Irreducibility of polynomials with random multiplicative coefficients revisited
Oleksiy Klurman, Vlad Matei · 2025
Read Irreducibility of polynomials with random multiplicative coefficients revisited by Oleksiy Klurman, Vlad Matei (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-10359.txt.
Introduction
Oleksiy Klurman and Vlad Matei revisit a question in random polynomials: how often is a polynomial with random multiplicative coefficients irreducible, meaning it cannot be factored into smaller polynomials? Their coefficients come from a random completely multiplicative function taking values plus or minus one, so each coefficient is forced by the values at primes. The polynomial's degree grows with a cutoff parameter.
The authors prove that the probability of irreducibility has limsup equal to one as the degree grows. In plain terms, along a suitable sequence of degrees, almost all such random polynomials are irreducible, even though the coefficients are far from independent. The result sharpens earlier understanding of factorization behavior in this rigid random model.
Catalog record
| CREATOR | Oleksiy Klurman, Vlad Matei |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 14 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-10359
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Irreducibility of polynomials with random multiplicative coefficients revisited Oleksiy Klurman, Vlad Matei Abstract . We show that for a random polynomial F ( X ) = ∑ n = 1 N f ( n ) X n − 1 , F(X)=\sum_{n=1}^{N}f(n)X^{n-1}, where f ( n ) f(n) is a random completely multiplicative function taking values in { ± 1 } \{\pm 1\} , one has lim sup N → ∞ ℙ [ F ( X ) is irreducible ] = 1 . \limsup_{N\to\infty}\mathbb{P}\big[F(X)\text{ is irreducible}\big]=1. 1. Introduction The study of the irreducibility of polynomials with integer coefficients has a long and rich history. One may ask this question for a specific polynomial or study various families of polynomials. The first model is the “large box model,” where we fix the degree of a polynomial and let the coefficients be drawn at random from the box [-H,H] as H\to\infty . The papers [ 8 ] , [ 13 ] , and the references therein describe some recent spectacular developments in this subject. Another setting that has led to fruitful developments considers polynomials whose bounded coefficients are sampled independently from a fixed distribution law, while the degree of the polynomials in the family grows. The works [ 4 ] , [ 5 ] , and [ 6 ] establish various exciting instances of this universality phenomenon for irreducibility—either when the coefficients are drawn from a sufficiently large finite set of integers or when they are near-uniformly distributed modulo enough primes. Conditionally on the Extended Riemann Hypothesis (ERH) over number fields, Breuillard and Varjú [ 10 ] proved that polynomials with coefficients that are independent \{0,1\} -valued random variables, each taking the value 1 with probability 1/2 , are almost surely irreducible. This conditionally resolved an old conjecture of Odlyzko and Poonen ( [ 33 ] ). Their method also yields sharp estimates for the probability that such a random polynomial is reducible. Our work is motivated by the general question of irreducibility when the coefficients exhibit some form of dependence. There are several natural frameworks in which such dependencies arise, particularly in the study of characteristic polynomials of various ensembles of random matrices (see [ 19 ] , [ 18 ] , [ 2 ] for recent results in this direction). Here we confine ourselves to a more arithmetic framework, where the dependencies are related to multiplicative functions. The most studied example of such families of polynomials is the well-known Fekete polynomials of degree |D|-1 , defined by F_{D}(z):=\sum_{n=1}^{|D|-1}\chi_{D}(n)z^{n}, where D is any fundamental discriminant and \chi_{D} is the corresponding quadratic character. The properties of the real and complex zeros of such polynomials, as well as conjectures concerning their irreducibility, have been the subject of extensive recent research [ 24 ] , [ 9 ] , [ 15 ] , [ 30 ] , [ 1 ] , [ 25 ] . To facilitate our discussion, let \{f(p)\}_{p} be a sequence of independent Bernoulli \pm 1 random variables indexed by the primes, and define f(n)=\prod_{p^{k}\|n}f^{k}(p) to be a random completely multiplicative function. The statistical behavior of the closely related Rademacher random multiplicative model has seen tremendous recent progress, most notably due to the work of Harper [ 22 ] on Helson’s conjecture. Let f:\mathbb{N}\to\{-1,1\} be a random completely multiplicative function, and consider the family of polynomials P_{f,N}(X)=\sum_{n=1}^{N}f(n)X^{n-1}. Analytic properties of such objects have been extensively studied in the recent works [ 21 ] and [ 7 ] . Very recently, Varjú and Xu [ 36 ] , conditionally on the Extended Riemann Hypothesis, proved that such polynomials are irreducible with probability 1-o_{N\to\infty}(1) . We instead unconditionally establish this result for a subsequence of degrees, inspired by the work of Bary-Soroker, Hokken, Kozma, and Poonen [ 3 ] , who addressed the analogous question for independent coefficients. Our main result is the following. Theorem 1.1 . Let N=2^{k} . Then \lim_{k\rightarrow\infty}\mathbb{P}[P_{f,N}(X)\hskip 5.69054pt\text{is irreducible}]=1. In the last section we briefly discuss application of this result to the deterministic family of Turyn polynomials. 2. Proof of Main Theorem In what follows, we shall always assume that all constants are sufficiently large unless stated otherwise. Lemma 2.1 . Consider \tilde{P}_{f,N}(X)=P_{f,N}(X+1)=\sum_{j\leq N}a_{j}x^{j} where N=2^{k} and j\geq 1 and fix large A\ll\log\log N. Then \mathbb{P}(a_{i}\equiv 0\pmod{4}\hskip 8.53581pt\forall i\in\mathcal{A})\leq\exp(-cA), where \mathcal{A}=\{2^{A+1}+2^{i}+1,|\hskip 8.53581pt1\leq i\leq A\}. Proof. First, we note that \tilde{P}_{f,N}(X)=\sum_{l=0}^{N-1}X^{l}\left(\sum_{l+1\leq n\leq N}f(n)\binom{n-1}{l}\right), and introduce \displaystyle g_{l}=\sum_{l+1\leq n\leq N}f(n)\binom{n-1}{l}. Denoting by \tilde{f}(n)=\cfrac{f(n)+1}{2} we rewrite g_{l}=2\sum_{l+1\leq n\leq N}\tilde{f}(n)\binom{n-1}{l}-\binom{N}{l+1}. For N=2^{k}, we have that \dbinom{2^{k}}{j} is even for 1\leq j\leq 2^{k}-1 . Thus we have that g_{l} is even for 0\leq l\leq N-1 and, moreover, its residue modulo 4 is determined by the parity of \sum_{l+1\leq n\leq N}\tilde{f}(n)\binom{n-1}{l}. For each a\in\mathcal{A} we can pick a prime p_{a} in \left[\cfrac{N}{2},N\right] such that p_{a}\equiv a\pmod{2^{A+1}} by the uniform version of the prime number theorem for arithmetic progressions (Theorem 1.1 in [ 28 ] ). Using Lucas’s theorem (see Theorem 2.5.14 in [ 31 ] ), we deduce that \dbinom{p_{a}-1}{a} is odd and \dbinom{p_{a}-1}{b} is even for any a\neq b\in\mathcal{A} . All that remains to note is that for coefficients g_{a} with a\in\mathcal{A} the values of \tilde{f}(p_{a}) behave like independent random variables taking values 0,1 and, moreover, g_{a}\equiv\tilde{f}(p_{a})+h_{a}\pmod{2} where the terms h_{a} do not involve any other term \tilde{f}(p_{b}) with b\neq a . Thus \mathbb{P}(a_{i}\equiv 0\pmod{4}\hskip 8.53581pt\forall i\in\mathcal{A})=\cfrac{1}{2^{A}} and the claim is proved. ∎ Lemma 2.2 . With probability (1-\exp(-cA)) we have that P_{f,N}(X) has an irreducible factor of degree \geq N-2^{A}. Proof. Consider the Newton polygon associated to the polynomial \tilde{P}_{f,N}\left(X\right) over \mathbb{Q}_{2} . With probability (1-\exp(-cA)) there exists a coefficient with 1\leq j\leq 2^{A} such that g_{j}\equiv 2\pmod{4} . Thus, the Newton polygon would have a segment of height 1 and length at least N-j , and hence an irreducible factor of degree at least N-j over \mathbb{Q}_{2} (see, for example, [ 20 ] , section 7.4). Consequently P_{f,N}(X) must have an irreducible factor of degree at least N-j over \mathbb{Q} with high probability claimed above. ∎ To finish the proof, we need to bound the probability that our random polynomial has a low degree factor. To this end, we work with an “independent” part of our polynomial given by \displaystyle\sum_{N/2\leq p\leq N,\hskip 2.84526ptp\hskip 2.84526pt\text{prime}}f(p)X^{p-1}. Lemma 2.3 . We have that for B\leq\log\log N \mathbb{P}[P_{f,N}(X)\hskip 8.53581pt\text{ has a factor of degree $\leq B$}\hskip 2.84526pt]=O(N^{-1/2+\varepsilon}). Proof. We will make use of Theorem 1.7 in [ 35 ] . In their notations, we take \Omega=\{z\in\mathbb{C}:|z|\leq 2\} , with M=2. More precisely, if we show that we can find p\in[0,1] such that \sup_{z\in\Omega}\mathbb{P}(P_{f,N}(z)=0)\leq p, then the probability that our polynomial P_{f,N}(X) has a factor of degree \leq B is bounded by p(2e)^{B^{2}}, since any root of P_{f,N}(X), say z_{0} satisfies the inequality |z_{0}|<2. We now focus on bounding probability \mathbb{P}(P_{f,N}(z)=0) for every z\in\Omega and distinguish between two cases depending on the size of |z|-1. Since the product of the roots is equal to \pm 1, we can restrict ourselves to looking only at the algebraic integers of modulus \geq 1 and estimating the probability. Case I. Suppose |z|>1+\cfrac{u_{N}}{N} where u_{N}\geq(\log N)^{2} and P_{f,N}(z)=0. Using Proposition 5.4 from [ 17 ] we can pick primes p_{1},\ldots,p_{r} in the corresponding intervals