{"id":"41f94a90-5ba6-4bcb-abe5-7931ed902789","arxiv_id":"2506.08910","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper proves central limit theorems for fluctuations of repeatedly differentiated random polynomials and a heavy-tailed analogue with random Appell limits, but the polynomial CLT has a factor error.","lead":"Repeatedly differentiating random polynomials with independent roots converges, after rescaling, to Hermite polynomials. This paper uses finite free cumulants to prove fluctuation theorems around that limit and to handle heavy-tailed roots, though one of the stated fluctuation formulas is wrong.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2.3 as stated is internally contradicted by its own proof: the sum in (3.14) equals -(ell(ell-1)/2) sqrt(m4-1) Z He_{ell-2}(x), not (ell/2) sqrt(m4-1) Z He_{ell-2}(x).","rationale":"I re-derived the proof of Theorem 2.3 from (1.8), Lemma 1.2, and Theorem 3.3. The reader's strongest_claim is precisely correct: the final simplification in (3.14) is wrong. The penultimate sum equals -(ell(ell-1)/2) sqrt(m4-1) Z He_{ell-2}(x), not (ell/2) sqrt(m4-1) Z He_{ell-2}(x). This is load-bearing because Theorem 2.3 is one of the two advertised main contributions, and it is an internal inconsistency rather than a disagreement with an external convention or a subtle domain-of-attraction issue. The reader's weakest_assumption about finite moments is a secondary concern: the amplitude error already invalidates the theorem under the paper's own hypotheses. The cumulant machinery and the proof of Theorem 2.6 appear plausible, and a corrected version of Theorem 2.3 with the factor -(ell(ell-1)/2) would likely follow from the same argument. Despite the apparent fixability, the main CLT as written is false, so the rejection verdict is supported.","tokens_in":11938,"tokens_out":16180,"duration_ms":170723,"concrete_test":"Analytic check: compare the coefficient of x^{ell-2} in the final sum of (3.14) with that in (ell/2) He_{ell-2}(x). For r=1 the sum gives -ell(ell-1)/2, while (ell/2) He_{ell-2} gives ell/2; these agree only for ell=0. For a direct numerical check, simulate ell=2, N=10^4 iid standard Gaussian roots, compute the x^0 coefficient of sqrt(N)[tilde p_{2,N}(x) - He_2(x)], and regress it on the sample fourth-moment fluctuation; the slope should be negative, matching -(m4-1)^{1/2} rather than +(m4-1)^{1/2}.","verdict_should_be":"REJECT","load_bearing_attack":"The central CLT claim is internally contradicted by its proof. In (3.14), after the Delta method the coefficient of (kappa^ell_2(tilde p)^r - ell^r) is (ell)_{2r}/(ell^{2r}(2r)!) (2r-1)!! (-1)^r ell^r. Since sqrt(N)(kappa^ell_2 - ell) => ell sqrt(m4-1) Z, the r-th term of sqrt(N)[tilde p_{ell,N} - He_ell] tends to (-1)^r r (ell)_{2r} (2r-1)!! / (2r)! sqrt(m4-1) Z x^{ell-2r} = (-1)^r ell! / ((ell-2r)! (r-1)! 2^r) sqrt(m4-1) Z x^{ell-2r}. Summing r = 1..floor(ell/2) gives -(ell(ell-1)/2) sqrt(m4-1) Z He_{ell-2}(x), using the definition of He_{ell-2} in (1.6). For ell=2 the correct coefficient of the constant term is -sqrt(m4-1)Z, whereas Theorem 2.3 predicts +sqrt(m4-1)Z. This is not a convention or domain-of-attraction issue: it fails under the paper's own finite-moment hypotheses. The theorem as stated is false, although the corrected statement follows from the same proof.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper uses finite free cumulants to analyse the roots of random polynomials under repeated differentiation. Its first contribution is a central limit theorem for the fluctuations of the rescaled differentiated polynomial around the limiting Hermite polynomial, together with a CLT for the fluctuations of the ordered roots. Its second contribution is a limit theorem for the differentiated polynomials when the iid roots are not in the Gaussian domain of attraction, with the limit described through a random Appell sequence and an infinitely divisible distribution. The proofs are based on finite free cumulants, moment–cumulant formulas, the Delta method, and convergence of point processes.","tokens_in":12246,"tokens_out":14740,"duration_ms":169776,"significance":"The finite free cumulant framework is well chosen for this problem and the paper makes a useful connection between repeated differentiation of random polynomials and finite free probability. The second result, Theorem 2.6, is a substantial extension of the Hoskins–Steinerberger theorem to infinite-variance and Poisson-type root distributions, and the Appell-sequence description is natural. The paper also gives compact proofs and explicitly identifies the fourth-moment dependence of the fluctuations. However, the central polynomial CLT, Theorem 2.3, is stated with an incorrect coefficient, and the proof contains a scaling-factor error that is load-bearing for the same theorem. Because the main advertised result is false as stated, the manuscript cannot be accepted in its current form, although the corrected version appears to follow from the authors' own proof.","major_comments":[{"comment":"The statement of Theorem 2.3 is contradicted by the proof. In Eq. (3.14), the Delta method gives the r-th term of the limit as (-1)^r * ell! / ((ell-2r)! (r-1)! 2^r) * sqrt(m4(mu)-1) Z x^{ell-2r}, and summing r=1,...,floor(ell/2) yields -ell(ell-1)/2 * sqrt(m4(mu)-1) Z He_{ell-2}(x), not (ell/2) sqrt(m4(mu)-1) Z He_{ell-2}(x). For ell=2, this changes the sign of the constant term. Thus Theorem 2.3 is false as stated; the coefficient in (2.6) and in the final line of (3.14) must be corrected to -ell(ell-1)/2.","section":"Theorem 2.3 and Eq. (3.14)"},{"comment":"The definition of the differentiation operator is normalized incorrectly. For a monic degree-N polynomial p_N, the (N-k)-th derivative has leading coefficient N!/k!, so the definition in (1.2) gives a leading coefficient of (N-k)!/k! for the resulting degree-k polynomial. After applying D_{\\sqrt N} in (1.4), the leading coefficient is N^{k/2}(N-k)!/k!, which diverges as N grows. This contradicts Proposition 1.1 and every limit in the paper. The intended normalization appears to be k!/N! (or an equivalent convention that makes the derivative monic after the standard scaling); please correct and re-verify the subsequent formulas under that convention.","section":"Eq. (1.2)"},{"comment":"The scaling formula in the proof of Theorem 3.3 has the wrong exponent. Eq. (3.12) states kappa^ell_j(\\tilde p_{ell,N}) = ell^{j-1} N^{j/2-1} kappa^N_j(p_N). Since D_{\\sqrt N} scales roots by \\sqrt N and Lemma 1.2 contributes a factor (ell/N)^{j-1}, the correct exponent is N^{1-j/2}. With the printed exponent, for j >= 3 the quantity \\sqrt{N} kappa^ell_j would diverge for generic mu with nonzero kappa_j(mu), contradicting the conclusion of Theorem 3.3 that only the j=2 entry has a non-zero limiting variance. Please correct the exponent and adjust the surrounding argument.","section":"Eq. (3.12)"}],"minor_comments":[{"comment":"There is a typo in the sentence introducing Theorem 2.6: 'See Theorem, 2.6 below' should read 'See Theorem 2.6 below'.","section":"Introduction"},{"comment":"The text contains a double comma in 'standard exponential random variables,, ε1'; this should be cleaned up.","section":"Example 2.7(2)"},{"comment":"The sentence 'the right-hand side of (2.1) vanishes at the roots of He_{ell-2}' is terse; (2.1) is the covariance matrix, and the intended statement is that the polynomial fluctuations are of smaller order at those points. Please clarify the wording.","section":"Remark 2.4"},{"comment":"Several cited works are preprints ([1], [6], [7]); if any have appeared or been updated, the reference entries should reflect that.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The error in Theorem 2.3 is serious but appears to be a coefficient/sign mistake, and the corrected statement follows from the authors' own computation. The normalization issues in Eq. (1.2) and Eq. (3.12) are also fixable. If the authors correct these points and re-verify the root CLT after the normalization change, the paper would be a solid contribution. I recommend major revision rather than outright rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper has a good idea and a serious misstatement. The finite free cumulant approach to repeated differentiation CLTs is new and, for the most part, works. Theorem 2.1 on root fluctuations looks right, and Theorem 2.6 extending to infinitely divisible root limits is a genuine extension with a credible proof outline. The paper is clearly written and the method is compact.\n\nThe problem is Theorem 2.3. As stated, the limit is sqrt(m4-1) Z (ell/2) He_{ell-2}(x). But the proof in (3.14) actually yields sqrt(m4-1) Z times S(x), where S(x) = sum_{r=1}^{floor(ell/2)} (-1)^r ell! / ((ell-2r)! (r-1)! 2^r) x^{ell-2r}. Direct summation using the definition of He_{ell-2} in (1.6) gives S = -(ell(ell-1)/2) He_{ell-2}(x). So the stated sign is wrong for all ell, and the amplitude is wrong for ell>2. For ell=2 the theorem predicts +sqrt(m4-1)Z but the proof gives -sqrt(m4-1)Z. This is not a convention issue; it fails under the paper's own hypotheses. The good news is the corrected statement follows from the same proof, so the error is localized to the statement of one theorem.\n\nThere is also a normalization typo in (1.2): with the factor (N-k)!/N! the leading coefficient of the k-th derivative object diverges; it should be k!/N! to make the leading coefficient 1. The later dilation in (1.4) does not fix that.\n\nThe fourth-moment dependence in the CLT is natural and the moment assumptions are stated correctly. The heavy-tailed extension in Theorem 2.6 is the most interesting part and does not rely on the faulty theorem.\n\nVerdict: this paper deserves a serious referee. The method is sound and the errors are fixable, but a referee should require the corrected version of Theorem 2.3 and the normalization fix before publication. I would not cite it in its current form, but I would read the revision.","headline":"A promising method with a misstated central CLT: the proof gives -ell(ell-1)/2 He_{ell-2}, not +ell/2 He_{ell-2}, so Theorem 2.3 is false as written but easily corrected.","tokens_in":12766,"tokens_out":12079,"would_cite":false,"duration_ms":108115,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60B20","60F05","46L54","30C15"],"pacs":[],"model":"deepseek-v4-flash","headline":"Repeatedly differentiating random polynomials gives Gaussian fluctuations around Hermite limits.","keywords":["random polynomials","critical points","finite free probability","finite free cumulants","Hermite polynomials","Appell sequences","infinitely divisible distributions","central limit theorem"],"falsifier":"Evaluate the coefficient of $x^{\\ell-2}$ on both sides of Theorem 2.3 for $\\ell=4$ using the paper's own expansion in equation (3.14): the stated limit has coefficient $(\\ell/2)\\sqrt{m_4(\\mu)-1}\\,Z = 2\\sqrt{m_4(\\mu)-1}\\,Z$, while equation (3.14) gives $-\\ell(\\ell-1)/2\\,\\sqrt{m_4(\\mu)-1}\\,Z = -6\\sqrt{m_4(\\mu)-1}\\,Z$; if this calculation is correct, the proof does not establish the stated amplitude for $\\ell>2$.","tokens_in":11728,"feed_emoji":"📊","tokens_out":9300,"duration_ms":103904,"temperature":0.7,"pith_summary":"It is known that repeatedly differentiating a random polynomial with independent mean-zero, variance-one roots stabilizes, after rescaling, to the Hermite polynomial. This paper establishes central limit theorems for the fluctuations around that deterministic limit: the rescaled differences for the polynomial and for its ordered roots converge to Gaussian distributions whose covariance depends on the root law only through its fourth moment. It also removes the assumption that the roots have a finite second moment, showing that when the roots belong to the domain of attraction of a general infinitely divisible law the Hermite limit is replaced by a random Appell sequence built from the Lévy triple. The proofs use finite free cumulants to track how root fluctuations propagate through differentiation.","feed_headline":"Gaussian noise governs derivatives of random polynomials","feed_subtitle":"Finite free cumulants show the fluctuation size is set by the fourth moment of the roots.","key_machinery":"Finite free cumulants $\\kappa^N_j(p)$, the coefficients of the finite $R$-transform $R_p(s) = -P'(Ns)/P(Ns)$ for $P(d/dx)x^N = p(x)$, linearize the finite free additive convolution $\\boxplus_N$. The load-bearing identity is $\\kappa^k_j(\\partial_{k|N}p) = (k/N)^{j-1}\\kappa^N_j(p)$, which lets the authors read off how cumulants of the original roots are scaled by differentiation; the moment-cumulant formula of Lemma 1.3 then converts the classical CLT for empirical moments into a CLT for finite free cumulants, and the $\\Delta$ method transfers that CLT from moments to polynomial coefficients and roots.","core_discovery":"On the paper's own terms, the central discovery is Theorem 2.3 together with Theorem 2.1: for fixed $\\ell$, $\\sqrt{N}[\\tilde{p}_{\\ell,N}(x) - \\mathrm{He}_\\ell(x)]$ converges in distribution to $\\sqrt{m_4(\\mu)-1}\\,Z\\,(\\ell/2)\\,\\mathrm{He}_{\\ell-2}(x)$, and the ordered roots satisfy $\\sqrt{N}[z(\\tilde{p}_{\\ell,N}) - z(\\mathrm{He}_\\ell)] \\Rightarrow \\mathcal{N}(0,\\Sigma^{(\\ell),z})$ with a covariance matrix built from the fourth moment of $\\mu$ and the Hermite moments. The second discovery, Theorem 2.6, replaces the Gaussian domain of attraction by an arbitrary Lévy triple $(c,\\sigma^2,\\nu)$: the limiting polynomial is $f(d/dx)x^\\ell$, where $f$ is a random entire function $e^{-Yz - \\sigma^2 z^2/2}\\prod_{j\\neq 0}(1-\\alpha_j z)e^{\\alpha_j z}$, expressed as a finite free convolution of a random monomial, a Hermite term, and a Poisson-point-process factor.","pith_inferences":["The factorization in equation (2.9) suggests a finite-free analogue of a Gaussian perturbation: to first order, differentiation acts on the limiting noise by finite free convolution with a quadratic polynomial, which could make the CLT coefficient a combinatorial statistic of pairings rather than a separate calculation.","Because Theorem 2.6 depends only on the Lévy triple, the same limiting Appell sequence should arise from any real-rooted polynomial ensemble whose roots converge to the same infinitely divisible law; the paper does not fully spell this out, but its method appears ready for it."],"forward_implications":["The fluctuation variance of the critical-point process is universal across root distributions sharing mean 0, variance 1, and fourth moment $m_4(\\mu)$.","For a fixed $\\ell$, the polynomial fluctuation limit is a Gaussian multiple of $\\mathrm{He}_{\\ell-2}$, so the zeros of the fluctuation field asymptotically sit at the roots of $\\mathrm{He}_{\\ell-2}$.","Removing the finite-variance assumption does not destroy the limit: the Hermite polynomial is replaced by a random Appell sequence determined by the Lévy triple of the root distribution.","In the Bernoulli/Poisson case the limiting polynomial is a Laguerre polynomial of random parameter, so discrete root models produce explicitly identifiable limits.","The same cumulant argument applies to other real-rooted random polynomial ensembles with understood root statistics, such as characteristic polynomials of random matrices."],"supporting_citations":[{"why":"Supplies the base deterministic limit (repeated differentiation to a Hermite polynomial) that the paper extends to fluctuations.","marker":"[15]"},{"why":"Introduces finite free cumulants and the moment-cumulant formulas used throughout the proofs.","marker":"[3]"},{"why":"Provides the finite free convolution framework used to identify the limiting Appell polynomials.","marker":"[21]"},{"why":"Gives the scaling identity for cumulants under differentiation used as Lemma 1.2.","marker":"[1]"},{"why":"Supplies the Poisson point process representation of infinitely divisible laws used to build the limiting random entire function.","marker":"[8]"},{"why":"Provides the convergence criteria for sums and point measures used in Lemma 4.1.","marker":"[17]"},{"why":"Gives the classical-free infinite divisibility bijection that frames the interpretation of Theorem 2.6.","marker":"[5]"}],"fun_headline_variants":["Fourth moment sets fluctuations in random polynomial derivatives","Free cumulants reveal root fluctuations in derivative limits","Finite free cumulants pin down random polynomial derivative fluctuations","Fourth moment sets scale of derivative limit fluctuations"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central limit theorems require that the empirical root moments of the original random polynomial fluctuate at scale $N^{-1/2}$ toward a Gaussian, which only holds when the root distribution has enough finite moments; if the fourth moment is infinite, the variance formulas in Theorems 2.1 and 2.3 fail.","fun_headline_variants_meta":{"raw":{"variants":["Fourth moment sets fluctuations in random polynomial derivatives","Free cumulants reveal root fluctuations in derivative limits","Finite free cumulants pin down random polynomial derivative fluctuations","Fourth moment sets scale of derivative limit fluctuations"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000764,"raw_usage":{"total_tokens":3384,"prompt_tokens":938,"completion_tokens":2446,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":554,"completion_tokens_details":{"reasoning_tokens":2387}},"tokens_in":554,"tokens_out":2446,"duration_ms":21504,"temperature":1.0,"reasoning_tokens":2387,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T05:03:51.387054+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Evaluate the coefficient of $x^{\\ell-2}$ on both sides of Theorem 2.3 for $\\ell=4$ using the paper's own expansion in equation (3.14): the stated limit has coefficient $(\\ell/2)\\sqrt{m_4(\\mu)-1}\\,Z = 2\\sqrt{m_4(\\mu)-1}\\,Z$, while equation (3.14) gives $-\\ell(\\ell-1)/2\\,\\sqrt{m_4(\\mu)-1}\\,Z = -6\\sqrt{m_4(\\mu)-1}\\,Z$; if this calculation is correct, the proof does not establish the stated amplitude for $\\ell>2$.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Poisson point process representation of infinitely divisible laws used to build the limiting random entire function."},{"cited_title":"Bercovici and V","cited_arxiv_id":null,"evidence_quote":"Gives the classical-free infinite divisibility bijection that frames the interpretation of Theorem 2.6."}],"review_version":1}