{"id":"fd1d3d29-e018-4d30-b7f0-1e6b6be88db3","arxiv_id":"1908.01474","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper claims that for the Kingman coalescent, the probability that the block count equals (2+v√t)/t at small time t is asymptotic to a Gaussian-type expression of order √t.","lead":"This paper derives a complex integral representation for the finite-time block-counting probabilities of the Kingman coalescent and uses steepest descent to claim a local central limit theorem at small times. The new integral representation is a technical contribution, but the proof of the main asymptotic contains inconsistencies in the expansion and constants.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 3.3's quadratic coefficient is -1/48, not -1/72; the steepest-descent expansion and the v^3/v^2 inconsistency leave Theorem 3.2 unsupported.","rationale":"The paper's central claim is a small-time local central limit theorem for the Kingman block-counting process, and the proof rests on the steepest-descent analysis in Lemma 3.3. That analysis is the weakest point: the displayed quadratic coefficient is arithmetically wrong. Replacing w by (z+π)/2 and keeping the fourth-order term of log cos shows the coefficient is -1/48, not -1/72. This is not a cosmetic typo, because it changes the variance of the limiting Gaussian and hence the normalization of the local limit theorem. The reader's verdict identified exactly this assumption, and the manuscript's own inconsistency between -3v^3/4 and -3v^2/4 reinforces the concern. I therefore do not find a reason to move the verdict: the conclusion REJECT remains appropriate. I would still recommend the concrete re-derivation as a decisive check, because if the corrected coefficient and prefactors happen to reproduce the theorem's constant, the paper could be salvageable with revised proof details; as written, however, the central claim is unsupported.","tokens_in":12713,"tokens_out":34044,"duration_ms":314026,"concrete_test":"Recompute Lemma 3.3 from scratch: set w = (z+π)/2, z(y) = -π + t^{1/4}e^{iπ/4}√y, expand (1/t)ψ_t(w) through order y^2, and then evaluate the Gaussian integral exactly. If the y^2 coefficient is -1/48 (as a direct expansion shows) rather than -1/72, the printed derivation must be corrected. Then recompute the prefactor of Theorem 3.2 using the corrected Gaussian and an independent Stirling evaluation of (2n-1+θ choose n)/2^{2n-1+θ}; check whether the resulting constant matches 3/(2√(2π)). If it does not, the theorem's normalization is not supported by the proof as written.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is Lemma 3.3's steepest-descent expansion of (1/t)ψ_t along L1,t. Substituting w = (z(y)+π)/2 = 2^{-1} t^{1/4} e^{iπ/4}√y into ψ_t(w) = -2w^2 - (2n+θ)t log cos w and using log cos w = -w^2/2 - w^4/12 + O(w^6) gives the y^2 coefficient from the fourth-order term: (1/t)(2n+θ)t·w^4/12 = (2n+θ)t·(-t y^2/16)/12 / t = -(2n+θ)t/192, whose limit is -4/192 = -1/48. The paper prints -(2n+θ)t/(12×24), i.e. -1/72. The two coefficients produce different Gaussian integrals: with -1/48, ∫ exp(ivy/4 - y^2/48)dy = 4√(3π)e^{-3v^2/4}, whereas the printed -1/72 would give √(72π)e^{-9v^2/8}. Ironically, the constant and exponent finally stated in Lemma 3.3, namely √(3π) and e^{-3v^2/4} (the latter appears as v^3 in the printed formula), correspond to the corrected -1/48 coefficient, not to the coefficient used in the derivation. Thus the proof of Lemma 3.3 is internally inconsistent, and Theorem 3.2 inherits this problem. Independently, the theorem statement and its proof disagree on the exponent: Theorem 3.2 displays e^{-3v^3/4}, while the proof's last line uses e^{-3v^2/4}. The contour deformation in Theorem 3.1 also relies on asserted cancellation along cut lines without a complete justification, but the arithmetic discrepancy in the steepest-descent expansion is the more direct obstruction to the central claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper starts from Tavaré's infinite-series expression for the probabilities d_θ^n(t) = P(D_t = n) of the block-counting process of the Kingman coalescent and rewrites it as a contour integral (Theorem 2.1). It then deforms the contour and uses the Jacobi triple product to obtain a second representation (Theorem 3.1). Finally, a steepest-descent analysis of this representation is used to claim a small-time local central limit theorem (Theorem 3.2): for n = floor((2 + √t v)/t), d_θ^n(t) ∼ (3/(2√(2π))) √t e^{-3v^3/4}. The main novelty claimed is the integral representation suited to asymptotic analysis and the new local CLT.","tokens_in":13138,"tokens_out":25250,"duration_ms":242064,"significance":"The approach is appealing: an explicit integral representation could become a useful tool for small-time asymptotics, and the derivation is non-circular, starting from Tavaré's independent series rather than assuming the conclusion. The paper also contains concrete, falsifiable asymptotic predictions. However, the core asymptotic calculation contains arithmetic inconsistencies that currently invalidate the proof; the result as stated is not established. If the constants and exponents are corrected and the contour deformation is justified, the method may still yield a publishable local CLT.","major_comments":[{"comment":"The quadratic coefficient in the steepest-descent expansion is miscalculated. Substituting w = (z(y)+π)/2 = 2^{-1} t^{1/4} e^{iπ/4} √y into ψ_t(w) = -2w^2 - (2n+θ)t log cos w and using log cos w = -w^2/2 - w^4/12 + O(w^6) gives (1/t)ψ_t(z+π/2) = ((2n+θ)t - 4)/(8√t) i y - (2n+θ)t/192 y^2 + O(√t log^6(1/t)). The printed coefficient -(2n+θ)t/(12·24) = -(2n+θ)t/288 differs from -(2n+θ)t/192, so the limiting Gaussian variance is -1/72 instead of -1/48. This changes the Gaussian integral and the final constant in Lemma 3.3. The displayed Bernoulli expansion in Eq. (2) also has the wrong sign for the z^4 term under the standard convention B_4 = -1/30: ψ_t should have coefficient +(2n+θ)t/12 for z^4, and the y^2 term in Lemma 3.3 is negative only because w^4 = -t y^2/16.","section":"§3.2, Lemma 3.3, Eq. (2)"},{"comment":"The theorem statement and its proof disagree on the exponential argument: Theorem 3.2 displays e^{-3v^3/4}, while the last line of the proof uses e^{-3v^2/4}. A probability approximation of the form e^{-3v^3/4} is invalid because it is unbounded as v → -∞; the correct Gaussian form must be e^{-3v^2/4}. This is not a purely cosmetic typo because the asymptotic statement as printed is not a probability density in v.","section":"§3.2, Lemma 3.3 and Theorem 3.2"},{"comment":"Even after correcting the quadratic coefficient, the constant 3/(2√(2π)) does not follow from the stated Lemma 3.3 and Eq. (1). With the corrected expansion, Lemma 3.3 gives ∫_{C+}(K_t(z)-K_t(-z))φ_t(z)dz ∼ -i√(3π) √t e^{-3v^2/4}. Using Eq. (1), which has the factor i/√(2πt), and the Stirling estimate (2n-1+θ choose n)/2^{2n-1+θ} ∼ √t/√(2π), one obtains d_θ^n(t) ∼ (√3/(2√π)) √t e^{-3v^2/4}, not 3/(2√(2π))√t e^{-3v^2/4}. The intermediate line in the proof of Theorem 3.2 and the displayed 'Therefore' are arithmetically incompatible; the proof must be recomputed.","section":"Proof of Theorem 3.2"},{"comment":"The displayed identity ∫_{C+}(K_t(z)-K_t(-z))φ_t(z)dz = -∫_{L1}K_t(z)φ_t(z)dz + ∫_{L1}K_t(z)φ_t(z)dz has two identical integrals on the right-hand side, so as printed it asserts that the left-hand side is zero. The intended relation among L1, L2, and the map z ↦ -z is not stated precisely, and the assertion that contributions along the branch cuts cancel is not proved. A complete contour-deformation argument is needed before Lemma 3.3 can be applied.","section":"§3.2, contour deformation after Eq. (1)"}],"minor_comments":[{"comment":"Citation numbers are inconsistent between the abstract and the body: Kingman is cited as [7] in the abstract but [9] in the text, and Tavaré is cited as [12] in the abstract but [13] in the text. There are also typographical errors such as 'bench-mark' and 'ininitely'.","section":"Abstract and Introduction"},{"comment":"The statement contains a redundant and confusing double equality: |∫_{L^c_{1,t}} K_t φ dz| = |∫_{L^c_{1,t}} K_t φ dz| ≤ ... . Only one integral should appear on the left-hand side.","section":"Lemma 3.2"},{"comment":"The phrase 'steep descent' should be 'steepest descent' throughout Sections 3.1 and 3.2, and the figures should be referred to at the points where the contours are first described.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The main theorem's constant appears to be wrong as printed; the proof of Lemma 3.3 and the contour deformation both need substantial reworking. I recommend a major revision rather than reject because the integral representation idea is promising and the corrected calculations may still yield a valid local CLT, but the authors must rederive the constants and justify the contour deformation before the paper can be considered."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nYou asked about arXiv:1908.01474. The honest summary: the paper has a sensible plan and one genuinely new representation, but the main theorem is not proved as written. The central local CLT may well be true; the proof just doesn't support it.\n\nWhat's new: Theorem 2.1 is, as the author says, essentially Griffiths' (2.16), so that's not new. Theorem 3.1's theta-function integral representation is new and could be a useful tool for small-time asymptotics. The paper also does a fair job of citing the relevant literature (Tavaré, Griffiths, Limic–Talarczyk, Depperschmidt et al.). The writing is clear, and the idea of using steepest descent on the integral representation is attractive.\n\nWhere it breaks down: Lemma 3.3 contains a load-bearing arithmetic error. The paper expands psi_t along L1,t with a quadratic coefficient -(2n+θ)t/(12×24) = -1/72 in the limit, but the actual expansion gives -(2n+θ)t/192, which in the limit is -1/48. This changes the Gaussian integral and the final constant. The proof of Lemma 3.3 then states a constant, sqrt(3π), that actually corresponds to the corrected -1/48 coefficient, not the -1/72 used in the derivation—so the argument is internally inconsistent. On top of that, Theorem 3.2 states the exponent as -3v^3/4, while the proof's last line and the derived Gaussian integral produce -3v^2/4. The v^3 form is almost certainly a typo, but it matters because the whole point is a precise asymptotic. The contour deformation in Theorem 3.1 ('integrations on cut lines are perfectly cancelled') is asserted rather than demonstrated; that may be fixable but it's not a proof as written.\n\nThe stress-test note, which I checked against the displayed formulas, is correct on the arithmetic. So the paper's central claim is unsupported at the moment. There is enough of a good idea here that I would not dismiss it, but it needs a substantial revision: correct Lemma 3.3, reconcile the exponents, and give a proper justification of the contour deformation.\n\nFor peer review: yes, I'd send it to a referee. The topic is relevant, the approach is potentially publishable, and a competent referee could verify whether the error is easily fixed. But I wouldn't accept the current version. If you're looking for a clean result to cite, wait until the revision.","headline":"The integral representation idea is worth a look, but the main local CLT is not proved: Lemma 3.3's quadratic coefficient is wrong and the exponents are inconsistent.","tokens_in":13634,"tokens_out":2125,"would_cite":false,"duration_ms":20822,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60F10","60C05"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that the block-counting process of the Kingman coalescent satisfies a small-time local central limit theorem: for $n=\\lfloor(2+\\sqrt{t}\\,v)/t\\rfloor$, $d_\\theta^n(t)\\sim \\frac{3}{2\\sqrt{2\\pi}}\\sqrt{t}\\,e^{-3v^3/4}$…","keywords":["Kingman coalescent","block-counting process","finite-time distribution","integral representation","local central limit theorem","steepest descent","small-time asymptotics","Jacobi theta function"],"falsifier":"Expand $\\frac1t\\psi_t((z(y)+\\pi)/2)$ at $z(y)=-\\pi+t^{1/4}e^{i\\pi/4}\\sqrt y$ using $\\psi_t(z)=(-2+(2n+\\theta)t/2)z^2+\\cdots$; compare the coefficient of $y^2$ with the paper's $-(2n+\\theta)/(12\\cdot24)$. The direct expansion gives $-(2n+\\theta)/192$, which on the scale $(2n+\\theta)t\\to4$ is $-1/48$ rather than $-1/72$. Evaluating the resulting Gaussian integral and comparing it with the original series for small $t$ and fixed $v$ would show whether the stated constant $3/(2\\sqrt{2\\pi})$ is correct.","tokens_in":12492,"feed_emoji":"🧬","tokens_out":13608,"duration_ms":124567,"temperature":0.7,"pith_summary":"The paper's aim is to convert the unwieldy explicit distribution of the Kingman coalescent into a form that supports asymptotic analysis. The author starts from the known infinite-series formula for $d_\\theta^n(t)=P(D_t=n)$, rewrites the combinatorial coefficients and exponential factors as contour integrals, and sums the resulting geometric series to get a clean integral representation. A second, deformed representation then feeds a steepest-descent argument whose payoff is a local central limit theorem in the small-time regime: the block count $D_t$ fluctuates around $2/t$ on the $\\sqrt{t}$ scale with a Gaussian-type law. If the argument is correct, it is the first local CLT for this process at small times, and the same integral representation is proposed as a route to large and moderate deviations.","feed_headline":"Kingman coalescent block counts obey a small-time local limit law","feed_subtitle":"A contour-integral rewrite of the unwieldy series exposes the small-time fluctuations of the block count.","key_machinery":"The argument hangs on the phase function $\\psi_t(z)=-2z^2-(2n+\\theta)t\\log\\cos z$, which appears as $\\exp(\\psi_t((z+\\pi)/2)/t)\\sin((z+\\pi)/2)\\varphi_t(z)$ in the integrand. The real part of $\\psi_t$ decreases monotonically along the deformed contour $L_1$ away from the critical point $z=-\\pi$, so the integral localizes near $-\\pi$; the Jacobi triple product identity converts the $\\theta$ series into the explicit infinite product $\\varphi_t(z)$, and the oddness of the full integrand allows the two contour pieces to be combined. On the short segment $L_{1,t}$ the integral is reduced to a Fourier-type integral in a variable $y$, which produces the claimed $\\sqrt t$ scale and exponential factor.","core_discovery":"The central claim is that a contour-integral representation makes the finite-time distribution of the Kingman coalescent analytically accessible. Theorem 2.1 writes $d_\\theta^n(t)$ as a one-dimensional complex integral whose integrand is $e^{-w^2/2t}\\,i\\sin(w/2)/(\\cos(w/2))^{2n+\\theta}$ times an explicit prefactor. Theorem 3.1 deforms the integration contour to a unit circle and uses the Jacobi triple product to absorb the $\\theta$ sum into an infinite product $\\varphi_t(z)$. The main theorem then claims that for $n=\\lfloor(2+\\sqrt t v)/t\\rfloor$, $d_\\theta^n(t)\\sim \\frac{3}{2\\sqrt{2\\pi}}\\sqrt t\\,e^{-3v^3/4}$ as $t\\to0$, describing the fluctuation of the block count around $2/t$ on the $\\sqrt t$ scale. The theorem as printed displays the exponent $-3v^3/4$, while the closing line of the proof writes $-3v^2/4$; this exponent discrepancy is one of the load-bearing points checked by the calculation in the falsifier below.","pith_inferences":["If the quadratic-coefficient discrepancy is real, the local-CLT shape may survive but the variance and prefactor would change; checking the $y^2$ coefficient directly is the cheapest way to test the stated constant.","A corrected Gaussian factor would read as $\\exp(-c v^2)$ for some constant $c$, not the cubic exponent printed in the theorem; a careful re-derivation of Lemma 3.3 would determine $c$.","The same contour machinery could be tried on $\\Lambda$-coalescents with a Kingman component, where small-time fluctuations have so far been studied with probabilistic rather than analytic methods; the analytic route would be a new comparison.","Controlling the remainder terms in the uniform convergence step (3) more carefully should yield higher-order asymptotic expansions in powers of $\\sqrt t$, extending the local CLT to a full small-time expansion."],"forward_implications":["If the theorem holds, the block count has the small-time fluctuation description $D_t\\approx 2/t+\\sqrt t\\,Z$ with $Z$ distributed according to the limiting density given by the formula, refining the known deterministic limit $tD_t\\to2$.","Because $d_\\theta^n(t)$ are the convex coefficients in the transition functions of Fleming-Viot and infinitely-many-neutral-alleles diffusion models, the local CLT provides a concrete small-time approximation to those transition probabilities.","The same integral representation is proposed as a basis for large-deviation and moderate-deviation estimates for $D_t$ at small times, going beyond the central-limit scale.","The explicit prefactor in the theorem gives a quantitative benchmark that numerical evaluations or other asymptotic methods should reproduce in the same regime."],"supporting_citations":[{"why":"supplies the explicit infinite-series formula for the finite-time probabilities that the paper rewrites as a contour integral.","marker":"[13]"},{"why":"contains the earlier integral form (2.16) to which Theorem 2.1 is described as essentially equivalent.","marker":"[8]"},{"why":"provides the Jacobi triple product identity used to factor the theta sum into the infinite product $\\varphi_t(z)$.","marker":"[1]"},{"why":"gives the small-time functional fluctuation theorem that the paper's local CLT complements and uses as a benchmark.","marker":"[11]"},{"why":"supplies the second-order small-time asymptotics for the same block-counting process, the comparison point for the new pointwise result.","marker":"[10]"}],"fun_headline_variants":["Small-time Kingman block count law via contour integrals","Contour integral yields Kingman block count law","Integral representation reveals Kingman small-time law","Steepest descent derives Kingman block count law"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument rests on the Taylor expansion of the exponent function along the deformed contour having exactly the quadratic coefficient stated in Lemma 3.3; a direct expansion of that same function appears to give a different coefficient, and if the direct expansion is right the final constant does not follow.","fun_headline_variants_meta":{"raw":{"variants":["Small-time Kingman block count law via contour integrals","Contour integral yields Kingman block count law","Integral representation reveals Kingman small-time law","Steepest descent derives Kingman block count law"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000773,"raw_usage":{"total_tokens":3392,"prompt_tokens":885,"completion_tokens":2507,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":501,"completion_tokens_details":{"reasoning_tokens":2446}},"tokens_in":501,"tokens_out":2507,"duration_ms":19445,"temperature":1.0,"reasoning_tokens":2446,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:17:05.182627+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Expand $\\frac1t\\psi_t((z(y)+\\pi)/2)$ at $z(y)=-\\pi+t^{1/4}e^{i\\pi/4}\\sqrt y$ using $\\psi_t(z)=(-2+(2n+\\theta)t/2)z^2+\\cdots$; compare the coefficient of $y^2$ with the paper's $-(2n+\\theta)/(12\\cdot24)$. The direct expansion gives $-(2n+\\theta)/192$, which on the scale $(2n+\\theta)t\\to4$ is $-1/48$ rather than $-1/72$. Evaluating the resulting Gaussian integral and comparing it with the original series for small $t$ and fixed $v$ would show whether the stated constant $3/(2\\sqrt{2\\pi})$ is correct.","supporting_citations":[{"cited_title":"Line-of-descent and genealogical processes, and their applications in population genetics models","cited_arxiv_id":null,"evidence_quote":"supplies the explicit infinite-series formula for the finite-time probabilities that the paper rewrites as a contour integral."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"contains the earlier integral form (2.16) to which Theorem 2.1 is described as essentially equivalent."},{"cited_title":"Diffusion limits at small times for Λ-coalescents with a Kingman com- ponent","cited_arxiv_id":null,"evidence_quote":"supplies the second-order small-time asymptotics for the same block-counting process, the comparison point for the new pointwise result."}],"review_version":1}