Overview

  • Which coefficient laws give irreducibility as degree grows?
  • When is Aₙ contained in the group, and when is the group Sₙ?
  • How do modular factorization and discriminant estimates enter the proofs?

Probability models

In each row the displayed limit is taken with the stated law or degree fixed. A conclusion for one row need not hold in another.

Model and independent dataLimit and conditioningConclusion and source
Monic xn+∑j<nξjxjx^n+\sum_{j<n}\xi_jx^j; all nn coordinates uniform in [−H,H]∩Z[-H,H]\cap\mathbb Z.Fixed n≥3n\ge3; H→∞H\to\infty. No conditioning.The number with group different from SnS_n, including repeated-root cases, is On(Hn−1)O_n(H^{n-1}). Bhargava, Theorem 1.
Monic polynomial; nonleading coefficients independent with a fixed finitely supported nonconstant law μ\mu.n→∞n\to\infty; condition on ξ0≠0\xi_0\ne0.For some θ=θ(μ)>0\theta=\theta(\mu)>0, no nonconstant factor of degree less than θn\theta n, with probability 1−o(1)1-o(1). Bary-Soroker–Koukoulopoulos–Kozma, Theorem 2.
Same monic model; μ\mu uniform on a fixed interval of at least 3535 consecutive integers.n→∞n\to\infty; condition on ξ0≠0\xi_0\ne0.Irreducible and group AnA_n or SnS_n, with probability 1−o(1)1-o(1). Same source, Theorems 1(a) and 6.
Monic polynomial; nonleading coefficients independent uniform in [−L(n),L(n)]∩Z[-L(n),L(n)]\cap\mathbb Z.n→∞n\to\infty, L(n)→∞L(n)\to\infty. No conditioning.Group SnS_n with probability 1−o(1)1-o(1). Bary-Soroker–Goldgraber, Theorem 1.1. No fixed-width conclusion follows.
Littlewood polynomial; all n+1n+1 coefficients independent fair signs.n=pr−1→∞n=p^r-1\to\infty, with p=2p=2, or prime p≥7p\ge7 for which 22 generates (Z/p2Z)×(\mathbb Z/p^2\mathbb Z)^\times. No conditioning.Irreducible with probability at least 1−n−c1-n^{-c}. Bary-Soroker–Hokken–Kozma–Poonen, Theorem 1.2.
Binary polynomial; leading and constant coefficients equal 11; internal coefficients independent uniform in {0,1}\{0,1\}.n→∞n\to\infty; no further conditioning.A recent preprint claims reducibility probability 2/(πn)+O(n−1)\sqrt{2/(\pi n)}+O(n^{-1}). Kogler, Theorem 1.1, v2.
Monic reciprocal polynomial of degree 2m2m; mm free coefficients independent uniform on a fixed interval of at least 3535 integers; reflected coefficients are equal.m→∞m\to\infty; no conditioning.Preprint: irreducible, with group among C2≀SmC_2\wr S_m and two specified index-two subgroups, with probability 1−o(1)1-o(1). Hokken–Koukoulopoulos, Theorem A, v2.

None of the displayed coefficient results assumes a Riemann hypothesis. For arbitrary fixed laws, however, the small-factor conclusion does not exclude two factors of comparable degree. If μ(0)>0\mu(0)>0, the event ξ0=0\xi_0=0 has fixed positive probability, so irreducibility cannot tend to one without addressing it. Fixing the constant coefficient to 11 defines a different model and must be justified separately.

Matrix models, multiplicative coefficients, and the function-field model have their own independence and arithmetic hypotheses. In the repeated-root theorem below, every coefficient is randomized, including the leading one.

References 6

Small factors and interval laws

Theorem · Bary-Soroker–Koukoulopoulos–Kozma

For every fixed, finitely supported, nonconstant law μ\mu, there is θ>0\theta>0 such that, conditional on ξ0≠0\xi_0\ne0, the probability of a nonconstant factor of degree less than θn\theta n tends to zero.

If μ\mu is uniform on a fixed interval of at least 3535 consecutive integers, then, under the same conditioning,

Pr⁡(Fn irreducible and Gal⁡(Fn/Q)⊇An)⟶1.\Pr(F_n\text{ irreducible and }\operatorname{Gal}(F_n/\mathbb Q)\supseteq A_n)\longrightarrow1.

The small-factor estimate leaves open the possibility of two factors of comparable degree, for example near n/2n/2. In the interval case, the group is AnA_n or SnS_n with probability tending to one.

References 1

Cyclotomic factors and error bounds

For coefficients uniform on a fixed interval of N≥108N\ge10^8 integers, Bazin’s Theorem 1.1.2 gives

Pr⁡(Fn reducible)≤Pr⁡(ξ0=0)+O(n−1/2).\Pr(F_n\text{ reducible})\le\Pr(\xi_0=0)+O(n^{-1/2}).

For each fixed C>0C>0, if the interval is sufficiently long in terms of CC, Theorem 1.1.4 gives

Pr⁡(Fn reducible)=Pr⁡(∃ω∈UC:Fn(ω)=0)+OC(n−C/2),\Pr(F_n\text{ reducible})=\Pr(\exists\omega\in U_C:F_n(\omega)=0)+O_C(n^{-C/2}),

where UCU_C consists of zero and the roots of unity of degree less than CC over Q\mathbb Q.

References 1
  • Bazin2025 · Theorems 1.1.2 and 1.1.4, v2The constant-term obstruction and cyclotomic refinement are explicit.Source

Repeated roots

Consider Pn(x)=∑j=0nηjxjP_n(x)=\sum_{j=0}^n\eta_jx^j, with all coefficients, including the leading coefficient, independent and of a fixed bounded integer law μ\mu with max⁡aμ(a)≤1/2\max_a\mu(a)\le1/2. Feldheim–Sen prove

Pr⁡(Pn has a repeated root)=Pr⁡(Pn has a repeated root in {0,1,−1})+o(n−2).\Pr(P_n\text{ has a repeated root})=\Pr(P_n\text{ has a repeated root in }\{0,1,-1\})+o(n^{-2}).

If μ(0)=0\mu(0)=0, this probability is O(n−2)O(n^{-2}). If μ(0)>0\mu(0)>0, then η0=η1=0\eta_0=\eta_1=0 has probability μ(0)2\mu(0)^2 and forces a repeated root at zero.

The proof separates possible roots by algebraic degree and size and uses anti-concentration for integer evaluations. Applying the theorem to a monic or conditioned ensemble requires a separate justification.

References 1
  • FeldheimSen2017 · Theorem 1.1 and §1.1The full iid model and the largest-atom condition are essential to the stated theorem.Source

Root separation and discriminant size

For independent, identically distributed mean-zero sub-Gaussian coefficients with no atom at zero, Michelen–Yakir prove that the minimum distance between roots, indexed with multiplicity, has scale n−5/4n^{-5/4} and a nondegenerate limiting law. Their Corollary 1.2 gives a vanishing probability of repeated roots under these hypotheses, without the largest-atom restriction for bounded integer laws. The earlier theorem for Littlewood polynomials gives a sharper quantitative bound in that model.

In a separate preprint, for Pn(x)=∑j=0nηjxjP_n(x)=\sum_{j=0}^n\eta_jx^j under the additional normalization E∣ηj∣2=1\mathbb E|\eta_j|^2=1, they report

log⁡∣disc⁡(Pn)∣−2nlog⁡nn→Pr⁡−D∗,D∗>0.\frac{\log|\operatorname{disc}(P_n)|-2n\log n}{n}\xrightarrow{\Pr}-D_*,\qquad D_*>0.

The proof uses a reciprocal form of the discriminant identity, Mahler measure, and root statistics. Distinguishing AnA_n from SnS_n requires an additional arithmetic argument about the square class. The no-zero-atom hypothesis removes the repeated-root obstruction at zero.

References 2

Fourier estimates and Frobenius

For a nonzero candidate divisor D∈Fp[x]D\in\mathbb F_p[x], divisibility is the event that a random sum vanishes in Fp[x]/(D)\mathbb F_p[x]/(D). Independence gives a product formula for its Fourier transform. Decay of the product controls deviation from uniformity, with exceptions determined by the characters and the structure of the quotient algebra. Possible rational factor degrees must occur among the subset sums of the modular factor degrees at each prime. Combinatorial estimates bound the probability of a common subset sum.

At primes of squarefree reduction, the modular factors give Frobenius cycle types. Probabilistic permutation-group estimates exclude most proper transitive groups. A second method averages root counts over primes and uses effective prime-ideal estimates to study global Galois orbits. Breuillard–Varjú use Riemann hypotheses for Dedekind zeta functions in this step.

References 2

The square-discriminant probability

Reported preprint result · Blachar–Breuillard–Kozma, v1

Let Fn=xn+∑j=1n−1ξjxj+1F_n=x^n+\sum_{j=1}^{n-1}\xi_jx^j+1, where ξ1,…,ξn−1\xi_1,\ldots,\xi_{n-1} are independent with a fixed finitely supported integer law μ\mu. If

H2(μ):=−log⁡ ⁣∑a∈Zμ(a)2>12,H_2(\mu):=-\log\!\sum_{a\in\mathbb Z}\mu(a)^2>12,

the preprint reports Pr⁡(disc⁡Fn is a square)=Oμ(1/log⁡n)\Pr(\operatorname{disc}F_n\text{ is a square})=O_\mu(1/\log n). This is the constant-term-one case of its stated Theorem 1.1.

Corollary 1.2 states that when all ξ0,…,ξn−1\xi_0,\ldots,\xi_{n-1} are independent and uniform on a fixed interval of N>e12N>e^{12} integers, xn+∑j=0n−1ξjxjx^n+\sum_{j=0}^{n-1}\xi_jx^j is irreducible with group SnS_n, conditional on ξ0≠0\xi_0\ne0, with probability 1−Oμ(1/log⁡n)1-O_\mu(1/\log n).

Remark on the scope of Theorem 1.1

The unqualified square-discriminant bound in v1 fails for some permitted constant-term laws. For Rn=xn+∑j=0n−1ξjxjR_n=x^n+\sum_{j=0}^{n-1}\xi_jx^j, if ξ0,ξ1\xi_0,\xi_1 are independent with law μ\mu and μ(0)>0\mu(0)>0, then

Pr⁡(disc⁡Rn=0)≥Pr⁡(ξ0=ξ1=0)=μ(0)2.\Pr(\operatorname{disc}R_n=0)\ge\Pr(\xi_0=\xi_1=0)=\mu(0)^2.

On this event, x2∣Rnx^2\mid R_n and the discriminant is the square zero. The obstruction occurs even for a uniform interval with N>e12N>e^{12} containing zero. The O(1/n)O(1/n) repeated-root estimate invoked at the start of Section 8 cannot hold in that setting. Fixing a nonzero constant term removes this particular obstruction. The restricted statements above remain reported preprint claims; this observation does not verify the rest of their proof.

The argument studies double roots modulo primes and their lifts modulo prime squares to bound discriminant valuations. Once AnA_n-containment is known, a nonsquare discriminant forces SnS_n. For a uniform law on NN values, H2=log⁡NH_2=\log N. The fair-sign law has H2=log⁡2H_2=\log2, below the required threshold.

References 1
  • BlacharBreuillardKozma2026 · Theorem 1.1, Corollary 1.2, and opening of §8, v1Reported preprint statements; the atlas excludes the zero-constant obstruction and records the scope issue in the original square-discriminant assertion.Source

Example

For coefficients uniform on {−17,…,17}\{-17,\ldots,17\}, the 35-point theorem gives irreducibility and AnA_n-containment with probability tending to one conditional on ξ0≠0\xi_0\ne0. Without conditioning, xx is a factor with probability 1/351/35.

For coefficients uniform on {−1,1}\{-1,1\}, the constant term is nonzero. The 35-consecutive-integer theorem and the entropy bound both require larger supports. For coefficients uniform on [−L(n),L(n)]∩Z[-L(n),L(n)]\cap\mathbb Z with L(n)→∞L(n)\to\infty, Bary-Soroker–Goldgraber give full SnS_n with probability tending to one. In this last model the probability of a zero constant term tends to zero.

An experiment should record the coefficient law, degree, sample size and group-certification method. It estimates a probability for that model; it does not prove the limiting theorem.

References 1

Multiplicative coefficients

Choose independent fair signs XpX_p for primes pp, put X1=1X_1=1, and extend completely multiplicatively by Xn=∏pa∥nXpaX_n=\prod_{p^a\parallel n}X_p^a. Define

Pd(x)=∑j=1dXjxj.P_d(x)=\sum_{j=1}^dX_jx^j.

Assuming RH for the Dedekind zeta functions of all number fields, Varjú–Xu prove that for every ε>0\varepsilon>0,

Pr⁡(Pd(x)/x irreducible over Q)≥1−Oε(d−1/2+ε).\Pr(P_d(x)/x\text{ irreducible over }\mathbb Q)\ge1-O_\varepsilon(d^{-1/2+\varepsilon}).

The theorem also covers the variant with Xj=0X_j=0 at nonsquarefree integers. In both models one removes the factor xx; in the second, the degree of the original polynomial may be less than dd. For example, the coefficients satisfy X6=X2X3X_6=X_2X_3. The proof combines finite-field root averages with arithmetic equidistribution. The corresponding deterministic questions for Fekete polynomials are separate.

References 1
  • VarjuXu2026 · Theorem 1.1; §1.1; Problem 1.2Both multiplicative models are specified in the introduction; the conditional conclusion concerns P_d/x.Source

Polynomials over a function field

Fix a prime power qq and set

f(x,y)=yn+∑i=0n−1ai(x)yi,ai independent and uniform in {a∈Fq[x]:deg⁡a≤d}.f(x,y)=y^n+\sum_{i=0}^{n-1}a_i(x)y^i,\qquad a_i\text{ independent and uniform in }\{a\in\mathbb F_q[x]:\deg a\le d\}.

Entin–Popov take the splitting field over Fq(x)\mathbb F_q(x). For fixed n,qn,q and d→∞d\to\infty, the probability of irreducibility, separability, and Galois group SnS_n tends to one. For fixed d≥1,qd\ge1,q and n→∞n\to\infty,

Pr⁡(f irreducible)⟶1−q−d,Pr⁡(Gf∈{An,Sn}∣f irreducible)⟶1.\Pr(f\text{ irreducible})\longrightarrow1-q^{-d},\qquad\Pr(G_f\in\{A_n,S_n\}\mid f\text{ irreducible})\longrightarrow1.

Entin subsequently removed the Chowla hypothesis for odd qq: with fixed d≥1d\ge1, the conditional probability Pr⁡(Gf=Sn∣f irreducible)\Pr(G_f=S_n\mid f\text{ irreducible}) tends to one. His method does not treat characteristic two. The base field here is Fq(x)\mathbb F_q(x). Over the finite field Fq\mathbb F_q itself, every irreducible degree-nn polynomial has cyclic Galois group of order nn.

References 2
  • EntinPopov2024 · Theorems 1.1, 1.2, 1.4 and 1.6; Conjecture 1.5Entin–Popov’s original full symmetric conclusion was conditional and assumed odd q; Entin’s later theorem removes that condition on Chowla.Source
  • Entin2025RationalFunction · Corollary 1.1 and Remark 1.4Unconditional full symmetric group in odd characteristic; the characteristic-two case lies outside the theorem.Source

Questions

Research directionCounting and probability

Can one prove full symmetric groups for small fixed coefficient laws?

Scope. Independent bounded coefficients over Q, with the zero-constant obstruction excluded.

The entropy hypothesis of the cited preprint excludes small-support laws, including fair signs. For these laws, the stated estimate gives no bound for the square-discriminant probability.

References 1
Discuss this question →
OpenCounting and probability

How short can the prime interval be?

Scope. A prime P sampled uniformly from (d,H(d)], where this interval contains at least one prime.

Set

Td,P(x)=∑a=1d(aP)xa.T_{d,P}(x)=\sum_{a=1}^d\left(\frac aP\right)x^a.

Varjú–Xu ask for the probability that this polynomial is irreducible after its linear factors are removed. Assuming RH for the Dedekind zeta functions of all number fields, they prove

Pr⁡(Td,P/x irreducible over Q)≥1−Oε(d−1/2+ε)\Pr\bigl(T_{d,P}/x\text{ irreducible over }\mathbb Q\bigr)\ge1-O_\varepsilon(d^{-1/2+\varepsilon})

when H(d)>22π(d)d4H(d)>2^{2\pi(d)}d^4; here π(d)\pi(d) counts primes at most dd. The behavior in shorter intervals is the open problem. The endpoint d=P−1d=P-1 gives the classical Fekete family, whose additional linear factors must also be removed.

References 1
  • VarjuXu2026 · Problem 1.2 and Corollary 1.3The random-prime truncated model is different from the all-prime classical Fekete conjecture.Source
Discuss this question →
OpenCounting and probability

Does the full symmetric conclusion hold in characteristic two?

Scope. Fixed q a power of two and d≥1; independent uniform coefficients in F_q[t] of degree at most d; polynomial degree n→∞.

Does Pr⁡(Gf=Sn∣f irreducible)→1\Pr(G_f=S_n\mid f\text{ irreducible})\to1? Entin proves this for odd qq. Remark 1.4 explains why the discriminant argument does not extend to characteristic two.

References 2
Discuss this question →

Related topics

References

  1. Galois groups of random integer polynomials and van der Waerden’s Conjecture

    Bhargava, Manjul. Annals of Mathematics, 201(2), 339–377, 2025.

    Read the source
  2. Irreducibility of random polynomials: general measures

    Bary-Soroker, Lior and Koukoulopoulos, Dimitris and Kozma, Gady. Inventiones Mathematicae, 233(3), 1041–1120, 2023.

    Read the source
  3. Full Galois groups of polynomials with slowly growing coefficients

    Bary-Soroker, Lior and Goldgraber, Noam. Bulletin of the London Mathematical Society, 57(3), 941–955, 2025.

    Read the source
  4. Irreducibility of Littlewood polynomials of special degrees

    Bary-Soroker, Lior and Hokken, David and Kozma, Gady and Poonen, Bjorn. International Mathematics Research Notices, 2025(21), rnaf326, 2025.

    Read the source
  5. The Odlyzko-Poonen Conjecture on Irreducibility of Random Polynomials

    Kogler, Constantin. 2026.

    Read the source
  6. Irreducibility and Galois groups of random reciprocal polynomials of large degree

    Hokken, David and Koukoulopoulos, Dimitris. 2025.

    Read the source
  7. Irreducibility of random polynomials of ℤ[X]

    Bazin, Pierre-Alexandre. International Mathematics Research Notices, 2025(11), rnaf136, 2025.

    Read the source
  8. Double roots of random polynomials with integer coefficients

    Feldheim, Ohad N. and Sen, Arnab. Electronic Journal of Probability, 22, Article 10, 1–23, 2017.

    Read the source
  9. Limit law for root separation in random polynomials

    Michelen, Marcus and Yakir, Oren. Advances in Mathematics, 496, Article 110990, 2026.

    Read the source
  10. Law of large numbers for the discriminant of random polynomials

    Michelen, Marcus and Yakir, Oren. 2025.

    Read the source
  11. Irreducibility of random polynomials of large degree

    Breuillard, Emmanuel and Varjú, Péter P. Acta Mathematica, 223(2), 195–249, 2019.

    Read the source
  12. Galois groups of random polynomials of large degree

    Blachar, Guy and Breuillard, Emmanuel and Kozma, Gady. 2026.

    Read the source
  13. A random polynomial with multiplicative coefficients is almost surely irreducible

    Varjú, Péter P. and Xu, Max Wenqiang. International Mathematics Research Notices, 2026(16), rnag178, 2026.

    Read the source
  14. Probabilistic Galois theory in function fields

    Entin, Alexei and Popov, Alexander. Finite Fields and Their Applications, 98, Article 102466, 2024.

    Read the source
  15. Galois groups of random polynomials over the rational function field

    Entin, Alexei. Journal of the London Mathematical Society, 111(1), Article e70061, 2025.

    Read the source