{"id":"36996f88-2905-4ae4-a01a-9ab14b87e9b8","arxiv_id":"2602.10151","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"N-point functions of locally non-Gaussian fields can be expanded in powers of the Gaussian correlation function using the Kibble-Slepian decomposition, with coefficients from one-point integrals; for an exponential-tail model the strong-non-Gaussian two-point function is exactly (ξ0/8β²)arcsin²(ξ_G/","lead":"This paper shows how to compute correlation functions of non-Gaussian random fields using a mathematical series expansion, even when the field is not a smooth function of a Gaussian seed. It gives exact results for a model with exponential tails, relevant to early-universe cosmology.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"For N≥3, the Kibble–Slepian expansion (3.9) may diverge for |ξ_ij|<ξ0; the claimed exactness for non-analytic F is unproven and the two-point convergence bound does not extend.","rationale":"The reader’s weakest assumption identified the same load-bearing issue: the unproven interchange of summation and integration and the resulting series convergence. My analysis sharpens this: the two-point series is actually safe, but for N≥3 the absolute convergence is not guaranteed and a heuristic estimate suggests the convergence radius in ψ can be as small as 1/2 for strongly non-Gaussian fields. This does not contradict the reader’s CONDITIONAL verdict, but it strengthens the necessity of a convergence proof or a qualification of the exactness claim. The paper’s own admission of poor behavior near |ξ_G/ξ0|→1 is consistent with this concern. The proposed numerical test would settle whether the series diverges in a concrete model.","tokens_in":18167,"tokens_out":43533,"duration_ms":402417,"concrete_test":"For the exponential-tail model with ¯β=1, compute G3(ξ12,ξ13,ξ23) by direct 3D integration of (2.14) for a configuration with ψ12=0 and ψ13=ψ23=ψ0, choosing ψ0=0.5 and ψ0=0.8. Then evaluate the series (3.39)-(3.40) truncated at total multiplicity ν_tot = 10, 20, 30. If the partial sums do not converge to the direct numerical integral but instead grow with truncation order, the exactness claim for N≥3 fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that (3.9) gives the exact N-point function for locally non-Gaussian fields with non-analytic F. For N=2, the series (3.35) is absolutely convergent for |ψ|≤1 whenever F has finite variance, because ∑ n! C_n² = ⟨F²⟩ < ∞ and |ψ|^n ≤ 1. For N≥3, the terms in (3.9) are products of C_{s_i} with combinatorial factors s_i!/(∏ ν_ij!), and no analogous bound controls their sum. Writing C_s = a_s/√(s!) with ∑ a_s² < ∞, the factor s_i!|C_{s_i}| = √(s_i!)|a_{s_i}| can grow super-exponentially. A saddle-point estimate for the exponential-tail model (where C_s ~ s^{-3/4}/√(s!)) indicates that the dominant balanced diagram at total order S scales as (√(2ψ))^S S^{-3/4}, which diverges when ψ>1/2. Thus the multivariate series likely has a radius of convergence smaller than 1 for N≥3, making it at best asymptotic, not exactly convergent over the physical domain |ψ_ij|<1. The paper provides no convergence proof for N≥3, and its numerical checks in Sec. 4 only address the two-point function. This directly undermines the claim that (3.9) is exact for non-analytic F.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a formalism for N-point functions of locally non-Gaussian fields ζ(x)=F(ζ_G(x)) without Taylor-expanding F. Using the Kibble–Slepian decomposition of the multivariate Gaussian distribution, Eq. (3.9) expresses the N-point function as a power series in the Gaussian correlation ξ_ij, with all F-dependence encoded in one-point coefficients C_s defined by Eq. (3.10). The authors give diagrammatic Feynman rules, explicit formulas for bispectra and trispectra, and apply the method to an exponential-tail model, deriving an exact strongly non-Gaussian two-point function, Eq. (4.9), and studying the resulting deformation of the power spectrum.","tokens_in":18657,"tokens_out":8062,"duration_ms":90129,"significance":"If the convergence question is settled, this is a valuable framework: it factorizes the model-dependent and spectrum-dependent parts of the computation, extends beyond analytic mappings F, and provides non-perturbative benchmarks for PBH and SIGW applications. The central identity is derived from a rigorous mathematical formula; the coefficients C_s are computed from F rather than fitted to the target correlation function; numerical checks for N=2 are reported; and the limitations of the exponential-tail model are stated explicitly. The main unresolved issue is whether the multivariate series (3.9) is an exact convergent representation for N≥3 and non-analytic F; the paper currently provides no proof of this.","major_comments":[{"comment":"The derivation of Eq. (3.9) integrates the Kibble–Slepian series term by term. For N=2, Eq. (3.35) is absolutely convergent for |ψ|≤1 because ∑ n!C_n²=⟨F²⟩<∞. For N≥3 no analogous bound is given: the terms contain products ∏ s_i!C_{s_i}; writing C_s=a_s/√(s!) with ∑a_s²<∞, the factors √(s_i)!|a_{s_i}| can grow. A saddle-point estimate for the exponential-tail model gives dominant balanced diagrams scaling as (√(2ψ))^S S^{-3/4}, suggesting divergence when ψ>1/2. Thus exactness of (3.9) for non-analytic F is not established; the numerical checks in Sec. 4 cover only the two-point function. Please either prove convergence under stated assumptions or explicitly qualify (3.9) as asymptotic/formal for N≥3 and provide practical convergence criteria.","section":"Sec. 3.1, Eq. (3.9)"},{"comment":"The exact strong-NG result Eq. (4.9) is derived in the footnote only as a 'rough outline' via derivatives of the hypergeometric function 2F1, with intermediate steps omitted. Since this is a headline exact analytic result, please provide a complete derivation, or alternatively state it as a conjecture supported by numerical evidence.","section":"Sec. 4, footnote 11 / Eq. (4.9)"}],"minor_comments":[{"comment":"The text refers to 'defining C_n in Eq. (3.35)', but Eq. (3.35) is the two-point expansion; the definition appears in Eq. (3.10). Cross-reference error.","section":"Sec. 3.1, near Eq. (3.14)"},{"comment":"The notation P^{*ν_ij}_G(q_ij) ≡ (2π²/q³) P^{*ν_ij}_G(q_ij) is self-referential; the convolution power should be denoted with a different symbol or explicitly defined.","section":"Eq. (3.23)"},{"comment":"Extra closing parenthesis in 'F(ζ_G)) = C^{-1}(C_G(ζ_G))'.","section":"Eq. (2.12)"},{"comment":"Typo: '4P: T rispectra' should be 'Trispectra'.","section":"Sec. 3.5 heading"},{"comment":"The text says the series (4.6) is asymptotic; it would help to also state clearly which of the subsequent series in the paper are exact and which are formal/asymptotic, especially in relation to (3.9) for N≥3.","section":"Sec. 4, Eq. (4.6)"}],"recommendation":"major_revision","confidential_remarks":"The convergence concern is the main substantive obstacle. The manuscript would be acceptable after the authors either prove a convergence theorem for Eq. (3.9) under stated conditions or carefully rephrase the exactness claim and add convergence diagnostics for N≥3. I see no circularity or citation-bias issue; the comparison with Ref. [33] is not used as input to the derivation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the Kibble–Slepian apparatus gives you a clean, one-point resummation of local non-Gaussianity, and the β→∞ two-point function for the exponential-tail model is a real result. But the paper's stronger claim—that the N-point expansion (3.9) is exact for non-analytic F—doesn't survive contact with convergence for N≥3. I'd send it out, but the revision has to address that.\n\nWhat is actually new: the reorganisation of the usual F_NL expansion into coefficients C_s that absorb all single-point non-linearities is elegant, and Eq. (3.13) is a useful bridge to the standard language. For N=2 the result is rock solid: the series (3.35) is absolutely convergent for |ξ_G|≤ξ0 whenever ⟨F²⟩<∞, exactly because the sum of n! C_n² is finite. The exponential-tail example is well-motivated, and the exact strong-NG limit (4.9) passes the smell test and is numerically checked. The power-spectrum calculations in Sec. 4 are careful and show the flattening plus the k^3 IR tail; that's a concrete payoff for the PBH/SIGW community.\n\nSoft spots: the convergence of (3.9) for n≥3 is neither proven nor discussed, and the stress-test argument is convincing that it fails. For n=3 with all Gaussian correlations equal to ρ, the multivariate Kibble–Slepian series has a natural radius |ρ|<1/2, because the density itself has a singularity at ρ=−1/2 (and ρ=1). After integrating against a non-analytic F with infinite C_s, that radius doesn't necessarily improve. The saddle-point estimate in the stress-test for the exponential-tail model—dominant diagrams growing like (√(2ψ))^S—indicates divergence for ψ>1/2. That means (3.9) is not an exact convergent representation over the whole physical domain for N≥3; it's an asymptotic expansion, useful but not 'exact'. The paper should say this plainly and prove the two-point case while hedging the N≥3 case.\n\nAlso minor: the derivation of (4.9) is sketched via a hypergeometric derivative; a few more steps would help. And the paper itself notes in Sec. 5 that the covariance parameter space grows quickly for n≥3, which limits practical application, but that's a limitation of the problem, not the method.\n\nBottom line: the paper deserves a serious referee. It's a genuine extension of the resummed vertex technique for SIGW (Ref. [26]) and the two-point results are exact and useful. But the advertised 'exact for non-analytic F' needs qualification. I'd recommend minor-to-major revision, not rejection.","headline":"Useful resummation for locally non-Gaussian fields, but the 'exact' N≥3 claim outruns the convergence proof—treat the expansion as asymptotic beyond two points.","tokens_in":19017,"tokens_out":6769,"would_cite":true,"duration_ms":75402,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"One formula gives exact N-point functions of non-Gaussian fields, even when the local mapping is non-analytic.","keywords":["locally non-Gaussian fields","Kibble–Slepian formula","N-point correlation functions","non-analytic mapping","exponential-tail model","primordial black holes","scalar-induced gravitational waves","power spectrum IR tail"],"falsifier":"Evaluate G_2(ξ_G/ξ_0) for the exponential-tail model at an intermediate β (say β√ξ_0 ≈ 1) by direct numerical double integration of the bivariate Gaussian average, and compare against truncated versions of Eq. (3.35) for |ξ_G/ξ_0| between 0.9 and 1; if the truncated series does not approach the exact integral as more terms are included, the claimed exactness of the expansion is falsified. Alternatively, test Eq. (4.9) directly at large β by comparing the closed-form arcsin² expression to Monte Carlo integration of the original bivariate average.","tokens_in":18129,"feed_emoji":"🌀","tokens_out":5313,"duration_ms":55724,"temperature":0.7,"pith_summary":"This paper claims that every N-point function of a locally non-Gaussian field ζ(x)=F(ζ_G(x))—including non-analytic mappings—can be written exactly as a sum over powers of the Gaussian correlation function, with all non-Gaussian information packed into one-point coefficients C_s. The derivation uses the Kibble–Slepian decomposition of the multivariate Gaussian density, replacing the usual Taylor expansion of F with a semi-perturbative series in the Gaussian correlation function. For a model with exponential tails, the paper derives an exact strongly non-Gaussian two-point function and shows that the normalized correlation saturates while the power spectrum develops a k^3 infrared tail. The result matters because it extends analytic control to strongly non-Gaussian regimes relevant to primordial black holes and scalar-induced gravitational waves.","feed_headline":"One formula gives exact N-point functions of non-Gaussian fields","feed_subtitle":"It works even for non-analytic local mappings, opening strongly non-Gaussian regimes to analytic control.","key_machinery":"The Kibble–Slepian decomposition: an identity that expands the n-dimensional Gaussian probability density in products of one-dimensional Hermite-polynomial densities, with powers of the correlation matrix serving as expansion coefficients. Applied to the local field ζ=F(ζ_G), it replaces the multidimensional path integral with sums of products of ξ_ij times resummed one-point coefficients C_s. These coefficients absorb all information about F and can be computed by a one-dimensional Gaussian integral even when F is non-analytic.","core_discovery":"The paper establishes the exact expansion (3.9): any N-point function of a locally non-Gaussian field is a sum over multiplicity matrices ν_ij of products (ξ_ij^{ν_ij}/ν_ij!) times s_i! C_{s_i}, where s_i is the total multiplicity at point i and C_s is a one-point Gaussian expectation value of the rescaled Hermite polynomial ar H_s with F. This remains meaningful when F(ζ_G) is not analytic, because C_s is defined as an integral rather than through a Taylor expansion. In the exponential-tail model, the paper evaluates the coefficients and derives the closed-form two-point function G_2 = ξ_0/(8β²) arcsin²(ξ_G/ξ_0) in the strong non-Gaussianity limit, from which the saturation of the normalize","pith_inferences":["If the exact expansion holds for non-analytic F, the standard practice of expanding curvature perturbations as a polynomial in the Gaussian field may be unnecessary for a whole class of models; this could change how strongly non-Gaussian SIGW and PBH constraints are computed.","The k^3 infrared tail is a generic prediction of local non-Gaussianity whenever G_2 is unrelated to ξ_G; this gives a target observable feature that could be searched for in induced gravitational-wave backgrounds.","The paper's method still leaves n≥3 practical: the parameter space of covariances grows rapidly, so fully non-perturbative bispectra and trispectra may need interpolation or compressed representations; the paper notes this as an obstacle.","One could test the formalism on a known analytic example (e.g., F quadratic) to benchmark truncation errors, then extend to F with a branch point; the paper does not perform such an explicit convergence study."],"forward_implications":["For any locally non-Gaussian field, the N-point function is determined entirely by the Gaussian correlation function and the coefficients C_s, so the mapping G_n can be precomputed independently of the shape of the power spectrum.","The formalism extends to non-analytic F, so strongly non-Gaussian models with exponential tails, logarithms, or piecewise continuations can be treated without a Taylor expansion.","For the exponential-tail model, the exact strong-NG two-point function gives a normalized correlation ξ(x)/ξ(0) = (4/π²) arcsin²(ξ_G(x)/ξ_0), independent of β, and a power spectrum with a k^3 infrared tail.","The series in C_s captures nonlinearities where the conventional F_NL,n expansion fails; the paper estimates that perturbativity breaks down when ξ_0 f_NL² ≲ O(10^-2)."],"fun_headline_variants":["Exact N-point functions even for non-analytic mappings","Non-perturbative framework goes beyond local expansions","Strongly non-Gaussian regimes now exactly solvable","Compute polyspectra without Taylor expansions"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that term-by-term integration of the Kibble–Slepian series against arbitrary non-analytic F is legitimate and that the resulting power series in the Gaussian correlation function represents the exact N-point function; the paper gives no convergence proof and notes the series is poorly behaved near full correlation.","fun_headline_variants_meta":{"raw":{"variants":["Exact N-point functions even for non-analytic mappings","Non-perturbative framework goes beyond local expansions","Strongly non-Gaussian regimes now exactly solvable","Compute polyspectra without Taylor expansions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00028,"raw_usage":{"total_tokens":1467,"prompt_tokens":681,"completion_tokens":786,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":425,"completion_tokens_details":{"reasoning_tokens":725}},"tokens_in":425,"tokens_out":786,"duration_ms":8693,"temperature":1.0,"reasoning_tokens":725,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T03:03:52.798835+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Evaluate G_2(ξ_G/ξ_0) for the exponential-tail model at an intermediate β (say β√ξ_0 ≈ 1) by direct numerical double integration of the bivariate Gaussian average, and compare against truncated versions of Eq. (3.35) for |ξ_G/ξ_0| between 0.9 and 1; if the truncated series does not approach the exact integral as more terms are included, the claimed exactness of the expansion is falsified. Alternatively, test Eq. (4.9) directly at large β by comparing the closed-form arcsin² expression to Monte Carlo integration of the original bivariate average.","supporting_citations":[],"review_version":1}