{"id":"531918c1-29d2-4789-8bd5-556e8a08d858","arxiv_id":"2610.27872","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The 2k-th moments of permanents and Hafnians of random matrices are computed by a replica-symmetric saddle point for k>2, including prefactors and first 1/N corrections.","lead":"Random matrices with independent entries have permanents and Hafnians whose high moments are hard to compute. This paper derives them in closed form using a replica method, giving exact growth rates and 1/N corrections with direct implications for boson-sampling statistics.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"The complex-case asymptotics rest on an unproved global-maximum premise for |per(y)| on the sphere; Eq. (50) and the Hafnian rates inherit this gap, since the implication from [40] is stated but not demonstrated.","rationale":"The central claim of the paper is the closed-form asymptotic control of the 2k-th moments, Eqs. (30) and (50). The real-Gaussian route to Eq. (30) has an analogous saddle-dominance step, but the paper cites a published source [38] for the global-minimum statement. The genuinely unprotected hinge is the complex case: the paper says the global-maximum statement 'should follow' from [40], gives numerical checks only for k=3,4,5, and then uses that statement to justify the exact asymptotics for arbitrary k. Since the integral (46) is finite-dimensional for each fixed k, the missing proof is plausibly available, and the exact identity (43) together with Table III gives strong empirical support; this is why I do not recommend escalation beyond the reader's CONDITIONAL verdict. The unhand-verified machine algebra in the Appendix is a secondary concern: the final 1/N coefficients are checked against exact moments, and if the dominance premise is supplied the algebra can be independently audited. The reader's weakest assumption and my attack coincide on the same load-bearing premise.","tokens_in":29274,"tokens_out":15831,"duration_ms":106138,"concrete_test":"Extract the Hadamard-type inequality from [40] and write out the implication: for every complex k-by-k matrix y with sum |y_ab|^2 = k, |per(y)| <= k! k^{-k/2}, with equality at y_ab = e^{i theta}/sqrt(k). In particular, verify that the hypotheses of [40] cover arbitrary complex y, not only nonnegative or positive-semidefinite matrices, and that the implied bound has the sharp constant. If the derivation cannot be completed, then Eqs. (49)-(52) and (63) should be labelled conditional on the global-maximum conjecture; a global optimization search for k=6,...,10 would then be a useful falsification test, with any value above k!/k^{k/2} disproving Eq. (50).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The rate, prefactor, and 1/N correction for complex Gaussian permanents, Eqs. (49)-(52), are obtained by saddle-point evaluation of the fixed-k finite-dimensional integral (46). That evaluation is valid only if the replica-symmetric point y_ab = y0 is the dominant critical point, i.e. if |per(y)| is globally maximized on the sphere sum |y_ab|^2 = k. The paper states this in Section IV.A: the RS saddle is 'the global maximum of |per(y)| on the sphere... as we checked numerically for k=3,4,5; this should follow from a Hadamard-type inequality for permanents [40].' No derivation from [40] is supplied, and numerical checks for k <= 5 do not cover arbitrary k. If a competitor critical point with larger |per(y)| existed for some k > 5, the rate Lambda_C(k) would be smaller than Eq. (49), and both the leading asymptotics and the 1/N formula would fail. The real-permanent/Hafnian case has an analogous dominance step, but it is supported by a cited published result [38]; for complex permanents the supporting implication is merely asserted. The Hafnian asymptotics (63), via the dual real-field integral (61), inherit the need for the same |per(s)| maximum, restricted to real s. This is a proof gap rather than an observed inconsistency: the exact identities (43) and (59) and the numerical tables provide real evidence, and a proof for all k is plausible. But the paper's own sentence marks the missing support explicitly, and a conditional verdict is the right response.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper develops a replica method for the moments of permanents and Hafnians of random matrices, starting from contour-integral representations of these matrix functions and a Hubbard-Stratonovich decoupling. For the normalized 2k-th moments of N x N matrices with independent entries, the authors derive exact finite-N integral representations: Eq. (22) for general real symmetric entry laws, Eq. (24) for real Gaussians, Eq. (43) for complex Gaussians, and Eqs. (59) and (61) for Hafnians of symmetric complex Gaussians. A large-N saddle-point analysis around a replica-symmetric saddle then yields closed-form rate functions, polynomial prefactors, and first 1/N corrections, Eqs. (30), (50), and (63), with the rate claimed to be universal within the real (even) and circular complex classes. Exact enumerations for N up to 13 (real and +/-1 entries) and up to N = 60 and N = 300 for complex permanents and Hafnians agree with the asymptotic formulas including the 1/N term. The paper also derives a power-law tail of the normalized permanent distribution, a k = 2 instability separating the bulk from the large-deviation tail, a moment-size duality for complex permanents, Eq. (44), and a duality between real-permanent moments and complex-Hafnian moments. The main technical gap, explicitly acknowledged by the authors, is that the complex saddle-point evaluation assumes a global-maximum property of |per(y)| on a sphere that is verified only numerically for k = 3, 4, 5.","tokens_in":29726,"tokens_out":3443,"duration_ms":22944,"significance":"If the claimed asymptotic formulas hold, this is a substantial advance: it provides closed-form leading asymptotics for random permanents and Hafnians, including prefactors and 1/N corrections, with no fitted parameters. The exact identities (22), (24), (43), and (59) are proven in the text, and the numerical agreement with exact enumerations in Tables I-IV is strong evidence that the saddle-point expansion is correct. The moment-size duality (44) and the Hafnian-permanent duality (59) are elegant and useful, and the derivation is considerably simpler than the representation-theoretic route in [39]. The authors also deserve credit for disclosing the machine-assisted algebraic calculations in the Appendix and for stating the limitations of the bulk-distribution and anti-concentration results. The main obstacle to full rigor is the unproved global-extremum premise for |per(y)| on the sphere for general k in the complex case, which affects the central asymptotic claim (50) and, through (62)-(63), the Hafnian rates.","major_comments":[{"comment":"","section":"Section IV.A, paragraphs around Eqs. (46)-(50)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"","section":"Section IV.B, Eqs. (61)-(64)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"","section":"Appendix, disclaimer after Eq. (103)","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"","section":"Section V, Eqs. (65)-(70)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"","section":"Section VI, Eqs. (73)-(74)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"","section":"Section III.B, Table I","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"","section":"Notation in Section IV.A, Eq. (40)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"","section":"Section I, paragraph on Tao-Vu","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision","confidential_remarks":"The paper is strong and potentially important, and the main gap is openly disclosed by the authors themselves rather than hidden. I would encourage the editor to treat the global-maximum premise as the key condition for acceptance: a proof from [40] or a rigorous verification for general k would move this to acceptance, while a refusal to address it would leave the central complex-case theorem conditional. The LLM-assisted algebra is a governance concern for reproducibility but not, on the present evidence, a correctness concern; the numerical tables provide a strong consistency check. The manuscript is squarely within the scope of the journal and should not be rejected on novelty grounds."},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"Punchline: worth a serious look, and the reader's conditional verdict is the right one. This paper derives closed-form asymptotics — exponential rate, prefactor, and first 1/N correction — for the 2k-th moments of random permanents and Hafnians via a Hubbard-Stratonovich replica treatment. The central identities (22), (43), and (59) are exact, the saddle-point expansion in the Appendix is explicit, and the formulas agree with exact enumerations and exact finite-N values, including the k=2 exact result ⟨Y²⟩=N+1 and the 1/N coefficient fits. That is real evidence, not numerology.\n\nWhat is new: the replica derivation itself, the real-Gaussian and Rademacher asymptotics, the Hafnian results, the 1/N corrections, and the k=2 transition with its power-law tail. The complex-permanent rate and the moment-size duality were already in Nezami [39]; the paper cites him and uses his results for comparison. The dominant-saddle step in the real case is supported by a cited published result [38]; that part is fine.\n\nTwo soft spots, in proportion. First, the complex-case asymptotics rest on the claim that the replica-symmetric point y_ab=y0 is the global maximum of |per(y)| on the sphere Σ|y|²=k. In Section IV.A the paper says it checked this numerically for k=3,4,5 and that it should follow from a Hadamard-type inequality in [40], but no derivation is supplied. For arbitrary k the premise is unproved, and Eqs. (49)–(52) plus the Hafnian rates via (61) inherit it. This is a proof gap rather than an observed inconsistency; the exact identities and tables make the statement plausible. The paper's own sentence flags the missing support, and a conditional verdict is the honest response.\n\nSecond, the Appendix discloses that much of the algebra was generated by Claude and not hand-verified. That disclosure is candid, and the cross-checks are real: propagators against direct Hessian inversion, Wick contractions against brute-force tensor sums, and final coefficients against exact moments. But “checked by an adversarial LLM” is not a mathematical check; the most delicate sums deserve hand verification before the formulas are treated as rigorous.\n\nThe photonics section is properly scoped — lossless, collision-free, many-mode assumptions are stated, and the authors explicitly leave the bulk distribution and anti-concentration alone. Audience: random matrix theory and boson-sampling people. It deserves a serious referee. My recommendation: send it to review, ask for a proof (or a clearly stated conjecture) of the complex-saddle global-maximum step, and request independent verification of the machine-generated pieces.","headline":"Closed-form moment asymptotics for random permanents and Hafnians via replicas, with solid numerics and one honest proof gap: the dominant complex saddle is only checked for k≤5.","tokens_in":30253,"tokens_out":3818,"would_cite":true,"duration_ms":24735,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"pith_extraction":null,"created_at":"2026-10-21T12:00:17.572807+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"falsifier":null,"supporting_citations":[],"review_version":1}