives F_{1,1}=ηε/2, and Eq. (18) sets η=1, while the homodyne comparison uses a common-loss model with a single ε. Under symmetric loss (η=ε), PNR also scales quadratically.","rationale":"The reader's weakest assumption correctly identifies the uncontrolled loss comparison as the most load-bearing concern. I independently checked the derivation of the central zero-phase formula F=kl/N in Appendix A; the derivation is internally consistent, including the multiplicity counting and the vanishing contribution of ∑γ_i=0 outcomes, and the small-N numerical tables agree. The loss comparison in Section IV, however, is genuinely flawed: Eq. (18) silently sets η=1, while the homodyne calculation uses a common ε for both modes. Under the symmetric loss model implied by the homodyne treatment, the PNR result becomes quadratic, eliminating the claimed qualitative advantage. The abstract-level phase-independence claim is also only numerically supported up to N=6, but that is secondary to the uncontrolled loss comparison because the zero-phase optimality of the balanced partition does not depend on full phase-independence. The reader's conditional verdict remains appropriate: the core analytical result appears sound, but the loss-robustness conclusions require correction and a fair comparison before acceptance.","tokens_in":18679,"tokens_out":17411,"duration_ms":176231,"concrete_test":"Set η=ε in Table II for the (N,k,l)=(2,1,1) row, giving F_{1,1}=ε²/2. Compare its ε→0 scaling with the homodyne result FH,ε(0)≈ε² from Eq. (C4). If both are quadratic in ε, the claimed linear scaling in Eq. (18) is not a property of the protocol under symmetric loss; the comparison requires an explicit justification for why the reference mode is lossless while the homodyne model attenuates both modes equally.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section IV's central practical claim is that PNR detection retains linear loss scaling (F_{1,1}=ε/2, Eq. (18)) whereas local homodyne detection degrades quadratically (FH,ε∼ε², Eq. (17)). The comparison is uncontrolled. Table II gives the lossy PNR result as F_{1,1}=ηε/2, and Eq. (18) is obtained only by setting the reference-mode transmissivity η=1. The homodyne loss model in Appendix C, Eq. (C1), instead applies a single common transmissivity ε to both modes of the two-mode state—equivalent to an all-or-nothing erasure of the single photon. If the PNR result is evaluated under the same common-loss assumption, i.e., η=ε in the paper's own two-parameter notation, then F_{1,1}=ε²/2, which is quadratic in ε, not linear. Thus the advertised linear-versus-quadratic advantage is an artifact of comparing different loss channels. The central zero-phase formula F=kl/N is unaffected, but the loss-robustness conclusion—a headline result of the paper—is not sustained as stated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies N single photons distributed over 2N optical modes arranged in N pairs, with k phase-encoding pairs and l reference pairs, followed by a linear-optical network, quantum Fourier transform, and photon-number-resolving detection. The central result is a closed-form expression for the classical Fisher information at zero phase, F_{k,l}=kl/N (Eq. (7) and Appendix A), showing that a balanced partition k=l maximizes sensitivity for every N and uniquely gives a phase-independent response. The paper then analyzes the effect of loss via independent transmissivities ε and η for phase and reference modes, compares the protocol with local homodyne detection, and extends the framework to multimode distributed sensing with a single photon shared among M receivers, including an analysis of the quantum Fisher information matrix and relative versus symmetric phase combinations.","tokens_in":18929,"tokens_out":5850,"duration_ms":62383,"significance":"If the central derivation is correct, the zero-phase CFI formula is a clean, exact, parameter-free result that offers a simple resource-allocation rule for linear-optical phase estimation with single-photon inputs. The Appendix A derivation is self-contained and is supported by the small-N numerical tables, and the paper contains no fitted parameters. The multimode QFIM analysis in Sec. V gives an appealing geometric explanation for why relative-phase combinations are better estimated than symmetric combinations. However, the paper's headline practical claim that photon-number-resolving detection retains a linear loss scaling whereas homodyne detection degrades quadratically is not supported by the analysis as written, and Table III contains an unphysical loss dependence. These issues are load-bearing for the advertised loss-robustness advantage and must be fixed before the practical claims can be accepted.","major_comments":[{"comment":"The comparison between PNR and homodyne detection is not performed on the same interferometric scheme. The homodyne analysis uses the single-photon state of Eq. (10), which has N=1 photon, whereas the PNR result F_{1,1}=ε/2 is the Table II entry for N=2, k=1, l=1 with η=1. For a single photon in a phase-encoding and a reference mode but with no reference photon (l=0), Eq. (7) gives F=0, so the statement that PNR 'in the same interferometric scheme' retains linear loss scaling is not demonstrated by the quoted entries.","section":"Sec. IV, Eq. (18) and Table II"},{"comment":"The loss models used for the two detection strategies are not matched. The homodyne loss in Appendix C, Eq. (C1), is a common erasure channel acting on the whole two-mode single-photon state with a single parameter ε, while the PNR result in Table II uses two independent transmissivities ε and η. Evaluating the paper's own two-parameter expression under the common-loss condition η=ε gives F_{1,1}=ε²/2, which is quadratic in ε, not linear. The advertised linear-versus-quadratic loss advantage is therefore an artifact of setting η=1 for the PNR comparison while using a common ε for homodyne detection.","section":"Sec. IV and Appendix C"},{"comment":"The variances in Table III are listed as scaling with η^l ε^k, meaning they decrease as loss increases. For example, the N=2, k=1, l=1 row gives Var(φRel)=3ηε/2, which tends to zero as η,ε→0. A variance lower bound must increase under loss, scaling as the inverse of the survival-probability factor, so these entries are unphysical and contradict the text stating that loss degrades precision. This affects the quantitative claims in Sec. V and requires either correcting the entries to inverse scaling or relabeling them as CFI values rather than variances.","section":"Table III and Sec. V"}],"minor_comments":[{"comment":"The text contains the typo 'typically typically'; later in the same section, 'detoriates' should be 'deteriorates'.","section":"Sec. IV, paragraph before Eq. (10)"},{"comment":"The caption begins with 'The table highlights...' but the object is a figure; this should be corrected.","section":"Fig. 2 caption"},{"comment":"The displayed matrix in Eq. (D1) appears garbled, with repeated and misplaced sine terms in the off-diagonal entries; it should be checked and rewritten.","section":"Appendix D, Eq. (D1)"},{"comment":"The reference list contains duplicate entries (Refs. 6 and 16; Refs. 7 and 12).","section":"References"}],"recommendation":"major_revision","confidential_remarks":"I agree with the reader's assessment that the central F=kl/N derivation is sound, but the loss-robustness comparison in Sec. IV and the loss dependence in Table III need substantial reworking. These issues are correctable within the manuscript's scope, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Modak et al. derive a clean, exact result: for N single photons over 2N modes, with k phase-encoding and l reference pairs, the zero-phase classical Fisher information is kl/N, so the balanced partition is optimal for every N. The Appendix A derivation is self-contained, uses only permanents and permutation sums, and the small-N numerics match. That is a genuinely useful design rule for linear-optical phase estimation, and it appears new relative to the cited QFT-interference and telescope work. No fitted parameters, no self-citation issues. Credit where it is due.\n\nThe problems sit in the loss and distributed-sensing sections, and they are real. The advertised linear-versus-quadratic advantage of photon-number-resolving detection over local homodyne detection is uncontrolled. Equation (18), F_1,1 = epsilon/2, comes from Table II with eta = 1, i.e., the reference mode is assumed lossless. The homodyne loss model in Appendix C instead erases the whole single-photon state with one common parameter epsilon. Compare like with like: if the reference suffers the same loss as the phase mode, PNR also gives quadratic scaling, epsilon^2/2. So the robustness claim is overstated; at best it holds for asymmetric loss. The paper should either use a common-loss model for both schemes or state the asymmetry explicitly and re-derive the comparison.\n\nTable III is more worrying. It lists variances such as 3 eta epsilon / 2 that vanish as loss increases. A variance should grow under loss. Equation (26) gives F_Rel = (4/3) F_k,l, so Var_Rel should be inverse in eta^l epsilon^k. The table appears to have inverted numerator and denominator. This needs a correction before anyone relies on the distributed-sensing numbers.\n\nTwo smaller issues. The abstract says the balanced configuration uniquely exhibits a phase-independent response, but the paper only reports numerical evidence up to N = 6. Either prove it or soften the claim. And the claim that PNR avoids quadratic loss scaling is only true in the eta = 1 case; the paper's own Eq. (18) is conditional. The M-mode extension is explicitly incomplete, which is acceptable because it is flagged.\n\nThe paper deserves a serious referee. The central formula is important enough to be in the literature, and the derivation is honest. But I would expect major revision: fix Table III, redo the homodyne comparison under a common loss model, and align abstract claims with the numerical evidence. My verdict is conditional, not rejection.","headline":"The zero-phase CFI formula F=kl/N is solid and new; the loss comparison and Table III are not.","tokens_in":19451,"tokens_out":3057,"would_cite":true,"duration_ms":30777,"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":"For N single photons, phase sensitivity is exactly kl/N and peaks at the balanced split k=l.","keywords":["quantum metrology","distributed phase estimation","classical Fisher information","photon-number-resolving detection","linear optical interferometry","multiphoton interference","optical loss","multimode interferometry"],"falsifier":"Evaluate Eq. (3) numerically at $\\varphi=10^{-6}$ for $N=5$ with partition $(k,l)=(2,3)$: the classical Fisher information must equal $6/5$. Separately, set $\\eta=\\varepsilon$ for the $N=2$, $(1,1)$ lossy calculation; if the Fisher information is $\\varepsilon^2/2$ rather than $\\varepsilon/2$, the claimed linear-loss advantage of photon-number-resolving detection over homodyne detection is an artifact of assuming a lossless reference.","tokens_in":18481,"feed_emoji":"🔬","tokens_out":11474,"duration_ms":117133,"temperature":0.7,"pith_summary":"The paper asks how to distribute $N$ single photons among phase-encoding and reference modes of a linear-optical interferometer to estimate an unknown phase when the output is read by photon-number-resolving detectors. It derives an exact zero-phase formula: the classical Fisher information is $F_{k,l}=kl/N$, where $k$ photons pass through the phase and $l=N-k$ through the reference. The optimal choice is always the balanced split $k=l$, giving $F=N/4$, and any imbalance costs sensitivity quadratically. The same framework is then applied to lossy channels and to distributed multi-receiver arrays, where relative phase combinations carry more information than symmetric ones. If correct, the result is a simple allocation rule for photonic phase sensing.","feed_headline":"Balanced photon split wins: F = kl/N","feed_subtitle":"One formula sets the optimal split of N photons and shows photon counting beats local homodyne under loss.","key_machinery":"The central object is the partition $(k,l)$ of $N$ single photons over $N$ mode pairs, followed by a quantum Fourier transform and photon-number-resolving detection. The argument runs through the exact output probability\n$$P_{\\vec n}\\propto \\left|\\sum_{\\$\\sigma$\\in S(N)} $e^{{i2\\pi \\vec\\gamma\\cdot\\vec\\sigma/N+i\\varphi M_{\\mu,\\sigma}}$}\\right|^2,$$\nwhich is a permanent-type sum over permutations of the detected photons. At zero phase the first derivative of every outcome probability vanishes, so the Fisher information receives contributions only from outcomes with zero probability at $\\varphi=0$; a sum rule over those outcomes collapses to $F_{k,l}=kl/N$. Loss is modeled by coupling each mode to an independent vacuum mode with transmissivities $\\varepsilon$ and $\\eta$, and the distributed-receiver analysis uses the quantum Fisher information matrix of the local number operators, whose eigenvalues are $4/M$ and $4/M^2$.","core_discovery":"The paper's central claim is that for $N$ single photons distributed over $2N$ modes, with $k$ phase-encoding and $l=N-k$ reference pairs, the classical Fisher information for estimating a small phase at $\\varphi=0$ is exactly $F_{k,l}=kl/N$. This formula is maximized uniquely by the balanced partition $k=l$, where $F=N/4$, and it decreases monotonically as the partition becomes asymmetric; the balanced case is also the only one whose response is independent of the phase. With photon loss, the zero-phase Fisher information equals the lossless value times the probability $\\eta^l\\varepsilon^k$ that no photon is lost. The paper further claims that in a lossy low-flux setting, photon-number-resolving detection retains a Fisher information linear in the phase-mode transmission, whereas local homodyne detection degrades quadratically, and that in a distributed single-photon interferometer the relative-phase combinations are estimated with variance $M$ times smaller than the symmetric combination.","pith_inferences":["Editorial extension: writing the formula as $kl/N$ suggests a pairwise picture in which each phase-encoding photon pairs with each reference photon to contribute $1/N$ of the Fisher information. That picture is a natural design heuristic for assigning roles in larger networks, although the paper does not foreground it.","Editorial extension: for distributed sensing, the variance ratio VarSym/VarRel $=M$ implies that as the number of receivers grows, common-mode phase combinations become relatively harder; a practical distributed sensor should therefore encode information in relative and baseline phases.","Editorial extension: an immediate testable consequence is that a photonic chip implementing the quantum Fourier transform network for $N=4$ should show zero-phase Fisher information $1$ for partition $(2,2)$ and $3/4$ for $(1,3)$, visible in the curvature of the output click statistics."],"forward_implications":["At any fixed photon number $N$, equal partitioning of photons between phase and reference modes is the unique maximizer of the zero-phase Fisher information, with $F=N/4$; any imbalance reduces $F$ by the squared imbalance factor.","Under loss, the zero-phase Fisher information is the lossless value times $\\eta^l\\varepsilon^k$, so the effect of loss at that point is purely the probability that no photon is lost.","In the low-transmission regime, photon-number-resolving detection with one phase photon and one reference photon gives a Fisher information linear in the phase-mode transmission, while local homodyne detection gives a quadratic one.","In a distributed $M$-receiver interferometer with one photon, the symmetric phase combination has estimation variance $M$ times that of a relative phase combination, identifying relative phases as the informative parameters.","For the three-receiver architecture, the zero-phase Fisher information for each partition retains the $kl/N$ form with prefactors $4/3$ (relative) and $4/9$ (symmetric), so the balanced partition remains optimal."],"supporting_citations":[{"why":"Motivates the photon-number-resolving measurements whose output statistics define the Fisher information.","marker":"[22]"},{"why":"Provides the permanent formalism used to write the output probabilities in Eq. (3).","marker":"[23]"},{"why":"Supplies the quantum Fourier transform circuits that realize the global mode-mixing before detection.","marker":"[33]"},{"why":"Underpins the realization of the interferometric unitary as a passive linear-optical network.","marker":"[34]"},{"why":"Supplies the beam-splitter-plus-environment model of photon loss used in Sec. III.","marker":"[36]"},{"why":"Gives the bound consistent with the quadratic loss scaling of local homodyne detection.","marker":"[39]"},{"why":"Provides the quantum-Fisher-information covariance formula used for the multi-receiver QFIM.","marker":"[40]"},{"why":"Provides the prior distributed multi-phase estimation architecture that the multi-receiver extension builds on.","marker":"[26]"}],"fun_headline_variants":["Balanced split gives exact Fisher info: F=kl/N","Optimal phase sensing: distribute N photons evenly","F=N/4: the balanced photon split wins","Multiphoton interferometry: balanced split is optimal","Loss-robust phase sensing with balanced photon splits"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise in the loss comparison is that the reference mode in the photon-counting calculation is lossless ($\\eta=1$); if the reference mode loses photons at the same rate as the phase mode, the claimed linear-in-loss advantage over homodyne detection becomes quadratic in the common transmission.","fun_headline_variants_meta":{"raw":{"variants":["Balanced split gives exact Fisher info: F=kl/N","Optimal phase sensing: distribute N photons evenly","F=N/4: the balanced photon split wins","Multiphoton interferometry: balanced split is optimal","Loss-robust phase sensing with balanced photon splits"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000231,"raw_usage":{"total_tokens":1472,"prompt_tokens":918,"completion_tokens":554,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":534,"completion_tokens_details":{"reasoning_tokens":476}},"tokens_in":534,"tokens_out":554,"duration_ms":6758,"temperature":1.0,"reasoning_tokens":476,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T04:23:11.170317+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Evaluate Eq. (3) numerically at $\\varphi=10^{-6}$ for $N=5$ with partition $(k,l)=(2,3)$: the classical Fisher information must equal $6/5$. Separately, set $\\eta=\\varepsilon$ for the $N=2$, $(1,1)$ lossy calculation; if the Fisher information is $\\varepsilon^2/2$ rather than $\\varepsilon/2$, the claimed linear-loss advantage of photon-number-resolving detection over homodyne detection is an artifact of assuming a lossless reference.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Motivates the photon-number-resolving measurements whose output statistics define the Fisher information."},{"cited_title":"Aaronson, A linear-optical proof that the permanent is p- hard, Proceedings of the Royal Society A: Mathematical, Phys- ical and Engineering Sciences467, 3393 (2011)","cited_arxiv_id":null,"evidence_quote":"Provides the permanent formalism used to write the output probabilities in Eq. (3)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the quantum Fourier transform circuits that realize the global mode-mixing before detection."},{"cited_title":"Tsang, Quantum nonlocality in weak-thermal-light interfer- ometry, Phys","cited_arxiv_id":null,"evidence_quote":"Gives the bound consistent with the quadratic loss scaling of local homodyne detection."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the quantum-Fisher-information covariance formula used for the multi-receiver QFIM."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the prior distributed multi-phase estimation architecture that the multi-receiver extension builds on."}],"review_version":1}