Overview

  • Which entry laws give unconditional irreducibility?
  • How do matrix ranks determine root counts modulo primes?
  • What does irreducibility imply about rational invariant subspaces?

The independent-entry model

Let Mn=(mij)M_n=(m_{ij}) have independent entries of common law μ\mu, a fixed finitely supported nonconstant law on Z\mathbb Z, and set χn(t)=det⁡(tI−Mn)\chi_n(t)=\det(tI-M_n). As the dimension grows, one asks whether χn\chi_n is irreducible with probability tending to one and whether its Galois group is SnS_n with probability tending to one.

Example

For M=(abcd)M=\begin{pmatrix}a&b\\c&d\end{pmatrix},

χM(t)=t2−(a+d)t+(ad−bc).\chi_M(t)=t^2-(a+d)t+(ad-bc).

If a,b,c,da,b,c,d are signs and a+d=2a+d=2, the determinant is 00 or 22. If a+d=0a+d=0, it is −2-2 or 00. For independent fair signs, the conditional laws of the determinant differ, so the trace and determinant are dependent.

The dependence prevents a direct application of the independent-coefficient theorems to χn\chi_n. Symmetric, tridiagonal, and fixed-size height-bounded matrices are separate ensembles.

Eberhard’s theorems

Theorem · Eberhard

In the independent-entry model above:

  • If μ\mu is uniform modulo a product of four distinct primes, irreducibility and Gal⁡(χn/Q)⊇An\operatorname{Gal}(\chi_n/\mathbb Q)\supseteq A_n hold with probability tending to one, unconditionally.
  • For any fixed finitely supported nonconstant law μ\mu, assume the Riemann hypothesis for Dedekind zeta functions of all number fields. Then Pr⁡(χn reducible)=O(e−cn)\Pr(\chi_n\text{ reducible})=O(e^{-cn}) and Pr⁡(Gal⁡(χn/Q)⊉An)=O(e−cn)\Pr(\operatorname{Gal}(\chi_n/\mathbb Q)\not\supseteq A_n)=O(e^{-c\sqrt n}), with constants depending on μ\mu.

The uniform law on {0,1,…,209}\{0,1,\ldots,209\} satisfies the first hypothesis, since 210=2⋅3⋅5⋅7210=2\cdot3\cdot5\cdot7. The fair-sign law fails this hypothesis. Both theorems leave the alternatives AnA_n and SnS_n.

References 1
  • Eberhard2022 · Theorems 1.2 and 1.3Unconditional four-prime theorem and conditional general theorem.Source

Finite-field eigenvalues and Galois orbits

The equality χM(a)=0\chi_M(a)=0 holds exactly when aI−MaI-M is singular. Root counts modulo a prime therefore depend on the rank of a random matrix over Fp\mathbb F_p. Counting tuples of roots requires joint singularity estimates for distinct values of aa.

For a squarefree polynomial with root set Ω\Omega, the orbit-counting lemma identifies the number of GG-orbits on Ω\Omega with the average number of fixed roots of elements of GG. At good primes, the number of roots modulo the prime equals the number fixed by Frobenius. Effective prime-ideal estimates relate these local counts to the Galois average. Higher root correlations correspond to actions on tuples.

References 2

Other matrix ensembles

In the first three rows, the dimension nn tends to infinity and the entry law is fixed. “Dedekind RH” concerns zeta functions of number fields, not only the Riemann zeta function.

Ensemble and independent dataArithmetic hypothesisConclusion and failure bound
Full matrix: all n2n^2 entries independent with a fixed finitely supported nonconstant integer law μ\mu.RH for Dedekind zeta functions of all number fields is a sufficient hypothesis.Pr⁡(χn reducible)=O(e−cn)\Pr(\chi_n\text{ reducible})=O(e^{-cn}); Pr⁡(Gal⁡(χn)⊉An)=O(e−cn)\Pr(\operatorname{Gal}(\chi_n)\not\supseteq A_n)=O(e^{-c\sqrt n}). Constants depend on μ\mu. Eberhard, Theorem 1.3.
Symmetric sign matrix: entries on and above the diagonal are independent fair signs; entries below are determined by symmetry.RH for Dedekind zeta functions of all number fields.Irreducibility failure at most 2e−cn1/42e^{-cn^{1/4}}, with absolute c>0c>0. Ferber–Jain–Sah–Sawhney, Theorem 1.1. This theorem does not assert containment of AnA_n.
Symmetric tridiagonal matrix: diagonal entries are independent with a fixed bounded nonconstant integer law; adjacent off-diagonal entries equal 1.RH for the root fields of every integer-diagonal matrix of this form, as in equations (2)–(3) of the source. RH for all number fields suffices.Irreducible with group AnA_n or SnS_n, except with probability Ce−cnCe^{-cn}. Bary-Soroker–Garzoni–Sodin, Theorem 1.
Fixed-size matrix: entries range over an integer box of height H.No analytic hypothesis; dimension is fixed and H→∞H\to\infty.Exceptional-group counts belong to a height-counting model. Anderson–O’Dorney.

The tridiagonal theorem also allows unbounded laws satisfying Pr⁡(∣V1∣≥k)≤B(log⁡k)−b\Pr(|V_1|\ge k)\le B(\log k)^{-b} for some B,b>0B,b>0, all k≥3k\ge3. Its constants depend only on these tail constants and the second-largest atom. A separate constant-diagonal ensemble in the same paper has additional symmetry and a smaller possible group.

References 4

Connections and evidence

Conditional

Tridiagonal matrices, Galois groups, and spectral uncertainty

Finite-field mixing and Dedekind RH yield an alternating or symmetric Galois group with exponentially high probability. Emmrich’s theorem then turns that arithmetic event into a simultaneous uncertainty inequality for every nonzero vector.

Scope. n≥3n\ge3; HnH_n is real symmetric tridiagonal with adjacent entries one and iid nonconstant integer diagonal entries VjV_j, satisfying Pr⁡(∣V1∣≥k)≤B(log⁡k)−b\Pr(|V_1|\ge k)\le B(\log k)^{-b} for some B,b>0B,b>0, all k≥3k\ge3. The root-field Dedekind RH hypothesis stated below is retained.

Tridiagonal matrices, Galois groups, and spectral uncertaintyFinite-field mixing and Dedekind RH yield an alternating or symmetric Galois group with exponentially high probability. Emmrich’s theorem then turns that arithmetic event into a simultaneous uncertainty inequality for every nonzero vector. Connection 1: Tridiagonal matrix Hₙ leads to Transfer-matrix products. The characteristic polynomial is the upper-left entry of the ordered product of the transfer matrices T(X-V_j). Connection 2: Transfer-matrix products is an input to Root moments and multiplicities. The transfer-matrix random walk supplies prime-averaged higher root-count moments, together with control of proper powers. Connection 3: Root moments and multiplicities is an input to AND. The quantitative root-count and multiplicity estimates enter the arithmetic argument. Connection 4: Dedekind RH for root fields is an input to AND. Dedekind RH supplies the effective prime-ideal estimates needed at the chosen prime range. Connection 5: AND leads to Irreducible; G contains Aₙ. Theorem 1 gives irreducibility and alternating-group containment on one event of probability at least 1-C exp(-cn). Connection 6: Irreducible; G contains Aₙ leads to All eigenvector minors nonzero. For this diagonalizable rational matrix, alternating-group containment forces every minor of an eigenvector matrix to be nonzero. Connection 7: All eigenvector minors nonzero is equivalent to Support uncertainty for every v ≠ 0. For a real orthogonal eigenvector matrix U, nonvanishing of every minor is equivalent to the support inequality for every nonzero complex vector.Tridiagonal matrix Hₙ — related articleTridiagonal matrix HₙTransfer-matrix products — related articleTransfer-matrix productsRoot moments and multiplicities — related articleRoot moments andmultiplicitiesDedekind RH for root fields — related articleDedekind RH for rootfieldsANDANDIrreducible; G contains Aₙ — related articleIrreducible; G contains AₙAll eigenvector minors nonzero — related articleAll eigenvector minorsnonzeroSupport uncertainty for every v ≠ 0 — related articleSupport uncertainty forevery v ≠ 0Connection 1: The characteristic polynomial is the upper-left entry of the ordered product of the transfer matrices T(X-V_j).Determinant recurrence[1]Connection 2: The transfer-matrix random walk supplies prime-averaged higher root-count moments, together with control of proper powers.Finite-field mixing [2]Connection 3: The quantitative root-count and multiplicity estimates enter the arithmetic argument.Joint input [3]Connection 4: Dedekind RH supplies the effective prime-ideal estimates needed at the chosen prime range.Joint input [4]Connection 5: Theorem 1 gives irreducibility and alternating-group containment on one event of probability at least 1-C exp(-cn).Failure ≤ C exp(-cn)[5]Connection 6: For this diagonalizable rational matrix, alternating-group containment forces every minor of an eigenvector matrix to be nonzero.Emmrich: Theorem 2 [6]Connection 7: For a real orthogonal eigenvector matrix U, nonvanishing of every minor is equivalent to the support inequality for every nonzero complex vector.Emmrich: Theorem 10 [7]
Linked nodes open articles; numbered connections open the evidence below.Open full-size diagram

deduction or information flow equivalenceAND joint hypotheses

Put (Hn)jj=Vj(H_n)_{jj}=V_j, (Hn)j,j+1=(Hn)j+1,j=1(H_n)_{j,j+1}=(H_n)_{j+1,j}=1, and all other entries zero. The diagonal law is fixed, nonconstant and integer-valued, with the logarithmic tail bound in the scope. Every bounded nonconstant integer law qualifies. Write Pn(X)=det⁡(XI−Hn)P_n(X)=\det(XI-H_n).

Published theorem, conditional on Dedekind RH. Assume RH for each Dedekind zeta function of Q(a)\mathbb Q(a), where aa is a root of the characteristic polynomial of any integer-diagonal matrix of this tridiagonal form. RH for all number fields is sufficient. Bary-Soroker–Garzoni–Sodin prove that there are constants C,c>0C,c>0, depending on the law, such that

Pr⁡(Pn irreducible and Gal⁡(Pn/Q)∈{An,Sn})≥1−Ce−cn.\Pr\bigl(P_n\text{ irreducible and }\operatorname{Gal}(P_n/\mathbb Q)\in\{A_n,S_n\}\bigr)\ge1-Ce^{-cn}.

The transfer identity is

Pn(X)=(T(X−Vn)⋯T(X−V1))11,T(u)=(u−110).P_n(X)=\bigl(T(X-V_n)\cdots T(X-V_1)\bigr)_{11},\qquad T(u)=\begin{pmatrix}u&-1\\1&0\end{pmatrix}.

Higher root-count moments, rather than just the expected number of roots, detect the required permutation-group action.

Let UU have real orthonormal eigenvectors of HnH_n as columns. On the same event, Emmrich’s theorems give nonvanishing of every minor and, simultaneously for all 0≠v∈Cn0\ne v\in\mathbb C^n,

∣supp⁡v∣+∣supp⁡(UTv)∣≥n+1.|\operatorname{supp}v|+|\operatorname{supp}(U^{\mathsf T}v)|\ge n+1.

The support is the set of nonzero coordinates. Thus a vector cannot be sparse in both the coordinate and eigenvector bases. No additional union bound over vectors is needed. The authors already record this consequence in §1.2; the hypothesis remains attached to the probability estimate.

Inspect all 7 connections and exact references
Evidence for each connection
ConnectionStatement and source
1Tridiagonal matrix Hₙ → Transfer-matrix productsDeduction

The characteristic polynomial is the upper-left entry of the ordered product of the transfer matrices T(X-V_j).

2Transfer-matrix products → Root moments and multiplicitiesConstruction / input

The transfer-matrix random walk supplies prime-averaged higher root-count moments, together with control of proper powers.

3Root moments and multiplicities → ANDConstruction / input

The quantitative root-count and multiplicity estimates enter the arithmetic argument.

4Dedekind RH for root fields → ANDConstruction / input

Dedekind RH supplies the effective prime-ideal estimates needed at the chosen prime range.

5AND → Irreducible; G contains AₙDeduction

Theorem 1 gives irreducibility and alternating-group containment on one event of probability at least 1-C exp(-cn).

6Irreducible; G contains Aₙ → All eigenvector minors nonzeroDeduction

For this diagonalizable rational matrix, alternating-group containment forces every minor of an eigenvector matrix to be nonzero.

7All eigenvector minors nonzero ↔ Support uncertainty for every v ≠ 0Equivalence

For a real orthogonal eigenvector matrix U, nonvanishing of every minor is equivalent to the support inequality for every nonzero complex vector.

Sources and scope checked .

Rational invariant subspaces

An irreducible characteristic polynomial excludes nonzero proper rational invariant subspaces.

Proof

If W⊂QnW\subset\mathbb Q^n is MM-invariant and 0<dim⁡W<n0<\dim W<n, a basis adapted to WW puts MM in block upper-triangular form. Then χM\chi_M is the product of the characteristic polynomials on WW and Qn/W\mathbb Q^n/W, both of positive degree.

Nonsingularity and distinct eigenvalues are insufficient: diag⁡(1,2)\operatorname{diag}(1,2) has both properties and preserves two rational lines.

Questions

OpenCounting and probability

Does irreducibility hold unconditionally for every fixed entry law?

Scope. Independent entries of any fixed nonconstant finitely supported law on Z, with n tending to infinity.

Eberhard’s Conjecture 1.1 concerns irreducibility for arbitrary fixed entry laws. His general theorem assumes ERH, and the unconditional theorem requires uniformity modulo four distinct primes. In particular, the latter theorem does not cover fair signs. Full SnS_n is a further question.

References 1
Discuss this question →
OpenCounting and probability

Are reducible characteristic polynomials the main exception for matrices of bounded height?

Scope. Fixed matrix size n≥2; integer entries of absolute value at most T; T→∞.

Let Mn(T)M_n(T) count matrices whose characteristic polynomial has group different from SnS_n, and Rn(T)R_n(T) those with reducible characteristic polynomial. Anderson–O’Dorney conjecture

Rn(T)∼Mn(T)≍nTn2−n+1log⁡T.R_n(T)\sim M_n(T)\asymp_n T^{n^2-n+1}\log T.

The lower bound comes from integer eigenvalues. Their general upper bound is Mn(T)≪nTn2−1/2log⁡TM_n(T)\ll_n T^{n^2-1/2}\log T. The size nn is fixed here; this is a different problem from bounded entries and growing dimension.

References 2
Discuss this question →

Related topics

References

  1. The characteristic polynomial of a random matrix

    Eberhard, Sean. Combinatorica, 42(4), 491–527, 2022.

    Read the source
  2. Chebotarëv and his density theorem

    Stevenhagen, Peter and Lenstra, Jr., Hendrik W. The Mathematical Intelligencer, 18(2), 26–37, 1996.

    Read the source
  3. Random symmetric matrices: rank distribution and irreducibility of the characteristic polynomial

    Ferber, Asaf and Jain, Vishesh and Sah, Ashwin and Sawhney, Mehtaab. Mathematical Proceedings of the Cambridge Philosophical Society, 174(2), 233–246, 2023.

    Read the source
  4. Irreducibility of the characteristic polynomials of random tridiagonal matrices

    Bary-Soroker, Lior and Garzoni, Daniele and Sodin, Sasha. Journal of Number Theory, 280, 973–998, 2026.

    Read the source
  5. Galois groups of random integer matrices

    Anderson, Theresa C. and O’Dorney, Evan M. Mathematical Proceedings of the Cambridge Philosophical Society, 1–16, 2026.

    Read the source
  6. Walks on groups, counting reducible matrices, polynomials, and surface and free group automorphisms

    Rivin, Igor. Duke Mathematical Journal, 142(2), 353–379, 2008.

    Read the source
  7. Nonvanishing minors of eigenvector matrices and consequences

    Emmrich, Tarek. Special Matrices, 12(1), Article 20240020, 2024.

    Read the source