{"id":"44918756-cbed-47aa-aab1-9685c9b5ff5e","arxiv_id":"2412.12319","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Using Mellin analysis of a limit exchangeable partition, the authors obtain full asymptotic expansions, the variance constant, the moment generating function, a CLT, and large deviation rates for the leaf height of the critical beta-splitting tree.","lead":"This math paper derives sharp asymptotic formulas, including constants computed from the digamma function, for the height of a random leaf in the critical beta-splitting random tree. It uses a limit-tree representation and Mellin transforms to go beyond earlier leading-term results.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central Mellin expansion of E[D_n] rests on identity (2.4), imported without proof from companion preprint [3]; if that moment identity fails, the identification of Υ's Mellin transform and all residue expansions collapse.","rationale":"The reader's weakest assumption is exactly (2.4), and my reading agrees: the Mellin transform of Υ, the inversion estimate Lemma 6.1, and all subsequent residue expansions depend on it. I checked whether any internal inconsistency is more serious than the external dependency. The D_n/L_n swap at (1.4)-(1.5) is real and should be fixed, but the body consistently uses D_n for the continuous-time height, so the formulas are coherent once that is recognized. The paper's proof of Lemma 6.1 is careful conditional on (4.8), and the residue calculus is internally consistent; hence the remaining risk is that the imported moment identity from [3] is wrong or holds only in a weaker form. This is a genuine load-bearing gap rather than a proven error, and it justifies keeping the verdict conditional. An independent derivation via k-leaf separation rates is concrete and should settle the issue; it also gives the paper a self-contained proof of the identity it currently borrows. I therefore recommend no change to the reader's conditional verdict.","tokens_in":37203,"tokens_out":26829,"duration_ms":228280,"concrete_test":"Derive (2.4) without [3] as follows. For fixed k≥2, follow k distinct leaves in CTCS(n) until they first separate. In a current clade of size m, the separation rate is R_k(m)=h_{m-1}∑_{i=1}^{m-1} q(m,i)(1−[(i)_k+(m−i)_k]/(m)_k). Compute R_k(m) explicitly (or as m→∞) and show R_k(m)=ψ(k)−ψ(1)=H_{k-1}. Then the survival probability of k tagged leaves is e^{-H_{k-1}t}, so E[P_{t,1}^{k-1}]=e^{-t(ψ(k)-ψ(1))}; this gives (2.4) for all integer s≥0, and a direct generator computation with x^s extends it to ℜs>-1. If the computed R_k(m) differs from H_{k-1} for any k=2,3,4, the central Mellin route fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Every asymptotic statement in Theorem 1.1 and its corollaries passes through the measure Υ defined at (4.1) and the Mellin transform (4.8), which is exactly (2.4). The paper cites [3, Theorem 4.5] for (2.4) and does not reprove it; [3] is a not-yet-peer-reviewed companion preprint. The identity is not a small lemma: it fixes the poles of ~Υ(s)=1/(ψ(s)-ψ(1)) at the roots s_i and hence every term in the expansions (7.13)-(7.14), the variance calculation in Theorem 1.3 via (11.4), and the MGF/large-deviation analysis in Section 12. If (2.4) held only for integer s, or with a different rate function, the analytic continuation to ℜs>-1 and the residue sums at the negative roots would be unjustified. The finite-n identities in Proposition 4.1 also rely on the consistency and paintbox results imported from [3]. The exact formula (4.13) provides a partial check but is derived from (2.4), so it is not independent. This is a self-containedness gap rather than an identified error; the paper's own derivations are careful and internally consistent once (2.4) is granted.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the critical beta-splitting random tree and derives sharp asymptotic results for the height of a uniform random leaf. Using the exchangeable partition representation of the infinite limit tree CTCS(∞) from the companion paper [3], the authors define a measure Υ by (4.1) and identify its Mellin transform as 1/(ψ(s)−ψ(1)) via the moment identity (2.4). From this, they obtain exact representations of E[D_n], E[L_n], and E[Λ_n], and then use Mellin inversion and residue calculus to prove a full asymptotic expansion of E[D_n] (Theorem 1.1), the analogous expansion for E[L_n] (Theorem 1.2), the variance asymptotics for D_n (Theorem 1.3), the moment generating function asymptotics (Theorem 1.4), and the resulting CLT and large deviation results (Theorems 1.5 and 1.6). The paper also develops an alternative Parseval-formula method in Section 10 and applies it to higher moments in Section 11.","tokens_in":37462,"tokens_out":15907,"duration_ms":128883,"significance":"If the results hold, they are a substantial advance over the earlier recurrence-method results in [5] and [2]: they give full asymptotic expansions with explicitly computable coefficients, a surprising spectrum of powers n^{-|s_j|-k} determined by the nontrivial roots of ψ(s)=ψ(1), and a new Mellin-transform route through the limit tree. The analytic work is detailed and careful: Lemma 6.1 is proved in full, the Parseval formula of Lemma 10.1 is proved in Appendix A with an explicit Fejér-kernel argument, and exact finite-n formulas such as (4.13) provide partial checks. The main risk to correctness is the dependence on the unproven identity (2.4) imported from the companion preprint [3]; modulo that input, the internal derivations are consistent and the constants are cross-checked numerically.","major_comments":[{"comment":"The definitions of D_n and L_n are inconsistent with the rest of the paper. Section 1.2 states that D_n is the hop-height of DTCS(n) and L_n is the height of CTCS(n), but equations (1.4)–(1.5) assign D_n to the absorption time of the continuous-time chain and L_n to the absorption time of the discrete-time chain. Moreover, the proof of (4.4) treats D_n as a continuous-time height, Theorem 1.2 gives E[L_n] ~ (3/π^2) log^2 n, which is the discrete hop-height, and equations (5.5)–(5.6) in §5.2 only make sense if D_n is the continuous height and L_n is the discrete hop-height. As written, Theorem 1.1 appears to contradict the stated definition of D_n. This is more than a typo: the authors must relabel the definitions or the equations so that the discrete and continuous quantities are consistently named throughout the paper.","section":"§1.2 and §5.2"},{"comment":"The entire Mellin analysis rests on the identity E[P_{t,1}^s] = exp(-t(ψ(s+1)−ψ(1))) imported without proof from [3, Theorem 4.5]. Equation (4.8) is exactly this identity rewritten for the Mellin transform of Υ, and every subsequent residue expansion—in Theorems 7.3, 8.1, 9.1, 11.1, 12.4, and all of their corollaries—uses the poles of 1/(ψ(s)−ψ(1)). The exact formula (4.13) is derived from (4.8), so it is not an independent check. Since [3] is a not-yet-peer-reviewed companion preprint and no proof of (2.4) is included here, the paper should either provide a proof of (2.4) in an appendix or explicitly state that the main theorems are conditional on [3, Theorem 4.5].","section":"§2, Eq. (2.4); §4, Eq. (4.8)"}],"minor_comments":[{"comment":"The displayed line 'We thus interpret -zΓ(-ρ(-z)) = ψ′(1) for z = 0' contains a sign error in the argument of ρ; it should be -zΓ(-ρ(z)) = ψ′(1) at z = 0.","section":"§12.1, Eq. (12.10)"},{"comment":"In the acknowledgments, 'Bénédicte Haas for his careful explanation' should use 'her' instead of 'his'.","section":"§13"},{"comment":"The title 'Asymptotics of Υ k' contains the typo 'analoguous'; it should be 'analogous'.","section":"§11.2"},{"comment":"The statement '(1.6) for some coefficients c_i and c_{j,k}' would benefit from an explicit note that the double sum is interpreted by grouping terms in decreasing order of n-exponent, as already explained in §3.1; adding a cross-reference would prevent confusion.","section":"§1.3, Theorem 1.1"},{"comment":"The exact formula (4.13) is interesting, but the sentence 'we will not use it' could also mention that it depends on (2.4) in the same way as the main analysis, so it does not offer independent verification; this would address a possible reader misconception.","section":"§4, Remark 4.3"}],"recommendation":"major_revision","confidential_remarks":"The main reason for major revision is the unproven reliance on identity (2.4) from the companion preprint [3]. If the editorial policy for this series permits results to rest on companion papers that are simultaneously under review, the paper could be substantially acceptable after the D_n/L_n relabeling and the minor corrections; otherwise, the authors should include a proof of (2.4) or clearly state the conditional nature of the theorems. The inconsistency in the definitions of D_n and L_n in §1.2 appears to be a slip, since equations (1.4)–(1.5) and all later sections consistently use D_n for the continuous-time height and L_n for the discrete hop-height, but it must be fixed before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read. The main theorems are solid, conditional on one imported identity, and the paper is worth serious engagement. Aldous and Janson push the Mellin program to a sharp level: explicit constant c0, the full expansion with the mixed integer/non-integer power spectrum, the variance constant, MGF asymptotics, and large deviations. The technical core — Lemma 6.1's inversion estimate, the residue expansions in Section 10.1, and the careful handling of conditionally convergent line integrals in Lemma 10.1 — is genuinely good. The trick of differentiating the Mellin transform to gain integrability is useful beyond this paper.\n\nThe soft spots are real but manageable. First, everything rests on (2.4), the moment identity imported from companion preprint [3]. The paper does not reprove it, and if it failed, the Mellin transform of Upsilon and all pole expansions collapse. That is a self-containedness gap, not an identified error — the identity is plausible, and the paper's own calculations are internally consistent once you grant it. A referee should ask the authors to either prove (2.4) in an appendix or make the dependency explicit and wait for [3] to clear review. Second, the D_n/L_n notation is swapped in the displayed definitions (1.4)-(1.5): the text defines D_n as hop-height, but (1.4) gives the continuous-time absorption time, and (1.5) gives the discrete-time one. Easy fix, but currently confusing. Third, the claim that this is the first application of Parseval's formula for Mellin transforms to combinatorial probability is overstated — the Mellin toolbox has been used in combinatorial settings for decades. That is a literature comment, not a mathematical flaw.\n\nMy overall assessment matches the reader's: the central argument holds up conditionally, the expansions are new relative to prior work, and the citation pattern is appropriate. Self-citation to [3] is legitimate because the companion theorem is the natural source, not a reflexive citation. For a random tree theorist, this is a strong tool; for someone wanting a self-contained proof, you'll want to wait for [3] or ask the authors to close the gap. I'd send it to a serious referee, with the reviewer focusing on the dependency on (2.4) and the notation, not on hunting for a hidden error — the computations are careful and check out.","headline":"Mellin expansion of E[D_n] is real and well-proved conditional on the moment identity imported from the companion paper; send it to a referee, with requests to fix the notation and close the self-containedness gap.","tokens_in":37989,"tokens_out":2703,"would_cite":true,"duration_ms":25330,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60C05","05C05","44A15","60G09"],"pacs":[],"model":"deepseek-v4-flash","headline":"The expected leaf height in the critical beta-splitting random tree has a complete asymptotic expansion whose exponents are set by the negative roots of the digamma equation, and the same Mellin machinery yields the variance, CLT, and…","keywords":["critical beta-splitting tree","leaf height","Mellin transform","exchangeable random partition","digamma function","asymptotic expansion","central limit theorem","large deviations"],"falsifier":"Run a high-precision simulation of the discrete model for $n$ near $10^6$, estimate $E[D_n]$ to about $10^{-6}$, and compare $E[D_n]-(6/\\pi^2)\\log n - c_0$ with the predicted first corrections $-3/(\\pi^2 n)-0.0943\\,n^{-1.567}-1/(2\\pi^2 n^2)$; if the remainder instead matches a purely integer-power series, the negative-root spectrum in Theorem 1.1 is wrong.","tokens_in":36988,"feed_emoji":"🌳","tokens_out":9783,"duration_ms":78943,"temperature":0.7,"pith_summary":"This paper studies the critical $\\beta$-splitting model, the random binary tree on $n$ leaves in which a clade of $m$ leaves is split with probability proportional to $1/(i(m-i))$. The authors establish a complete asymptotic expansion for $D_n$, the hop-height of a uniform random leaf: $E[D_n] \\sim (6/\\pi^2)\\log n + c_0 + c_1 n^{-1} + \\sum_{j,k} c_{j,k} n^{-|s_j|-k}$, where $s_j$ are the negative roots of $\\psi(s)=\\psi(1)$ and $c_0=\\zeta(3)/\\zeta(2)^2+\\gamma/\\zeta(2)$. This goes beyond earlier results that supplied only the first few terms and had conjectured that only integer powers of $1/n$ would appear. The same Mellin machinery yields the variance, a central limit theorem, and large-deviation exponents for $D_n$, plus asymptotic expansions for the continuous-time height $L_n$ and the total tree length $\\Lambda_n$.","feed_headline":"Leaf height in random splitting tree gains full expansion","feed_subtitle":"Mellin analysis yields every term, including non-integer powers from digamma roots.","key_machinery":"The central object is the infinite measure $\\Upsilon$ on $(0,1)$ defined by $\\Upsilon=\\int_0^\\infty \\mathcal{L}(P_{t,1})\\,dt$, where $P_{t,1}$ is the asymptotic size-biased proportion of the clade of leaf 1 at time $t$ in the infinite limit tree; in practical terms this means $\\int_0^1 f\\,d\\Upsilon = \\int_0^\\infty E[f(P_{t,1})]\\,dt$ for nonnegative $f$. Its Mellin transform is the reciprocal of $\\psi(s)-\\psi(1)$, and the paper's inversion estimate shows that the density of $\\Upsilon$ is $\\upsilon(x)=\\frac{6}{\\pi^2 x}+\\sum_i \\frac{1}{\\psi'(s_i)}x^{|s_i|}+r_N(x)$, with the negative roots $s_i$ of $\\psi(s)=\\psi(1)$ controlling all non-integer powers. This density estimate, together with the integral identity for $E[D_n]$, is what converts a limit-tree representation into sharp finite-$n$ asymptotics; a parallel Parseval-formula argument reaches the same expansions by shifting complex contours and collecting residues at the corresponding poles.","core_discovery":"On the paper's own terms, the central discovery is that the expected leaf height is encoded in one infinite measure $\\Upsilon$ on $(0,1)$, built from the occupation time of the size-biased clade proportion in the infinite limit tree. The exact identity $E[D_n]=\\int_0^1 (1-(1-x)^{n-1})\\,d\\Upsilon(x)$ holds for every $n$, and $\\Upsilon$ has Mellin transform $\\int_0^1 x^{s-1}\\,d\\Upsilon(x)=1/(\\psi(s)-\\psi(1))$ for $\\Re s>1$. Because the denominator vanishes at $s=1$ and at the negative roots $s_j$ of $\\psi(s)=\\psi(1)$, shifting the Mellin inversion contour turns this representation into an asymptotic expansion labeled by those roots: $E[D_n] \\sim (6/\\pi^2)\\log n + c_0 + c_1 n^{-1} + \\sum_{j\\ge1}\\sum_{k\\ge1} c_{j,k}n^{-|s_j|-k}$, with $c_0=\\zeta(3)/\\zeta(2)^2+\\gamma/\\zeta(2)$ and $c_1=-3/\\pi^2$. The same mechanism extends to higher moments and to the moment generating function, giving the variance, the CLT with mean $6/\\pi^2$ and variance $2\\zeta(3)/\\zeta(2)^3$, and the large-deviation rate function.","pith_inferences":["This suggests that any fragmentation model whose infinite limit has an explicit size-biased moment function should admit the same Mellin treatment, with the analogue of $1/(\\psi(s)-\\psi(1))$ determining the spectrum of powers in its height expansion.","The paper leaves the variance of the continuous-time hop-height $L_n$ open, explicitly because no representation of its higher moments analogous to Proposition 4.1 is available; finding such a representation is the natural next step, after which the present machinery would apply directly.","A concrete, untested prediction of the expansion is that the non-integer correction $-0.0943\\,n^{-1.567}$ will be visible in high-precision simulations of $E[D_n]$ for $n$ around $10^5$ to $10^6$; observing only integer-power corrections would indicate that the inversion estimate misses essential structure."],"forward_implications":["The expected leaf height has a computable expansion to any order: after the leading $(6/\\pi^2)\\log n$, the constant is $0.795155660439$ and the $n^{-1}$ coefficient is $-3/\\pi^2$, with later coefficients determined by the roots $s_j$.","The variance of $D_n$ is $(2\\zeta(3)/\\zeta(2)^3)\\log n$ plus a known constant, with error $O((\\log n)/n)$, so the fluctuations grow only logarithmically.","The centered leaf height converges to a normal distribution with mean $6/\\pi^2$ and variance $2\\zeta(3)/\\zeta(2)^3$ after normalization by $\\sqrt{\\log n}$.","Large-deviation probabilities for $D_n/(\\log n)$ decay as $n^{-\\Lambda^*(x)+o(1)}$ with an explicit rate function, with threshold $x_0=6/\\pi^2$ at the center.","The same Mellin method gives $E[\\Lambda_n]=(6/\\pi^2)n+O(n^{-|s_1|})$ for the continuous tree's total length and the earlier occupation-probability asymptotics for the harmonic descent chain."],"supporting_citations":[{"why":"Supplies the exchangeable partition representation of the infinite limit tree and the key moment identity (2.4) on which every Mellin transform in the paper rests.","marker":"[3]"},{"why":"Provides the earlier height results and the recurrence method, including the first terms and the h-ansatz question that this paper sharpens into a full expansion.","marker":"[5]"},{"why":"Develops the exchangeable random partition framework used to model the infinite limit tree and its clade proportions.","marker":"[12]"},{"why":"Gives the paintbox construction and existence of asymptotic clade proportions used in Section 2.","marker":"[6]"},{"why":"Supplies the digamma function identities, root locations, and Stirling expansions that underlie the Mellin inversion estimates.","marker":"[19]"},{"why":"Supplies the standard Mellin inversion and residue techniques adapted in the paper's asymptotic arguments.","marker":"[9]"},{"why":"Provides the Parseval/Plancherel formula in the form needed for the alternative contour-shifting proofs in Section 10.","marker":"[22]"}],"fun_headline_variants":["Mellin transform yields exact leaf-height expansion","Leaf height expansion from Mellin inversion: full series","Exact leaf height via Mellin yields all terms","Digamma roots give leaf-height asymptotics to all orders","Mellin analysis: exact leaf height, all moments, CLT"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the moment identity $E[P_{t,1}^s]=e^{-t(\\psi(s+1)-\\psi(1))}$ for the infinite limit tree, imported from the companion preprint [3] rather than proved here; if that identity fails or holds only approximately, the Mellin transform of $\\Upsilon$, the inversion estimate, and every asymptotic expansion built on them would not follow.","fun_headline_variants_meta":{"raw":{"variants":["Mellin transform yields exact leaf-height expansion","Leaf height expansion from Mellin inversion: full series","Exact leaf height via Mellin yields all terms","Digamma roots give leaf-height asymptotics to all orders","Mellin analysis: exact leaf height, all moments, CLT"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000847,"raw_usage":{"total_tokens":3735,"prompt_tokens":1045,"completion_tokens":2690,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":661,"completion_tokens_details":{"reasoning_tokens":2610}},"tokens_in":661,"tokens_out":2690,"duration_ms":17718,"temperature":1.0,"reasoning_tokens":2610,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T14:14:26.714809+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a high-precision simulation of the discrete model for $n$ near $10^6$, estimate $E[D_n]$ to about $10^{-6}$, and compare $E[D_n]-(6/\\pi^2)\\log n - c_0$ with the predicted first corrections $-3/(\\pi^2 n)-0.0943\\,n^{-1.567}-1/(2\\pi^2 n^2)$; if the remainder instead matches a purely integer-power series, the negative-root spectrum in Theorem 1.1 is wrong.","supporting_citations":[{"cited_title":"The critical beta-split ting random tree III: The exchangeable partition representation and the fringe tree","cited_arxiv_id":null,"evidence_quote":"Supplies the exchangeable partition representation of the infinite limit tree and the key moment identity (2.4) on which every Mellin transform in the paper rests."},{"cited_title":"The Critical Beta-splitting Random Tree: Heights and Related Results","cited_arxiv_id":"2302.05066","evidence_quote":"Provides the earlier height results and the recurrence method, including the first terms and the h-ansatz question that this paper sharpens into a full expansion."},{"cited_title":"Continuum tree asymptotics of discrete fragmentations and applications t o phylogenetic models","cited_arxiv_id":null,"evidence_quote":"Develops the exchangeable random partition framework used to model the infinite limit tree and its clade proportions."},{"cited_title":"Random Fragmentation and Coagulation Processes , volume 102 of Cambridge Studies in Advanced Mathematics","cited_arxiv_id":null,"evidence_quote":"Gives the paintbox construction and existence of asymptotic clade proportions used in Section 2."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the digamma function identities, root locations, and Stirling expansions that underlie the Mellin inversion estimates."},{"cited_title":"Mellin transforms and asymptotics: harmonic sums","cited_arxiv_id":null,"evidence_quote":"Supplies the standard Mellin inversion and residue techniques adapted in the paper's asymptotic arguments."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the Parseval/Plancherel formula in the form needed for the alternative contour-shifting proofs in Section 10."}],"review_version":1}