REVIEW 3 major objections 3 minor 1 cited by
Limit law for root separation in random polynomials
T0 review · 3 major / 3 minor · reviewed 2026-08-16 · deepseek-v4-flash
Pith's one-line read At scale $n^{-5/4}$, the pairwise gaps among roots of a random polynomial converge to a Poisson process with an explicit cubic intensity.
desk verdict A substantial paper that likely resolves the n^{-5/4} root-separation problem for sub-Gaussian Kac polynomials, but the printed event A_z(U) has a sign error that makes the proof as written inconsistent until fixed. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The argument is carried by a net reduction in the bulk annulus $\Omega_K=\mathbb{H}\cap A(1-K/n,1+K/n)$, where almost all roots lie. At scale $\delta=n^{-5/4-\beta}$, each net point $z$ is interrogated through the random triple $(f_n(z),f_n'(z),f_n''(z))$: the linear root prediction $z - f_n(z)/f_n'(z)$ must land in the polar cell $R_z^\circ$, and the quadratic second-root prediction $2|f_n'(z)|/|f_n''(z)|$ must land in $n^{-5/4}U$. A close pair of roots forces both events, and conversely the quadratic approximation recovers the pair from them, giving the two-sided counting identity that reduces the Poisson limit to a method-of-moments computation over the net. To control small-ball probabilities for arbitrary coefficient laws, net points are split into 'smooth' points (angles $\theta$ with no $p\theta/\pi$ close to an integer for small $p$, so that the coefficient vector is genuinely high-dimensional) where a Konyagin-Schlag / Cook-Nguyen style local Gaussian comparison (Theorem 3.12) shows asymptotic agreement with Gaussian coefficients, and 'rough' points handled by cruder arithmetic small-ball bounds. The limiting intensity emerges from an explicit Gaussian computation of the event $A_z(U)$: the function $F(x)$ of (78) integrates to give $c_*(K)=\tfrac14\int_{-K}^K F(x)\,dx$ and $c_*=\lim_{K\to\infty}c_*(K)$. Pairs of roots well inside the unit disk are killed by the almost-sure no-double-zero theorem for random Taylor series, so the entire limiting process is driven by the $1/n$-neighborhood of the unit circle.
What would settle it
Run the theorem's own simulation: for Rademacher or standard-Gaussian coefficients at degree $n=10^4$, sample the unnormalized pair-distance process and check that the expected number of pairs with $n^{5/4}|\alpha_j-\alpha_j'|\le s$ grows like $c_*s^4/4$ and that $P[n^{5/4}m_n\ge s]$ converges to $\exp(-c_*s^4/4)$; any statistically clear deviation from the quartic power law refutes the $t^3$ intensity and the minimal-gap corollary. A second test: replace real coefficients with i.i.d. complex Gaussians and see whether the same Poisson limit survives, which would settle whether the real-coefficient conjugation step in Section 3 is essential.
Extended reading notes
Core claim
The central claim (Theorem 1.1) is that for $f_n(z)=\sum_{k=0}^n \xi_k z^k$ with i.i.d. mean-zero sub-Gaussian coefficients and $P[\xi_0=0]=0$, the point process $\{n^{5/4}|\alpha_j-\alpha_j'| : 1\le j<j'\le n\}$ converges vaguely to a non-homogeneous Poisson point process on $\mathbb{R}_{\ge 0}$ with intensity $c_* t^3\,dt$. The cubic shape of the intensity is universal: it depends on the coefficient law only through the positive constant $c_* = \tfrac{1}{4}\int_{-\infty}^{\infty}F(x)\,dx$, where $F$ is given by an explicit Gaussian computation. An immediate corollary is that the minimal separation $m_n$ between roots obeys $\lim_{n\to\infty} P[n^{5/4}m_n\ge s] = \exp(-c_*s^4/4)$, so double roots become asymptotically improbable for every coefficient law covered by the theorem. En route, Theorem 1.3 proves that a random Taylor series with i.i.d. coefficients satisfying $E\log(1+|a_0|)<\infty$ satisfies $P[\exists\alpha\in\mathbb{D}: F(\alpha)=F'(\alpha)=0] = (P[a_0=0])^2$, i.e. a double zero can occur only at the origin and only when the first two coefficients vanish.
Load-bearing premise
The coefficients must be real-valued for the proof as written: the upper-half-plane reduction and the identification of the limiting distance set both use the conjugation identity $f_n(z)=\overline{f_n(\bar z)}$, yet Theorem 1.1 as stated never says that the coefficients are real.
Editorial extensions
If this is right
- The minimal separation $m_n$ between roots satisfies $P[n^{5/4}m_n\ge s]\to\exp(-c_*s^4/4)$: a nontrivial, universal limit law for how close the two closest roots can be.
- Double roots become asymptotically improbable for every mean-zero sub-Gaussian coefficient law with no atom at zero, which the paper notes is the first such statement for Kac polynomials beyond Gaussian or integer-valued coefficient distributions.
- The entire pairwise-distance spectrum is universal: only the constant $c_*$ depends on the coefficient law, while the $t^3$ intensity shape is fixed by the geometry of the bulk annulus.
- Roots that remain a fixed distance inside the unit disk are uniformly separated with high probability, so the limiting statistics are governed entirely by roots within distance $K/n$ of the unit circle.
- As a byproduct, Theorem 1.3 states that a random Taylor series with i.i.d. coefficients has no double zero away from the origin almost surely, extending a previously known Gaussian special case to all coefficient laws with finite $E\log(1+|a_0|)$.
Reading between the lines
- A direct extension the paper leaves implicit: for i.i.d. complex coefficients the same $t^3$ Poisson law should hold with a possibly different constant, since the upper-half-plane reduction would be unnecessary; a complex-Gaussian simulation would show whether the conjugation symmetry in Section 3 is essential or merely organizational.
- The $(n\varepsilon)^2$ repulsion factor is a transferable template: any planar point field whose pair correlation carries such a factor at scales $\varepsilon\gg n^{-1}$ should exhibit the same $n^{-5/4}$ close-pair scale and $t^3$ distance intensity, and this paper supplies the first rigorous instance for random roots.
- The constant $c_*=\tfrac14\int_{-\infty}^{\infty}F(x)\,dx$, explicit in principle through (78), is never evaluated numerically; computing it to a few digits would turn Theorem 1.1 into a quantitative prediction that numerical experiments on the minimal gap could confirm to several digits.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies Kac polynomials with i.i.d. mean-zero sub-Gaussian coefficients and proves that the set of n^{5/4}-scaled pairwise root distances converges vaguely to a non-homogeneous Poisson point process on R_{\ge 0} with intensity c_* t^3 dt, for an explicit positive constant c_*. The proof reduces the problem to net counts in the annulus near the unit circle, uses small-ball probability estimates for (f_n, f'_n, f''_n) at 'smooth' points, compares these statistics with Gaussian ones for separated tuples, and computes the intensity by a Gaussian calculation. A separate result (Theorem 1.3) states that a random Taylor series with log-integrability has no double zeros inside the unit disk almost surely, except possibly at the origin with probability P[a_0=0]^2.
Significance. If the identified issues are corrected, this is a substantial universality result: root separation is one of the few genuinely pairwise quantities of random polynomials for which a full distributional limit is obtained beyond Gaussian models, and the explicit form c_* t^3 yields a limit law for the minimal separation (Corollary 1.2). The proof is a coherent chain of reductions, and Theorem 1.3 is self-contained and of independent interest. The intensity c_* is defined by an explicit positive integral rather than fitted, and the small-ball/Gaussian-comparison machinery is developed in detail. The current version, however, contains a sign inconsistency in the central event and a mismatch in the definition of the counting statistic; these are correctable but load-bearing.
major comments (3)
- [Section 5, Eqs. (50), (55), (56); see also (67) and (133)] The events A_z(U), A^+_z(U), and A^-_z(U) are defined with f_n(z)/f'_n(z) - z in R^\circ_z or the analogous rectangles, but every root-locating statement in Section 5 requires the opposite sign: Claim 5.5 concludes z - f_n(z)/f'_n(z) in R^\sharp_z, while Claim 5.6 and Lemma 5.7 use z - f_n(z)/f'_n(z) in R_z. On the good event G from (53) together with |f'_n(z)| \ge n^{5/4}/\log n, the printed event forces |f_n(z)| = |f_n(z)/f'_n(z)| |f'_n(z)| \asymp 2|z| n^{5/4}/\log n, contradicting the bound |f_n(z)| \le n^{1/2} \log^2 n. Moreover, the Gaussian computation in Section 10, especially (136), conditions on F_n(w)/F'_n(w) lying in a small rectangle around 0, which is the event z - f_n(z)/f'_n(z) \in R^\circ_z and not the displayed event. Thus, as written, Proposition 5.3 and the moment computation in Section 6 establish Poisson limits for two different counting events; the reduction to the net is valid only after the sign in (50), (55), (56), (67), and (133) is corrected.
- [Section 3, Eq. (23); proof of Theorem 1.1] The definition of X_n(U) in (23) counts the number of roots \alpha \in \Omega_K that have some root \alpha' at distance in n^{-5/4}U, so each close pair contributes two to X_n(U). The proof of Proposition 5.3 and the proof of Theorem 1.1 treat X_n(U) as the number of unordered pairs of roots at such a distance: X^\pm_n(U) in (57) sum over unordered pairs, and the proof of Theorem 1.1 identifies \mu^K_n(U) with X_n(U). These two quantities differ by a factor of two. Since the intensity c_*(K) is defined and computed for the pair count (see Claim 6.6 and the factor 2^{-m} in (84)), the statement of Theorem 3.2 is inconsistent with definition (23). The definition should be corrected to count unordered pairs, or all prefactors and the limiting intensity must be rescaled consistently.
- [Theorem 1.1 and Section 3] Theorem 1.1 does not state that the coefficients are real-valued, but the proof relies on this: Section 3 begins with the symmetry statement for real coefficients and defines X_n(U) using only roots in the upper half-plane H, and the proof of Theorem 1.1 uses only roots in H. Without the real-coefficient assumption, the symmetry argument and the identification of the limiting process with distances between H-roots fail. Either add the real-coefficient hypothesis to Theorem 1.1, the abstract, and Corollary 1.2, or provide an argument that covers complex coefficients.
minor comments (3)
- [Proof of Lemma 4.3] The subscripts in the four sums in the proof are inconsistent with the definitions in (34) and (35): the sums labelled N^{(1)}_s, N^{(2)}_s, and N^{(3)}_s should be over the non-smooth parts N^{(1)}_{ns}, N^{(2)}_{ns}, and N^{(3)}_{ns}.
- [Section 3] The sentence 'Since f_n has real coefficients, f_n(z) = f_n(z)' should read f_n(z) = \overline{f_n(\bar z)}; as printed it is either a typographical error or an empty identity.
- [Abstract] The abstract states only that the coefficients are independent and identically distributed, omitting the mean-zero, sub-Gaussian, and real-coefficient hypotheses used in Theorem 1.1; the abstract should state the assumptions that are actually proved.
Circularity Check
No significant circularity: the Poisson intensity is computed from an explicit Gaussian integral, not fitted, and the proof is self-contained against external benchmarks.
full rationale
The derivation of Theorem 1.1 is not circular. The limit object is reduced to Theorem 3.2, where the intensity lambda_{K,U} = c_*(K) integral_U t^3 dt is computed from first principles: c_*(K) is defined by (68) as an explicit integral of the function F(x) from (78), and Section 10 evaluates the Gaussian small-ball probability (133)-(137) directly; no parameter is fitted to the separation data, and the t^3 factor arises from the change of variables in the Gaussian computation rather than being imported from the conclusion. The net reduction in Sections 5-6 is justified by Rouché-type claims (Claims 5.5-5.7, Lemma 5.8), and the moment bounds use Gaussian comparison and small-ball estimates proved in Sections 7-9. Cited results by the same authors are used as technical tools, not as premises containing the separation theorem: in particular, [Coo+23, Lemma 3.5] is invoked only as a routine Gaussian comparison for densities after the covariance block structure is derived in (128), and [MS20] and [Cam+24] serve as background or technical input. Theorem 1.3 is proved independently in Section 2. The apparent sign mismatch between the displayed event (50) and the z - fn/f' condition used in Claims 5.5-5.6, as well as the unspecified real-coefficient assumption, are correctness risks rather than circularity: they do not make the derived quantity equal to an input by definition.
Assumptions & free parameters
free parameters (2)
- tau =
10^{-4}
- beta =
tau/20 = 5*10^{-6}
assumptions (5)
- domain assumption The coefficients xi_0,...,xi_n are i.i.d., mean-zero, sub-Gaussian, and P[xi_0 = 0] = 0.
- domain assumption The coefficients are real-valued, so f_n(z) = f_n(bar z) and roots are conjugate-symmetric.
- standard math Classical tools: Esseen inequality, Levy-Kolmogorov-Rogozin anti-concentration, Rouche's theorem, Hurwitz's theorem, Jensen's bound, Salem-Zygmund inequality.
- domain assumption The effective degree d_n(z) = min{n, (1-|z|)^{-1}} controls the local behavior of f_n at z.
- standard math The Riemann sum approximation of the integral of F in Claim 6.6 converges uniformly for K in compact sets.
Cite this review
Pith. "Pith review of Limit law for root separation in random polynomials." pith.science (2026). https://pith.science/paper/UJSBUBRR
@misc{pith2026250502723,
author = {Pith},
title = {Pith review of: Limit law for root separation in random polynomials},
year = {2026},
howpublished = {\url{https://pith.science/paper/UJSBUBRR}},
note = {Machine review of arXiv:2505.02723}
}
abstract
Let $f_n$ be a random polynomial of degree $n\ge 2$ whose coefficients are independent and identically distributed random variables. We study the separation distances between roots of $f_n$ and prove that the set of these distances, normalized by $n^{-5/4}$, converges in distribution as $n\to \infty$ to a non-homogeneous Poisson point process. As a corollary, we deduce that the minimal separation distance between roots of $f_n$, normalized by $n^{-5/4}$ has a non-trivial limit law. In the course of the proof, we establish a related result which may be of independent interest: a Taylor series with random i.i.d. coefficients almost-surely does not have a double zero anywhere other than the origin.
Figures
Forward citations
Cited by 1 Pith paper
-
Law of large numbers for the discriminant of random polynomials
Random Kac polynomials have discriminant |Δ(f_n)| = n^{2n} e^{-D_* n(1+o(1))} with an explicit universal constant D_* ≈ 5.92947.
Reference graph
Works this paper leans on
-
[1]
[BBL96] E. Bogomolny, O. Bohigas, and P. Leboeuf. Quantum chaotic dynamics and random polynomials. J. Statist. Phys. 85.5-6 (1996), pp. 639–679. [BP31] A. Bloch and G. P´ olya. On the Roots of Certain Algebraic Equations. Proc. London Math. Soc. (2) 33.2 (1931), pp. 102–114. [BR10] R. N. Bhattacharya and R. R. Rao. Normal approximation and asymptotic expa...
work page 1996
-
[5]
Cambridge University Press, Cam- bridge, 1985, pp. xiv+305. [Kal17] O. Kallenberg. Random measures, theory and applications. Vol
work page 1985
-
[23]
[Hal77] G. Hal´ asz. Estimates for the concentration function of combinatorial number theory and probability. Period. Math. Hungar. 8.3-4 (1977), pp. 197–211. [Ham56] J. M. Hammersley. The zeros of a random polynomial. Proceedings of the Third Berkeley Symposium on Mathematical Statistics and Probability, 1954–1955, vol. II . 1956, pp. 89–111. [Hou+09] J....
work page 1977
-
[25]
Bolyai Soc. Math. Stud. J´ anos Bolyai Math. Soc., 2013, pp. 409–463. [Pic96] B. Picinbono. Second-order complex random vectors and normal distributions. IEEE Transactions on Signal Processing 44.10 (1996), pp. 2637–2640. [PSZ16] R. Peled, A. Sen, and O. Zeitouni. Double roots of random Littlewood polynomials. Israel J. Math. 213.1 (2016), pp. 55–77. [PV0...
work page 1996
-
[45]
Cambridge University Press, New York, 1957, pp
Cambridge Tracts in Mathematics and Mathematical Physics. Cambridge University Press, New York, 1957, pp. x+166. [CN21] N. A. Cook and H. H. Nguyen. Universality of the minimum modulus for random trigonometric polynomials. Discrete Anal. (2021), Paper No. 20,
work page 2021
-
[46]
[Coo+23] N. A. Cook, H. H. Nguyen, O. Yakir, and O. Zeitouni. Universality of Poisson limits for moduli of roots of Kac polynomials. Int. Math. Res. Not. IMRN 8 (2023), pp. 6648–6683. [EO56] P. Erd¨ os and A. C. Offord. On the number of real roots of a random algebraic equation. Proc. London Math. Soc. (3) 6 (1956), pp. 139–160. [Ess66] C. G. Esseen. On t...
work page 2023
-
[47]
Cambridge University Press, Cambridge, 2018, pp. xiv+284. 77
work page 2018
- [64]
Show all 11 references
-
[77]
Springer, Cham, 2017, pp
Probability Theory and Stochas- tic Modelling. Springer, Cham, 2017, pp. xiii+694. [Kon94] S. V. Konyagin. On the minimum modulus of random trigonometric polynomials with coefficients ±1. Mat. Zametki 56.3 (1994), pp. 80–101,
1994
- [158]
-
[2009]
Ibragimov and D
[IZ13] I. Ibragimov and D. Zaporozhets. On distribution of zeros of random polynomials in complex plane. Prokhorov and contemporary probability theory. Springer Proc. Math. Stat. 2013, pp. 303–323. 76 [IZ97] I. Ibragimov and O. Zeitouni. On roots of random polynomials. Trans. ...
1997
Reviewed August 16, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.